Blog

3D cephalometric landmark detection using AI

3D cephalometric landmark detection by multiple stage deep reinforcement learning

Sung Ho Kang, Kiwan Jeon, Sang‑Hoon Kang & Sang‑Hwy Lee3

Summary

Manual landmarking in 3D cephalometry is accurate, but it takes a lot of time — which limits its use in daily practice and research. This study, published in Scientific Reports (2021), proposed an automatic system for 3D cephalometric landmark detection using multi-stage deep reinforcement learning (DRL). The model simulates how experts sequentially identify landmarks and applies that logic to CT images. Tested on a dataset of 28 patients, the system reached a mean error of 1.96 ± 0.78 mm and high detection rates within 2.5 to 4 mm of accuracy. Unlike other methods, it does not require prior segmentation or 3D mesh reconstruction, which makes the process faster and more direct.

DOI: https://doi.org/10.1038/s41598-021-97116-7

Key Words

AI in Dentistry, Cephalometry, Deep Reinforcement Learning, Landmark Detection, Dental AI

Extracted Data

Year: 2021

Modality: Computerized Tomography - CT Scan

Dataset: 28 normal Korean adults with skeletal class I occlusion volunteered

Dataset Split: divided into two groups, the training group (n = 20) and the test group (n = 8)

Network Architecture: Double DQN implementation is based on the open-source framework for landmark detection

Metrics: total mean error of the detected landmarks was 1.96 ± 0.78 mm… detection rate within 2.5 mm = 75.39%, 4 mm = 95.70%

AP – Strategy = Not mentioned

AP – Professional Qty: 2

AP – Supervisor Presence: Not mentioned

AP – Experience Level: 10 years

AP – Expertise Area: Orthodontists

AP – Tool or System: StoA software (Korea Copyright Commission No. C-2019-032537 - Daejeon, Korea)

Task: Point (landmark) detection

Project Objective: To develop an automatic 3D cephalometric annotation system using volume-rendered imaging and selective single- or multi-stage DRL

Clinical Relevance

This work is significant for three main reasons:

Efficiency: It reduces the time and complexity of 3D cephalometric analysis, making it closer to real clinical use.
Innovation: By applying deep reinforcement learning (DRL), it introduces a new AI strategy that mimics human annotation behavior.
Future potential: With larger CT datasets, the model could achieve even higher accuracy and support orthodontic and surgical planning with more confidence.

For dentistry and AI, this study shows a concrete step toward automating cephalometric tasks that today still depend heavily on specialists, opening the door for broader clinical adoption.

3D cephalometric landmark detection using AI

3D cephalometric landmark detection by multiple stage deep reinforcement learning

Sung Ho Kang, Kiwan Jeon, Sang‑Hoon Kang & Sang‑Hwy Lee3

Summary

DOI: https://doi.org/10.1038/s41598-021-97116-7

Key Words

AI in Dentistry, Cephalometry, Deep Reinforcement Learning, Landmark Detection, Dental AI

Extracted Data

Year: 2021

Modality: Computerized Tomography - CT Scan

Dataset: 28 normal Korean adults with skeletal class I occlusion volunteered

Dataset Split: divided into two groups, the training group (n = 20) and the test group (n = 8)

Network Architecture: Double DQN implementation is based on the open-source framework for landmark detection

Metrics: total mean error of the detected landmarks was 1.96 ± 0.78 mm… detection rate within 2.5 mm = 75.39%, 4 mm = 95.70%

AP – Strategy = Not mentioned

AP – Professional Qty: 2

AP – Supervisor Presence: Not mentioned

AP – Experience Level: 10 years

AP – Expertise Area: Orthodontists

AP – Tool or System: StoA software (Korea Copyright Commission No. C-2019-032537 - Daejeon, Korea)

Task: Point (landmark) detection

Project Objective: To develop an automatic 3D cephalometric annotation system using volume-rendered imaging and selective single- or multi-stage DRL

Clinical Relevance

This work is significant for three main reasons:

Efficiency: It reduces the time and complexity of 3D cephalometric analysis, making it closer to real clinical use.
Innovation: By applying deep reinforcement learning (DRL), it introduces a new AI strategy that mimics human annotation behavior.
Future potential: With larger CT datasets, the model could achieve even higher accuracy and support orthodontic and surgical planning with more confidence.

For dentistry and AI, this study shows a concrete step toward automating cephalometric tasks that today still depend heavily on specialists, opening the door for broader clinical adoption.

3D cephalometric landmark detection using AI

3D cephalometric landmark detection by multiple stage deep reinforcement learning

Summary

Key Words

Extracted Data

Year: 2021

Modality: Computerized Tomography - CT Scan

Dataset: 28 normal Korean adults with skeletal class I occlusion volunteered

Dataset Split: divided into two groups, the training group (n = 20) and the test group (n = 8)

Network Architecture: Double DQN implementation is based on the open-source framework for landmark detection

Metrics: total mean error of the detected landmarks was 1.96 ± 0.78 mm… detection rate within 2.5 mm = 75.39%, 4 mm = 95.70%

AP – Strategy = Not mentioned

AP – Professional Qty: 2

AP – Supervisor Presence: Not mentioned

AP – Experience Level: 10 years

AP – Expertise Area: Orthodontists

AP – Tool or System: StoA software (Korea Copyright Commission No. C-2019-032537 - Daejeon, Korea)

Task: Point (landmark) detection

Project Objective: To develop an automatic 3D cephalometric annotation system using volume-rendered imaging and selective single- or multi-stage DRL

Clinical Relevance

For dentistry and AI, this study shows a concrete step toward automating cephalometric tasks that today still depend heavily on specialists, opening the door for broader clinical adoption.

Related Video

3D cephalometric landmark detection using AI

3D cephalometric landmark detection by multiple stage deep reinforcement learning

Summary

Key Words

Extracted Data

Year: 2021

Modality: Computerized Tomography - CT Scan

Dataset: 28 normal Korean adults with skeletal class I occlusion volunteered

Dataset Split: divided into two groups, the training group (n = 20) and the test group (n = 8)

Network Architecture: Double DQN implementation is based on the open-source framework for landmark detection

Metrics: total mean error of the detected landmarks was 1.96 ± 0.78 mm… detection rate within 2.5 mm = 75.39%, 4 mm = 95.70%

AP – Strategy = Not mentioned

AP – Professional Qty: 2

AP – Supervisor Presence: Not mentioned

AP – Experience Level: 10 years

AP – Expertise Area: Orthodontists

AP – Tool or System: StoA software (Korea Copyright Commission No. C-2019-032537 - Daejeon, Korea)

Task: Point (landmark) detection

Project Objective: To develop an automatic 3D cephalometric annotation system using volume-rendered imaging and selective single- or multi-stage DRL

Clinical Relevance

For dentistry and AI, this study shows a concrete step toward automating cephalometric tasks that today still depend heavily on specialists, opening the door for broader clinical adoption.

Related Video