Abstract
The integration of advanced computer vision techniques (CV) and biomedical image processing has significantly advanced the early detection of skin cancers, particularly through dermoscopic imaging. Dermoscopy technique, also known as a non-invasive diagnostic method, offers improved vision of skin abnormalities, facilitating precise clinical evaluation. Convolutional neural networks have accomplished outstanding results in skin cancer classification, and recent advancements in deep transfer learning (DTL) have further enhanced automated disease detection. This work presents an automated clinical decision support system (CDSS) and a classification framework using CSO (cat swarm optimization) and deep transfer learning (ACDSS-CSODTL). This approach utilizes bilateral filtering (BF) to minimize noise, employs a U-Net for accurate lesion segmentation, leverages Efficient-Net for strong feature extraction, and integrates a DTL-based classifier that is fine-tuned through the CSO technique for hyperparameter optimization. Experimental results demonstrate that ACDSS-CSODTL outperforms existing approaches, achieving an accuracy of 99.79%, precision of 98.70%, recall of 98.66%, and F 1-score of 98.66%. These findings establish the proposed framework as an effective and reliable computer-aided diagnostic tool, enabling improved skin cancer detection and supporting early clinical intervention.

This publication has 41 references indexed in Scilit: