Skip to content
DraconiticResearch
Search Sources API
226,742 studies31 sourcesUpdated 42 min ago
Peer-reviewedCC BY 4.0Journal of Clinical Medicine

Accuracy of Artificial Intelligence-Based Models in Predicting Postoperative Soft Tissue Outcomes Following Orthognathic Surgery: A Systematic Review and Meta-Analysis

  1. 1Department of Oral and Maxillofacial Surgery, King Khaled University Hospital, King Saud University Medical City, College of Dentistry, King Saud University, Riyadh 11362, Saudi Arabia

Abstract

Background/Objectives: Predicting postoperative facial soft tissue change supports planning and patient counselling in orthognathic surgery. The geometric accuracy of artificial intelligence (AI)-based models has not previously been pooled. This review estimated it and compared the models with conventional methods. Methods: MEDLINE (PubMed), Scopus, IEEE Xplore, arXiv and the Europe PMC preprint index (which covers medRxiv) were searched to 23 October 2026, supplemented by citation tracking (PROSPERO CRD420261536010). Eligible studies compared AI predictions driven by the planned or achieved skeletal movement with the actual postoperative face. Three-dimensional (3D) surface, 3D landmark and two-dimensional (2D) landmark errors were synthesised separately. Cohort-level errors were summarised descriptively; random-effects syntheses, run only when at least three independent cohorts were available and with one evaluation per potentially overlapping cohort, were reported as exploratory because the within-cohort variances could be verified for one cohort only. Risk of bias was assessed with PROBAST + AI and certainty with GRADE. Results: Twelve studies (16 reports) were included. In six independent cohorts (152 patients), none driven by a confirmed preoperative plan, the cohort mean 3D surface errors ranged from 0.70 to 1.77 mm (median 1.06 mm) under four definitions of the surface distance; the exploratory random-effects estimate, a summary of these heterogeneous error measures and not a directly comparable, clinically validated or universally applicable measure of accuracy, was 1.10 mm (95% confidence interval [CI] 0.72–1.49; 95% prediction interval 0.04–2.17), with intervals that are approximate because the between-patient basis of the within-cohort variances could be verified for one cohort only. The only evaluation driven by a confirmed preoperative plan reported 1.55 mm. Regional errors were lowest at the nose (cohort means 0.55–0.77 mm; exploratory pooled estimate 0.68 mm) and highest at the chin (1.13–1.60 mm; pooled 1.37 mm, 95% CI 1.01–1.74), where about one patient in four exceeded 2 mm in the two studies reporting patient-level proportions. In the two independent same-patient comparisons, the AI model was more accurate than the finite element model by 0.18 mm (95% CI 0.09–0.27) and 0.29 mm (95% CI 0.20–0.37), differences of uncertain clinical importance that do not establish equivalence or non-inferiority; three further, non-independent comparisons gave −0.19, +0.09 and −0.08 mm. Nine of 12 studies were at high risk of bias, eight used the achieved rather than the planned movement, and one was externally validated. Certainty was very low. Conclusions: In the developers’ retrospective evaluations, mostly driven by the achieved skeletal movement, AI models predicted the postoperative facial surface with cohort mean errors of about 1 mm, largest at the chin; two small independent comparisons favoured AI over biomechanical simulation by 0.2 to 0.3 mm, a difference of uncertain clinical importance. These estimates describe how closely a model reproduces the face after a known skeletal movement, not the accuracy of a prediction made from a surgical plan. Externally validated, plan-driven studies are needed before clinical reliance.

Related

  1. Before Sepsis Strikes: Artificial Intelligence for Early Prediction of Postoperative Sepsis and Sepsis-Related Complications After Major Abdominal Surgery—A Systematic Review
    Wasim Iyad Ibrahim Alghoul, Ibrahim Alabid, Malak Moones Abedi +9
    Medical sciences (Basel, Switzerland)Peer-reviewedCC BYFull textGeneral medicineEurope PMC
  2. Artificial intelligence and machine learning for prediction of postoperative complications: A systematic review focused on anesthesiology, perioperative risk stratification, and clinical applicability.
    Hamza Hafiani, Hamza Kassimi, Salma Bekkour +2
    Saudi journal of anaesthesiaPeer-reviewedCC BY-NC-SAFull textCardiovascular medicineEurope PMCPMC
  3. Artificial intelligence-enhanced electrocardiography for predicting mortality and major adverse cardiovascular events: A systematic review and narrative synthesis
    Elhaitham Yasir Awad Musa, Amar Yasir Awad Musa
    PLOS OnePeer-reviewedCC BY 4.0Full textCardiovascular medicinePLOS
  4. Predicting acute radiation dermatitis in breast cancer patients using artificial intelligence: a systematic review.
    Hang Yu, Pan Gong, Xiangbin Zhang +2
    Quantitative imaging in medicine and surgeryPeer-reviewedCC BY-NC-NDFull textMedical imaging & radiologyEurope PMCPMC
  5. The role of artificial intelligence in predicting late-onset intrauterine growth restriction in singleton pregnancies: a systematic review and meta-analysis
    Miriam Turiel Miranda, Alba Etxeandia Atienza, José Nicolás Marín Santos +7
    UltrasonographyPeer-reviewedCC BY-NCFull textObstetrics, gynaecology & paediatricsDOAJ
BibTeX
@article{alotaibi2026accuracy,
  title = {{Accuracy of Artificial Intelligence-Based Models in Predicting Postoperative Soft Tissue Outcomes Following Orthognathic Surgery: A Systematic Review and Meta-Analysis}},
  author = {Fawaz Alotaibi and Bader Fatani},
  journal = {Journal of Clinical Medicine},
  year = {2026},
  month = nov,
  doi = {10.3390/jcm15228817},
  url = {https://www.mdpi.com/2077-0383/15/22/8817},
  note = {CC BY 4.0},
}
RIS
TY  - JOUR
TI  - Accuracy of Artificial Intelligence-Based Models in Predicting Postoperative Soft Tissue Outcomes Following Orthognathic Surgery: A Systematic Review and Meta-Analysis
AU  - Fawaz Alotaibi
AU  - Bader Fatani
PY  - 2026
DA  - 2026/11/11
JO  - Journal of Clinical Medicine
DO  - 10.3390/jcm15228817
UR  - https://www.mdpi.com/2077-0383/15/22/8817
AB  - Background/Objectives: Predicting postoperative facial soft tissue change supports planning and patient counselling in orthognathic surgery. The geometric accuracy of artificial intelligence (AI)-based models has not previously been pooled. This review estimated it and compared the models with conventional methods. Methods: MEDLINE (PubMed), Scopus, IEEE Xplore, arXiv and the Europe PMC preprint index (which covers medRxiv) were searched to 23 October 2026, supplemented by citation tracking (PROSPERO CRD420261536010). Eligible studies compared AI predictions driven by the planned or achieved skeletal movement with the actual postoperative face. Three-dimensional (3D) surface, 3D landmark and two-dimensional (2D) landmark errors were synthesised separately. Cohort-level errors were summarised descriptively; random-effects syntheses, run only when at least three independent cohorts were available and with one evaluation per potentially overlapping cohort, were reported as exploratory because the within-cohort variances could be verified for one cohort only. Risk of bias was assessed with PROBAST + AI and certainty with GRADE. Results: Twelve studies (16 reports) were included. In six independent cohorts (152 patients), none driven by a confirmed preoperative plan, the cohort mean 3D surface errors ranged from 0.70 to 1.77 mm (median 1.06 mm) under four definitions of the surface distance; the exploratory random-effects estimate, a summary of these heterogeneous error measures and not a directly comparable, clinically validated or universally applicable measure of accuracy, was 1.10 mm (95% confidence interval [CI] 0.72–1.49; 95% prediction interval 0.04–2.17), with intervals that are approximate because the between-patient basis of the within-cohort variances could be verified for one cohort only. The only evaluation driven by a confirmed preoperative plan reported 1.55 mm. Regional errors were lowest at the nose (cohort means 0.55–0.77 mm; exploratory pooled estimate 0.68 mm) and highest at the chin (1.13–1.60 mm; pooled 1.37 mm, 95% CI 1.01–1.74), where about one patient in four exceeded 2 mm in the two studies reporting patient-level proportions. In the two independent same-patient comparisons, the AI model was more accurate than the finite element model by 0.18 mm (95% CI 0.09–0.27) and 0.29 mm (95% CI 0.20–0.37), differences of uncertain clinical importance that do not establish equivalence or non-inferiority; three further, non-independent comparisons gave −0.19, +0.09 and −0.08 mm. Nine of 12 studies were at high risk of bias, eight used the achieved rather than the planned movement, and one was externally validated. Certainty was very low. Conclusions: In the developers’ retrospective evaluations, mostly driven by the achieved skeletal movement, AI models predicted the postoperative facial surface with cohort mean errors of about 1 mm, largest at the chin; two small independent comparisons favoured AI over biomechanical simulation by 0.2 to 0.3 mm, a difference of uncertain clinical importance. These estimates describe how closely a model reproduces the face after a known skeletal movement, not the accuracy of a prediction made from a surgical plan. Externally validated, plan-driven studies are needed before clinical reliance.
SN  - 2077-0383
ER  -