Vascular networks dataset generation used in deep learning methods for enhancement of photoacoustic images
Photoacoustic imaging (PAI) has become a popular biomedical imaging method in the last few years. Likewise, machine/deep learning methods have found many applications in biomedical systems such as photoacoustic microscopy. The implementation of deep learning in PAI has several applications such as solving the optical and acoustic inverse problems and segmentation of tumors. Meanwhile, the majority of the efforts are directed towards enhancing the quality of images produced by the PAI systems. The training of a deep learning model (architecture) will require a dataset of images. Unlike commercially available imaging systems, such as magnetic resonance imaging (MRI) and computed tomography (CT), PAI systems suffer from lack of known and standard datasets that can be used for deep learning and/or machine learning applications. In this study, we have simulated (generated) vascular networks dataset of photoacoustic microscopy (PAM) images for use in deep learning models. Also, we have tested our dataset with a simple known deep learning model known as the U-Net architecture and deblurred images successfully. The generated dataset can be used in deep learning models for other biomedical vascular imaging systems including photoacoustic tomography (PAT), photoacoustic endoscopy (PAE), and so forth.
File with abstract
Laser and Plasma Research Institute, Shahid Beheshti University, Tehran, Iran