Lung cancer detection system using neural networks and image processing

LUNG CANCER DETECTION SYSTEM USING NEURAL NETWORKS AND IMAGE PROCESSING

Presented By:

H.N.Gunasinghe

AS2010379

CSC 364 1.5 Seminar IIDepartment of Computer Science and Statistics , USJP

2

H.N.Gunasinghe

ACKNOWLEDGEMENT

This project was done By : K.A.G. Udeshani AS2006612

Supervisors : Dr. T.G.I. Fernando Dr. R.G.N. Meegama

Publication: Statistical Feature-based Neural Network

Approach for the Detection of Lung Cancer in Chest X-Ray Images

International Journal of Image Processing (IJIP), Volume (5) : Issue (4) : 2011 425 [link]

AS2010379

3

H.N.Gunasinghe

OVERVIEW

Introduction Literature review Methodology Results and discussion Conclusion Summery

AS2010379

4

H.N.Gunasinghe

INTRODUCTION

Lung cancer is a one of major cancers in Sri Lanka

Still a exact treatment is not found Early detection of is Lung cancer important

for a successful treatment. chest X-rays are considered to be the most

widely used technique for the detection of lung cancer.

It is a complex task to analyze these images as they are projected images.

A medical expert has to make extensive knowledge of anatomy and imaging techniques.

AS2010379

5

H.N.Gunasinghe

X-Ray image

System

Nodule

Non- Nodule

PROBLEM STATEMENT Aim-

To develop a lung cancer detection system using chest X-Ray images as input

This is a Hybrid method-

System

Neural Networks

Feature extraction

Image processing

AS2010379

6

H.N.Gunasinghe

WHAT IS A LUNG CANCER ?

Tumors arising from cells lining the airways of the respiratory system and it will expand into airways.

These cells are often in bright contrast in chest X-rays and take the shape of a round object.

Other diseases that can be seen in chest x-ray Pneumonia Tuberculosis Lung abscess

AS2010379

7

H.N.Gunasinghe

OBJECTIVES

1. To develop a lung cancer detection system using Neural Network and Image Processing.

2. To generate more accurate results.3. To minimize the computational time to get

an output from the system.4. To help the people in Sri Lanka to Identify

lung cancer in early stages.

AS2010379

8

H.N.Gunasinghe

MAJOR ACHIEVEMENTS

The system was successfully built using Neural Networks and image processing techniques

This system in cooperates an effective GUI with ease of interaction

AS2010379

9

H.N.Gunasinghe

LITERATURE REVIEW

Digital photographs of chest X-rays and CT scans have been used to identify lung cancer.

Principles of neural networks have been widely used for the detection of lung cancer in medical images with simulated lung nodules [1, 2] massive training artificial neural networks[4] two level neural classifiers [5] hybrid lung nodule detection [6] ladder structured decision trees [3, 7]

AS2010379

10

H.N.Gunasinghe

No Other researches Lung cancer detection system using Neural Network and Image Processing techniques

1 Computational effect is high

Computational effect is less, since considering only10 features.

2 Mainly focused on one approach

Two approaches were used. Pixel-based and feature-based detection with Image Processing techniques. View the suspicious areas of the lungs those can be lung nodules.

3 Use Neural Networks Use Neural Network with different architecture.

4 Not flexible This system is flexible and practical and may one can easily use this system for further improvements.

RESEARCH GAP

AS2010379

11

H.N.Gunasinghe

RESEARCH QUESTIONS

Identify the lung cancer and what are the methods that can be used to design software solution.

How doctor identifies a lung cancer using a chest X-ray.

Identify suitable methods and Image Processing techniques to extract the features from a digital image of a chest X-ray.

The optional features that can be considered as the input to the Neural Network.

Identify how Neural Network using Matlab. Identify suitable method to design the optimum

architecture of a Neural Network. Identify the use of Graphical User Interface in

Matlab.

AS2010379

12

H.N.Gunasinghe

METHODOLOGY (1)49

49

Features Pixels

Results

System Overview

AS2010379

13

H.N.Gunasinghe

METHODOLOGY (2)-SYSTEM DESIGN- Median Filtering

Sharpening

Histogram Equalization

AS2010379

14

H.N.Gunasinghe

METHODOLOGY (3)-FEATURE EXTRACTION-

Features extracted from an image1. Average gray level2. Uniformity3. Entropy4. Standard deviation 5. Skewness

Tackled noise removed by preprocessing image Most projects used integrated preprocessing

techniques to extract the lung region Extracted region was used to apply for further image processing techniques.

6. Smoothness

7. Contrast

8. Homogeneity

9. Energy

10. correlation

AS2010379

15

H.N.Gunasinghe

This system uses a neural network with one hidden layer containing 1000 neurons, and an output layer with 1 neuron.

pixel-based intensity input vectors - purelin and tansig transfer functions

feature-based inputs vectors - two tansig transfer functions

METHODOLOGY (4) -NEURAL NETWORK - BACKPROPAGATION

AS2010379

16

H.N.Gunasinghe

METHODOLOGY (5) – CONNECTED COMPONENT ANALYSIS

AS2010379

original image median filtering Histogram equalization

Labeling connected components

binary threshold image

17

H.N.Gunasinghe

RESULTS AND DISCUSSION (1)

Results obtained when testing for the accuracy of the system

Two main areas that has to be considered1. Neural network

a. Pixel based technique (2401 pixels)b. Feature based technique (10 features)

2. Connected component analysisa. View suspicious areasb. Calculate the roundness of the connected

components

AS2010379

18

H.N.Gunasinghe

RESULTS AND DISCUSSION (2) – SELECT NN

pixel-based feature-based

Training graph of the neural network

R 1 0.737

MSE 1.2682e-009 0.2684

AS2010379

19

H.N.Gunasinghe

RESULTS AND DISCUSSION (3) – RESULTS OF NN

managed to achieve a high recognition rate for a nodule when the neural network was trained using pixel-based intensity values.

Recognizing a non-nodule was 16% lower with statistical feature-based training of the neural network.

AS2010379

20

H.N.Gunasinghe

RESULTS AND DISCUSSION (3) – FEATURE EXTRACTION -

AS2010379

21

H.N.Gunasinghe

CONCLUSION This research was completed with good background

knowledge of lung cancer detection systems using computer intelligence

The detection rate of Feature based technique – 88% Pixel based technique – 96%

Successfully developed a solution using Neural Networks and image processing techniques with a GUI

A user has only to select the digital chest x ray as input and system will show suspicious areas of the chest x ray and the presence of lung nodules

It is considered only the visible area of the chest x – ray for the nodule detection

AS2010379

22

H.N.Gunasinghe

ASSUMPTIONS AND LIMITATIONS

Low subtlety images were used Any algorithm wasn’t used to avoid rib

shadows

Try the system with many preprocessing techniques

Develop lung region segmentation algorithm to use with many databases.

Try with different NN architectures Develop algorithms to overcome rib

shadows. Apply these techniques to identify other

cancers AS2010379

FUTURE WORKS

23

H.N.Gunasinghe

RESENT RESEARCHES

Soft Tool Development for Characterization of Lung Nodule from Chest X-ray Image International Journal of Image Processing and Vision

Sciences ISSN (Print): 2278 – 1110, Volume-2, Issue-1, 2012 [link]

Feature Extraction and Principal Component Analysis for Lung Cancer Detection in CT scan Images Ada et al., International Journal of Advanced Research

in Computer Science and Software Engineering 3(3), March - 2013, pp. 187-190 [link]

Oral cancer prognosis based on clinicopathologic

and genomic markers using a hybrid of feature

selection and machine learning methods Chang et al. BMC Bioinformatics 2013, 14:170 [link]

AS2010379

24

H.N.Gunasinghe

REFERENCES [1] P.R. Snoeren, G.J.S. Litjens, B.V. Ginneken and N. Karssemeijer, Training a

computer aided detection system with simulated lung nodules in chest radiographs, Proc. 3rd International Workshop on Pulmonary Image Analysis, Beijing, 2010.

[2] G. Coppini, S. Diciotti, M. Falchini, N. Villari and G. Valli, Neural networks for computer aided diagnosis: detection of lung nodules in chest radiograms, IEEE Trans. On Information Technology in Biomedicine, vol. 4, pp. 344-357, 2003.

[3] M.G. Penedo, M.J. Carreira, A. Mosquera and D. Cabello, Computer aided diagnosis: A neural network based approach to lung nodule detection, IEEE Trans. on Medical Imaging, vol. 17, N 6. pp. 872-880, 1998.

[4] K. Suzuki, J. Shiraishi, H. Abe, H. MacMahon and K. Doi, False-positive reduction in computer-aided diagnostic scheme for detecting nodules in chest radiographs by means of massive training artificial neural network, Academi Radiology, vol. 12, N 2, pp. 191-201, 2003.

[5] J.S. Lin, S.B. Lo, A. Hasegawa, M.T. Freedman and S.K. Mun, Reduction of false positives in lung nodule detection using a two-level neural classification, IEEE Trans.

On Medical Imaging, vol. 15, pp. 206-216, 1996. [6] Y.S.P. Chiou, Y.M.F. Lure and P.A. Ligomenides, Neural networks image analysis

and classification in hybrid lung nodule detection (HLND) system, IEEE Workshop on Neural Networks for Signal Processing, pp. 517-526, 1993.

[7] D.H. Ballard and J. Sklansky, A ladder-structured decision tree for recognizing tumors in chest radiographs, IEEE Trans. on Computers, vol. C-25, pp. 503-513, 1976.

AS2010379

THANK YOU

Q&A