Siddharth Krishnakumar, TJHSST

Skin Cancer is the most common cancer diagnosed in the United States and Melanoma is the deadliest of all skin cancers. Out of the 5 million incidents of Skin Cancer diagnosed annually in the U.S., melanoma causes the most deaths. Melanoma kills an estimated 9,320 people in the US annually, and it is curable if diagnosed and treated early. The problem with an early diagnosis is that only highly trained dermatologists can accurately recognize melanoma skin lesions. Even expert dermatologists can diagnose melanoma with an accuracy of only 66%. Therefore, there is a need for automated dermoscopy analysis to identify melanomic lesions. Dermoscopy involves obtaining high-resolution magnified images by controlling light and removing surface skin reflectance. Many deep machine learning models based on Neural networks, such as Convolutional Neural Network (CNN), are currently deployed for automatic detection of melanoma from dermoscopy images. In this paper, I propose a deep learning neural network architecture based on the modified AlexNet Convolutional Network to diagnose melanoma. My project uses the Keras Python library on top of Google’s Tensor flow deep learning library, and I have used skin lesion image data released by the ISIC. My proposed deep learning neural network architecture achieved a training accuracy of 97.5% and a validation accuracy of 74%. My project was able to diagnose Melanoma from dermoscopy  images with better accuracy than any other neural network architecture studied. My project can be used as a prescreening tool by dermatologist to rapidly diagnose Melanoma with better accuracy.