Ultrasound Imaging based on the Variance of a Diffusion Restoration Model

TL;DR

Ultrasound imaging is limited by speckle and artefacts that degrade SNR and contrast. The authors propose a hybrid physics-informed diffusion reconstruction framework that uses a pre-trained diffusion restoration model (DDRMs) as a prior, with unsupervised fine-tuning for ultrasound, and a diffusion-variance imaging modality (DRUSvar) to estimate echogenicity maps from posterior variance under multiplicative noise. They derive an empirical stochastic model for diffusion reconstructions and demonstrate that DRUSvar yields despeckled reconstructions with higher SNR and contrast than beamforming, diffusion-denoising baselines, and despeckling approaches across synthetic phantoms and PICMUS datasets (in vitro and in vivo). The approach enables robust echogenicity mapping from single plane-wave acquisitions and is compatible with unsupervised model adaptation, offering a practical tool for ultrasound echogenicity assessment and advancing diffusion-based priors in inverse problems.

Abstract

Despite today's prevalence of ultrasound imaging in medicine, ultrasound signal-to-noise ratio is still affected by several sources of noise and artefacts. Moreover, enhancing ultrasound image quality involves balancing concurrent factors like contrast, resolution, and speckle preservation. Recently, there has been progress in both model-based and learning-based approaches addressing the problem of ultrasound image reconstruction. Bringing the best from both worlds, we propose a hybrid reconstruction method combining an ultrasound linear direct model with a learning-based prior coming from a generative Denoising Diffusion model. More specifically, we rely on the unsupervised fine-tuning of a pre-trained Denoising Diffusion Restoration Model (DDRM). Given the nature of multiplicative noise inherent to ultrasound, this paper proposes an empirical model to characterize the stochasticity of diffusion reconstruction of ultrasound images, and shows the interest of its variance as an echogenicity map estimator. We conduct experiments on synthetic, in-vitro, and in-vivo data, demonstrating the efficacy of our variance imaging approach in achieving high-quality image reconstructions from single plane-wave acquisitions and in comparison to state-of-the-art methods. The code is available at: https://github.com/Yuxin-Zhang-Jasmine/DRUSvar

Figures (5)

• Figure 1: Echogenicity maps of the synthetic occlusion (left) and scatterer (right) phantoms. Metrics are calculated within the colored boundary regions.
• Figure 2: Quantitative and qualitative comparison of the synthetic occlusion phantom-based images under varying levels of additive noise. Images are in decibels with a dynamic range [-60,0]. Left $std({\mathbf{n}}\xspace)=0.02$, right $std({\mathbf{n}}\xspace)=0.08$.
• Figure 3: Quantitative and qualitative comparison of the synthetic scatterer phantom-based images under varying levels of additive noise. Images are in decibels with a dynamic range [-60,0]. Left $std({\mathbf{n}}\xspace)=0.018$, right $std({\mathbf{n}}\xspace)=0.1$.
• Figure 4: Comparison of reconstructed images on the PICMUS in vitro (row [1-2]) and in vivo (row [3-4]) datasets. All images are in decibels with a dynamic range [-60,0]. The colored boundaries outline the regions where the in vitro evaluation metrics are calculated.
• Figure 5: Visual comparison of despeckled images on the CC dataset, in decibels [-60,0] dynamic range.