Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Improved 3D Whole Heart Geometry from Sparse CMR Slices

Yiyang Xu, Hao Xu, Matthew Sinclair, Esther Puyol-Antón, Steven A Niederer, Amedeo Chiribiri, Steven E Williams, Michelle C Williams, Alistair A Young

TL;DR

This work tackles reconstructing accurate 3D whole-heart geometry from sparse CMR slices, a challenge amplified by respiratory motion and limited dense data. It introduces and evaluates six model combinations that fuse Slice Shifting Algorithm (SSA) for motion correction, Spatial Transformer Network (STN) for atlas-based deformation, and Label Transformer Network (LTN) for dense geometry reconstruction, using CT-derived ground truth to generate synthetic CMR data. Key findings show that SSA-LTN achieves the highest Dice and Hausdorff distance performance but can introduce topological errors, STN effectively fixes topology with minimal performance loss, and DSTN further reduces topology errors, with SSA acting as a versatile plug-in. The results offer practical guidance for robust, topology-consistent 3D whole-heart reconstructions from sparse CMR data, enabling non-invasive imaging workflows with improved downstream shape analysis and clinical applicability.

Abstract

Cardiac magnetic resonance (CMR) imaging and computed tomography (CT) are two common non-invasive imaging methods for assessing patients with cardiovascular disease. CMR typically acquires multiple sparse 2D slices, with unavoidable respiratory motion artefacts between slices, whereas CT acquires isotropic dense data but uses ionising radiation. In this study, we explore the combination of Slice Shifting Algorithm (SSA), Spatial Transformer Network (STN), and Label Transformer Network (LTN) to: 1) correct respiratory motion between segmented slices, and 2) transform sparse segmentation data into dense segmentation. All combinations were validated using synthetic motion-corrupted CMR slice segmentation generated from CT in 1699 cases, where the dense CT serves as the ground truth. In 199 testing cases, SSA-LTN achieved the best results for Dice score and Huasdorff distance (94.0% and 4.7 mm respectively, average over 5 labels) but gave topological errors in 8 cases. STN was effective as a plug-in tool for correcting all topological errors with minimal impact on overall performance (93.5% and 5.0 mm respectively). SSA also proves to be a valuable plug-in tool, enhancing performance over both STN-based and LTN-based models. The code for these different combinations is available at https://github.com/XESchong/STACOM2024.

Improved 3D Whole Heart Geometry from Sparse CMR Slices

TL;DR

This work tackles reconstructing accurate 3D whole-heart geometry from sparse CMR slices, a challenge amplified by respiratory motion and limited dense data. It introduces and evaluates six model combinations that fuse Slice Shifting Algorithm (SSA) for motion correction, Spatial Transformer Network (STN) for atlas-based deformation, and Label Transformer Network (LTN) for dense geometry reconstruction, using CT-derived ground truth to generate synthetic CMR data. Key findings show that SSA-LTN achieves the highest Dice and Hausdorff distance performance but can introduce topological errors, STN effectively fixes topology with minimal performance loss, and DSTN further reduces topology errors, with SSA acting as a versatile plug-in. The results offer practical guidance for robust, topology-consistent 3D whole-heart reconstructions from sparse CMR data, enabling non-invasive imaging workflows with improved downstream shape analysis and clinical applicability.

Abstract

Cardiac magnetic resonance (CMR) imaging and computed tomography (CT) are two common non-invasive imaging methods for assessing patients with cardiovascular disease. CMR typically acquires multiple sparse 2D slices, with unavoidable respiratory motion artefacts between slices, whereas CT acquires isotropic dense data but uses ionising radiation. In this study, we explore the combination of Slice Shifting Algorithm (SSA), Spatial Transformer Network (STN), and Label Transformer Network (LTN) to: 1) correct respiratory motion between segmented slices, and 2) transform sparse segmentation data into dense segmentation. All combinations were validated using synthetic motion-corrupted CMR slice segmentation generated from CT in 1699 cases, where the dense CT serves as the ground truth. In 199 testing cases, SSA-LTN achieved the best results for Dice score and Huasdorff distance (94.0% and 4.7 mm respectively, average over 5 labels) but gave topological errors in 8 cases. STN was effective as a plug-in tool for correcting all topological errors with minimal impact on overall performance (93.5% and 5.0 mm respectively). SSA also proves to be a valuable plug-in tool, enhancing performance over both STN-based and LTN-based models. The code for these different combinations is available at https://github.com/XESchong/STACOM2024.
Paper Structure (18 sections, 10 equations, 4 figures, 2 tables, 1 algorithm)

This paper contains 18 sections, 10 equations, 4 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related work
  3. Contributions
  4. Methodology
  5. Slice Shifting Algorithm (SSA)
  6. Spatial Transformer Network (STN)
  7. Label Transformer Network (LTN)
  8. Experiments
  9. Synthetic CMR dataset
  10. Implementation details
  11. Evaluation metrics
  12. Results
  13. Conclusion and Discussion
  14. Acknowledgments.
  15. Supplementary
  16. ...and 3 more sections

Figures (4)

  • Figure 1: Overview of experiments: six combinations of SSA, LTN and STN (sparse and dense versions) were examined. The yellow circle denotes a topological error from LTN, corrected by STN (zoom in for better view).
  • Figure 2: STN architecture (left) and Transformation Computation Module (right) sinclair2020atlasistn
  • Figure 3: Visualisation of the performance for SSA-STN, SSA-LTN and SSA-LTN-DSTN, where GT is ground truth dense volume from CT data.
  • Figure 4: SSA for synthetic CMR data: it shows the intersection between the current slice and other slices under each iteration. Convergence is achieved after 3 iterations.