Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

UltrasODM: A Dual Stream Optical Flow Mamba Network for 3D Freehand Ultrasound Reconstruction

Mayank Anand, Ujair Alam, Surya Prakash, Priya Shukla, Gora Chand Nandi, Domenec Puig

TL;DR

UltrasODM addresses drift and reliability issues in 3D freehand ultrasound reconstruction by coupling a dual-stream motion model with a contrastive frame embedding and a Dual Mamba temporal encoder. A HITL safety loop leverages Monte Carlo uncertainty, saliency explanations, and clinician-guided prompts to guide reacquisition, aiming for safer and more trustworthy clinical workflows. The approach yields reductions in drift, distance error, and Hausdorff distance on a clinical freehand dataset, while providing per-frame uncertainty and explainability. The work emphasizes system-level integration, real-time considerations, and ethical compliance, and releases public code to facilitate adoption and benchmarking in healthcare AI.

Abstract

Clinical ultrasound acquisition is highly operator-dependent, where rapid probe motion and brightness fluctuations often lead to reconstruction errors that reduce trust and clinical utility. We present UltrasODM, a dual-stream framework that assists sonographers during acquisition through calibrated per-frame uncertainty, saliency-based diagnostics, and actionable prompts. UltrasODM integrates (i) a contrastive ranking module that groups frames by motion similarity, (ii) an optical-flow stream fused with Dual-Mamba temporal modules for robust 6-DoF pose estimation, and (iii) a Human-in-the-Loop (HITL) layer combining Bayesian uncertainty, clinician-calibrated thresholds, and saliency maps highlighting regions of low confidence. When uncertainty exceeds the threshold, the system issues unobtrusive alerts suggesting corrective actions such as re-scanning highlighted regions or slowing the sweep. Evaluated on a clinical freehand ultrasound dataset, UltrasODM reduces drift by 15.2%, distance error by 12.1%, and Hausdorff distance by 10.1% relative to UltrasOM, while producing per-frame uncertainty and saliency outputs. By emphasizing transparency and clinician feedback, UltrasODM improves reconstruction reliability and supports safer, more trustworthy clinical workflows. Our code is publicly available at https://github.com/AnandMayank/UltrasODM.

UltrasODM: A Dual Stream Optical Flow Mamba Network for 3D Freehand Ultrasound Reconstruction

TL;DR

UltrasODM addresses drift and reliability issues in 3D freehand ultrasound reconstruction by coupling a dual-stream motion model with a contrastive frame embedding and a Dual Mamba temporal encoder. A HITL safety loop leverages Monte Carlo uncertainty, saliency explanations, and clinician-guided prompts to guide reacquisition, aiming for safer and more trustworthy clinical workflows. The approach yields reductions in drift, distance error, and Hausdorff distance on a clinical freehand dataset, while providing per-frame uncertainty and explainability. The work emphasizes system-level integration, real-time considerations, and ethical compliance, and releases public code to facilitate adoption and benchmarking in healthcare AI.

Abstract

Clinical ultrasound acquisition is highly operator-dependent, where rapid probe motion and brightness fluctuations often lead to reconstruction errors that reduce trust and clinical utility. We present UltrasODM, a dual-stream framework that assists sonographers during acquisition through calibrated per-frame uncertainty, saliency-based diagnostics, and actionable prompts. UltrasODM integrates (i) a contrastive ranking module that groups frames by motion similarity, (ii) an optical-flow stream fused with Dual-Mamba temporal modules for robust 6-DoF pose estimation, and (iii) a Human-in-the-Loop (HITL) layer combining Bayesian uncertainty, clinician-calibrated thresholds, and saliency maps highlighting regions of low confidence. When uncertainty exceeds the threshold, the system issues unobtrusive alerts suggesting corrective actions such as re-scanning highlighted regions or slowing the sweep. Evaluated on a clinical freehand ultrasound dataset, UltrasODM reduces drift by 15.2%, distance error by 12.1%, and Hausdorff distance by 10.1% relative to UltrasOM, while producing per-frame uncertainty and saliency outputs. By emphasizing transparency and clinician feedback, UltrasODM improves reconstruction reliability and supports safer, more trustworthy clinical workflows. Our code is publicly available at https://github.com/AnandMayank/UltrasODM.

Paper Structure

This paper contains 22 sections, 18 equations, 3 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Method
  3. Inner Mamba: Local Causal Encoding.
  4. Dual Mamba: Global Aggregation.
  5. Human-in-the-Loop Clinical Integration
  6. Uncertainty Quantification.
  7. Uncertainty-Aware Safety Gate.
  8. Saliency-Based Visual Explanations.
  9. Actionable Operator Feedback.
  10. Closed-Loop Reacquisition.
  11. User-Friendly Interface Design Principles
  12. Real-Time Deployment Considerations.
  13. Ethical and Clinical Compliance.
  14. Results
  15. Quantitative Comparison with Prior Work
  16. ...and 7 more sections

Figures (3)

  • Figure 1: UltrasODM network architecture.
  • Figure 2: Visual Comparison of 3D Trajectory Reconstruction Results. The figure shows the reconstructed 3D trajectories (red dashed lines) overlaid on the Ground Truth (GT) trajectories (green dashed lines) for multiple sequences. The visualization compares three methods: Left: Baseline method (EfficientNet), Center: Optical Flow method, and Right: Our proposed method (Optical Flow + Mamba Comparison/UltrasODM).
  • Figure 3: UltrasODM Human-in-the-Loop (HITL) clinical safety loop. Incoming ultrasound frames are processed by the Optical Flow + Dual-Mamba model, followed by Monte-Carlo uncertainty estimation. A two-threshold safety gate classifies each frame into safe, caution, or critical levels. Caution/critical states trigger saliency-based visualization and actionable operator prompts, forming a closed feedback loop for corrected reacquisition.