Automated Identification of Cephalometric Landmarks Using Cascade CNNs

Accuracy of Automated Identification of Lateral Cephalometric Landmarks Using Cascade CNNs

Authors: Jaerong Kim; Inhwan Kim; Yoon‑Ji Kim; Minji Kim; Jin‑Hyoung Cho; Mihee Hong; Kyung‑Hwa Kang; Sung‑Hoon Lim; Su‑Jung Kim; Young Ho Kim; Namkug Kim; Sang‑Jin Sung; Seung‑Hak Baek

Summary:

This nationwide multi-centre study developed and validated a cascade CNN model for automatic identification of 20 cephalometric landmarks on lateral cephalograms. Using RetinaNet for region detection and U-Net for precise landmark localisation, the system achieved a mean error of 1.36 ± 0.98 mm, comparable to experienced orthodontists. Performance was evaluated across sensors, hospitals and machine vendors, demonstrating high generalizability.

Check the original publication:

Key Words:

Cephalometry; CNN; Cascade Network; Landmark Detection; RetinaNet; U‑Net; Lateral Cephalogram

Extracted Data

Year: 2021

Modality: Lateral Cephalometric Radiographs

Dataset: 3,150 training images (10 centres) + 100 external validation images

Dataset Split: Training + Internal Validation + External Validation

Network Architecture: Cascade CNN (RetinaNet ROI detector + U‑Net Landmark Model)

Metrics: Mean detection error: 1.36 ± 0.98 mm

AP - Professional Qty: 4

AP - Supervisor Presence: No information

AP - Experience Level: 10–30 years

AP - Expertise Area: Orthodontics

AP - Tool or System: V-Ceph

ML Task: Object detection - Landmark Detection (20 points)

Project Objective: Develop a generalizable and accurate automatic cephalometric landmark detection model.

Clinical Relevance:

Automates cephalometric landmarking to support diagnosis, treatment planning and mid‑treatment evaluation across multiple imaging systems.

Watch full video on YouTube