tfMRI Data Research Articles

TS-AI: A deep learning pipeline for multimodal subject-specific parcellation with task contrasts synthesis

Accurate mapping of brain functional subregions at an individual level is crucial. Task-based functional MRI (tfMRI) captures subject-specific activation patterns during various functions and behaviors, facilitating the individual localization of functionally distinct subregions. However, acquiring high-quality tfMRI is time-consuming and resource-intensive in both scientific and clinical settings. The present study proposes a two-stage network model, TS-AI, to individualize an atlas on cortical surfaces through the prediction of tfMRI data. TS-AI first synthesizes a battery of task contrast maps for each individual by leveraging tract-wise anatomical connectivity and resting-state networks. These synthesized maps, along with feature maps of tract-wise anatomical connectivity and resting-state networks, are then fed into an end-to-end deep neural network to individualize an atlas. TS-AI enables the synthesized task contrast maps to be used in individual parcellation without the acquisition of actual task fMRI scans. In addition, a novel feature consistency loss is designed to assign vertices with similar features to the same parcel, which increases individual specificity and mitigates overfitting risks caused by the absence of individual parcellation ground truth. The individualized parcellations were validated by assessing test-retest reliability, homogeneity, and cognitive behavior prediction using diverse reference atlases and datasets, demonstrating the superior performance and generalizability of TS-AI. Sensitivity analysis yielded insights into region-specific features influencing individual variation in functional regionalization. Additionally, TS-AI identified accelerated shrinkage in the medial temporal and cingulate parcels during the progression of Alzheimer's disease, suggesting its potential in clinical research and applications.

A hybrid spatiotemporal deep belief network and sparse representation-based framework reveals multilevel core functional components in decoding multitask fMRI signals.

Decoding human brain activity on various task-based functional brain imaging data is of great significance for uncovering the functioning mechanism of the human mind. Currently, most feature extraction model-based methods for brain state decoding are shallow machine learning models, which may struggle to capture complex and precise spatiotemporal patterns of brain activity from the highly noisy fMRI raw data. Moreover, although decoding models based on deep learning methods benefit from their multilayer structure that could extract spatiotemporal features at multiscale, the relatively large populations of fMRI datasets are indispensable, and the explainability of their results is elusive. To address the above problems, we proposed a computational framework based on hybrid spatiotemporal deep belief network and sparse representations to differentiate multitask fMRI (tfMRI) signals. Using a relatively small cohort of tfMRI data as a test bed, our framework can achieve an average classification accuracy of 97.86% and define the multilevel temporal and spatial patterns of multiple cognitive tasks. Intriguingly, our model can characterize the key components for differentiating the multitask fMRI signals. Overall, the proposed framework can identify the interpretable and discriminative fMRI composition patterns at multiple scales, offering an effective methodology for basic neuroscience and clinical research with relatively small cohorts.

Constrained Decoupled Dictionary Learning of Task fMRI data

AbstractBackgroundDictionary learning techniques have the potential to efficiently detect functional connectivity impairments in distributed brain networks associated with Alzheimer’s Disease (AD) when applied to task functional magnetic resonance imaging (tfMRI) data. However, traditional methods are constrained in what brain networks they can identify due to the statistical independence assumptions they make, and they do not leverage the structure of the tfMRI task when identifying networks. We used a new sparse dictionary learning technique to identify brain functional network correlates of cardiometabolic risk variables in a community‐based cohort.MethodThe Bogalusa Heart Study (BHS) is community‐based cohort study of atherosclerosis and risk factors for cardiovascular disease. fMRI data was collected from 100 participants of BHS during performance of a Stroop task. Spatially concatenated tfMRI data was factorized into a group‐wise temporal dictionary matrix and a set of scan‐level spatial patterns while biasing a subset of learned temporal patterns to be paradigm related. Scan‐level spatial patterns obtained as a result were aggregated and factorized to derive a group‐level spatial dictionary matrix and sparse loading coefficients. The sparse loading coefficients obtained from this Constrained Decoupled Dictionary Learning (CDDL) method were used as descriptive biomarkers, the relevance of which was established by using them to predict cardiometabolic variables including fasting glucose level, homeostatic model assessment for insulin resistance (HOMA‐IR) index and fasting insulin level using support vector regression and a leave‐one out cross‐validation. The performance of the algorithm was calculated in terms of Mean Absolute Percentage Error (MAPE) between actual and predicted cardiometabolic variable valuesResultThe SVM regressor was able to predict the cardiometabolic variables with low MAPE. The prediction of glucose values achieved the least MAPE for all kernel types (Table 1).ConclusionCombining prior knowledge, decoupling of spatial and temporal patterns, and high‐dimensional regression, we generated brain functional network patterns from real‐world tfMRI data that associated with known cardiometabolic indicators of brain health. These brain functional network patterns thus have the potential to serve as indicators of brain health.

Multitask fMRI Data Classification via Group-Wise Hybrid Temporal and Spatial Sparse Representations.

Task-based functional magnetic resonance imaging (tfMRI) has been widely used to induce functional brain activities corresponding to various cognitive tasks. A relatively under-explored question is whether there exist fundamental differences in fMRI signal composition patterns that can effectively classify the task states of tfMRI data, furthermore, whether there exist key functional components in characterizing the diverse tfMRI signals. Recently, fMRI signal composition patterns of multiple tasks have been investigated via deep learning models, where relatively large populations of fMRI datasets are indispensable and the neurologic meaning of their results is elusive. Thus, the major challenges arise from the high dimensionality, low signal-to-noise ratio, interindividual variability, a small sample size of fMRI data, and the explainability of classification results. To address the above challenges, we proposed a computational framework based on group-wise hybrid temporal and spatial sparse representations (HTSSR) to identify and differentiate multitask fMRI signal composition patterns. Using relatively small cohorts of Human Connectome Project (HCP) tfMRI data as test-bed, the experimental results demonstrated that the multitask of fMRI data can be successfully classified with an average accuracy of 96.67%, where the key components in differentiating the multitask can be characterized, suggesting the effectiveness and explainability of the proposed method. Moreover, both task-related components and resting-state networks (RSNs) can be reliably detected. Therefore, our study proposed a novel framework that identifies the interpretable and discriminative fMRI composition patterns and can be potentially applied for controlling fMRI data quality and inferring biomarkers in brain disorders with small sample neuroimaging datasets.

Improving the accuracy of brain activation maps in the group-level analysis of fMRI data utilizing spatiotemporal Gaussian process model

ObjectiveAccuracy and precision of the statistical analysis methods used for brain activation maps are essential. Adjusting models to consider spatiotemporal correlation embedded in fMRI data may increase their accuracy, but it also introduces a high computational cost. The present study aimed to apply and assess the spatiotemporal Gaussian process (STGP) model to improve accuracy and reduce cost. MethodsWe applied the spatiotemporal Gaussian process (STGP) model for both simulated and experimental memory tfMRI data and compared the findings with fast, fully Bayesian, and General Linear Models (GLM). To assess their accuracy and precision, the models were fitted to the simulated data (1000 voxels,100 times point for 50 people), and an average of accuracy indexes of 100 repetitions was computed. Functional and activation maps for all models were calculated in experimental data analysis. ResultsSTGP model resulted in a higher Z-score in the whole brain, in the 1000 most activated voxels, and in the frontal lobe as the approved memory area. Based on the simulated data, the STGP model showed more accuracy and precision than the other two models. However, its computational time was more than the GLM, as the price of model correction, but much less than that of the fast, fully Bayesian model. ConclusionSpatiotemporal correlation further improved the accuracy of the STGP compared to the GLM and fast, fully Bayesian model. This can result in more accurate activation maps. Moreover, the STGP model’s computational speed appears to be reasonable for model application.

Supervised Brain Network Learning Based on Deep Recurrent Neural Networks

Task-based functional magnetic resonance imaging (tfMRI) is a widely used neuroimaging technique in exploring brain networks and functions associated with cognitive behaviors. Traditionally, the general linear model (GLM) is the most popular method in tfMRI data analysis due to its simpleness and robustness. This model-driven method adopts a canonical hemodynamic response function (HRF) and its various derivatives to construct regressors in the design matrix and estimate changes in the tfMRI data. However, a possible limitation of current model-driven methods is that the HRF is fixed and non-adaptive which may overlook other diverse and concurrent brain networks. In order to overcome these limitations, we proposed a novel hybrid framework, supervised brain network learning based on deep recurrent neural networks (SUDRNN), to reconstruct the diverse and concurrent functional brain networks. Specifically, this hybrid framework first takes advantage of the great capacity of deep recurrent neural networks (DRNN) in modeling sequential data to learn the diverse regressors from real tfMRI data. After that, it utilizes the effective supervised dictionary learning (SDL) method to reconstruct both the task-related functional brain networks and other latent brain networks simultaneously. Extensive experiment results on different tfMRI datasets from Human connectome project (HCP) demonstrated the superiority of the proposed framework.

Manifold Learning of Dynamic Functional Connectivity Reliably Identifies Functionally Consistent Coupling Patterns in Human Brains.

Most previous work on dynamic functional connectivity (dFC) has focused on analyzing temporal traits of functional connectivity (similar coupling patterns at different timepoints), dividing them into functional connectivity states and detecting their between-group differences. However, the coherent functional connectivity of brain activity among the temporal dynamics of functional connectivity remains unknown. In the study, we applied manifold learning of local linear embedding to explore the consistent coupling patterns (CCPs) that reflect functionally homogeneous regions underlying dFC throughout the entire scanning period. By embedding the whole-brain functional connectivity in a low-dimensional manifold space based on the Human Connectome Project (HCP) resting-state data, we identified ten stable patterns of functional coupling across regions that underpin the temporal evolution of dFC. Moreover, some of these CCPs exhibited significant neurophysiological meaning. Furthermore, we apply this method to HCP rsfMR and tfMRI data as well as sleep-deprivation data and found that the topological organization of these low-dimensional structures has high potential for predicting sleep-deprivation states (classification accuracy of 92.3%) and task types (100% identification for all seven tasks).In summary, this work provides a methodology for distilling coherent low-dimensional functional connectivity structures in complex brain dynamics that play an important role in performing tasks or characterizing specific states of the brain.

Modeling Brain Diverse and Complex Hemodynamic Response Patterns via Deep Recurrent Autoencoder

For decades, a variety of task-based functional MRI (tfMRI) data analysis approaches have been developed, including the general linear model (GLM), sparse representations, and independent component analysis (ICA). However, these methods are mainly shallow models and are limited in faithfully modeling the complex, diverse, and concurrent spatial–temporal functional brain activities. Recently, recurrent neural networks (RNNs) have demonstrated great superiority in modeling temporal dependency of signals, while autoencoder models have been proven to be effective in automatically estimating the optimal representations of the original data. These characteristics of RNNs and autoencoders naturally meet the requirement of modeling hemodynamic response patterns in tfMRI data. Thus, we propose a novel unsupervised framework of deep recurrent autoencoder (DRAE) for modeling hemodynamic response patterns in this article. The basic idea of the DRAE model is to combine the deep RNN and the autoencoder to automatically characterize the meaningful functional brain networks and corresponding diverse and complex hemodynamic response patterns simultaneously. The experimental results demonstrate the superiority of the proposed DRAE model in automatically estimating the diverse and complex hemodynamic response patterns.

Fast and scalable distributed deep convolutional autoencoder for fMRI big data analytics.

In recent years, analyzing task-based fMRI (tfMRI) data has become an essential tool for understanding brain function and networks. However, due to the sheer size of tfMRI data, its intrinsic complex structure, and lack of ground truth of underlying neural activities, modeling tfMRI data is hard and challenging. Previously proposed data modeling methods including Independent Component Analysis (ICA) and Sparse Dictionary Learning only provided shallow models based on blind source separation under the strong assumption that original fMRI signals could be linearly decomposed into time series components with corresponding spatial maps. Given the Convolutional Neural Network (CNN) successes in learning hierarchical abstractions from low-level data such as tfMRI time series, in this work we propose a novel scalable distributed deep CNN autoencoder model and apply it for fMRI big data analysis. This model aims to both learn the complex hierarchical structures of the tfMRI big data and to leverage the processing power of multiple GPUs in a distributed fashion. To deploy such a model, we have created an enhanced processing pipeline on the top of Apache Spark and Tensorflow, leveraging from a large cluster of GPU nodes over cloud. Experimental results from applying the model on the Human Connectome Project (HCP) data show that the proposed model is efficient and scalable toward tfMRI big data modeling and analytics, thus enabling data-driven extraction of hierarchical neuroscientific information from massive fMRI big data.

Modeling Task fMRI Data Via Deep Convolutional Autoencoder.

Task-based functional magnetic resonance imaging (tfMRI) has been widely used to study functional brain networks under task performance. Modeling tfMRI data is challenging due to at least two problems: the lack of the ground truth of underlying neural activity and the highly complex intrinsic structure of tfMRI data. To better understand brain networks based on fMRI data, data-driven approaches have been proposed, for instance, independent component analysis (ICA) and sparse dictionary learning (SDL). However, both ICA and SDL only build shallow models, and they are under the strong assumption that original fMRI signal could be linearly decomposed into time series components with their corresponding spatial maps. As growing evidence shows that human brain function is hierarchically organized, new approaches that can infer and model the hierarchical structure of brain networks are widely called for. Recently, deep convolutional neural network (CNN) has drawn much attention, in that deep CNN has proven to be a powerful method for learning high-level and mid-level abstractions from low-level raw data. Inspired by the power of deep CNN, in this paper, we developed a new neural network structure based on CNN, called deep convolutional auto-encoder (DCAE), in order to take the advantages of both data-driven approach and CNN's hierarchical feature abstraction ability for the purpose of learning mid-level and high-level features from complex, large-scale tfMRI time series in an unsupervised manner. The DCAE has been applied and tested on the publicly available human connectome project tfMRI data sets, and promising results are achieved.

Interpreting Intervention Induced Neuroplasticity with fMRI: The Case for Multimodal Imaging Strategies

Direct measurement of recovery from brain injury is an important goal in neurorehabilitation, and requires reliable, objective, and interpretable measures of changes in brain function, referred to generally as “neuroplasticity.” One popular imaging modality for measuring neuroplasticity is task-based functional magnetic resonance imaging (t-fMRI). In the field of neurorehabilitation, however, assessing neuroplasticity using t-fMRI presents a significant challenge. This commentary reviews t-fMRI changes commonly reported in patients with cerebral palsy or acquired brain injuries, with a focus on studies of motor rehabilitation, and discusses complexities surrounding their interpretations. Specifically, we discuss the difficulties in interpreting t-fMRI changes in terms of their underlying causes, that is, differentiating whether they reflect genuine reorganisation, neurological restoration, compensation, use of preexisting redundancies, changes in strategy, or maladaptive processes. Furthermore, we discuss the impact of heterogeneous disease states and essential t-fMRI processing steps on the interpretability of activation patterns. To better understand therapy-induced neuroplastic changes, we suggest that researchers utilising t-fMRI consider concurrently acquiring information from an additional modality, to quantify, for example, haemodynamic differences or microstructural changes. We outline a variety of such supplementary measures for investigating brain reorganisation and discuss situations in which they may prove beneficial to the interpretation of t-fMRI data.

Supervised dictionary learning for inferring concurrent brain networks.

Task-based fMRI (tfMRI) has been widely used to explore functional brain networks via predefined stimulus paradigm in the fMRI scan. Traditionally, the general linear model (GLM) has been a dominant approach to detect task-evoked networks. However, GLM focuses on task-evoked or event-evoked brain responses and possibly ignores the intrinsic brain functions. In comparison, dictionary learning and sparse coding methods have attracted much attention recently, and these methods have shown the promise of automatically and systematically decomposing fMRI signals into meaningful task-evoked and intrinsic concurrent networks. Nevertheless, two notable limitations of current data-driven dictionary learning method are that the prior knowledge of task paradigm is not sufficiently utilized and that the establishment of correspondences among dictionary atoms in different brains have been challenging. In this paper, we propose a novel supervised dictionary learning and sparse coding method for inferring functional networks from tfMRI data, which takes both of the advantages of model-driven method and data-driven method. The basic idea is to fix the task stimulus curves as predefined model-driven dictionary atoms and only optimize the other portion of data-driven dictionary atoms. Application of this novel methodology on the publicly available human connectome project (HCP) tfMRI datasets has achieved promising results.

Open Neuroscience Solutions for the Connectome-wide Association Era

The neuroimaging community is at a crossroads. Long characterized by individualism, the data and computational and analytic needs of the connectome-wide association era necessitate cultural reform. Emerging initiatives have demonstrated the feasibility and utility of adopting an open neuroscience model to accelerate the pace and success of scientific discovery.