Enhancement of the depth of field of photoacoustic microscopy based on deep learning

Abstract

As a noninvasive biomedical imaging modality, photoacoustic imaging has shown great potential for clinical application recently with simultaneous high contrast and penetration depth. Optical-resolution photoacoustic microscopy is one important branch of photoacoustic imaging, which can achieve high spital resolution. However, this technique suffers from limited depth of field due to the strongly focused Gaussian beam used. With the development of deep learning in various medical imaging techniques, deep learning also has had a significant impact on the field of photoacoustic imaging in recent years. This paper presents a novel method to improve the depth of field of photoacoustic microscope by integrating the Unet semantic segmentation model with the simulation platform of photoacoustic microscopy based on k-Wave. First, we imaged the blood vessels on the simulation platform to obtain the vascular slice B-scan images and the corresponding ground truth images, and then the dataset was randomly divided into the training dataset and the testing dataset in a ratio of 7:1. During the U-Net model training process, the B-scan images serve as the input to the model while corresponding ground truth images serve as the labels. Finally, this study demonstrates the potential of the U-Net model to improve the depth of field of photoacoustic imaging. And this method effectively improves the accuracy of obtaining structural features of tissues, which has significant implications for the diagnosis and treatment of diseases.