Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

RoCoSDF: Row-Column Scanned Neural Signed Distance Fields for Freehand 3D Ultrasound Imaging Shape Reconstruction

Hongbo Chen, Yuchong Gao, Shuhang Zhang, Jiangjie Wu, Yuexin Ma, Rui Zheng

TL;DR

This paper tackles freehand multi-view 3D ultrasound surface reconstruction under elevation distortion from thick transducers. It introduces RoCoSDF, a row-column scanned neural SDF framework that learns two SDFs from row and column views in a normalized space, fuses them with constructive solid geometry to form an initial distance field, and then refines the surface through supervised sampling and targeted regularizers to produce high-fidelity implicit surfaces without ground-truth SDFs. The approach shows quantitative and qualitative improvements over a neural UNSR baseline across twelve vertebra shapes and two ultrasound probes, validating its generalization and robustness, with ablations highlighting the importance of the refinement stage and surface regularizers. The method eliminates reliance on large-scale shape supervision and is adaptable to robotic 3D US scenarios, offering a practical, extensible path for accurate freehand ultrasound geometry reconstruction; code is publicly available.

Abstract

The reconstruction of high-quality shape geometry is crucial for developing freehand 3D ultrasound imaging. However, the shape reconstruction of multi-view ultrasound data remains challenging due to the elevation distortion caused by thick transducer probes. In this paper, we present a novel learning-based framework RoCoSDF, which can effectively generate an implicit surface through continuous shape representations derived from row-column scanned datasets. In RoCoSDF, we encode the datasets from different views into the corresponding neural signed distance function (SDF) and then operate all SDFs in a normalized 3D space to restore the actual surface contour. Without requiring pre-training on large-scale ground truth shapes, our approach can synthesize a smooth and continuous signed distance field from multi-view SDFs to implicitly represent the actual geometry. Furthermore, two regularizers are introduced to facilitate shape refinement by constraining the SDF near the surface. The experiments on twelve shapes data acquired by two ultrasound transducer probes validate that RoCoSDF can effectively reconstruct accurate geometric shapes from multi-view ultrasound data, which outperforms current reconstruction methods. Code is available at https://github.com/chenhbo/RoCoSDF.

RoCoSDF: Row-Column Scanned Neural Signed Distance Fields for Freehand 3D Ultrasound Imaging Shape Reconstruction

TL;DR

This paper tackles freehand multi-view 3D ultrasound surface reconstruction under elevation distortion from thick transducers. It introduces RoCoSDF, a row-column scanned neural SDF framework that learns two SDFs from row and column views in a normalized space, fuses them with constructive solid geometry to form an initial distance field, and then refines the surface through supervised sampling and targeted regularizers to produce high-fidelity implicit surfaces without ground-truth SDFs. The approach shows quantitative and qualitative improvements over a neural UNSR baseline across twelve vertebra shapes and two ultrasound probes, validating its generalization and robustness, with ablations highlighting the importance of the refinement stage and surface regularizers. The method eliminates reliance on large-scale shape supervision and is adaptable to robotic 3D US scenarios, offering a practical, extensible path for accurate freehand ultrasound geometry reconstruction; code is publicly available.

Abstract

The reconstruction of high-quality shape geometry is crucial for developing freehand 3D ultrasound imaging. However, the shape reconstruction of multi-view ultrasound data remains challenging due to the elevation distortion caused by thick transducer probes. In this paper, we present a novel learning-based framework RoCoSDF, which can effectively generate an implicit surface through continuous shape representations derived from row-column scanned datasets. In RoCoSDF, we encode the datasets from different views into the corresponding neural signed distance function (SDF) and then operate all SDFs in a normalized 3D space to restore the actual surface contour. Without requiring pre-training on large-scale ground truth shapes, our approach can synthesize a smooth and continuous signed distance field from multi-view SDFs to implicitly represent the actual geometry. Furthermore, two regularizers are introduced to facilitate shape refinement by constraining the SDF near the surface. The experiments on twelve shapes data acquired by two ultrasound transducer probes validate that RoCoSDF can effectively reconstruct accurate geometric shapes from multi-view ultrasound data, which outperforms current reconstruction methods. Code is available at https://github.com/chenhbo/RoCoSDF.
Paper Structure (18 sections, 7 equations, 4 figures, 1 table)

This paper contains 18 sections, 7 equations, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methodology
  3. Overview
  4. RoCoSDF: Multi-view Neural Shape Reconstruction
  5. Row-Column Neural SDFs Prediction
  6. Signed Distance Fields Fusion
  7. SDF Sampling and Refinement
  8. Model Training and Loss Functions
  9. Self-supervised Learning
  10. Supervised Learning
  11. Experiments and Results
  12. Data Acquisition and Preparation
  13. Evaluation Metrics
  14. Implementation Details
  15. Qualitative and Quantitative Results
  16. ...and 3 more sections

Figures (4)

  • Figure 1: Two typical data acquisition manners for freehand 3D US imaging using one ultrasound transducer (UT). (a) and (b) Single-view scanning and shape reconstruction in row and column directions from UNSR chenNeuralImplicitSurface2024. (c) Proposed row-column scan and multi-view shape reconstruction. The shape contour (dark) is optimized from row-column scan. In practical use for handheld scans, our row-column scan is not necessarily orthogonal, as indicated by the dashed blue lines and red lines in the left circle.
  • Figure 2: An overview of the proposed framework for shape reconstruction. (a) Row-Column neural SDFs prediction from point cloud $P_{ro}$ and $P_{co}$. (b) SDFs fusion using constructive solid geometry (CSG). (c) SDF sampling and refinement.
  • Figure 3: Visualization of thoracic vertebra T4. The top and bottom are the 3D meshes and colorized error maps from CAD model, UNSR (row-scan), UNSR (column-scan) and RoCoSDF (ours).
  • Figure 4: Ablation study on the effectiveness of step (c) and two regularizers.