|
Alsufyani et al. (2016)
|
10 CBCT |
Semi-automatic Segura tool |
Manual segmentation |
ICC (Volume): 0.96; ICC (Surface area): 0.97 |
Strong agreement with manual segmentation; time-efficient |
No disease- specific outcome |
|
Zhang et al. (2019)
|
10 CBCT |
Airway Segmentor (AS) Mimics 19.0 & INVIVO 5 |
Phantom gold standard |
ICC: 0.899–0.966; Random error: 3.9 to 4.5% |
Volumetric estimates close to gold standard |
Small pilot study |
|
Leonardi et al. (2021)
|
40 CBCT |
CNN-based segmentation |
Manual segmentation |
ICC: 0.921; DSC difference: 3.3–5.8% |
Comparable performance to experienced reader |
No clinical outcome validation |
|
Liu et al. (2021)
|
170 CT/CBCT |
Multi-stage CNN (SkullEngine) |
Manual annotation |
DSC > 0.90; Landmark error below clinical threshold |
Reduced processing time from 12 hrs to <3 min |
Not airway- specific focus |
|
Shujaat et al. (2021)
|
103 CT/CBCT |
3D U-Net |
Manual segmentation |
DSC: 0.97; IoU: 0.93; Precision: 0.97; Recall: 0.96 |
Strong agreement across airway regions |
No diagnostic utility assessed |
|
Sin et al. (2021)
|
306 CBCT |
Deep-learning CNN |
Manual (ITK-SNAP) |
DSC: 0.919; IoU:0993 Accuracy: 0.961 |
High segmentation accuracy |
OSA status not reported |
|
Cho et al. (2022)
|
216 CBCT |
3D U-Net |
Expert manual segmentation |
DSC: 0.928 Precision: 0.925; Recall: 0.921 |
Reduced precision in hypopharyngeal region |
No OSA validation |
|
De Bataille et al. (2022)
|
60 CBCT |
Semi-automated 3D volumetric |
Manual segmentation |
ICC > 0.7 |
Strong correlation with manual measures |
Restricted population |
|
Orhan et al. (2022)
|
200 adults (PSG-confirmed OSA vs controls) |
CNN-based segmentation (Diagnocat) |
Manual radiologist segmentation |
ICC: 0.954–0.972 |
Strong volumetric agreement |
No diagnostic outcome tested |
|
Balashova et al. (2023)
|
40 pediatric |
Diagnocat AI segmentation |
Visual exam + CBCT |
Significant airway differences between groups |
Effective 3D visualization |
Not benchmarked to PSG |
|
Chen et al. (2023)
|
30 CBCT |
Semi-automatic segmentation (Amira) |
Phantom gold standard |
ICC: 0.948–0.997; Overestimation: 1.1–3.1% |
Reliable nasal cavity volume measurement |
Semi-automatic, limited automation |
