Orthodontic Diagnosis and AI

Study Notes

The most important parts of orthodontic treatment are determining the treatment plan, outcomes, and monitoring patients.
AI can automate manual work, speed up diagnosis, treatment planning, and assessment of growth patterns.

Advances in digital data have increased the demand for automating cephalometric analysis and diagnostic tasks.
Automation aims to reduce time required for analysis, improve accuracy of landmark identification, and reduce errors due to clinicians' subjectivity.

Automatic identification of landmarks has been undertaken using computer vision, AI, and deep learning techniques for 20 years.
Trained AI algorithms can analyze and annotate cephalometric landmarks on radiological images in a fraction of a second with comparable precision to experienced human examiners.
However, the usefulness of 3D cephalometric analysis based on 2D cephalometric measurement and landmarks has been questioned.

More accurate and reliable methods to perform cephalometric superimposition, such as voxel-based superimposition and surface-to-surface matching, have been introduced.
These methods have resulted in cephalometric landmarking becoming outdated.

Machine learning and deep learning techniques have been applied for fully automatic segmentation of maxillary and mandibular bones, upper airway, and for skeletal bone age assessment.
Segmentation is the construction of 3D virtual surface models to match volumetric data, allowing for evaluation of size, shape, and volume of anatomic structures.
Manual segmentation is time-consuming, tedious, and requires expertise, making automated methods desirable.

Completely fully automated systems to segment any structure from CBCT images have been developed using AI, particularly convolutional neural networks (CNNs).
CNNs have led to breakthroughs in CBCT segmentation, especially when compared to previous methods employing general hand-crafted features.