DermLens: The Enhanced Skin Lesion Classification Using Deep Learning

Abstract

Skin lesion classification plays a critical role in the early detection and accurate diagnosis of dermatological diseases, particularly melanoma, one of the most life-threatening skin cancers. Though widely used, traditional CNN-based models often face challenges handling complex cases, leading to misclassification and reduced diagnostic reliability. To overcome these limitations, this research integrates MobileNetV2 and Inceptionv3, two highly efficient and lightweight deep learning architectures trained on the HAM10000 dataset—a benchmark dataset containing diverse and well-labelled skin lesion images. By leveraging these advanced models, the proposed framework enhances classification precision, ensuring more accurate and reliable identification of seven skin lesions. The model can recognise complex patterns in lesion images by optimising feature extraction during training, improving diagnostic results. Regarding classification accuracy, comparative tests on the HAM10000 dataset show that our method performs noticeably better than conventional CNN-based models, making it a reliable and scalable AI-driven option for automated dermatological analysis. The findings demonstrate how deep learning can transform skin disease identification by providing a precise, effective, and interpretable framework to help medical practitioners make quicker and better clinical judgments.