Skin lesion classification presents significant challenges in medicine, especially due to the increasing complexity of data. Traditional methods often face difficulties when dealing with heterogeneous datasets, particularly those in the medical domain, which frequently exhibit class imbalance and uneven distribution of disease types. In this paper, we propose a novel approach that integrates Energy-Based Models (EBMs) with Energy Correlation to classify skin lesion images. Our method optimizes the relationship between features and labels by minimizing intra-class energy while maximizing inter-class differences. Experiments conducted on the ISIC 2019 dataset demonstrated promising results, achieving 72.94% accuracy, 53.45% sensitivity, and 95.30% specificity. This approach significantly enhances classification performance and introduces a new paradigm for addressing complex medical classification problems.