HM-3DCE-Net for Superior 3D Photoacoustic Imaging Enhancement and Segmentation

Abstract

Photoacoustic tomography is a highly sensitive modality for imaging optical absorption contrast in biological tissues over a range of spatial scales at high speed. Over the past decade, the quality of photoacoustic imaging has improved by multiple folds, but it still cannot meet the needs of clinical application. One main challenge is related to the 3D PAI data set where substantial variance in signal intensity and signal-to-noise ratio exists at different depths, making it difficult to automate the data analysis. Here we propose a Hessian matrix-based three-dimensional context encoder network to analysis of 3D data set. With superior generalization ability, 3D PAI in humans and animals were simultaneously enhancement and segmentation with significantly improved. Furthermore, wide-field and ultra-dense exogenous 3D PAI of mouse brain vasculature were enhanced and segmented for the first time.  Therefore, we believe the proposed technique would enable new clinical application and basic research in photoacoustic imaging.