Table 3.

Overview of the Model Performance of the Convolutional Neural Network in Relation to the Main Diagnostic Classes from the Independent Test Set (n = 479).

	True Positives		True Negatives		False Positives		False Negatives		Diagnostic Performance
Detection of Cavitation	n	%	n	%	n	%	n	%	ACC	SE	SP	NPV	PPV	AUC
Results from caries-free teeth or surfaces (n = 180 test images)
25% of the images	156	86.7	0	0.0	0	0.0	24	13.3	86.7	86.7	UC	0.0	100.0	UC
50% of the images	148	82.2	0	0.0	0	0.0	32	17.8	82.2	82.2	UC	0.0	100.0	UC
75% of the images	159	88.3	0	0.0	0	0.0	21	11.7	88.3	88.3	UC	0.0	100.0	UC
100% of the images	163	90.6	0	0.0	0	0.0	17	9.4	90.6	90.6	UC	0.0	100.0	UC
Results from noncavitated caries lesions (n = 216 test images)
25% of the images	170	78.7	0	0.0	0	0.0	46	21.3	78.7	78.7	UC	0.0	100.0	UC
50% of the images	187	86.6	0	0.0	0	0.0	29	13.4	86.6	86.6	UC	0.0	100.0	UC
75% of the images	183	84.7	0	0.0	0	0.0	33	15.3	84.7	84.7	UC	0.0	100.0	UC
100% of the images	184	85.2	0	0.0	0	0.0	32	14.8	85.2	85.2	UC	0.0	100.0	UC
Results from cavitated caries lesions (n = 83 test images)
25% of the images	53	63.9	0	0.0	0	0.0	30	36.1	63.9	63.9	UC	0.0	100.0	UC
50% of the images	61	73.5	0	0.0	0	0.0	22	26.5	73.5	73.5	UC	0.0	100.0	UC
75% of the images	61	73.5	0	0.0	0	0.0	22	26.5	73.5	73.5	UC	0.0	100.0	UC
100% of the images	66	79.5	0	0.0	0	0.0	17	20.5	79.5	79.5	UC	0.0	100.0	UC

The calculations included all types of teeth or surfaces, which were classified into each diagnostic category by the independent expert evaluation. As the reference standard served as selection criteria, true-negative and false-positive rates appear as zero values and, in consequence, SP and AUC became uncalculable.

ACC, accuracy; AUC, area under the receiver operating characteristic curve; SE, sensitivity; SP, specificity; NPV, negative predictive value; PPV, positive predictive value; UC, uncalculable.