Examining real-world Alzheimer’s disease heterogeneity using neuroanatomical normative modelling

Abstract

Alzheimer′s disease (AD) has been traditionally associated with episodic memory impairment and medial temporal lobe atrophy. However, recent literature has highlighted the existence of atypical forms of AD, presenting with different cognitive and radiological profiles. Failure to appreciate the heterogeneity of AD in the past has led to misdiagnoses, diagnostic delays, clinical trial failures and risks limiting our understanding of the disease. AD research requires the incorporation of new analytic methods that are as free as possible from the intragroup homogeneity assumption underlying case-control approaches according to which patients belonging to the same group are comparable to each other. Neuroanatomical normative modelling is a promising technique allowing for modelling the variation in neuroimaging profiles and then assessing individual deviations from the respective distribution. Here, neuroanatomical normative modelling was applied for the first time to a real-worldclinical cohortof Alzheimer′s disease patients (n=86) who had a positive amyloid PET scan and a T1-weighted MR performed as part of their diagnostic workup. The model indexed normal cortical thickness distributions using a separate healthy reference dataset (n= 33,072), employing hierarchical Bayesian regression to predict cortical thickness per region using age and sex. Transfer learning was used to recalibrate the normative model on avalidation cohort(n=20) of scanner-matched cognitively normal individuals. Brain heterogeneity was quantified as z-scores at each of the 148 ROIs generated within each AD patient. Z-scores < -1.96 defined as outliers. Clinical features including disease severity, presenting phenotypes and comorbidities were collected from health records to explore their association with outlier profiles. Amyloid quantification was performed using an automated PET-only driven method to examine the association between amyloid burden and outliers. The total number of individual outliers (total outlier count) in biomarker-confirmed AD clinical patients ranged between 1 and 120 out of 148 (median 21.5). The superior temporal sulcus was the region with the highest count of outliers (60%) in AD patients. The mean proportion of outliers was higher in the temporal (31.5%) than in the extratemporal (19.1%) regions and up to 20% of patients had no temporal outliers. We found higher mean outlier count in patients with non-amnestic phenotypes, at more advanced disease stages and without depressive symptoms. Amyloid burden was negatively associated with outlier count. This study corroborates the heterogeneity of brain atrophy in AD and provides evidence that this approach can be used to explore anatomo-clinical correlations at an individual level.

Personalising Alzheimer’s Disease progression using brain atrophy markers

“…Similarly, atrophy patterns were shown to be related to disease severity, presenting phenotypes and comorbidities in amyloid-positive AD patients in a study using real-world memory clinic data. 18 These studies corroborate the heterogeneity of brain atrophy in AD and provide evidence that neuroanatomical normative modelling can be used to explore anatomicalclinical correlations at an individual level. However, the insights from these studies were based only on cross-sectional neuroimaging data.…”

Analyzing heterogeneity in Alzheimer Disease using multimodal normative modeling on imaging-based ATN biomarkers

“…Recent normative modelling works have used Gaussian regression models as normative models for each brain region and used that to analyze the heterogeneity in MRI brain volume/cortical thickness outliers. 20,21,56 Our work differs from the above studies in two aspects. First, we adopted a deep learning based normative model (like Pinaya et al) using a variational autoencoder which learned a single normative model for all features instead of separate normative models for individual features.…”

“…20,21,56 Our work differs from the above studies in two aspects. First, we adopted a deep learning based normative model (like Pinaya et al) using a variational autoencoder which learned a single normative model for all features instead of separate normative models for individual features.…”

A neuroimaging measure to capture heterogeneous patterns of atrophy in Parkinson’s disease and dementia with Lewy bodies

“…Neuroanatomical normative modelling has recently been successfully applied in Alzheimer's disease (AD) [13][14][15]. As expected, patients with AD had more outlier regions than mild cognitive impairment (MCI) and healthy control groups [14], though neuroanatomical patterns of outliers were variable even within AD groups [14,15]. Importantly, total outlier count correlated with poorer cognitive performance, fluid biomarker-measures of Alzheimer's pathology, and predicted future conversion from MCI to dementia [13,14].…”

“…Neuroanatomical normative modelling has recently been successfully applied in Alzheimer's disease (AD) [13][14][15]. As expected, patients with AD had more outlier regions than mild cognitive impairment (MCI) and healthy control groups [14], though neuroanatomical patterns of outliers were variable even within AD groups [14,15].…”