Skin Cancer Classification from Skin Lesion Images Using Modified Depthwise Convolution Neural Network

Abstract

Nowadays, skin diseases are among the most common health issues faced by people. Skin cancer (SC) is one of these diseases, and its detection relies on skin biopsy results and the expertise of doctors. However, this process is time-consuming and has poor accuracy. Detecting SC at an early stage is challenging, as it can quickly spread throughout the body, leading to higher mortality rates. Early detection of SC is crucial for successful treatment. The critical task in achieving accurate SC classification lies in identifying and classifying SC based on various features such as shape, size, color, symmetry, etc., which are also present in many other skin diseases. Selecting relevant features from a SC dataset image poses a significant challenge. Therefore, an automated SC detection and classification framework is required to improve diagnostic accuracy and address the shortage of human experts. In this paper, we implement a modified depth-wise Convolutional Neural Network (D-CNN) and compare its performance with other CNN frameworks, namely Deep Belief Network (DBN) and CNN-based cascaded ensemble network. We evaluate the effectiveness of SC identification using depth-wise CNN technique by employing performance metrics such as precision, recall, accuracy, sensitivity, specificity, and F-measure. The proposed technique not only improves classification accuracy but also reduces computational complexities and time consumption.