Optimising for the Unknown: Domain Alignment for Cephalometric Landmark Detection

Julian Wyatt; Irina Voiculescu

cs.CVOct 6, 2024

Optimising for the Unknown: Domain Alignment for Cephalometric Landmark Detection

Authors: Julian Wyatt, Irina Voiculescu

Organizations: Department of Computer Science, University of Oxford

Abstract

Cephalometric Landmark Detection is the process of identifying key areas for cephalometry. Each landmark is a single GT point labelled by a clinician. A machine learning model predicts the probability locus of a landmark represented by a heatmap. This work, for the 2024 CL-Detection MICCAI Challenge, proposes a domain alignment strategy with a regional facial extraction module and an X-ray artefact augmentation procedure. The challenge ranks our method's results as the best in MRE of 1.186mm and third in the 2mm SDR of 82.04% on the online validation leaderboard. The code is available at https://github.com/Julian-Wyatt/OptimisingfortheUnknown.

Figures & tables

Figure 1: Abstraction of the RCNN regional face extraction module.

Figure 2: ConvNeXt V2 MLP feature pyramid prediction architecture.

Panel of Figure 3 — Figure 3: Ablation studies trained with the nano encoder. (a) Impact of RCNN padding around landmarks for bounding box generation during training from 16-128 pixels. Models are trained on patients 1-400 and evaluated on 401-600. Here, y is a deterministic image pad and crop method and z is a standard pad to aspect ratio and resize. (b) Effect of varying the frequency of X-ray artefact simulation on model performance (Fold 1). (c) Influence of the number of hottest heatmap values averaged to determine the final coordinates (Fold 1).

Figure 4: Qualitative results for four cases: the worst overall SDR for the tiny encoder ( 062.bmp ), the worst for the baseline ( 194.bmp ), and two hand-picked failure cases ( 447.bmp & 535.bmp ). Blue dots represent predictions within $2mm$ , red dots are outside $2mm$ , and green dots are ground truth, with yellow lines connecting predictions to ground truth. The top-left number indicates the MRE for each image, and all images are cropped to square to highlight landmarks rather than by the RCNN module.

Method	Fold 1		Fold 2		Fold 3		Fold 4		Ensemble
Method	MRE	SDR	MRE	SDR	MRE	SDR	MRE	SDR	MRE	SDR
Baseline [ 11 ] (CHH)	1.20	83.12	1.21	82.44	1.29	81.56	1.25	82.19	1.22	82.42
Nano Encoder	1.16	84.13	1.16	83.87	1.24	82.78	1.20	83.40	0.97	87.85
Tiny Encoder	1.18	84.13	1.16	83.68	1.24	82.49	1.20	83.20	0.93	88.61

Table 1: Four fold cross validation performance with ensembling performance across every training image.

Explore similar work

Automatic Cephalometric Landmark Localization on CBCT-Derived Digitally Reconstructed Radiographs for Skeletal Malocclusion Classification
Aug 17, 2026Benjamin Hou, Konstantinia Almpani, Janice S. Lee +1Medical Image Localization Vision Transformer
Division of Intramural Research, National Library of Medicine · Craniofacial Anomalies and Regeneration Section, National Institute of Dental and Craniofacial Research, National Institutes of Health, Bethesda, MD, USA.
Towards Unified Dynamic Face Landmark Detection
Aug 11, 2026Sebastian Regalado, Varshanth R. Rao, Ruowei Jiang +2
University of Toronto · 2ModiFace · †Work completed while employed at ModiFace +1
CDPM-Align: Multi-Scale Guidance-Aligned Diffusion Pretraining for Robust Few-Shot Anatomical Landmark Detection
Jun 3, 2026Roberto Di Via, Irina Voiculescu, Francesca Odone +1Diffusion Model Alignment Diffusion Model Guidance
MaLGa, DIBRIS, University of Genoa, Italy · Department of Computer Science, University of Oxford, UK

Aug 17, 2026J/K move · Enter open · S save

Automatic Cephalometric Landmark Localization on CBCT-Derived Digitally Reconstructed Radiographs for Skeletal Malocclusion Classification

Benjamin Hou, Konstantinia Almpani, Janice S. Lee, Zhiyong Lu

Division of Intramural Research, National Library of Medicine · Craniofacial Anomalies and Regeneration Section, National Institute of Dental and Craniofacial Research, National Institutes of Health, Bethesda, MD, USA.

Medical Image Localization Vision Transformer

Open paper AskPDF arXiv

Manual cephalometric landmark annotation is important for craniofacial assessment but is labor-intensive and difficult to scale. We introduce CephViT, a Vision Transformer-based model for automated 2D lateral cephalometric landmark localization, and evaluate its use in downstream skeletal malocclusion classification. CephViT was trained and benchmarked on a public lateral cephalogram dataset, achieving a mean radial error of 1.28 +/- 1.42 mm and a successful detection rate of 92.0% at 3.0 mm. Because the private evaluation cohort consisted of 3D CBCT scans, lateral cephalogram-like digitally reconstructed radiographs (DRRs) were generated from each volume and used as 2D inputs to the landmark localization model. Landmark coordinates were normalized into a common coordinate frame, and skeletal malocclusion classification was performed using landmarks shared between the reference and DRR-based pipelines. Classification performance using DRR-localized landmarks was comparable to that obtained using manually annotated reference landmarks, with accuracies of 70.0% and 68.3%, respectively. These results support the feasibility of automated cephalometric analysis on CBCT-derived DRRs for skeletal malocclusion assessment.

Abstract

Explore similar work

Automatic Cephalometric Landmark Localization on CBCT-Derived Digitally Reconstructed Radiographs for Skeletal Malocclusion Classification

Towards Unified Dynamic Face Landmark Detection

CDPM-Align: Multi-Scale Guidance-Aligned Diffusion Pretraining for Robust Few-Shot Anatomical Landmark Detection