Table 4 Comparison of tooth and bone level mask prediction results across different models. Evaluation metrics include average precision at the IoU threshold of 0.5 (AP50) and AP across IoU thresholds from 0.5 to 0.95. The p-values denote statistically significant differences relative to YOLOv8x-seg on the test dataset.
From: AI-assisted radiographic analysis in detecting alveolar bone-loss severity and patterns
Task | Model name | Data set type | AP50 | AP50:95 | p-value (vs YOLOv8x-seg) |
|---|---|---|---|---|---|
Tooth Mask | YOLOv8x-seg | Train | 0.995 | 0.962 | - |
Validation | 0.984 | 0.895 | - | ||
Test | 0.978 | 0.900 | - | ||
Mask RCNN | Train | 0.916 | 0.566 | - | |
Validation | 0.903 | 0.532 | - | ||
Test | 0.885 | 0.548 | \(3.45 \times 10^{-4}\) | ||
Bone Level Mask | YOLOv8x-seg | Train | 0.697 | 0.202 | - |
Validation | 0.555 | 0.144 | - | ||
Test | 0.525 | 0.135 | - | ||
Mask RCNN | Train | 0.033 | 0.009 | - | |
Validation | 0.011 | 0.004 | - | ||
Test | 0.005 | 0.001 | \(1.11 \times 10^{-2}\) |
