Table 3.

Comparison of diagnostic performance indicators at representative cutoffs generated from the respective confusion matrices of observed against predicted by the trained classifier algorithm pixel status in the test subset (n = 19,593).

Cutoff	Confusion matrices of pixel status				Diagnostic performance
0.050			Predicted		Accuracy	0.98173
			Negative	Positive	Sensitivity	0.99217
	Observed	Negative	14,798	323	Specificity	0.97864
		Positive	35	4,437	Difference	0.01353
0.135			Predicted		Accuracy	0.98836
			Negative	Positive	Sensitivity	0.98882
	Observed	Negative	14,943	178	Specificity	0.98823
		Positive	50	4,422	Difference	0.00059
0.500			Predicted		Accuracy	0.99433
			Negative	Positive	Sensitivity	0.98323
	Observed	Negative	15,085	36	Specificity	0.99767
		Positive	75	4,397	Difference	−0.01439
0.800			Predicted		Accuracy	0.99234
			Negative	Positive	Sensitivity	0.96847
	Observed	Negative	15,112	9	Specificity	0.99940
		Positive	141	4,331	Difference	−0.03093
0.995			Predicted		Accuracy	0.98816
			Negative	Positive	Sensitivity	0.94879
	Observed	Negative	15,118	3	Specificity	0.99980
		Positive	229	4,243	Difference	−0.05101