Non-Cochrane · DTA · PM41764765
Efficacy of Automatic 3D Segmentation of the Upper Airway in CBCT or CT Scans via Artificial Intelligence Versus Manual Segmentation by Human Experts: A Systematic Review and Meta-Analysis.
Sobouti Farhad, Aryana Mehdi, Mohammad-Rahimi Hossein, et al. · Clinical and experimental dental research · 2026 · PubMed · doi:10.1002/cre2.70314
Findings
21 findings
| Population | ||||
|---|---|---|---|---|
| Studies assessing the upper airway on CBCT or CT scans | Automatic 3D segmentation by artificial intelligence (deep learning or machine learning) | Manual segmentation by human experts | Dice similarity score CBCT or CT upper airway segmentation · Leave-one-out: omitting Jin (2023) | Weighted mean 0.94 [0.91, 0.97] GRADE not reported 3 studies |
| Studies assessing the upper airway (nasal and pharyngeal) on CBCT or CT scans | AI-based (deep learning or machine learning) 3D segmentation | Manual segmentation by human experts | Precision of AI segmentation Upper airway CBCT or CT scans · Sensitivity: leave-one-out omitting Süküt (2024) | Weighted mean 0.95 [0.92, 0.97] GRADE not reported 3 studies |
| Studies assessing the upper airway (nasal and pharyngeal) on CBCT or CT scans | AI-based (deep learning or machine learning) 3D segmentation | Manual segmentation by human experts | Recall (sensitivity) of AI segmentation Upper airway CBCT or CT scans · Sensitivity: leave-one-out omitting Cho (2022) | Weighted mean 0.94 [0.88, 1.0] GRADE not reported 3 studies |
| Studies assessing the upper airway (nasal and pharyngeal) on CBCT or CT scans | AI-based (deep learning or machine learning) 3D segmentation | Manual segmentation by human experts | Precision of AI segmentation Upper airway CBCT or CT scans · Sensitivity: leave-one-out omitting Jin (2023) | Weighted mean 0.96 [0.93, 0.99] GRADE not reported 3 studies |
| Studies assessing the upper airway on CBCT or CT scans | Automatic 3D segmentation by artificial intelligence (deep learning or machine learning) | Manual segmentation by human experts | Intersection over union (IoU) CBCT or CT upper airway segmentation · Leave-one-out: omitting Shujaat (2021) | Weighted mean 0.93 [0.92, 0.94] GRADE not reported 1 study |
| Studies assessing the upper airway on CBCT or CT scans | Automatic 3D segmentation by artificial intelligence (deep learning or machine learning) | Manual segmentation by human experts | Recall (sensitivity) CBCT or CT upper airway segmentation · Leave-one-out: omitting Süküt (2024) | Weighted mean 0.95 [0.92, 0.99] GRADE not reported 3 studies |
| Studies assessing the upper airway (nasal and pharyngeal) on CBCT or CT scans | AI-based (deep learning or machine learning) 3D segmentation | Manual segmentation by human experts | Recall (sensitivity) of AI segmentation Upper airway CBCT or CT scans · Sensitivity: leave-one-out omitting Jin (2023) | Weighted mean 0.92 [0.87, 0.97] GRADE not reported 3 studies |
| Studies assessing the upper airway (nasal and pharyngeal) on CBCT or CT scans | AI-based (deep learning or machine learning) 3D segmentation | Manual segmentation by human experts | Precision of AI segmentation Upper airway CBCT or CT scans | Weighted mean 0.95 [0.93, 0.97] GRADE not reported 4 studies |
| Studies assessing the upper airway on CBCT or CT scans | Automatic 3D segmentation by artificial intelligence (deep learning or machine learning) | Manual segmentation by human experts | Dice similarity score CBCT or CT upper airway segmentation · Leave-one-out: omitting Cho (2022) | Weighted mean 0.95 [0.92, 0.98] GRADE not reported 3 studies |
| Studies assessing the upper airway on CBCT or CT scans | Automatic 3D segmentation by artificial intelligence (deep learning or machine learning) | Manual segmentation by human experts | Recall (sensitivity) CBCT or CT upper airway segmentation · Leave-one-out: omitting Shujaat (2021) | Weighted mean 0.93 [0.87, 0.99] GRADE not reported 3 studies |
| Studies assessing the upper airway (nasal and pharyngeal) on CBCT or CT scans | AI-based (deep learning or machine learning) 3D segmentation | Manual segmentation by human experts | Recall (sensitivity) of AI segmentation Upper airway CBCT or CT scans | Weighted mean 0.94 [0.89, 0.98] GRADE not reported 4 studies |
| Studies assessing the upper airway on CBCT or CT scans | Automatic 3D segmentation by artificial intelligence (deep learning or machine learning) | Manual segmentation by human experts | Precision CBCT or CT upper airway segmentation · Leave-one-out: omitting Cho (2022) | Weighted mean 0.96 [0.94, 0.98] GRADE not reported 3 studies |
| Studies assessing the upper airway on CBCT or CT scans | Automatic 3D segmentation by artificial intelligence (deep learning or machine learning) | Manual segmentation by human experts | Dice similarity score CBCT or CT upper airway segmentation · Leave-one-out: omitting Süküt (2024) | Weighted mean 0.95 [0.93, 0.98] GRADE not reported 3 studies |
| Studies assessing the upper airway on CBCT or CT scans | Automatic 3D segmentation by artificial intelligence (deep learning or machine learning) | Manual segmentation by human experts | Total volume difference CBCT or CT upper airway segmentation · Leave-one-out: omitting Alsufyani (2016) | Weighted mean 1.93 [1.61, 2.25] GRADE not reported 1 study |
| Studies assessing the upper airway on CBCT or CT scans | Automatic 3D segmentation by artificial intelligence (deep learning or machine learning) | Manual segmentation by human experts | Dice similarity score CBCT or CT upper airway segmentation | Weighted mean 0.94 [0.92, 0.97] GRADE not reported 4 studies |
| Studies assessing the upper airway on CBCT or CT scans | Automatic 3D segmentation by artificial intelligence (deep learning or machine learning) | Manual segmentation by human experts | Intersection over union (IoU) CBCT or CT upper airway segmentation · Leave-one-out: omitting Jin (2023) | Weighted mean 0.93 [0.92, 0.94] GRADE not reported 1 study |
| Studies assessing the upper airway on CBCT or CT scans | Automatic 3D segmentation by artificial intelligence (deep learning or machine learning) | Manual segmentation by human experts | Dice similarity score CBCT or CT upper airway segmentation · Leave-one-out: omitting Shujaat (2021) | Weighted mean 0.94 [0.91, 0.96] GRADE not reported 3 studies |
| Studies assessing the upper airway (nasal and pharyngeal) on CBCT or CT scans | AI-based (deep learning or machine learning) 3D segmentation | Manual segmentation by human experts | Total volume difference of AI segmentation Upper airway CBCT or CT scans | MD 1.93 [1.63, 2.23] GRADE not reported 2 studies |
| Studies assessing the upper airway (nasal and pharyngeal) on CBCT or CT scans | AI-based (deep learning or machine learning) 3D segmentation | Manual segmentation by human experts | Intersection over union (IoU) of AI segmentation Upper airway CBCT or CT scans | Weighted mean 0.93 [0.92, 0.94] GRADE not reported 2 studies |
| Studies assessing the upper airway on CBCT or CT scans | Automatic 3D segmentation by artificial intelligence (deep learning or machine learning) | Manual segmentation by human experts | Precision CBCT or CT upper airway segmentation · Leave-one-out: omitting Shujaat (2021) | Weighted mean 0.95 [0.92, 0.97] GRADE not reported 3 studies |
| Studies assessing the upper airway on CBCT or CT scans | Automatic 3D segmentation by artificial intelligence (deep learning or machine learning) | Manual segmentation by human experts | Total volume difference CBCT or CT upper airway segmentation · Leave-one-out: omitting Leonardi (2021) | Weighted mean 1.9 [1.03, 2.77] GRADE not reported 1 study |
Included primary studies 10
Open in Studies| Study | Year | Also included by | Outcomes it informs here | Trial data |
|---|---|---|---|---|
| Alsufyani 2016 New algorithm for semiautomatic segmentation of nasal cavity and pharyngeal airway in comparison with manual segmentation using cone-beam… PMID:27692428 | 2016 | this review only | Total volume difference | not extracted yet |
| Cho 2022 Accuracy of convolutional neural networks-based automatic segmentation of pharyngeal airway sections according to craniofacial skeletal… PMID:35654686 | 2022 | this review only | Precision; Dice similarity score; Recall (sensitivity) of AI segmentation | not extracted yet |
| Chu 2023 Deep Learning Models for Automatic Upper Airway Segmentation and Minimum Cross-Sectional Area Localisation in Two-Dimensional Images. PMID:37627800 | 2023 | this review only | not stated | not extracted yet |
| Jin 2023 Automatic three-dimensional nasal and pharyngeal airway subregions identification via Vision Transformer. PMID:37343616 | 2023 | this review only | Precision; Dice similarity score; Recall (sensitivity) of AI segmentation; Intersection over union (IoU) of AI segmentation | not extracted yet |
| Leonardi 2021 Fully automatic segmentation of sinonasal cavity and pharyngeal airway based on convolutional neural networks. PMID:34059213 | 2021 | this review only | Total volume difference | not extracted yet |
| Orhan 2022 AI-based automatic segmentation of craniomaxillofacial anatomy from CBCT scans for automatic detection of pharyngeal airway evaluations in… PMID:35831451 | 2022 | this review only | not stated | not extracted yet |
| Park 2021 Deep Learning Based Airway Segmentation Using Key Point Prediction DOI:10.3390/app11083501 | 2021 | this review only | not stated | not extracted yet |
| Shujaat 2021 Automatic segmentation of the pharyngeal airway space with convolutional neural network. PMID:34077802 | 2021 | this review only | Precision; Dice similarity score; Recall (sensitivity) of AI segmentation; Intersection over union (IoU) of AI segmentation | not extracted yet |
| Sin 2021 A deep learning algorithm proposal to automatic pharyngeal airway detection and segmentation on CBCT images. PMID:33619828 | 2021 | this review only | not stated | not extracted yet |
| Süküt 2025 Accuracy of deep learning-based upper airway segmentation. PMID:39244033 | 2025 | this review only | Precision; Dice similarity score; Recall (sensitivity) of AI segmentation | not extracted yet |
Review question
| Element | Planned (methods) | Found (results) |
|---|---|---|
| Population | Studies assessing the upper airway (nasal and pharyngeal airways) on CBCT or CT scans | Upper airway CBCT or CT scans (n=11 studies; 6 in meta-analysis) |
| Intervention | Any AI tool (deep learning, machine learning) for image analysis | AI segmentation models (U-Net, 3D U-Net CNN, Vision Transformer, MONAI Label and others) |
| Comparison | Manual analysis of CBCT or CT images by humans | Manual or semi-automatic segmentation by human experts |
| Outcomes | Accuracy, precision, dice similarity score, total volume difference, intersection over union, recall, or other segmentation parameters | Precision (4 studies), dice score (4), total volume difference (2), recall (4), IoU (2) |
| Study designs | Original studies; reviews, case reports, editorials, guidelines, letters and conference abstracts excluded; English language | Observational segmentation studies; 11 qualitative, 6 meta-analyzed |
| Timing | Not specified | Not specified |
AMSTAR 2 items
Two independent extractors rated each item. Their ratings are shown side by side because they have not been reconciled yet. The overall rating follows Shea et al. 2017: one critical flaw gives Low, more than one gives Critically Low.
| Item | Extractor A | Extractor B | Agree |
|---|---|---|---|
| 1Research question includes PICO components | NR | Yes | no |
| 2Protocol established before the reviewcritical | NR | No | no |
| 3Choice of study designs explained | NR | No | no |
| 4Comprehensive literature searchcritical | NR | Partial Yes | no |
| 5Study selection in duplicate | NR | No | no |
| 6Data extraction in duplicate | NR | Yes | no |
| 7List of excluded studies with reasonscritical | NR | No | no |
| 8Included studies described in detail | NR | Partial Yes | no |
| 9Risk of bias assessed satisfactorilycritical | NR | NRSI: Yes · RCT: N/A | no |
| 10Funding of included studies reported | NR | No | no |
| 11Appropriate meta-analysis methodscritical | NR | NRSI: No · RCT: N/A | no |
| 12Impact of risk of bias on the pooled results | NR | No | no |
| 13Risk of bias accounted for in interpretationcritical | NR | No | no |
| 14Heterogeneity explained | NR | Yes | no |
| 15Publication bias investigatedcritical | NR | Yes | no |
| 16Conflicts of interest reported | NR | Yes | no |
| Overall | NR | Critically Low | no |