The future of skin cancer diagnosis: a comprehensive systematic literature review of machine learning and deep learning models

Abstract

Skin cancer, a life-threatening disease, necessitates early detection and accurate classification for successful treatment. Misdiagnoses can lead to significant consequences for patients, highlighting the critical need for improved accuracy. Despite advancements in machine learning (ML) and deep learning (DL) techniques, accurate diagnosis remains challenging due to its complex nature. This comprehensive Systematic Literature Review (SLR) aims to examine the use of ML and DL models in skin cancer detection and classification, including convolutional neural networks (CNNs), recurrent neural networks (RNNs), and transfer learning. Model performance is evaluated based on accuracy, sensitivity, specificity, and precision. Key findings reveal the dominance of DL models, with SVM-PSO emerging as a top-performing hybrid model with 97.50% accuracy. Tailored models, such as M-SVM and FCN-ResAlexNet, demonstrate high accuracy, emphasizing the importance of customization for dermatology tasks. Deep neural networks, such as ResNet-50, ResNet34, Inception V3, and ResNet 152, consistently exhibit strong performance, highlighting the impact of architectural depth. Traditional ML algorithms like Random Forest, KNN, and Naive Bayes face challenges compared to DL models. Furthermore, the analysis explores correlations between dataset size and accuracy, revealing varied model responses. Temporal trends and model-specific analyses uncover outliers, anomalies, and the influence of specific datasets (e.g. imbalanced classes), providing valuable insights for future research and model development. This multifaceted nature of model performance, influenced by factors beyond dataset size, underscores the need for nuanced considerations in dermatology image classification. Overall, the findings of this SLR offer valuable insights for researchers and practitioners, serving as a crucial step towards developing even more accurate and reliable tools for skin cancer diagnosis.

Keywords

• Classification
• machine learning
• deep learning
• skin
• cancer
• Artificial Intelligence