Fig. 7
From: Multimodal dual-stage feature refinement for robust skin lesion classification
Illustration of (a) correct and incorrect classifications with the probability values by the proposed DualRefNet and the best Transformer model on PAD-UFES2043 and ISIC-201944,45,46, and (b) a comparative analysis of both datasets in terms of classification accuracy (ACC), balanced accuracy (BACC), and area under the ROC curve (AUC). For PAD-UFES20, the top row shows predicted labels correctly classified by DualRefNet’s best model, while the bottom row displays the classes predicted for the same samples misclassified by the Convolutional Vision Transformer (CVT). For ISIC-2019, the top row presents predicted classes correctly classified by DualRefNet, whereas the bottom row highlights the predicted values of the same images misclassified by the Swin Transformer.
