Modeling longitudinal imaging biomarkers with parametric Bayesian multi‐task learning

Abstract

Longitudinal imaging biomarkers are invaluable for understanding the course of neurodegeneration, promising the ability to track disease progression and to detect disease earlier than cross‐sectional biomarkers. To properly realize their potential, biomarker trajectory models must be robust to both under‐sampling and measurement errors and should be able to integrate multi‐modal information to improve trajectory inference and prediction. Here we present a parametric Bayesian multi‐task learning based approach to modeling univariate trajectories across subjects that addresses these criteria. Our approach learns multiple subjects' trajectories within a single model that allows for different types of information sharing, that is, coupling , across subjects. It optimizes a combination of uncoupled, fully coupled and kernel coupled models. Kernel‐based coupling allows linking subjects' trajectories based on one or more biomarker measures. We demonstrate this using Alzheimer's Disease Neuroimaging Initiative (ADNI) data, where we model longitudinal trajectories of MRI‐derived cortical volumes in neurodegeneration, with coupling based on APOE genotype, cerebrospinal fluid (CSF) and amyloid PET‐based biomarkers. In addition to detecting established disease effects, we detect disease related changes within the insula that have not received much attention within the literature. Due to its sensitivity in detecting disease effects, its competitive predictive performance and its ability to learn the optimal parameter covariance from data rather than choosing a specific set of random and fixed effects a priori , we propose that our model can be used in place of or in addition to linear mixed effects models when modeling biomarker trajectories. A software implementation of the method is publicly available.

Predicting Alzheimer's disease progression using deep recurrent neural networks

“…First, the “preprocessing” approach handles the missing data issue in a separate preprocessing step, by imputing the missing data (e.g., using the missing variable’s mean or more sophisticated machine learning strategies; Azur et al, 2011 ; Rehfeld et al, 2011 ; Stekhoven and Buhlmann, 2011 ; White et al, 2011 ; Zhou et al, 2013 ), and then using the imputed data for subsequent modeling. Second, the “integrative” approach is to integrate the missing data issue directly into the models or training strategies, e.g., marginalizing the missing data in Bayesian approaches ( Marquand et al, 2014 ; Wang et al, 2014 ; Goyal et al, 2018 ; Aksman et al, 2019 ).…”

Predicting Alzheimer’s disease progression using deep recurrent neural networks

“…First, the "preprocessing" approach handles the missing data issue in a separate preprocessing step, by imputing the missing data (e.g., using the missing variable's mean or more sophisticated machine learning strategies; Azur et al, 2011;Rehfeld et al, 2011;Stekhoven and Bühlmann, 2011;White et al, 2011;Zhou et al, 2013), and then using the imputed data for subsequent modeling. Second, the "integrative" approach is to integrate the missing data issue directly into the models or training strategies, e.g., marginalizing the missing data in Bayesian approaches (Marquand et al, 2014;Wang et al, 2014;Aksman et al, 2019).…”

A survey on machine and statistical learning for longitudinal analysis of neuroimaging data in Alzheimer’s disease

“…Multi-task learning [52][53][54][55][56][57][58] Deep Learning [15,40,[59][60][61][62] Event-based models [63,64] Manifold learning [65][66][67] Mixed-effect models [18,20,38,39,[68][69][70][71][72] Shape analysis models [46,[73][74][75][76][77] Gaussian processes [29,[78][79][80] Data-based progression scores [19,47,[81][82][83] Support Vector Machine [24,25,27,33,45,[84][85]…”

“…Defining cognitive scores as separate prediction tasks and training them jointly creates a more robust predictive model. Such models have shown to have many advantages: they can use a variable number of follow-ups [55,58,90] and can provide direct information between cognitive scores and imaging markers [90,99,[102][103][104]. Apart from multi-task learning, other methods have been used for this task, such as probabilistic models [105], regression models [106], or learning ensemble models [41,100,107], which combine different, smaller models, and can be defined in flexible ways to integrate missing follow-ups into the model.…”