Table 1.
Overview of the Model Performance of the Convolutional Neural Network When the Independent Test Set (n = 479 with 180 Healthy Tooth Surfaces, 216 Noncavitated Carious Lesions, and 83 Cavitations) Was Used for Overall Caries Detection.
| True Positives | True Negatives | False Positives | False Negatives | Diagnostic Performance | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Overall Caries Detection | n | % | n | % | n | % | n | % | ACC | SE | SP | NPV | PPV | AUC |
| Results from all the included teeth and surfaces (n = 479 test images) | ||||||||||||||
| 25% of the images | 156 | 32.6 | 258 | 53.8 | 24 | 5.0 | 41 | 8.6 | 86.4 | 79.2 | 91.5 | 86.7 | 86.3 | 0.924 |
| 50% of the images | 148 | 30.9 | 280 | 58.4 | 32 | 6.7 | 19 | 4.0 | 89.4 | 88.6 | 89.7 | 82.2 | 93.6 | 0.950 |
| 75% of the images | 159 | 33.2 | 276 | 57.6 | 21 | 4.4 | 23 | 4.8 | 90.8 | 87.4 | 92.9 | 88.3 | 92.3 | 0.955 |
| 100% of the images | 163 | 34.0 | 280 | 58.5 | 17 | 3.5 | 19 | 4.0 | 92.5 | 89.6 | 94.3 | 90.6 | 93.6 | 0.964 |
| Results from anterior surfaces—incisors/canines (n = 153 test images) | ||||||||||||||
| 25% of the images | 63 | 41.2 | 65 | 42.5 | 4 | 2.6 | 21 | 13.7 | 83.7 | 75.0 | 94.2 | 94.0 | 75.6 | 0.911 |
| 50% of the images | 59 | 38.6 | 79 | 51.6 | 8 | 5.2 | 7 | 4.6 | 90.2 | 89.4 | 90.8 | 88.1 | 91.9 | 0.953 |
| 75% of the images | 64 | 41.8 | 73 | 47.7 | 3 | 2.0 | 13 | 8.5 | 89.5 | 83.1 | 96.1 | 95.5 | 84.9 | 0.947 |
| 100% of the images | 64 | 41.8 | 80 | 52.3 | 3 | 2.0 | 6 | 3.9 | 94.1 | 91.4 | 96.4 | 95.5 | 93.0 | 0.965 |
| Results from posterior surfaces—molars/premolars (n = 326 test images) | ||||||||||||||
| 25% of the images | 93 | 28.6 | 193 | 59.2 | 20 | 6.1 | 20 | 6.1 | 87.7 | 82.3 | 90.6 | 82.3 | 90.6 | 0.932 |
| 50% of the images | 89 | 27.3 | 201 | 61.6 | 24 | 7.4 | 12 | 3.7 | 89.0 | 88.1 | 89.3 | 78.8 | 94.4 | 0.945 |
| 75% of the images | 95 | 29.1 | 203 | 62.3 | 18 | 5.5 | 10 | 3.1 | 91.4 | 90.5 | 91.9 | 84.1 | 95.3 | 0.961 |
| 100% of the images | 99 | 30.4 | 200 | 61.3 | 14 | 4.3 | 13 | 4.0 | 91.7 | 88.4 | 93.5 | 87.6 | 93.9 | 0.964 |
| Results from vestibular and oral surfaces—anterior/posterior teeth (n = 225 test images) | ||||||||||||||
| 25% of the images | 65 | 28.9 | 126 | 56.0 | 8 | 3.6 | 26 | 11.5 | 84.9 | 71.4 | 94.0 | 89.0 | 82.9 | 0.910 |
| 50% of the images | 62 | 27.6 | 143 | 63.5 | 11 | 4.9 | 9 | 4.0 | 91.1 | 87.3 | 92.9 | 84.9 | 94.1 | 0.954 |
| 75% of the images | 67 | 29.8 | 137 | 60.9 | 6 | 2.7 | 15 | 6.6 | 90.7 | 81.7 | 95.8 | 91.8 | 90.1 | 0.952 |
| 100% of the images | 67 | 29.8 | 142 | 63.1 | 6 | 2.7 | 10 | 4.4 | 92.9 | 87.0 | 95.9 | 91.8 | 93.4 | 0.964 |
| Results from occlusal surfaces—molars/premolars (n = 253 test images) | ||||||||||||||
| 25% of the images | 91 | 36.0 | 131 | 51.8 | 16 | 6.3 | 15 | 5.9 | 87.7 | 85.8 | 89.1 | 85.0 | 89.7 | 0.943 |
| 50% of the images | 86 | 34.0 | 136 | 53.7 | 21 | 8.3 | 10 | 4.0 | 87.7 | 89.6 | 86.6 | 80.4 | 93.2 | 0.949 |
| 75% of the images | 92 | 36.4 | 138 | 54.5 | 15 | 5.9 | 8 | 3.2 | 90.9 | 92.0 | 90.2 | 86.0 | 94.5 | 0.961 |
| 100% of the images | 96 | 37.9 | 137 | 54.2 | 11 | 4.3 | 9 | 3.6 | 92.1 | 91.4 | 92.6 | 89.7 | 93.8 | 0.968 |
The calculations were performed for different types of teeth, surfaces, and training steps, which resulted in different subsamples.
ACC, accuracy; AUC, area under the receiver operating characteristic curve; SE, sensitivity; SP, specificity; NPV, negative predictive value; PPV, positive predictive value.