Information Theory Applied to Classifying Skin Lesions in Supporting the Medical Diagnosis of Melanomas

In this paper, we propose a classification technique for melanocytic lesions based on fundamentals of Information Theory. In particular, we evaluate the accuracy level of the Correntropy Coefficient as a similarity measure in classifying Melanoma and (common and atypical) nevi. These lesions were chosen because they are generally similar to each other, then it can be difficult to carry out a differential diagnosis of the melanocytic lesion type. The effectiveness of the proposed approach was verified through a case study using a public dermoscopic image dataset. The obtained results for performance evaluation and comparison show very high accuracy for melanoma classification which outperforms state-of-the-art approaches. Besides, considering the simplicity of the proposed technique and the results obtained, it is possible to use this approach in developing computational systems to support the medical diagnosis of melanomas.