Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Bonnet: Ultra-fast whole-body bone segmentation from CT scans

Hanjiang Zhu, Pedro Martelleto Rezende, Zhang Yang, Tong Ye, Bruce Z. Gao, Feng Luo, Siyu Huang, Jiancheng Yang

TL;DR

Bonnet tackles the bottleneck of slow whole-body bone segmentation by introducing a sparse-volume pipeline that converts CT scans into a sparse representation using HU thresholds, then performs patch-wise inference with a lightweight spconv U-Net and fuses predictions across windows. It achieves around $2.69$ seconds per scan on an RTX $A6000$, roughly a $25$-fold speedup over dense voxel baselines, while maintaining high Dice accuracy on ribs, pelvis, and spine. Crucially, a model trained only on TotalSegmentator generalizes to external rib, pelvic, and spine datasets without retraining, demonstrating cross-domain robustness. The work enables time-critical clinical workflows and large-scale analyses, and the authors release the toolkit and pre-trained models.

Abstract

This work proposes Bonnet, an ultra-fast sparse-volume pipeline for whole-body bone segmentation from CT scans. Accurate bone segmentation is important for surgical planning and anatomical analysis, but existing 3D voxel-based models such as nnU-Net and STU-Net require heavy computation and often take several minutes per scan, which limits time-critical use. The proposed Bonnet addresses this by integrating a series of novel framework components including HU-based bone thresholding, patch-wise inference with a sparse spconv-based U-Net, and multi-window fusion into a full-volume prediction. Trained on TotalSegmentator and evaluated without additional tuning on RibSeg, CT-Pelvic1K, and CT-Spine1K, Bonnet achieves high Dice across ribs, pelvis, and spine while running in only 2.69 seconds per scan on an RTX A6000. Compared to strong voxel baselines, Bonnet attains a similar accuracy but reduces inference time by roughly 25x on the same hardware and tiling setup. The toolkit and pre-trained models will be released at https://github.com/HINTLab/Bonnet.

Bonnet: Ultra-fast whole-body bone segmentation from CT scans

TL;DR

Bonnet tackles the bottleneck of slow whole-body bone segmentation by introducing a sparse-volume pipeline that converts CT scans into a sparse representation using HU thresholds, then performs patch-wise inference with a lightweight spconv U-Net and fuses predictions across windows. It achieves around seconds per scan on an RTX , roughly a -fold speedup over dense voxel baselines, while maintaining high Dice accuracy on ribs, pelvis, and spine. Crucially, a model trained only on TotalSegmentator generalizes to external rib, pelvic, and spine datasets without retraining, demonstrating cross-domain robustness. The work enables time-critical clinical workflows and large-scale analyses, and the authors release the toolkit and pre-trained models.

Abstract

This work proposes Bonnet, an ultra-fast sparse-volume pipeline for whole-body bone segmentation from CT scans. Accurate bone segmentation is important for surgical planning and anatomical analysis, but existing 3D voxel-based models such as nnU-Net and STU-Net require heavy computation and often take several minutes per scan, which limits time-critical use. The proposed Bonnet addresses this by integrating a series of novel framework components including HU-based bone thresholding, patch-wise inference with a sparse spconv-based U-Net, and multi-window fusion into a full-volume prediction. Trained on TotalSegmentator and evaluated without additional tuning on RibSeg, CT-Pelvic1K, and CT-Spine1K, Bonnet achieves high Dice across ribs, pelvis, and spine while running in only 2.69 seconds per scan on an RTX A6000. Compared to strong voxel baselines, Bonnet attains a similar accuracy but reduces inference time by roughly 25x on the same hardware and tiling setup. The toolkit and pre-trained models will be released at https://github.com/HINTLab/Bonnet.
Paper Structure (13 sections, 7 equations, 2 figures, 2 tables)

This paper contains 13 sections, 7 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Overview
  4. Sparse Voxel Sliding-Window and Fusion
  5. Sparse U-Net Architecture and Sparse Convolutions
  6. Learning Objective
  7. Experiments
  8. Datasets
  9. Experimental Setup
  10. Baseline Methods
  11. Results and Discussion
  12. Conclusion
  13. Acknowledgements

Figures (2)

  • Figure 1: Overview of the Bonnet pipeline. Bonnet first applies simple HU thresholding to obtain a sparse CT volume, then segments bone structures patch-wise using a U-Net backbone built with sparse 3D convolutions, and finally aggregates the patch predictions into the final segmentation result. Bonnet delivers $\sim\!25\times$ faster inference while maintaining anatomical accuracy and robust generalization.
  • Figure 2: Cross-dataset qualitative comparison. From left to right: input CT, ground-truth (GT), Bonnet (ours), PointNet, PVCNN, and SPVCNN. Each row shows an example from the in-domain TotalSegmentator test set and from three external datasets (RibSeg, CT-Spine1K, and CT-Pelvic1K).