Abstract:Despite its wide use in medicine, ultrasound imaging faces several challenges related to its poor signal-to-noise ratio and several sources of noise and artefacts. Enhancing ultrasound image quality involves balancing concurrent factors like contrast, resolution, and speckle preservation. In recent years, there has been progress both in model-based and learning-based approaches to improve ultrasound image reconstruction. Bringing the best from both worlds, we propose a hybrid approach leveraging advances in diffusion models. To this end, we adapt Denoising Diffusion Restoration Models (DDRM) to incorporate ultrasound physics through a linear direct model and an unsupervised fine-tuning of the prior diffusion model. We conduct comprehensive experiments on simulated, in-vitro, and in-vivo data, demonstrating the efficacy of our approach in achieving high-quality image reconstructions from a single plane wave input and in comparison to state-of-the-art methods. Finally, given the stochastic nature of the method, we analyse in depth the statistical properties of single and multiple-sample reconstructions, experimentally show the informativeness of their variance, and provide an empirical model relating this behaviour to speckle noise. The code and data are available at: (upon acceptance).

What problem does this paper attempt to address?

### Problems the Paper Attempts to Solve This paper aims to address some key challenges in ultrasound imaging, particularly its poor signal-to-noise ratio (SNR) and various sources of noise and artifacts. Specifically, the paper focuses on the following aspects: 1. **Improving Image Quality**: Enhancing the quality of ultrasound images requires balancing contrast, resolution, and speckle preservation. 2. **High-Quality Reconstruction from Single Plane Wave Input**: The paper proposes a hybrid approach that leverages the advantages of Diffusion Models to achieve high-quality image reconstruction from single plane wave input. 3. **Statistical Uncertainty Analysis**: Due to the stochastic nature of the method, the paper delves into the statistical properties of single-sample and multi-sample reconstructions, demonstrating the usefulness of their variance, particularly in despeckling and improving SNR and contrast. ### Main Contributions 1. **Adaptation of Diffusion Reconstruction Model (DDRM)**: Extending DDRM from natural image restoration tasks (such as denoising, inpainting, super-resolution) to the task of reconstructing B-mode ultrasound images from raw radio frequency (RF) channel data. 2. **Quantitative Evaluation**: Extensive quantitative evaluation on the PICMUS dataset and self-collected in vivo datasets, showing performance improvements over traditional DAS, state-of-the-art beamforming methods, and recently proposed methods using diffusion models as denoisers. 3. **Multi-Sample Variance Analysis**: For the first time, revealing that despeckling can be achieved by computing the variance of multiple DRUS samples, generating images with higher SNR and contrast without sacrificing resolution. ### Method Overview 1. **Forward Ultrasound Imaging Model**: Describes the linear time-invariant system model of ultrasound imaging, including transmit pulse, received signal, and propagation time. 2. **From Classical Beamforming to Inverse Problem**: Introduces the traditional DAS algorithm and its application in beamforming, linking it to the inverse problem. 3. **Diffusion Restoration Model (DDRM)**: Utilizes a pre-trained diffusion model as prior knowledge, transforming the inverse problem into a denoising problem through a linear observation model. 4. **Diffusion Reconstruction of Ultrasound Images**: Combines the physical characteristics and constraints of ultrasound imaging, reduces problem dimensions through projection data, and performs iterative reconstruction using DDRM. ### Experimental Validation 1. **Phantom Performance Evaluation**: Evaluates the convergence and performance of Deno and DRUS through experiments with different iteration steps. 2. **Statistical Behavior Analysis**: Demonstrates the generative uncertainty and speckle preservation capability of the method through multi-sample analysis. 3. **Multi-Sample Imaging**: Further explores the positive correlation between variance and pixel value change rate, showcasing its advantages in improving image quality and despeckling through multi-sample reconstruction. Overall, this paper proposes a novel ultrasound image reconstruction technique by combining model-driven and data-driven approaches, not only improving image quality but also providing an in-depth analysis of uncertainty.

Diffusion Reconstruction of Ultrasound Images with Informative Uncertainty

Ultrasound Imaging based on the Variance of a Diffusion Restoration Model

Ultrasound Image Reconstruction with Denoising Diffusion Restoration Models

Ultrasound Image Enhancement with the Variance of Diffusion Models

Denoising Plane Wave Ultrasound Images Using Diffusion Probabilistic Models

Dehazing Ultrasound using Diffusion Models

Deep Ultrasound Denoising Using Diffusion Probabilistic Models

Diffusion as Sound Propagation: Physics-inspired Model for Ultrasound Image Generation

Super-resolution reconstruction of ultrasound image using a modified diffusion model

Fast Sampling generative model for Ultrasound image reconstruction

Toward Smart Ultrasound Image Augmentation to Advance Tumor Treatment Monitoring: Exploring the Potential of Diffusion Generative Model

Minutia reconstruction in sonar images with diffusion probabilistic models

Synomaly Noise and Multi-Stage Diffusion: A Novel Approach for Unsupervised Anomaly Detection in Ultrasound Imaging

General Intelligent Imaging and Uncertainty Quantification by Deterministic Diffusion Model

An Efficient Multi-Scale Wavelet Approach for Dehazing and Denoising Ultrasound Images Using Fractional-Order Filtering

A Nonlinear Diffusion Equation-Based Model for Ultrasound Speckle Noise Removal

Controllable Deep Learning Denoising Model for Ultrasound Images Using Synthetic Noisy Image.

PHOCUS: Physics-Based Deconvolution for Ultrasound Resolution Enhancement

Ultrasound Image Denoising Based on the Contourlet Transform and Anisotropic Diffusion

Sparse-view reconstruction for photoacoustic tomography combining diffusion model with model-based iteration