Deep Learning-Based Multi-Class Classification Of Skin Lesion Images

The classification of skin lesion images in medical image analysis remains a complex task due to the heterogeneous nature of skin diseases. A key challenge arises from the strong visual resemblance among different lesion categories, as benign and malignant cases often exhibit similar color distributions, textures, and structural characteristics. Traditional machine learning techniques rely heavily on handcrafted feature extraction, which is often inadequate for capturing subtle and discriminative patterns in such data. In contrast, deep learning approaches, particularly convolutional neural networks (cnns), have shown significant potential in automating image classification tasks. However, these models are often constrained by high computational requirements and issues related to generalization across diverse datasets. Addressing these limitations is essential for developing efficient, accurate, and deployable automated diagnostic systems. This study investigates multiple deep learning strategies, including transfer learning-based models and a custom-designed cnn architecture, using the ham10000 dataset. The performance of the models is evaluated using standard metrics such as accuracy, precision, recall, and f1-score. Experimental findings demonstrate that the proposed cnn model achieves a strong balance between computational efficiency and classification performance, making it a reliable approach for skin lesion image classification.