딥러닝 기반 기능적 생체 의학 영상 가속화 기법

Abstract

Functional imaging modalities have become essential in preclinical and clinical studies to unveil physiological activities within living organisms. However, a major technical drawback is that a tradeoff between imaging speed and quality is often inevitable in such systems. Although studies have focused on developing advanced hardware systems to improve both the temporal and spatial resolutions, such adjustments are usually cost-prohibitive and may pose limitations for practical clinical system implementations. Recently, deep learning (DL)-based methods have gained attention in improving the temporal resolution for a wide range of biomedical imaging systems by reconstructing high-quality images from spatially or temporally undersampled data. Motivated by the previous studies, we present deep learning-based methods to enhance the temporal resolution for functional biomedical imaging modalities by reconstructing dense information from sparsely sampled data. We present two application studies of optical coherence tomography angiography (OCTA) and localization-based superresolution photoacoustic computed tomography (PACT). The results demonstrate that our method can accelerate the imaging modalities by ten to a few hundred folds while preserving spatial information. Our DL framework can thus provide a convenient software-only solution to enhance preclinical and clinical studies.