Alzheimer's disease (AD) is highly heterogenous, with significant variations in both clinical presentation and neurobiology. To explore this, we used normative modelling on multimodal neuroimaging data to index spatial patterns of neuroanatomical and neuropathological variability in AD participants. Furthermore, we quantify group differences in between-participant dissimilarity for each modality. Finally, we proposed a disease severity index based on outlier deviations, assessed the relationships between the severity index and cognitive function and examined whether the disease index was predictive of disease progression.Multimodal regional brain data were obtained in the form of gray matter volumes from T1-weighted MRI scans, amyloid SUVR (Standardized Uptake Value Ratio) from Florbetapir 18F AV45 amyloid PET and tau SUVR from Florbetapir 18F AV1451 tau PET scans respectively. A multimodal variational autoencoder-based normative model, adjusted on age and sex, was trained on cognitively unimpaired subjects (Clinical Dementia Rating [CDR®] = 0 and without amyloid positivity). The trained model was subsequently used to estimate regional Z-scores brain map for each individual with AD, measuring the deviation from the norm for each modality. Finally, statistical outliers were identified and a disease severity index was calculated for each AD participant.Subjects in the advanced AD stages (a) have more morphological and pathological brain changes, (b) have more within-group heterogeneity and (c) have a higher proportion of outlier regional deviations than people with preclinical AD and healthy controls. It was also observed that subject-level heterogeneity in MRI atrophy (neuroanatomical heterogeneity) and amyloid and tau deposition (neuropathological heterogeneity) are (a) significantly associated with cognitive performance and (b) can be potential markers to predict survival time before AD progression to advanced CDR stages.Individualized normative maps of brain atrophy, amyloid and tau loading highlight the heterogeneous effect of AD on the brain. The disease severity index based on regional outlier estimates can be potentially used to track an individual's disease progression or treatment response in clinical trials.

Analyse patient-level heterogeneity in Alzheimer’s Disease using multimodal normative modelling