HYBRID DEEP LEARNING MODEL OF SKIN LESION DIAGNOSIS BASED ON RESNET 50 AND CAPSULE NETWORK

A common and potentially fatal condition, skin cancer requires early and preciseidentification in order to be effectively treated. This paper introduces a hybrid deep learning modelbased Resnet 50 and Capsule Network for automated skin lesion classification using dermoscopicimages. The model follows multiple stages, including preprocessing, segmentation, featureextraction, and classification. Preprocessing involves resizing, data augmentation, andnormalization to enhance model performance. The active contour snake model is applied forsegmentation to isolate lesion areas, improving feature extraction accuracy. Segmented images arethen processed using ResNet50 for deep feature extraction, capturing key Lesion characteristicsinclude convexity, circularity, irregularity, textural patterns, and color aspects. A ConvolutionBlock Attention Module (CBAM) is integrated to refine these features, enhancing lesion patternfocus. The final classification module categorizes lesions into eight types, including melanoma andbasal cell carcinoma. This framework enhances diagnostic accuracy, supporting dermatologists inearly skin cancer detection and reducing subjectivity in manual diagnosis through advanced deeplearning techniques.