Attenuation Tomography Using Low-Frequency Ultrasound With Variational Autoencoder for Thorax Imaging: Experimental...

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Ultrasound Technology

Background:

• Accurate imaging of the human thorax is crucial for respiratory system monitoring.
• Traditional methods face challenges with ill-posed inverse problems and noise.
• Low-frequency ultrasound offers potential for functional imaging but requires robust inversion techniques.

Purpose of the Study:

• To develop a novel inversion methodology for human thorax attenuation tomography using low-frequency ultrasound.
• To leverage variational autoencoders (VAE) for improved image reconstruction by incorporating structural priors.
• To validate the feasibility and effectiveness of this VAE-based approach in numerical simulations and human subject experiments.

Main Methods:

• A variational autoencoder (VAE) was trained using computed tomography (CT) derived data to learn human thorax structural priors.
• High-dimensional attenuation distributions were compressed into a lower-dimensional latent space.
• The VAE's decoder was used to reconstruct the attenuation model by optimizing the latent code during inversion.
• The VAE-based method was compared against the conventional pixel-based Gauss-Newton method.

Main Results:

• The VAE-based inversion methodology demonstrated robust generalization and noise resilience in numerical simulations.
• The proposed method outperformed the conventional Gauss-Newton method in simulation results.
• Human subject experiments confirmed the effectiveness of the VAE approach for low-frequency ultrasound thorax imaging.
• This study provides the first experimental validation for low-frequency ultrasound functional imaging of the human thorax.

Conclusions:

• The VAE-based inversion technique effectively integrates prior information, mitigating the ill-posedness of inverse problems in ultrasound tomography.
• The method shows significant potential for accurate and robust functional imaging of the human thorax using low-frequency ultrasound.
• This approach holds promise for continuous monitoring of the human respiratory system.