Abstract—Skin lesion classification presents significant challenges in the medical field, primarily due to the increasing complexity and diversity of datasets. Traditional methods often struggle with issues like class imbalance and unseen distribution of disease types, which are common in medical data. In this paper, we introduce a novel approach: the integration of the Energy Correlation (EC) method into Energy-Based Models (EBMs), hereafter abbreviated as EC-EBMs, which combines EBMs with EC to enhance the classification of skin lesion images. This method effectively addresses the challenges of class imbalance and data heterogeneity by optimizing the relationship between features and labels, minimizing intraclass energy, and maximizing interclass differences. Experimental results from two datasets demonstrate the effectiveness of EC-EBMs. On the ISIC 2020 dataset, our approach achieved an accuracy of 98.24% in distinguishing between malignant melanoma and benign non-melanoma lesions. Additionally, it showed promising results on the ISIC 2019 dataset, successfully classifying eight diagnostic labels with an accuracy of 72.33%, sensitivity of 55.09%, and specificity of 95.18%. These findings highlight the potential of EC-EBMs to improve classification performance and provide a new solution to complex challenges in medical classification.
 
Keywords—Energy-Based Models (EBMs), Energy Correlation (EC), skin lesion classification, ISIC datasets, nonlinear relationships, loss function optimization


Copyright © 2025 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).