[Paper Review] Compressive Deconvolution in Medical Ultrasound Imaging

Zhouye Chen, Adrian Basarab+1 more|arXiv (Cornell University)|Jul 1, 2015

Sparse and Compressive Sensing Techniques54 references62 citations

TL;DR

This paper proposes a novel compressive deconvolution framework for medical ultrasound imaging that jointly reduces data volume and enhances image quality by combining compressive sampling with deconvolution. Using an alternating direction method of multipliers (ADMM) optimization, the method reconstructs tissue reflectivity functions from compressed measurements while enforcing sparsity in the transformed domain and generalized Gaussian priors on the reflectivity, achieving superior resolution and SNR compared to existing methods.

ABSTRACT

The interest of compressive sampling in ultrasound imaging has been recently extensively evaluated by several research teams. Following the different application setups, it has been shown that the RF data may be reconstructed from a small number of measurements and/or using a reduced number of ultrasound pulse emissions. Nevertheless, RF image spatial resolution, contrast and signal to noise ratio are affected by the limited bandwidth of the imaging transducer and the physical phenomenon related to US wave propagation. To overcome these limitations, several deconvolution-based image processing techniques have been proposed to enhance the ultrasound images. In this paper, we propose a novel framework, named compressive deconvolution, that reconstructs enhanced RF images from compressed measurements. Exploiting an unified formulation of the direct acquisition model, combining random projections and 2D convolution with a spatially invariant point spread function, the benefit of our approach is the joint data volume reduction and image quality improvement. The proposed optimization method, based on the Alternating Direction Method of Multipliers, is evaluated on both simulated and in vivo data.

Motivation & Objective

• To address the limitations of conventional ultrasound imaging, such as reduced spatial resolution, contrast, and SNR due to transducer bandwidth and wave propagation effects.
• To overcome the trade-off between data volume and image quality by integrating compressive sampling (CS) and deconvolution into a unified framework.
• To develop a novel optimization method that jointly reconstructs high-quality RF images from compressed measurements while improving resolution and noise robustness.
• To validate the method on both simulated and in vivo ultrasound data, demonstrating its superiority over existing CS and deconvolution techniques.

Proposed method

• Formulates the compressive deconvolution problem as a convex optimization problem: y = ΦHx + n, where y is compressed measurements, Φ is the CS matrix, H is the convolution operator with a spatially invariant PSF, and x is the tissue reflectivity function.
• Employs the Alternating Direction Method of Multipliers (ADMM) to solve the optimization problem, alternating between updating the sparse representation a and the reflectivity function x.
• Imposes sparsity on the RF image r = Ψa in a transformed domain (e.g., wavelet or Fourier) via an ℓ1-norm regularization on a.
• Applies a generalized Gaussian distribution (GGD) prior on x via an ℓp-norm regularization (1 ≤ p ≤ 2) to model tissue reflectivity.
• Estimates the PSF from RF data using the method in [35] as a preprocessing step to construct the matrix H.
• Uses a two-step ADMM scheme: first recover the blurred image r from compressed data, then deconvolve to estimate x.

Experimental results

Research questions

• RQ1 Can compressive sampling and deconvolution be jointly optimized to reduce data volume while improving ultrasound image quality?
• RQ2 How does the proposed compressive deconvolution framework compare to standard CS or deconvolution alone in terms of resolution, SNR, and reconstruction accuracy?
• RQ3 What is the impact of using a generalized Gaussian prior (ℓp-norm) on the tissue reflectivity function compared to standard sparsity priors?
• RQ4 How does the ADMM-based optimization perform in terms of convergence speed and computational efficiency on real and simulated ultrasound data?

Key findings

• The proposed method achieves a PSNR improvement of 0.5 to 2 dB over the state-of-the-art method in [26] across all tested CS ratios (20% to 80%) and SNR levels (30 dB and 40 dB).
• At 40 dB SNR and 80% CS ratio, the proposed method achieves a PSNR of 26.91 dB, outperforming CD Amizic’s 25.51 dB.
• The method reduces reconstruction time compared to CD Amizic, with faster convergence across all CS ratios, as shown in computational time comparisons.
• The method successfully enhances image resolution and contrast by jointly exploiting compressed measurements and deconvolution, even in the presence of noise.
• The use of generalized Gaussian priors (ℓp-norm) for reflectivity modeling leads to better reconstruction fidelity than standard ℓ1 or total variation priors.
• The framework demonstrates robust performance on both simulated Shepp-Logan phantoms and in vivo ultrasound data, confirming its practical applicability.