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

Area of Science:

• Biomedical Imaging
• Optical Imaging
• Ultrasound Technology

Background:

• Photoacoustic computed tomography (PACT) offers combined optical contrast and ultrasound detection advantages.
• Larger ultrasound transducers enhance sensitivity but degrade resolution in analytic reconstruction methods.
• Optimization-based methods improve accuracy but are computationally intensive, especially in 3D.

Purpose of the Study:

• To develop a rapid and accurate 3D PACT image reconstruction framework.
• To introduce a learned spatial impulse response (SIR) compensation method operating in the data domain.
• To enable computationally efficient reconstruction by compensating for transducer SIR effects.

Main Methods:

• A data-domain learned compensation framework was established, mapping corrupted PACT data to compensated data.
• Two compensation models, U-Net and Deconv-Net, were investigated.
• A fast, analytical training data generation procedure was developed and integrated.

Main Results:

• The framework demonstrated improved resolution and robustness to noise, complexity, and sound speed variations in virtual studies.
• Learned compensation models successfully revealed fine structures obscured by SIR artifacts in in-vivo breast imaging.
• This represents the first demonstration of learned SIR compensation for 3D PACT.

Conclusions:

• The proposed learned SIR compensation framework significantly enhances 3D PACT image quality and reconstruction speed.
• This approach overcomes limitations of traditional methods, offering a practical solution for advanced biomedical imaging.
• The method shows great potential for clinical applications, particularly in breast imaging.