landmarker: a Toolkit for Anatomical Landmark Localization in 2D/3D Images

TL;DR

The paper tackles the challenge of precise anatomical landmark localization in 2D/3D medical images, where generic pose-estimation tools fall short in modularity and medical-format support. It introduces landmarker, a PyTorch-based toolkit with modular components for data handling, heatmap generation (Gaussian/Laplace with potential learnable covariances), and decoding (argmax, weighted mean, local soft-argmax), plus loss functions like GaussianHeatmapL2Loss, all integrated with MONAI for healthcare imaging. Key contributions include a GaussianHeatmapGenerator, diverse dataset types (e.g., LandmarkDataset, HeatmapDataset), end-to-end evaluation/visualization utilities, and benchmarks showing improved accuracy (PE) and SDR on 2D pelvis X-ray and 3D skull CT datasets. The framework accelerates development of medical landmark localization methods and sets the stage for future work in uncertainty quantification within a unified, extensible platform $H_i$ and $\hat{y}_i$.

Abstract

Anatomical landmark localization in 2D/3D images is a critical task in medical imaging. Although many general-purpose tools exist for landmark localization in classical computer vision tasks, such as pose estimation, they lack the specialized features and modularity necessary for anatomical landmark localization applications in the medical domain. Therefore, we introduce landmarker, a Python package built on PyTorch. The package provides a comprehensive, flexible toolkit for developing and evaluating landmark localization algorithms, supporting a range of methodologies, including static and adaptive heatmap regression. landmarker enhances the accuracy of landmark identification, streamlines research and development processes, and supports various image formats and preprocessing pipelines. Its modular design allows users to customize and extend the toolkit for specific datasets and applications, accelerating innovation in medical imaging. landmarker addresses a critical need for precision and customization in landmark localization tasks not adequately met by existing general-purpose pose estimation tools.

Figures (4)

• Figure 1: Taxonomy of (deep learning) landmark localization approaches. The frequently used taxonomy in the literature is highlighted in yellow, while our extended taxonomy of the problem is highlighted in blue.
• Figure 2: Flowchart of the different sub-packages of the landmarker pacakge.
• Figure 3: Example of results of running the inspection_plot on the ISBI2015 dataset.
• Figure 4: Examples of landmark predictions using the prediction_inspection_plot tool for different datasets: (a) Endoscopic images, (b) ISBI2015, and (c) Pelvis X-rays.