Skinnet: A lightweight deep learning model for skin cancer diagnosis

Skin cancer is among the most widespread and life-threatening cancers globally. If not detected at an early stage, skin cancer may result in severe tissue damage, costly treatment, and even death. Traditional diagnosis depends on physical examination by dermatologists. This process can be time-consuming and subjective. Recently, computer aided diagnostic tools have gained popularity due to their speed and consistent performance. These tools support dermatologists in identifying early signs of cancer. Deep Learning (DL) models have shown strong performance in image analysis. Research on DL based skin cancer detection continues to expand. This paper proposes a lightweight DL framework for skin cancer diagnosis by integrating Improved ShuffleNet V2 and Multilayer Network (MLN), named SKINNet. The proposed SKINNet involves three phases: image preprocessing, feature extraction, and classification. Initially, dermoscopic images are preprocessed to remove noise and improve the image quality. Subsequently, Improved ShuffleNet V2 is utilized to extract deep features from the preprocessed dermoscopic images. Finally, these features are passed to MLN for final classification. Performance of the SKINNet is validated using two benchmark datasets and results are computed across various performance metrics. SKINNet demonstrated a superior recognition accuracy of 98.91 %for HAM1000 and 99.01% for ISIC 2019. These findings highlight the potential of DL for accurate skin cancer diagnosis. The continued fusion of artificial intelligence into clinical practice has the potential to strength diagnostic reliability and enhance patient outcomes.