Development of Hybrid Convolutional Neural Network and Auto-Regressive Integrated Moving Average for Skin Cancer Classification

KA CHIN, CHEE (2022) Development of Hybrid Convolutional Neural Network and Auto-Regressive Integrated Moving Average for Skin Cancer Classification. Masters thesis, Universiti Malaysia Sarawak.

PDF (Please get the password by email to repository@unimas.my , or call ext: 082-583914/3973/3933) Chee.pdf Restricted to Registered users only Download (2MB)

Abstract

Skin cancer is one of the most lethal illnesses in humans. Dermatologists spend much more time investigating these lesions due to the high similarities between different skin cancer forms. The existing methods, such as ABCDE rule and 7-point checklist, can guide the dermatologists to analyse the skin lesion, but these techniques can only distinguish benign and malignant of skin cancer. The use of Artificial Intelligence (AI) based Computer-Aided Decision (CAD) systems on automating the classification of skin cancer would save time, effort, and human lives. Although machine learning and deep learning are widely used in skin cancer diagnosis, machine learning is still unable to get the deep features from network flow. Deep learning also has the complex network with an enormous number of parameters that may resulting in the low and limited classification accuracy performance. In this research, the process of AI-based CAD system for skin cancer classification is introduced. The suitable image pre-processing technique and image augmentation method are identified to avoid the degradation of classification accuracy in this research. Other than that, this research also focuses on the use of Convolutional Neural Network (CNN) as a feature extractor and Auto-Regressive Integrated Moving Average (ARIMA) behaved as a skin cancer classifier. Thus, the new hybrid CNN-ARIMA is proposed, which is able to classify skin cancer images successfully with test accuracy, average sensitivity, average specificity, average precision, and AUC of 96.00%, 96.02%, 97.98%, 96.13%, and 0.995, respectively.

Actions (For repository members only: login required)

View Item