Dental Radiography and Imaging

State of the evidence

Deep learning models (CNNs, YOLO, SSD) demonstrate high accuracy in automating cephalometric landmarking across 2D and 3D imaging 1,2,3,10. Cervical Vertebral Maturation (CVM) is established as a viable biologic indicator for predicting pubertal growth spurts, often compared against the hand-wrist gold standard 4,5,9. While automated CVM classification is feasible using CNNs 6,7, clinical adoption of 3D landmarking is hindered by concerns over robustness and generalizability 3.

Key findings

• YOLOv3 and SSD are effective deep-learning architectures for identifying cephalometric landmarks in 2D images 1. 1
• CVM staging can effectively replace hand-wrist radiographs as a proxy for skeletal maturity and pubertal growth spurts 5,9. 5 9
• Fully automated CVM classification is achievable using custom CNNs with directional filters to highlight vertebral edges 6,7. 6 7
• Multi-stage CNNs can accurately identify landmarks on posteroanterior (PA) cephalometric images synthesized from CBCT scans 10. 10
• 3D cephalometric landmarking via deep learning is promising but lacks the robustness required for widespread clinical use 3. 3

Where the evidence disagrees

• CVM reliability varies across studies, with some systematic reviews noting a high proportion of low-score evidence regarding its accuracy 4.
• High model accuracy in research settings does not yet translate to clinical robustness or generalizability in 3D imaging 3.

Open opportunities

What the research supports that no idea in the catalog takes yet.

• 3D-to-2D Surgical Planning Bridge

  Evidence shows 3D landmarking is possible 3 and PA images can be synthesized from CBCT 10, but no idea bridges the gap between 3D planning and 2D radiographic verification.

• CVM Longitudinal Tracking Dashboard

  CVM is a proxy for growth spurts 5, but current ideas are single-point classifiers; a longitudinal tool would track the rate of change in vertebral morphology over time.

• Cross-Modality Landmark Validator

  Given the robustness concerns in 3D 3 and the accuracy of 2D 1, a tool that validates 3D landmarks against 2D projections would increase clinical trust.

References

1. 1.Automated identification of cephalometric landmarks: Part 1—Comparisons between the latest deep-learning methods YOLOV3 and SSDJihoon Park, Hyewon Hwang, Jun‐Ho Moon et al. · 2019 · 233 citations
2. 2.Deep learning for cephalometric landmark detection: systematic review and meta-analysisFalk Schwendicke, Akhilanand Chaurasia, Lubaina T. Arsiwala-Scheppach et al. · 2021 · 179 citations
3. 3.Automatic 3-Dimensional Cephalometric Landmarking via Deep LearningGauthier Dot, Thomas Schouman, Shang‐Hung Chang et al. · 2022 · 70 citations
4. 4.Cervical vertebral maturation as a biologic indicator of skeletal maturityRodrigo César Santiago, Luiz Felipe de Miranda Costa, Robert Willer Farinazzo Vitral et al. · 2012 · 114 citations
5. 5.Validity of the assessment method of skeletal maturation by cervical vertebrae: a systematic review and meta-analysisGraziela Oro Cericato, Marcos Alan Vieira Bittencourt, Luiz Renato Paranhos · 2014 · 126 citations
6. 6.Fully automated determination of the cervical vertebrae maturation stages using deep learning with directional filtersSalih Furkan Atici, Rashid Ansari, Veerasathpurush Allareddy et al. · 2022 · 51 citations
7. 7.Convolutional neural network-based automatic cervical vertebral maturation classification methodHaizhen Li, Yanlong Chen, Qing Wang et al. · 2022 · 35 citations
8. 8.A semi-automated method for bone age assessment using cervical vertebral maturationRoberto Silva Baptista, Camila Leite Quaglio, Laila M. E. H. Mourad et al. · 2011 · 32 citations
9. 9.Correlation between Chronological Age, Cervical Vertebral Maturation and Fishman's Skeletal Maturity Indicators in Southern ChineseH Alkhal, Rwk Wong, A. Bakr M. Rabie · 2008 · 107 citations
10. 10.Evaluation of a multi-stage convolutional neural network-based fully automated landmark identification system using cone-beam computed tomographysynthesized posteroanterior cephalometric imagesMin-Jung Kim, Yi Liu, Song Hee Oh et al. · 2021 · 32 citations