Health Care Intelligent System: Deep Residual Network Powered by Data Augmentation for Automatic Melanoma Image Classification

Melanoma, one of the most dangerous types of skin cancer, is a significant global health issue. Globally, 325,000 new melanoma cases are expected to be detected in 2020, according to the International Agency for Research on Cancer (IARC), with a predicted mortality rate of 57,000. Early detection of melanoma is crucial, and artificial intelligence (AI) algorithms have shown promise in automated detection. In this research, our goal was to develop a representative and accurate training dataset for an automated melanoma detection system using neural networks. To address the problem of overfitting images collected from open sources, we used transfer learning and data augmentation techniques. With an accuracy of 85.29%, our suggested model, which was based on the Fast AI framework and 1-cycle policy, outperformed other state-of-the-art methods. This result suggests that the model can help accurately analyze skin cancer situations, potentially preventing future risks and deaths. The proposed automated melanoma detection system can detect malignant melanoma and benign melanoma using simple numerical images. An automated technique for identifying melanoma has been developed as a result of our research. This approach has the potential to improve skin cancer detection, enabling prompt treatment and enhancing patient outcomes. This study emphasizes how AI algorithms may be used to address the rising global health problem of skin cancer.