BodyGPS: Anatomical Positioning System

TL;DR

BodyGPS addresses the need for fast, interpretable anatomical localization in medical imaging without runtime atlas registration. It introduces a two-step approach: 3.5D sparse descriptor extraction and atlas-coordinate regression via a 16-layer residual network to produce absolute, navigable anatomical coordinates. Across CT segmentation, longitudinal matching, and MRI landmarking, it demonstrates competitive accuracy with substantial speed advantages and cross-modality applicability, enabling real-time query-style navigation of anatomy. The framework opens opportunities for cross-modality alignment, retrieval, and abnormality detection, with future work on hierarchical refinement and broader modality extension.

Abstract

We introduce a new type of foundational model for parsing human anatomy in medical images that works for different modalities. It supports supervised or unsupervised training and can perform matching, registration, classification, or segmentation with or without user interaction. We achieve this by training a neural network estimator that maps query locations to atlas coordinates via regression. Efficiency is improved by sparsely sampling the input, enabling response times of less than 1 ms without additional accelerator hardware. We demonstrate the utility of the algorithm in both CT and MRI modalities.

Figures (4)

• Figure 1: BodyGPS is a deep learning regressor that maps any query point to reference atlas coordinates
• Figure 2: Application of BodyGPS on segmentation
• Figure 3: Performance of Methods For Longitudinal Matching
• Figure 4: Landmark location estimation with supervised BodyGPS