Artificial Neural Networks applications in Computer Aided Diagnosis. System design and use as an educational tool. Jorge Hernández Rodríguez Education in the Knowledge Society PhD Program Hernández Rodríguez, Jorge; Rodríguez Conde, María José; Cabrero Fraile, Francisco Javier Track 16: Doctoral Consortium
23
Embed
Artificial Neural Networks applications in Computer Aided Diagnosis. System design and use as an educational tool
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Artificial Neural Networks applications in Computer Aided Diagnosis. System design and use as an educational tool.
Jorge Hernández RodríguezEducation in the Knowledge Society PhD Program
Hernández Rodríguez, Jorge; Rodríguez Conde, María José; Cabrero Fraile, Francisco Javier
Track 16: Doctoral Consortium
2
INDEX
● Context and motivation that drives the dissertation research
● State of the art
● Hypothesis and problem statement
● Research objective and goals
● Research approaches and methods
● Results to date and their validity. Dissertation status.
● Current and expected contributions
Track 16: Artificial Neural Networks application in Computer Aided Diagnosis. System design and use as an educational tool
3
Track 16: Artificial Neural Networks application in Computer Aided Diagnosis. System design and use as an educational tool
CONTEXT AND MOTIVATION THAT DRIVES THE DISSERTATION RESEARCH (I)
Computer Aided Detection and
Diagnosis (CAD)
Software and algorithms (image processing, lesion detection
and classification algorithms)“Second opinion” for the
radiologist
CADe CADx
Software specialized in lesion and pathological
features detection
Software specialized in pathology characterization ,
classification or diagnosis
Radiological image modalities
Computed TomographyMammography
Conventional RadiologyMagnetic Resonance Imaging
Ultrasound…
4
Track 16: Artificial Neural Networks application in Computer Aided Diagnosis. System design and use as an educational tool
CONTEXT AND MOTIVATION THAT DRIVES THE DISSERTATION RESEARCH (II)
CAD UTILITY
● Increasing number of high technology equipment in hospitals and clinics● Radiological examinations with a high number of images and rising number of patients scanned (CT, MRI, screening, interventional radiology…)● Growth of clinical indications for different types of modalities
¡ Huge amount of workload for the
radiologist!
CAD: very useful tool to ease workload and improve detection
and diagnosis and therefore posterior treatment prescription.
Reported in scientific and technical literature:► The utility of these systems has been confirmed in numerous articles
► Reduction of inter-observer variability associated with image interpretation ► Sensitivity and specificity improvement associated with its use ► Improvement in specialists’ Receiver Operating Characteristic curves ► They provide supplementary information for managing a problem ► They have been successfully integrated into Picture Archiving and
Communication Systems and Radiology Information Systems
OBJECTIVE ■ Assisting radiologists in
detection and diagnosis
■ Great potential for medical specialists’ training and as an educational tool
5
Track 16: Artificial Neural Networks application in Computer Aided Diagnosis. System design and use as an educational tool
STATE OF THE ART (I)
Artificial Intelligence
Artificial Neural
Networks(ANNs)
6
Track 16: Artificial Neural Networks application in Computer Aided Diagnosis. System design and use as an educational tool
STATE OF THE ART (II)
Evolution of ANNs in Medical Imaging↓
Convolutional Neural Networks (CNNs)Neural Networks based in Deep Learning Methods
↓Highly non-linear systems
Designed to work directly with imagesDeeper architectures (higher number of layers)
Reduction in the amount of adjustable network parametersLimited size of training and validation datasets
↓ Calculation of empirical features of segmented lesions in images (manually or automatically)
Use of a classification algorithm (for example, linear discriminant analysis, support vector machines, artificial neural networks
“Traditional CAD schemes”
7
Track 16: Artificial Neural Networks application in Computer Aided Diagnosis. System design and use as an educational tool
STATE OF THE ART (III): Mammography
Arévalo et al. (2016). Computers and Methods in Biomedicine
CNNs based in Deep learning techniquesMammography database: BCDR
Preprocessing
• Image cropping• Data augmentation• Global Contrast Normalization• Local Contrast Normalization
Sample of marked lesion
Supervised learning through CNN and
classification
Architecture of the CNN that performed better
Evaluation of results Comparison with other
state-of-the art representations for
lesion classification and image analysis
8
Track 16: Artificial Neural Networks application in Computer Aided Diagnosis. System design and use as an educational tool
STATE OF THE ART (IV): Pulmonary node classification
Cheng et al. (2016). Scientific reports - Nature
Deep learning techniques
SDAE classifier achieved better results
Patterns from classifier’s
hidden layers
Examples of rated nodules
Shen et al. (2015). Information processing in medical imaging Multi-scale CNNs
Classification accuracy of 86,4% on LIDC-IDRI database
Track 16: Artificial Neural Networks application in Computer Aided Diagnosis. System design and use as an educational tool
STATE OF THE ART (V): Computer-Aided Detection
Roth et.al. (2016). IEEE Transactions on Medical Imaging
Convolutional Neural Networks
● Hierarchical two-tiered CADe system● New approach: 2,5 D image decomposition● Wide range of applications: