Modeling motor task activation from resting-state fMRI using machine learning in individual subjects

Neural Representation of Collective Self-esteem in Resting-state Functional Connectivity and its Validation in Task-dependent Modality

INTRODUCTION

Collective self-esteem (CSE) is an important personality variable, defined as self-worth derived from membership in social groups. A study explored the neural basis of CSE using a task-based functional magnetic resonance imaging (fMRI) paradigm; however, task-independent neural basis of CSE remains to be explored, and whether the CSE neural basis of resting-state fMRI is consistent with that of task-based fMRI is unclear.

METHODS

We built support vector regression (SVR) models to predict CSE scores using topological metrics measured in the resting-state functional connectivity network (RSFC) as features. Then, to test the reliability of the SVR analysis, the activation pattern of the identified brain regions from SVR analysis was used as features to distinguish collective self-worth from other conditions by multivariate pattern classification in task-based fMRI dataset.

RESULTS

SVR analysis results showed that leverage centrality successfully decoded the individual differences in CSE. The ventromedial prefrontal cortex, anterior cingulate cortex, posterior cingulate gyrus, precuneus, orbitofrontal cortex, posterior insula, postcentral gyrus, inferior parietal lobule, temporoparietal junction, and inferior frontal gyrus, which are involved in self-referential processing, affective processing, and social cognition networks, participated in this prediction. Multivariate pattern classification analysis found that the activation pattern of the identified regions from the SVR analysis successfully distinguished collective self-worth from relational self-worth, personal self-worth and semantic control.

CONCLUSION

Our findings revealed CSE neural basis in the whole-brain RSFC network, and established the concordance between leverage centrality and the activation pattern (evoked during collective self-worth task) of the identified regions in terms of representing CSE.

3T vs. 7T fMRI: capturing early human memory consolidation after motor task utilizing the observed higher functional specificity of 7T

Objective

Functional magnetic resonance imaging (fMRI) visualizes brain structures at increasingly higher resolution and better signal-to-noise ratio (SNR) as field strength increases. Yet, mapping the blood oxygen level dependent (BOLD) response to distinct neuronal processes continues to be challenging. Here, we investigated the characteristics of 7 T-fMRI compared to 3 T-fMRI in the human brain beyond the effect of increased SNR and verified the benefits of 7 T-fMRI in the detection of tiny, highly specific modulations of functional connectivity in the resting state following a motor task.

Methods

18 healthy volunteers underwent two resting state and a stimulus driven measurement using a finger tapping motor task at 3 and 7 T, respectively. The SNR for each field strength was adjusted by targeted voxel size variation to minimize the effect of SNR on the field strength specific outcome. Spatial and temporal characteristics of resting state ICA, network graphs, and motor task related activated areas were compared. Finally, a graph theoretical approach was used to detect resting state modulation subsequent to a simple motor task.

Results

Spatial extensions of resting state ICA and motor task related activated areas were consistent between field strengths, but temporal characteristics varied, indicating that 7 T achieved a higher functional specificity of the BOLD response than 3 T-fMRI. Following the motor task, only 7 T-fMRI enabled the detection of highly specific connectivity modulations representing an "offline replay" of previous motor activation. Modulated connections of the motor cortex were directly linked to brain regions associated with memory consolidation.

Conclusion

These findings reveal how memory processing is initiated even after simple motor tasks, and that it begins earlier than previously shown. Thus, the superior capability of 7 T-fMRI to detect subtle functional dynamics promises to improve diagnostics and therapeutic assessment of neurological diseases.

Generalizing prediction of task-evoked brain activity across datasets and populations

Predictions of task-based functional magnetic resonance imaging (fMRI) from task-free resting-state (rs) fMRI have gained popularity over the past decade. This method holds a great promise for studying individual variability in brain function without the need to perform highly demanding tasks. However, in order to be broadly used, prediction models must prove to generalize beyond the dataset they were trained on. In this work, we test the generalizability of prediction of task-fMRI from rs-fMRI across sites, MRI vendors and age-groups. Moreover, we investigate the data requirements for successful prediction. We use the Human Connectome Project (HCP) dataset to explore how different combinations of training sample sizes and number of fMRI datapoints affect prediction success in various cognitive tasks. We then apply models trained on HCP data to predict brain activations in data from a different site, a different MRI vendor (Phillips vs. Siemens scanners) and a different age group (children from the HCP-development project). We demonstrate that, depending on the task, a training set of approximately 20 participants with 100 fMRI timepoints each yields the largest gain in model performance. Nevertheless, further increasing sample size and number of timepoints results in significantly improved predictions, until reaching approximately 450-600 training participants and 800-1000 timepoints. Overall, the number of fMRI timepoints influences prediction success more than the sample size. We further show that models trained on adequate amounts of data successfully generalize across sites, vendors and age groups and provide predictions that are both accurate and individual-specific. These findings suggest that large-scale publicly available datasets may be utilized to study brain function in smaller, unique samples.

BolT: Fused window transformers for fMRI time series analysis

Deep-learning models have enabled performance leaps in analysis of high-dimensional functional MRI (fMRI) data. Yet, many previous methods are suboptimally sensitive for contextual representations across diverse time scales. Here, we present BolT, a blood-oxygen-level-dependent transformer model, for analyzing multi-variate fMRI time series. BolT leverages a cascade of transformer encoders equipped with a novel fused window attention mechanism. Encoding is performed on temporally-overlapped windows within the time series to capture local representations. To integrate information temporally, cross-window attention is computed between base tokens in each window and fringe tokens from neighboring windows. To gradually transition from local to global representations, the extent of window overlap and thereby number of fringe tokens are progressively increased across the cascade. Finally, a novel cross-window regularization is employed to align high-level classification features across the time series. Comprehensive experiments on large-scale public datasets demonstrate the superior performance of BolT against state-of-the-art methods. Furthermore, explanatory analyses to identify landmark time points and regions that contribute most significantly to model decisions corroborate prominent neuroscientific findings in the literature.

Machine learning may predict individual hand motor activation from resting-state fMRI in patients with brain tumors in perirolandic cortex

OBJECTIVE

The study aimed to evaluate the predictive validity of the neural network (NN) method for presurgical mapping of motor areas using resting-state functional MRI (rs-fMRI) data of patients with brain tumor located in the perirolandic cortex (PRC).

METHODS

A total of 109 patients with brain tumors occupying PRC underwent rs-fMRI and hand movement task-based fMRI (tb-fMRI) scans. Using a NN model trained on fMRI data of 47 healthy controls, individual task activation maps were predicted from their rs-fMRI data. NN-predicted maps were compared with task activation and independent component analysis (ICA)-derived maps. Spatial Pearson's correlation coefficients (CC) matrices and Dice coefficients (DC) between task activation and predicted activation using NN (DC_{NN_Act}) and ICA (DC_{ICA_Act}) were calculated and compared using non-parametric tests. The effects of tumor types and head motion on predicted maps were demonstrated.

RESULTS

The CC matrix of NN-predicted maps showed higher diagonal values compared with ICA-derived maps (p < 0.001). DC_{NN_Act} were higher than DC_{ICA_Act} (p < 0.001) for patients with or without motor deficits. Lower DCs were found in subjects with head motion greater than one voxel. DCs were higher on the nontumor side than on the tumor side (p < 0.001), especially in the glioma group compared with meningioma and metastatic groups.

CONCLUSIONS

This study indicated that the NN approach could predict individual motor activation using rs-fMRI data and could have promising clinical applications in brain tumor patients with anatomical and functional reorganizations.

KEY POINTS

• The neural network machine learning approach successfully predicted hand motor activation in patients with a tumor in the perirolandic cortex, despite space-occupying effects and possible functional reorganization. • Compared to the conventional independent component analysis, the neural network approach utilizing resting-state fMRI data yielded a higher correlation to the active task hand activation data. • The Dice coefficient of machine learning-predicted activation vs. task fMRI activation was different between tumor and nontumor side, also between tumor types, which might indicate different effects of possible neurovascular uncoupling on resting-state and task fMRI.

Jumping over baselines with new methods to predict activation maps from resting-state fMRI

Cognitive fMRI research primarily relies on task-averaged responses over many subjects to describe general principles of brain function. Nonetheless, there exists a large variability between subjects that is also reflected in spontaneous brain activity as measured by resting state fMRI (rsfMRI). Leveraging this fact, several recent studies have therefore aimed at predicting task activation from rsfMRI using various machine learning methods within a growing literature on 'connectome fingerprinting'. In reviewing these results, we found lack of an evaluation against robust baselines that reliably supports a novelty of predictions for this task. On closer examination to reported methods, we found most underperform against trivial baseline model performances based on massive group averaging when whole-cortex prediction is considered. Here we present a modification to published methods that remedies this problem to large extent. Our proposed modification is based on a single-vertex approach that replaces commonly used brain parcellations. We further provide a summary of this model evaluation by characterizing empirical properties of where prediction for this task appears possible, explaining why some predictions largely fail for certain targets. Finally, with these empirical observations we investigate whether individual prediction scores explain individual behavioral differences in a task.