Modeling longitudinal imaging biomarkers with parametric Bayesian multi-task learning

Bridging Imaging and Clinical Scores in Parkinson's Progression via Multimodal Self-Supervised Deep Learning

Neurodegenerative diseases pose a formidable challenge to medical research, demanding a nuanced understanding of their progressive nature. In this regard, latent generative models can effectively be used in a data-driven modeling of different dimensions of neurodegeneration, framed within the context of the manifold hypothesis. This paper proposes a joint framework for a multi-modal, common latent generative model to address the need for a more comprehensive understanding of the neurodegenerative landscape in the context of Parkinson's disease (PD). The proposed architecture uses coupled variational autoencoders (VAEs) to joint model a common latent space to both neuroimaging and clinical data from the Parkinson's Progression Markers Initiative (PPMI). Alternative loss functions, different normalization procedures, and the interpretability and explainability of latent generative models are addressed, leading to a model that was able to predict clinical symptomatology in the test set, as measured by the unified Parkinson's disease rating scale (UPDRS), with R2 up to 0.86 for same-modality and 0.441 cross-modality (using solely neuroimaging). The findings provide a foundation for further advancements in the field of clinical research and practice, with potential applications in decision-making processes for PD. The study also highlights the limitations and capabilities of the proposed model, emphasizing its direct interpretability and potential impact on understanding and interpreting neuroimaging patterns associated with PD symptomatology.

Gaussian Process-based prediction of memory performance and biomarker status in ageing and Alzheimer's disease-A systematic model evaluation

Neuroimaging markers based on Magnetic Resonance Imaging (MRI) combined with various other measures (such as genetic covariates, biomarkers, vascular risk factors, neuropsychological tests etc.) might provide useful predictions of clinical outcomes during the progression towards Alzheimer's disease (AD). The use of multiple features in predictive frameworks for clinical outcomes has become increasingly prevalent in AD research. However, many studies do not focus on systematically and accurately evaluating combinations of multiple input features. Hence, the aim of the present work is to explore and assess optimal combinations of various features for MR-based prediction of (1) cognitive status and (2) biomarker positivity with a multi-kernel learning Gaussian process framework. The explored features and parameters included (A) combinations of brain tissues, modulation, smoothing, and image resolution; (B) incorporating demographics & clinical covariates; (C) the impact of the size of the training data set; (D) the influence of dimensionality reduction and the choice of kernel types. The approach was tested in a large German cohort including 959 subjects from the multicentric longitudinal study of cognitive impairment and dementia (DELCODE). Our evaluation suggests the best prediction of memory performance was obtained for a combination of neuroimaging markers, demographics, genetic information (ApoE4) and CSF biomarkers explaining 57% of outcome variance in out-of-sample predictions. The highest performance for Aβ42/40 status classification was achieved for a combination of demographics, ApoE4, and a memory score while usage of structural MRI further improved the classification of individual patient's pTau status.

Cognitive Trajectories in Preclinical and Prodromal Alzheimer's Disease Related to Amyloid Status and Brain Atrophy: A Bayesian Approach

Background

Cognitive decline is a key outcome of clinical studies in Alzheimer's disease (AD).

Objective

To determine effects of global amyloid load as well as hippocampus and basal forebrain volumes on longitudinal rates and practice effects from repeated testing of domain specific cognitive change in the AD spectrum, considering non-linear effects and heterogeneity across cohorts.

Methods

We included 1,514 cases from three cohorts, ADNI, AIBL, and DELCODE, spanning the range from cognitively normal people to people with subjective cognitive decline and mild cognitive impairment (MCI). We used generalized Bayesian mixed effects analysis of linear and polynomial models of amyloid and volume effects in time. Robustness of effects across cohorts was determined using Bayesian random effects meta-analysis.

Results

We found a consistent effect of amyloid and hippocampus volume, but not of basal forebrain volume, on rates of memory change across the three cohorts in the meta-analysis. Effects for amyloid and volumetric markers on executive function were more heterogeneous. We found practice effects in memory and executive performance in amyloid negative cognitively normal controls and MCI cases, but only to a smaller degree in amyloid positive controls and not at all in amyloid positive MCI cases.

Conclusions

We found heterogeneity between cohorts, particularly in effects on executive functions. Initial increases in cognitive performance in amyloid negative, but not in amyloid positive MCI cases and controls may reflect practice effects from repeated testing that are lost with higher levels of cerebral amyloid.

Regional Deep Atrophy: a Self-Supervised Learning Method to Automatically Identify Regions Associated With Alzheimer's Disease Progression From Longitudinal MRI

Longitudinal assessment of brain atrophy, particularly in the hippocampus, is a well-studied biomarker for neurodegenerative diseases, such as Alzheimer's disease (AD). In clinical trials, estimation of brain progressive rates can be applied to track therapeutic efficacy of disease modifying treatments. However, most state-of-the-art measurements calculate changes directly by segmentation and/or deformable registration of MRI images, and may misreport head motion or MRI artifacts as neurodegeneration, impacting their accuracy. In our previous study, we developed a deep learning method DeepAtrophy that uses a convolutional neural network to quantify differences between longitudinal MRI scan pairs that are associated with time. DeepAtrophy has high accuracy in inferring temporal information from longitudinal MRI scans, such as temporal order or relative inter-scan interval. DeepAtrophy also provides an overall atrophy score that was shown to perform well as a potential biomarker of disease progression and treatment efficacy. However, DeepAtrophy is not interpretable, and it is unclear what changes in the MRI contribute to progression measurements. In this paper, we propose Regional Deep Atrophy (RDA), which combines the temporal inference approach from DeepAtrophy with a deformable registration neural network and attention mechanism that highlights regions in the MRI image where longitudinal changes are contributing to temporal inference. RDA has similar prediction accuracy as DeepAtrophy, but its additional interpretability makes it more acceptable for use in clinical settings, and may lead to more sensitive biomarkers for disease monitoring in clinical trials of early AD.

MC-RVAE: Multi-channel recurrent variational autoencoder for multimodal Alzheimer's disease progression modelling

The progression of neurodegenerative diseases, such as Alzheimer's Disease, is the result of complex mechanisms interacting across multiple spatial and temporal scales. Understanding and predicting the longitudinal course of the disease requires harnessing the variability across different data modalities and time, which is extremely challenging. In this paper, we propose a model based on recurrent variational autoencoders that is able to capture cross-channel interactions between different modalities and model temporal information. These are achieved thanks to its multi-channel architecture and its shared latent variational space, parametrized with a recurrent neural network. We evaluate our model on both synthetic and real longitudinal datasets, the latter including imaging and non-imaging data, with N=897 subjects. Results show that our multi-channel recurrent variational autoencoder outperforms a set of baselines (KNN, random forest, and group factor analysis) for the task of reconstructing missing modalities, reducing the mean absolute error by 5% (w.r.t. the best baseline) for both subcortical volumes and cortical thickness. Our model is robust to missing features within each modality and is able to generate realistic synthetic imaging biomarkers trajectories from cognitive scores.

Multimodal machine learning in precision health: A scoping review

Machine learning is frequently being leveraged to tackle problems in the health sector including utilization for clinical decision-support. Its use has historically been focused on single modal data. Attempts to improve prediction and mimic the multimodal nature of clinical expert decision-making has been met in the biomedical field of machine learning by fusing disparate data. This review was conducted to summarize the current studies in this field and identify topics ripe for future research. We conducted this review in accordance with the PRISMA extension for Scoping Reviews to characterize multi-modal data fusion in health. Search strings were established and used in databases: PubMed, Google Scholar, and IEEEXplore from 2011 to 2021. A final set of 128 articles were included in the analysis. The most common health areas utilizing multi-modal methods were neurology and oncology. Early fusion was the most common data merging strategy. Notably, there was an improvement in predictive performance when using data fusion. Lacking from the papers were clear clinical deployment strategies, FDA-approval, and analysis of how using multimodal approaches from diverse sub-populations may improve biases and healthcare disparities. These findings provide a summary on multimodal data fusion as applied to health diagnosis/prognosis problems. Few papers compared the outputs of a multimodal approach with a unimodal prediction. However, those that did achieved an average increase of 6.4% in predictive accuracy. Multi-modal machine learning, while more robust in its estimations over unimodal methods, has drawbacks in its scalability and the time-consuming nature of information concatenation.

Bayesian multisource data integration for explainable brain-behavior analysis

Different data sources can provide complementary information. Moving from a simple approach based on using one data source at a time to a systems approach that integrates multiple data sources provides an opportunity to understand complex brain disorders or cognitive processes. We propose a data fusion method, called Bayesian Multisource Data Integration, to model the interactions among data sources and behavioral variables. The proposed method generates representations from data sources and uses Bayesian network modeling to associate representations with behavioral variables. The generated Bayesian network is transparent and easy to understand. Bayesian inference is used to understand how the perturbation of representation is related to behavioral changes. The proposed method was assessed on the simulated data and data from the Adolescent Brain Cognitive Development study. For the Adolescent Brain Cognitive Development study, we found diffusion tensor imaging and resting-state functional magnetic resonance imaging were synergistic in understanding the fluid intelligence composite and the total score composite in healthy youth (9-11 years of age).

Susceptibility-weighted imaging at high-performance 0.5T magnetic resonance imaging system: Protocol considerations and experimental results

The high-performance low-field magnetic resonance imaging (MRI) system, equipped with modern hardware and contemporary imaging capabilities, has garnered interest within the MRI community in recent years. It has also been proven to have unique advantages over high-field MRI in both physical and cost aspects. However, for susceptibility weighted imaging (SWI), the low signal-to-noise ratio and the long echo time inherent at low field hinder the SWI from being applied to clinical applications. This work optimized the imaging protocol to select suitable parameters such as the values of time of echo (TE), repetition time (TR), and the flip angle (FA) of the RF pulse according to the signal simulations for low-field SWI. To improve the signal-to-noise ratio (SNR) performance, averaging multi-echo magnitude images and BM4D phase denoising were proposed. A comparison of the SWI in 0.5T and 1.5T was carried out, demonstrating the capability to identify magnetic susceptibility differences between variable tissues, especially, the blood veins. This would open the possibility to extend SWI applications in the high-performance low field MRI.

Evaluating the performance of Bayesian and frequentist approaches for longitudinal modeling: application to Alzheimer's disease

Linear mixed effects (LME) modelling under both frequentist and Bayesian frameworks can be used to study longitudinal trajectories. We studied the performance of both frameworks on different dataset configurations using hippocampal volumes from longitudinal MRI data across groups—healthy controls (HC), mild cognitive impairment (MCI) and Alzheimer’s disease (AD) patients, including subjects that converted from MCI to AD. We started from a big database of 1250 subjects from the Alzheimer’s disease neuroimaging initiative (ADNI), and we created different reduced datasets simulating real-life situations using a random-removal permutation-based approach. The number of subjects needed to differentiate groups and to detect conversion to AD was 147 and 115 respectively. The Bayesian approach allowed estimating the LME model even with very sparse databases, with high number of missing points, which was not possible with the frequentist approach. Our results indicate that the frequentist approach is computationally simpler, but it fails in modelling data with high number of missing values.

Predicting Alzheimer's disease progression using deep recurrent neural networks

Early identification of individuals at risk of developing Alzheimer’s disease (AD) dementia is important for developing disease-modifying therapies. In this study, given multimodal AD markers and clinical diagnosis of an individual from one or more timepoints, we seek to predict the clinical diagnosis, cognition and ventricular volume of the individual for every month (indefinitely) into the future. We proposed and applied a minimal recurrent neural network (minimalRNN) model to data from The Alzheimer’s Disease Prediction Of Longitudinal Evolution (TADPOLE) challenge, comprising longitudinal data of 1677 participants (Marinescu et al., 2018) from the Alzheimer’s Disease Neuroimaging Initiative (ADNI). We compared the performance of the minimalRNN model and four baseline algorithms up to 6 years into the future. Most previous work on predicting AD progression ignore the issue of missing data, which is a prevalent issue in longitudinal data. Here, we explored three different strategies to handle missing data. Two of the strategies treated the missing data as a “preprocessing” issue, by imputing the missing data using the previous timepoint (“forward filling”) or linear interpolation (“linear filling). The third strategy utilized the minimalRNN model itself to fill in the missing data both during training and testing (“model filling”). Our analyses suggest that the minimalRNN with “model filling” compared favorably with baseline algorithms, including support vector machine/regression, linear state space (LSS) model, and long short-term memory (LSTM) model. Importantly, although the training procedure utilized longitudinal data, we found that the trained minimalRNN model exhibited similar performance, when using only 1 input timepoint or 4 input timepoints, suggesting that our approach might work well with just cross-sectional data. An earlier version of our approach was ranked 5th (out of 53 entries) in the TADPOLE challenge in 2019. The current approach is ranked 2nd out of 63 entries as of June 3rd, 2020.

Association of tumor growth rates with molecular biomarker status: a longitudinal study of high-grade glioma

To determine the association of molecular biomarkers with tumor growth in patients with high-grade gliomas (HGGs), the tumor growth rates and molecular biomarker status in 109 patients with HGGs were evaluated. Mean tumor diameter was assessed on at least two pre-surgical T₂-weighted and contrast-enhancement T_1-weighted magnetic resonance images (MRIs). Tumor growth rates were calculated based on tumor volume and diameter using various methods. The association of biomarkers with increased or decreased tumor growth was calculated using linear mixed-effects models. HGGs exhibited rapid growth rates, with an equivalent volume doubling time of 63.4 days and an equivalent velocity of diameter expansion of 51.6 mm/year. The WHO grade was an independent clinical factor of eVDEs. TERT promoter mutation C250T and MGMT promoter methylation was significantly associated with tumor growth in univariable analysis but not in multivariable analysis. Molecular groups of IDH1, TERT, and 1p/19q and IDH1 and MGMT were independently associated with tumor growth. In addition, tumor enhanced area had a faster growth rate than a tumor entity in incomplete enhanced HGGs (p = 0.006). Our findings provide crucial information for the prediction of preoperative tumor growth in HGGs, and aided in the decision making for aggressive resection and adjuvant treatment strategies.

Educational fMRI: From the Lab to the Classroom

Functional MRI (fMRI) findings hold many potential applications for education, and yet, the translation of fMRI findings to education has not flowed. Here, we address the types of fMRI that could better support applications of neuroscience to the classroom. This 'educational fMRI' comprises eight main challenges: (1) collecting artifact-free fMRI data in school-aged participants and in vulnerable young populations, (2) investigating heterogenous cohorts with wide variability in learning abilities and disabilities, (3) studying the brain under natural and ecological conditions, given that many practical topics of interest for education can be addressed only in ecological contexts, (4) depicting complex age-dependent associations of brain and behaviour with multi-modal imaging, (5) assessing changes in brain function related to developmental trajectories and instructional intervention with longitudinal designs, (6) providing system-level mechanistic explanations of brain function, so that useful individualized predictions about learning can be generated, (7) reporting negative findings, so that resources are not wasted on developing ineffective interventions, and (8) sharing data and creating large-scale longitudinal data repositories to ensure transparency and reproducibility of fMRI findings for education. These issues are of paramount importance to the development of optimal fMRI practices for educational applications.

Modeling longitudinal imaging biomarkers with parametric Bayesian multi-task learning

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.