A Learning-based Framework for Spatial Impulse Response Compensation in 3D Photoacoustic Computed Tomography

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Computational Imaging

Background:

• Photoacoustic computed tomography (PACT) offers molecular contrast and high spatial resolution.
• Larger ultrasound transducers improve sensitivity but analytic reconstruction neglecting spatial impulse responses (SIRs) compromises resolution.
• Optimization-based reconstruction methods account for SIRs but are computationally expensive, especially in 3D.

Purpose of the Study:

• To develop a rapid and accurate 3D PACT image reconstruction framework using learned data-domain SIR compensation.
• To enable efficient reconstruction by mapping SIR-corrupted data to data from idealized point-like transducers.
• To validate the proposed method in virtual and in-vivo imaging scenarios.

Main Methods:

• A learned data-domain SIR compensation framework was developed, including a purely data-driven model and a physics-inspired Deconv-Net model.
• Training data was generated using a fast, analytical procedure.
• The compensated data was used with a computationally efficient reconstruction method that neglects SIR effects.

Main Results:

• The framework demonstrated improved resolution and robustness to noise, object complexity, and sound speed heterogeneity in virtual studies.
• Application to in-vivo breast imaging revealed fine structures obscured by SIR artifacts.
• At least a 30% reduction in relative squared error was achieved compared to the baseline.

Conclusions:

• The proposed learned SIR compensation framework enables accurate and rapid 3D PACT image reconstruction.
• This approach effectively mitigates SIR-induced artifacts, enhancing image quality.
• This work represents the first demonstration of learned SIR compensation in 3D PACT imaging.