Convolutional sparse coding for compressed sensing photoacoustic CT reconstruction with partially known support.

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Signal Processing

Background:

• Photoacoustic tomography (PAT) imaging speed is limited by laser repetition rate and data acquisition card (DAQ) channels.
• Reconstructing images with fewer elements reduces hardware costs but causes artifacts due to undersampling.
• Existing dictionary learning methods overlook pixel uniformity in overlapping blocks.

Purpose of the Study:

• To develop a compressive sensing (CS) reconstruction algorithm for circular array PAT.
• To address image quality degradation caused by undersampling in PAT.
• To reduce hardware costs and improve imaging speed in PAT.

Main Methods:

• Proposed a gradient domain convolutional sparse coding (CSCGR) algorithm for PAT.
• Utilized partially known support (PKS) from sparsely encoded feature maps.
• Implemented CS-CSCGR-PKS for reconstruction with reduced ultrasound transducer elements.

Main Results:

• The CS-CSCGR-PKS algorithm enables signal acquisition with fewer transducers while maintaining image fidelity.
• Demonstrated effective sparse imaging in experiments on mouse torso, brain, and human fingers.
• Showcased significant reduction in equipment hardware costs and improved imaging speed.

Conclusions:

• The proposed CS-CSCGR-PKS algorithm effectively reduces the number of array elements in PAT without compromising image quality.
• This method offers a cost-effective and faster alternative for PAT systems.
• Sparse imaging in PAT can be achieved with reduced hardware, enhancing practical applications.