Fusion of Dermoscopic Datasets: A Method to Promote Performance of Deep Learningin Melanoma Detection

Skin cancer is one of the diseases that has spread a lot in recent years. The abnormal growth of skin cells causesthis cancer. If this disease is diagnosed in the early stages, it can be easily treated and prevent the possible death ofthe patient. Therefore, artificial intelligence experts have made great efforts to identify and diagnose this diseasewith the help of computer-aided detection (CAD) systems. Based on this, machine learning-based approacheshave become an efficient way to classify skin lesions. This paper presents a CAD system for classifying melanomaand nevus. We used the combination of two ISIC ۲۰۱۹ and ISIC ۲۰۲۰ datasets to overcome the data imbalance.A pre-trained model with extra layers is used to feature extraction and classify skin lesions. The proposedmethod achieves ۹۴% sensitivity, ۹۷% specificity, and ۹۵% accuracy on test data.

کلیدواژه ها:

Skin cancer ، Melanoma ، Deep learning ، Classification ، Dataset fusion