Segmentation and Classification of skin Lesion for disease diagnosis

Project Title

Segmentation and Classification of skin Lesion for disease diagnosis

Project Area of Specialization Artificial IntelligenceProject Summary

This project consists of three main stages: pre-processing, segmentation of skin lesion, and classification of lesion into pre-defined classes.  At the first stage, images are pre-processed to remove unwanted hairs and reduce the noise. In the next stage, the pre-processed images are decomposed into segments based on the texture distinctiveness called as texture segments. Then, the feature extraction is performed from each texture segment where texture based features, shape based features and color space based features are extracted. Using the ground truth information of skin images lesion and non-lesion texture segments along with their class label information, the conventional machine learning algorithms like Support Vector Machine, Decision Trees, and Distance based classifiers are trained on the extracted features of lesion and non-lesion segments. The trained machine learning algorithms are used to predict the class of skin lesion in the test images. In order to compare the performance of the traditional machine learning algorithms with the deep learning algorithms, we use the pre-trained convolutional neural networks from the medical images domain for segmentation and classification of skin lesion images. The available skin lesion datasets like ISIC and dermquest as are used to evaluate the performance of the segmentation and classification of the proposed framework.

Project Objectives

The following are the set of objectives for the proposed project:

1) collection of the skin lesion images

2) proposing new techniques for preprocessing the skin images removing the image artifacts

3) evaluating the existing segmentation techinques and improving them for enhancing the accuracy

4) exploring the different features sets describing the skin lesion characterisitcs

5) exploring the available machine learing algorithms for high classification of skin lesions into pre-defined classes

6) using the CNNs pre-trained  on the medical images for skin lesion segmentation and classification

7) adapting the conventional and deep learning techniques for improving the segmentation and classification accuracy 

Project Implementation Method

1) Using the off-the-shelf available codes tailoring to fit the needs of project.

2) Implementing the newly proposed approaches in Python 

3) Using the available conventional machine learning algorithms in Python

4) Using the available deep learning nets libraries in Python, tensor flow, keras etc. executing on GPUs with parallel processing stages

5) Implementing the framework, translating the code into a format to low level language achieving the real time performance

Benefits of the Project

1) Computer aided diagnosis (CAD) of the skin diseases

2) skin diagnosis software available for clinical support to the dermatologists

3) useful for the telemedicine 

4) Computer aided diagnosis for the consumer grade camera images which can be used by the patients for early diagnosis and to send a practioner for early treatment suggestions

5) mobile apps development 

6) Tuning of the CAD system for the specific skin diseases classificaition prevailing in the region

Technical Details of Final Deliverable

1) Theory and Python based code of the developed CAD framework

2) Theory and code of the novel techniques for pre-processing of skin lesions, segmentation and classification of region-specific skin lesion diseases

3) A skin database of the region-specific skin diseases images

4) Report of the project containing the explanation of the implemented techniques and execution of the software

5) Relaibality evaluation of the CAD system 

Final Deliverable of the Project Software SystemType of Industry Health Technologies Artificial Intelligence(AI)Sustainable Development Goals Good Health and Well-Being for PeopleRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	70000
GPU	Equipment	1	45000	45000
Desktop CPU	Equipment	1	25000	25000