Table 2.

Comparison of the results of the sliding window method (SWM) and mask region-based convolutional neural network (Mask-RCNN) method

	Sliding window	Mask R-CNN
Number of detections	320	37
Percentage of incorrectly detected from backgrounds	0.3031 (97/320)	0.027 (1/37)
Average number of undetectable cell nuclear regions per cell nucleus	1.8611 (67/36)	0 (0/36)
Average number of regions detected for each cell nucleus	6.1944 (223/36)	1 (36/36)

Ten images were randomly selected from Classes II and III. Mask-RCNN showed lower background detection rate and lower average number of undetected regions per cell nucleus compared with SWM. The average number of detected regions for each cell nucleus in Mask-RCNN was 1. This indicates that the detected cell nuclei were not duplicated