Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Deep Learning Techniques for Automatic Lateral X-ray Cephalometric Landmark Detection: Is the Problem Solved?

Hongyuan Zhang, Ching-Wei Wang, Hikam Muzakky, Juan Dai, Xuguang Li, Chenglong Ma, Qian Wu, Xianan Cui, Kunlun Xu, Pengfei He, Dongqian Guo, Xianlong Wang, Hyunseok Lee, Zhangnan Zhong, Zhu Zhu, Bingsheng Huang

TL;DR

This paper introduces the "Cephalometric Landmark Detection (CL-Detection)" dataset, which is the largest publicly available and comprehensive dataset for cephalometric landmark detection and identifies scenarios for which deep learning methods are still failing.

Abstract

Localization of the craniofacial landmarks from lateral cephalograms is a fundamental task in cephalometric analysis. The automation of the corresponding tasks has thus been the subject of intense research over the past decades. In this paper, we introduce the "Cephalometric Landmark Detection (CL-Detection)" dataset, which is the largest publicly available and comprehensive dataset for cephalometric landmark detection. This multi-center and multi-vendor dataset includes 600 lateral X-ray images with 38 landmarks acquired with different equipment from three medical centers. The overarching objective of this paper is to measure how far state-of-the-art deep learning methods can go for cephalometric landmark detection. Following the 2023 MICCAI CL-Detection Challenge, we report the results of the top ten research groups using deep learning methods. Results show that the best methods closely approximate the expert analysis, achieving a mean detection rate of 75.719% and a mean radial error of 1.518 mm. While there is room for improvement, these findings undeniably open the door to highly accurate and fully automatic location of craniofacial landmarks. We also identify scenarios for which deep learning methods are still failing. Both the dataset and detailed results are publicly available online, while the platform will remain open for the community to benchmark future algorithm developments at https://cl-detection2023.grand-challenge.org/.

Deep Learning Techniques for Automatic Lateral X-ray Cephalometric Landmark Detection: Is the Problem Solved?

TL;DR

This paper introduces the "Cephalometric Landmark Detection (CL-Detection)" dataset, which is the largest publicly available and comprehensive dataset for cephalometric landmark detection and identifies scenarios for which deep learning methods are still failing.

Abstract

Localization of the craniofacial landmarks from lateral cephalograms is a fundamental task in cephalometric analysis. The automation of the corresponding tasks has thus been the subject of intense research over the past decades. In this paper, we introduce the "Cephalometric Landmark Detection (CL-Detection)" dataset, which is the largest publicly available and comprehensive dataset for cephalometric landmark detection. This multi-center and multi-vendor dataset includes 600 lateral X-ray images with 38 landmarks acquired with different equipment from three medical centers. The overarching objective of this paper is to measure how far state-of-the-art deep learning methods can go for cephalometric landmark detection. Following the 2023 MICCAI CL-Detection Challenge, we report the results of the top ten research groups using deep learning methods. Results show that the best methods closely approximate the expert analysis, achieving a mean detection rate of 75.719% and a mean radial error of 1.518 mm. While there is room for improvement, these findings undeniably open the door to highly accurate and fully automatic location of craniofacial landmarks. We also identify scenarios for which deep learning methods are still failing. Both the dataset and detailed results are publicly available online, while the platform will remain open for the community to benchmark future algorithm developments at https://cl-detection2023.grand-challenge.org/.
Paper Structure (33 sections, 2 equations, 9 figures, 6 tables)

This paper contains 33 sections, 2 equations, 9 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Previous works
  3. Previous X-ray cephalometric datasets
  4. Non-deep learning methods
  5. Deep learning-based methods
  6. Challenge description
  7. Organization
  8. Image datasets
  9. Reference detection and annotation protocol
  10. Evaluation metrics and and ranking scheme
  11. Competing solutions
  12. T1: Chohotech Technology Co., Ltd (CTT)
  13. T2: Singapore University of Technology and Design (SUTD)
  14. T3: Wuhan University (WHU)
  15. T4: Peking University (PKU)
  16. ...and 18 more sections

Figures (9)

  • Figure 1: Challenges in cephalometric landmark detection: (a) Overlapping craniofacial structures (red box) and poor contrast (orange box): In cephalometric X-ray images, craniofacial structures overlap, and soft tissue-related region is low-contrast, making it hard to distinguish individual components. (b) Landmark occlusion: Landmarks could be obscured by dental braces (blue box), implants (green box) or lead shields (purple box). (c) Site and individual variability: t-SNE visualization of the challenge data from three medical centers reveals anatomical variations leading to differences in landmark appearance and location, not only across centers but also within the same center.
  • Figure 2: The workflow of the MICCAI CL-Detection2023 challenge consists of four stages: (1) Data preparation, (2) Training phase, (3) Validation and testing phase, and (4) Result analysis.
  • Figure 3: Comparative analysis of the differences between our previous work ISBI challenge and CL-Detection2023 challenge. The green highlights represent the anatomical landmarks featured in the ISBI 2015 challenge. In CL-Detection2023 challenge, we have extended dataset from a single center to a multi-center, multi-vendor and more landmark annotations, which are highlighted in orange.
  • Figure 4: Summary of CL-Detection challenge participants and submissions. There were 348 teams registering on the official grand-challenge website and 171 of them were approved before the end of the training phase. Finally, 46 teams submitted validation results and 37 teams submitted Docker containers for test leaderboard.
  • Figure 5: Dot- and boxplot visualization (a and c) and statistical significance maps (b and d) for the MRE and SDR@2.0mm metrics of the top 10 teams. (a) and (c), (b) and (d) are the results for MRE and SDR@2.0mm, respectively. In the statistical significance map, light yellow shading indicates that the MRE scores of the teams on the $x$-axis are significantly superior to the scores of the teams on the $y$-axis (p-value $<$ 0.05) whereas light blue shading indicates they are not significantly superior.
  • ...and 4 more figures