Abstract: Skin cancer presents a significant global health challenge, necessitating early and accurate diagnosis for patient survival. However, clinical evaluation of skin lesions is hindered by long waiting times and subjective interpretations. To address these issues, deep learning techniques have been leveraged to assist dermatologists in making more precise diagnoses. In this project, we aimed to develop reliable deep learning prediction models for skin cancer classification, addressing class imbalance and facilitating model interpretation. Initially, a Convolutional Neural Network (CNN) was optimized using the HAM10000 dataset, achieving 81% accuracy with the combination of Swish activation function and RMSprop optimization. To further enhance performance, we explored advanced models such as Xception and DenseNet, anticipating an accuracy of 90% or higher. Additionally, we propose extending the project by integrating these models into a user-friendly interface using the Flask framework, enabling user testing with authentication. This comprehensive approach holds promise for improving early detection and treatment of skin cancer, ultimately reducing its morbidity and mortality.
Index Terms: Skin cancer, Optimised CNN, Optimization functions, Activation functions
|
DOI:
10.17148/IJARCCE.2024.134147
