Diagnosis of Skin Cancer via Transfer Learning with Combined Channel attention and Spatial Attention

Early detection of skin cancer significantly increases survival rates. However, even experienced dermatologists find it challenging to identify early-stage malignant tumors from skin images. Consequently, several methods for classifying dermatoscopic images have been developed. Unfortunately, many of these methods are either inadequate for accurate skin cancer detection or require significant computational resources. This study evaluates and enhances the performance of VGG16 and IRV2 models with channel attention and spatial attention mechanisms for detecting skin cancer. By incorporating these attention mechanisms, our approach achieves notable accuracy rates of 88.5% with VGG16 model and 94.68% with IRv2 model on the HAM10000 dataset. These conclusions emphasize the efficiency of using attention-augmented models in improving the accuracy of skin cancer detection and suggesting a valuable tool for early diagnosis and clinical decision-making.