Dehazing Ultrasound Using Diffusion Models.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Echocardiography is crucial for cardiac diagnosis but often suffers from poor image quality due to acoustic clutter (haze).
• Haze, caused by multipath reflections, significantly impedes accurate diagnosis, particularly in difficult-to-image patients (e.g., obese individuals).
• Existing methods like tissue harmonic imaging and traditional denoising algorithms are insufficient for severe haze removal.

Purpose of the Study:

• To develop an advanced method for removing structured acoustic clutter (haze) from cardiac ultrasound images.
• To improve the diagnostic utility of B-Mode ultrasound imaging, especially for challenging patient populations.
• To leverage the power of diffusion models for unsupervised noise modeling and removal in ultrasound data.

Main Methods:

• A joint posterior sampling framework combining two diffusion models was developed to model clean ultrasound and haze distributions.
• Techniques for training diffusion models on radio-frequency (RF) ultrasound data were explored, highlighting advantages over image data.
• The framework was trained in an unsupervised manner on both in-vitro and in-vivo cardiac ultrasound datasets.

Main Results:

• The proposed dehazing method effectively removed haze from cardiac ultrasound images.
• The technique successfully preserved signals from weakly reflected tissues, maintaining diagnostic information.
• Experiments demonstrated the efficacy of the diffusion model approach on both in-vitro and in-vivo data.

Conclusions:

• The novel diffusion model framework offers a powerful solution for haze removal in cardiac ultrasound imaging.
• This approach significantly enhances image quality, potentially reducing the need for alternative imaging modalities or contrast agents.
• The study underscores the potential of RF ultrasound data and diffusion models for advancing cardiac ultrasound diagnostics.