ensemble convolutional neural networks for skin cancer detection 5336

Project TitleProject Area of Specialization Artificial IntelligenceProject Summary

Project Summary:

Skin cancer is an uncontrolled growth of abnormal cells. Skin damage with unrepaired DNA causing malignant tumors. It has ability to spread from one part of the body to other by the time passes. Melanoma is one of the common kind of skin cancer, which causes high number of causalities worldwide. The common types of melanomas are of skin-colored, brown or black, pink, red and many others, Melanoma is difficult to recognize at earlier stages, which is the major challenging problem for the dermatologist. There are two major stages of melanoma such as benign and malignant. The common and serious melanoma is benign, which is noncancerous tumor and do not have spread ability. However, it is treated as pre-cancerous signs most of the times. The second type of Melanoma is malignant form is very dangerous with no clear symptoms in advance. It is due to abnormal cell growth and can spread over the body. Fortunately, it can be clinically diagnosed and mostly curable up to 5% at premature stage. Therefore, skin cancer detection at an early stage is a significant requirement. Yet, some factors like irregularities, color and border area of cancerous cells make its detection complicated.  Hence, the objective of this research is to propose an ensemble convolutional neural network model for the detection of melanoma at an early stage and save patient survival rate.

Project Objectives

Project Objectives:

• This system will detect/classify the skin cancer at an early stage.
• This system will have a user interface in which user can insert image as an input and after complete processing on image system will shows a grade of the disease (i.e., benign and malignant).
• A new deep learning model is proposed for classification of skin cancer.

Project Implementation Method

Software Implementation

Output

Input Images

Benign

Feature Maps

Feature Maps

Feature Maps

              

Benign

                                                                                                                                                                   

Convolutions

Malignant

Subsampling

Convolutions

Subsampling

                                          Figure1. General steps of the proposed deep learning model

Hardware Implementation: 

Output Input      Deploying On Ultra 96 Benign/Malignant

Figure 2. Flow diagram of hardware implementation

        Output

        Input Images

Benign

Feature Maps

Feature Maps

Feature Maps

Benign

Benign

  Convolutions

Malignant

   Subsampling

Convolutions

    Subsampling

     

Output

Input

Deploying On Ultra 96

Benign/Malignant

Output

Output

Input

Input

Deploying On Ultra 96

     Deploying On Ultra 96

Benign/Malignant

Benign/Malignant

Benefits of the Project

• This system will detect the beginning skin cancer.
• This system will have a user interface in which user can insert image as an input and after complete processing on image system will shows a name of disease.
• In processing include contrast enhancement, segmentation and features extraction

Technical Details of Final Deliverable

Technical Details:

Matlab 2019b

Deep learning toolkit (DLTK)

Final Deliverable of the Project HW/SW integrated systemCore Industry MedicalOther Industries IT Core Technology Artificial Intelligence(AI)Other TechnologiesSustainable Development Goals Good Health and Well-Being for PeopleRequired Resources

Input Images