Innovation science and technologiy Volume 2 Issue 6 (2026)
COMPARATIVE ANALYSIS OF DEEP LEARNING ALGORITHMS FOR AUTOMATED SKIN DISEASE DIAGNOSIS FROM DERMOSCOPIC IMAGES
Nazirova, Elmira, Abdusalomova, Shokhista
Abstract
A comparative analysis of CNN, ResNet-50, and EfficientNet-B4 architectures for automated skindisease classification from dermoscopic images was conducted. The study was performed using the HAM10000dataset comprising 10,015 images across seven nosological classes. Model performance was evaluated usingaccuracy, sensitivity, specificity, F1-score, AUC-ROC, inference time, and memory footprint. EfficientNet-B4demonstrated the best performance, achieving 93.71 % accuracy and an AUC-ROC of 0.978. The obtainedresults indicate the potential of this architecture for integration into clinical decision-support systems
deep learning; skin disease diagnosis; convolutional neural network (CNN); ResNet; EfficientNet; dermoscopy; HAM10000; medical image analysis; transfer learning; AUC-ROC
Metadata source: the journal's OAI-PMH archive · Sindex does not store the full text; it links to the source.