Anatomical modeling of brain vasculature in two-photon microscopy by generalizable deep learning

Abstract

Segmentation of blood vessels from two-photon microscopy (2PM) angiograms of brains has important applications in hemodynamic analysis and disease diagnosis. Here we develop a generalizable deep-learning technique for accurate 2PM vascular segmentation of sizable regions in mouse brains acquired from multiple 2PM setups. In addition, the technique is computationally efficient, making it ideal for large-scale neurovascular analysis. Introduction: Vascular segmentation from 2PM angiograms is usually an important first step in hemodynamic modeling of brain vasculature. Existing state-of-the-art segmentation methods based on deep learning either lack the ability to generalize to data from various imaging systems, or are computationally infeasible for large-scale angiograms. In this work, we present a method which improves upon both these limitations by being generalizable to various imaging systems, and also being able to segment very large-scale angiograms. Methods: We employ a computationally efficient deep learning framework based on a semi-supervised learning strategy, whose effectiveness we demonstrate on experimentally acquired in-vivo angiograms from mouse brains of dimensions up to 808x808x702 micrometers. Results: After training on data from only one 2PM microscope, we perform vascular segmentation on data from another microscope without any network tuning. Our method demonstrates 10x faster computation in terms of voxels-segmented-persecond and 3x larger depth compared to the state-of-the-art. Conclusion: Our work provides a generalizable and computationally efficient anatomical modeling framework for the brain vasculature, which consists of deeplearning based vascular segmentation followed by graphing. It paves the way for future modeling and analysis of hemodynamic response at much greater scales that were inaccessible before.

Three‐dimensional microvascular network reconstruction from in vivo images with adaptation of the regional inhomogeneity in the signal‐to‐noise ratio

“…With reference to the raw images, adding and deleting a segment, connecting/disconnecting pixels for correction of a segment, and correcting the position of the vessel center points were performed using the custom‐written interactive software in accordance with visual inspection. Other study proposed machine learning methods for automatic segmentation of the 2PLSM angiographic images 51 . If the accuracy of the identification is equivalent to or even exceeds that of the visual inspection, the machine learning methods will allow the fully automated analysis of the imaged microvascular networks captured with 2PLSM.…”

Deep Learning in Biomedical Optics

“…Techniques, such as Noise2Void, Noise2Self, and their variants, can be directly trained on noisy data set without the need for paired noisy images [63–65]. In addition, semi‐supervised and unsupervised DL approaches have also been developed to reduce or completely remove the need for labeled training data during training, which have been demonstrated for vessel segmentation [66,67]. Lastly, physics‐embedded DL opens up a new avenue for reducing training requirements by incorporating the physical model of the microscopy technique [68,69].…”

Enhancing electron transfer of a semiconducting polymer for type I photodynamic and photothermal synergistic therapy

“…Phototherapy, including photodynamic and photothermal therapy, holds tremendous potential for cancer treatment because it utilizes the photogeneration of cytotoxic reactive oxygen species (ROS) or heat, respectively, to induce cell apoptosis and further leads to tumor suppression ( Zhou et al, 2016 ; Xu et al, 2017 ; Huang et al, 2018 ; Li et al, 2019a ; Liu et al, 2019a ; Liu et al, 2019b ; Li et al, 2020a ; Li et al, 2020b ; Yao et al, 2020 ; Zhang et al, 2020b ; Tahir et al, 2021 ; Wang et al, 2021 ). Continuous irradiation, however, inevitably leads to tumor hypoxia in the photoinduced ROS generation, especially the oxygen-dependent type II process ( Zou et al, 2020 ; Zou et al, 2021a ).…”