Multimodal MRI data boosts the performance of brain-based cognition prediction
Poster No:
1098
Submission Type:
Abstract Submission
Authors:
Jianxiao Wu1, Jingwei Li1, Kyesam Jung2, Simon Eickhoff2, Sarah Genon1
Institutions:
1Research Center Jülich, Jülich, Germany, 2Research Centre Jülich, Jülich, Germany
First Author:
Co-Author(s):
Introduction:
Brain-based behavior prediction is an increasingly popular field, with promising prospects as well as serious challenges (Wu et al., 2023). An important limitation to address is the overall low prediction accuracies. Potentially, combining a range of neuroimaging features probing different, albeit complementary neurobiological aspects could help to improve the power of brain-based prediction models (Wu et al., 2023). Nevertheless, existing studies utilizing multimodal neuroimaging data do not always report improvements in prediction performance, nor agree on what features are useful (Jiang et al., 2020; Xiao et al., 2021; Ooi et al., 2022). In this study, we aim to investigate the usefulness of different neuroimaging features from different imaging modalities for the prediction of cognition, in heterogeneous populations.
Methods:
Data from the Human Connectome Project Aging (HCP-A) (Bookheimer et al., 2019) and Human Connectome Project Development (Somerville et al., 2018) were used, representing cohorts with distinct age ranges (36-100 and 5-22 years old respectively). The imaging data were processed with the HCP minimal processing pipeline and ICA-AROMA. A total of 24 types of features were extracted from each dataset, based on functional, anatomical, and diffusion MRI. Each imaging feature is computed based on brain regions represented by the 300-parcel Schaefer cortical atlas (Schaefer et al., 2018) and the 50-parcel Melbourne subcortex atlas (Tian et al., 2020). Nine cognition measures available in both datasets were used as prediction targets, representing either overall cognition or specific aspects of cognition.
First, feature-wise predictions were carried out with elastic net using each type of features separately. Then, the feature-wise prediction outcomes were combined in a stacking model using Random Forest regression, implemented separately for different number of features where only the top 2 to 24 feature types (based on training accuracies) were used respectively. Prediction accuracy was evaluated by the Pearson correlation between predicted and true values, averaged across 10 repeats of 10-fold cross-validation. For each prediction target, we defined a best feature set based on the stacking model with the numerically highest accuracy, and a necessary feature set based on the model with the fewest feature types that achieved statistically comparable accuracy as the best feature set.
First, feature-wise predictions were carried out with elastic net using each type of features separately. Then, the feature-wise prediction outcomes were combined in a stacking model using Random Forest regression, implemented separately for different number of features where only the top 2 to 24 feature types (based on training accuracies) were used respectively. Prediction accuracy was evaluated by the Pearson correlation between predicted and true values, averaged across 10 repeats of 10-fold cross-validation. For each prediction target, we defined a best feature set based on the stacking model with the numerically highest accuracy, and a necessary feature set based on the model with the fewest feature types that achieved statistically comparable accuracy as the best feature set.
Results:
Prediction accuracies increase monotonically with the number of feature types in general, albeit mostly plateauing around 5 feature types (Fig 1A). Making use of all features, most cognition measures can be predicted with accuracies of 0.4 to 0.6 (Fig 1B). In both datasets, composite measures showed better prediction performance compared to individual cognitive test measures.
Comparing the best feature sets and the necessary feature sets shows that while varying number of features are required to achieve the numerically best accuracies, mostly fewer than 5 feature types are required to achieve comparable performance (Fig 2A). The composition of these sparse necessary feature sets across prediction targets is shown in Fig 2B. The most commonly required feature types are resting-state and task based connectivity features. Other often required feature types include cortical surface area related features and structural connectivity.
Comparing the best feature sets and the necessary feature sets shows that while varying number of features are required to achieve the numerically best accuracies, mostly fewer than 5 feature types are required to achieve comparable performance (Fig 2A). The composition of these sparse necessary feature sets across prediction targets is shown in Fig 2B. The most commonly required feature types are resting-state and task based connectivity features. Other often required feature types include cortical surface area related features and structural connectivity.
Conclusions:
We examined the contribution of a large range of neuroimaging features from multiple imaging modalities to the prediction of cognition phenotypes in developmental and aging populations. Overall, we found that only a sparse set of features is necessary for accurate prediction of cognition, achieving similar accuracies as the state-of-the-art studies. Contrary to previous studies (Xiao et al., 2021; Ooi et al., 2022), our results suggest that multimodal data from not only functional MRI, but also anatomical and diffusion MRI are useful for prediction of cognition in developmental and aging populations.
Modeling and Analysis Methods:
Classification and Predictive Modeling 1
Connectivity (eg. functional, effective, structural) 2
fMRI Connectivity and Network Modeling
Keywords:
Cognition
FUNCTIONAL MRI
Machine Learning
MRI
Open Data
STRUCTURAL MRI
WHITE MATTER IMAGING - DTI, HARDI, DSI, ETC
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
For human MRI, what field strength scanner do you use?
Which processing packages did you use for your study?
Provide references using APA citation style.
• Jiang R., Calhoun V.D., Cui Y., Qi S., Zhuo C., Li J., Jung R., Yang J., Du Y., Jiang T., Sui J. (2020). Multimodal data revealed different neurobiological correlates of intelligence between males and females. Brain Imaging and Behavior, 14, 1979-1993.
• Ooi L.Q.R., Chen J., Zhang S., Kong R., Tam A., Li J., Dhamala E., Zhou J.H., Holmes A.J., Yeo B.T.T. (2022). Comparison of individualized behavioral predictions across anatomical, diffusion and functional connectivity MRI. NeuroImage, 263, 119636.
• Schaefer A., Kong R., Gordon E.M., Laumann T.O., Zuo X-N., Holmes A.J., Eickhoff S.B., Yeo B.T.T. (2018). Local-global parcellation of the human cerebral cortex from intrinsic functional connectivity MRI. Cerebral Cortex, 28, 3095-3114.
• Somerville L.H., Bookheimer S.Y., Buckner R.L., Burges G.C., Curtiss S.W., Dapretto M., Elam J.S., Gaffrey M.S., Harms M.P., Hodge C., et al. (2018). The Lifespan Human Connectome Project in development: A large-scale study of brain connectivity development in 5-21 year olds. NeuroImage, 183, 456-468.
• Tian Y., Margulies D.S., Breakspear M., Zalesky A. (2020). Topographic organization of the human subcortex unveiled with functional connectivity gradients. Nature Neuroscience, 23, 1421-1432.
• Wu J., Eickhoff S.B., Scheinost D., Genon S. (2023). The challenges and prospects of brain-based prediction of behaviour. Nature Human Behaviour, 7, 1255-1264.
• Xiao Y., Lin Y., Ma J., Qian J., Ke Z., Li L., Yi Y., Zhang J., Cam-CAN, Dai Z. (2021). ‘Predicting visual working memory with multimodal magnetic resonance imaging’. Human Brain Mapping, 42, 1446-1462.