Predicting sex, age, general cognition and mental health with machine learning on brain structural connectomes

A simple but tough-to-beat baseline for fMRI time-series classification

Current neuroimaging studies frequently use complex machine learning models to classify human fMRI data, distinguishing healthy and disordered brains, often to validate new methods or enhance prediction accuracy. Yet, where prediction accuracy is a concern, our results suggest that precision in prediction does not always require such sophistication. When a classifier as simple as logistic regression is applied to feature-engineered fMRI data, it can match or even outperform more sophisticated recent models. Classification of the raw time series fMRI data generally benefits from complex parameter-rich models. However, this complexity often pushes them into the class of black-box models. Yet, we found that a relatively simple model can consistently outperform much more complex classifiers in both accuracy and speed. This model applies the same multi-layer perceptron repeatedly across time and averages the results. Thus, the complexity and black-box nature of the parameter rich models, often perceived as a necessary trade-off for higher performance, do not invariably yield superior results on fMRI. Given the success of straightforward approaches, we challenge the merit of research that concentrates solely on complex model development driven by classification. Instead, we advocate for increased focus on designing models that prioritize the explainability of fMRI data or pursue applicable objectives beyond mere classification accuracy, unless they significantly outperform logistic regression or our proposed model. To validate our claim, we explore possible reasons for the superior performance of our straightforward model by examining the innate characteristics of fMRI time series data. Our findings suggest that the sequential information hidden in the temporal order may be far less important for the accurate fMRI classification than the stand-alone pieces of information scattered across the frames of the time series.

Brain network fingerprints of Alzheimer's disease risk factors in mouse models with humanized APOE alleles

Alzheimer's disease (AD) presents complex challenges due to its multifactorial nature, poorly understood etiology, and late detection. The mechanisms through which genetic and modifiable risk factors influence disease susceptibility are under intense investigation, with APOE being the major genetic risk factor for late onset AD. Yet the impact of unique risk factors on brain networks is difficult to disentangle, and their interactions remain unclear. To model multiple risk factors, including APOE genotype, age, sex, diet, and immunity we used a cross sectional design, leveraging mice expressing human APOE and NOS2 genes, conferring a reduced immune response compared to mouse Nos2. We used network topological and GraphClass analyses of brain connectomes derived from accelerated diffusion-weighted MRI to assess the global and local impact of risk factors, in the absence of AD pathology. Aging and a high-fat diet impacted extensive networks comprising AD-vulnerable regions, including the temporal association cortex, amygdala, and the periaqueductal gray, involved in stress responses. Sex impacted networks including sexually dimorphic regions (thalamus, insula, hypothalamus) and key memory-processing areas (fimbria, septum). APOE genotypes modulated connectivity in memory, sensory, and motor regions, while diet and immunity both impacted the insula and hypothalamus. Notably, these risk factors converged on a circuit comprising 63 of 54,946 total connections (0.11% of the connectome), highlighting shared vulnerability amongst multiple AD risk factors in regions essential for sensory integration, emotional regulation, decision making, motor coordination, memory, homeostasis, and interoception. APOE genotype specific immune signatures support the design of interventions tailored to risk profiles. Sparse Canonical Correlation Analysis (CCA) including spatial memory as a risk factor resulted in a network comprising 80 edges, showing significant overlap with risk-associated networks from GraphClass. The largest overlaps were observed with networks impacted by diet (47 edges), immunity (39 edges), APOE3 vs 4 (26 edges), sex (23 edges), and age (19 edges), the resulting networks supporting the use of sensory cues in spatial memory retrieval. These network-based biomarkers hold translational value for distinguishing high-risk versus low-risk participants at preclinical AD stages, suggest circuits as potential therapeutic targets, and advance our understanding of network fingerprints associated with AD risk.

APOE, Immune Factors, Sex, and Diet Interact to Shape Brain Networks in Mouse Models of Aging

Alzheimer's disease (AD) presents complex challenges due to its multifactorial nature, poorly understood etiology, and late detection. The mechanisms through which genetic, fixed and modifiable risk factors influence susceptibility to AD are under intense investigation, yet the impact of unique risk factors on brain networks is difficult to disentangle, and their interactions remain unclear. To model multiple risk factors including APOE genotype, age, sex, diet, and immunity we leveraged mice expressing the human APOE and NOS2 genes, conferring a reduced immune response compared to mouse Nos2. Employing graph analyses of brain connectomes derived from accelerated diffusion-weighted MRI, we assessed the global and local impact of risk factors in the absence of AD pathology. Aging and a high-fat diet impacted extensive networks comprising AD-vulnerable regions, including the temporal association cortex, amygdala, and the periaqueductal gray, involved in stress responses. Sex impacted networks including sexually dimorphic regions (thalamus, insula, hypothalamus) and key memory-processing areas (fimbria, septum). APOE genotypes modulated connectivity in memory, sensory, and motor regions, while diet and immunity both impacted the insula and hypothalamus. Notably, these risk factors converged on a circuit comprising 63 of 54,946 total connections (0.11% of the connectome), highlighting shared vulnerability amongst multiple AD risk factors in regions essential for sensory integration, emotional regulation, decision making, motor coordination, memory, homeostasis, and interoception. These network-based biomarkers hold translational value for distinguishing high-risk versus low-risk participants at preclinical AD stages, suggest circuits as potential therapeutic targets, and advance our understanding of network fingerprints associated with AD risk.

Significance Statement

Current interventions for Alzheimer's disease (AD) do not provide a cure, and are delivered years after neuropathological onset. Addressing the impact of risk factors on brain networks holds promises for early detection, prevention, and revealing putative therapeutic targets at preclinical stages. We utilized six mouse models to investigate the impact of factors, including APOE genotype, age, sex, immunity, and diet, on brain networks. Large structural connectomes were derived from high resolution compressed sensing diffusion MRI. A highly parallelized graph classification identified subnetworks associated with unique risk factors, revealing their network fingerprints, and a common network composed of 63 connections with shared vulnerability to all risk factors. APOE genotype specific immune signatures support the design of interventions tailored to risk profiles.

TractGeoNet: A geometric deep learning framework for pointwise analysis of tract microstructure to predict language assessment performance

We propose a geometric deep-learning-based framework, TractGeoNet, for performing regression using diffusion magnetic resonance imaging (dMRI) tractography and associated pointwise tissue microstructure measurements. By employing a point cloud representation, TractGeoNet can directly utilize tissue microstructure and positional information from all points within a fiber tract without the need to average or bin data along the streamline as traditionally required by dMRI tractometry methods. To improve regression performance, we propose a novel loss function, the Paired-Siamese Regression loss, which encourages the model to focus on accurately predicting the relative differences between regression label scores rather than just their absolute values. In addition, to gain insight into the brain regions that contribute most strongly to the prediction results, we propose a Critical Region Localization algorithm. This algorithm identifies highly predictive anatomical regions within the white matter fiber tracts for the regression task. We evaluate the effectiveness of the proposed method by predicting individual performance on two neuropsychological assessments of language using a dataset of 20 association white matter fiber tracts from 806 subjects from the Human Connectome Project Young Adult dataset. The results demonstrate superior prediction performance of TractGeoNet compared to several popular regression models that have been applied to predict individual cognitive performance based on neuroimaging features. Of the twenty tracts studied, we find that the left arcuate fasciculus tract is the most highly predictive of the two studied language performance assessments. Within each tract, we localize critical regions whose microstructure and point information are highly and consistently predictive of language performance across different subjects and across multiple independently trained models. These critical regions are widespread and distributed across both hemispheres and all cerebral lobes, including areas of the brain considered important for language function such as superior and anterior temporal regions, pars opercularis, and precentral gyrus. Overall, TractGeoNet demonstrates the potential of geometric deep learning to enhance the study of the brain's white matter fiber tracts and to relate their structure to human traits such as language performance.

Differential predictability of cognitive profiles from brain structure in older males and females

Structural brain imaging parameters may successfully predict cognitive performance in neurodegenerative diseases but mostly fail to predict cognitive abilities in healthy older adults. One important aspect contributing to this might be sex differences. Behaviorally, older males and females have been found to differ in terms of cognitive profiles, which cannot be captured by examining them as one homogenous group. In the current study, we examined whether the prediction of cognitive performance from brain structure, i.e. region-wise grey matter volume (GMV), would benefit from the investigation of sex-specific cognitive profiles in a large sample of older adults (1000BRAINS; N = 634; age range 55-85 years). Prediction performance was assessed using a machine learning (ML) approach. Targets represented a) a whole-sample cognitive component solution extracted from males and females, and b) sex-specific cognitive components. Results revealed a generally low predictability of cognitive profiles from region-wise GMV. In males, low predictability was observed across both, the whole sample as well as sex-specific cognitive components. In females, however, predictability differences across sex-specific cognitive components were observed, i.e. visual working memory (WM) and executive functions showed higher predictability than fluency and verbal WM. Hence, results accentuated that addressing sex-specific cognitive profiles allowed a more fine-grained investigation of predictability differences, which may not be observable in the prediction of the whole-sample solution. The current findings not only emphasize the need to further investigate the predictive power of each cognitive component, but they also emphasize the importance of sex-specific analyses in older adults.

Prediction of cognitive performance differences in older age from multimodal neuroimaging data

Differences in brain structure and functional and structural network architecture have been found to partly explain cognitive performance differences in older ages. Thus, they may serve as potential markers for these differences. Initial unimodal studies, however, have reported mixed prediction results of selective cognitive variables based on these brain features using machine learning (ML). Thus, the aim of the current study was to investigate the general validity of cognitive performance prediction from imaging data in healthy older adults. In particular, the focus was with examining whether (1) multimodal information, i.e., region-wise grey matter volume (GMV), resting-state functional connectivity (RSFC), and structural connectivity (SC) estimates, may improve predictability of cognitive targets, (2) predictability differences arise for global cognition and distinct cognitive profiles, and (3) results generalize across different ML approaches in 594 healthy older adults (age range: 55-85 years) from the 1000BRAINS study. Prediction potential was examined for each modality and all multimodal combinations, with and without confound (i.e., age, education, and sex) regression across different analytic options, i.e., variations in algorithms, feature sets, and multimodal approaches (i.e., concatenation vs. stacking). Results showed that prediction performance differed considerably between deconfounding strategies. In the absence of demographic confounder control, successful prediction of cognitive performance could be observed across analytic choices. Combination of different modalities tended to marginally improve predictability of cognitive performance compared to single modalities. Importantly, all previously described effects vanished in the strict confounder control condition. Despite a small trend for a multimodal benefit, developing a biomarker for cognitive aging remains challenging.

Structure-decoupled functional connectome-based brain age prediction provides higher association to cognition

Brain age prediction as well as the prediction difference has been well examined to be a potential biomarker for brain disease or abnormal aging process. However, less knowledge was reported for the cognitive association within normal population. In this study, we proposed a novel approach to brain age prediction by structure-decoupled functional connectome. The original functional connectome was decomposed and decoupled into a structure-decoupled functional connectome using structural connectome harmonics. Our method was applied to a large dataset of normal aging individuals and achieved a high correlation between predicted and chronological age (r = 0.77). Both the original FC and structure-decoupled FC could be well-trained in a brain age prediction model. Significant remarkable relationships between the brain age prediction difference (predicted age minus chronological age) and cognitive scores were discovered. However, the brain age-predicted difference driven by structure-decoupled FC showed a stronger correction to the two cognitive scores (MMSE: r = -0.27, P -value = 0.002; MoCA: r = -0.32, P -value = 0.0003). Our findings suggest that our structure-decoupled functional connectivity approach could provide a more individual-specific functional network, leading to improved brain age prediction performance and a better understanding of cognitive decline in aging.

Shared and distinct structural brain networks related to childhood maltreatment and social support: connectome-based predictive modeling

Childhood maltreatment (CM) has been associated with changes in structural brain connectivity even in the absence of mental illness. Social support, an important protective factor in the presence of childhood maltreatment, has been positively linked to white matter integrity. However, the shared effects of current social support and CM and their association with structural connectivity remain to be investigated. They might shed new light on the neurobiological basis of the protective mechanism of social support. Using connectome-based predictive modeling (CPM), we analyzed structural connectomes of N = 904 healthy adults derived from diffusion-weighted imaging. CPM predicts phenotypes from structural connectivity through a cross-validation scheme. Distinct and shared networks of white matter tracts predicting childhood trauma questionnaire scores and the social support questionnaire were identified. Additional analyses were applied to assess the stability of the results. CM and social support were predicted significantly from structural connectome data (all rs ≥ 0.119, all ps ≤ 0.016). Edges predicting CM and social support were inversely correlated, i.e., positively correlated with CM and negatively with social support, and vice versa, with a focus on frontal and temporal regions including the insula and superior temporal lobe. CPM reveals the predictive value of the structural connectome for CM and current social support. Both constructs are inversely associated with connectivity strength in several brain tracts. While this underlines the interconnectedness of these experiences, it suggests social support acts as a protective factor following adverse childhood experiences, compensating for brain network alterations. Future longitudinal studies should focus on putative moderating mechanisms buffering these adverse experiences.