Enhancement of photoacoustic imaging systems for biomedical applications

Photoacoustic imaging (PAI) is a hybrid biomedical imaging modality that combines the advantages of optical illumination and acoustic detection to provide label-free, high resolution images. Among different embodiments of PAI, photoacoustic tomography/ photoacoustic computed tomography (PAT/PACT) is widely used for in vivo studies which require deep tissue imaging. Alternatively, photoacoustic microscopy (PAM) is preferred to attain high resolution images at the cost of decreased imaging depth. Due to its advantages like good resolution, high contrast, label-free imaging, deep imaging depth, and the ability to perform multiscale structural and functional imaging, PAI has grown exponentially in the last two decades. However, there still exist some limitations in the current imaging systems in terms of cost, size, acquisitions time, and image quality, that need to be addressed. Therefore, further improvement and optimization of imaging systems is needed so that PAI can be utilized to its highest potential for pre-clinical and clinical applications. In a typical PAT system, photoacoustic (PA) waves are recorded using an ultrasound transducer rotating around the sample. Among different system configurations, being economical and easily available, single element transducer (SET) based PAT systems are commonly used for biomedical imaging. In these systems, the SET acquires one time resolved PA signal (A-line) for each laser pulse. Cross sectional images are obtained by rotating the SET around the sample and acquiring A-lines at different angles. Acquisition of A-lines can be done either by stop-and-go or by continuous scanning method. By comparing the different types of scanning methods, it was exhibited how the imaging speed of SET-based PAT systems can be improved without compromising the image quality. Mathematical calculations and PA experiments were conducted to show that a 2-4 folds and a 7-12 folds improvement in scan time can be achieved for lasers with low (10 Hz) and high (7 kHz) pulse repetition rate, respectively, while maintaining similar signal to noise ratio, spatial accuracy, and spatial resolution in the images. Following this, improvement in imaging depth for different PA imaging systems was proposed. Although, PAI provides higher imaging depth compared to optical imaging modalities, depth of PA imaging systems can further be improved by using near-infrared (NIR) waves for illuminating the sample, instead of visible light. By providing lower optical absorption and scattering in biological tissues, and higher maximum permissible exposure (MPE), irradiation by laser pulses in this range can enhance the depth of imaging for PAI. Using Monte Carlo simulations, imaging depth after irradiation in visible, NIR-I, and NIR-II regions were compared. Both PAT and acoustic resolution PAM (AR-PAM) experiments were conducted to validate the simulation results. The results demonstrated that improvement in imaging depth in breast tissue by NIR-II waves is due to the increased MPE in this region. To further enhance PA images, deep learning (DL) method was suggested to improve the out-of-focus resolution of AR-PAM systems. DL relies on supervised learning, i.e., the network gets trained on pre-defined dataset to solve practical problems on new data. In the last decade, there has been a surge in using DL al- gorithms for biomedical imaging. A fully dense U-net based Convolutional neural network (CNN) architecture was trained on simulated AR-PAM images and tested on phantom as well as in vivo experimental images. Applying the trained model on experimental images showed that the variation in resolution is