Double AMIS-ensemble deep learning for skin cancer classification

This study aims to create a precise skin cancer classification system (SC-CS) able to distinguish various skin cancer types. Targeted categories include melanoma, vascular lesions, melanocytic nevus, cutaneous fibromas, benign keratosis, and different carcinomas and skin moles. The system employs image segmentation and convolutional neural network (CNN) algorithms within a double artificial multiple intelligence system (AMIS) ensemble model, optimizing results weighting for superior solution quality. Evaluation of the SC-CS involved assessing metrics like accuracy, precision, AUC, and F1-score, and gathering feedback from 31 volunteers, including dermatologists, medical professionals and other medical staff with expertise in skin cancer diagnosis. The system accurately classified varied skin cancer types using HAM10000 and malignant vs. benign datasets, chosen for their broad representation and research utility. With over 99.4% accuracy, it surpassed state-of-the-art models by 2.1% for larger and 15.7% for smaller sizes. The high system usability scale (SUS) score of 96.85% signals strong user satisfaction and recommendation potential. High data security ensured patient privacy, with no access to personal information. Uploaded data and images are promptly deleted after processing. The system aids dermatologists in accurate skin cancer diagnosis, integrating into workflows with resource requirements. Performance varies with image quality, dataset diversity, and population-specific variations.