Shape Analysis of Subcortical Brain Structures for Tracking & Predicting Hallmarks of Alzheimer’s Disease

Introduction

Subcortical structures, such as the hippocampus, are core elements in Alzheimer’s disease. Indeed, the hippocampus is typically associated with AD-characteristic episodic memory syndrome and downstream neurodegeneration, and hippocampal atrophy is closely linked to tau-related medial temporal pathology. Apart from the hippocampus, there has been a recent interest in studying the role of other subcortical structures in the progression of AD pathology and cognitive decline. For instance, several studies have recently identified the amygdala as a region showing early neurofibrillary tau pathology and describe the thalamus as a key region in dementia due to thalamic pathology that might contribute to cognitive decline and behavioral symptoms. However, very little is known about the ability of using morphological shape analysis of these structures to track the disease progression and detect significant changes in longitudinal settings. 

We defined the projection of the of the nonlinear template-to-subject transformation onto the direction of the normal vector to the surface as our local measure of inward or outward displacement. Using this metric, we showed that such quantification of the local deformations in the shape of subcortical structures, like the hippocampus and amygdala, is a sensitive biomarker for tracking the longitudinal progression of morphological changes and atrophy in AD. We also determined that local deformations in the shape of the hippocampus are associated with other imaging and non-imaging hallmarks of AD, including CSF and plasma biomarkers. Importantly, we show that deformations in hippocampus surface can predict the Tau PET SUVR in the hippocampus itself and in the entorhinal cortex, a very early hallmark of AD.

Patient Demographics & Clinical Features at Baseline

† Diagnosis at Baseline

The data used for analysis was obtained from the ADNI database. N=1592 MRI scans from Cognitively Normal (CN), Mild Cognitive Impairment (MCI), and Alzheimer’s Disease (AD) participants were included in the longitudinal analysis of subcortical deformations, where each subject had at least two time points including a baseline visit. After filtering this cohort by image and processing quality, the resulting scans were labeled as having an initial baseline visit as well as time visits for 24 and 48 months after baseline. Overall, the average age of the participants at baseline visit was approximately 73 years and there was no statistically significant difference in age among the three diagnosis groups. There was a statistically significant association between Sex and Clinical Diagnosis, which can be explained by the larger number of males than females in the MCI group. Similarly, there was a statistically significant association between APOEε4 and Clinical Diagnosis due to the larger number of APOEε4 non carriers in the CN group. 

† Diagnosis at Baseline

The cross-sectional and predictive model analyses were performed on a cohort of N=1291 scans from CN and MCI subjects at their first available visit. For correlational analysis with other imaging (e.g. Tau PET, FDG PET) and non-imaging (e.g. plasma, CSF) biomarkers, this sample was filtered to match the acquisition dates (within a 90-day window) between the MRI scans and the corresponding biomarkers. The average age at baseline visit for the participants in this cohort was approximately 73 years and there was no statistically significant age difference between CN and MCI. There was a statistically significant association between Sex and Clinical Diagnosis, as well as between APOEε4 and Clinical Diagnosis. 

Image Processing

This illustration provides a high-level overview of the various steps involved in PIANO™, Biospective’s automated structural MRI processing "pipeline". PIANO™ is a configurable, modular, pipeline-based system for fully automated processing of multi-modality images. PIANO™ is designed for high-throughput processing of large-scale, multi-center, neuroimaging data. Briefly 3D T1-weighted MRI scans underwent: 

• Image non-uniformity correction using the N3 algorithm, brain masking, linear spatial transformation to “stereotaxic space”, and nonlinear transformation to a pre-defined “template space”. 

• Computation of the nonlinear deformation fields associated with the spatial transformation to the template space. 

• Specific VOIs were defined anatomically, explicitly on a template, or generated implicitly using a deep-learning model, and were combined with subject-specific tissue classification, as needed. 

• The outer surfaces of the several VOIs, including the Hippocampus, Amygdala, Caudate, Thalamus, and Putamen were automatically extracted using the marching cubes algorithm. 

• The local deformation on the surfaces was quantified with the projection of the of the nonlinear template-to-subject transformation onto the normal vector to the template surface. 

• Vertexwise shape analysis of the subcortical VOIs was derived from the underlying deformation maps along the normal to the outer surfaces. 

Longitudinal Analysis of Hippocampus Deformations in MCI 

A mixed-effects longitudinal model was fitted to the surface deformation data to assess the main effect of change over time for each of the groups under study, namely Cognitive Normal, MCI, and AD. The set of covariates included age at baseline and sex. 

After correction for multiple comparisons using a False Discovery Rate (FDR) approach, statistically significant longitudinal effects in local surface deformations of the hippocampus were detected in MCI subjects. Most of these significant changes were located in the head of the hippocampus and reflected an inward displacement along the surface’s normal, providing a local assessment of atrophy progression over time. The significant changes of deformations relative to the baseline time point appeared to become stronger over time.

Longitudinal Analysis of Hippocampus Deformations in Cognitively Normal and AD Subjects

Association of Hippocampus Surface Deformations with
Tau PET SUVR

A cross-sectional linear model was fitted to the hippocampus surface deformation data in CN and MCI subjects at baseline to assess its association with Tau PET SUVR. The set of covariates included age, sex, and APOEε4 genotype. 

After multiple comparisons correction, some local shape deformations in the head of the hippocampus had a statistically significant association with the accumulation of tau in both the hippocampus and the entorhinal cortex for the population of MCI subjects only. Remarkably, the significant clusters of association are mainly located in the head of the hippocampus and anatomically close to the transentorhinal cortex, a well-known region for early localization of tau pathology (Braak stages I/II).

Association of Hippocampus Surface Deformations with CSF & Plasma pTau181 

A cross-sectional linear model was fitted to the hippocampus surface deformation data in CN and MCI subjects at baseline to assess its association with CSF pTau181 and plasma pTau181. The set of covariates included age, sex, and APOEε4 genotype. 

After multiple comparisons correction, the local shape deformations in several regions of the hippocampus had a statistically significant association with both CSF and plasma pTau81, although the strength and spatial extent appeared to be higher for CSF pTau181. Likewise, the relationship between hippocampus surface deformations and CSF/plasma pTau181 appears to be stronger in MCI subjects. 

The negative sign in these maps indicates that the increase of the strength of the inward direction deformations in the head and body of the hippocampus is associated with the increase of both CSF and plasma pTau181 levels. In other words, the hippocampus surface inward deformations are closely related to Aβ pathology and phosphorylated tau.

Association of Hippocampus Volume with CSF & Plasma pTau181 

A Pearson’s correlation analysis was computed between the hippocampus volume and both CSF pTau181 and plasma pTau181 for CN and MCI subjects. 

For MCI subjects we obtained a moderate correlation between the hippocampus volume and each of the two fluid biomarkers of pathology. However, no significant correlations were found for the case of cognitively normal subjects. As compared with our data on the previous slide, this finding illustrates that the hippocampus surface deformations are more sensitive than volume for revealing associations of the hippocampus shape with both CSF and plasma pTau181 in cognitively normal and MCI subjects. 

Association of Hippocampus Surface Deformations with Plasma NfL & GFAP Biomarkers 

A cross-sectional linear model was fitted to the hippocampus surface deformation data in CN and MCI subjects at baseline to assess its association with plasma neurofilament light chain (NfL) and plasma GFAP biomarkers. The set of covariates included age, sex, and APOEε4 genotype. 

After multiple comparisons correction, the local shape deformations in several regions of the hippocampus had statistically significant associations with with both plasma NfL and plasma GFAP, although the strength and spatial extent appeared to be higher in MCI subjects. The negative sign in the significant associations means that the increase of the strength of the inward direction deformations in some regions of the hippocampus is correlated with the increase of both plasma NFL and GFAP levels. In other words, the hippocampus surface inward deformations are closely related to both axonal degeneration and astrogliosis.

Association of Hippocampus Surface Deformations with Plasma Aβ42/Aβ40 Ratio & FDG PET Biomarkers

A cross-sectional linear model was fitted to the hippocampus surface deformation data in CN and MCI subjects at baseline to assess its association with plasma Aβ42/Aβ40 ratio and FDG PET biomarkers. The set of covariates included age, sex, and APOEε4 genotype. 

After multiple comparisons correction, the local shape deformations in several regions of the hippocampus had statistically significant associations with both plasma Aβ42/Aβ40 ratio and FDG PET in MCI subjects only. The positive sign in the significant associations means that the increase of the strength of the inward direction deformations in some regions of the hippocampus is correlated with the decrease of both plasma Aβ42/Aβ40 ratio and FDG PET SUVR levels. In other words, the hippocampus surface inward deformations in MCI are related with both Aβ plaque accumulation and cerebral glucose hypometabolism.

Accuracy of Tau PET SUVR Predictions 

A leave-one-out cross-validation analysis produced an optimal number of 220 Partial Least Squares (PLS) components with a squared cross-validated correlation coefficient (Q2) value of 0.74 for the prediction of Tau PET SUVR in the hippocampus itself, and 200 PLS components (Q2=0.81) for the prediction of Tau PET SUVR in the entorhinal cortex. In both cases, the predicted Tau SUVR PET values are highly correlated with the observed ones. 

Longitudinal Analysis of Several Surface Deformations

Very few to no statistically significant clusters were detected for the longitudinal effects of the amygdala and putamen surface deformations at 24 months in the MCI group. In contrast, the caudate and thalamus revealed large clusters of statistically significant changes in deformation. For the case of the caudate, the deformation changes occurred in both positive and negative signs, which is a typical example where local deformations occurred in both inward and outward directions along the surface normal, revealing subtle morphological changes that is probably not affecting the overall volume of the anatomical structure.

Association of Amygdala Surface Deformations with Tau PET SUVR 

A cross-sectional linear model was fitted to the amygdala surface deformation data in CN and MCI subjects at baseline to assess its association with Tau PET SUVR in both the entorhinal cortex and the amygdala itself. The set of covariates included age, sex, and APOEε4 genotype. 

After multiple comparisons correction, the local shape deformations in several regions of the amygdala had statistically significant associations with Tau PET SUVR in both regions for the MCI group only. However, this association seems to be stronger and more spatially extended for the case of tau in the amygdala itself.

Association of Amygdala Surface Deformations with CSF pTau181 & Plasma GFAP Biomarkers

A cross-sectional linear model was fitted to the amygdala surface deformation data in CN and MCI subjects at baseline to assess its association with CSF pTau181 and plasma GFAP biomarkers. The set of covariates included age, sex, and APOEε4 genotype. 

After multiple comparisons correction, the local shape deformations in several regions of the amygdala had a statistically significant association with both CSF pTau181 and plasma GFAP, although the strength and spatial extension appeared to be higher in MCI subjects. However, in MCI, the amygdala surface deformations were more highly correlated with CSF pTau181 than with plasma GFAP, revealing a greater implication of the amygdala in the accumulation of Aβ plaques than with astrogliosis. 

Association of Amygdala Volume with CSF pTau181 & Plasma GFAP 

A Pearson’s correlation analysis was computed between the amygdala volume and both CSF pTau181 and plasma GFAP for CN and MCI subjects. 

For MCI subjects, we obtained a moderate correlation between the amygdala volume and each of the two fluid biomarkers. However, no significant correlations were found for the case of cognitively normal subjects. As compared with our data on the previous slide, this finding illustrates that the amygdala surface deformations might be more sensitive than volume for revealing associations of the hippocampus shape with both CSF pTau181 and plasma GFAP in cognitively normal and MCI subjects.

Summary

We examined the feasibility of the local deformations on the surface of subcortical structures (i.e. “shape analysis”) as an alternative biomarker to the traditional volumetric metrics for the diagnosis and tracking progression of Alzheimer’s’ disease.

Our key findings:

• Local deformations in the shape of subcortical structures, such as the hippocampus, is a sensitive biomarker for tracking the progression of atrophy in MCI.

• Surface-based deformations in the hippocampus and the amygdala are highly correlated with other imaging and fluid biomarkers of Alzheimer’s disease.

• Surface-based deformations in the hippocampus can be used as accurate predictors of Tau PET SUVR in Cognitively Normal and MCI subjects.

By accurately identifying subcortical regions with significant morphological changes, our approach enhances our ability to track disease progression and evaluate therapeutic interventions. Our predictive models will allow us to use anatomical T1-weighted MRI data as a surrogate for Tau PET SUVR for eligibility screening in clinical trials of disease-modifying therapeutics.