Frontiers in Neuroimaging

Single photon emission computed tomography (SPECT) is a highly specialized imaging modality that enables measurement of regional cerebral perfusion and, in particular, resting cerebral blood flow (rCBF). Recent technological advances have improved SPECT quantification and reliability, making it increasingly useful for studying rCBF abnormalities and perfusion network alterations in psychiatric and neurological disorders. To characterize large scale functional organization in SPECT data, data driven decomposition methods such as independent component analysis (ICA) have been used to extract covarying perfusion patterns that map onto interpretable brain networks. Blind ICA provides a data driven approach to estimate these networks without strong prior assumptions. More recently, a hybrid approach that leverages spatial priors to guide a spatially constrained ICA (sc ICA) have been used to fully automate the ICA analysis while also providing participant-specific network estimates. While this has been reliably demonstrated in fMRI with the NeuroMark template, there is currently no comparable SPECT template. A SPECT template would enable automatic estimation of functional SPECT networks with participant-specific expressions that correspond across participants and studies. The current study introduces a new replicable NeuroMark SPECT template for estimating canonical perfusion covariance patterns (networks). We first identify replicable SPECT networks using blind ICA applied to two large sample SPECT datasets. We then demonstrate the use of the resulting template by applying sc-ICA to an independent schizophrenia dataset. In sum, this work presents and shares the first NeuroMark SPECT template and demonstrating its utility in an independent cohort, providing a scalable and robust framework for network-based analyses.

Resting-state fMRI (rs-fMRI) data are typically filtered at different frequency bins between 0.008[~]0.2 Hz (varies across the literature) prior to analysis to mitigate nuisance variables (e.g., drift, motion, cardiac, and respiratory) and maximize the sensitivity to neuronal-mediated BOLD signal. However, multiple lines of evidence suggest meaningful BOLD signal may also be parsed at higher frequencies. To test this notion, a functional network connectivity (FNC) analysis based on a spatially informed independent component analysis (ICA) was performed at seven different bandpass frequency bins to examine FNC matrices across spectra. Further, eyes open (EO) vs. eyes closed (EC) resting-state acquisitions from the same participants were compared across frequency bins to examine if EO vs. EC FNC matrices and randomness estimations of FNC matrices are distinguishable at different frequencies. Results show that FNCs in higher-frequency bins display modular FNC similar to the lowest frequency bin, while r-to-z FNC and FNC-based measures indicating matrix non-randomness were highest in the 0.31-0.46 Hz range relative to all frequency bins above and below this range. As such, the FNC within this range appears to be the most temporally correlated, but the mechanisms facilitating this coherence require further analyses. Compared to EO, EC displayed greater FNC (involved in visual, cognitive control, somatomotor, and auditory domains) and randomness values at lower frequency bins, but this phenomenon flipped (EO > EC) at frequency bins greater than 0.46 Hz, particularly within visual regions. While the effect sizes range from small to large specific to frequency range and resting state (EO vs. EC), with little influence from common artifacts. These differences indicate that unique information can be derived from FNC between BOLD signals at different frequencies relative to a given restingstate acquisition and support the hypothesis meaningful BOLD signal is present at higher frequency ranges.

PurposePrior work has shown that whole-band linear regression of nuisance signals can introduce artifactual connectivity in high-frequency resting-state fMRI. Errors of motion regressors and non-stationarity of nuisance signals exacerbate artifacts. Here, we introduce spectral-temporal segmentation of regression vectors to decouple regression in different frequency bands to reduce motion artifacts. MethodsAn alternative approach to whole-band linear nuisance regression is introduced in the present work relying on spectral segmentation of the motion parameters into k-bands using non-causal or FIR filters, with whole-band regression of the filtering residual, and temporal segmentation of regression vectors. The methodology was tested in computer simulations and in-vivo data. Resting-state networks in five healthy controls and two brain tumor patients using high-speed fMRI (TR >= 205 ms) were mapped using the present approach combined with spectrally constrained regression of physiological noise and the results were compared to the conventional whole band regression approach. ResultsComputer simulations showed high tolerance to frequency dependent errors in regression vectors. Motion and physiological noise artifacts in-vivo were substantially reduced without introducing artifactual connectivity. Artifactual connectivity decreased asymptotically with increasing number of frequency bands without decreasing connectivity in major resting-state networks. Connectivity above 0.3 Hz in-vivo was consistent with that in traditional low-frequency networks. ConclusionsSpectral-temporal segmentation of regression vectors is a powerful approach to reduce artifacts from non-stationary high-bandwidth nuisance signals.

The objective diagnostic and stratification biomarkers developed with resting-state functional magnetic resonance imaging (rs-fMRI) data are expected to contribute to more effective treatment for mental disorders. Unfortunately, there are currently no widely accepted biomarkers, partially due to the large variety of analysis pipelines for developing them. In this study we comprehensively evaluated analysis pipelines using a large-scale, multi-site fMRI dataset for major depressive disorder (MDD) (1162 participants from eight imaging sites). We explored the combinations of options in four subprocesses of analysis pipelines: six types of brain parcellation, four types of estimations of functional connectivity (FC), three types of site difference harmonization, and five types of machine learning methods. 360 different MDD diagnostic biomarkers were constructed using the SRPBS dataset acquired with unified protocols (713 participants from four imaging sites) as a discovery dataset and evaluated with datasets from other projects acquired with heterogeneous protocols (449 participants from four imaging sites) for independent validation. To identify the optimal options regardless of the discovery dataset, we repeated the same procedure after swapping the roles of the two datasets. We found pipelines that included Glassers parcellation, tangent-covariance, no harmonization, and non-sparse machine learning methods tended to result in high classification performance. The diagnosis results of the top 10 biomarkers showed high similarity, and weight similarity was also observed between eight of the biomarkers, except two that used both data-driven parcellation and FC computation. We applied the top 10 pipelines to the datasets of other mental disorders (autism spectral disorder: ASD and schizophrenia: SCZ) and eight of the ten biomarkers showed sufficient classification performances for both disorders, except two pipelines that included Pearson correlation, ComBat harmonization and random forest classifier combination. HighlightsO_LIWe evaluated the analysis pipelines of rsFC biomarker development. C_LIO_LIFour subprocesses in them were investigated with two multi-site datasets. C_LIO_LIGlassers parcellation, tangent covariance, and non-sparse methods were preferred. C_LIO_LIThe weight patterns of eight of the top 10 biomarkers showed high commonality. C_LIO_LIEight of the top 10 pipelines were successful for developing SCZ/ASD biomarkers. C_LI

BackgroundRegional neural response and network property used to be treated separately. However, evidence has suggested an intimate relationship between the regional and inter-regional profiles. This research aimed to investigate the influence of functional connectivity on regional spontaneous activity. MethodsThirty-six and sixty datasets of structural magnetic resonance imaging (sMRI) and resting state functional MRI (rsfMRI) were selected from the NKI and CAN-BIND database, respectively. The cerebral cortex in rsfMRI was parcellated by MOSI (modular analysis and similarity measurements), which enables multi-resolution exploration. For each parcellated cluster, the mean amplitude of low-frequency fluctuation (ALFF) and its average functional connectivity strength with the remaining cortical analogs were computed. Correlation analyses were exploited to examine their relationship. Supplementary analysis was applied to CAN-BIND EEG data (1 to 30 Hz). ResultsNegative correlation coefficients between inter-regional interaction and regional power were noticed in both MRI datasets. One-sample t-tests revealed robust statistics across different analytic resolutions yielded by MOSI, with individual P values at the level 10^-4 to 10^-5. The results suggested that the more intense crosstalk a neural node is embedded in, the less regional power it manifests, and vice versa. The negative relationship was replicated in EEG analysis but limited to delta (1 to 4 Hz) and theta (4 to 8 Hz) frequencies. ConclusionsWe postulate that inhibitory coupling is the mechanism that bridges the local and inter-regional properties, which is more prominent in the lower spectra. The interpretation warrants particular caution since noise may also contribute to the observation.

BackgroundAdvances in functional magnetic resonance imaging (fMRI) have led to the ability to study the brain across many contexts. However, the large number of features generated by functional connectivity approaches may overfit the data. These problems can be overcome with either feature selection (FS) or dimensionality reduction (DR), which can be applied to less complex models. We utilize two open-source datasets to compare the performance of DR/FS methods on cognitive task decoding using a suite of ML classifiers. New MethodWhile DR and FS methods have been used previously in decoding research, no systematic comparison of their performance has been undertaken. Here, we compare available methods using commonly utilized machine learning libraries to establish which methods provide the best predictive performance. We then conduct statistical tests to examine the relative contributions of DR and FS methods and classifiers on decoding accuracy. ResultsNeither DR or FS was found to be superior. However, differences were identified across datasets and tasks. In the majority of methods and datasets, a peak in predictive performance was found using a small percentage of the total number of original features. Comparison with existing methodsSome methods perform better than the baseline method of prediction with all available features or selecting features randomly. Decoding performance utilizing the HCP datasets with certain DR/FS methods exceeds that of deep learning approaches. ConclusionsSimple machine learning models with DR/FS have competitive decoding performance. These results suggest a "sweet spot" for the tradeoff between the retention of features and predictive accuracy.

In functional magnetic resonance imaging (fMRI) studies, it is common to evaluate the brains functional network connectivity (FNC) which captures the temporal coupling between hemodynamic signals derived from whole brain networks. FNC has been linked to various psychological phenomena. However, analysis of FNCs mainly focuses on linear statistical relationships, which may not capture the full complexity of the interactions among brain intrinsic connectivity networks (ICNs). Therefore, it is important to explore approaches that can better account for possible intricate nonlinear interactions involved in cognitive operations and the changes observed in psychiatric conditions such as schizophrenia. This exploration can lead to a better understanding of brain function and provide new insights into neural links to various psychological and psychiatric conditions. In this paper, we present an innovative approach which utilizes a deep convolutional neural network (DCNN) to extract nonlinear heatmaps from FNC matrices. By analyzing the heatmaps, multi-level nonlinear interactions can be derived from the corresponding input FNC data. Our results show these networks represent a significant improvement over previous approaches and offer a robust framework for understanding the complex interactions between brain regions. By incorporating two stages in the training process, our method ensures optimal efficiency and effectiveness. In the initial stage, a deep convolutional neural network is trained to create heatmaps from various convolution layers of the network. In the next stage, by utilizing a t-test-based feature selection method, we can effectively analyze heatmaps from different convolution layers. This approach ensures that we are able to functional connectivity with varying degrees of nonlinearity with a focus on the heatmaps that play an important role in distinguishing different groups. We used a large dataset consisting of both schizophrenia patients and healthy controls, which were divided into separate training and validation sets to evaluate this approach. Results showed patients with increases in default mode networks connections to itself and cognitive control regions and controls with increases within visual and between visual, motor, and auditory domains. We also find significantly increased cross-validated classification accuracy (at 92.8%) compared to several competing approaches. Our approach shows the potential to accurately distinguish differences between the schizophrenia and healthy control groups with high accuracy.

BackgroundFunctional near-infrared spectroscopy (fNIRS) has recently gained momentum as a reliable and accurate tool for assessing brain states. This increase in popularity is due to its robustness to movement, non-invasive nature, portability, and user-friendly application. However, compared to functional magnetic resonance imaging (fMRI), fNIRS is less sensitive to deeper brain activity and offers less coverage. Additionally, due to fewer advancements in method development, the performance of fNIRS-based brain-state classification still lags behind more prevalent methods like fMRI. MethodsWe introduce a novel classification approach grounded in Riemannian geometry for the classification of kernel matrices, leveraging the temporal and spatial channel relationships and inherent duality of fNIRS signals--more specifically, oxygenated and deoxygenated hemoglobin. For the Riemannian geometry-based models, we compared different kernel matrix estimators and two classifiers: Riemannian Support Vector Classifier and Tangent Space Logistic Regression. These were benchmarked against four models employing traditional feature extraction methods. Our approach was tested in two brain-state classification scenarios based on the same fNIRS dataset: an 8-choice classification, which includes seven established plus an individually selected imagery task, and a 2-choice classification of all possible 28 2-task combinations. ResultsThe novel approach achieved a mean 8-choice classification accuracy of 65%, significantly surpassing the mean accuracy of 42% obtained with traditional methods. Additionally, the best-performing model achieved an average accuracy of 96% for 2-choice classification across all possible 28 task combinations, compared to 78% with traditional models. ConclusionTo our knowledge, we are the first to demonstrate that the proposed Riemannian geometry-based classification approach is both powerful and viable for fNIRS data, considerably increasing the accuracy in binary and multi-class classification of brain activation patterns.

Mouse resting-state functional magnetic resonance imaging (rs-fMRI) is an increasingly popular tool for probing brain activity under experimental manipulations; however, there remains considerable variability in data quality throughout the field. There is a need for an accessible set of acquisition guidelines such that a baseline level of data quality can be attained regardless of domain expertise or specialized equipment (i.e. in anesthetized, free-breathing mice). In particular, there is a gap in the literature regarding the interpretation of physiological parameters as markers of anesthetic depth, and ultimately, data quality. To this end, we developed a set of acquisition guidelines after examining whether continuous physiological variables predict network detectability above and beyond categorical external variables (anesthetic dose, session, time) in C57Bl/7 and C3HeB/FeJ mice anesthetized with isoflurane-dexmedetomidine. Standard physiological metrics (respiration rate and heart rate) did not predict network detectability above and beyond anesthetic dose but instead depended strongly on strain and subject, thus we advise against tuning anesthesia based on respiration or heart rate when the goal is obtaining clear resting-state networks. The most important predictor of improved network detectability was a low isoflurane dose of 0.23%, hence we recommend that researchers prioritize piloting the minimal possible isoflurane dose for their mouse model. In summary, our work examines the contributions from sources of variability that impact rs-fMRI data quality and synthesizes the findings into practical guidelines to help experimenters adapt their acquisition protocols and improve data quality.

Resting state functional connectivity (rsFC) and resting state effective connectivity (rsEC) are two of the most common measures that can be extracted from resting state functional magnetic resonance imaging (rs-fMRI) data. RSFC is often used to indicate the statistical dependencies among different brain regions of interest, whereas rsEC describes the causal influences among them. Many studies have explored utilities of rsFC and rsEC measures for classifying psychiatric conditions. Several studies showed that rsEC were better than rsFC features for classifying major depression (Frassle et al., 2020; Geng et al., 2018) and schizophrenia ((Brodersen et al., 2014)). However, no study to-date has investigated whether rsEC is inherently better than rsFC for classifying psychiatric conditions or the impact of autocorrelation on classifying rsFC, even though autocorrelation is known to be present in rs-fMRI data. To fill these gaps, we performed a series of computational experiments, by varying the size of the network and the number of participants, to gain some insight into these two aspects of supervised classification with resting state connectivity. Contrary to what has been reported in the literature, the results from our study suggest that rsEC cannot be, in principle, better than rsFC features for classification. In fact, rsEC measures led to systematically worse classification results, compared to rsFC features. In terms of the impact of autocorrelation, we found that lag-one autocorrelation could lead to both false negative and false positive classification results for studies with a small sample size.

Removal of nuisance signals (such as motion) from the BOLD time series is an important aspect of preprocessing to obtain meaningful resting-state functional connectivity (rs-FC). The nuisance signals are commonly removed using denoising procedures at the finest resolution, i.e. the voxel time series. Typically the voxel-wise time series are then aggregated into predefined regions or parcels to obtain a rs-FC matrix as the correlation between pairs of regional time series. Computational efficiency can be improved by denoising the aggregated regional time series instead of the voxel time series. However, a comprehensive comparison of the effects of denoising on these two resolutions is missing. In this study, we systematically investigate the effects of denoising at different time series resolutions (voxel- and region-level) in 370 unrelated subjects from the HCP-YA dataset. Alongside the time series resolution, we considered additional factors such as aggregation method (Mean and first eigenvariate [EV]) and parcellation granularity (100, 400, and 1,000 regions). To assess the effect of those choices on the utility of the resulting whole-brain rs-FC, we evaluated the individual specificity (fingerprinting) and the capacity to predict age and three cognitive scores. Our findings show generally equal or better performance for region-level denoising with notable differences depending on the aggregation method. Using mean aggregation yielded equal individual specificity and prediction performance for voxel- and region-level denoising. When EV was employed for aggregation, the individual specificity of voxel-level denoising was reduced compared to region-level denoising. Increasing parcellation granularity generally improved individual specificity. For the prediction of age and cognitive test scores, only fluid intelligence indicated worse performance for voxel-level denoising in the case of aggregating with the EV. Based on these results, we recommend the adoption of region-level denoising for brain-behavior investigations when using mean aggregation. This approach offers equal individual specificity and prediction capacity with reduced computational resources for the analysis of rs-FC patterns.

Functional connectivity (FC) is a widely used imaging parameter of functional magnetic resonance imaging (fMRI). However, low reliability has been a concern among researchers, particularly in small-sample-size studies. Previous studies have shown that FC based on longer fMRI scans was more reliable, therefore, a feasible solution is to predict long-scan FCs using existing short-scan FCs. This study explored three different generalized linear models (GLMs) using the human connectome project (HCP) dataset. We found that the GLM based on individual short-scan FC could effectively predict long-scan individual FC value, while GLMs based on whole-brain FCs and dynamic FC performed better in predicting long-scan summed FC value of whole brain. The models were explained through visualization of weights in models. Besides, the differences in three GLMs could be explained as differences in distribution features of FC matrices predicted by them. Results were validated in different datasets, including the Consortium for Reliability and Reproducibility (CoRR) project and our local dataset. These models could be applied to improve the test-retest reliability of FC and to improve the performance of connectome-based predictive models (CPM). In conclusion, we developed three GLMs that could be used to predict long-scan FC from short-scan FC, and these models were robust across different datasets and could be applied to improve the test-retest reliability of FC and the performance of CPM.

SignificanceCurrent techniques for data analysis in functional near-infrared spectroscopy (fNIRS), such as artifact correction, do not allow to integrate the information originating from both wavelengths, considering only temporal and spatial dimensions of the signals structure. Parallel factor analysis (PARAFAC) has previously been validated as a multidimensional decomposition technique in other neuroimaging fields. AimWe aimed to introduce and validate the use of PARAFAC for the analysis of fNIRS data, which is inherently multidimensional (time, space, wavelength). ApproachWe used data acquired in 17 healthy adults during a verbal fluency task to compare the efficacy of PARAFAC for motion artifact correction to traditional 2D decomposition techniques, i.e. target principal (tPCA) and independent component analysis (ICA). Correction performance was further evaluated under controlled conditions with simulated artifacts and hemodynamic response functions. ResultsPARAFAC achieved significantly higher improvement in data quality as compared to tPCA and ICA. Correction in several simulated signals further validated its use and promoted it as a robust method independent of the artifacts characteristics. ConclusionsThis study describes the first implementation of PARAFAC in fNIRS and provides validation for its use to correct artifacts. PARAFAC is a promising data-driven alternative for multidimensional data analyses in fNIRS and this study paves the way for further applications.

The brain connectivity of resting-state fMRI (rs-fMRI) represents an intrinsic state of brain architecture, and it has been used as a useful neural marker for detecting psychiatric conditions as well as for predicting psychosocial characteristics. However, most studies using brain connectivity have focused more on the strength of functional connectivity over time (static-FC) but less attention to temporal characteristics of connectivity changes (FC-variability). The primary goal of the current study was to investigate the effectiveness of using the FC-variability in classifying an individuals pathological characteristics from others and predicting psychosocial characteristics. In addition, the current study aimed to prove that benefits of the FC-variability are reliable across various analysis procedures. To this end, three open public large resting-state fMRI datasets including individuals with Autism Spectrum Disorder (ABIDE; N = 1249), Schizophrenia disorder (COBRE; N = 145), and typical development (NKI; N = 672) were utilized for the machine learning (ML) classification and prediction based on their static-FC and the FC-variability metrics. To confirm the robustness of FC-variability utility, we benchmarked the ML classification and prediction with various brain parcellations and sliding window parameters. As a result, we found that the ML performances were significantly improved when the ML included FC-variability features in classifying pathological populations from controls (e.g., individuals with autism spectrum disorder vs. typical development) and predicting psychiatric severity (e.g., score of autism diagnostic observation schedule), regardless of parcellation selection and sliding window size. Additionally, the ML performance deterioration was significantly prevented with FC-variability features when excessive features were inputted into the ML models, yielding more reliable results. In conclusion, the current finding proved the usefulness of the FC-variability and its reliability.

Although neuroimaging provides powerful tools for assessing brain structure and function, their utility for elucidating mechanisms underlying neuropsychiatric disorders is limited by their sensitivity to head motion. Several publications have shown that standard retrospective motion correction and arduous quality assessment are insufficient to fully remove the deleterious impacts of motion on functional (fMRI) and structural (sMRI) neuroimaging data. These residual errors tend to be correlated with age and clinical diagnosis, resulting in artifactual findings in studies of clinical, developmental, and aging populations. As such there is a continued need to explore and evaluate novel methods for reducing head motion, and their applicability in these populations. Recently, a custom-fitted styrofoam head mold was reported to reduce motion across a range of ages, mostly adolescents, during a resting state fMRI scan. In the present study, we tested the efficacy of these head molds in a sample exclusively of young children (N = 19; mean age = 7.9 years) including those with ADHD (N = 6). We evaluated the head molds impact on head motion, data quality, and analysis results derived from the data. Importantly, we also evaluated whether the head molds were tolerated by our population. We also assessed the extent to which the head molds efficacy was related to anxiety levels and ADHD symptoms. In addition to fMRI, we examined the head molds impact on sMRI by using a specialized sequence with embedded volumetric navigators (vNAV) to determine head motion during sMRI. We evaluated the head molds impact on head motion, data quality, and analysis results derived from the data. Additionally, we conducted acoustic measurements and analyses to determine the extent to which the head mold reduced the noise dosage from the scanner. We found that some individuals benefited while others did not improve significantly. One individuals sMRI motion was made worse by the head mold. We were unable to identify predictors of the head mold response due to the smaller sample size. The head molds were tolerated well by young children, including those with ADHD, and they provided ample hearing protection. Although the head mold was not a universal solution for reducing head motion and improving data quality, we believe the time and cost required for using the head mold may outweigh the potential loss of data from excessive head motion for developmental studies.

Independent component analysis (ICA) is a widely used data-driven technique for investigating brain structure and function to extract intrinsic networks. However, the ability of ICA, a linear mixing model, to capture nonlinear relationships is inherently limited. While nonlinear ICA can be used to estimate nonlinear+ linear mixtures, it can be useful to study the degree to which there is nonlinearity above and beyond the widely studied linear resting networks. Here, we propose a way to divide the data into sources exhibiting linear-only or explicitly nonlinear dependencies in resting functional magnetic resonance imaging (fMRI) data. Such an approach can be very informative as it allows us to evaluate the degree to which a given network might be linear, nonlinear, or both linear and nonlinear. Here, we present an enhanced connectivity-domain ICA approach, connectivity-matrix ICA, incorporating normalized mutual information (NMI) after canceling the linear effects to measure explicitly nonlinear (EN) relationships within voxel connectivity. This integration enables the identification of brain spatial maps that exhibit pronounced explicitly nonlinear dependencies while excluding linear relationships. By eliminating linear dependencies and utilizing NMI, we discover highly structured resting networks that conventional functional connectivity methods would typically overlook. The results indicate that several maps show only linear or EN relationships, and the rest of the components display both linear and nonlinear patterns. We categorized these maps as linear-only, EN-only, and linear-EN maps. We also evaluate differences in the identified networks in a schizophrenia dataset. A significant global difference has been discovered between schizophrenia and controls in some linear-EN maps, such as the frontal lobe. Moreover, the temporal lobe and thalamus display linear group differences, while the visual and motor cortex display global differences in nonlinear relationships as their primary driver of these disparities. In sum, our findings emphasize the significance of accounting for explicitly nonlinear dependencies in functional connectivity analysis and demonstrate the effectiveness of the extended cmICA approach in revealing previously unrecognized brain dynamics.

Accurate registration of regions of interest (ROIs) from standard atlases to participants native spaces is a critical step in fMRI studies, as it directly affects the reliability of sampled BOLD signals. While T1-weighted (T1w) image-based ROI registration is well validated and widely adopted in cortical fMRI, its performance degrades in brainstem studies due to the small size, dense packing, and poor visibility of brainstem nuclei on T1w contrast. We hypothesized that incorporating diffusion MR images, containing more information about internal brainstem architecture, should improve ROI registration accuracy. To test this, we developed four registration pipelines that either included or excluded diffusion-based alignment components and evaluated their performance using data from n=20 healthy participants. Registration accuracy was assessed using Dice coefficient for the red nucleus (RN) and the substantia nigra (SN), and mis-registration fraction--a metric developed for nuclei that cannot be manually delineated--for the dorsal raphe nucleus (DRN). The results showed that diffusion-based pipelines, using fractional anisotropy (FA) images, non-diffusion-weighted (b0) images, and multivariate combination, outperformed the T1w-only baseline. Probabilistic maps derived from inverse-transformed native ROIs further supported improved sensitivity to inter-individual anatomical variability in the diffusion-augmented pipelines. In addition, analysis of gradient magnitude maps from the Jacobian determinants revealed associations between localized deformation and image modality-specific landmarks. These findings demonstrate the potential of diffusion-augmented pipelines for improving brainstem ROI registration, which could enhance the robustness of fMRI studies on brainstem disorders characterized by functional dysregulation.

Dynamic functional connectivity (dFC) studies the time-varying coordination between brain regions measured with fMRI and is a potential biomarker for understanding cognitive dynamics and tracking the development of neurological disorders. However, a critical methodological challenge lies in the variability of dFC estimates across different dFC assessment methods, raising concerns about the reliability and interpretation of downstream findings. While deep learning (DL) models have demonstrated the ability to capture traditionally inaccessible data patterns in many disciplines, they encounter challenges when applied to neuroimaging data. For instance, the high dimensionality, noise, and temporal complexity inherent in dFC makes it challenging for DL models to extract meaningful and interpretable insights. In this study, we investigated how DL architectures can be developed and adapted to predict task presence over time from task-based dFC data, and additionally, how the choice of dFC assessment method influences the predictive performance of DL models. We developed and compared a convolutional neural network (CNN), a node-level classification graph convolutional network (GCN), and a graph-level classification GCN based on their ability to predict time points at which subjects were performing a cognitive task or at rest. In our study, the results indicate that both DL model architecture and dFC estimation methodology significantly impact task presence prediction capacity, while the specific task paradigm had minimal influence out of the limited types that were explored. This work offers a powerful benchmark for understanding the dynamics of underlying task-driven cognitive state transitions and the analytical flexibility limitations of dFC estimation methods and DL architectures.

BackgroundWhile functional connectivity is widely studied, there has been little work studying functional connectivity at different spatial scales. Likewise, the relationship of functional connectivity between spatial scales is unknown. MethodsWe proposed an independent component analysis (ICA) - based approach to capture information at multiple model orders (component numbers) and to evaluate functional network connectivity (FNC) both within and between model orders. We evaluated the approach by studying group differences in the context of a study of resting fMRI (rsfMRI) data collected from schizophrenia (SZ) individuals and healthy controls (HC). The predictive ability of FNC at multiple spatial scales was assessed using support vector machine (SVM)-based classification. ResultsIn addition to consistent predictive patterns at both multiple-model orders and single model orders, unique predictive information was seen at multiple-model orders and in the interaction between model orders. We observed that the FNC between model order 25 and 50 maintained the highest predictive information between HC and SZ. Results highlighted the predictive ability of the somatomotor and visual domains both within and between model orders compared to other functional domains. Also, subcortical-somatomotor, temporal-somatomotor, and temporal-subcortical FNCs had relatively high weights in predicting SZ. ConclusionsIn sum, multi-model order ICA provides a more comprehensive way to study FNC, produces meaningful and interesting results which are applicable to future studies. We shared the spatial templates from this work at different model orders to provide a reference for the community, which can be leveraged in regression-based or fully automated (spatially constrained) ICA approaches. Impact StatementMulti-model order ICA provides a comprehensive way to study brain functional network connectivity within and between multiple spatial scales, highlighting findings that would have been ignored in single model order analysis. This work expands upon and adds to the relatively new literature on resting fMRI-based classification and prediction. Results highlighted the differentiating power of specific intrinsic connectivity networks on classifying brain disorders of schizophrenia patients and healthy participants, at different spatial scales. The spatial templates from this work provide a reference for the community, which can be leveraged in regression-based or fully automated ICA approaches.

Functional magnetic resonance imaging (fMRI) in rodents holds great potential for advancing our understanding of brain networks. Unlike the human fMRI community, there remains no standardized resource in rodents for image processing, analysis and quality control, posing significant reproducibility limitations. Our software platform, Rodent Automated Bold Improvement of EPI Sequences (RABIES), is a novel pipeline designed to address these limitations for preprocessing, quality control, and confound correction, along with best practices for reproducibility and transparency. We demonstrate the robustness of the preprocessing workflow by validating performance across multiple acquisition sites and both mouse and rat data. Building upon a thorough investigation into data quality metrics across acquisition sites, we introduce guidelines for the quality control of network analysis and offer recommendations for addressing issues. Taken together, the RABIES software will allow the emerging community to adopt reproducible practices and foster progress in translational neuroscience.