Enhancing Skin Cancer Detection Using AlexNet Empowered Transfer Learning

Abstract

Skin diseases necessitate early detection for effective treatment, and deep learning technology has proven instrumental in their classification. In this study, we introduce an innovative approach that employs transfer learning for skin disease prediction, leveraging a comprehensive dataset of skin images. Our model, the Skin Cancer Detection Transfer Learning Algorithm (SCTLA), demonstrates a commendable average accuracy of 98.28% on the dataset, representing a substantial progression in skin disease detection. While our findings may not be considered groundbreaking, they contribute significantly to the timely and precise diagnosis of skin diseases, ultimately enhancing patient care and outcomes, particularly in cases where rapid intervention is essential. The SCTLA was developed and tested on MATLAB 2018 using a single GPU NVIDIA 840m with for processing during training. The dataset, sourced from Kaggle, focuses on binary classification of skin lesions into benign and malignant categories, with 80% allocated for training and 20% for validation. Evaluation metrics such as sensitivity, specificity, precision, accuracy, and the Matthews Correlation Coefficient demonstrate the SCTLA's robust performance. The robustness of SCTLA underscores its potential as a powerful tool in the hands of medical practitioners, contributing to a paradigm shift in the landscape of skin disease diagnosis. The integration of deep learning techniques underscores the study's pivotal role in advancing early detection practices for skin diseases.