Segmentation and quantification of mesoscopic subcortical vessels using post mortem MRI at 50 micron

2164 

Abstract Submission 

Kevin Sitek¹, Marshall Xu², Omer Faruk Gulban³, Saskia Bollmann⁴

¹Northwestern University, Evanston, IL, ²The University of Queensland, Brisbane, Queensland, ³Maastricht University, Maastricht, Netherlands, ⁴School of Electrical Engineering and Computer Science, The University of Queensland, Brisbane, Queensland

Kevin Sitek  
Northwestern University
Evanston, IL

Marshall Xu  
The University of Queensland
Brisbane, Queensland

Omer Faruk Gulban, Ph.D.  
Maastricht University
Maastricht, Netherlands

Saskia Bollmann  
School of Electrical Engineering and Computer Science, The University of Queensland
Brisbane, Queensland

Recent work has mapped human neurovasculature in neocortex using anatomical MRI (Bollman et al., 2022; Gulban et al., 2022), but such work has not yet investigated subcortical vessels comprehensively (see Fig. 5 from Sitek & Gulban et al., 2019, for a preliminary investigation). While subcortical vessels were laboriously delineated by Duvernoy (Duvernoy, 1978), no quantitative characterisation of the whole three-dimensional vasculature in human subcortex has been performed at this point. Subcortical vascular mapping is important within the context of fMRI because T2*–weighted images-including BOLD functional MRI-are sensitive to blood vessels and distort the immediate vicinity of vessels, causing activation patterns to spatially shift (e.g. "draining veins" (Havlicek et al., 2020). However, currently it is difficult to directly image the subcortical vessels at mesoscopic scale in vivo (e.g., below 250 µm). Therefore, here we use high contrast post mortem MRI at 50 µm isotropic resolution to segment vasculature in a portion of the human brainstem-dorsal midbrain, including superior and inferior colliculus-and quantify the radius, length, and tortuosity of these subcortical vessels.

T2*-weighted anatomical MRI was collected with 50 µm isotropic resolution from a post mortem human brainstem (65-year-old male, no neurological conditions) using a small-bore 7 Tesla MRI (Calabrese et al., 2015).

To segment the vasculature, we first created a binary mask of the midbrain with fslmaths, which required thresholding the image, eroding the mask to remove "islands" in the mask, dilating the mask to return boundaries for most parts of the image, and filling holes. We next inverted the T2*–weighted image such that low intensity parts of the original image (such as blood vessels) were now high intensity and vice versa. To return the non-brain parts of the image back to 0, we then masked the inverted image using the binary mask in step 1.

Vessels were extracted using a self-learning algorithm (Bollman, 2023 ISMRM), which used a manually cleaned thresholding segmentation as the training label. Vessel statistics were quantified using VesselVio, which reports the Number of Segments per Radius Bin, Mean Length of Segments per Radius Bin, and Mean Segment Tortuosity per Radius Bin.

Using VesselVio (Bumgarner and Nelson, 2022), we identified 8368 segments corresponding to vasculature in the dorsal midbrain of our T2*-weighted sample. These segments had a mean radius of 59.4 µm and a mean length of 440.5 µm. The smallest vessels identified had a radius between 20–30 µm (corresponding to the 50µm spatial resolution of our data), while over 50% of the vessels were between 40–70 µm. In terms of segment length, vessels with a radius between 60–90 µm had the highest estimated length (>500 µm; over 10 voxels). Tortuosity generally decreased as a function of segment length, with highest tortuosity in vessels with a 30–50 µm radius.

Using 50 µm isotropic T2*-weighted MRI in a post mortem sample, we provide the first characterization of key properties of vasculature in human dorsal midbrain. By using uniquely high resolution anatomical MRI, we found vessels as small as 20–30 µm in radius. Smaller radius vessels generally had shorter trajectories similar to class 1-3 intracortical arteries and class 1 veins based on Duvernoy's classification (Duvernoy, 1981; see also Gulban et al., 2022, Sup. Fig. 9), with the longest vessels having radii that correspond to class 5 and 6 arteries and class 5 veins. Future work will expand this analysis to all of the human brainstem and thalamus.

Image Registration and Computational Anatomy

Segmentation and Parcellation

Subcortical Structures ¹

Cerebral Metabolism and Hemodynamics ²

Physiology, Metabolism and Neurotransmission Other

ANGIOGRAPHY

Brainstem

HIGH FIELD MR

MR ANGIOGRAPHY

STRUCTURAL MRI

Sub-Cortical