Answers are grounded in this paper and cite their sources.
Evaluates skin lesion classification accuracy using AI and new ensemble weighting method, indicating improved diagnostics.
Skin lesions whether minor or life threatening cancerous can be detected early, thus treated effectively. The process of development of lesions into serious ailments, such as cancer, may be defected by delayed diagnosis. To fulfill this urgent requirement, we introduce a new AI-based solution that is based on Transfer Learning, and has a new architecture: Stacked RegNet Fusion (SRNF) with an Attention-Triad mechanism. This set of triad integrates soft attention, channel attention, and squeeze-excitation attention to better identify affected regions. Our SRNF system combines both versions of RegNetX and RegNetY models using the concatenation method, which forms a powerful hybrid system. Although Ensemble Learning has found extensive use in both Machine Learning and Deep Learning, the overall allocation of weights in ensemble predictions is a problem. We propose to solve this by the Inverse Kullback–Leibler Divergence Weighting (iKLDW) method which is then improved to Incremental Inverse Kullback–Leibler Divergence Weighting (I-iKLDW) to provide better ensemble with multiple layers. Experiments on the HAM10000 dataset prove that the proposed framework is effective and attains an accuracy of 94.08%, which is better than the existing state-of-the-art methodologies. Gradient-weighted Class Activation Mapping (Grad-CAM) is used to enhance interpretability such that the most discriminative regions are visualized to thereby improve model transparency. The proposed methodology also addresses the problem of accessibility, cost, and time, in addition to enhancing diagnostic accuracy, which is why it is an efficient approach to skin lesion early detection.
No takes yet. Share an insight, caveat, or question.
Efat et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: