nnLandmark: A Self-Configuring Method for 3D Medical Landmark Detection

TL;DR

nnLandmark introduces a self-configuring, heatmap-based approach for 3D medical landmark detection by adapting the nnU-Net framework. It converts landmark annotations into per-channel Gaussian heatmaps (with $\sigma=4$) and trains with Mean Squared Error using Adam, followed by channel-wise maximum extraction to recover landmark coordinates, with five-fold cross-validation and model ensembling. Across AFIDs (brain MRI) and complete MML (dental CT) datasets, it achieves MRE around $1.25$–$1.5$ mm and SDR within $2$–$3$ mm, approaching inter-rater variability and surpassing several baselines, while validating strong cross-dataset generalization. The work establishes nnLandmark as a robust, open-source baseline to standardize benchmarking and accelerate translation of precise anatomical localization in clinical workflows.

Abstract

Landmark detection plays a crucial role in medical imaging tasks that rely on precise spatial localization, including specific applications in diagnosis, treatment planning, image registration, and surgical navigation. However, manual annotation is labor-intensive and requires expert knowledge. While deep learning shows promise in automating this task, progress is hindered by limited public datasets, inconsistent benchmarks, and non-standardized baselines, restricting reproducibility, fair comparisons, and model generalizability. This work introduces nnLandmark, a self-configuring deep learning framework for 3D medical landmark detection, adapting nnU-Net to perform heatmap-based regression. By leveraging nnU-Net's automated configuration, nnLandmark eliminates the need for manual parameter tuning, offering out-of-the-box usability. It achieves state-of-the-art accuracy across two public datasets, with a mean radial error (MRE) of 1.5 mm on the Mandibular Molar Landmark (MML) dental CT dataset and 1.2 mm for anatomical fiducials on a brain MRI dataset (AFIDs), where nnLandmark aligns with the inter-rater variability of 1.5 mm. With its strong generalization, reproducibility, and ease of deployment, nnLandmark establishes a reliable baseline for 3D landmark detection, supporting research in anatomical localization and clinical workflows that depend on precise landmark identification. The code will be available soon.

Figures (3)

• Figure 1: Overview of the proposed nnLandmark approach, leveraging key characteristics of the nnU-Net for heatmap-based 3D medical landmark detection by adjusting the respective fixed parameters.
• Figure 2: The landmark segmentations are transformed to heatmaps at the end of the data-augmentation pipeline. Thereby each landmark is represented by a Gaussian blob in a dedicated channel. In the postprocessing, the exact positions of the landmarks are then identified by taking the channel-wise maximum.
• Figure 3: Qualitative results for each test subset with ground truth (green), model prediction (red) and error (yellow). If the label and the prediction were at the exact voxel, it was only colored green. The AFIDs images show results from our stratified split.