Non-Cochrane · DTA · PM42433437
Artificial intelligence for skeletal classification in orthodontics: A systematic review and meta-analysis of lateral cephalometric studies.
Dashti Mahmood, Khosraviani Farshad, Esmaeili Sara, et al. · Journal of Taibah University Medical Sciences · 2026 · PubMed · doi:10.1016/j.jtumed.2026.06.003
Findings
Population Patients undergoing orthodontic assessment with lateral cephalometric radiographs
15 findings
| AI/deep-learning model classifying sagittal skeletal Class I/II/III from lateral cephalograms | Reference skeletal class from conventional cephalometric analysis | AUC Lateral cephalometric radiographs · All classes (multiclass) | AUC 0.94 [0.90, 0.98] GRADE not reported 3 studies · 2743 participants |
| AI/deep-learning model classifying sagittal skeletal Class I/II/III from lateral cephalograms | Reference skeletal class from conventional cephalometric analysis | Sensitivity Lateral cephalometric radiographs · Skeletal Class I | Sensitivity (%) 85.50 [80.82, 90.17] GRADE not reported 2 studies |
| AI, machine-learning or deep-learning models classifying sagittal skeletal Class I, II, III from lateral cephalograms | Reference skeletal class from standard cephalometric diagnostic criteria | AUC of AI skeletal classification Skeletal Class II, as printed in Figure 7 | AUC 0.97 [0.96, 0.98] GRADE not reported 2 studies |
| AI/deep-learning model classifying sagittal skeletal Class I/II/III from lateral cephalograms | Reference skeletal class from conventional cephalometric analysis | AUC Lateral cephalometric radiographs · Skeletal Class I (Figure 7 subgroup plot) | AUC 0.94 [0.91, 0.96] GRADE not reported 2 studies |
| AI/deep-learning model classifying sagittal skeletal Class I/II/III from lateral cephalograms | Reference skeletal class from conventional cephalometric analysis | Sensitivity Lateral cephalometric radiographs · Skeletal Class III | Sensitivity (%) 91.20 [88.84, 93.57] GRADE not reported 2 studies |
| AI, machine-learning or deep-learning models classifying sagittal skeletal Class I, II, III from lateral cephalograms | Reference skeletal class from standard cephalometric diagnostic criteria | AUC of AI skeletal classification Skeletal Class III | AUC 0.99 [0.98, 1.00] GRADE not reported 2 studies |
| AI/deep-learning model classifying sagittal skeletal Class I/II/III from lateral cephalograms | Reference skeletal class from conventional cephalometric analysis | Sensitivity Lateral cephalometric radiographs · All classes combined (overall diamond of Figure 3 subgroup plot) | Sensitivity (%) 88.86 [87.11, 90.62] GRADE not reported 2 studies |
| AI, machine-learning or deep-learning models classifying sagittal skeletal Class I, II, III from lateral cephalograms | Reference skeletal class from standard cephalometric diagnostic criteria | AUC of AI skeletal classification Overall (model-level), as printed in Figure 6 | AUC 0.95 [0.91, 0.98] GRADE not reported 3 studies · 2743 participants |
| AI/deep-learning model classifying sagittal skeletal Class I/II/III from lateral cephalograms | Reference skeletal class from conventional cephalometric analysis | Sensitivity Lateral cephalometric radiographs · All classes (multiclass) | Sensitivity (%) 87.02 [85.64, 88.41] GRADE not reported 3 studies · 2743 participants |
| AI/deep-learning model classifying sagittal skeletal Class I/II/III from lateral cephalograms | Reference skeletal class from conventional cephalometric analysis | AUC Lateral cephalometric radiographs · All classes combined (overall diamond of Figure 7 subgroup plot) | AUC 0.97 [0.95, 0.98] GRADE not reported 2 studies |
| AI/deep-learning model classifying sagittal skeletal Class I/II/III from lateral cephalograms | Reference skeletal class from conventional cephalometric analysis | Sensitivity Lateral cephalometric radiographs · Skeletal Class II | Sensitivity (%) 89.13 [86.52, 91.74] GRADE not reported 2 studies |
| AI/deep-learning model classifying sagittal skeletal Class I/II/III from lateral cephalograms | Reference skeletal class from conventional cephalometric analysis | Specificity Lateral cephalometric radiographs · All classes (multiclass) | Specificity (%) 91.08 [87.30, 94.87] GRADE not reported 2 studies |
| AI, machine-learning or deep-learning models classifying sagittal skeletal Class I, II, III from lateral cephalograms | Reference skeletal class from standard cephalometric diagnostic criteria | AUC of AI skeletal classification Skeletal Class I, as printed in Figure 7 | AUC 0.94 [0.91, 0.96] GRADE not reported 2 studies |
| AI/deep-learning model classifying sagittal skeletal Class I/II/III from lateral cephalograms | Reference skeletal class from conventional cephalometric analysis | Accuracy Lateral cephalometric radiographs · All classes (multiclass) | Accuracy (%) 85.25 [79.85, 90.66] GRADE not reported 3 studies · 2743 participants |
| AI/deep-learning model classifying sagittal skeletal Class I/II/III from lateral cephalograms | Reference skeletal class from conventional cephalometric analysis | AUC Lateral cephalometric radiographs · Skeletal Class II | AUC 0.97 [0.95, 1.00] GRADE not reported 2 studies |
Included primary studies 4
Open in Studies| Study | Year | Also included by | Outcomes it informs here | Trial data |
|---|---|---|---|---|
| Jeong 2021 Deep-Learning-Based Detection of Cranio-Spinal Differences between Skeletal Classification Using Cephalometric Radiography. PMID:33806132 | 2021 | this review only | not stated | not extracted yet |
| Li 2022 Automatic Classification for Sagittal Craniofacial Patterns Based on Different Convolutional Neural Networks. PMID:35741169 | 2022 | this review only | not stated | not extracted yet |
| Nan 2023 Automated Sagittal Skeletal Classification of Children Based on Deep Learning. PMID:37238203 | 2023 | this review only | not stated | not extracted yet |
| Yu 2020 Automated Skeletal Classification with Lateral Cephalometry Based on Artificial Intelligence. PMID:31977286 | 2020 | this review only | not stated | not extracted yet |
Review question
| Element | Planned (methods) | Found (results) |
|---|---|---|
| Population | Human participants undergoing orthodontic assessment with lateral cephalometric radiographs | Orthodontic patients with lateral cephalograms (n=2,743 effective test sample in 3 pooled studies) |
| Intervention | AI, machine-learning or deep-learning model classifying sagittal skeletal Class I, II and/or III using lateral cephalogram as primary input | CNN models: DenseNet-121/161, ResNet-101/152, VGG16, GoogleNet, ConvNeXt-T, Swin-T |
| Comparison | Reference skeletal class by standard cephalometric diagnostic criteria | Reference skeletal class from conventional cephalometric analysis |
| Outcomes | Accuracy, sensitivity, specificity, precision, F1-score, AUC | Pooled sensitivity, specificity, accuracy, AUC (exploratory) |
| Study designs | Original peer-reviewed research articles reporting diagnostic performance | 4 diagnostic accuracy studies included (3 pooled; Jeong 2021 not pooled) |
| Timing | Not specified | Not specified |