Deep Learning Solutions for Skin Cancer Detection and Diagnosis

Abstract

Skin cancer, a concerning public health predicament, with over 5,000,000 newly identified cases every year, just in the United States. Generally, skin cancer is of two types: melanoma and non-melanoma. Melanoma also called as Malignant Melanoma is the 19th most frequently occurring cancer in women and men. It is the deadliest form of skin cancer [1]. In the year 2015, the global occurrence of melanoma was approximated to be over 350,000 cases, with around 60,000 deaths. The most prevalent non-melanoma tumours are squamous cell carcinoma and basal cell carcinoma. Non-melanoma skin cancer is the 5th most frequently occurring cancer, with over 1 million diagnoses worldwide in 2018 [2]. As of 2019, greater than 1.7 Million new cases are expected to be diagnosed [3]. Even though the mortality is significantly high, but when detected early, survival rate exceeds 95%. This motivates us to come up with a solution to save millions of lives by early detection of skin cancer. Convolutional Neural Network (CNN) or ConvNet, are a class of deep neural networks, basically generalized version of multi-layer perceptrons. CNNs have given highest accuracy in visual imaging tasks [4]. This project aims to develop a skin cancer detection CNN model which can classify the skin cancer types and help in early detection [5]. The CNN classification model will be developed in Python using Keras and Tensorflow in the backend. The model is developed and tested with different network architectures by varying the type of layers used to train the network including but not limited to Convolutional layers, Dropout layers, Pooling layers and Dense layers. The model will also make use of Transfer Learning techniques for early convergence. The model will be tested and trained on the dataset collected from the International Skin Imaging Collaboration (ISIC) challenge archives.