Fig. 2: Examples of co-localization of annotation data and predictive tiles. | Modern Pathology

Fig. 2: Examples of co-localization of annotation data and predictive tiles.

From: Deep-learning based classification distinguishes sarcomatoid malignant mesotheliomas from benign spindle cell mesothelial proliferations

Fig. 2

(A, C) areas with tiles highly predictive of malignancy and (B, D) areas annotated as sarcomatoid malignancy (circled in red) on H&E stain. Representative benign and malignant areas from the H&E images are shown at 20X. For (A, C), tile scores closer to 1 (yellow) were most predictive of malignancy. Areas scoring 0.90 or less (low likelihood of malignancy) are shown in purple.

Back to article page