Diagnosis of Melanoma with Artificial Intelligence Techniques: An Evaluation of Hybrid Models Using Deep Learning and Classical Machine Learning Techniques

Cancer remains one of the most significant global health challenges due to its high mortality rates and the limited understanding surrounding its progression. Early diagnosis is critical to improving patient outcomes, especially for skin cancer, where timely detection can dramatically enhance recovery rates. Histopathological analysis is a common diagnostic method for skin cancer, but it is a cumbersome process that relies heavily on the expertise of highly trained specialists. Recent advancements in Artificial Intelligence have demonstrated promising results in image classification tasks, suggesting its potential as a support tool for medical diagnosis. This study explores applying hybrid models for melanoma diagnosis using histopathological images. We propose a two-step framework that leverages an Autoencoder for dimensionality reduction and feature extraction, followed by a classification algorithm for melanoma-nevi discrimination, where we tested Support Vector Machines, Random Forest, Multilayer Perceptron, and K-Nearest Neighbours. Among the models evaluated, the Autoencoder combined with K-Nearest Neighbours or Random Forest achieved the best performance, with an accuracy of approximately 97.95%.