Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Blind Restoration of High-Resolution Ultrasound Video

Chu Chen, Kangning Cui, Pasquale Cascarano, Wei Tang, Elena Loli Piccolomini, Raymond H. Chan

TL;DR

This paper addresses the challenge of restoring high-resolution ultrasound videos without paired HR/LR data. It introduces Deep Ultrasound Prior (DUP), a self-supervised, frame-wise optimization framework that leverages a CNN prior with Weight Inheritance and dual regularization (Total Variation and higher-order terms) to produce SR frames from noisy LR US videos. Through experiments on EchoNet-LVH and EchoNet-Dynamic, DUP achieves superior PSNR/SSIM and robustness to noise, while also improving downstream cardiac EF predictions using EchoCLIP in a zero-shot setting. The work demonstrates the practical impact of blind, video-specific restoration for clinical ultrasound analysis and EF-related diagnostics.

Abstract

Ultrasound imaging is widely applied in clinical practice, yet ultrasound videos often suffer from low signal-to-noise ratios (SNR) and limited resolutions, posing challenges for diagnosis and analysis. Variations in equipment and acquisition settings can further exacerbate differences in data distribution and noise levels, reducing the generalizability of pre-trained models. This work presents a self-supervised ultrasound video super-resolution algorithm called Deep Ultrasound Prior (DUP). DUP employs a video-adaptive optimization process of a neural network that enhances the resolution of given ultrasound videos without requiring paired training data while simultaneously removing noise. Quantitative and visual evaluations demonstrate that DUP outperforms existing super-resolution algorithms, leading to substantial improvements for downstream applications.

Blind Restoration of High-Resolution Ultrasound Video

TL;DR

This paper addresses the challenge of restoring high-resolution ultrasound videos without paired HR/LR data. It introduces Deep Ultrasound Prior (DUP), a self-supervised, frame-wise optimization framework that leverages a CNN prior with Weight Inheritance and dual regularization (Total Variation and higher-order terms) to produce SR frames from noisy LR US videos. Through experiments on EchoNet-LVH and EchoNet-Dynamic, DUP achieves superior PSNR/SSIM and robustness to noise, while also improving downstream cardiac EF predictions using EchoCLIP in a zero-shot setting. The work demonstrates the practical impact of blind, video-specific restoration for clinical ultrasound analysis and EF-related diagnostics.

Abstract

Ultrasound imaging is widely applied in clinical practice, yet ultrasound videos often suffer from low signal-to-noise ratios (SNR) and limited resolutions, posing challenges for diagnosis and analysis. Variations in equipment and acquisition settings can further exacerbate differences in data distribution and noise levels, reducing the generalizability of pre-trained models. This work presents a self-supervised ultrasound video super-resolution algorithm called Deep Ultrasound Prior (DUP). DUP employs a video-adaptive optimization process of a neural network that enhances the resolution of given ultrasound videos without requiring paired training data while simultaneously removing noise. Quantitative and visual evaluations demonstrate that DUP outperforms existing super-resolution algorithms, leading to substantial improvements for downstream applications.

Paper Structure

This paper contains 16 sections, 5 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Works.
  3. Contributions.
  4. Method
  5. Problem Formulation
  6. Deep Ultrasound Prior
  7. WIn Strategy
  8. Early Stopping and Input Update
  9. Experiments
  10. Datasets
  11. Implement Details
  12. Image-based Evaluations
  13. Downstream Evaluations
  14. Conclusion
  15. Acknowledgments.
  16. ...and 1 more sections

Figures (4)

  • Figure 1: Schematic of the DUP algorithm.
  • Figure 2: Visual comparisons of $4\times$ SR results. The paired HR and LR example frames are shown on the far left. The upper row displays the SR frames from various methods, with quantitative results (PSNR (dB)/SSIM) shown underneath, while the lower row is the corresponding error maps.
  • Figure 3: Quantitative evaluations on $4\times$SR videos corrupted by multi-level noise.
  • Figure 4: Visualization of maximum diastole (upper) and minimum systole (lower) detected within cardiac cycles.