Automated analysis of whole brain vasculature using machine learning

Abstract

Tissue clearing methods enable imaging of intact biological specimens without sectioning. However, reliable and scalable analysis of such large imaging data in 3D remains a challenge. Towards this goal, we developed a deep learning-based framework to quantify and analyze the brain vasculature, named Vessel Segmentation & Analysis Pipeline (VesSAP). Our pipeline uses a fully convolutional network with a transfer learning approach for segmentation. We systematically analyzed vascular features of the whole brains including their length, bifurcation points and radius at the micrometer scale by registering them to the Allen mouse brain atlas. We reported the first evidence of secondary intracranial collateral vascularization in CD1-Elite mice and found reduced vascularization in the brainstem as compared to the cerebrum. VesSAP thus enables unbiased and scalable quantifications for the angioarchitecture of the cleared intact mouse brain and yields new biological insights related to the vascular brain function.

Multi-Fluorescence High-Resolution Episcopic Microscopy (MF-HREM) for Three-Dimensional Imaging of Adult Murine Organs

“…This distribution is similar to the mean radius measured in other serial sectioning modalities e.g. MOST (49) and for clearing techniques (50) although no large vessels (>10 μm) are present in MF-HREM data, due to the preferential binding of lectin to microvasculature over larger vessels as noted previously (50).…”

“…[53] It should be noted that no large vessels (>22 µm) are visible in MF-HREM data, likely due to the preferential binding of lectin to microvasculature over larger vessels as noted previously. [54]…”

Cellular and Molecular Probing of Intact Human Organs

“…Deep learning approaches could be expanded to classify the cells and analyze new structures such as vessels, nerves, and muscles-tasks that cannot be achieved easily with traditional software packages. Thanks to the adaptability of deep learning approaches, new algorithms can be trained with a small amount of training data to perform previously unknown segmentation tasks at high accuracy and speed (Belthangady and Royer, 2019;Moen et al, 2019;Todorov et al, 2019). Deep learning methods can also be parallelized on multiple GPUs (such as using cloud computing) to quickly scale up the processing speed for data size of hundreds of terabytes.…”

“…Recent deep learning approaches have proven to be superior for the analysis of large imaging data compared to prior methods both in terms of segmentation accuracy and computational power requirements (Belthangady and Royer, 2019;Kermany et al, 2018;Moen et al, 2019;Wainberg et al, 2018). To analyze largescale data from the cleared human tissue in a scalable and unbiased manner, we adopted a deep learning approach based on CNNs (convolutional neural networks) (Tetteh et al, 2018;Todorov et al, 2019) (Figure 7A). To generalize the efforts, we chose to focus on reliable detection, segmentation, and counting millions of cells in cleared human brain tissues.…”

Three-dimensional single-cell imaging for the analysis of RNA and protein expression in intact tumour biopsies

“…VesSAP is a deep-learning-based method that enables segmentation and analysis of the brain vasculature of a whole mouse brain. It can serve as a vasculature brain atlas leading to progress in understanding the brain (Todorov et al, 2019). In the future, the generation of atlases for the whole mouse as well as human organs would be of great interest.…”