Dictionary learning sparse-sampling reconstruction method for in-vivo 3D photoacoustic computed tomography.

Area of Science:

• Biomedical Imaging
• Medical Physics
• Computational Imaging

Background:

• Current model-based reconstruction in photoacoustic computed tomography (PACT) uses predefined sparse transforms.
• These predefined transforms inadequately capture specific data features, limiting photoacoustic image quality.
• This necessitates advanced reconstruction techniques for high-fidelity PACT imaging.

Purpose of the Study:

• To develop and validate an advanced PACT reconstruction model using K-VSD dictionary learning.
• To adapt and evaluate this model within a 3D PACT system for *in vivo* imaging.
• To demonstrate improvements in image accuracy, contrast-to-noise ratio, and imaging speed.

Main Methods:

• Implementation of a K-VSD dictionary learning technique for sparse representation in PACT reconstruction.
• Adaptation of the model for a 3D PACT system.
• Conducting *in vivo* experiments on human hand and rat models with sparse-sampling (50%).

Main Results:

• The K-VSD method improved reconstructed photoacoustic image accuracy by an average of 3.7 times compared to traditional sparse transforms at 50% sparse-sampling.
• Contrast-to-noise ratio was enhanced by an average of 1.8 times under the same conditions.
• Imaging speed was 60% faster than other tested reconstruction approaches.

Conclusions:

• The proposed K-VSD dictionary learning model offers a superior method for PACT image reconstruction.
• This technique enhances image quality metrics and imaging speed, crucial for clinical translation.
• The approach facilitates the development of high-speed, low-cost 3D PACT systems for diverse biomedical applications.