Dehazing Ultrasound using Diffusion Models

Echocardiography has been a prominent tool for the diagnosis of cardiac disease. However, these diagnoses can be heavily impeded by poor image quality. Acoustic clutter emerges due to multipath reflections imposed by layers of skin, subcutaneous fat, and intercostal muscle between the transducer and heart. As a result, haze and other noise artifacts pose a real challenge to cardiac ultrasound imaging. In many cases, especially with difficult-to-image patients such as patients with obesity, a diagnosis from B-Mode ultrasound imaging is effectively rendered unusable, forcing sonographers to resort to contrast-enhanced ultrasound examinations or refer patients to other imaging modalities. Tissue harmonic imaging has been a popular approach to combat haze, but in severe cases is still heavily impacted by haze. Alternatively, denoising algorithms are typically unable to remove highly structured and correlated noise, such as haze. It remains a challenge to accurately describe the statistical properties of structured haze, and develop an inference method to subsequently remove it. Diffusion models have emerged as powerful generative models and have shown their effectiveness in a variety of inverse problems. In this work, we present a joint posterior sampling framework that combines two separate diffusion models to model the distribution of both clean ultrasound and haze in an unsupervised manner. Furthermore, we demonstrate techniques for effectively training diffusion models on radio-frequency ultrasound data and highlight the advantages over image data. Experiments on both in-vitro and in-vivo cardiac datasets show that the proposed dehazing method effectively removes haze while preserving signals from weakly reflected tissue.

Qualitative results on in-vivo cardiac ultrasound data.

Try dragging the slider to compare the input and dehazed images.

Comparison on in-vivo data

Varying dehazing strength

Use the slider to change the dehazing strength through the gamma parameter. The higher the gamma (γ), the stronger the dehazing.

Comparison on phantom data

In-vitro results using the phantom data. The ground truth ultrasound x and haze signals h which are used to construct measurement y are shown on the left. RF-based dehazing results are on the right, for both the baseline methods (BM3D and NCSNv2) and the proposed diffusion method. For the latter, we show posterior solutions for both the ultrasound and haze signals (lower inset plot to the right). For all methods, we show error plots that highlight the difference between ground truth and dehazed images, with the proposed diffusion method notably dropping less signal.

Flick through the images below to see the results on the phantom data. Notice how the proposed method is able to remove the haze while preserving the signal. The baselines either are unable to remove the haze or remove too much signal (see error plots).

Overview of the algorithm

The dehazing diffusion process, where the reverse diffusion trajectory is displayed from left to right for both signal (top) and haze (bottom) in parallel. During each step of the posterior sampling process, data consistency is enforced through the measurement model.

More examples TBA...

[1] A. Fatemi, E. A. R. Berg, and A. Rodriguez-Molares. Studying the Origin of Reverberation Clutter in Echocardiography: In Vitro Experiments and In Vivo Demonstrations. Ultrasound in Medicine & Biology, 2019.
[2] Jahren, Tollef Struksnes and Sornes, Anders Rasmus and Denarie, Bastien and Steen, Erik and Bjastad, Tore and Solberg, Anne H Schistad. Reverberation Suppression in Echocardiography Using a Causal Convolutional Neural Network. IEEE, 2023
[3] Stevens, Tristan S. W. and van Gorp, Hans and Meral, Faik C. and Shin, Junseob and Yu, Jason and Robert, Jean-Luc and van Sloun, Ruud J. G. Removing Structured Noise with Diffusion Models. ArXiv, 2023.

BibTeX

@article{stevens2023dehazing,
  title={Dehazing Ultrasound using Diffusion Models},
  author={Stevens, Tristan SW and Meral, Faik C and Yu, Jason and Apostolakis, Iason Z and Robert, Jean-Luc and van Sloun, Ruud JG},
  journal={arXiv preprint arXiv:2307.11204},
  year={2023}
}

Dehazing Ultrasound using Diffusion Models

Abstract