Iterative approach of dual regression with a sparse prior enhances the performance of independent component analysis for group functional magnetic resonance imaging (fMRI) data

Transfer Learning of Deep Neural Networks Pretrained using the ABCD Dataset for General Psychopathology Prediction in Korean Adolescents

BACKGROUND

This study examines whether a deep neural network (DNN), trained to predict the general psychopathology factor (p-factor) using functional magnetic resonance imaging (fMRI) data from adolescents in the Adolescent Brain Cognitive Development (ABCD) study, generalizes to Korean adolescents.

METHOD

We trained a scanner-generalization neural network (SGNN) to predict p-factor scores from resting-state functional connectivity (RSFC) data of 6,905 ABCD adolescents, controlling for MRI scanner-related confounds. We then transferred the pretrained SGNN to a DNN to predict p-factor scores for 125 adolescents, including healthy individuals and those with major depressive disorder, using data from Seoul National University Hospital (SNUH). We compared the transferred DNN's performance with that of kernel ridge regression (KRR) and a baseline DNN.

RESULTS

The transferred DNN outperformed KRR (0.17 ± 0.16; 0.60 ± 0.07) and the baseline DNN (0.17 ± 0.16; 0.69 ± 0.11), achieving a higher Pearson's correlation coefficient (0.29 ± 0.18) and lower mean absolute error (0.59 ± 0.09; p < 0.005). We identified the default mode network (DMN) and visual network (VIS) as crucial functional networks (FNs) for predicting p-factors across both datasets. The dorsal attention network was specific to ABCD, while the cingulo-opercular and ventral attention networks were specific to SNUH.

CONCLUSION

The transferred SGNN successfully generalized to Korean adolescents. Altered RSFC in the DMN and VIS may serve as promising biomarkers for p-factor prediction across diverse populations, addressing heterogeneity in demographics, diagnoses, and MRI scanner characteristics.

Multidimensional inhibitory signatures of sentential negation in behavioral variant frontotemporal dementia

BACKGROUND

Processing of linguistic negation has been associated to inhibitory brain mechanisms. However, no study has tapped this link via multimodal measures in patients with core inhibitory alterations, a critical approach to reveal direct neural correlates and potential disease markers.

METHODS

Here we examined oscillatory, neuroanatomical, and functional connectivity signatures of a recently reported Go/No-go negation task in healthy controls and behavioral variant frontotemporal dementia (bvFTD) patients, typified by primary and generalized inhibitory disruptions. To test for specificity, we also recruited persons with Alzheimer's disease (AD), a disease involving frequent but nonprimary inhibitory deficits.

RESULTS

In controls, negative sentences in the No-go condition distinctly involved frontocentral delta (2-3 Hz) suppression, a canonical inhibitory marker. In bvFTD patients, this modulation was selectively abolished and significantly correlated with the volume and functional connectivity of regions supporting inhibition (e.g. precentral gyrus, caudate nucleus, and cerebellum). Such canonical delta suppression was preserved in the AD group and associated with widespread anatomo-functional patterns across non-inhibitory regions.

DISCUSSION

These findings suggest that negation hinges on the integrity and interaction of spatiotemporal inhibitory mechanisms. Moreover, our results reveal potential neurocognitive markers of bvFTD, opening a new agenda at the crossing of cognitive neuroscience and behavioral neurology.

Modeling and augmenting of fMRI data using deep recurrent variational auto-encoder

Objective. Recently, deep learning models have been successfully applied in functional magnetic resonance imaging (fMRI) modeling and associated applications. However, there still exist at least two challenges. Firstly, due to the lack of sufficient data, deep learning models tend to suffer from overfitting in the training process. Secondly, it is still challenging to model the temporal dynamics from fMRI, due to that the brain state is continuously changing over scan time. In addition, existing methods rarely studied and applied fMRI data augmentation.Approach. In this work, we construct a deep recurrent variational auto-encoder (DRVAE) that combined variational auto-encoder and recurrent neural network, aiming to address all of the above mentioned challenges. The encoder of DRVAE can extract more generalized temporal features from assumed Gaussian distribution of input data, and the decoder of DRVAE can generate new data to increase training samples and thus partially relieve the overfitting issue. The recurrent layers in DRVAE are designed to effectively model the temporal dynamics of functional brain activities. LASSO (least absolute shrinkage and selection operator) regression is applied on the temporal features and input fMRI data to estimate the corresponding spatial networks.Main results. Extensive experimental results on seven tasks from HCP dataset showed that the DRVAE and LASSO framework can learn meaningful temporal patterns and spatial networks from both real data and generated data. The results on group-wise data and single subject suggest that the brain activities may follow certain distribution. Moreover, we applied DRVAE on four resting state fMRI datasets from ADHD-200 for data augmentation, and the results showed that the classification performances on augmented datasets have been considerably improved.Significance. The proposed method can not only derive meaningful temporal features and spatial networks from fMRI, but also generate high-quality new data for fMRI data augmentation and associated applications.

Dual Temporal and Spatial Sparse Representation for Inferring Group-Wise Brain Networks From Resting-State fMRI Dataset

Recently, sparse representation has been successfully used to identify brain networks from task-based fMRI dataset. However, when using the strategy to analyze resting-state fMRI dataset, it is still a challenge to automatically infer the group-wise brain networks under consideration of group commonalities and subject-specific characteristics. In the paper, a novel method based on dual temporal and spatial sparse representation (DTSSR) is proposed to meet this challenge. First, the brain functional networks with subject-specific characteristics are obtained via sparse representation with online dictionary learning for the fMRI time series (temporal domain) of each subject. Next, based on the current brain science knowledge, a simple mathematical model is proposed to describe the complex nonlinear dynamic coupling mechanism of the brain networks, with which the group-wise intrinsic connectivity networks (ICNs) can be inferred by sparse representation for these brain functional networks (spatial domain) of all subjects. Experiments on Leiden_2180 dataset show that most group-wise ICNs obtained by the proposed DTSSR are interpretable by current brain science knowledge and are consistent with previous literature reports. The robustness of DTSSR and the reproducibility of the results are demonstrated by experiments on three different datasets (Leiden_2180, Leiden_2200, and our own dataset). The present work also shed new light on exploring the coupling mechanism of brain networks from perspective of information science.

Using Dual Regression to Investigate Network Shape and Amplitude in Functional Connectivity Analyses

Independent Component Analysis (ICA) is one of the most popular techniques for the analysis of resting state FMRI data because it has several advantageous properties when compared with other techniques. Most notably, in contrast to a conventional seed-based correlation analysis, it is model-free and multivariate, thus switching the focus from evaluating the functional connectivity of single brain regions identified a priori to evaluating brain connectivity in terms of all brain resting state networks (RSNs) that simultaneously engage in oscillatory activity. Furthermore, typical seed-based analysis characterizes RSNs in terms of spatially distributed patterns of correlation (typically by means of simple Pearson's coefficients) and thereby confounds together amplitude information of oscillatory activity and noise. ICA and other regression techniques, on the other hand, retain magnitude information and therefore can be sensitive to both changes in the spatially distributed nature of correlations (differences in the spatial pattern or "shape") as well as the amplitude of the network activity. Furthermore, motion can mimic amplitude effects so it is crucial to use a technique that retains such information to ensure that connectivity differences are accurately localized. In this work, we investigate the dual regression approach that is frequently applied with group ICA to assess group differences in resting state functional connectivity of brain networks. We show how ignoring amplitude effects and how excessive motion corrupts connectivity maps and results in spurious connectivity differences. We also show how to implement the dual regression to retain amplitude information and how to use dual regression outputs to identify potential motion effects. Two key findings are that using a technique that retains magnitude information, e.g., dual regression, and using strict motion criteria are crucial for controlling both network amplitude and motion-related amplitude effects, respectively, in resting state connectivity analyses. We illustrate these concepts using realistic simulated resting state FMRI data and in vivo data acquired in healthy subjects and patients with bipolar disorder and schizophrenia.

Task-specific feature extraction and classification of fMRI volumes using a deep neural network initialized with a deep belief network: Evaluation using sensorimotor tasks

Feedforward deep neural networks (DNNs), artificial neural networks with multiple hidden layers, have recently demonstrated a record-breaking performance in multiple areas of applications in computer vision and speech processing. Following the success, DNNs have been applied to neuroimaging modalities including functional/structural magnetic resonance imaging (MRI) and positron-emission tomography data. However, no study has explicitly applied DNNs to 3D whole-brain fMRI volumes and thereby extracted hidden volumetric representations of fMRI that are discriminative for a task performed as the fMRI volume was acquired. Our study applied fully connected feedforward DNN to fMRI volumes collected in four sensorimotor tasks (i.e., left-hand clenching, right-hand clenching, auditory attention, and visual stimulus) undertaken by 12 healthy participants. Using a leave-one-subject-out cross-validation scheme, a restricted Boltzmann machine-based deep belief network was pretrained and used to initialize weights of the DNN. The pretrained DNN was fine-tuned while systematically controlling weight-sparsity levels across hidden layers. Optimal weight-sparsity levels were determined from a minimum validation error rate of fMRI volume classification. Minimum error rates (mean±standard deviation; %) of 6.9 (±3.8) were obtained from the three-layer DNN with the sparsest condition of weights across the three hidden layers. These error rates were even lower than the error rates from the single-layer network (9.4±4.6) and the two-layer network (7.4±4.1). The estimated DNN weights showed spatial patterns that are remarkably task-specific, particularly in the higher layers. The output values of the third hidden layer represented distinct patterns/codes of the 3D whole-brain fMRI volume and encoded the information of the tasks as evaluated from representational similarity analysis. Our reported findings show the ability of the DNN to classify a single fMRI volume based on the extraction of hidden representations of fMRI volumes associated with tasks across multiple hidden layers. Our study may be beneficial to the automatic classification/diagnosis of neuropsychiatric and neurological diseases and prediction of disease severity and recovery in (pre-) clinical settings using fMRI volumes without requiring an estimation of activation patterns or ad hoc statistical evaluation.

Artifact removal in the context of group ICA: A comparison of single-subject and group approaches

Independent component analysis (ICA) has been widely applied to identify intrinsic brain networks from fMRI data. Group ICA computes group-level components from all data and subsequently estimates individual-level components to recapture intersubject variability. However, the best approach to handle artifacts, which may vary widely among subjects, is not yet clear. In this work, we study and compare two ICA approaches for artifacts removal. One approach, recommended in recent work by the Human Connectome Project, first performs ICA on individual subject data to remove artifacts, and then applies a group ICA on the cleaned data from all subjects. We refer to this approach as Individual ICA based artifacts Removal Plus Group ICA (IRPG). A second proposed approach, called Group Information Guided ICA (GIG-ICA), performs ICA on group data, then removes the group-level artifact components, and finally performs subject-specific ICAs using the group-level non-artifact components as spatial references. We used simulations to evaluate the two approaches with respect to the effects of data quality, data quantity, variable number of sources among subjects, and spatially unique artifacts. Resting-state test-retest datasets were also employed to investigate the reliability of functional networks. Results from simulations demonstrate GIG-ICA has greater performance compared with IRPG, even in the case when single-subject artifacts removal is perfect and when individual subjects have spatially unique artifacts. Experiments using test-retest data suggest that GIG-ICA provides more reliable functional networks. Based on high estimation accuracy, ease of implementation, and high reliability of functional networks, we find GIG-ICA to be a promising approach.

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.

The Inclusion of Functional Connectivity Information into fMRI-based Neurofeedback Improves Its Efficacy in the Reduction of Cigarette Cravings

Real-time fMRI (rtfMRI) neurofeedback (NF) facilitates volitional control over brain activity and the modulation of associated mental functions. The NF signals of traditional rtfMRI-NF studies predominantly reflect neuronal activity within ROIs. In this study, we describe a novel rtfMRI-NF approach that includes a functional connectivity (FC) component in the NF signal (FC-added rtfMRI-NF). We estimated the efficacy of the FC-added rtfMRI-NF method by applying it to nicotine-dependent heavy smokers in an effort to reduce cigarette craving. ACC and medial pFC as well as the posterior cingulate cortex and precuneus are associated with cigarette craving and were chosen as ROIs. Fourteen heavy smokers were randomly assigned to receive one of two types of NF: traditional activity-based rtfMRI-NF or FC-added rtfMRI-NF. Participants received rtfMRI-NF training during two separate visits after overnight smoking cessation, and cigarette craving score was assessed. The FC-added rtfMRI-NF resulted in greater neuronal activity and increased FC between the targeted ROIs than the traditional activity-based rtfMRI-NF and resulted in lower craving score. In the FC-added rtfMRI-NF condition, the average of neuronal activity and FC was tightly associated with craving score (Bonferroni-corrected p = .028). However, in the activity-based rtfMRI-NF condition, no association was detected (uncorrected p &gt; .081). Non-rtfMRI data analysis also showed enhanced neuronal activity and FC with FC-added NF than with activity-based NF. These results demonstrate that FC-added rtfMRI-NF facilitates greater volitional control over brain activity and connectivity and greater modulation of mental function than activity-based rtfMRI-NF.

Deep neural network with weight sparsity control and pre-training extracts hierarchical features and enhances classification performance: Evidence from whole-brain resting-state functional connectivity patterns of schizophrenia

Functional connectivity (FC) patterns obtained from resting-state functional magnetic resonance imaging data are commonly employed to study neuropsychiatric conditions by using pattern classifiers such as the support vector machine (SVM). Meanwhile, a deep neural network (DNN) with multiple hidden layers has shown its ability to systematically extract lower-to-higher level information of image and speech data from lower-to-higher hidden layers, markedly enhancing classification accuracy. The objective of this study was to adopt the DNN for whole-brain resting-state FC pattern classification of schizophrenia (SZ) patients vs. healthy controls (HCs) and identification of aberrant FC patterns associated with SZ. We hypothesized that the lower-to-higher level features learned via the DNN would significantly enhance the classification accuracy, and proposed an adaptive learning algorithm to explicitly control the weight sparsity in each hidden layer via L1-norm regularization. Furthermore, the weights were initialized via stacked autoencoder based pre-training to further improve the classification performance. Classification accuracy was systematically evaluated as a function of (1) the number of hidden layers/nodes, (2) the use of L1-norm regularization, (3) the use of the pre-training, (4) the use of framewise displacement (FD) removal, and (5) the use of anatomical/functional parcellation. Using FC patterns from anatomically parcellated regions without FD removal, an error rate of 14.2% was achieved by employing three hidden layers and 50 hidden nodes with both L1-norm regularization and pre-training, which was substantially lower than the error rate from the SVM (22.3%). Moreover, the trained DNN weights (i.e., the learned features) were found to represent the hierarchical organization of aberrant FC patterns in SZ compared with HC. Specifically, pairs of nodes extracted from the lower hidden layer represented sparse FC patterns implicated in SZ, which was quantified by using kurtosis/modularity measures and features from the higher hidden layer showed holistic/global FC patterns differentiating SZ from HC. Our proposed schemes and reported findings attained by using the DNN classifier and whole-brain FC data suggest that such approaches show improved ability to learn hidden patterns in brain imaging data, which may be useful for developing diagnostic tools for SZ and other neuropsychiatric disorders and identifying associated aberrant FC patterns.

Joint sparse representation of brain activity patterns in multi-task fMRI data

A single-task functional magnetic resonance imaging (fMRI) experiment may only partially highlight alterations to functional brain networks affected by a particular disorder. Multivariate analysis across multiple fMRI tasks may increase the sensitivity of fMRI-based diagnosis. Prior research using multi-task analysis in fMRI, such as those that use joint independent component analysis (jICA), has mainly assumed that brain activity patterns evoked by different tasks are independent. This may not be valid in practice. Here, we use sparsity, which is a natural characteristic of fMRI data in the spatial domain, and propose a joint sparse representation analysis (jSRA) method to identify common information across different functional subtraction (contrast) images in data from a multi-task fMRI experiment. Sparse representation methods do not require independence, or that the brain activity patterns be nonoverlapping. We use functional subtraction images within the joint sparse representation analysis to generate joint activation sources and their corresponding sparse modulation profiles. We evaluate the use of sparse representation analysis to capture individual differences with simulated fMRI data and with experimental fMRI data. The experimental fMRI data was acquired from 16 young (age: 19-26) and 16 older (age: 57-73) adults obtained from multiple speech comprehension tasks within subjects, where an independent measure (namely, age in years) can be used to differentiate between groups. Simulation results show that this method yields greater sensitivity, precision, and higher Jaccard indexes (which measures similarity and diversity of the true and estimated brain activation sources) than does the jICA method. Moreover, superiority of the jSRA method in capturing individual differences was successfully demonstrated using experimental fMRI data.

Template based rotation: a method for functional connectivity analysis with a priori templates

Functional connectivity magnetic resonance imaging (fcMRI) is a powerful tool for understanding the network level organization of the brain in research settings and is increasingly being used to study large-scale neuronal network degeneration in clinical trial settings. Presently, a variety of techniques, including seed-based correlation analysis and group independent components analysis (with either dual regression or back projection) are commonly employed to compute functional connectivity metrics. In the present report, we introduce template based rotation,(1) a novel analytic approach optimized for use with a priori network parcellations, which may be particularly useful in clinical trial settings. Template based rotation was designed to leverage the stable spatial patterns of intrinsic connectivity derived from out-of-sample datasets by mapping data from novel sessions onto the previously defined a priori templates. We first demonstrate the feasibility of using previously defined a priori templates in connectivity analyses, and then compare the performance of template based rotation to seed based and dual regression methods by applying these analytic approaches to an fMRI dataset of normal young and elderly subjects. We observed that template based rotation and dual regression are approximately equivalent in detecting fcMRI differences between young and old subjects, demonstrating similar effect sizes for group differences and similar reliability metrics across 12 cortical networks. Both template based rotation and dual-regression demonstrated larger effect sizes and comparable reliabilities as compared to seed based correlation analysis, though all three methods yielded similar patterns of network differences. When performing inter-network and sub-network connectivity analyses, we observed that template based rotation offered greater flexibility, larger group differences, and more stable connectivity estimates as compared to dual regression and seed based analyses. This flexibility owes to the reduced spatial and temporal orthogonality constraints of template based rotation as compared to dual regression. These results suggest that template based rotation can provide a useful alternative to existing fcMRI analytic methods, particularly in clinical trial settings where predefined outcome measures and conserved network descriptions across groups are at a premium.

Integration of structural and functional magnetic resonance imaging improves mild cognitive impairment detection

The identification of mild cognitive impairments (MCI) via either structural magnetic resonance imaging (sMRI) or functional MRI (fMRI) has great potential due to the non-invasiveness of the techniques. Furthermore, these techniques allow longitudinal follow-ups of single subjects via repeated measurements. sMRI- or fMRI-based biomarkers have been adopted separately to diagnose MCI; however, there has not been a systematic effort to integrate sMRI- and fMRI-based features to increase MCI detection accuracy. This study investigated whether the detection of MCI can be improved via the integration of biomarkers identified from both sMRI and fMRI modalities. Regional volume sizes and neuronal activity levels of brains from MCI subjects were compared with those from healthy controls and used to identify biomarkers from sMRI and fMRI data, respectively. In the subsequent classification phase, MCI was automatically detected using a support vector machine algorithm that employed the identified sMRI- and fMRI-based biomarkers as an input feature vector. The results indicate that the fMRI-based biomarkers provided more information for detecting MCI than the sMRI-based biomarkers. Moreover, the integrated feature sets using the sMRI- and fMRI-based biomarkers consistently showed greater detection accuracy than the feature sets based only on the fMRI-based biomarkers. The results demonstrate that integration of sMRI and fMRI modalities can provide supplemental information to improve the diagnosis of MCI relative to either the sMRI or fMRI modalities alone.

Hippocampus-precuneus functional connectivity as an early sign of Alzheimer's disease: a preliminary study using structural and functional magnetic resonance imaging data

Alzheimer's disease (AD) is characterized by structural atrophies in the hippocampus (HP) and aberrant patterns of functional connectivities (FC) between the hippocampus and the rest of the brain. However, the relationship between cortical atrophy levels and corresponding degrees of aberrant FC patterns has not been systematically examined. In this study, we investigated whether there was an explicit link between structural abnormalities and corresponding functional aberrances associated with AD using structural and functional magnetic resonance imaging (fMRI) data. To this end, brain regions with cortical atrophies that are associated with AD were identified in the HP in the left (L) and right (R) hemispheres using structural MRI data from volume analyses (p<0.03 for L-HP; p<0.04 for R-HP) and voxel-based morphometry analyses (p<4×10(-4) for L-HP; p<2×10(-3) for R-HP). Aberrantly reduced FC levels between the HP (with atrophy) and precuneus were also consistently observed in fMRI data from AD than HC brains that were analyzed by the Pearson's correlation coefficients (p<3×10(-4) for L-HP; and p<8×10(-5) for R-HP). In addition, the substantial negative FC levels from the HC brains between the precuneus and post central gyrus (PoCG) without structural atrophy were also significantly diminished from the AD brains (p<5×10(-5) for L-PoCG; and p<6×10(-5) for R-PoCG). The effect sizes of these aberrant FC levels associated with AD were greater than that of cortical atrophy levels when comparing using normalized Z score and Cohen's d measures, which indicates that an aberrant FC level may precede cortical atrophy.