1. How has AI-assisted cephalometric analysis evolved?
AI-assisted cephalometric analysis has evolved from manual cephalometric analysis to improve diagnostic value by reducing measurement errors and saving clinical time. The integration of AI in medicine, driven by deep learning algorithms, computing hardware advances, and data growth, has led to its application in orthodontics for evaluating post-treatment results and predicting growth. The YOLOv3 method and cascaded convolutional neural networks have been used to evaluate landmark detection accuracy, with results showing mean detection errors between AI and orthodontists. However, challenges such as reproducibility due to inter-and intra-variability errors in landmark annotation and the need for a sufficient quantity of learning data sets have been identified. Commercially available AI-assisted programs have shown promise in improving accuracy and process automation, but further assessment of their impact on accuracy and comparison with manual tracing is necessary.
read more
2. What ethical standards were followed in the trial design?
The trial design followed the ethical standards of the Clinical Research Ethics Committee of Chongqing Medical University, with approval number 2022-077. Patient data were handled according to the CONSORT Statement and Helsinki Declaration. Informed consent was obtained from patients and their parents or guardians, ensuring transparency and respect for participants' rights. This adherence to ethical guidelines ensures the integrity and credibility of the research findings.
read more
3. What are the 32 commonly used cephalometric landmarks?
The 32 commonly used cephalometric landmarks include 21 hard tissue and 11 soft tissue landmarks. These landmarks are essential for accurate cephalometric analysis and are marked on radiographs. They are defined, positioned, and abbreviated in Table 1. The landmarks are manually digitized by observers and confirmed by mutual agreement. Three observers calibrate with respect to the landmark definitions before registration. The 33 radiographs are coded and presented to the observers in random order for landmark identification. The manual group uses a mouse-controlled cursor, while the automated detection group utilizes commercial software for auto-identification. The AI-assisted group digitizes landmarks after automated identification. To prevent recognition of previous landmarks, markings are removed, and digitization is conducted after a 2-week interval. Intra-observer error is evaluated by conducting a second digitization after a 2-week interval.
read more
4. How are landmarks identified in evaluation matrices?
Landmarks are identified using x-and y-coordinates. The positions of the landmarks are determined by the coordinates from MyOrthoX, Angelalign, and Fig. 1, which consist of 32 anatomical landmarks. These coordinates are used to calculate the Distance Error (DE) between manually annotated landmark coordinates and estimated landmark coordinates by AI. The DE is calculated using the formula (xi - x1)^2 + (yi - y1)^2 (mm), where xi and yi denote the coordinates from the orthodontist group. The successful detection ratio (SDR) is then calculated for 1.0-, 1.5-, and 2.0-mm ranges using the equations SDR = n * (DEi - MDEi) / n (mm), where n represents the number of test images. This process allows for the evaluation of the performance of submitted methods in the context of orthodontic landmark identification.
read more