Table 7.

Comparison between the convolutional neural network (CNN) predictive score and the clinicians’ feature for the fibrous crescent (F-Cre)

		No. of clinicians out of 5 who scored positive
		0	1	2	3	4	5
CNN’s predictive score over 4 folds	0.75–1.00	9a	4	2	7	6	0b
	0.50–0.75	11a	5	1	2	1	0b
	0.25–0.50	20c	2	1	3	0	1d
	0.00–0.25	17c	4	1	0	1	0d

This confusion matrix shows the comparison between the CNN predictive score and the clinicians’ feature for the F-Cre. Rows show the CNN’s predictive score for an image, which was calculated as an average of the CNN’s softmax probability over the 4 folds. Columns show the number of clinicians out of 5 who scored it as positive. Each number shows the number of images.

^aImages in which the CNN predicted it to be positive, but no clinician scored it as positive (CNN’s completely false-positive result).

^bImages in which the CNN predicted it to be positive and all of the clinicians scored it as positive (the CNN’s completely true-positive result).

^cImages in which the CNN predicted it to be negative and no clinician scored it as positive (the CNN’s completely true-negative result).

^dImages in which the CNN predicted it to be negative, but all clinicians scored it as positive (the CNN’s completely false-negative result).