Pixel Based Multi Class Skin Cancer Classification using Convolutional Neural Network

Abstract

Skin cancer poses a significant societal concern, as pigments responsible for skin colour can become carcinogenic, leading to the development of the disease. Detecting lesions early is vital for effective skin cancer treatment, yet diagnosis is challenging due to the similarity of many pigments. To aid dermatologists, substantial advancements have been achieved in developing automated tools utilizing artificial intelligence. One notable tool is a Computer-Aided Diagnosis (CAD) system employing a Convolutional Neural Network (CNN) to classify seven types of skin cancer from atypical lesion images. In recent years, there has been significant progress in the development of automated cancer classification systems utilizing deep learning techniques. However, these systems often exhibit bias towards fair skin tones due to datasets skewed towards lighter skin complexions. This project addresses this challenge by proposing a model comparing among Convolution Neural Network and Multi - Layer Perceptron deep learning architecture specifically designed for fair skin cancer classification. The proposed model incorporates with a pre - trained model on an imbalanced pixel dataset to mitigate bias on skin colour and improve classification accuracy for fair skin cancers. The model is evaluated on a HAM10000 pixel dataset for fair skin cancer classification, demonstrating competitive performance while promoting fairness in cancer diagnosis using minimal resources..