バイオマーカー

Understanding pathological contribution in mixed etiology dementia and atypical Alzheimer's disease remains challenging. However, in-vivo MRI and neuropathology studies are hampered by the years of delay between MRI and brain autopsy. Postmortem in-situ MRI is a good proxy to mitigate this problem (Frigerio et al. 2021) but is logistically challenging. Ex-vivo MRI offers significant higher anatomical resolution but is influenced by fixation and possible deformation. Therefore, we propose a whole-hemisphere ex-vivo to in-situ/in-vivo parcellation and registration pipeline "purple-mri" (Khandelwal et al. 2024) to cater multi-resolution, translation, and clinical relevance for T2w 7T 0.3 mm³ exvivo and T1w 3T 1x0.5x0.5 mm³ insitu MRI in 21 (age: 57-88 years; F=9/M=12) control specimens (Table 1).

We combine a deep learning-based segmentation architecture with classical surface-based modeling techniques for topology correction, cortical modeling, inflation, and registration to provide accurate anatomical parcellations of T2w 7T 0.3 mm³ exvivo cerebral hemispheres using the DKT atlas. Next, we develop a two-stage inter-modality diffeomorphic image registration method between corresponding pairs of exvivo hemispheres and insitu whole-brain MRI that first aligns segmentations of the two modalities and then applies the obtained warp to perform intensity-based MRI alignment. FreeSurfer was used to provide DKT atlas-based segmentations for insitu MRI.

Figure 1 shows qualitative parcellations of exvivo hemisphere and corresponding registered in-situ MRI. Our method can segment brain regions involving imaging artifacts with signal dropout in anterior/posterior areas. Figure 2 shows Spearman's correlation for region-wise volume and mean thickness between exvivo and insitu MRI. We observe significant correlations between the two modalities in inferior/middle/superior temporal, insula, lateralorbitofrontal, post/pre central regions for volume-based correlation. For mean thickness, we observe strong correlation in cuneus, fusiform, medial/lateral orbitofrontal, postcentral gyrus but an overall slight decrease in correlation strength primarily due to noisy thickness measurements due to lack of segmentation in areas with damaged/missing tissues during physical handling.

Our proposed pipeline (purple-mri) accurately segments high-resolution exvivo MRI and performs registration to give one-to-one correspondence between multi-modal postmortem MRI. The developed pipeline sets the stage for exvivo to invivo registrations, which would enable stronger association studies between morphometry and histopathological examination.