A Robust Multi-head Self-attention-Based Framework for Melanoma Detection

Melanoma has the potential to spread to several body areas if it is not found on time, which makes it one of the world’s most serious illnesses. Of all skin tumors, melanoma is one of the most deadly and quickly spreading conditions. Recently, a lot of research has been focused on convolutional neural networks (CNNs), which comprise the majority of deep learning methods, for their ability to detect skin malignancies in nearly identical images. With the development of Artificial Intelligence (AI) systems with Deep Learning and Machine Learning, the healthcare system now has impressive automation and cutting-edge options. AI-driven automated diagnosis tools help the medical field identify the illness they are treating. The suggested strategy for early melanoma detection from the phase of the image is to stop the spread of the virus. The suggested technique uses a multi-head self-attention-based transformer architecture to extract more pertinent information from melanoma images. The model was made more robust and generalized in the proposed study through data augmentation, enabling deployment in real-time applications. For the PH2 dataset, the proposed multi-head attention-based technique obtained 99.11% outstanding accuracy.