fMRI Time Series Research Articles

Classifying ASD based on time-series fMRI using spatial–temporal transformer

As the prevalence of autism spectrum disorder (ASD) increases globally, more and more patients need to receive timely diagnosis and treatment to alleviate their suffering. However, the current diagnosis method of ASD still adopts the subjective symptom-based criteria through clinical observation, which is time-consuming and costly. In recent years, functional magnetic resonance imaging (fMRI) neuroimaging techniques have emerged to facilitate the identification of potential biomarkers for diagnosing ASD. In this study, we developed a deep learning framework named spatial–temporal Transformer (ST-Transformer) to distinguish ASD subjects from typical controls based on fMRI data. Specifically, a linear spatial–temporal multi-headed attention unit is proposed to obtain the spatial and temporal representation of fMRI data. Moreover, a Gaussian GAN-based data balancing method is introduced to solve the data unbalance problem in real-world ASD datasets for subtype ASD diagnosis. Our proposed ST-Transformer is evaluated on a large cohort of subjects from two independent datasets (ABIDE I and ABIDE II) and achieves robust accuracies of 71.0% and 70.6%, respectively. Compared with state-of-the-art methods, our results demonstrate competitive performance in ASD diagnosis.

Beyond Linear Dynamic Functional Connectivity: A Vine Copula Change Point Model

To estimate dynamic functional connectivity for functional magnetic resonance imaging (fMRI) data, two approaches have dominated: sliding window and change point methods. While computationally feasible, the sliding window approach has several limitations. In addition, the existing change point methods assume a Gaussian distribution for and linear dependencies between the fMRI time series. In this work, we introduce a new methodology called Vine Copula Change Point (VCCP) to estimate change points in the functional connectivity network structure between brain regions. It uses vine copulas, various state-of-the-art segmentation methods to identify multiple change points, and a likelihood ratio test or the stationary bootstrap for inference. The vine copulas allow for various forms of dependence between brain regions including tail, symmetric and asymmetric dependence, which has not been explored before in the dynamic analysis of neuroimaging data. We apply VCCP to various simulation datasets and to two fMRI datasets: a reading task and an anxiety inducing experiment. In particular, for the former dataset, we illustrate the complexity of textual changes during the reading of Chapter 9 in Harry Potter and the Sorcerer’s Stone and find that change points across subjects are related to changes in more than one type of textual attributes. Further, the graphs created by the vine copulas indicate the importance of working beyond Gaussianity and linear dependence. Finally, the R package vccp implementing the methodology from the article is available from CRAN. Supplementary Materials for this article are available online.

Inferring Effective Connectivity Networks From fMRI Time Series With a Temporal Entropy-Score.

Inferring brain-effective connectivity networks from neuroimaging data has become a very hot topic in neuroinformatics and bioinformatics. In recent years, the search methods based on Bayesian network score have been greatly developed and become an emerging method for inferring effective connectivity. However, the previous score functions ignore the temporal information from functional magnetic resonance imaging (fMRI) series data and may not be able to determine all orientations in some cases. In this article, we propose a novel score function for inferring effective connectivity from fMRI data based on the conditional entropy and transfer entropy (TE) between brain regions. The new score employs the TE to capture the temporal information and can effectively infer connection directions between brain regions. Experimental results on both simulated and real-world data demonstrate the efficacy of our proposed score function.

WHOCARES: WHOle-brain CArdiac signal REgression from highly accelerated simultaneous multi-Slice fMRI acquisitions

Objective. To spatio-temporally resolve cardiac signals in functional magnetic resonance imaging (fMRI) time-series of the human brain using neither external physiological measurements nor ad hoc modelling assumptions. Approach. Cardiac pulsation is a physiological confound of fMRI time-series that introduces spurious signal fluctuations in proximity to blood vessels. fMRI alone is not sufficiently fast to resolve cardiac pulsation. Depending on the ratio between the instantaneous heart-rate and the acquisition sampling frequency (, with TR being the repetition time), the cardiac signal may alias into the frequency band of neural activation so that its removal through spectral filtering techniques is generally not possible. In this paper, we show that it is feasible to temporally and spatially resolve cardiac signals throughout the brain even when cardiac aliasing occurs by combining fMRI hyper-sampling with simultaneous multislice (SMS) imaging. The technique, which we name WHOle-brain CArdiac signal REgression from highly accelerated simultaneous multi-Slice fMRI acquisitions (WHOCARES), was developed on 695 healthy subjects selected from the Human Connectome Project and its performance validated against the RETROICOR, HAPPY and the pulse oxymeter signal regression methods. Main results. WHOCARES is capable of retrieving voxel-wise cardiac signal regressors. This is achieved without employing external physiological recordings nor through ad hoc modelling assumptions. The performance of WHOCARES was, on average, superior to RETROICOR, HAPPY and the pulse oxymeter regression methods. Significance. WHOCARES holds basis for the reliable mapping of cardiac activity in fMRI time-series. WHOCARES can be employed for the retrospective removal of cardiac noise in publicly available fMRI datasets where physiological recordings are not available. WHOCARES is freely available at https://github.com/gferrazzi/WHOCARES.

A simplicial analysis of the fMRI data from human brain dynamics under functional cognitive tasks.

The topological analysis of fMRI time series data has recently been used to characterize the identification of patterns of brain activity seen during specific tasks carried out under experimentally controlled conditions. This study uses the methods of algebraic topology to characterize time series networks constructed from fMRI data measured for adult and children populations carrying out differentiated reading tasks. Our pilot study shows that our methods turn out to be capable of identifying distinct differences between the activity of adult and children populations carrying out identical reading tasks. We also see differences between activity patterns seen when subjects recognize word and nonword patterns. The results generalize across different populations, different languages and different active and inactive brain regions.

On the Spatial Distribution of Temporal Complexity in Resting State and Task Functional MRI

Measuring the temporal complexity of functional MRI (fMRI) time series is one approach to assess how brain activity changes over time. In fact, hemodynamic response of the brain is known to exhibit critical behaviour at the edge between order and disorder. In this study, we aimed to revisit the spatial distribution of temporal complexity in resting state and task fMRI of 100 unrelated subjects from the Human Connectome Project (HCP). First, we compared two common choices of complexity measures, i.e., Hurst exponent and multiscale entropy, and observed a high spatial similarity between them. Second, we considered four tasks in the HCP dataset (Language, Motor, Social, and Working Memory) and found high task-specific complexity, even when the task design was regressed out. For the significance thresholding of brain complexity maps, we used a statistical framework based on graph signal processing that incorporates the structural connectome to develop the null distributions of fMRI complexity. The results suggest that the frontoparietal, dorsal attention, visual, and default mode networks represent stronger complex behaviour than the rest of the brain, irrespective of the task engagement. In sum, the findings support the hypothesis of fMRI temporal complexity as a marker of cognition.

Sustained attention induces altered effective connectivity of the ascending thalamo-cortical relay in obsessive-compulsive disorder.

Abnormal function of the thalamo-cortical relay is considered a hallmark of obsessive-compulsive disorder (OCD) and aberrant network interactions may underpin many of the clinical and cognitive symptoms that characterize the disorder. Several statistical approaches have been applied to in vivo fMRI data to support the general loss of thalamo-cortical connectivity in OCD. However, (a) few studies have assessed the contextual constraints under which abnormal network interactions arise or (b) have used methods of effective connectivity to understand abnormal network interactions. Effective connectivity is a particularly valuable method as it describes the putative causal influences that brain regions exert over each other, as opposed to the largely statistical consistencies captured in functional connectivity techniques. Here, using dynamic causal modeling (DCM), we evaluated how attention demand induced inter-group differences (HC ≠ OCD) in effective connectivity within a motivated thalamo-cortical network. Of interest was whether these effects were observed on the ascending thalamo-cortical relay, essential for the sensory innervation of the cortex. fMRI time series data from sixty-two participants (OCD, 30; HC, 32) collected using an established sustained attention task were submitted to a space of 162 competing models. Across the space, models distinguished between competing hypotheses of thalamo-cortical interactions. Bayesian model selection (BMS) identified marginally differing likely generative model architectures in OCD and HC groups. Bayesian model averaging (BMA), was used to weight connectivity parameter estimates across all models, with each parameter weighted by each model’s posterior probability, thus providing more stable estimates of effective connectivity. Inferential statistical analyses of estimated parameters revealed two principal results: (1) Significantly reduced intrinsic connectivity of the V1 → SPC pathway in OCD, suggested connective weakness in the early constituents of the dorsal visual pathway; (2) More pertinent with the discovery possibilities afforded by DCM, sustained attention in OCD patients induced significantly reduced contextual modulation of the ascending relay from the thalamus to the prefrontal cortex. These results form an important complement to our understanding of the contextual bases of thalamo-cortical network deficits in OCD, emphasizing vulnerability of the ascending relay.

Decision thresholding on fMRI activation maps using the Hilbert–Huang transform

Objective. Functional magnetic resonance imaging (fMRI) requires thresholds by which to identify brain regions with significant activation, particularly for experiments involving real-life paradigms. One conventional non-parametric approach to generating surrogate data involves decomposition of the original fMRI time series using the Fourier transform, after which the phase is randomized without altering the magnitude of individual frequency components. However, it has been reported that spontaneous brain signals could be non-stationary, which, if true, could lead to false-positive results. Approach. This paper introduces a randomization procedure based on the Hilbert–Huang transform by which to account for non-stationarity in fMRI time series derived from two fMRI datasets (stationary or non-stationary). The significance of individual voxels was determined by comparing the distribution of empirical data versus a surrogate distribution. Main results. In a comparison with conventional phase-randomization and wavelet-based permutation methods, the proposed method proved highly effective in generating activation maps indicating essential brain regions, while filtering out noise in the white matter. Significance. This work demonstrated the importance of considering the non-stationary nature of fMRI time series when selecting resampling methods by which to probe brain activity or identify functional networks in real-life fMRI experiments. We propose a statistical testing method to deal with the non-stationarity of continuous brain signals.

The dynamic shaping of local cortical circuitry by sex and age, and its relation to pattern comparison processing speed.

Previous resting-state functional magnetic resonance imaging (fMRI) studies have shown that the strength of local neural interactions decreases with distance. Here, we extend that line of research to evaluate effects of sex and age on local cortical circuitry in six cortical areas (superior frontal, precentral, postcentral, superior parietal, inferior parietal, and lateral occipital) using data acquired from 1,054 healthy young adults who participated in the Human Connectome Project. We confirmed previous findings that the strength of zero-lag correlations between prewhitened, resting-state, blood level oxygenation-dependent (BOLD) fMRI time series decreased with distance locally and documented that the rate of decrease with distance (spatial steepness) 1) was progressively lower from anterior to posterior areas, 2) was greater in women, especially in anterior areas, 3) increased with age, particularly for women, 4) was significantly correlated with percent inhibition, and 5) was positively and highly significantly correlated with pattern comparison processing speed (PCPS). A hierarchical tree clustering analysis of this dependence of PCPS on spatial steepness revealed a differential organization in processing that information between the two hemispheres, namely, a more independent vs. a more integrative processing in the left and right hemispheres, respectively. These findings document sex and age differences in dynamic local cortical interactions and provide evidence that spatial sharpening of these interactions may underlie cognitive processing speed differently organized in the two hemispheres.NEW & NOTEWORTHY Sex and age significantly affect shaping of local cortical interactions that are more limited in women and older brains. The net result is an increase in local spatial steepness of interactions, leading to a reduction of overlap among local ensembles and, hence, a more efficient information processing and an increase in the number of independent local cortical "processors." Remarkably, cognitive processing speed was positively associated with local spatial steepness, in keeping with the reasoning earlier.

The optimized combination of aCompCor and ICA-AROMA to reduce motion and physiologic noise in task fMRI data

One of the main challenges in fMRI processing is filtering the task BOLD signals from the noise. Independent component analysis with automatic removal of motion artifacts (ICA-AROMA) reduces motion artifacts by identifying ICA noise components based on their location at the brain edges and cerebrospinal fluid (CSF), high frequency content and correlation with motion regressors. In anatomical component correction (aCompCor), physiological noise regressors extracted from CSF were regressed out from the fMRI time series. In this study, we compared three methods to combine aCompCor and ICA-AROMA denoising in one denoising step. In the first analysis, we regressed the temporal signals of the ICA components identified as noise by ICA-AROMA together with the noise signals determined by aCompCor from the fMRI signals. For the second and third analyses, the correlation between the temporal signals of the ICA components and the aCompCor noise signals was used as an additional criterion to identify the noise components. In the second analysis, the temporal signals of the ICA components classified as noise were regressed from the fMRI signals. In the third analysis, the noise components were removed. To compare the denoising strategies, we examined the fractional amplitude of low-frequency fluctuations (fALFF) and the overlap between the contrast maps. Our results revealed that including the aCompCor noise signals as regressors in ICA-AROMA resulted in more correctly identified noise components, higher fALFF values, and larger activation maps. Moreover, combining the temporal signals of the noise components identified by ICA-AROMA with the aCompCor signals in a noise regression matrix resulted in deactivations. These results suggest that using the correlation between the ICA component temporal signals and the aCompCor signals as noise identification criteria in ICA-AROMA is the best approach for combining both denoising methods.

Change Point Detection Methods for Locating Activations in Functional Neuronal Images

The most common analysis for fMRI images is activation detection, in which the purpose is to find the locations in the brain that respond to specific functions, such as visual processing or motor functions by providing related stimuli as tasks in the experiment. On the other hand, it is also important to detect the instance the activation is triggered. One of the powerful techniques that can analyze the abnormal behavior of any data is change point (CP) analysis. We suggest that CP detection algorithms also can be used to locate the activations in functional magnetic resonance imaging (fMRI) sequences, as well. Our paper presents a two-fold innovative study in that respect. First, we propose to use CP detection algorithms to locate the activations in fMRI signals as a state-of-art topic. Furthermore, we propose and compare a set of change point analysis methods, a regression-based method (RBM), a statistical method (SM), and a mean difference of double sliding windows method (MDSW)) to locate such points. Second, we apply these methods to the fMRI signals, which are acquired from the real subjects, while they were performing fMRI tasks. Proposed methods were applied to three different fMRI experiments with a motor task, a visual task, and a linguistic task. The analysis shows that the methods find activations in accordance with established methods such as statistical parametric maps (SPM). The acquired up to 94 % results also show that the proposed methods can be used effectively to locate the activation times on fMRI time series.

Learning brain representation using recurrent Wasserstein generative adversarial net

Background and objectiveTo understand brain cognition and disorders, modeling the mapping between mind and brain has been of great interest to the neuroscience community. The key is the brain representation, including functional brain networks (FBN) and their corresponding temporal features. Recently, it has been proven that deep learning models have superb representation power on functional magnetic resonance imaging (fMRI) over traditional machine learning methods. However, due to the lack of high-quality data and labels, deep learning models tend to suffer from overfitting in the training process. MethodsIn this work, we applied a recurrent Wasserstein generative adversarial net (RWGAN) to learn brain representation from volumetric fMRI data. Generative adversarial net (GAN) is widely used in natural image generation and is able to capture the distribution of the input data, which enables the extraction of generalized features from fMRI and thus relieves the overfitting issue. The recurrent layers in RWGAN are designed to better model the local temporal features of the fMRI time series. The discriminator of RWGAN works as a deep feature extractor. With LASSO regression, the RWGAN model can decompose the fMRI data into temporal features and spatial features (FBNs). Furthermore, the generator of RWGAN can generate high-quality new data for fMRI augmentation. ResultsThe experimental results on seven tasks from the HCP dataset showed that the RWGAN can learn meaningful and interpretable temporal features and FBNs, compared to HCP task designs and general linear model (GLM) derived networks. Besides, the results on different training datasets showed that the RWGAN performed better on small datasets than other deep learning models. Moreover, we used the generator of RWGAN to yield fake subjects. The result showed that the fake data can also be used to learn meaningful representation compared to those learned from real data. ConclusionsTo our best knowledge, this work is among the earliest attempts of applying generative deep learning for modeling fMRI data. The proposed RWGAN offers a novel methodology for learning brain representation from fMRI, and it can generate high-quality fake data for the potential use of fMRI data augmentation.

Brain Synchronization and Multivariate Autoregressive (MVAR) Modeling in Cognitive Neurodynamics.

This paper is a review of cognitive neurodynamics research as it pertains to recent advances in Multivariate Autoregressive (MVAR) modeling. Long-range synchronization between the frontoparietal network (FPN) and forebrain subcortical systems occurs when multiple neuronal actions are coordinated across time (Chafee and Goldman-Rakic, 1998), resulting in large-scale measurable activity in the EEG. This paper reviews the power and advantages of the MVAR method to analyze long-range synchronization between brain regions (Kaminski et al., 2016; Kaminski and Blinowska, 2017). It explores the synchronization expressed in neurocognitive networks that is observable in the local field potential (LFP), an EEG-like signal, and in fMRI time series. In recent years, the surge in MVAR modeling in cognitive neurodynamics experiments has highlighted the effectiveness of the method, particularly in analyzing continuous neural signals such as EEG and fMRI (Pereda et al., 2005). MVAR modeling has been particularly useful in identifying causality, a multichannel time-series measure that can only be consistently computed with multivariate processes. Due to the multivariate nature of neuronal communication, multiple non-linear multivariate-analysis models are successful, presenting results with much greater accuracy and speed than non-linear univariate-analysis methods. Granger’s framework provides causal information about neuronal flow using neural time and frequency analysis, comprising the basis of the MVAR model. Recent advancements in MVAR modeling have included Directed Transfer Function (DTF) and Partial Directed Coherence (PDC), multivariate methods based on MVAR modeling that are capable of determining causal influences and directed propagation of EEG activity. The related Granger causality is an increasingly popular tool for measuring directed functional interactions from time series data.

Resting-state network organisation in children with traumatic brain injury

Children with traumatic brain injury are at risk of neurocognitive and behavioural impairment. Although there is evidence for abnormal brain activity in resting-state networks after TBI, the role of resting-state network organisation in paediatric TBI outcome remains poorly understood. This study is the first to investigate the impact of paediatric TBI on resting-state network organisation using graph theory, and its relevance for functional outcome.Participants were 8–14 years and included children with (i) mild TBI and risk factors for complicated TBI (mildRF+, n = 20), (ii) moderate/severe TBI (n = 15), and (iii) trauma control injuries (n = 27). Children underwent resting-state functional magnetic resonance imaging (fMRI), neurocognitive testing, and behavioural assessment at 2.8 years post-injury. Graph theory was applied to fMRI timeseries to evaluate the impact of TBI on global and local organisation of the resting-state network, and relevance for neurocognitive and behavioural functioning.Children with TBI showed atypical global network organisation as compared to the trauma control group, reflected by lower modularity (mildRF + TBI and moderate/severe TBI), higher smallworldness (mildRF + TBI) and lower assortativity (moderate/severe TBI ps < .04, Cohen's ds: > .6). Regarding local network organisation, the relative importance of hub regions in the network did not differ between groups. Regression analyses showed relationships between global as well as local network parameters with neurocognitive functioning (i.e., working memory, memory encoding; R2 = 23.3 - 38.5%) and behavioural functioning (i.e., externalising problems, R2 = 36.1%).Findings indicate the impact of TBI on global functional network organisation, and the relevance of both global and local network organisation for long-term neurocognitive and behavioural outcome after paediatric TBI. The results suggest potential prognostic value of resting-state network organisation for outcome after paediatric TBI.

Alzheimer's Disease Diagnosis and Biomarker Analysis Using Resting-State Functional MRI Functional Brain Network With Multi-Measures Features and Hippocampal Subfield and Amygdala Volume of Structural MRI.

Accurate diagnosis of the initial phase of Alzheimer’s disease (AD) is essential and crucial. The objective of this research was to employ efficient biomarkers for the diagnostic analysis and classification of AD based on combining structural MRI (sMRI) and resting-state functional MRI (rs-fMRI). So far, several anatomical MRI imaging markers for AD diagnosis have been identified. The use of cortical and subcortical volumes, the hippocampus, and amygdala volume, as well as genetic patterns, has proven to be beneficial in distinguishing patients with AD from the healthy population. The fMRI time series data have the potential for specific numerical information as well as dynamic temporal information. Voxel and graphical analyses have gained popularity for analyzing neurodegenerative diseases, such as Alzheimer’s and its prodromal phase, mild cognitive impairment (MCI). So far, these approaches have been utilized separately for the diagnosis of AD. In recent studies, the classification of cases of MCI into those that are not converted for a certain period as stable MCI (MCIs) and those that converted to AD as MCIc has been less commonly reported with inconsistent results. In this study, we verified and validated the potency of a proposed diagnostic framework to identify AD and differentiate MCIs from MCIc by utilizing the efficient biomarkers obtained from sMRI, along with functional brain networks of the frequency range .01–.027 at the resting state and the voxel-based features. The latter mainly included default mode networks (amplitude of low-frequency fluctuation [ALFF], fractional ALFF [ALFF], and regional homogeneity [ReHo]), degree centrality (DC), and salience networks (SN). Pearson’s correlation coefficient for measuring fMRI functional networks has proven to be an efficient means for disease diagnosis. We applied the graph theory to calculate nodal features (nodal degree [ND], nodal path length [NL], and between centrality [BC]) as a graphical feature and analyzed the connectivity link between different brain regions. We extracted three-dimensional (3D) patterns to calculate regional coherence and then implement a univariate statistical t-test to access a 3D mask that preserves voxels showing significant changes. Similarly, from sMRI, we calculated the hippocampal subfield and amygdala nuclei volume using Freesurfer (version 6). Finally, we implemented and compared the different feature selection algorithms to integrate the structural features, brain networks, and voxel features to optimize the diagnostic identifications of AD using support vector machine (SVM) classifiers. We also compared the performance of SVM with Random Forest (RF) classifiers. The obtained results demonstrated the potency of our framework, wherein a combination of the hippocampal subfield, the amygdala volume, and brain networks with multiple measures of rs-fMRI could significantly enhance the accuracy of other approaches in diagnosing AD. The accuracy obtained by the proposed method was reported for binary classification. More importantly, the classification results of the less commonly reported MCIs vs. MCIc improved significantly. However, this research involved only the AD Neuroimaging Initiative (ADNI) cohort to focus on the diagnosis of AD advancement by integrating sMRI and fMRI. Hence, the study’s primary disadvantage is its small sample size. In this case, the dataset we utilized did not fully reflect the whole population. As a result, we cannot guarantee that our findings will be applicable to other populations.

Biclustering fMRI time series: a comparative study

BackgroundThe effectiveness of biclustering, simultaneous clustering of rows and columns in a data matrix, was shown in gene expression data analysis. Several researchers recognize its potentialities in other research areas. Nevertheless, the last two decades have witnessed the development of a significant number of biclustering algorithms targeting gene expression data analysis and a lack of consistent studies exploring the capacities of biclustering outside this traditional application domain.ResultsThis work evaluates the potential use of biclustering in fMRI time series data, targeting the Region × Time dimensions by comparing seven state-in-the-art biclustering and three traditional clustering algorithms on artificial and real data. It further proposes a methodology for biclustering evaluation beyond gene expression data analysis. The results discuss the use of different search strategies in both artificial and real fMRI time series showed the superiority of exhaustive biclustering approaches, obtaining the most homogeneous biclusters. However, their high computational costs are a challenge, and further work is needed for the efficient use of biclustering in fMRI data analysis.ConclusionsThis work pinpoints avenues for the use of biclustering in spatio-temporal data analysis, in particular neurosciences applications. The proposed evaluation methodology showed evidence of the effectiveness of biclustering in finding local patterns in fMRI time series data. Further work is needed regarding scalability to promote the application in real scenarios.

Toward Precise Localization of Abnormal Brain Activity: 1D CNN on Single Voxel fMRI Time-Series.

Functional magnetic resonance imaging (fMRI) is one of the best techniques for precise localization of abnormal brain activity non-invasively. Machine-learning approaches have been widely used in neuroimaging studies; however, few studies have investigated the single-voxel modeling of fMRI data under cognitive tasks. We proposed a hybrid one-dimensional (1D) convolutional neural network (1D-CNN) based on the temporal dynamics of single-voxel fMRI time-series and successfully differentiated two continuous task states, namely, self-initiated (SI) and visually guided (VG) motor tasks. First, 25 activation peaks were identified from the contrast maps of SI and VG tasks in a blocked design. Then, the fMRI time-series of each peak voxel was transformed into a temporal-frequency domain by using continuous wavelet transform across a broader frequency range (0.003–0.313 Hz, with a step of 0.01 Hz). The transformed time-series was inputted into a 1D-CNN model for the binary classification of SI and VG continuous tasks. Compared with the univariate analysis, e.g., amplitude of low-frequency fluctuation (ALFF) at each frequency band, including, wavelet-ALFF, the 1D-CNN model highly outperformed wavelet-ALFF, with more efficient decoding models [46% of 800 models showing area under the curve (AUC) > 0.61] and higher decoding accuracies (94% of the efficient models), especially on the high-frequency bands (>0.1 Hz). Moreover, our results also demonstrated the advantages of wavelet decompositions over the original fMRI series by showing higher decoding performance on all peak voxels. Overall, this study suggests a great potential of single-voxel analysis using 1D-CNN and wavelet transformation of fMRI series with continuous, naturalistic, steady-state task design or resting-state design. It opens new avenues to precise localization of abnormal brain activity and fMRI-guided precision brain stimulation therapy.

Multigroup recognition of dementia patients with dynamic brain connectivity under multimodal cortex parcellation

ObjectiveTo accurately predict Alzheimer’s disease (AD) in its early stage of cognitive impairment is crucial to clinical diagnosis and intervention. However, there is no consensus over which parts of brain areas are responsible for cognitive decline due to the incompatible and single-modal-based parcellation methods employed by researchers. MethodsA novel dynamic brain connectivity processing method (DBCP) is proposed based on the human connectome project multimodal parcellation (HCP MMP) to explore the spatial–temporal characteristics of the brain in different stages of mild cognitive impairment (MCI) and Alzheimer’s disease. First, dynamic connectivity under HCP MMP is constructed to divide the whole fMRI time series into hundreds of segmentations. Then, graph-based topological measures are calculated, followed by statistical outlier examinations implemented by the K-means method. ResultsA superior performance (accuracy = 86%, recall = 87%, precision = 86%, F1-score = 86%) in the four groups (healthy control vs. early MCI vs. late MCI vs. AD) recognition is achieved by training an effective but uncomplicated deep learning model. ConclusionDynamic connectivity within the most fine-grained multimodal human cortex parcellation can reveal more useful details to distinguish brain dysfunctional patients compared with static connectivity or single modal based parcellation, and the proposed method can suppress the outliers well among fragmented fMRI signals. SignificanceProviding more evidence on the primary responsibility of DMN and DAN for cognitive impairment of the brain, 64 cortex regions with significant topological alterations are suggested as the most prominent and fine-grained biomarker for further longitudinal AD studies.

Spatio-temporal directed acyclic graph learning with attention mechanisms on brain functional time series and connectivity

We develop a deep learning framework, spatio-temporal directed acyclic graph with attention mechanisms (ST-DAG-Att), to predict cognition and disease using functional magnetic resonance imaging (fMRI). This ST-DAG-Att framework comprises of two neural networks, (1) spatio-temporal graph convolutional network (ST-graph-conv) to learn the spatial and temporal information of functional time series at multiple temporal and spatial graph scales, where the graph is represented by the brain functional network, the spatial convolution is over the space of this graph, and the temporal convolution is over the time dimension; (2) functional connectivity convolutional network (FC-conv) to learn functional connectivity features, where the functional connectivity is derived from embedded multi-scale fMRI time series and the convolutional operation is applied along both edge and node dimensions of the brain functional network. This framework also consists of an attention component, i.e., functional connectivity-based spatial attention (FC-SAtt), that generates a spatial attention map through learning the local dependency among high-level features of functional connectivity and emphasizing meaningful brain regions. Moreover, both the ST-graph-conv and FC-conv networks are designed as feed-forward models structured as directed acyclic graphs (DAGs). Our experiments employ two large-scale datasets, Adolescent Brain Cognitive Development (ABCD, n=7693) and Open Access Series of Imaging Study-3 (OASIS-3, n=1786). Our results show that the ST-DAG-Att model is generalizable from cognition prediction to age prediction. It is robust to independent samples obtained from different sites of the ABCD study. It outperforms the existing machine learning techniques, including support vector regression (SVR), elastic net's mixture with random forest, spatio-temporal graph convolution, and BrainNetCNN.

NAS-optimized topology-preserving transfer learning for differentiating cortical folding patterns

Increasing evidence suggests that cortical folding patterns of human cerebral cortex manifest overt structural and functional differences. However, for interpretability, few studies leverage advanced techniques (e.g., deep learning) to investigate the difference among cortical folds, resulting in more differences yet to be extensively explored. To this end, we proposed an effective topology-preserving transfer learning framework to differentiate cortical fMRI time series extracted from cortical folds. Our framework consists of three main parts: (1) Neural architecture search (NAS), which is used to devise a well-performing network structure based on an initialized hand-designed super-graph in an image dataset; (2) Topology-preserving transfer, which takes the model searched by NAS as the source network, keeping the topological connectivity in the network unchanged, while transforming all 2D operations including convolution and pooling into 1D, therefore resulting in a topology-preserving network, named TPNAS-Net; (3) Classification and correlation analysis, which involves using the TPNAS-Net to classify 1D cortical fMRI time series for each individual brain, and performing a group difference analysis between autism spectrum disorder (ASD) and healthy control (HC) and correlation analysis with clinical information (i.e., age). Extensive experiments on two ASD datasets obtain consistent results, demonstrating that the TPNAS-Net not only discriminates cortical folding patterns at high classification accuracy, but also captures subtle differences between ASD and HC (p-value=0.042). In addition, we discover that there is a positive correlation between the classification accuracy and age in ASD (r=0.39, p-value=0.04). These findings together suggest that structural and functional differences in cortical folding patterns between ASD and HC may provide a potentially useful biomarker for the diagnosis of ASD.