Automatic Skin Cancer Diagnosis in Dermoscopic Imaging Using Brute Force and Xception

Skin cancer is a serious health issue that, if not identified and treated promptly, might have disastrous effects. Traditional diagnostic approaches rely mainly on dermatologists performing visual inspections, which can be subjective and time-consuming. Deep learning architectures have demonstrated significant potential in the domain of medical diagnostics, specifically in the identification and classification of skin cancer. They can automatically learn and extract complex information from medical images, resulting in more precise and efficient diagnoses. This paper proposes a model for automatically detecting skin cancer. The proposed model utilizes the Xception deep learning architecture to distinguish benign and malignant dermoscopic images. It consists of two primary phases: data preprocessing and classification. In data preprocessing, various data augmentation techniques, followed by data oversampling, are applied to the original dataset. Finally, in the classification phase, brute force for hyperparameter optimization of Xception is applied. ISIC2020 is adopted for evaluating the proposed model. The proposed model obtained an accuracy of 95.2%, precision of 98.9%, sensitivity of 90%, and F1-score of 95.2%. Furthermore, the results demonstrated how competitive the suggested model is when compared to the state-of-the-art models.