Improve Radiologists Productivity in Hospitals Based on ... · Improve Radiologists Productivity in Hospitals Based on Data Mining Techniques تايقت مادتساب تايفت
Post on 26-Jun-2020
4 Views
Preview:
Transcript
Improve Radiologists Productivity in Hospitals
Based on Data Mining Techniques
تحسين إنتاجية أطباء األشعة في المستشفيات باستخدام تقنيات البيانات تنقيب
By
Mona Abdul-Fattah El-Sibakhi
Supervised by
Dr. Tawfiq Barhoom
Associate prof. of Applied Computer Technology
A thesis submitted in partial fulfillment
of the requirements for the degree of
Master of Information Technology
September, 2017
زةــغب ةــالميــــــة اإلســـــــــامعـالج
البحث العلمي والدراسات العليا عمادة
تكـــنولوجيـــــا المعلومـــــاتة ليــــــك
المعلومــــات اتكـــنولوجيــــ ماجستير
The Islamic University of Gaza
Deanship of Research and Postgraduate
Faculty of Information Technology
Master of Information Technology
I
إقــــــــــــــرار
أنا الموقع أدناه مقدم الرسالة التي تحمل العنوان:
Improve Radiologists Productivity in Hospitals
Based on Data Mining Techniques
تحسين إنتاجية أطباء األشعة في المستشفيات باستخدام تقنيات تنقيب البيانات
أقر بأن ما اشتملت عليه هذه الرسالة إنما هو نتاج جهدي الخاص، باستثناء ما تمت اإلشارة إليه حيثما ورد، وأن
لنيل درجة أو لقب علمي أو بحثي لدى أي مؤسسة اآلخرينهذه الرسالة ككل أو أي جزء منها لم يقدم من قبل
تعليمية أو بحثية أخرى.
Declaration
I understand the nature of plagiarism, and I am aware of the University’s policy on
this.
The work provided in this thesis, unless otherwise referenced, is the researcher's own
work, and has not been submitted by others elsewhere for any other degree or
qualification.
:Student's name السباخي الفتاح عبد منى الطالب:اسم
:Signature التوقيع:
:Date التاريخ:
III
Abstract
Modern radiology departments have enormous databases of images and text. Like any
databases, which are rich in data content, but poor in information content. Data Mining
is an effective tool that extracts useful information from this enormous database of
images and text which helps decision makers in departments and hospitals to take
proper decisions.
In this research, the idea investigates some problems in radiology departments at
hospitals based on applying Data Mining techniques and conducting Data Mining
model to improve radiologists productivity by assigning the appropriate cases to
appropriate radiologists within tele-radiology environment. Due to the heavy load of
work assigned to radiologists, there is significant delay in writing radiology reports by
them.
Data with seven feature sets were collected from four hospitals in Saudi Arabia
covering eight radiologists (two from each hospital) with varying productivity and
specialisation with emphasis on CT, MRI and Mammography modalities. Four
different classifiers were applied for the dataset to predict and assign the suitable cases
for each radiologist to improve radiologists productivity.
The model was evaluated by presenting its results to an expert in one of the four
hospitals for his opinion. He declared that the results of the model are very good as
they take into account the subspecialty of each procedure in assigning the cases. He
also believes that applying the model in hospitals will achieve good results and
improve the radiologists productivity.
Accuracy and F-measure evaluation performance measures were applied to compare
among the classifiers. The results show that the Naïve Bayes was the best classifier in
improving the productivity of radiologists, it improved the productivity by up to 24%
as it assigned the appropriate case to the appropriate radiologist. Naïve Bayes had the
highest value in Accuracy and F-measure by up to 8% in accuracy and 4% in F-
measure.
Keywords: Radiology, Data Mining, Classification, Productivity.
IV
الملخص
، تكون قواعد أي قواعد بياناتلدى أقسام األشعة الحديثة قواعد بيانات ضخمة من الصور والنصوص. وكما
في محتوى البيانات ولكنها ضعيفة في استخراج المعلومات. البيانات هذه غنية
لذلك يمكن اعتبار تنقيب البيانات أداة ذات كفاءة وفعالية الستخراج المعلومات من قواعد البيانات الضخمة وتقديمها
بشكل مفيد يدعم اتخاذ القرار في تلك األقسام والمستشفيات.
اكل أقسام األشعة في المستشفيات وتقديم الحل على أساس تطبيق تستند فكرة هذا البحث على دراسة إحدى مش
بهدف تحسين إنتاجية أطباء األشعة من خالل إسناد الحاالت وإنشاء نموذج تنقيب البيانات تقنيات تنقيب البيانات
المناسبة لكل طبيب أشعة وذلك في بيئة قراءة األشعة عن بعد.
إنجاز التقارير خالل الوقت المناسب بسبب زيادة عبء بعض أطباء األشعة.تواجه أقسام األشعة حالًيا تأخيًرا في
من كل طبيبان“ثمانية أطباء تتضمنتم جمع البيانات من أربع مستشفيات مختلفة في المملكة العربية السعودية
وتمت مراعاة تفاوت األطباء في التخصص واإلنتاجية.” مستشفى
تصوير الثدي، وتضمنت لمقطعي، التصوير بالرنين المغناطيسي و اقتصر البحث على فحوصات التصوير ا
سبع صفات. البيانات
تم تطبيق أربع أدوات مختلفة لتصنيف البيانات على مجموعة البيانات، وتنبأ النموذج بالحالة المناسبة لكل طبيب
إلى توزيع الحاالت بالشكل األمثل لتحسين إنتاجية األطباء. أدىأشعة مما
بأن قالو ، النتائج على خبير في واحدة من المستشفيات األربع إلبداء رأيهم النموذج من خالل عرض تم تقيي
على األطباء الحاالت توزيعاالعتبار التخصص الفرعي لكل إجراء في بعينذ تأخألنها نتائج النموذج جيدة
جيدة وسيعمل على تحسين إنتاجية األطباء. نتائج ويعتقد أنه عند تطبيق النموذج في المستشفيات سوف يحقق
. وأظهرت النتائج أن F-measureو Accuracyم تقييم نتائج أدوات التصنيف المستخدمة باستخدام أيضا ت
Naive Bayes حيث أدت إلى تحسن أداء كانت أفضل أداة في إسناد الحاالت المناسبة لكل طبيب أشعة ،
بنسبة تصل Accuracyالوقت حصلت على أعلى نسبة في تقييم . وفي نفس%24األطباء بنسبة تصل إلى
% مقارنة بأدوات التصنيف األخرى.4بنسبة تصل إلى F-measureو% 8إلى
V
Epigraph Page
VI
Acknowledgment
Thanks to Almighty Allah for giving me strength and ability to understand, learn and
complete this research.
With great pleasure, I would like to express my deepest gratitude to my supervisor Dr.
Tawfiq Barhoom for his unwavering support and mentorship throughout this research.
I also greatly thank my Mum and Dad who paved the path for me and upon whose
shoulders I stand. This is dedicated to my family and the many friends who supported
me during this journey, Thank you.
Special thanks to my dear husband for his direct and indirect support to complete this
research.
VII
Table of Contents
Declaration .................................................................................................................. I
Abstract ..................................................................................................................... III
Epigraph Page ............................................................................................................ V
Acknowledgment ...................................................................................................... VI
Table of Contents .................................................................................................... VII
List of Tables .............................................................................................................. X
List of Figures ........................................................................................................... XI
List of Abbreviations .............................................................................................. XII
Chapter 1 Introduction .............................................................................................. 1
1.1 Background and Context .................................................................................... 2
1.2 Statement of the Problem .................................................................................... 3
1.3 Objectives ........................................................................................................... 3
1.3.1 Main Objective ................................................................................................ 3
1.3.2 Specific Objectives .......................................................................................... 3
1.4 Importance of the research .................................................................................. 4
1.5 Scope and Limitations ........................................................................................ 4
1.6 Methodology ....................................................................................................... 4
1.7 Overview of research .......................................................................................... 5
Chapter 2 Background ............................................................................................... 6
2.1 Overview of Data Mining ................................................................................... 7
2.2 Rapid Miner ........................................................................................................ 8
2.3 Data Mining Classification Techniques .............................................................. 8
2.3.1 Decision Tree ................................................................................................... 8
2.3.2 Naïve Bayes ..................................................................................................... 8
2.3.3 Random Forest ................................................................................................. 9
2.4 Performance Evaluation ...................................................................................... 9
2.4.1 Confusion Matrix ............................................................................................. 9
2.4.2 Accuracy (AC) ............................................................................................... 10
2.4.3 Precision (P) ................................................................................................... 10
2.4.4 Recall ............................................................................................................. 10
2.4.5 F-measure ....................................................................................................... 10
2.5 Data Mining in Healthcare ................................................................................ 10
2.6 Data Mining versus Statistics ........................................................................... 11
2.7 Radiology Information System ......................................................................... 11
VIII
2.8 Radiology and Tele-radiology in hospitals ....................................................... 12
2.9 Relative Value Units (RVUs) ........................................................................... 13
2.9.1 RVUs in Radiology ........................................................................................ 13
Chapter 3 Related Works ......................................................................................... 14
3.1 Radiologists Productivity Measurements ......................................................... 15
3.2 Data Mining for Measuring Indicators ............................................................. 16
3.3 Data Mining for Diagnosis Diseases ................................................................ 17
3.4 Related Work Discussion .................................................................................. 19
3.5 Summary ........................................................................................................... 19
Chapter 4 The Data Mining Model ......................................................................... 20
4.1 General View of Model .................................................................................... 21
4.2 Model Details .................................................................................................... 22
4.3 Model Iterations ................................................................................................ 22
4.3.1 The Initial Iteration ........................................................................................ 22
4.3.2 The Next Iteration (one or more) ................................................................... 25
4.4 Summary ........................................................................................................... 26
Chapter 5 Methodology ............................................................................................ 27
5.1 Methodology Steps ........................................................................................... 28
5.2 Data Acquisition and Collection ....................................................................... 28
5.3 Data Pre-processing and Feature Sets Selections ............................................. 30
5.3.1 Generating New Columns .............................................................................. 30
5.3.2 Combining Data ............................................................................................. 32
5.3.3 Feature Set Selection ..................................................................................... 32
5.4 Testing Data ...................................................................................................... 33
5.5 Implementation ................................................................................................. 34
5.5.1 Tools .............................................................................................................. 34
5.6 Evaluation ......................................................................................................... 35
5.7 Summary ........................................................................................................... 35
Chapter 6 Results, Discussion and Evaluation ....................................................... 36
6.1 Classification Methods Settings ........................................................................ 37
6.2 Experimental Results ........................................................................................ 37
6.3 Evaluation ......................................................................................................... 40
6.3.1 Performance Evaluation Results .................................................................... 40
6.4 Summary ........................................................................................................... 42
Chapter 7 Conclusions and Future Work .............................................................. 43
7.1 Conclusion ........................................................................................................ 44
IX
7.2 Future Work ...................................................................................................... 45
References .................................................................................................................. 46
References .................................................................................................................. 47
Appendices ................................................................................................................. 50
Appendix A: Reported Cases Statistics .................................................................. 51
Appendix B: Sample of Exam Code Dictionary ..................................................... 53
X
List of Tables
Table (3.1): Summary of the Most Related Works to this Work ............................... 19
Table (4.1): Previous Productivity ............................................................................. 24
Table (4.2): Joined Data ............................................................................................. 24
Table (4.3): Calculating Current Productivity ........................................................... 24
Table (4.4): The Work of the Loop ............................................................................ 25
Table (5.1): Description of Figure 5.2 Columns ........................................................ 29
Table (5.2): FDA Age Classifications (FDA, 2014) .................................................. 31
Table (6.1): Classifiers Settings ................................................................................. 37
Table (6.2): Auto Assigned Cases ............................................................................. 39
Table (6.3): Performance Evaluation Results ............................................................ 40
Table (6.4): Wrongly Assigned Cases ....................................................................... 41
XI
List of Figures
Figure (2.1): Confusion Matrix .................................................................................... 9
Figure (4.1): The Model ............................................................................................. 21
Figure (4.2): The Initial Iteration of the Model ......................................................... 23
Figure (4.3): The Next Iteration of the Model ........................................................... 26
Figure (5.1): Methodology Steps ............................................................................... 28
Figure (5.2): Data Before Pre-processing .................................................................. 29
Figure (5.3): Exam Code Dictionary ......................................................................... 30
Figure (5.4): Generate Age Group ............................................................................. 30
Figure (5.5): Generate Time to Report ...................................................................... 31
Figure (5.6): Generate Reporting Time ..................................................................... 31
Figure (5.7): Generate Exam Day .............................................................................. 32
Figure (5.8): Combined Data ..................................................................................... 32
Figure (5.9): Training Data Set .................................................................................. 32
Figure (5.10): Testing Data Before Pre-processing ................................................... 33
Figure (5.11): Combined Data ................................................................................... 33
Figure (5.12): Testing Data ........................................................................................ 34
Figure (5.13): Rest Data ............................................................................................. 34
Figure (6.1): The Initial Iteration in Rapid Miner ..................................................... 38
Figure (6.2): Auto Assigned Cases ............................................................................ 38
Figure (6.3): Next Iteration in Rapid Miner .............................................................. 39
Figure (6.4(: Productivity Comparison ...................................................................... 40
Figure (6.5): Performance Evaluation Results ........................................................... 41
Figure (6.6): Naive Bayes Classifier Results ............................................................. 41
Figure (7.1): Productivity Comparison ...................................................................... 45
XII
List of Abbreviations
AC Accuracy
ANN Artificial Neural Network
CAD Coronary Artery Disease
CT Computerized Tomography
CV Coefficient of Variation
FDA Food and Drugs Administration
FTE Full-Time Equivalent
HIS Hospital Information Systems
KDD Knowledge Data Discovery
KNN K-Nearest Neighbour
LAD Left Anterior Descending
LCX Left Circumflex
LOS Length of Stay
L-RVU Local Relative Value Units
MRI Magnetic Resonance Imaging
P Precision
PACS Picture Archiving and Communication System
RCA Right Coronary Artery
RIS Radiology Information System
ROC Receiver Operating Characteristic
RVUs Relative Value Units
SCI Spinal Cord Injuries
SVM Support Vector Machines
TASH Total Available Staffed Hours
TCM Traditional Chinese Medicine
US Ultrasonography
X-Ray Conventional Radiography
Chapter 1
Introduction
2
Chapter 1
Introduction
1.1 Background and Context
Nowadays Healthcare industry produces massive amounts of complex data about
patients, hospitals resources and disease diagnosis. This massive amount of data is a
key resource to be processed and analyzed for knowledge extraction that enables
support for cost-savings and decision making (Desikan, Hsu, & Srivastava, 2011).
Previous studies solved the problems that concentrated on the prediction and diagnosis
of heart diseases and breast cancer in addition to measure RVUs and productivity for
radiologists. Each study has its own Data Mining techniques which gives it points of
strength but at the same time some limits. In this research, the idea investigates some
problems in radiology departments at hospitals with the aim to construct a model to
improve radiologists productivity which helps in identifying the hospitals that are in
actual need for radiologists and to apply auto assigning in Tele-radiology. Due to the
heavy load of work assigned to radiologists, there is significant delay in writing
radiology reports by them and some reports have been pending for days or weeks.
However, the conventional Tele-radiology procedure provided by hospitals is limited
and has never overcome such delay. Data Mining has been utilized to improve the
radiologists productivity by assigning the appropriate cases to targeted radiologist. The
main objective of this research is to construct a model to improve radiologists
productivity in hospitals. This research focuses on applying Data Mining techniques
on data from different hospitals in different areas of Saudi Arabia. Data Mining brings
a set of tools and techniques that can be applied to the processed data to discover
hidden patterns that provide healthcare professionals with an additional source of
knowledge for making decisions (Desikan, Hsu, & Srivastava, 2011). Radiologists
productivity was measured, then all cases were reassigned to all radiologists to apply
auto assign Tele-radiology environment, Tele-radiology means that the radiologist
could conduct writing reports for different hospitals.
3
1.2 Statement of the Problem
Nowadays, radiology has a vital role in medical diagnosis process. Computerized
Tomography (CT), Magnetic Resonance Imaging (MRI) and Mammography are the
most popular modalities of radiology. Radiologists encountered delays in writing
radiology reports and some reports were kept pending for a long time, mainly those
related to CT, MRI and Mammography. Radiologists productivity percentages vary
and may be above the acceptable range. The conventional Tele-radiology procedure
provided by hospitals is limited and has never overcome such delays. Data Mining has
been utilized to improve the radiologists productivity by assigning the appropriate
cases to targeted radiologist.
1.3 Objectives
1.3.1 Main Objective
The main objective of this research is to construct a model to improve radiologists
productivity in hospitals.
1.3.2 Specific Objectives
The specific objectives of the research are:
1. Data aggregation from different hospitals.
2. Pre-processing and Analyzing data to prepare it for implementing Data Mining
techniques.
3. Data Modelling: implementing Data Mining classification techniques such as
Decision Tree, Naïve Bayes, K-NN and Random Forest.
4. Measuring Radiology productivity percentage by using the traditional way before
applying Data Mining techniques.
5. Assigning appropriate radiology procedure for appropriate radiologist by applying
Data Mining classification techniques within Tele-radiology environment
4
7. Evaluating the model by presenting the results to an expert from one of the four
hospitals for his opinion about the model. Also, an evaluation to classification
techniques was done by using different evaluation measures to evaluate the
performance and to compare among them.
1.4 Importance of the research
Despite the differences and inconsistency in approaches, the radiology system is in
more need for Data Mining today. There are some arguments that could support the
use of Data Mining in the radiology system (D. & Jr., 2009). Radiologists who must
write radiology reports delay writing such reports or keep them pending for quite a
long time. Together with the radiologists productivity problems, this led to implement
Data Mining techniques which have ability to improve the productivity of radiologists.
1.5 Scope and Limitations
The research focused on applying Data Mining techniques on data for eight
radiologists from four different hospitals in different areas of Saudi Arabia.
Radiologists productivity was measured, then all cases were reassigned to all
radiologists to apply auto assigned Tele-radiology environment.
Hospitals face a lot of problems and delay of reports in CT, MRI and Mammography
because of the large number of cases and patients. So, in this research the proposed
model concentrates on these modalities.
The radiologists workload is limited to clinical work only, other work like
administration, teaching and conferences are not considered.
1.6 Methodology
The methodology of this research consists of the following phases:
Phase 1: Data Acquisition and Collection
Data were collected for eight radiologists from four hospitals in Saudi Arabia; one data
set for a year was collected for training data and another for a month for testing data.
5
The training and testing datasets contained data of the three modalities which the
research concentrated on, i.e. CT, MRI and Mammography.
Phase 2: Data Pre-processing
Data were pre-processed, and a dataset of training and testing data was built. Seven
feature sets were created: Radiologist as a label, Visit_Class, Age_Group, Body_Part,
Reporting_time, Exam_Day and wRVU. For training data, 13,142 records were
selected for a year and for testing data, 1,290 records were selected for a month.
Phase 3: Implementation
Rapid Miner Programme was used to implement Data Mining techniques. Decision
Tree, Naïve Bayes, K-NN and Random Forest classification Data Mining techniques
were applied, data for a year used as training data set and data for a month used as
testing data set.
Phase 4: Evaluation
After applying classification methods, the model was evaluated by presenting the
results to an expert from one of the four hospitals for his opinion about the model
Accuracy and F-measure evaluated the performance of the model.
1.7 Overview of research
The research is divided into seven chapters; chapter one includes the Introduction,
chapter two provides Literature Review, chapter three includes related works in Data
Mining healthcare, chapter four provides the model and its component, chapter five
provides description of the methodology, chapter six includes the analysis of
experiments results and evaluation and chapter seven talks about conclusion and future
work.
Chapter 2
Background
7
Chapter 2
Background
This chapter presents the background and theoretical concepts of the Data Mining
techniques applied in this research. It starts by discussing the importance of Data
Mining techniques in healthcare domain, and clarifies the statistics versus Data
Mining. The last part presents a background about Radiology Systems in hospitals,
Tele-radiology system and Relative Value Units (RVUs).
2.1 Overview of Data Mining
Data Mining is considered to be a recently developed methodology and technology,
starting in 1994 aiming to identify valid, new, useful, and understandable data
correlations and patterns (Koh & Tan, 2005).
It can be considered as the process of extracting important information from large set
of data utilizing the relationship between the data. It is also exemplified as Knowledge
Data Discovery (KDD) and many consider that it is almost impossible to distinguish
between the two. Many also consider Data Mining to be a very vital step in KDD (
Tapedia & Wagh , 2016).
Data Mining is a variety of methods and techniques utilized in different analytical
patterns to address a range of organizational needs (SAS Institute Inc, 2012).
Descriptive Modelling: It reveals similarities or groupings in historical data to know
success or failure causes such as Clustering which groups similar records together and
Association Rule Learning which shows relationships between records (SAS Institute
Inc, 2012).
Predictive Modelling: This modelling classifies future events or estimates unknown
outcomes – for example, using credit scoring to determine an individual's likelihood
of repaying a loan. Similar to Decision Tree, it is tree-shaped diagrams in which each
branch represents a probable occurrence and Support Vector Machine which is
supervised learning models with related learning algorithms (SAS Institute Inc, 2012).
8
2.2 Rapid Miner
Rapid Miner is a data science software developed to provide a cohesive environment
for data preparation, machine learning, deep learning, text mining, and predictive
analytics. It used for business and commercial applications and for research, education,
training. Rapid Miner supports all steps of the machine learning process including data
preparation, results visualization, model validation and optimization.
2.3 Data Mining Classification Techniques
Classification is the most commonly applied technique in Data Mining, which utilizes
a set of pre-classified examples to develop a model through which the population of
records can be classified at large. This method usually employs Decision Tree or
Neural Network-based classification algorithms. The data classification process
involves learning and classification. Learning analyses the training data by
classification algorithm. Classification uses test data to estimate the accuracy of the
classification rules. These pre-classified examples are used by the classifier-training
algorithm to determine the set of parameters required for proper discrimination. These
parameters are then encoded by the algorithm into a model called a classifier (Mlambo
, 2016). In this research, a variety of classification methods with different feature sets
were used such as: Decision Tree, Naïve Bayes, K-NN and Random Forest.
2.3.1 Decision Tree
Decision Tree is a graphical representation of the relations existing between the data
and is used for data classification. The result is displayed as a tree, hence the name of
this technique. Decision Tree is a simple and a powerful way of representing
knowledge and is mainly used in the classification and prediction. The models
obtained from the decision tree are represented as a tree structure (Milovic & Milovic,
2012).
2.3.2 Naïve Bayes
The Bayesian Classification represents a supervised learning method and a statistical
method for classification. It supposes a probabilistic model and allows to capture
9
uncertainty about the model by determining probabilities of the outcomes. It solves
diagnostic and predictive problems (Hachesu, Ahmadi, Alizadeh, & Sadoughi, 2013).
2.3.3 Random Forest
The Random Forest is a group of unpruned classification trees. It generates many
classification trees where each tree is constructed by a different sample from the
original data using a tree classification algorithm. After the forest is formed, a new
object that needs to be classified is put down each tree in the forest for classification.
Each tree gives a vote indicating its decision about the class of the object. Then the
forest chooses the class with the most votes for the object (Al Mehedi , Nasser, Pal, &
Shamim, 2014).
2.4 Performance Evaluation
Performance Evaluation aims to provide an equitable measurement of a model to
produce accurate evaluation and to obtain a high level of quality and quantity in the
results produced (Capko, 2003).
2.4.1 Confusion Matrix
A confusion matrix is a simple performance analysis tool used in supervised learning.
It is used to represent the test result of a prediction model. Each row of the matrix
represents the instances in a predicted class, while each column represents the
instances in an actual class (M, 2012).
Figure (2.1): Confusion Matrix
10
2.4.2 Accuracy (AC)
Accuracy is the proportion of the total number of predictions that were correct. It is
determined using the equation:
(2.1)
2.4.3 Precision (P)
Precision is the proportion of the predicted positive cases that were correct. It is
calculated using the equation:
(2.2)
2.4.4 Recall
The recall is the positive cases that were correctly identified. It is calculated using the
equation:
(2.3)
2.4.5 F-measure
A measure that combines precision and recall is the harmonic mean of precision and
recall. It is calculated using the equation:
(2.4)
2.5 Data Mining in Healthcare
Many organizations have used Data Mining, Data Mining becoming very popular in
healthcare. Its applications can greatly benefit all those involved in the healthcare
True Positive (TP): If the instance is positive and it is classified as positive
False Positive (FP): If the instance is negative but it is classified as positive
True Positive (TP): If the instance is positive and it is classified as positive
False Negative (FN): If the instance is positive but it is classified as negative
11
industry. They can help healthcare insurers detect fraud and abuse, healthcare
organizations make customer relationship management decisions, physicians identify
effective treatments and best practices and patients receive better and more affordable
services. Healthcare transactions produce great amounts of data which are usually
complicated and huge in size. It is difficult to process and analyze these data by usual
methods. Data Mining can provide the means and technology to change the data into
useful information to help decision makers to take proper decisions (Koh & Tan,
2005).
2.6 Data Mining versus Statistics
Data Mining is the process of extracting unidentified information from large databases
and using it to make decisions. It is a set of methods used in the knowledge discovery
process to distinguish previously unknown relationships and patterns within data. Data
Mining that provides the tools and analytics techniques for dealing with huge amounts
of data contains statistics. It is the science of learning from data and includes
everything from collecting and organizing to analyzing and presenting data. Data
Mining and Statistics are related to learning from data. They are about discovering and
identifying structures in them, thus aim to turn data to information. Both techniques
have different approaches. Although their aims overlap. Statistics is only about
quantifying data. While it uses tools to find relevant properties of data, it is very much
like math. It provides the tools necessary for Data Mining. On the other hand, Data
Mining builds models to detect patterns and relationships in data, particularly from
large data bases (GS, 2015).
2.7 Radiology Information System
A Radiology Information System (RIS) is the main system for the management of
imaging departments. Its major functions include scheduling patients, managing
resources, tracking examination performance, interpreting examinations, distributing
results, and billing procedure. RIS complements Hospital Information Systems (HIS)
and Picture Archiving and Communication System (PACS), and is critical to efficient
workflow to radiology practices.
12
Radiology in healthcare provides diagnostic imaging services for patients using
Computed Tomography (CT), Magnetic Resonance Imaging (MRI), radionuclide
imaging (nuclear medicine), UltraSonography (US), Conventional Radiography (X-
ray) and interventional procedures using advanced image-guided techniques. These
types of images are archived in PACS and radiology workflow is managed by RIS.
2.8 Radiology and Tele-radiology in hospitals
The Department of Radiology has an ongoing programme for monitoring, evaluating,
and assuring quality services by the department. The programme has integrated in the
hospital's overall quality assessment and improvement plan. The Radiology
Department has a quality system that ensures compliance with accreditation
requirements. Quality planning and evaluating the effectiveness and efficiency of the
Quality system is conducted through scheduled leadership reviews. The Hospital
Director is very firmly committed to support all activities, participations, and
implementation of the Quality system. The department supports the Quality system
and contributes data collection for quality reports. The Department provides services
like conventional X-Rays, Fluoroscopy, Colour Doppler Studies, and Ultra-
Sonography (Radiology | Al Yousuf Hospital, Al Khobar, KSA, 2013).
Tele-Radiology is the electronic transmission of radiographic images from one
geographical location to another for interpretation and consultation. It allows various
users in different locations to view images simultaneously. The main applications of
Tele-Radiology provide radiological expertise at remote sites more quickly than would
otherwise be possible (Ahmed & Aldosh, 2014).
Need of Tele Radiology in Saudi Arabia: Saudi Arabia, faces a variety of health
challenges. There is a shortage of Community Health Centres and specialists at these
Centres and hospitals. Particularly, Radiology departments are understaffed where
staffing shortages are occurring at a time when radiology volume generally is
increasing. The gap between demand and supply of quality radiologist is always
increasing. There are various underlying reasons why the supply of radiologists is
insufficient to meet the demand in many areas. Tele-Radiology services in Saudi
Arabia are going to setup across various district hospitals in the southern region where
13
Tele-Radiology solution will help to manage the data and streamline the image flow
from all district hospitals. Tele-Radiology system, offers a comprehensive enterprise
class hospital. (Ahmed & Aldosh, 2014).
2.9 Relative Value Units (RVUs)
Relative value unit (RVU) is a measure of value used in the United States Medicare
for radiologist services, it is considered to be the primary measure of a radiologist’s
productivity. In the past, there were several ways of calculating radiologist
productivity, nowadays these ways focus on models based on Relative Value Units
(RVUs). RVUs which reflect the relative level of time and skill required of a
radiologist to provide a given service. RVUs are a method for calculating the volume
of work or effort expended by a radiologist in treating patient (MERRITT HAWKINS
an AMN Healthcare Company, 2011).
2.9.1 RVUs in Radiology
In the past, tracking productivity was not a big issue due to the surplus of radiologists.
Today radiology practices aim to have the most qualified staff to interpret the volume
of images and not lose money to competitors. RVUs have a correct and accurate
measurement on radiologists (Forrest, 2007). There are three components of a
Medicare RVU: Work RVU (wRVU) ≈ 52%, which is Relative time, effort, and skill
needed by a radiologist in providing a procedure, Practice Expense RVU (peRVU) ≈
44%, it is Costs associated with maintaining a practice, and Malpractice Expense RVU
(mRVU) ≈ 4% m, it is Professional liability insurance (Kuehn, 2009).
Chapter 3
Related Works
15
Chapter 3
Related Works
Many of researches works concentrate on Data Mining in Healthcare. Data mining is
becoming increasingly popular and essential (Koh & Tan, 2005). The Following are
related works that uses Data Mining techniques in Healthcare.
3.1 Radiologists Productivity Measurements
Researchers (dora , Faccin , & Fogliatto , 2016) developed a local RVU (L-RVU)
system to measure radiologists reporting productivity and workload focusing on CT
exams. A method to normalize exams according to the anatomical region (Body Part)
was developed. A time-based measure of radiologist reporting workload was built, a
query that searched for all CT reports (from July 1st, 2013 to February 28th, 2015)
was performed from Radiology Information System (RIS). The query resulted in
42,382 instances for 24 tests performed by the CT Unit. A list of 17 categories
(anatomical region) was proposed, then these categories of tests were normalized with
the shortest reporting time as the reference test. The result total RVUs (Productivity)
did not have target value.
Researchers (Cowan, MacDonald, & Floyd, 2013) focused on measuring radiologists
RVUs based on reporting times using data from a Radiology Information System (RIS)
for all reports generated from 1 January 2010 to 30 June 2012 including CT, MRI, US
and X-Ray modalities. A technique for semi-automated measurement of radiologist
reporting time and measuring the time required for radiologists to produce reports
during normal work was created. A sample of reporting times was recorded by the
Radiology Information System using voice recognition system with the description of
each examination and placed in a database. The study was limited to consultant
radiologists. Relative Value Units (RVUs) were calculated using the reporting time for
a single view chest X-ray of 1 min 38 s. The researchers categorized the data based on
modality and exam description e.g. CT abdomen pancreas, an examination is defined
by a single modality in the same visit, even though it covered more than one anatomical
area. So, CT of head plus chest is one examination. This led to wrong RVUs results.
16
The researcher (Brady, 2011) reported a survey of Consultant Radiologist workload
in Ireland in 2009 for measuring radiologist workload, Relative Value Units (RVUs)
were assigned. Hospitals’ data were collected for the full calendar year of 2009. The
2006 Australian survey recommended 40,000 RVU per radiologist. In 2009 the same
methodology to measure RVU in a larger and broader sample was applied, the results
found that the RVU level had risen to 45,000 RVU per radiologist. On the other hand,
the researcher recalculated the RVU value and his results showed that the value was
57,659.1, but with taking into account the non-clinical work (teaching and
administration), the value rose to 103,987. These results showed that radiologist
staffing levels were already more than appropriate international value of RVU. This
survey did not suggest any solutions to overcome the overload of radiologists.
Researchers (Radiology, 2013) measured the required RVU which should be done by
radiologist per a staffed hour. The study took into account the radiologists vacations,
non-clinical work and on-call duties. Some definitions and formulas were applied to
get results, the result of the calculations yielded to some indexes: productivity index is
the average professional component work RVUs per available staffed hour,
availability index is a measure of the time of radiologists availability relative to the
number of working hours in a business year, and intensity indicator is a measure of the
degree of difficulty of the procedures performed by the practice. The study did not
consider the time taken to write reports.
3.2 Data Mining for Measuring Indicators
The researcher (Lai, 2015) analyzed the use of TCM by employed Complete datasets
of Traditional Chinese medicine (TCM) outpatient from 2005 to 2007, the
characteristics of TCM patients, and the disease categories that were treated by TCM
in Taiwan. The result of this study showed that female use TCM more frequently than
male. The reasons for this female majority were not fully explained in previous reports.
It was suggested that independent females or females of good social status, had higher
expectations and belief in TCM in respect of postpartum conditions, menopause and
chronic diseases. The age distribution of TCM users peaked in the 20-29 group,
followed by the 10-19 group and 31-39 group.
17
Researchers (Hachesu, Ahmadi, Alizadeh, & Sadoughi, 2013) provided a model to
predict the Length of Stay (LOS) of heart patients. Data were collected from patients
with Coronary Artery Disease (CAD). Records of 4,948 patients who had suffered
from CAD were included in the analysis. Classification techniques were used with
three algorithms: Decision Tree, Support Vector Machines (SVM) and Artificial
Neural Network (ANN). LOS was the target variable. The overall accuracy of SVM
was 96.4% in the training set. single patients had an LOS ≤5 days with percentage of
64.3%, whereas 41.2% of married patients who had an LOS >10 days.
3.3 Data Mining for Diagnosis Diseases
Researchers (Shukla, Gupta, & Prasad, 2016) presents the importance and usefulness
of different Data Mining techniques such as Classification, Clustering, Decision trees
and Naïve Bayes. Comparison is done of different Data Mining techniques used for
prediction of cancer disease with different accuracy. The techniques are effective for
identifying hidden cancer aggregation pattern and for classification of familiar risk by
the help of providing better accuracy in many cases as compared to other techniques.
Researchers (Dubey & chandrakar , 2015) presented a systematic review of the
application of Data Mining methods to solve the problems in healthcare domain, the
research aimed to use Data Mining techniques for the diagnosis and prognosis of
different heart diseases. The study discussed how different types of Data Mining
techniques were used for diagnosis of heart diseases and how to perform better results
when it applied on different data sets. Each technique was unique, which might be
suitable for different applications. Hybrid Data Mining techniques showed promising
results in the diagnosis of heart diseases.
Researchers ( Bellaachia & Guven , 2006) Used Data Mining techniques, Experiments
to predict the survivability rate of breast cancer was presented. Experiments were
conducted using Naïve Bayes, Neural Network and the C4.5 Decision Tree algorithms.
The accuracy of three Data Mining techniques was compared. The goal was to have
high accuracy, besides high precision and recall metrics, C4.5 algorithm had a much
better performance than the other two techniques ( Bellaachia & Guven , 2006)
18
In (Lundina , et al., 1999) ANN was applied on 951 instances dataset of Turku
University Central Hospital and City Hospital of Turku to evaluate the accuracy of
neural networks in 5, 10 and 15 years’ for predict breast cancer specific survival. The
values of Receiver Operating Characteristic ROC curve for 5 years were evaluated as
0.909, for 10 years 0.086 and for 15 years 0.883. These values were used as a measure
of accuracy of the prediction model. They found that ANN predicted survival with
higher accuracy.
Researchers (Choi, Han, & Park, 2009) compared the performance of an Artificial
Neural Network, a Bayesian Network and a Hybrid Network used to predict breast
cancer prognosis. The hybrid Network combined both ANN and Bayesian Network.
The accuracy of ANN (88.8%) and Hybrid Network (87.2%) were very similar and
they both outperformed the Bayesian Network. The proposed Hybrid model can also
be useful to take decisions.
In (Shouman, Turner, & Stocker, 2012) KNN was applied to help healthcare in the
diagnosis of heart disease. It was also integrated with voting to enhance the accuracy
in the diagnosis of heart disease patients. The results showed that KNN achieved a
higher accuracy more than neural network in the diagnosis of heart disease patients.
The results also showed that applying voting could not enhance the KNN accuracy in
the diagnosis of heart disease.
The study in (Alizadehsani, et al., 2013). aimed to use Data Mining algorithms to
predict the stenosis of arteries. Among many people who were referred to hospitals
due to chest pain, a great number of them were normal and as such did not need
angiography. The objective of this study was to predict patients who were most
probably normal using features with the highest correlations with coronary artery
disease (CAD) with a view to obviate angiography costs and complications, Bagging
and C4.5 classification algorithms were applied to analyse the data, the accuracy rates
of 79.54%, 61.46%, and 68.96% for the diagnosis of the stenosis of the Left Anterior
Descending (LAD), Left Circumflex (LCX), and Right Coronary Artery (RCA),
respectively. The accuracy to predict the LAD stenosis was attained via feature
selection.
19
3.4 Related Work Discussion
Table (3.1): Summary of the Most Related Works to this Work Research Name Description Short come
The use of relative value
units to monitor
radiologists’ reporting
productivity and workload
This work aimed to develop
local RVU (L-RVU) system to
measure radiologists reporting
productivity and workload.
The result total RVUs
(Productivity) did not have
target value.
Measuring and managing
radiologist workload
This study focused on
measuring radiologists RVUs
based on reporting times using
data from a Radiology
Information System (RIS),
concentrated on CT, MRI, US
and X-Ray modalities. A
technique measuring the time
required for radiologists to
produce reports during normal
work was created.
The researchers categorized
the data based on modality
and exam description e.g
CT abdomen pancreas, an
examination was defined by
a single modality in the
same visit, even though it
covered more than one
anatomical area. So, CT of
head plus chest was one
examination, this will may
be led to wrong RVUs
results.
Measuring Consultant
Radiologist workload:
method and results from a
national survey
This study reported a survey of
Consultant Radiologist
workload in Ireland in 2009 for
measuring Radiologist
workload, Relative Value
Units (RVUs) were assigned.
This survey did not suggest
any solutions to overcome
the overload of radiologists.
Radiologist Productivity
Measurement
This study aimed to measure
the required RVU which
should be done by radiologist
per a staffed hour. The study
took into account the
radiologists vacations, non-
clinical work and on-call
duties.
The study did not consider
the time taken to write
reports.
3.5 Summary
This chapter presents a number of related works in radiology departments and Data
Mining in healthcare. Table (3.1) shows the most related works to this work. These
works concentrated on calculating RVUs and measuring productivity of radiologists
but suffer from some gaps which differentiate between them and the work of this
research. The related works calculated the RVUs without considering reported times
and did not provide solutions for the overload of radiologists. These gaps are
considered in this research that overcome the overload of radiologists.
Chapter 4
The Data Mining Model
21
Chapter 4
The Data Mining Model
In this chapter, a Data Mining model of this work is presented. It aims to improve the
productivity of radiologists by assigning the appropriate case to the appropriate
radiologist. Figure 4.1 shows the model; data from different hospitals were collected
and contained radiology cases which were assigned using the Data Mining model to
different radiologists in different hospitals based on Tele-radiology system. The model
was applied four times with different classification method in each one. The four
classification methods which were applied are: Decision Tree, Naïve Bayes, K-NN,
Random Forest.
4.1 General View of Model
Hospital B
Hospital A Hospital C
Radiologists in Hospitals
Assigning Cases
Assigning Cases
Assigning Cases
Figure (4.1): The Model
22
4.2 Model Details
Data from different hospitals were collected, it contained radiology cases to be
assigned to the radiologists using the Data Mining model. The model consists of some
iterations, the need of iterations is to assign the cases, the initial iteration assign the
cases to the appropriate radiologists, the need to more iterations is to reassign the cases
which caused the radiologists productivity to exceed 100 to the radiologists who did
not reach 100. The model was applied four times with using four classification
techniques. In this research, each classification technique has four iterations to assign
all cases to the radiologists.
4.3 Model Iterations
The model consists of some iterations to assign the cases. The initial iteration assign
the cases to the appropriate radiologists. The cases which caused the radiologists
productivity to exceed 100 need to more iterations to reassign them to the radiologists
who did not reach 100. The iterations will continue until all cases are distributed.
4.3.1 The Initial Iteration
The flow chart of the initial iteration is shown in Figure 4.2. Training and Testing Data
are inputs for apply model to apply the classification method to assign the appropriate
cases to the appropriate radiologists. The previous productivity for other modalities
was calculated from the Rest Data then it was joined to the assigned data to calculate
the total productivity of the radiologists after auto assigning. The total productivity
joined a loop and was checked one by one; if it is <=100 the cases which were assigned
will be added to a final result, and if the total productivity exceeds 100 the cases which
cause such exceeding will be added to reassign data as input in the next iteration.
23
Figure (4.2): The Initial Iteration of the Model
• Training Data which contain radiologists cases for one year including three
modalities CT, MRI and Mammography.
• Testing Data which contain radiologists cases for one month including three
modalities CT, MRI and Mammography.
• Apply Model which applied the classification technique e.g. Decision Tree to
auto assign cases to all radiologists.
• Assigned Cases which are the result of apply model (the assigned cases).
• Rest Data which contain radiologists cases for the same month of Testing Data
including other modality types such as (US, X-Ray, Fluro. …etc.).
• Calculate Productivity which calculates the radiologists productivity for the
modalities in Rest Data in order to achieve the target of total productivity that
the radiologist must reach.
• Previous Productivity which is the result of calculating productivity
(radiologists productivity in Rest Data).
24
Table (4.1): Previous Productivity
Radiologist Previous Productivity
Radiologist 2 98.3
Radiologist 4 13.9
• Joined Data which attach the value of previous productivity for each
radiologist to the assigned cases which are the result of apply model.
Table (4.2): Joined Data
Case ID Radiologist wRVU Previous Productivity
Case 1 Radiologist 2 0.6 98.3
Case 2 Radiologist 4 0.4 13.9
Case 3 Radiologist 2 0.6 98.3
Case 4 Radiologist 2 0.3 98.3
Case 5 Radiologist 4 0.2 13.9
The Loop:
The need for the loop is to identify which cases will be reassigned and which cases
will go to the final result by checking the total productivity. Table 4.4 shows the work
of the loop.
• Current Productivity is the value of RVU for the current assigned case +
RVUs of the previous assigned cases for the same radiologist.
Table (4.3): Calculating Current Productivity
Case ID Radiologist wRVU Sum wRVU Prev. Productivity
Case 1 Radiologist 2 0.6 0.6 98.3
Case 2 Radiologist 4 0.4 0.4 13.9
Case 3 Radiologist 2 0.6 1.2 98.3
Case 4 Radiologist 2 0.6 1.8 98.3
Case 5 Radiologist 4 0.2 0.6 13.9
• Sum Previous Productivity and Current Productivity which calculates the
current productivity then sums its value with the previous productivity of the
radiologist to produce the total productivity.
25
• Checking Total Productivity: The total productivity was checked; if it is
<=100 the cases will be added to a final result, otherwise it will be added to
reassign data as input in the next iteration.
Table (4.4): The Work of the Loop
Case ID Radiologist wRVU Sum
wRVU
Previous
Prod.
Total
Prod. Condition
Case 1 Radiologist 2 0.6 0.6 98.3 98.9 Final
Case 2 Radiologist 4 0.4 0.4 13.9 14.3 Final
Case 3 Radiologist 2 0.6 1.2 98.3 99.5 Final
Case 4 Radiologist 2 0.6 1.8 98.3 100.1 Re-Assign
Case 5 Radiologist 4 0.2 0.6 13.9 14.5 Final
4.3.2 The Next Iteration (one or more)
Figure 4.3 shows the flow chart of the next iteration of the model; the testing data are
replaced with reassigned data (the result of the previous iteration) and the Training
Data are filtered from the radiologists whose productivity became around 100. The
iterations will continue until all cases are distributed.
The inputs of the next iteration:
• Training Data which were filtered from the radiologists whose productivity
became around 100
• Reassigned Cases which were the result of the initial iteration (cases which
were assigned to radiologists and made their productivity >100).
• Rest Data in addition to initial iteration result.
26
Figure (4.3): The Next Iteration of the Model
4.4 Summary
This chapter presents the Data Mining model of this work and describes its
components. The model consists of iterations which will continue until all cases are
distributed. In this research, each classifier of the four applied classifiers had four
iterations to assign all cases to the radiologists. The model aims to improve the
productivity of radiologists in hospitals utilizing Data Mining classification
techniques.
Chapter 5
Methodology
28
Chapter 5
Methodology
In this chapter, the methodology for enhancing radiologists productivity in hospitals is
presented. The chapter is divided into six sections, section one introduces the
methodology steps, section two contains the process of collection and acquisition of
data which were collected from different hospitals, section three contains data pre-
processing, section four provides feature sets selections and extraction, section five
provides the classifications algorithms which were used and section six provides the
evaluation of the model.
5.1 Methodology Steps
Figure (5.1): Methodology Steps
5.2 Data Acquisition and Collection
Data were collected for eight radiologists from different four hospitals in Saudi Arabia.
Two data sets were collected: One for training data for a year (From 1 May 2016 to 30
April 2017) and another for testing data for a month (May 2017). The training and
testing data sets contain data of the three modalities which the research concentrated
on, i.e. CT, MRI and mammography. Figure 5.2 shows a data set for training from a
hospital before pre-processing, EXM_NUMBER is a unique number for each
radiology procedure, Visit_Class is the class of the patient where there are three classes
for a patient (OutPatient, InPatient, Emergency). Radiologist is the radiologist name
and PAT_DOB is the patient Date of Birth. EXM_DONE_STAMP is the exam date
and time when sent to the radiologist to write report, EXM_APPROVED_STAMP is
Data Acquisition
Data Preprocessing
and Building up Dataset
Feature Sets Selection
Training Process Implementation Evaluation
29
the date and time of writing the report by radiologist, and EXM_CODE is exam
procedure identification.
Figure (5.2): Data Before Pre-processing
Table (5.1): Description of Figure 5.2 Columns
Column Name Description
EXM_NUMBER number of exam procedure
Visit_Class class of the patient (OutPatient, InPatient, Emergency).
Radiologist Radiologist Name
PAT_DOB patient Date of Birth
EXM_DONE_STAMP exam date and time when sent to the radiologist to write
report.
EXM_APPROVED_STAMP date and time of writing the report
EXM_CODE exam identification procedure.
Figure 5.3 shows the exam code dictionary; it is a schedule to describe radiology
procedure with details of body parts and modality type. Code is the exam identification
procedure, Description is a full description for radiology procedure, Modality is the
radiology machine type (CT, MRI …… etc.), Body part is the part of body is exposed
to radiation, Subspecialty is the exact specialization of the radiology procedure,
Exam_Group is a radiology procedure grouped by modality type and (body_part or
Subspecialty, wRVU is it is Relative time, effort, and skill needed by a radiologist in
providing a report (Kuehn, 2009).
30
Figure (5.3): Exam Code Dictionary
5.3 Data Pre-processing and Feature Sets Selections
In this phase, data have been pre-processed, a data set of training and testing was built
and six feature sets were created: Radiologist as a label, Visit_Class, Age_Group,
Body_Part, Reporting_time, and Exam_Day. training data were selected for a year was
selected (13,142 records) and testing data were selected for a month (1,290 records).
Data named as rest data to be taken into account.
5.3.1 Generating New Columns
Using Excel 2016 Age column was generated using EXM_DONE_STAMP and
PAT_DOB columns in Figure 5.2, then according to Food and Drugs Administration
(FDA) age classifications, age was grouped into four groups (Infant, Child,
Adolescent, Adult) (FDA, 2014). Table 5.2 shows FDA age classifications.
Figure (5.4): Generate Age Group
31
Table (5.2): FDA Age Classifications (FDA, 2014)
Pediatric Subgroup Approximate Age Range
Infant greater than 1 month to 2 years of age
Child greater than 2 to 12 years of age
Adolescent greater than 12 through 21 years of age
Adult Greater than 22
Time to Report was generated by subtracting EXM_DONE_STAMP from
EXM_APPROVED_STAMP. Figure 5.5 shows generate Time to Report.
Figure (5.5): Generate Time to Report
Reporting Time was generated from Time to Report column, if Time to report <=24
then the Reporting Time is 24, in case it is <=48 then frame is 48 and if it >48 the
frame is 100. Figure 5.6 shows generate Reporting Time.
Figure (5.6): Generate Reporting Time
Exam_Day was generated from EXM_DONE_STAMP column and the Date is
converted into Day. Figure 5.7 shows generate Exam Day.
32
Figure (5.7): Generate Exam Day
5.3.2 Combining Data
Using Excel 2016 data from figure 5.2 and 5.3 were combined by comparing
EXM_CODE in figure 5.2 with Code in figure 5.3 and taking the value of
EXAM_GROUP.
Figure (5.8): Combined Data
5.3.3 Feature Set Selection
After data combining, seven feature sets were selected from figure 4.9 to use as
training data in rapid miner programme. Figure 5.9 shows the final training data set
with seven feature sets.
Figure (5.9): Training Data Set
33
Radiologist is the name of the radiologist who writes the report, Visit_class is the
patient class (OutPatient, InPatient, Emergency), Age_Group is the patient age,
Reporting Time is the time which the radiologist takes to write the report, Exam_Day
is the day on which the radiology procedure was done and wRVU is the time, effort,
and skill needed by a radiologist to provide a report.
5.4 Testing Data
Testing Data were collected for the same eight radiologists for one month (May 2017).
Figure 5.10 show testing data before pre-processing.
Figure (5.10): Testing Data Before Pre-processing
After generating new columns and combining figure 5.10 with 5.3, figure 5.11 was
obtained.
Figure (5.11): Combined Data
Data in Figure 5.11 contains all types of modalities which the research concentrated
on, i.e. CT, MRI and Mammography in addition to other types such as US, X-RAY,
Fluoroscopy…. etc.) which were taken into account. Therefore, data were divided into
two data sets: Figure 5.12 shows the first dataset which contains the research
modalities (Testing Data) and Figure 5.13 shows the second dataset which contains
other modalities and which is named (Rest Data).
34
Figure (5.12): Testing Data
Figure (5.13): Rest Data
ID column is a unique number to distinguish each row and Reporting Time column is
generated based on Visit Class. The radiological studies are reported by the radiologist
within defined time limits. Urgent cases (Inpatient, Emergency) are reported within 24
hours and routine cases (Outpatient) are reported within 48 hours ((CBAHI), 2016).
5.5 Implementation
After preparing the data sets with seven different feature sets, four types of
classification methods were applied using Rapid Miner programme. The model
consists of some iterations which will be explained in the next chapter.
5.5.1 Tools
The tools which were used to apply implementation are:
• Rapid Miner: it was used to apply classifiers of this model, also it was used to
test the model with different performance measures.
• Microsoft Excel 2016: it is a spreadsheet developed by Microsoft for Windows,
Mac OS X, and iOS. Its features are calculation, graphing tools a pivot tables.
35
5.6 Evaluation
After applying the model, the results of the model were presented to an expert from
one of the four hospitals and his feedback about the model was obtained. Also,
accuracy and F-measure evaluate the performance of the classification methods to
compare among them.
5.7 Summary
In this chapter, a methodology of this work was presented. Data were collected from
different four hospitals contained eight radiologists, data were pre-processed by
generating new columns to extract useful data sets. Data after pre-processing were
combined, seven feature sets were selected for training and testing data sets to apply
the model and classification methods.
Chapter 6
Results, Discussion and
Evaluation
37
Chapter 6
Results, Discussion and Evaluation
In this chapter, the classification methods settings and the results of the model are
presented, also the evaluation which were conducted on the model is provided,
accuracy and F-measure performance evaluation were used to evaluate the model.
6.1 Classification Methods Settings
Table (6.1): Classifiers Settings
Classifier Property Value
K-NN The k value of the
classifier
1
Naïve Bayes Laplace Correction unchecked
Decision Tree
Criterion gain_ratio
Minimize size of split 4
Minimal leaf size 2
Minimal gain 0.1
Maximal depth 20
Confidence 0.25
Pruning and pre-pruning checked
Random Forest
Number of trees 10
Criterion gain_ratio
Maximal depth 20
Confidence 0.25
Minimal gain 0.1
Minimal leaf size 2
Minimal size for split 3
6.2 Experimental Results
The model consists of iterations which are applied to four classification methods.
Figure 6.3 shows the initial iteration in Rapid Miner.
38
Figure (6.1): The Initial Iteration in Rapid Miner
After applying the initial iteration, Figure 6.4 shows the cases which are auto assigned
to different radiologists.
Figure (6.2): Auto Assigned Cases
39
Figure (6.3): Next Iteration in Rapid Miner
Table 6.2 shows the number of cases distributed in each iteration in different
classifiers. Decision Tree, Random Forest and K-NN assigned all cases but in Naïve
Bayes 61 cases are still not assigned.
Table (6.2): Auto Assigned Cases
Iteration 1 Iteration 2 Iteration 3 Iteration 4 Total
Decision Tree 789 217 161 123 1290
Random Forest 408 296 401 185 1290
K-NN 725 288 104 173 1290
Naive Bayes 773 335 93 28 1229
The 61 cases consist of six exam groups which are CT abd-pelvis, CT brain, CT Chest,
CTA, MR brain and MR MSK. The radiologists which have the lowest productivity
are Radiologist 4 and Radiologist 6. From the training data Radiologist 4 and
Radiologist 6 did not write reports for these exam groups. Therefore, the Naïve Bayes
classifier did no assign any of these cases to them whereas in other classifiers some of
these cases were assigned to those radiologists
Figure 6.4 shows the differences between productivity of radiologists before and after
Data Mining with the four different classification methods, the differences between
productivity in different classifiers are may due to the wrongly assigned cases which
will be explained in table 6.4
40
Figure (6.4(: Productivity Comparison
6.3 Evaluation
Model Evaluation is an essential part of model development process. It helps to select
the best model that represents the data and how well the selected model will work in
the future (sayad, 2014).
6.3.1 Performance Evaluation Results
The model was evaluated by presenting its results to an expert in one of the four
hospitals for his opinion. He declared that the results of the model are very good as
they take into account the subspecialty of each procedure in assigning the cases. He
also believes that applying the model in hospitals will achieve good results and
improve the radiologists productivity. Also, an evaluation to classification techniques
was done by using different evaluation measures to evaluate the performance and to
compare among them.
The performance evaluation of the classifiers was based on matching the assigned
cases to radiologists with their last year cases (training data). Four classification
methods were applied to a data set. The number of samples was 1,290 in all classifiers
except Naïve Bayes, the Naïve Bayes classifier was higher in both accuracy & F-
measure, as shown in table 6.3 and Figure 6.7
Table (6.3): Performance Evaluation Results
Classifier Accuracy F-measure
Decision Tree 93.88 96.84
K-NN 88.68 94.00
Random Forest 86.74 92.90
Naïve Bayes 100 100
Naïve Bayes* 95.27 97.58
41
Naïve Bayes*: Performance measurements with the unassigned 61 cases.
Figure (6.5): Performance Evaluation Results
Figure (6.6): Naive Bayes Classifier Results
The performance evaluation was measured based on the assigned cases. Naïve Bayes
has the highest value in both accuracy and F-measure because it did not have any
wrongly assigned cases but other classifiers have wrongly assigned cases which means
that the classifiers assigned new cases to the radiologists although they did not deal
with these cases in the last year (training data). Table 6.4 shows the number and the
percentage of the wrong cases in each classifier.
Table (6.4): Wrongly Assigned Cases
Classifier Wrong assigned cases Percentage
Decision Tree 79 6.12%
K-NN 146 11.31%
Random Forest 171 13.25%
Naïve Bayes 0 0%
80
82
84
86
88
90
92
94
96
98
100
Decision Tree K-NN Random Forest Naïve Bayes
Accuracy F-measure
9293949596979899
100101
Accuracy F-measure
Naïve Bayes (with unassinged cases)
Naïve Bayes (without unassigned cases)
42
6.4 Summary
In this chapter, the experimental and evaluation results were reviewed. Four classifiers
were applied for the data set. To evaluate the model the results were presented to an
expert from one of the four hospitals, he declared that the results of the model are good
as they take into account the subspecialty of each procedure in assigning the cases and
he believes that applying the model in hospitals will achieve good results. Accuracy
and F-measure performance evaluation were used to compare among these classifiers.
The results show that the Naïve Bayes was the best classifier by up to 8% in accuracy
and 4% in F-measure due to its way of assigning cases. It assigned the appropriate case
to the appropriate radiologist. Naïve Bayes had the highest value in accuracy and F-
measure.
Chapter 7
Conclusions and Future
Work
44
Chapter 7
Conclusions and Future Work
7.1 Conclusion
Today, radiology departments have moved away from a patient-focused to concentrate
on other areas as well. Several processes of procedures need interpretation by the
radiologists.
The idea of this research investigates some problems in radiology departments at
hospitals based on the delay of writing reports by radiologists which is due to the heavy
load of work assigned to them. A Data Mining model was conducted to overcome this
problem and improve the radiologists productivity by assigning the appropriate cases
to the appropriate radiologists within Tele-radiology procedure.
Data were collected from four hospitals in different areas in Saudi Arabia covering
eight radiologists (two from each hospital) with varying productivity and
specialization with emphasis on CT, MRI and Mammography modalities. The data
were pre-processed and combined to produce data sets to apply the model, one data
set for a year for training data and another for a month for testing data. The training
and testing datasets contained data of the three modalities which the research
concentrated on, i.e. CT, MRI and Mammography.
Data Mining model was conducted to improve the productivity of radiologists by
assigning the appropriate case to the appropriate radiologist. Data from different
hospitals were collected and contained radiology cases which were assigned using the
model to different radiologists in different hospitals based on Tele-radiology system.
Four different classifiers were applied, each classifier has four iterations to predict and
assign the suitable cases to each radiologist to improve radiologists productivity.
The results showed the differences between radiologists productivity before and after
Data Mining model with the four different classification methods, these differences
are may due to the wrongly assigned cases. The productivity of each radiologist has
improved as the aim of this research is to ensure that the productivity of each
radiologist is within an acceptable range (around 100) and that he has a fair load.
45
Figure (7.1): Productivity Comparison
Naïve Bayes did not have any wrongly assigned cases but other classifiers had wrongly
assigned cases which means that the classifiers assigned new cases to the radiologists
although they did not deal with these cases in the last year (training data). The accuracy
and F-measure performance evaluation were used to compare among the four
classifiers. The results showed that the Naïve Bayes was the best classifier due to its
way of assigning cases. It assigned the appropriate case to the appropriate radiologist
and had the highest value in accuracy and F-measure.
7.2 Future Work
This model which is conducted for the first time achieved its objectives and improved
radiologists productivity.
The following operations can be carried out to improve the performance of the model:
Unassigned cases in Naïve Bayes results are due the limited number of radiologists.
This led to help the decision makers to determine the accurate need of radiologists and
their specialties when applying the model in hospitals.
Improving the model requires bigger training dataset, which can be obtained from the
historical data in hospitals and from daily workload.
Improving RVU calculations to take into account non- clinical and teaching work of
radiologists.
The model can be applied on one or several hospitals in case of Tele-radiology by
making it as a part of RIS "Radiology Information Systems". The model can also be
developed as a layer on top of these systems since they are provided by different
vendors.
References
47
References
Bellaachia, A., & Guven , E. (2006). Predicting Breast Cancer Survivability Using
Data Mining Techniques. Washington. Retrieved October 15, 2016, from
http://www.siam.org/meetings/sdm06/workproceed/Scientific%20Datasets/be
llaachia.pdf?q=data-mining-techniques
Tapedia, K., & Wagh , A. M. (2016). Data Mining for Various Internets of Things
Applications. International Journal of Research in Advent Technology.
Retrieved April 6, 2017, from
http://www.ijrat.org/downloads/ncpci2016/ncpci-27.pdf
(CBAHI), S. C. (2016). CBAHI Standerds. Saudi Arabia. Retrieved August 15, 2017
Ahmed, M. E., & Aldosh, M. S. (2014). Telemedicine and Teleradiology in Saudi
Arabia. IOSR Journal of Dental and Medical Sciences (IOSR-JDMS), 13(2).
Retrieved February 13, 2017, from http://www.iosrjournals.org/iosr-
jdms/papers/Vol13-issue2/Version-2/V013228890.pdf
Al Mehedi , H. M., Nasser, M., Pal, B., & Shamim, A. (2014). Support Vector
Machine and Random Forest Modeling for Intrusion Detection System (IDS).
Journal of Intelligent Learning Systems and Applications, 6, 45-52. Retrieved
June 6, 2017, from http://file.scirp.org/pdf/JILSA_2014021411471330.pdf
Alizadehsani, R., Habibi, J., Zahra Sani, Z. A., Mashayekhi, H., Boghrati, R.,
Ghandeharioun, A., . . . Alizadeh-Sani, F. (2013). Diagnosing Coronary
Artery Disease via Data Mining Algorithms by Considering Laboratory and
Echocardiography Features. Res Cardiovasc Med, 2(3), 133–139. Retrieved
October 10, 2016, from
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4253773/
Brady, A. P. (2011). Measuring Consultant Radiologist workload: method and results
from a national survey. Insights Imaging, 247–260. Retrieved August 20,
2017, from http://link.springer.com/article/10.1007/s13244-011-0094-3
Capko, J. (2003). 5 Steps to a Performance Evaluation System. Retrieved September
27, 2017, from aafp.org: http://www.aafp.org/fpm/2003/0300/p43.html
Choi, J. P., Han, T. H., & Park, R. W. (2009). A Hybrid Bayesian Network Model
for Predicting Breast Cancer Prognosis. J Korean Soc Med Inform, 15(1), 49-
57. Retrieved October 13, 2016, from
http://synapse.koreamed.org/DOIx.php?id=10.4258/jksmi.2009.15.1.49&vmo
de=FULL
Cowan, I. A., MacDonald, S. L., & Floyd, R. A. (2013). Measuring and managing
radiologist workload: Measuring radiologist reporting times using data from a
Radiology Information System. Journal of Medical Imaging and Radiation
Oncology, 57(5), 558–566. Retrieved August 14, 2017, from
http://onlinelibrary.wiley.com/doi/10.1111/1754-9485.12092/full
D., R., & Jr., C. (2009). Data Mining in Healthcare: Current Applications and Issues.
Australia. Retrieved March 5, 2017, from
http://www.cclfi.org/files/Data_Mining_Health.pdf
Desikan, P., Hsu, K.-W., & Srivastava, J. (2011). Data Mining for HealthCare
Management. Retrieved October 10, 2016, from
http://www.siam.org/meetings/sdm11/dmhm.pdf
48
dora , J., Faccin , C., & Fogliatto , F. (2016). The use of relative value units to
monitor radiologists’ reporting productivity and workload. RAdIOlOGY
wORklOAd MANAGeMeNt. Retrieved July 20, 2017
Dubey, R., & chandrakar , S. (2015). Review on Hybrid Data Mining Techniques for
The Diagnosis of Heart Diseases in Medical Ground. INDIAN JOURNAL OF
APPLIED RESEARCH, 5(8). Retrieved September 20, 2016, from
http://www.worldwidejournals.in/ojs/index.php/ijar/article/view/6366/6404
FDA. (2014). U.S FOOD & DRUGS ADMINSTRATION. Retrieved August 8, 2017,
from www.fda.gov:
https://www.fda.gov/RegulatoryInformation/Guidances/ucm089740.htm#ft3
Forrest, W. (2007). RVUs key to tracking radiologist and physician productivity.
Retrieved August 10, 2017, from .auntminnie.com:
http://www.auntminnie.com/index.aspx?sec=ser&sub=def&pag=dis&ItemID
=76228
Forsberg, D., Rosipko, B., & Sunshine, J. L. (2016). Variation of Reading Times for
Radiologists over Different Examination Types. The Reading Room:
Workflow and Efficiency. Retrieved August 8, 2017, from
https://c.ymcdn.com/sites/siim.org/resource/resmgr/siim2016abstracts/Readin
g_Rm_Forsberg.pdf
GS, P. (2015). Data Mining vs. Statistics - How Are They Different? Retrieved April
4, 2017, from simplilearn.com: https://www.simplilearn.com/data-mining-vs-
statistics-article
Hachesu, P. R., Ahmadi, M., Alizadeh, S., & Sadoughi, F. (2013). Use of Data
Mining Techniques to Determine and Predict Length of Stay of Cardiac
Patients. Healthc Inform Res, 19(2), 121-129. Retrieved September 20, 2016,
from
http://synapse.koreamed.org/search.php?where=aview&id=10.4258/hir.2013.
19.2.121&code=1088HIR&vmode=FULL
Kaur, S., & Singh, R. (2016). A Review of Data Mining Based Breast Cancer
Detection and Risk Assessment Techniques. International Journal of
Computer Science and Information Security, 14(5). Retrieved September 15,
2016
Koh, H. C., & Tan, G. (2005). Data Mining Applications in Healthcare. Journal of
Healthcare Information Management, 19(2). Retrieved March 6, 2017
Kraft, M. R., Desouza, K. C., & Androwich, I. (2003). Data mining in healthcare
information systems: case study of a veterans' administration spinal cord
injury population:. Proceeding of the 36th Hawaii International Conference
on System Sciences (HICSS'03). Retrieved September 19, 2016, from
https://www.computer.org/csdl/proceedings/hicss/2003/1874/06/187460159a.
Kuehn, L. (2009). Relative Value Unit (RVU) Data Analysis. American Health
Information Management Association. Retrieved August 15, 2017
Lai, Y.-H. (2015). Applying Data Mining Technology on the Using of Traditional
Chinese Medicine in Taiwan. International Journal of Computer and
Information Technology, 04(01). Retrieved November 19, 2016, from
http://www.ijcit.com/archives/volume4/issue1/Paper040108.pdf
Lundina , M., Lundina , J., Burked , H., Toikkanenb , S., Pylkkänenc , L., & Joensuu,
H. (1999). “Artificial Neural Networks Applied to Survival Prediction in
49
Breast Cancer. Oncology International Journal for Cancer Resaerch and
Treatment, 57. Retrieved November 19, 2016
M, S. E. (2012). Design and development of data mining models for the prediction of
manpower placement in the technical domain. India: Cochin University of
Science and Technology. Retrieved July 6, 2017
MERRITT HAWKINS an AMN Healthcare Company. (2011). RVU BASED
PHYSICIAN COMPENSATION AND PRODUCTIVITY. Retrieved July 4,
2017, from merritthawkins.com:
https://www.merritthawkins.com/pdf/mharvuword.pdf
Milovic, B., & Milovic, M. (2012). Prediction and Decision Making in Health Care
using Data Mining. International Journal of Public Health Science (IJPHS),
1(2), 69-78. Retrieved February 2, 2017, from
http://www.iaescore.com/journals/index.php/IJPHS/article/view/4593/3465
Mlambo , N. (2016). Data Mining: Techniques, Key Challenges and Approaches for
Improvement. International Journal of Advanced Research in Computer
Science and Software Engineering, 6(3). Retrieved May 7, 2017, from
https://www.ijarcsse.com/docs/papers/Volume_6/3_March2016/V6I3-
01305.pdf
Radiology | Al Yousuf Hospital, Al Khobar, KSA. (2013). Retrieved May 15, 2017,
from alyousif-hospital: http://alyousif-hospital.com/EN/radiology.html
Radiology, T. A. (2013). Radiologist Productivity Measurement. Retrieved August
20, 2017, from https://www.acr.org/Membership/Residents-and-
Fellows/Resident-
Resources/~/media/2DF7FF7E1D444175B4BAD8C79019C2CF.pdf
SAS Institute Inc. (2012). Data Mining what it is and why it matters. Retrieved
August 7, 2017, from SAS website:
https://www.sas.com/en_us/insights/analytics/data-mining.html
sayad, s. (2014). Model Evaluation. saed sayad. Retrieved September 27, 2017
Shouman, M., Turner, T., & Stocker, R. (2012). Applying k-Nearest Neighbour in
Diagnosing Heart Disease Patients. International Conference on Knowledge
Discovery (ICKD), XX. Retrieved November 19, 2016, from
https://www.scribd.com/document/215428649/Applying-K-Nearest-
Neighbour-in-Diagnosing-Heart-Disease-Patients-Libre
Shukla, S., Gupta, D. L., & Prasad, B. R. (2016). Comparative Study of Recent
Trends on Cancer Disease Prediction using Data Mining Techniques.
Internation Journal of Database Theory and Application, 9(9), 107-118.
Retrieved September 15, 2016
Srinivas, K., Rani , B. K., & Govrdhan , A. (2010). Applications of Data Mining
Techniques in Healthcare and Prediction of Heart Attacks. International
Journal on Computer Science and Engineering (IJCSE), 02(02), 250-255.
Retrieved January 15, 2017
Staff, I. a. (2014, March 24). U.S FOOD & DRUGS ADMINSTRATIOn. Retrieved
from www.fda.gov:
https://www.fda.gov/RegulatoryInformation/Guidances/ucm089740.htm#ft3
50
Appendices
51
Appendix A: Reported Cases Statistics
Hospital 1 CT MR US MG Grand Total
Delayed 506 353 180 107 1146
On-Time 4810 975 4693 170 10648
Grand Total 5316 1328 4873 277 11794
Hospital 2 CT MR US MG Grand Total
Delayed 565 591 104 8 1268
On-Time 11175 3777 5771 112 20835
Grand Total 11740 4368 5875 120 22103
Hospital 3 CT MR US MG Grand Total
Delayed 4314 1313 1466 65 7158
On-Time 9217 1692 12302 697 23908
Grand Total 13531 3005 13768 762 31066
Hospital 4 CT MR US MG Grand Total
Delayed 6042 1962 1972 200 10176
On-Time 11540 3311 10837 73 25761
Grand Total 17582 5273 12809 273 35937
52
Table A.1: Delayed Reported Cases Percentage
CT MR MG US
Hospital 1 9.52% 26.58% 38.63% 3.69%
Hospital 2 4.81% 13.53% 6.67% 1.77%
Hospital 3 40.01% 33.77% 37.43% 1.63%
Hospital 4 34.36% 37.21% 73.26% 15.40%
Figure A.1: Delayed Cases Chart
0.00%
10.00%
20.00%
30.00%
40.00%
50.00%
60.00%
70.00%
80.00%
Hospital 1 Hospital 2 Hospital 3 Hospital 4
CT MR MG US
53
Appendix B: Sample of Exam Code Dictionary
CODE DESCRIPTION MOD BODYPART SUBSPECIALTY Exam_Group wRVU
CT0001 CT Abdomen w/ + w/o Contrast CT ABDOMEN ABDOMINAL CT abd-pelvis 0.278
CT0002 CT Abdomen w/ Contrast CT ABDOMEN ABDOMINAL CT abd-pelvis 0.278
CT0003 CT Abdomen w/o Contrast CT ABDOMEN ABDOMINAL CT abd-pelvis 0.278
CT0004 CT Angiography Abd Aorta +
Iliofemoral CT AORTA CTANGIO CTA 0.442
CT0005 CT Angiography Abdomen CT ABDOMEN CTANGIO CTA 0.442
CT0006 CT Angiography Aorta CT AORTA CTANGIO CTA 0.442
CT0007 CT Angiography Chest w/ + w/o
Contrast CT CHEST CTANGIO CTA 0.442
CT0008 CT Angiography Head w/ + w/o
Contrast CT BRAIN CTANGIO CTA 0.442
CT0009 CT Angiography Lower
Extremity Bilat CT BLEXTREMITY CTANGIO CTA 0.442
CT0010 CT Angiography Lower
Extremity Left CT LLEXTREMITY CTANGIO CTA 0.442
CT0011 CT Angiography Lower
Extremity Right CT RLEXTREMITY CTANGIO CTA 0.442
CT0012 CT Angiography Neck w/ + w/o
Contrast CT NECK CTANGIO CTA 0.442
CT0013 CT Angiography Pelvis w/ + w/o
Contrast CT PELVIS CTANGIO CTA 0.442
CT0014 CT Angiography Upper
Extremity Bilat CT BUEXTREMITY CTANGIO CTA 0.442
CT0015 CT Angiography Upper
Extremity Left CT LUEXTREMITY CTANGIO CTA 0.442
CT0016 CT Angiography Upper
Extremity Right CT RUEXTREMITY CTANGIO CTA 0.442
CT0017 CT Ankle w/ + w/o Contrast
Bilateral CT BANKLE MSK CT MSK 0.189
CT0018 CT Ankle w/ + w/o Contrast Left CT LANKLE MSK CT MSK 0.189
CT0019 CT Ankle w/ + w/o Contrast
Right CT RANKLE MSK CT MSK 0.189
CT0020 CT Ankle w/ Contrast Bilateral CT BANKLE MSK CT MSK 0.189
CT0021 CT Ankle w/ Contrast Left CT LANKLE MSK CT MSK 0.189
CT0022 CT Ankle w/ Contrast Right CT RANKLE MSK CT MSK 0.189
CT0023 CT Ankle w/o Contrast Bilateral CT BANKLE MSK CT MSK 0.189
CT0024 CT Ankle w/o Contrast Left CT LANKLE MSK CT MSK 0.189
CT0025 CT Ankle w/o Contrast Right CT RANKLE MSK CT MSK 0.189
CT0026 CT Aspiration CT ASPIRATION ASPIRATION CT chest 0.227
CT0027 CT Aspiration Renal Left CT ASPIRATION ASPIRATION CT chest 0.227
CT0028 CT Aspiration Renal Right CT ASPIRATION ASPIRATION CT chest 0.227
CT0029 CT Biopsy CT BIOPSY BIOPSY CT abd-pelvis 0.278
CT0030 CT Biopsy Abdomen CT BIOPSY BIOPSY CT abd-pelvis 0.278
CT0031 CT Biopsy Bone CT BIOPSY BIOPSY CT abd-pelvis 0.278
CT0032 CT Biopsy Liver CT BIOPSY BIOPSY CT abd-pelvis 0.278
CT0033 CT Biopsy Lung Left CT BIOPSY BIOPSY CT abd-pelvis 0.278
CT0034 CT Biopsy Lung Right CT BIOPSY BIOPSY CT abd-pelvis 0.278
CT0035 CT Biopsy Pancreas CT BIOPSY BIOPSY CT abd-pelvis 0.278
CT0036 CT Biopsy Pleura Left CT BIOPSY BIOPSY CT abd-pelvis 0.278
CT0037 CT Biopsy Pleura Right CT BIOPSY BIOPSY CT abd-pelvis 0.278
54
CT0038 CT Biopsy Renal Left CT BIOPSY BIOPSY CT abd-pelvis 0.278
CT0039 CT Biopsy Renal Right CT BIOPSY BIOPSY CT abd-pelvis 0.278
CT0040 CT Brain Perfusion Study CT BRAIN NEURO CT brain 0.189
CT0041 CT Bronchoscopy CT CHEST CHEST CT Chest 0.227
CT0042 CT Consultation Outside Film CT OUTFILMS CT Chest 0.227
CT0043 CT Dental CT DENTAL DENTAL CT head & neck 0.253
CT0044 CT Drainage - Abscess or Cyst CT DRAINAGE DRAINAGE CT abd-pelvis 0.278
CT0045 CT Drainage Liver CT DRAINAGE DRAINAGE CT abd-pelvis 0.278
CT0046 CT Drainage Lung Bilateral CT DRAINAGE DRAINAGE CT abd-pelvis 0.278
CT0047 CT Drainage Lung Left CT DRAINAGE DRAINAGE CT abd-pelvis 0.278
CT0048 CT Drainage Lung Right CT DRAINAGE DRAINAGE CT abd-pelvis 0.278
CT0049 CT Drainage Pancreas CT DRAINAGE DRAINAGE CT abd-pelvis 0.278
CT0050 CT Drainage Peritoneal CT DRAINAGE DRAINAGE CT abd-pelvis 0.278
CT0051 CT Drainage Renal Bilateral CT DRAINAGE DRAINAGE CT abd-pelvis 0.278
CT0052 CT Drainage Renal Left CT DRAINAGE DRAINAGE CT abd-pelvis 0.278
CT0053 CT Drainage Renal Right CT DRAINAGE DRAINAGE CT abd-pelvis 0.278
CT0054 CT Drainage Retroperitoneal
Abscess CT DRAINAGE DRAINAGE CT abd-pelvis 0.278
CT0055 CT Drainage
Subdiaphragm/Subphrenic CT DRAINAGE DRAINAGE CT abd-pelvis 0.278
CT0056 CT Elbow w/ + w/o Contrast
Bilateral CT BELBOW MSK CT MSK 0.189
CT0057 CT Elbow w/ + w/o Contrast
Left CT LELBOW MSK CT MSK 0.189
CT0058 CT Elbow w/ + w/o Contrast
Right CT RELBOW MSK CT MSK 0.189
CT0059 CT Elbow w/ Contrast Bilateral CT BELBOW MSK CT MSK 0.189
CT0060 CT Elbow w/ Contrast Left CT LELBOW MSK CT MSK 0.189
CT0061 CT Elbow w/ Contrast Right CT RELBOW MSK CT MSK 0.189
CT0062 CT Elbow w/o Contrast Bilateral CT BELBOW MSK CT MSK 0.189
CT0063 CT Elbow w/o Contrast Left CT LELBOW MSK CT MSK 0.189
CT0064 CT Elbow w/o Contrast Right CT RELBOW MSK CT MSK 0.189
CT0065 CT Femur w/ + w/o Contrast
Bilateral CT BFEMUR MSK CT MSK 0.189
CT0066 CT Femur w/ + w/o Contrast
Left CT LFEMUR MSK CT MSK 0.189
CT0067 CT Femur w/ + w/o Contrast
Right CT RFEMUR MSK CT MSK 0.189
CT0068 CT Femur w/ Contrast Bilateral CT BFEMUR MSK CT MSK 0.189
CT0069 CT Femur w/ Contrast Left CT LFEMUR MSK CT MSK 0.189
CT0070 CT Femur w/ Contrast Right CT RFEMUR MSK CT MSK 0.189
CT0071 CT Femur w/o Contrast
Bilateral CT BFEMUR MSK CT MSK 0.189
CT0072 CT Femur w/o Contrast Left CT LFEMUR MSK CT MSK 0.189
CT0073 CT Femur w/o Contrast Right CT RFEMUR MSK CT MSK 0.189
CT0074 CT Fistula or Sinus Tract
Abscess Study CT FISTULA CT MSK 0.189
CT0075 CT Foot w/ + w/o Contrast
Bilateral CT BFOOT MSK CT MSK 0.189
CT0076 CT Foot w/ + w/o Contrast Left CT LFOOT MSK CT MSK 0.189
CT0077 CT Foot w/ + w/o Contrast
Right CT RFOOT MSK CT MSK 0.189
CT0078 CT Foot w/ Contrast Bilateral CT BFOOT MSK CT MSK 0.189
55
CT0079 CT Foot w/ Contrast Left CT LFOOT MSK CT MSK 0.189
CT0080 CT Foot w/ Contrast Right CT RFOOT MSK CT MSK 0.189
CT0081 CT Foot w/o Contrast Bilateral CT BFOOT MSK CT MSK 0.189
CT0082 CT Foot w/o Contrast Left CT LFOOT MSK CT MSK 0.189
CT0083 CT Foot w/o Contrast Right CT RFOOT MSK CT MSK 0.189
CT0084 CT Forearm w/ + w/o Contrast
Bilateral CT BFOREARM MSK CT MSK 0.189
CT0085 CT Forearm w/ + w/o Contrast
Left CT LFOREARM MSK CT MSK 0.189
CT0086 CT Forearm w/ + w/o Contrast
Right CT RFOREARM MSK CT MSK 0.189
CT0087 CT Forearm w/ Contrast
Bilateral CT BFOREARM MSK CT MSK 0.189
CT0088 CT Forearm w/ Contrast Left CT LFOREARM MSK CT MSK 0.189
CT0089 CT Forearm w/ Contrast Right CT RFOREARM MSK CT MSK 0.189
CT0090 CT Forearm w/o Contrast
Bilateral CT BFOREARM MSK CT MSK 0.189
CT0091 CT Forearm w/o Contrast Left CT LFOREARM MSK CT MSK 0.189
CT0092 CT Forearm w/o Contrast Right CT RFOREARM MSK CT MSK 0.189
CT0093 CT Guidance Tissue Ablation CT GUIDANCE INTERVENTIONAL CT MSK 0.189
CT0094 CT Guide for Stereotactic Loc CT GUIDANCE INTERVENTIONAL CT MSK 0.189
CT0095 CT Hand w/ + w/o Contrast
Bilateral CT BHAND MSK CT MSK 0.189
CT0096 CT Hand w/ + w/o Contrast Left CT LHAND MSK CT MSK 0.189
CT0097 CT Hand w/ + w/o Contrast
Right CT RHAND MSK CT MSK 0.189
CT0098 CT Hand w/ Contrast Bilateral CT BHAND MSK CT MSK 0.189
CT0099 CT Hand w/ Contrast Left CT LHAND MSK CT MSK 0.189
CT0100 CT Hand w/ Contrast Right CT RHAND MSK CT MSK 0.189
MA0001 MA Additional Projections Left MG LMAMMARY BREAST Mammography 0.189
MA0002 MA Additional Projections
Right MG RMAMMARY BREAST Mammography 0.189
MA0003 MA Breast Ndl Loc Placement
Bilat MG BMAMMARY BREAST Mammography 0.189
MA0004 MA Breast Ndl Loc Placement
Left MG LMAMMARY BREAST Mammography 0.189
MA0005 MA Breast Ndl Loc Placement
Right MG RMAMMARY BREAST Mammography 0.189
MA0006 MA Consultation Outside Film MG OUTFILMS Mammography 0.189
MA0007 MA Core Biopsy Breast L MG LMAMMARY BREAST Mammography 0.189
MA0008 MA Core Biopsy Breast R MG RMAMMARY BREAST Mammography 0.189
MA0009 MA Ductogram or Galactogram
Multi Bilat MG BMAMMARY BREAST Mammography 0.189
MA0010 MA Ductogram or Galactogram
Multi Left MG LMAMMARY BREAST Mammography 0.189
MA0011 MA Ductogram or Galactogram
Multi Right MG RMAMMARY BREAST Mammography 0.189
MA0012 MA Ductogram or Galactogram
Single Bilat MG BMAMMARY BREAST Mammography 0.189
MA0013 MA Ductogram or Galactogram
Single Left MG LMAMMARY BREAST Mammography 0.189
MA0014 MA Ductogram or Galactogram
Single Right MG RMAMMARY BREAST Mammography 0.189
56
MA0015 MA Mammogram Diagnostic
Bilateral MG BMAMMARY BREAST Mammography 0.189
MA0016 MA Mammogram Left MG LMAMMARY BREAST Mammography 0.189
MA0017 MA Mammogram Right MG RMAMMARY BREAST Mammography 0.189
MA0018 MA Mammogram Routine
Screening Bilat MG BMAMMARY BREAST Mammography 0.189
MA0019 MA Radiological Specimen MG MASPECIMEN BREAST Mammography 0.189
MA0020 MA Stereotactic Localization
Bilateral MG BMAMMARY BREAST Mammography 0.189
MA0021 MA Stereotactic Localization
Left MG LMAMMARY BREAST Mammography 0.189
MA0022 MA Stereotactic Localization
Right MG RMAMMARY BREAST Mammography 0.189
MA0023 MA Previous Films for
Comparison MG Mammography 0.189
MA0023 MA Stereotactic VAB Right MG RMAMMARY BREAST Mammography 0.189
MA0024 MA Stereotactic VAB Left MG LMAMMARY BREAST Mammography 0.189
MA0025 MA Stereotactic VAB Bilateral MG BMAMMARY BREAST Mammography 0.189
MA0026 MA Stereotactic FNA Right MG RMAMMARY BREAST Mammography 0.189
MA0027 MA Stereotactic FNA Left MG LMAMMARY BREAST Mammography 0.189
MA0028 MA Stereotactic FNA Bilateral MG BMAMMARY BREAST Mammography 0.189
MR0001 MRA Abdomen MR AORTA MR-
ANGIOGRAPHY MRA 0.568
MR0002 MRA AO with LE Run off MR AORTA MR-
ANGIOGRAPHY MRA 0.568
MR0003 MRA Aortic Arch w + w/o
Contrast MR AORTA MR-
ANGIOGRAPHY MRA 0.568
MR0004 MRA Aortic Arch w/ Contrast MR AORTA MR-
ANGIOGRAPHY MRA 0.568
MR0005 MRA Aortic Arch w/o Contrast MR AORTA MR-
ANGIOGRAPHY MRA 0.568
MR0006 MRA Chest MR AORTA MR-
ANGIOGRAPHY MRA 0.568
MR0007 MRA Head w/ + w/o Contrast MR BRAIN MR-
ANGIOGRAPHY MRA 0.568
MR0008 MRA Head w/ Contrast MR BRAIN MR-
ANGIOGRAPHY MRA 0.568
MR0009 MRA Head w/o Contrast MR BRAIN MR-
ANGIOGRAPHY MRA 0.568
MR0010 MRA Iliac Vessels w/ Contrast MR PELVIS MR-
ANGIOGRAPHY MRA 0.568
MR0011 MRA Lower Extremity Bilat MR BLEXTREMITY MR-
ANGIOGRAPHY MRA 0.568
MR0012 MRA Lower Extremity Left MR LLEXTREMITY MR-
ANGIOGRAPHY MRA 0.568
MR0013 MRA Lower Extremity Right MR RLEXTREMITY MR-
ANGIOGRAPHY MRA 0.568
MR0014 MRA Lower Extremity w/ + w/o
Bilat MR BLEXTREMITY MR-
ANGIOGRAPHY MRA 0.568
MR0015 MRA Lower Extremity w/ + w/o
Left MR LLEXTREMITY MR-
ANGIOGRAPHY MRA 0.568
MR0016 MRA Lower Extremity w/ + w/o
Right MR RLEXTREMITY MR-
ANGIOGRAPHY MRA 0.568
57
MR0017 MRA Lower Extremity w/ Bilat MR BLEXTREMITY MR-
ANGIOGRAPHY MRA 0.568
MR0018 MRA Lower Extremity w/ Left MR LLEXTREMITY MR-
ANGIOGRAPHY MRA 0.568
MR0019 MRA Lower Extremity w/ Right MR RLEXTREMITY MR-
ANGIOGRAPHY MRA 0.568
MR0020 MRA Lower Extremity w/o Bilat MR BLEXTREMITY MR-
ANGIOGRAPHY MRA 0.568
MR0021 MRA Lower Extremity w/o Left MR LLEXTREMITY MR-
ANGIOGRAPHY MRA 0.568
MR0022 MRA Lower Extremity w/o
Right MR RLEXTREMITY MR-
ANGIOGRAPHY MRA 0.568
MR0023 MRA Neck w/ + w/o Contrast MR NECK MR-
ANGIOGRAPHY MRA 0.568
MR0024 MRA Neck w/ Contrast MR NECK MR-
ANGIOGRAPHY MRA 0.568
MR0025 MRA Neck w/o Contrast MR NECK MR-
ANGIOGRAPHY MRA 0.568
MR0026 MRA Pelvis MR PELVIS MR-
ANGIOGRAPHY MRA 0.568
MR0027 MRA Pelvis w/ + w/o Contrast MR PELVIS MR-
ANGIOGRAPHY MRA 0.568
MR0028 MRA Pelvis w/ Contrast MR PELVIS MR-
ANGIOGRAPHY MRA 0.568
MR0029 MRA Pelvis w/o Contrast MR PELVIS MR-
ANGIOGRAPHY MRA 0.568
MR0030 MRA Popliteal w/ Contrast MR KNEE MR-
ANGIOGRAPHY MRA 0.568
MR0031 MRA Popliteal w/+w/o Contrast MR KNEE MR-
ANGIOGRAPHY MRA 0.568
MR0032 MRA Popliteal w/o Contrast MR KNEE MR-
ANGIOGRAPHY MRA 0.568
MR0033 MRA Portal Vessels w/contrast MR LIVER MR-
ANGIOGRAPHY MRA 0.568
MR0034 MRA Pulmonary Vessels w/
Contrast MR CHEST MR-
ANGIOGRAPHY MRA 0.568
MR0035 MRA Renal Artery w/+w/o
Contrast MR RENAL MR-
ANGIOGRAPHY MRA 0.568
MR0036 MRA Renal Artery w/ Contrast MR RENAL MR-
ANGIOGRAPHY MRA 0.568
MR0037 MRA Renal Artery w/o Contrast MR RENAL MR-
ANGIOGRAPHY MRA 0.568
MR0038 MRA Spinal Canal + Contents MR SPINE MR-
ANGIOGRAPHY MRA 0.568
MR0039 MRA Subclavian Artery w/+w/o
Contrast MR CHEST MR-
ANGIOGRAPHY MRA 0.568
MR0040 MRA Subclavian Artery w/o
Contrast MR CHEST MR-
ANGIOGRAPHY MRA 0.568
MR0041 MRA Subclavian Artery w/
Contrast MR CHEST MR-
ANGIOGRAPHY MRA 0.568
MR0042 MRA Superior Mesenteric
Vessels w/ Contr MR ABDOMEN MR-
ANGIOGRAPHY MRA 0.568
58
MR0043 MRA Superior Mesenteric
Vessels w/+w/o C MR ABDOMEN MR-
ANGIOGRAPHY MRA 0.568
MR0044 MRA Superior Mesenteric
Vessels w/o Cont MR ABDOMEN MR-
ANGIOGRAPHY MRA 0.568
MR0045 MRA Upper Extremity Bilat MR BUEXTREMITY MR-
ANGIOGRAPHY MRA 0.568
MR0046 MRA Upper Extremity Left MR LUEXTREMITY MR-
ANGIOGRAPHY MRA 0.568
MR0047 MRA Upper Extremity Right MR RUEXTREMITY MR-
ANGIOGRAPHY MRA 0.568
MR0048 MRI Face w/o Contrast MR FACE MR body 0.379
MR0049 MRI Abdomen w/ + w/o
Contrast MR ABDOMEN ABDOMINAL MR body 0.379
MR0050 MRI Abdomen w/ Contrast MR ABDOMEN ABDOMINAL MR body 0.379
MR0051 MRI Abdomen w/o Contrast MR ABDOMEN ABDOMINAL MR body 0.379
MR0052 MRI Adrenal Gland w/ +w/o
Contrast MR ADRENAL ABDOMINAL MR body 0.379
MR0053 MRI Adrenal Gland w/ Contrast MR ADRENAL ABDOMINAL MR body 0.379
MR0054 MRI Adrenal Gland w/o
Contrast MR ADRENAL ABDOMINAL MR body 0.379
MR0055 MRI Ankle w/ + w/o Contrast
Bilateral MR BANKLE MSK MR MSK 0.253
MR0056 MRI Ankle w/ + w/o Contrast
Left MR LANKLE MSK MR MSK 0.253
MR0057 MRI Ankle w/ + w/o Contrast
Right MR RANKLE MSK MR MSK 0.253
MR0058 MRI Ankle w/ Contrast Bilateral MR BANKLE MSK MR MSK 0.253
MR0059 MRI Ankle w/ Contrast Left MR LANKLE MSK MR MSK 0.253
MR0060 MRI Ankle w/ Contrast Right MR RANKLE MSK MR MSK 0.253
MR0061 MRI Ankle w/o Contrast
Bilateral MR BANKLE MSK MR MSK 0.253
MR0062 MRI Ankle w/o Contrast Left MR LANKLE MSK MR MSK 0.253
MR0063 MRI Ankle w/o Contrast Right MR RANKLE MSK MR MSK 0.253
MR0064 MRI Axilla w/ Contrast MR AXILLA CHEST MR body 0.379
MR0065 MRI Axilla w/+w/o Contrast MR AXILLA CHEST MR body 0.379
MR0066 MRI Axilla w/o Contrast MR AXILLA CHEST MR body 0.379
MR0067 MRI Brachial Plexus w/
Contrast MR NECK NEURO
MR brain/spine + additional
sequences 0.505
MR0068 MRI Brachial Plexus w/+w/o
Contrast MR NECK NEURO
MR brain/spine + additional
sequences 0.505
MR0069 MRI Brachial Plexus w/o
Contrast MR NECK NEURO
MR brain/spine + additional
sequences 0.505
MR0070 MRI Brain CSF Flow Study MR BRAIN NEURO MR brain 0.316
MR0071 MRI Brain Neuronavigation MR BRAIN NEURO MR brain 0.316
MR0072 MRI Brain Perfusion MR BRAIN NEURO MR brain 0.316
MR0073 MRI Brain Stroke MR BRAIN NEURO MR brain 0.316
MR0074 MRI Brain w/ + w/o Contrast MR BRAIN NEURO MR brain 0.316
MR0075 MRI Brain w/ Contrast MR BRAIN NEURO MR brain 0.316
MR0076 MRI Brain w/o Contrast MR BRAIN NEURO MR brain 0.316
59
MR0077 MRI Breast w/ + w/o Contrast
Bilateral MR BMAMMARY BREAST MR breast 0.568
MR0078 MRI Breast w/ + w/o Contrast
Left MR LMAMMARY BREAST MR breast 0.568
MR0079 MRI Breast w/ + w/o Contrast
Right MR RMAMMARY BREAST MR breast 0.568
MR0080 MRI Breast w/ Contrast
Bilateral MR BMAMMARY BREAST MR breast 0.568
MR0081 MRI Breast w/ Contrast Left MR LMAMMARY BREAST MR breast 0.568
MR0082 MRI Breast w/ Contrast Right MR RMAMMARY BREAST MR breast 0.568
MR0083 MRI Breast w/o Contrast
Bilateral MR BMAMMARY BREAST MR breast 0.568
MR0084 MRI Breast w/o Contrast Left MR LMAMMARY BREAST MR breast 0.568
MR0085 MRI Breast w/o Contrast Right MR RMAMMARY BREAST MR breast 0.568
MR0086 MRI Calf Left w/ Contrast MR LTIBFIB MSK MR MSK 0.253
MR0087 MRI Calf Left w/+w/o Contrast MR LTIBFIB MSK MR MSK 0.253
MR0088 MRI Calf Left w/o Contrast MR LTIBFIB MSK MR MSK 0.253
MR0089 MRI Calf Right w/ Contrast MR RTIBFIB MSK MR MSK 0.253
MR0090 MRI Calf Right w/+w/o Contrast MR RTIBFIB MSK MR MSK 0.253
MR0091 MRI Calf Right w/o Contrast MR RTIBFIB MSK MR MSK 0.253
MR0092 MRI Cardiac Function Complete MR CARDIAC CARDIOLOGY MR cardiac 0.758
MR0093 MRI Cardiac Function Limited MR CARDIAC CARDIOLOGY MR cardiac 0.758
MR0094 MRI Cardiac Morphology w/
Contrast MR CARDIAC CARDIOLOGY MR cardiac 0.758
MR0095 MRI Cardiac Morphology w/o
Contrast MR CARDIAC CARDIOLOGY MR cardiac 0.758
MR0096 MRI Cardiac Velocity Flow
Mapping MR CARDIAC CARDIOLOGY MR cardiac 0.758
MR0097 MRI Cervico-Thoracic Spine w/
Contrast MR CTSPINE NEURO MR spine 0.253
MR0098 MRI Cervico-Thoracic Spine
w/+w/o Contra MR CTSPINE NEURO MR spine 0.253
MR0099 MRI Cervico-Thoracic Spine
w/o Contrast MR CTSPINE NEURO MR spine 0.253
MR0100 MRI Cervix w/ Contrast MR PELVIS GYNECOLOGY MR body 0.379
top related