Figure 4.

A visual example of HR-related lesions segmentation steps. Where figure (a) shows the shallow CNN architecture with the Random Forest (RF) classifier used for sematic-based segmentation, figure (b) represents the malignant-HR image, figure (c) shows the corresponding mask, figure (d) shows the design architecture used for semantic-based segmentation through the U-Net model to refine detection results and, finally, figure (e) indicates the image processing steps to get the final regions of HR-related lesions.