Skip to main content
. Author manuscript; available in PMC: 2022 Feb 23.
Published in final edited form as: IEEE Access. 2021 Dec 6;9:163526–163541. doi: 10.1109/ACCESS.2021.3132958

FIGURE 4.

FIGURE 4.

(a) shows different labeling methods, including our soft label method, for an pT1a skin biopsy image with three tissue slices and one representative region of interest (red box) that helped expert pathologists in diagnosing the image. (b) compares the performance of different labeling methods. Our soft labeling method is simple and effective; it reduces the ambiguity that arises during training because of multiple tissue slices in a WSI that do not have a ROI and helps improve the performance. In (b), we do not report sensitivity and specificity, because their values are the same as accuracy.