Normative modeling using multimodal variational autoencoders to identify abnormal brain volume deviations in Alzheimer's disease

Abstract

Normative modelling is a method for understanding the underlying heterogeneity within brain disorders like Alzheimer Disease (AD), by quantifying how each patient deviates from the expected normative pattern that has been learned from a healthy control distribution. Existing deep learning based normative models have been applied on only single modality Magnetic Resonance Imaging (MRI) neuroimaging data. However, these do not take into account the complementary information offered b y m ultimodal M RI, w hich i s e ssential for understanding a multifactorial disease like AD. To address this limitation, we propose a multi-modal variational autoencoder (mmVAE) based normative modelling framework that can capture the joint distribution between different modalities to identify abnormal brain volume deviations due to AD. Our multi-modal framework takes as input Freesurfer processed brain region volumes from T1-weighted (cortical and subcortical) and T2-weighed (hippocampal) scans of cognitively normal participants to learn the morphological characteristics of the healthy brain. The estimated normative model is then applied on AD patients to quantify the deviation in brain volumes and identify abnormal brain pattern deviations due to the progressive stages of AD. We compared our proposed mmVAE with a baseline unimodal VAE having a single encoder and decoder and the two modalities concatenated as unimodal input. Our experimental results show that deviation maps generated by mmVAE are more sensitive to disease staging within AD, have a better correlation with patient cognition and result in higher number of brain regions with statistically significant d eviations c ompared t o t he u nimodal b aseline model.

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

“…Specifically, they calculated either the average deviation in MRI gray matter volumes across all regions or the total number of MRI outlier deviations (TOC) across all regions. 20,21,30 . There are also studies quantifying the spatial spread of tau loading (TSS) for predicting cognitive impairment and disease progression.…”

“…17,20,21 In our work, we adopted a multimodal variational autoencoder (mmVAE) as the normative model, validated in previous works. 21,30,31…”

“…Ours is the first study to examine individual patterns of variation in brain structure in individuals with AD using normative modelling across multiple neuroimaging modalities. We applied our previous implementation 30,31 of normative modelling framework based on multimodal variational autoencoders (mmVAE), which used the ATN (amyloid-tau-neurodegeneration) 32 biomarker data as multimodal input. First, we trained mmVAE on a reference control group (i.e., cognitively unimpaired (CU) subjects) from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset to learn the healthy (normative) brain patterns across all modalities.…”

“…Together, these biomarkers compose the National Institute on Aging and Alzheimer's Association research framework that helps with defining AD as a biological construct and facilitates a more comprehensive understanding of individual differences in cognitive performance and clinical progression. [17][18][19][20][21] However, despite recent progress on developing deep learning models for normative modeling across multiple modalities [22][23][24], these efforts have primarily been methodological and have not investigated AD heterogeneity by taking into account core AD biomarkers (i.e., amyloid and tau). [25] In this study, we aimed to identify individual patterns of neuroanatomical and neuropathological variation in the brains of individuals with AD using a deep learning based normative modeling framework across amyloid-tau-neurodegeneration (ATN) [26] imaging biomarker data from Alzheimer's Disease Neuroimaging Initiative (ADNI).…”

“…[25] In this study, we aimed to identify individual patterns of neuroanatomical and neuropathological variation in the brains of individuals with AD using a deep learning based normative modeling framework across amyloid-tau-neurodegeneration (ATN) [26] imaging biomarker data from Alzheimer's Disease Neuroimaging Initiative (ADNI). Accordingly, we trained our previously validated normative modeling framework, [22,24] which is based on multimodal variational autoencoders (mmVAE), on data from a reference control group (i.e., amyloid negative cognitively unimpaired (CU) subjects). We subsequently used the trained model to estimate the extent to which individuals spanning the AD spectrum (ADS) deviate from the normative distribution.…”