Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Implicit Neural Representations for Breathing-compensated Volume Reconstruction in Robotic Ultrasound

Yordanka Velikova, Mohammad Farid Azampour, Walter Simson, Marco Esposito, Nassir Navab

TL;DR

This study introduces a method to compensate for breathing motion in 3D US compounding by leveraging implicit neural representations and demonstrates that the proposed pipeline facilitates robust automated robotic acquisition, mitigating artifacts from breathing motion, and yields smoother 3D reconstructions for enhanced screening and medical diagnosis.

Abstract

Ultrasound (US) imaging is widely used in diagnosing and staging abdominal diseases due to its lack of non-ionizing radiation and prevalent availability. However, significant inter-operator variability and inconsistent image acquisition hinder the widespread adoption of extensive screening programs. Robotic ultrasound systems have emerged as a promising solution, offering standardized acquisition protocols and the possibility of automated acquisition. Additionally, these systems enable access to 3D data via robotic tracking, enhancing volumetric reconstruction for improved ultrasound interpretation and precise disease diagnosis. However, the interpretability of 3D US reconstruction of abdominal images can be affected by the patient's breathing motion. This study introduces a method to compensate for breathing motion in 3D US compounding by leveraging implicit neural representations. Our approach employs a robotic ultrasound system for automated screenings. To demonstrate the method's effectiveness, we evaluate our proposed method for the diagnosis and monitoring of abdominal aorta aneurysms as a representative use case. Our experiments demonstrate that our proposed pipeline facilitates robust automated robotic acquisition, mitigating artifacts from breathing motion, and yields smoother 3D reconstructions for enhanced screening and medical diagnosis.

Implicit Neural Representations for Breathing-compensated Volume Reconstruction in Robotic Ultrasound

TL;DR

This study introduces a method to compensate for breathing motion in 3D US compounding by leveraging implicit neural representations and demonstrates that the proposed pipeline facilitates robust automated robotic acquisition, mitigating artifacts from breathing motion, and yields smoother 3D reconstructions for enhanced screening and medical diagnosis.

Abstract

Ultrasound (US) imaging is widely used in diagnosing and staging abdominal diseases due to its lack of non-ionizing radiation and prevalent availability. However, significant inter-operator variability and inconsistent image acquisition hinder the widespread adoption of extensive screening programs. Robotic ultrasound systems have emerged as a promising solution, offering standardized acquisition protocols and the possibility of automated acquisition. Additionally, these systems enable access to 3D data via robotic tracking, enhancing volumetric reconstruction for improved ultrasound interpretation and precise disease diagnosis. However, the interpretability of 3D US reconstruction of abdominal images can be affected by the patient's breathing motion. This study introduces a method to compensate for breathing motion in 3D US compounding by leveraging implicit neural representations. Our approach employs a robotic ultrasound system for automated screenings. To demonstrate the method's effectiveness, we evaluate our proposed method for the diagnosis and monitoring of abdominal aorta aneurysms as a representative use case. Our experiments demonstrate that our proposed pipeline facilitates robust automated robotic acquisition, mitigating artifacts from breathing motion, and yields smoother 3D reconstructions for enhanced screening and medical diagnosis.
Paper Structure (14 sections, 9 equations, 7 figures, 1 table)

This paper contains 14 sections, 9 equations, 7 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methodology
  3. Image acquisition and segmentation
  4. Navigation pipeline
  5. 3D Volume Reconstruction via the INR Model
  6. Setup
  7. Hardware
  8. Training Details
  9. Evaluation
  10. Breath-hold with button
  11. Free breathing
  12. Second volunteer
  13. Results and Discussion
  14. CONCLUSIONS

Figures (7)

  • Figure 1: Overview of the proposed system: a) we use a robotic arm manipulator with a convex ultrasound probe attached to the end-effector to scan the patient, b) Live B-mode, c) Intermediate image, d) Aorta segmentation, e) 3D Reconstruction.
  • Figure 2: Overview of the pipeline. Phase 1: Robotic ultrasound acquisition with real-time image segmentation; the robot's trajectory is adjusted based on the segmentation. Phase 2: The acquired sweep and segmentation train the INR model, which is sampled to produce a dense aorta point cloud. Post-processing of this cloud yields the final mesh.
  • Figure 3: Robot transformations.
  • Figure 4: Schematic of the INR model. The model accepts a normalized 3D voxel position (ranging from $-1$ to $1$) as input and produces two outputs: the intensity and the semantic label of the voxel. For training, we leverage the ultrasound (US) sweep alongside its corresponding segmentation. During inference, we sample across all points, yielding a dense point cloud from the semantic results. This is subsequently post-processed to generate the final mesh.
  • Figure 5: Left:breath-hold mode, right: free breathing mode.
  • ...and 2 more figures