ArticlePublisher preview available

Connectome-based predictive models using resting-state fMRI for studying brain aging

Authors:
Article

Connectome-based predictive models using resting-state fMRI for studying brain aging

If you want to read the PDF, try requesting it from the authors.

Abstract and Figures

Changes in the brain with age can provide useful information regarding an individual’s chronological age. studies have suggested that functional connectomes identified via resting-state functional magnetic resonance imaging (fMRI) could be a powerful feature for predicting an individual’s age. We applied connectome-based predictive modeling (CPM) to investigate individual chronological age predictions via resting-state fMRI using open-source datasets. The significant feature for age prediction was confirmed in 168 subjects from the Southwest University Adult Lifespan Dataset. The higher contributing nodes for age production included a positive connection from the left inferior parietal sulcus and a negative connection from the right middle temporal sulcus. On the network scale, the subcortical–cerebellum network was the dominant network for age prediction. The generalizability of CPM, which was constructed using the identified features, was verified by applying this model to independent datasets that were randomly selected from the Autism Brain Imaging Data Exchange I and the Open Access Series of Imaging Studies 3. CPM via resting-state fMRI is a potential robust predictor for determining an individual’s chronological age from changes in the brain.
This content is subject to copyright. Terms and conditions apply.
Vol.:(0123456789)
1 3
Experimental Brain Research (2022) 240:2389–2400
https://doi.org/10.1007/s00221-022-06430-7
RESEARCH ARTICLE
Connectome‑based predictive models using resting‑state fMRI
forstudying brain aging
EunjiKim1,2· SeunghoKim2· YunheungKim2· HyunsilCha2· HuiJoongLee3,4· TaekwanLee5· YongminChang2,4,6
Received: 23 February 2022 / Accepted: 26 July 2022 / Published online: 4 August 2022
© The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2022
Abstract
Changes in the brain with age can provide useful information regarding an individual’s chronological age. studies have sug-
gested that functional connectomes identified via resting-state functional magnetic resonance imaging (fMRI) could be a
powerful feature for predicting an individual’s age. We applied connectome-based predictive modeling (CPM) to investigate
individual chronological age predictions via resting-state fMRI using open-source datasets. The significant feature for age
prediction was confirmed in 168 subjects from the Southwest University Adult Lifespan Dataset. The higher contributing
nodes for age production included a positive connection from the left inferior parietal sulcus and a negative connection from
the right middle temporal sulcus. On the network scale, the subcortical–cerebellum network was the dominant network for
age prediction. The generalizability of CPM, which was constructed using the identified features, was verified by applying
this model to independent datasets that were randomly selected from the Autism Brain Imaging Data Exchange I and the
Open Access Series of Imaging Studies 3. CPM via resting-state fMRI is a potential robust predictor for determining an
individual’s chronological age from changes in the brain.
Keywords Prediction model· Resting-state functional magnetic resonance imaging· Functional connectivity·
Connectome-based predictive modeling· Feature selection
Abbreviations
ABIDE I Autism brain imaging data exchange I
CP Connectome-based predictive
CPM Connectome-based predictive modeling
FD Framewise head displacement
fMRI Functional magnetic resonance imaging
LOOCV Leave-one-out cross-validation
OASIS 3 Open access series of imaging studies 3
rs-fMRI Resting-state fMRI
SALD Southwest University Adult Lifespan Dataset
Introduction
Human brain undergoes anatomical and functional changes
throughout the lifespan. Several neuroimaging studies have
demonstrated that changes in the brain structure exhibit dis-
tinct patterns, including cortical thinning during adulthood
(Fjell etal. 2009; Hogstrom etal. 2013; Lamballais etal.
2020). Certain sections of the brain show regional and global
volume changes that reflect age effects (Good etal. 2001;
Farokhian etal. 2017; Bajaj etal. 2017), whereas other sec-
tions exhibit nonlinear changes (Walhovd etal. 2005; Ziegler
Communicated by Melvyn A. Goodale.
* Taekwan Lee
tklee@kbri.re.kr
* Yongmin Chang
ychang@knu.ac.kr
1 Department ofKorea Radioisotope Center
forPharmaceuticals, Korea Institute ofRadiological
andMedical Sciences, Seoul, Korea
2 Department ofMedical andBiological Engineering,
Kyungpook National University, Daegu, Korea
3 Department ofRadiology, Kyungpook National University
School ofMedicine, Daegu, Korea
4 Department ofRadiology, Kyungpook National University
Hospital, Daegu, Korea
5 Korea Brain Research Institute, Chumdanro 61, Dong-gu,
Daegu41021, RepublicofKorea
6 The Department ofMolecular Medicine andRadiology,
Kyungpook National University School ofMedicine, 200
Dongduk-Ro Jung-Gu, Daegu, Korea
Content courtesy of Springer Nature, terms of use apply. Rights reserved.
ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
Resting-state functional connectivity (RSFC) has widely been used to examine reorganization of functional brain networks during normal aging. The extraction of generalizable age trends, however, is hampered by differences in methodological approaches, study designs and sample characteristics. Distinct age ranges of study samples thereby represent an important aspect between studies especially due to the increase in inter-individual variability over the lifespan. The current review focuses on comparing age-related differences in RSFC in the course of the whole adult lifespan versus later decades of life. We summarize and compare studies assessing age-related differences in within- and between-network RSFC of major resting-state brain networks. Differential effects of the factor age on resting-state networks can be identified when comparing studies focusing on younger versus older adults with studies investigating effects within the older adult population. These differential effects pertain to higher order and primary processing resting-state networks to a varying extent. Especially during later decades of life, other factors beyond age might come into play to understand the high inter-individual variability in RSFC.
Article
Full-text available
Previous studies have demonstrated relations between spontaneous neural activity evaluated by resting-state functional magnetic resonance imaging (fMRI) and symptom severity in post-traumatic stress disorder. However, few studies have used brain-based measures to identify imaging associations with illness severity at the level of individual patients. This study applied connectome-based predictive modeling (CPM), a recently developed data-driven and subject-level method, to identify brain function features that are related to symptom severity of trauma survivors. Resting-state fMRI scans and clinical ratings were obtained 10–15 months after the earthquake from 122 earthquake survivors. Symptom severity of post-traumatic stress disorder features for each survivor was evaluated using the Clinician Administered Post-traumatic Stress Disorder Scale (CAPS-IV). A functionally pre-defined atlas was applied to divide the human brain into 268 regions. Each individual’s functional connectivity 268 × 268 matrix was created to reflect correlations of functional time series data across each pair of nodes. The relationship between CAPS-IV scores and brain functional connectivity was explored in a CPM linear model. Using a leave-one-out cross-validation (LOOCV) procedure, findings showed that the positive network model predicted the left-out individual’s CAPS-IV scores from resting-state functional connectivity. CPM predicted CAPS-IV scores, as indicated by a significant correspondence between predicted and actual values ( r = 0.30, P = 0.001) utilizing primarily functional connectivity between visual cortex, subcortical-cerebellum, limbic, and motor systems. The current study provides data-driven evidence regarding the functional brain features that predict symptom severity based on the organization of intrinsic brain networks and highlights its potential application in making clinical evaluation of symptom severity at the individual level.
Article
Full-text available
Deep learning has huge potential for accurate disease prediction with neuroimaging data, but the prediction performance is often limited by training-dataset size and computing memory requirements. To address this, we propose a deep convolutional neural network model, Simple Fully Convolutional Network (SFCN), for accurate prediction of brain age using T1-weighted structural MRI data. Compared with other popular deep network architectures, SFCN has fewer parameters, so is more compatible with small dataset size and 3D volume data. The network architecture was combined with several techniques for boosting performance, including data augmentation, pre-training, model regularization, model ensemble and prediction bias correction. We compared our overall SFCN approach with several widely-used machine learning models. It achieved state-of-the-art performance in UK Biobank data (N = 14,503), with mean absolute error (MAE) = 2.14y in brain age prediction and 99.5% in sex classification. SFCN also won (both parts of) the 2019 Predictive Analysis Challenge for brain age prediction, involving 79 competing teams (N = 2,638, MAE = 2.90y). We describe here the details of our approach, and its optimisation and validation. Our approach can easily be generalised to other tasks using different image modalities, and is released on GitHub.
Preprint
Full-text available
OASIS-3 is a compilation of MRI and PET imaging and related clinical data for 1098 participants who were collected across several ongoing studies in the Washington University Knight Alzheimer Disease Research Center over the course of 15 years. Participants include 605 cognitively normal adults and 493 individuals at various stages of cognitive decline ranging in age from 42 to 95 years. The OASIS-3 dataset contains over 2000 MR sessions, including multiple structural and functional sequences. PET metabolic and amyloid imaging includes over 1500 raw imaging scans and the accompanying post-processed files from the PET Unified Pipeline (PUP) are also available in OASIS-3. OASIS-3 also contains post-processed imaging data such as volumetric segmentations and PET analyses. Imaging data is accompanied by dementia and APOE status and longitudinal clinical and cognitive outcomes. OASIS-3 is available as an open access data set to the scientific community to answer questions related to healthy aging and dementia.
Article
Full-text available
Decrement in processing speed (PS) is a primary cognitive morbidity in clinical populations and could significantly influence other cognitive functions, such as attention and memory. Verifying the usefulness of connectome-based models for predicting neurocognitive abilities has significant translational implications on clinical and aging research. In this study, we verified that resting-state functional connectivity could be used to predict PS in 99 older adults by using connectome-based predictive modeling (CPM). We identified two distinct connectome patterns across the whole brain: the Fast-PS and Slow-PS networks. Relative to the slow-PS network, the fast-PS network showed more within-network connectivity in the motor and visual networks and less between-network connectivity in the motor-visual, motor-subcortical/cerebellum and motor-frontoparietal networks. We further verified that the connectivity patterns for prediction of PS were also useful for predicting attention and memory in the same sample. To test the generalizability and specificity of the connectome-based predictive models, we applied these two connectome models to an independent sample of three age groups (101 younger adults, 103 middle-aged adults and 91 older adults) and confirmed these models could specifically be generalized to predict PS of the older adults, but not the younger and middle-aged adults. Taking all the findings together, the identified connectome-based predictive models are strong for predicting PS in older adults. The application of CPM to predict neurocognitive abilities can complement conventional neurocognitive assessments, bring significant clinical benefits to patient management and aid the clinical diagnoses, prognoses and management of people undergoing the aging process.
Article
Full-text available
Federated learning (FL) is a distributed deep learning method that enables multiple participants, such as mobile and IoT devices, to contribute a neural network while their private training data remains in local devices. This distributed approach is promising in the mobile systems where have a large corpus of decentralized data and require high privacy. However, unlike the common datasets, the data distribution of the mobile systems is imbalanced which will increase the bias of model. In this paper, we demonstrate that the imbalanced distributed training data will cause an accuracy degradation of FL applications. To counter this problem, we build a self-balancing FL framework named Astraea, which alleviates the imbalances by 1) Z-score-based data augmentation, and 2) Mediator based multi-client rescheduling. The proposed framework relieves global imbalance by adaptive data augmentation and downsampling, and for averaging the local imbalance, it creates the mediator to reschedule the training of clients based on Kullback-Leibler divergence (KLD) of their data distribution. Compared with FedAvg, the most popular FL framework, Astraea shows +4.39% and +6.51% improvement of top-1 accuracy on the imbalanced EMNIST and imbalanced CINIC-10 datasets, respectively. Meanwhile, the communication traffic of Astraea is reduced by 75% compared to FedAvg.
Article
Full-text available
Although the feeling of stress is ubiquitous, the neural mechanisms underlying this affective experience remain unclear. Here, we investigate functional hippocampal connectivity throughout the brain during an acute stressor and use machine learning to demonstrate that these networks can specifically predict the subjective feeling of stress. During a stressor, hippocampal connectivity with a network including the hypothalamus (known to regulate physiological stress) predicts feeling more stressed, whereas connectivity with regions such as dorsolateral prefrontal cortex (associated with emotion regulation) predicts less stress. These networks do not predict a subjective state unrelated to stress, and a nonhippocampal network does not predict subjective stress. Hippocampal networks are consistent, specific to the construct of subjective stress, and broadly informative across measures of subjective stress. This approach provides opportunities for relating hypothesis-driven functional connectivity networks to clinically meaningful subjective states. Together, these results identify hippocampal networks that modulate the feeling of stress. Although the feeling of being stressed is ubiquitous and clinically significant, the underlying neural mechanisms are unclear. Using a novel predictive modeling approach, the authors show that functional hippocampal networks specifically and consistently predict the feeling of stress.
Article
Full-text available
Brain imaging has been used to predict language skills during development and neuropathology but its accuracy in predicting language performance in healthy adults has been poorly investigated. To address this shortcoming, we studied the ability to predict reading accuracy and single‐word comprehension scores from rest‐ and task‐based functional magnetic resonance imaging (fMRI) datasets of 424 healthy adults. Using connectome‐based predictive modeling, we identified functional brain networks with >400 edges that predicted language scores and were reproducible in independent data sets. To simplify these complex models we identified the overlapping edges derived from the three task‐fMRI sessions (language, working memory, and motor tasks), and found 12 edges for reading recognition and 11 edges for vocabulary comprehension that accounted for 20% of the variance of these scores, both in the training sample and in the independent sample. The overlapping edges predominantly emanated from language areas within the frontoparietal and default‐mode networks, with a strong precuneus prominence. These findings identify a small subset of edges that accounted for a significant fraction of the variance in language performance that might serve as neuromarkers for neuromodulation interventions to improve language performance or for presurgical planning to minimize language impairments.
Article
Full-text available
The brain-age paradigm is proving increasingly useful for exploring ageing-related disease and can predict important future health outcomes. Most brain-age research utilises structural neuroimaging to index brain volume. However, ageing affects multiple aspects of brain structure and function, which can be examined using multi-modality neuroimaging. Using UK Biobank, brain-age was modelled in n=2,205 healthy people with T1-weighted MRI, T2-FLAIR, T2*, diffusion-MRI, task fMRI and resting-state fMRI. In a held-out healthy validation set (n=520), chronological age was accurately predicted (r=0.78, mean absolute error=3.55 years) using LASSO regression, higher than using any modality separately. Thirty-four neuroimaging phenotypes were deemed informative by the regression (after bootstrapping); predominantly grey-matter volume and white-matter microstructure measures. When applied to new individuals from UK Biobank (n=14,701), significant associations with multi-modality brain-predicted age difference (brain-PAD) were found for: stroke history, diabetes diagnosis, smoking, alcohol intake and some, but not all, cognitive measures (corrected p<0.05). Multi-modality neuroimaging can improve brain-age prediction, and derived brain-PAD values are sensitive to biomedical and lifestyle factors that negatively impact brain and cognitive health.
Article
Deep learning has emerged as a powerful approach to constructing imaging signatures of normal brain ageing as well as of various neuropathological processes associated with brain diseases. In particular, MRI-derived brain age has been used as a comprehensive biomarker of brain health that can identify both advanced and resilient ageing individuals via deviations from typical brain ageing. Imaging signatures of various brain diseases, including schizophrenia and Alzheimer's disease, have also been identified using machine learning. Prior efforts to derive these indices have been hampered by the need for sophisticated and not easily reproducible processing steps, by insufficiently powered or diversified samples from which typical brain ageing trajectories were derived, and by limited reproducibility across populations and MRI scanners. Herein, we develop and test a sophisticated deep brain network (DeepBrainNet) using a large (n = 11 729) set of MRI scans from a highly diversified cohort spanning different studies, scanners, ages and geographic locations around the world. Tests using both cross-validation and a separate replication cohort of 2739 individuals indicate that DeepBrainNet obtains robust brain-age estimates from these diverse datasets without the need for specialized image data preparation and processing. Furthermore, we show evidence that moderately fit brain ageing models may provide brain age estimates that are most discriminant of individuals with pathologies. This is not unexpected as tightly-fitting brain age models naturally produce brain-age estimates that offer little information beyond age, and loosely fitting models may contain a lot of noise. Our results offer some experimental evidence against commonly pursued tightly-fitting models. We show that the moderately fitting brain age models obtain significantly higher differentiation compared to tightly-fitting models in two of the four disease groups tested. Critically, we demonstrate that leveraging DeepBrainNet, along with transfer learning, allows us to construct more accurate classifiers of several brain diseases, compared to directly training classifiers on patient versus healthy control datasets or using common imaging databases such as ImageNet. We, therefore, derive a domain-specific deep network likely to reduce the need for application-specific adaptation and tuning of generic deep learning networks. We made the DeepBrainNet model freely available to the community for MRI-based evaluation of brain health in the general population and over the lifespan.