Classifying brain states and determining the discriminating activation patterns: Support Vector Machine on functional MRI data

Enhanced generalization and specialization of brain representations of semantic knowledge in healthy aging

Aging is often associated with a decrease in cognitive capacities. However, semantic memory appears relatively well preserved in healthy aging. Both behavioral and neuroimaging studies support the view that changes in brain networks contribute to this preservation of semantic cognition. However, little is known about the role of healthy aging in the brain representation of semantic categories. Here we used pattern classification analyses and computational models to examine the neural representations of living and non-living word concepts. The results demonstrate that brain representations of animacy in healthy aging exhibit increased similarity across categories, even across different task contexts. This pattern of results aligns with the neural dedifferentiation hypothesis that proposes that aging is associated with decreased specificity in brain activity patterns and less efficient neural resource allocation. However, the loss in neural specificity for different categories was accompanied by increased dissimilarity of item-based conceptual representations within each category. Taken together, the age-related patterns of increased generalization and specialization in the brain representations of semantic knowledge may reflect a compensatory mechanism that enables a more efficient coding scheme characterized by both compression and sparsity, thereby helping to optimize the limited neural resources and maintain semantic processing in the healthy aging brain.

Classifying Unconscious, Psychedelic, and Neuropsychiatric Brain States with Functional Connectivity, Graph Theory, and Cortical Gradient Analysis

Accurate and generalizable classification of brain states is essential for understanding their neural underpinnings and improving clinical diagnostics. Traditionally, functional connectivity patterns and graph-theoretic metrics have been utilized. However, cortical gradient features, which reflect global brain organization, offer a complementary approach. We hypothesized that a machine learning model integrating these three feature sets would effectively discriminate between baseline and atypical brain states across a wide spectrum of conditions, even though the underlying neural mechanisms vary. To test this, we extracted features from brain states associated with three meta-conditions including unconsciousness (NREM2 sleep, propofol deep sedation, and propofol general anesthesia), psychedelic states induced by hallucinogens (subanesthetic ketamine, lysergic acid diethylamide, and nitrous oxide), and neuropsychiatric disorders (attention-deficit hyperactivity disorder, bipolar disorder, and schizophrenia). We used support vector machine with nested cross-validation to construct our models. The soft voting ensemble model marked the average balanced accuracy (average of specificity and sensitivity) of 79% (62-98% across all conditions), outperforming individual base models (70-76%). Notably, our models exhibited varying degrees of transferability across different datasets, with performance being dependent on the specific brain states and feature sets used. Feature importance analysis across meta-conditions suggests that the underlying neural mechanisms vary significantly, necessitating tailored approaches for accurate classification of specific brain states. This finding underscores the value of our feature-integrated ensemble models, which leverage the strengths of multiple feature types to achieve robust performance across a broader range of brain states. While our approach offers valuable insights into the neural signatures of different brain states, future work is needed to develop and validate even more generalizable models that can accurately classify brain states across a wider array of conditions.

Abnormal brain activation during speech perception and production in children and adults with reading difficulty

Reading difficulty (RD) is associated with phonological deficits; however, it remains unknown whether the phonological deficits are different in children and adults with RD as reflected in foreign speech perception and production. In the current study, using functional Near-infrared spectroscopy (fNIRS), we found less difference between Chinese adults and Chinese children in the RD groups than the control groups in the activation of the right inferior frontal gyrus (IFG) and the dorsolateral prefrontal cortex (DLPFC) during Spanish speech perception, suggesting slowed development in these regions associated with RD. Furthermore, using multivariate pattern analysis (MVPA), we found that activation patterns in the left middle temporal gyrus (MTG), premotor, supplementary motor area (SMA), and IFG could serve as reliable markers of RD. We provide both behavioral and neurological evidence for impaired speech perception and production in RD readers which can serve as markers of RD.

Hue and orientation pinwheels in macaque area V4

Area V4 is an intermediate-level area of the macaque visual cortical hierarchy that serves key functions in visual processing by integrating inputs from lower areas such as V1 and V2 and providing feedforward inputs to many higher cortical areas. Previous V4 imaging studies have focused on differential responses to color, orientation, disparity, and motion stimuli, but many details of the spatial organization of significant hue and orientation tuning have not been fully described. We used support vector machine (SVM) decoding of intrinsic cortical single-condition responses to generate high-resolution maps of hue and orientation tuning and to describe the organization of hue and orientation pinwheels in V4. Like V1 and V2, V4 contains maps of orientation that are organized as pinwheels. V4 also contains maps of hue that are organized as pinwheels, whose circular organization more closely represents the perception of hue than is observed in antecedent cortical areas. Unlike V1, where orientation is continuously mapped across the surface, V4 hue and orientation pinwheels are organized in limited numbers of pinwheel sequences. The organization of these sequences and the distance between pinwheels may provide insight into the functional organization of V4. Regions significantly tuned for hue occupy roughly four times that of the orientation, are largely separated from each other, and overlap by roughly 5%. This spatial organization is largely consistent with segregated inputs arising from V2 thin and interstripes. This modular organization of V4 suggests that further integration of color and shape might occur in higher areas in inferotemporal cortical.NEW & NOTEWORTHY The representation of hue and orientation in macaque monkey area V4 was determined by intrinsic cortical imaging of responses to isoluminant hues and achromatic grating stimuli. Vector summation of support vector machine (SVM) decoded single-condition responses was used to generate hue and orientation maps that, like V1 orientation maps, were both characterized by distinct pinwheel patterns. These data suggest that pinwheels are an important structure to represent different stimulus features across multiple visual cortical areas.

Heritability of functional gradients in the human subcortico-cortical connectivity

The human subcortex plays a pivotal role in cognition and is widely implicated in the pathophysiology of many psychiatric disorders. However, the heritability of functional gradients based on subcortico-cortical functional connectivity remains elusive. Here, leveraging twin functional MRI (fMRI) data from both the Human Connectome Project (n = 1023) and the Adolescent Brain Cognitive Development study (n = 936) datasets, we construct large-scale subcortical functional gradients and delineate an increased principal functional gradient pattern from unimodal sensory/motor networks to transmodal association networks. We observed that this principal functional gradient is heritable, and the strength of heritability exhibits a heterogeneous pattern along a hierarchical unimodal-transmodal axis in subcortex for both young adults and children. Furthermore, employing a machine learning framework, we show that this heterogeneous pattern of the principal functional gradient in subcortex can accurately discern the relationship between monozygotic twin pairs and dizygotic twin pairs with an accuracy of 76.2% (P < 0.001). The heritability of functional gradients is associated with the anatomical myelin proxied by MRI-derived T1-weighted/T2-weighted (T1w/T2w) ratio mapping in subcortex. This study provides new insights into the biological basis of subcortical functional hierarchy by revealing the structural and genetic properties of the subcortical functional gradients.

Age-related differences in finger interdependence during complex hand movements

The well-known decrease in finger dexterity during healthy aging leads to a significant reduction in quality of life. Still, the exact patterns of altered finger kinematics of older adults in daily life are fairly unexplored. Finger interdependence is the unintentional comovement of fingers that are not intended to move, and it is known to vary across the lifespan. Nevertheless, the magnitude and direction of age-related differences in finger interdependence are ambiguous across studies and tasks and have not been explored in the context of daily life finger movements. We investigated five different free and daily-life-inspired finger movements of the right, dominant hand as well as a sequential finger tapping task of the thumb against the other fingers, in 17 younger (22-37 yr) and 17 older (62-80 yr) adults using an exoskeleton data glove for data recording. Using inferential statistics, we found that the unintentional comovement of fingers generally decreases with age in all performed daily-life-inspired movements. Finger tapping, however, showed a trend towards higher finger interdependence for older compared with younger adults. Using machine learning, we predicted the age group of a person from finger interdependence features of single movement trials significantly better than chance level for the daily-life-inspired movements, but not for finger tapping. Taken together, we show that for specific tasks, decreased finger interdependence (i.e., less comovement) could potentially act as a marker of human aging that specifically characterizes older adults' complex finger movements in daily life.NEW & NOTEWORTHY Kinematic finger movement data were analyzed with regard to age-related differences. Extensive analyses of complex and daily-life-inspired movements reveal that the direction of age effects is not uniform but task-dependent: Although older adults generally show more finger interdependence than younger adults in a simple finger tapping task, this effect is reversed for daily-life-inspired movement tasks. For these tasks, finger interdependence indices offer potential new markers to predict the age group of an individual using machine learning approaches.

Decreased intrinsic neural timescale in treatment-naïve adolescent depression

BACKGROUND

Major depressive disorder (MDD) is mainly characterized by its core dysfunction in higher-order brain cortices involved in emotional and cognitive processes, whose neurobiological basis remains unclear. In this study, we applied a relatively new developed resting-state functional magnetic resonance imaging (rs-fMRI) method of intrinsic neural timescale (INT), which reflects how long neural information is stored in a local brain area and reflects an ability of information integration, to investigate the local intrinsic neural dynamics using univariate and multivariate analyses in adolescent depression.

METHOD

Based on the rs-fMRI data of sixty-six treatment-naïve adolescents with MDD and fifty-two well-matched healthy controls (HCs), we calculated an INT by assessing the magnitude of autocorrelation of the resting-state brain activity, and then compared the difference of INT between the two groups. Correlation between abnormal INT and clinical features was performed. We also utilized multivariate pattern analysis to determine whether INT could differentiate MDD patients from HCs at the individual level.

RESULT

Compared with HCs, patients with MDD showed shorter INT widely distributed in cortical and partial subcortical regions. Interestingly, the decreased INT in the left hippocampus was related to disease severity of MDD. Furthermore, INT can distinguish MDD patients from HCs with the most discriminative regions located in the dorsolateral prefrontal cortex, angular, middle occipital gyrus, and cerebellar posterior lobe.

CONCLUSION

Our research aids in advancing understanding the brain abnormalities of treatment-naïve adolescents with MDD from the perspective of the local neural dynamics, highlighting the significant role of INT in understanding neurophysiological mechanisms. This study shows that the altered intrinsic timescales of local neural signals widely distributed in higher-order brain cortices regions may be the neurodynamic basis of cognitive and emotional disturbances in MDD patients, and provides preliminary support for the suggestion that these could be used to aid the identification of MDD patients in clinical practice.

Decoding fMRI data with support vector machines and deep neural networks

BACKGROUND

Multivoxel pattern analysis (MVPA) examines fMRI activation patterns associated with different cognitive conditions. Support vector machines (SVMs) are the predominant method in MVPA. While SVM is intuitive and easy to apply, it is mainly suitable for analyzing data that are linearly separable. Convolutional neural networks (CNNs) are known to have the ability to approximate nonlinear relationships. Applications of CNN to fMRI data are beginning to appear with increasing frequency, but our understanding of the similarities and differences between CNN models and SVM models is limited.

NEW METHOD

We compared the two methods when they are applied to the same datasets. Two datasets were considered: (1) fMRI data collected from participants during a cued visual spatial attention task and (2) fMRI data collected from participants viewing natural images containing varying degrees of affective content.

RESULTS

We found that (1) both SVM and CNN are able to achieve above-chance decoding accuracies for attention control and emotion processing in both the primary visual cortex and the whole brain, (2) the CNN decoding accuracies are consistently higher than that of the SVM, (3) the SVM and CNN decoding accuracies are generally not correlated, and (4) the heatmaps derived from SVM and CNN are not significantly overlapping.

COMPARISON WITH EXISTING METHODS

By comparing SVM and CNN we pointed out the similarities and differences between the two methods.

CONCLUSIONS

SVM and CNN rely on different neural features for classification. Applying both to the same data may yield a more comprehensive understanding of neuroimaging data.

Neuroimaging Methods for MRI Analysis in CSF Biomarkers Studies

Among others, the existence of pathophysiological biomarkers such as cerebrospinal fluid (CSF) Aβ-42, t-tau, and p-tau preceding the onset of Alzheimer's disease (AD) symptomatology has shifted the conceptualization of AD as a continuum. In addition, magnetic resonance imaging (MRI) enables the study of structural and functional cross-sectional correlates and longitudinal changes in vivo, and therefore, the combination of CSF data and imaging analyses emerges as a synergistic approach to understand the structural correlates related with specific AD-related biomarkers. In this chapter, we describe the methods used in neuroimaging that will allow researchers to combine data on CSF metabolites with imaging analyses.

Gaussian Process-based prediction of memory performance and biomarker status in ageing and Alzheimer's disease-A systematic model evaluation

Neuroimaging markers based on Magnetic Resonance Imaging (MRI) combined with various other measures (such as genetic covariates, biomarkers, vascular risk factors, neuropsychological tests etc.) might provide useful predictions of clinical outcomes during the progression towards Alzheimer's disease (AD). The use of multiple features in predictive frameworks for clinical outcomes has become increasingly prevalent in AD research. However, many studies do not focus on systematically and accurately evaluating combinations of multiple input features. Hence, the aim of the present work is to explore and assess optimal combinations of various features for MR-based prediction of (1) cognitive status and (2) biomarker positivity with a multi-kernel learning Gaussian process framework. The explored features and parameters included (A) combinations of brain tissues, modulation, smoothing, and image resolution; (B) incorporating demographics & clinical covariates; (C) the impact of the size of the training data set; (D) the influence of dimensionality reduction and the choice of kernel types. The approach was tested in a large German cohort including 959 subjects from the multicentric longitudinal study of cognitive impairment and dementia (DELCODE). Our evaluation suggests the best prediction of memory performance was obtained for a combination of neuroimaging markers, demographics, genetic information (ApoE4) and CSF biomarkers explaining 57% of outcome variance in out-of-sample predictions. The highest performance for Aβ42/40 status classification was achieved for a combination of demographics, ApoE4, and a memory score while usage of structural MRI further improved the classification of individual patient's pTau status.

Abnormal dynamic functional connectivity in the hippocampal subregions of patients with untreated moderate-to-severe obstructive sleep apnea

OBJECTIVE

To investigate the dynamic change characteristics of dynamic functional connectivity (dFC) between the hippocampal subregions (anterior and posterior) and other brain regions in obstructive sleep apnoea (OSA) and its relationship with cognitive function, and to explore whether these characteristics can be used to distinguish OSA from healthy controls (HCs).

METHODS

Eighty-five patients with newly diagnosed moderate-to-severe OSA and 85 HCs were enrolled. All participants underwent resting-state functional magnetic resonance imaging (fMRI). The difference between dFC values between the hippocampal subregions and other brain regions in OSA patients and HCs was compared using the two-sample t tests. Correlation analyses were used to assess the relationship between dFC, clinical data, and cognitive functions in OSA patients. dFC values from different brain regions were used as classification features to distinguish between the two groups using a support vector machine.

RESULTS

Compared with HCs, the dFC values between the left anterior hippocampus and right culmen of the cerebellum anterior lobe, right anterior hippocampus and left lingual gyrus, and left posterior hippocampus and left precentral gyrus were significantly lower, and the dFC values between the left posterior hippocampus and precuneus were significantly higher in OSA patients. The dFC values between the left posterior hippocampus and the precuneus of OSA patients were associated with sleep-related indicators and Montreal Cognitive Assessment scores. Support vector machine analysis results showed that dFC values in different brain regions could distinguish OSA patients from HCs.

CONCLUSION

dFC patterns between the hippocampal subregions and other brain regions were altered in patients with OSA, including the cerebellum, default mode networks, sensorimotor networks, and visual function networks, which is possibly associated with cognitive decline. In addition, the dFC values of different brain regions could effectively distinguish OSA patients from HCs. These findings provide new perspectives on neurocognition in these patients.

Genetic risks of Alzheimer's by APOE and MAPT on cortical morphology in young healthy adults

Genetic risk factors such as APOE ε4 and MAPT (rs242557) A allele are associated with amyloid and tau pathways and grey matter changes at both early and established stages of Alzheimer's disease, but their effects on cortical morphology in young healthy adults remain unclear. A total of 144 participants aged from 18 to 24 underwent 3T MRI and genotyping for APOE and MAPT to investigate unique impacts of these genetic risk factors in a cohort without significant comorbid conditions such as metabolic and cardiovascular diseases. We segmented the cerebral cortex into 68 regions and calculated the cortical area, thickness, curvature and folding index for each region. Then, we trained machine learning models to classify APOE and MAPT genotypes using these morphological features. In addition, we applied a growing hierarchical self-organizing maps algorithm, which clustered the 68 regions into 4 subgroups representing different morphological patterns. Then, we performed general linear model analyses to estimate the interaction between APOE and MAPT on cortical patterns. We found that the classifiers using all cortical features could accurately classify individuals carrying genetic risks of dementia outperforming each individual feature alone. APOE ε4 carriers had a more convoluted and thinner cortex across the cerebral cortex. A similar pattern was found in MAPT A allele carriers only in the regions that are vulnerable for early tau pathology. With the clustering analysis, we found a synergetic effect between APOE ε4 and MAPT A allele, i.e. carriers of both risk factors showed the most deviation of cortical pattern from the typical pattern of that cluster. Genetic risk factors of dementia by APOE ε4 and MAPT (rs242557) A allele were associated with variations of cortical morphology, which can be observed in young healthy adults more than 30 years before Alzheimer's pathology is likely to occur and 50 years before dementia symptoms may begin.

Groupwise structural sparsity for discriminative voxels identification

This paper investigates the selection of voxels for functional Magnetic Resonance Imaging (fMRI) brain data. We aim to identify a comprehensive set of discriminative voxels associated with human learning when exposed to a neutral visual stimulus that predicts an aversive outcome. However, due to the nature of the unconditioned stimuli (typically a noxious stimulus), it is challenging to obtain sufficient sample sizes for psychological experiments, given the tolerability of the subjects and ethical considerations. We propose a stable hierarchical voting (SHV) mechanism based on stability selection to address this challenge. This mechanism enables us to evaluate the quality of spatial random sampling and minimizes the risk of false and missed detections. We assess the performance of the proposed algorithm using simulated and publicly available datasets. The experiments demonstrate that the regularization strategy choice significantly affects the results' interpretability. When applying our algorithm to our collected fMRI dataset, it successfully identifies sparse and closely related patterns across subjects and displays stable weight maps for three experimental phases under the fear conditioning paradigm. These findings strongly support the causal role of aversive conditioning in altering visual-cortical activity.

Differential processing of intrinsic vs. extrinsic coordinates in wrist movement: connectivity and chronometry perspectives

This study explores brain-network differences between the intrinsic and extrinsic motor coordinate frames. A connectivity model showing the coordinate frames difference was obtained using brain fMRI data of right wrist isometric flexions and extensions movements, performed in two forearm postures. The connectivity model was calculated by machine-learning-based neural representation and effective functional connectivity using psychophysiological interaction and dynamic causal modeling analyses. The model indicated the network difference wherein the inferior parietal lobule receives extrinsic information from the rostral lingual gyrus through the superior parietal lobule and transmits intrinsic information to the Handknob, whereas extrinsic information is transmitted to the Handknob directly from the rostral lingual gyrus. A behavioral experiment provided further evidence on the difference between motor coordinate frames showing onset timing delay of muscle activity of intrinsic coordinate-directed wrist movement compared to extrinsic one. These results suggest that, if the movement is externally directed, intrinsic coordinate system information is bypassed to reach the primary motor area.

Decoding fMRI Data: A Comparison Between Support Vector Machines and Deep Neural Networks

Multivoxel pattern analysis (MVPA) examines the differences in fMRI activation patterns associated with different cognitive conditions and provides information not possible with the conventional univariate analysis. Support vector machines (SVMs) are the predominant machine learning method in MVPA. SVMs are intuitive and easy to apply. The limitation is that it is a linear method and mainly suitable for analyzing data that are linearly separable. Convolutional neural networks (CNNs), a class of AI models originally developed for object recognition, are known to have the ability to approximate nonlinear relationships. CNNs are rapidly becoming an alternative to SVMs. The purpose of this study is to compare the two methods when they are applied to the same datasets. Two datasets were considered: (1) fMRI data collected from participants during a cued visual spatial attention task (the attention dataset) and (2) fMRI data collected from participants viewing natural images containing varying degrees of affective content (the emotion dataset). We found that (1) both SVM and CNN are able to achieve above chance level decoding accuracies for attention control and emotion processing in both the primary visual cortex and the whole brain with, (2) the CNN decoding accuracies are consistently higher than that of the SVM, (3) the SVM and CNN decoding accuracies are generally not correlated with each other, and (4) the heatmaps derived from SVM and CNN are not significantly overlapping. These results suggest that (1) there are both linearly separable features and nonlinearly separable features in fMRI data that distinguish cognitive conditions and (2) applying both SVM and CNN to the same data may yield a more comprehensive understanding of neuroimaging data.

Key points

We compared the performance and characteristics of SVM and CNN, two major methods in MVPA analysis of neuroimaging data, by applying them to the same two fMRI datasets.Both SVM and CNN achieved decoding accuracies above chance level for both datasets in the chosen ROIs and the CNN decoding accuracies were consistently higher than those of SVM.The heatmaps derived from SVM and CNN, which assess the contribution of voxels or brain regions to MVPA decoding performance, showed no significant overlap, providing evidence that the two methods depend on distinct brain activity patterns for decoding cognitive conditions.

Variational relevance evaluation of individual fMRI data enables deconstruction of task-dependent neural dynamics

In neuroimaging research, univariate analysis has always been used to localize "representations" at the microscale, whereas network approaches have been applied to characterize transregional "operations". How are representations and operations linked through dynamic interactions? We developed the variational relevance evaluation (VRE) method to analyze individual task fMRI data, which selects informative voxels during model training to localize the "representation", and quantifies the dynamic contributions of single voxels across the whole-brain to different cognitive functions to characterize the "operation". Using 15 individual fMRI data files for higher visual area localizers, we evaluated the characterization of selected voxel positions of VRE and revealed different object-selective regions functioning in similar dynamics. Using another 15 individual fMRI data files for memory retrieval after offline learning, we found similar task-related regions working in different neural dynamics for tasks with diverse familiarities. VRE demonstrates a promising horizon in individual fMRI research.

Multiscale functional connectivity patterns of the aging brain learned from harmonized rsfMRI data of the multi-cohort iSTAGING study

To learn multiscale functional connectivity patterns of the aging brain, we built a brain age prediction model of functional connectivity measures at seven scales on a large fMRI dataset, consisting of resting-state fMRI scans of 4186 individuals with a wide age range (22 to 97 years, with an average of 63) from five cohorts. We computed multiscale functional connectivity measures of individual subjects using a personalized functional network computational method, harmonized the functional connectivity measures of subjects from multiple datasets in order to build a functional brain age model, and finally evaluated how functional brain age gap correlated with cognitive measures of individual subjects. Our study has revealed that functional connectivity measures at multiple scales were more informative than those at any single scale for the brain age prediction, the data harmonization significantly improved the brain age prediction performance, and the data harmonization in the functional connectivity measures' tangent space worked better than in their original space. Moreover, brain age gap scores of individual subjects derived from the brain age prediction model were significantly correlated with clinical and cognitive measures. Overall, these results demonstrated that multiscale functional connectivity patterns learned from a large-scale multi-site rsfMRI dataset were informative for characterizing the aging brain and the derived brain age gap was associated with cognitive and clinical measures.

The causal mechanisms underlying analogical reasoning performance improvement by executive attention intervention

Analogical reasoning is important for human. We have found that a short executive attention intervention improved analogical reasoning performance in healthy young adults. Nevertheless, previous electrophysiological evidence was limited for comprehensively characterizing the neural mechanisms underlying the improvement. And although we hypothesized that the intervention improved active inhibitory control and attention shift first and then relation integration, it is still unclear whether there are two sequential cognitive neural activities were indeed changed during analogical reasoning. In the present study, we combined hypothesis with multivariate pattern analysis (MVPA) to explore the effects of the intervention on electrophysiology. Results showed that in the resting state after the intervention, alpha and high gamma power and the functional connectivity between the anterior and middle in the alpha band could discriminate the experimental group from the active control group, respectively. These indicated that the intervention influenced the activity of multiple bands and the interaction of frontal and parietal regions. In the analogical reasoning, alpha, theta, and gamma activities could also fulfill such discrimination, and furthermore, they were sequential (alpha first, theta, and gamma later). These results directly supported our previous hypothesis. The present study deepens our understanding about how executive attention contributes to higher-order cognition.

Global diversity in individualized cortical network topography

Individualized cortical network topography (ICNT) varies between people and exhibits great variability in the association networks in the human brain. However, these findings were mainly discovered in Western populations. It remains unclear whether and how ICNT is shaped by the non-Western populations. Here, we leveraged a multisession hierarchical Bayesian model to define individualized functional networks in White American and Han Chinese populations with data from both US and Chinese Human Connectome Projects. We found that both the size and spatial topography of individualized functional networks differed between White American and Han Chinese groups, especially in the heteromodal association cortex (including the ventral attention, control, language, dorsal attention, and default mode networks). Employing a support vector machine, we then demonstrated that ethnicity-related ICNT diversity can be used to identify an individual's ethnicity with high accuracy (74%, pperm < 0.0001), with heteromodal networks contributing most to the classification. This finding was further validated through mass-univariate analyses with generalized additive models. Moreover, we reveal that the spatial heterogeneity of ethnic diversity in ICNT correlated with fundamental properties of cortical organization, including evolutionary cortical expansion, brain myelination, and cerebral blood flow. Altogether, this case study highlights a need for more globally diverse and publicly available neuroimaging datasets.

Framework for Accurate Classification of Self-Reported Stress From Multisession Functional MRI Data of Veterans With Posttraumatic Stress

Background: Posttraumatic stress disorder (PTSD) is a significant burden among combat Veterans returning from the wars in Iraq and Afghanistan. While empirically supported treatments have demonstrated reductions in PTSD symptomatology, there remains a need to improve treatment effectiveness. Functional magnetic resonance imaging (fMRI) neurofeedback has emerged as a possible treatment to ameliorate PTSD symptom severity. Virtual reality (VR) approaches have also shown promise in increasing treatment compliance and outcomes. To facilitate fMRI neurofeedback-associated therapies, it would be advantageous to accurately classify internal brain stress levels while Veterans are exposed to trauma-associated VR imagery. Methods: Across 2 sessions, we used fMRI to collect neural responses to trauma-associated VR-like stimuli among male combat Veterans with PTSD symptoms (N = 8). Veterans reported their self-perceived stress level on a scale from 1 to 8 every 15 s throughout the fMRI sessions. In our proposed framework, we precisely sample the fMRI data on cortical gray matter, blurring the data along the gray-matter manifold to reduce noise and dimensionality while preserving maximum neural information. Then, we independently applied 3 machine learning (ML) algorithms to this fMRI data collected across 2 sessions, separately for each Veteran, to build individualized ML models that predicted their internal brain states (self-reported stress responses). Results: We accurately classified the 8-class self-reported stress responses with a mean (± standard error) root mean square error of 0.6 (± 0.1) across all Veterans using the best ML approach. Conclusions: The findings demonstrate the predictive ability of ML algorithms applied to whole-brain cortical fMRI data collected during individual Veteran sessions. The framework we have developed to preprocess whole-brain cortical fMRI data and train ML models across sessions would provide a valuable tool to enable individualized real-time fMRI neurofeedback during VR-like exposure therapy for PTSD.

Explainable AI: A review of applications to neuroimaging data

Deep neural networks (DNNs) have transformed the field of computer vision and currently constitute some of the best models for representations learned via hierarchical processing in the human brain. In medical imaging, these models have shown human-level performance and even higher in the early diagnosis of a wide range of diseases. However, the goal is often not only to accurately predict group membership or diagnose but also to provide explanations that support the model decision in a context that a human can readily interpret. The limited transparency has hindered the adoption of DNN algorithms across many domains. Numerous explainable artificial intelligence (XAI) techniques have been developed to peer inside the "black box" and make sense of DNN models, taking somewhat divergent approaches. Here, we suggest that these methods may be considered in light of the interpretation goal, including functional or mechanistic interpretations, developing archetypal class instances, or assessing the relevance of certain features or mappings on a trained model in a post-hoc capacity. We then focus on reviewing recent applications of post-hoc relevance techniques as applied to neuroimaging data. Moreover, this article suggests a method for comparing the reliability of XAI methods, especially in deep neural networks, along with their advantages and pitfalls.

A real-time fMRI neurofeedback system for the clinical alleviation of depression with a subject-independent classification of brain states: A proof of principle study

Most clinical neurofeedback studies based on functional magnetic resonance imaging use the patient's own neural activity as feedback. The objective of this study was to create a subject-independent brain state classifier as part of a real-time fMRI neurofeedback (rt-fMRI NF) system that can guide patients with depression in achieving a healthy brain state, and then to examine subsequent clinical changes. In a first step, a brain classifier based on a support vector machine (SVM) was trained from the neural information of happy autobiographical imagery and motor imagery blocks received from a healthy female participant during an MRI session. In the second step, 7 right-handed female patients with mild or moderate depressive symptoms were trained to match their own neural activity with the neural activity corresponding to the “happiness emotional brain state” of the healthy participant. The training (4 training sessions over 2 weeks) was carried out using the rt-fMRI NF system guided by the brain-state classifier we had created. Thus, the informative voxels previously obtained in the first step, using SVM classification and Effect Mapping, were used to classify the Blood-Oxygen-Level Dependent (BOLD) activity of the patients and converted into real-time visual feedback during the neurofeedback training runs. Improvements in the classifier accuracy toward the end of the training were observed in all the patients [Session 4–1 Median = 6.563%; Range = 4.10–27.34; Wilcoxon Test (0), 2-tailed p = 0.031]. Clinical improvement also was observed in a blind standardized clinical evaluation [HDRS CE2-1 Median = 7; Range 2 to 15; Wilcoxon Test (0), 2-tailed p = 0.016], and in self-report assessments [BDI-II CE2-1 Median = 8; Range 1–15; Wilcoxon Test (0), 2-tailed p = 0.031]. In addition, the clinical improvement was still present 10 days after the intervention [BDI-II CE3-2_Median = 0; Range −1 to 2; Wilcoxon Test (0), 2-tailed p = 0.50/ HDRS CE3-2 Median = 0; Range −1 to 2; Wilcoxon Test (0), 2-tailed p = 0.625]. Although the number of participants needs to be increased and a control group included to confirm these findings, the results suggest a novel option for neural modulation and clinical alleviation in depression using noninvasive stimulation technologies.

Sex differences in the functional topography of association networks in youth

Prior work has shown that there is substantial interindividual variation in the spatial distribution of functional networks across the cerebral cortex, or functional topography. However, it remains unknown whether there are sex differences in the topography of individualized networks in youth. Here, we leveraged an advanced machine learning method (sparsity-regularized non-negative matrix factorization) to define individualized functional networks in 693 youth (ages 8 to 23 y) who underwent functional MRI as part of the Philadelphia Neurodevelopmental Cohort. Multivariate pattern analysis using support vector machines classified participant sex based on functional topography with 82.9% accuracy (P < 0.0001). Brain regions most effective in classifying participant sex belonged to association networks, including the ventral attention, default mode, and frontoparietal networks. Mass univariate analyses using generalized additive models with penalized splines provided convergent results. Furthermore, transcriptomic data from the Allen Human Brain Atlas revealed that sex differences in multivariate patterns of functional topography were spatially correlated with the expression of genes on the X chromosome. These results highlight the role of sex as a biological variable in shaping functional topography.

Incorporating CNN Features for Optimizing Performance of Ensemble Classifier for Cardiovascular Disease Prediction

Cardiovascular diseases (CVDs) have been regarded as the leading cause of death with 32% of the total deaths around the world. Owing to the large number of symptoms related to age, gender, demographics, and ethnicity, diagnosing CVDs is a challenging and complex task. Furthermore, the lack of experienced staff and medical experts, and the non-availability of appropriate testing equipment put the lives of millions of people at risk, especially in under-developed and developing countries. Electronic health records (EHRs) have been utilized for diagnosing several diseases recently and show the potential for CVDs diagnosis as well. However, the accuracy and efficacy of EHRs-based CVD diagnosis are limited by the lack of an appropriate feature set. Often, the feature set is very small and unable to provide enough features for machine learning models to obtain a good fit. This study solves this problem by proposing the novel use of feature extraction from a convolutional neural network (CNN). An ensemble model is designed where a CNN model is used to enlarge the feature set to train linear models including stochastic gradient descent classifier, logistic regression, and support vector machine that comprise the soft-voting based ensemble model. Extensive experiments are performed to analyze the performance of different ratios of feature sets to the training dataset. Performance analysis is carried out using four different datasets and results are compared with recent approaches used for CVDs. Results show the superior performance of the proposed model with 0.93 accuracy, and 0.92 scores each for precision, recall, and F1 score. Results indicate both the superiority of the proposed approach, as well as the generalization of the ensemble model using multiple datasets.

Central stress processing, T cell responsivity to stress hormones, and disease severity in multiple sclerosis

Epidemiological, clinical and neuroscientific studies support a link between psychobiological stress and multiple sclerosis. Neuroimaging suggests that blunted central stress processing goes along with higher multiple sclerosis severity, neuroendocrine studies suggest that blunted immune system sensitivity to stress hormones is linked to stronger neuroinflammation. Until now, however, no effort has been made to elucidate whether central stress processing and immune system sensitivity to stress hormones are related in a disease-specific fashion, and if so, whether this relation is clinically meaningful. Consequently, we conducted two functional MRI analyses based on a total of 39 persons with multiple sclerosis and 25 healthy persons. Motivated by findings of an altered interplay between neuroendocrine stress processing and T-cell glucocorticoid sensitivity in multiple sclerosis, we searched for neural networks whose stress task-evoked activity is differentially linked to peripheral T-cell glucocorticoid signalling in patients versus healthy persons as a potential indicator of disease-specific CNS–immune crosstalk. Subsequently, we tested whether this activity is simultaneously related to disease severity. We found that activity of a network comprising right anterior insula, right fusiform gyrus, left midcingulate and lingual gyrus was differentially coupled to T-cell glucocorticoid signalling across groups. This network’s activity was simultaneously linked to patients’ lesion volume, clinical disability and information-processing speed. Complementary analyses revealed that T-cell glucocorticoid signalling was not directly linked to disease severity. Our findings show that alterations in the coupling between central stress processing and T-cell stress hormone sensitivity are related to key severity measures of multiple sclerosis.

fMRI Brain Decoding and Its Applications in Brain-Computer Interface: A Survey

Brain neural activity decoding is an important branch of neuroscience research and a key technology for the brain-computer interface (BCI). Researchers initially developed simple linear models and machine learning algorithms to classify and recognize brain activities. With the great success of deep learning on image recognition and generation, deep neural networks (DNN) have been engaged in reconstructing visual stimuli from human brain activity via functional magnetic resonance imaging (fMRI). In this paper, we reviewed the brain activity decoding models based on machine learning and deep learning algorithms. Specifically, we focused on current brain activity decoding models with high attention: variational auto-encoder (VAE), generative confrontation network (GAN), and the graph convolutional network (GCN). Furthermore, brain neural-activity-decoding-enabled fMRI-based BCI applications in mental and psychological disease treatment are presented to illustrate the positive correlation between brain decoding and BCI. Finally, existing challenges and future research directions are addressed.

A developmental reduction of the excitation:inhibition ratio in association cortex during adolescence

Adolescence is hypothesized to be a critical period for the development of association cortex. A reduction of the excitation:inhibition (E:I) ratio is a hallmark of critical period development; however, it has been unclear how to assess the development of the E:I ratio using noninvasive neuroimaging techniques. Here, we used pharmacological fMRI with a GABAergic benzodiazepine challenge to empirically generate a model of E:I ratio based on multivariate patterns of functional connectivity. In an independent sample of 879 youth (ages 8 to 22 years), this model predicted reductions in the E:I ratio during adolescence, which were specific to association cortex and related to psychopathology. These findings support hypothesized shifts in E:I balance of association cortices during a neurodevelopmental critical period in adolescence.

Discriminating subclinical depression from major depression using multi-scale brain functional features: A radiomics analysis

BACKGROUND

The diagnosis of subclinical depression (SD) currently relies exclusively on subjective clinical scores and structured interviews, which shares great similarities with major depression (MD) and increases the risk of misdiagnosis of SD and MD. This study aimed to develop a method of disease classification for SD and MD by resting-state functional features using radiomics strategy.

METHODS

Twenty-six SD, 36 MD subjects and 33 well-matched healthy controls (HC) were recruited and underwent resting-state functional magnetic resonance imaging (rs-fMRI). A novel radiomics analysis was proposed to discriminate SD from MD. Multi-scale brain functional features were extracted to explore a comprehensive representation of functional characteristics. A two-level feature selection strategy and support vector machine (SVM) were employed for classification.

RESULTS

The overall classification accuracy among SD, MD and HC groups was 84.21%. Particularly, the model excellently distinguished SD from MD with 96.77% accuracy, 100% sensitivity, and 92.31% specificity. Moreover, features with high discriminative power to distinguish SD from MD showed a strong association with default mode network, frontoparietal network, affective network, and visual network regions.

LIMITATION

The sample size was relatively small, which may limit the application in clinical translation to some extent.

CONCLUSION

These findings demonstrated that a valid radiomics approach based on functional measures can discriminate SD from MD with a high classification performance, facilitating an objective and reliable diagnosis individually in clinical practice. Features with high discriminative power may provide insight into a profound understanding of the brain functional impairments and pathophysiology of SD and MD.

Predicting Attentional Vulnerability to Sleep Deprivation: A Multivariate Pattern Analysis of DTI Data

BACKGROUND

Large individual differences exist in sleep deprivation (SD) induced sustained attention deterioration. Several brain imaging studies have suggested that the activities within frontal-parietal network, cortico-thalamic connections, and inter-hemispheric connectivity might underlie the neural correlates of vulnerability/resistance to SD. However, those traditional approaches are based on average estimates of differences at the group level. Currently, a neuroimaging marker that can reliably predict this vulnerability at the individual level is lacking.

METHODS

Efficient transfer of information relies on the integrity of white matter (WM) tracts in the human brain, we therefore applied machine learning approach to investigate whether the WM diffusion metrics can predict vulnerability to SD. Forty-nine participants completed the psychomotor vigilance task (PVT) both after resting wakefulness (RW) and after 24 h of sleep deprivation (SD). The number of PVT lapse (reaction time > 500 ms) was calculated for both RW condition and SD condition and participants were categorized as vulnerable (24 participants) or resistant (25 participants) to SD according to the change in the number of PVT lapses between the two conditions. Diffusion tensor imaging were acquired to extract four multitype WM features at a regional level: fractional anisotropy, mean diffusivity, axial diffusivity, and radial diffusivity. A linear support vector machine (LSVM) learning approach using leave-one-out cross-validation (LOOCV) was performed to assess the discriminative power of WM features in SD-vulnerable and SD-resistant participants.

RESULTS

LSVM analysis achieved a correct classification rate of 83.67% (sensitivity: 87.50%; specificity: 80.00%; and area under the receiver operating characteristic curve: 0.85) for differentiating SD-vulnerable from SD-resistant participants. WM fiber tracts that contributed most to the classification model were primarily commissural pathways (superior longitudinal fasciculus), projection pathways (posterior corona radiata, anterior limb of internal capsule) and association pathways (body and genu of corpus callosum). Furthermore, we found a significantly negative correlation between changes in PVT lapses and the LSVM decision value.

CONCLUSION

These findings suggest that WM fibers connecting (1) regions within frontal-parietal attention network, (2) the thalamus to the prefrontal cortex, and (3) the left and right hemispheres contributed the most to classification accuracy.

Abnormal Spontaneous Brain Activity and Cognitive Impairment in Obstructive Sleep Apnea

PURPOSE

This study aimed to explore the alterations in spontaneous brain activity in obstructive sleep apnea (OSA) using percent amplitude of fluctuation (PerAF) and investigate the relationship between abnormal spontaneous brain activity and cognitive impairment in OSA.

PATIENTS AND METHODS

Overall, 52 patients with moderate to severe OSA and 61 healthy controls (HCs) were eventually enrolled in this study. All participants underwent resting-state functional magnetic resonance (rs-fMRI) and T1-weighted imaging. The PerAF was calculated and compared between patients with OSA and HCs, with voxel level P < 0.001 and cluster level P < 0.05 corrected with Gaussian Random Field was be considered statistically different. A partial correlation analysis was used to assess the relationship between altered PerAF and clinical assessments in patients with OSA.

RESULTS

Compared to HCs, patients with OSA had significantly lower PerAF values in the right rectal gyrus and left superior frontal gyrus, but higher PerAF values in the right cerebellum posterior lobe and left middle frontal gyrus. The PerAF values of some specific regions in patients with OSA correlated with sleep efficiency and Montreal Cognitive Assessment scores. Additionally, support vector machine analysis showed that PerAF values in all differential brain regions could differentiate patients with OSA from HCs with good accuracy.

CONCLUSION

Specific brain areas in OSA patients may exhibit aberrant neuronal activity, and these anomalies may be linked to decreased cognitive performance. This discovery offers fresh perspectives on these patients' neurocognition.

Self-Reported Complaints as Prognostic Markers for Outcome After Mild Traumatic Brain Injury in Elderly: A Machine Learning Approach

Self-reported complaints are common after mild traumatic brain injury (mTBI). Particularly in the elderly with mTBI, the pre-injury status might play a relevant role in the recovery process. In most mTBI studies, however, pre-injury complaints are neither analyzed nor are the elderly included. Here, we aimed to identify which individual pre- and post-injury complaints are potential prognostic markers for incomplete recovery (IR) in elderly patients who sustained an mTBI. Since patients report many complaints across several domains that are strongly related, we used an interpretable machine learning (ML) approach to robustly deal with correlated predictors and boost classification performance. Pre- and post-injury levels of 20 individual complaints, as self-reported in the acute phase, were analyzed. We used data from two independent studies separately: UPFRONT study was used for training and validation and ReCONNECT study for independent testing. Functional outcome was assessed with the Glasgow Outcome Scale Extended (GOSE). We dichotomized functional outcome into complete recovery (CR; GOSE = 8) and IR (GOSE ≤ 7). In total 148 elderly with mTBI (median age: 67 years, interquartile range [IQR]: 9 years; UPFRONT: N = 115; ReCONNECT: N = 33) were included in this study. IR was observed in 74 (50%) patients. The classification model (IR vs. CR) achieved a good performance (the area under the receiver operating characteristic curve [ROC-AUC] = 0.80; 95% CI: 0.74-0.86) based on a subset of only 8 out of 40 pre- and post-injury complaints. We identified increased neck pain (p = 0.001) from pre- to post-injury as the strongest predictor of IR, followed by increased irritability (p = 0.011) and increased forgetfulness (p = 0.035) from pre- to post-injury. Our findings indicate that a subset of pre- and post-injury physical, emotional, and cognitive complaints has predictive value for determining long-term functional outcomes in elderly patients with mTBI. Particularly, post-injury neck pain, irritability, and forgetfulness scores were associated with IR and should be assessed early. The application of an ML approach holds promise for application in self-reported questionnaires to predict outcomes after mTBI.

Neural Processes of Psychological Stress and Relaxation Predict the Future Evolution of Quality of Life in Multiple Sclerosis

Health-related quality of life (HRQoL) is an essential complementary parameter in the assessment of disease burden and treatment outcome in multiple sclerosis (MS) and can be affected by neuropsychiatric symptoms, which in turn are sensitive to psychological stress. However, until now, the impact of neurobiological stress and relaxation on HRQoL in MS has not been investigated. We thus evaluated whether the activity of neural networks triggered by mild psychological stress (elicited in an fMRI task comprising mental arithmetic with feedback) or by stress termination (i.e., relaxation) at baseline (T0) predicts HRQoL variations occurring between T0 and a follow-up visit (T1) in 28 patients using a robust regression and permutation testing. The median delay between T0 and T1 was 902 (range: 363–1,169) days. We assessed HRQoL based on the Hamburg Quality of Life Questionnaire in MS (HAQUAMS) and accounted for the impact of established HRQoL predictors and the cognitive performance of the participants. Relaxation-triggered activity of a widespread neural network predicted future variations in overall HRQoL (t = 3.68, p_{family−wise error [FWE]}-corrected = 0.008). Complementary analyses showed that relaxation-triggered activity of the same network at baseline was associated with variations in the HAQUAMS mood subscale on an α_FWE = 0.1 level (t = 3.37, p_FWE = 0.087). Finally, stress-induced activity of a prefronto-limbic network predicted future variations in the HAQUAMS lower limb mobility subscale (t = −3.62, p_FWE = 0.020). Functional neural network measures of psychological stress and relaxation contain prognostic information for future HRQoL evolution in MS independent of clinical predictors.

Machine Learning Models Prognosticate Functional Outcomes Better than Clinical Scores in Spontaneous Intracerebral Haemorrhage

OBJECTIVE

This study aims to develop and compare the use of deep neural networks (DNN) and support vector machines (SVM) to clinical prognostic scores for prognosticating 30-day mortality and 90-day poor functional outcome (PFO) in spontaneous intracerebral haemorrhage (SICH).

MATERIALS AND METHODS

We conducted a retrospective cohort study of 297 SICH patients between December 2014 and May 2016. Clinical data was collected from electronic medical records using standardized data collection forms. The machine learning workflow included imputation of missing data, dimensionality reduction, imbalanced-class correction, and evaluation using cross-validation and comparison of accuracy against clinical prognostic scores.

RESULTS

32 (11%) patients had 30-day mortality while 177 (63%) patients had 90-day PFO. For prognosticating 30-day mortality, the class-balanced accuracies for DNN (0.875; 95% CI 0.800-0.950; McNemar's p-value 1.000) and SVM (0.848; 95% CI 0.767-0.930; McNemar's p-value 0.791) were comparable to that of the original ICH score (0.833; 95% CI 0.748-0.918). The c-statistics for DNN (0.895; DeLong's p-value 0.715), and SVM (0.900; DeLong's p-value 0.619), though greater than that of the original ICH score (0.862), were not significantly different. For prognosticating 90-day PFO, the class-balanced accuracies for DNN (0.853; 95% CI 0.772-0.934; McNemar's p-value 0.003) and SVM (0.860; 95% CI 0.781-0.939; McNemar's p-value 0.004) were better than that of the ICH-Grading Scale (0.706; 95% CI 0.600-0.812). The c-statistic for SVM (0.883; DeLong's p-value 0.022) was significantly greater than that of the ICH-Grading Scale (0.778), while the c-statistic for DNN was 0.864 (DeLong's p-value 0.055).

CONCLUSION

We showed that the SVM model performs significantly better than clinical prognostic scores in predicting 90-day PFO in SICH.

Multi-site clustering and nested feature extraction for identifying autism spectrum disorder with resting-state fMRI

Brain functional connectivity (FC) derived from resting-state functional magnetic resonance imaging (rs-fMRI) has been widely employed to study neuropsychiatric disorders such as autism spectrum disorder (ASD). Existing studies usually suffer from (1) significant data heterogeneity caused by different scanners or studied populations in multiple sites, (2) curse of dimensionality caused by millions of voxels in each fMRI scan and a very limited number (tens or hundreds) of training samples, and (3) poor interpretability, which hinders the identification of reproducible disease biomarkers. To this end, we propose a Multi-site Clustering and Nested Feature Extraction (MC-NFE) method for fMRI-based ASD detection. Specifically, we first divide multi-site training data into ASD and healthy control (HC) groups. To model inter-site heterogeneity within each category, we use a similarity-driven multiview linear reconstruction model to learn latent representations and perform subject clustering within each group. We then design a nested singular value decomposition (SVD) method to mitigate inter-site heterogeneity and extract FC features by learning both local cluster-shared features across sites within each category and global category-shared features across ASD and HC groups, followed by a linear support vector machine (SVM) for ASD detection. Experimental results on 609 subjects with rs-fMRI from the ABIDE database with 21 imaging sites suggest that the proposed MC-NFE outperforms several state-of-the-art methods in ASD detection. The most discriminative FCs identified by the MC-NFE are mainly located in default mode network, salience network, and cerebellum region, which could be used as potential biomarkers for fMRI-based ASD analysis.

Construction of a 5-feature gene model by support vector machine for classifying osteoporosis samples

Osteoporosis is a progressive bone disease in the elderly and lacks an effective classification method of patients. This study constructed a gene signature for an accurate prediction and classification of osteoporosis patients. Three gene expression datasets of osteoporosis samples were acquired from the Gene Expression Omnibus database with pre-set criteria. Differentially expressed genes (DEGs) between normal and diseased osteoporosis samples were screened using Limma package in R language. Protein–protein interaction (PPI) network was established based on interaction data of the DEGs from the Human Protein Reference Database. Classification accuracy of the classifier was assessed with sensitivity, specificity and area under curve (AUC) using the pROC package in the R. Pathway enrichment analysis was performed on feature genes with clusterProfiler. A total of 310 differentially expressed genes between two samples were associated with positive regulation of protein secretion and cytokine secretion, neutrophil-mediated immunity, and neutrophil activation. PPI network of DEGs consisted of 12 genes. A SVM classifier based on five feature genes was developed to classify osteoporosis samples, showing a higher prediction accuracy and AUC for GSE35959, GSE62402, GSE13850, GSE56814, GSE56815 and GSE7429 datasets. A SVM classifier with a high accuracy was developed for predicting osteoporosis. The genes included may be the potential feature genes in osteoporosis development.AbbreviationsDEGs: Differentially expressed genes; PPI: protein–protein interaction; WHO: World Health Organization; SVM: Support vector machine; GEO: Gene Expression Omnibus; KEGG: Kyoto Encyclopedia of Genes and Genomes; GO: Gene Ontology; BP: Biological Process; CC: Cellular Component; MF: Molecular Function; SVM: Support vector machines

Autism Spectrum Disorder Studies Using fMRI Data and Machine Learning: A Review

Machine learning methods have been frequently applied in the field of cognitive neuroscience in the last decade. A great deal of attention has been attracted to introduce machine learning methods to study the autism spectrum disorder (ASD) in order to find out its neurophysiological underpinnings. In this paper, we presented a comprehensive review about the previous studies since 2011, which applied machine learning methods to analyze the functional magnetic resonance imaging (fMRI) data of autistic individuals and the typical controls (TCs). The all-round process was covered, including feature construction from raw fMRI data, feature selection methods, machine learning methods, factors for high classification accuracy, and critical conclusions. Applying different machine learning methods and fMRI data acquired from different sites, classification accuracies were obtained ranging from 48.3% up to 97%, and informative brain regions and networks were located. Through thorough analysis, high classification accuracies were found to usually occur in the studies which involved task-based fMRI data, single dataset for some selection principle, effective feature selection methods, or advanced machine learning methods. Advanced deep learning together with the multi-site Autism Brain Imaging Data Exchange (ABIDE) dataset became research trends especially in the recent 4 years. In the future, advanced feature selection and machine learning methods combined with multi-site dataset or easily operated task-based fMRI data may appear to have the potentiality to serve as a promising diagnostic tool for ASD.

Regression and Classification of Alzheimers Disease Diagnosis using NMF-TDNet Features from 3D Brain MR Image

With the development of deep learning and medical imaging technology, many researchers use convolutional neural network(CNN) to obtain deep-level features of medical image in order to better classify Alzheimer's disease (AD) and predict clinical scores. The principal component analysis network (PCANet) is a lightweight deep-learning network that mainly uses principal component analysis (PCA) to generate multilevel filter banks for the centralized learning of samples and then performs binarization and generates blockwise histograms to obtain image features. However, the dimensions of the extracted PCANet features reaching tens of thousands or even hundreds of thousands, and the formation of the multilevel filter banks is sample data dependent, reducing the flexibility of PCANet. In this paper, based on the idea of PCANet, we propose a data-independent network called the nonnegative matrix factorization tensor decomposition network (NMF-TDNet), which improves the computational efficiency and solves the data dependence problem of PCANet. In this network, we use nonnegative matrix factorization (NMF) instead of PCA to create multilevel filter banks for sample learning, then uses the learning results to build a higher-order tensor and perform tensor decomposition (TD) to achieve data dimensionality reduction, producing the final image features. Finally, our method use these features as the input of the support vector machine (SVM) for AD classification diagnosis and clinical score prediction. The performance of our method is extensively evaluated on the ADNI-1, ADNI-2 and OASIS datasets. The experimental results show that NMF-TDNet can achieve data dimensionality reduction (the dimensionality of the extracted features numbers only a few hundred dimensions, far less than the hundreds of thousands required by PCANet) and the NMF-TDNet features as input achieved superior performance than using PCANet features as input.

Characterizing the hyper- and hypometabolism in temporal lobe epilepsy using multivariate machine learning

Mesial temporal lobe epilepsy (MTLE) is the most common type of focal epilepsy, presenting both structural and metabolic abnormalities in the ipsilateral mesial temporal lobe. While it has been demonstrated that the metabolic abnormalities in MTLE actually extend beyond the epileptogenic zone, how such multidimensional information is associated with the diagnosis of MTLE remains to be tested. Here, we explore the whole-brain metabolic patterns in 23 patients with MTLE and 24 healthy controls using [¹⁸ F]fluorodeoxyglucose PET imaging. Based on a multivariate machine learning approach, we demonstrate that the brain metabolic patterns can discriminate patients with MTLE from controls with a superior accuracy (>95%). Importantly, voxels showing the most extreme contributing weights to the classification (i.e., the most important regional predictors) distribute across both hemispheres, involving both ipsilateral negative weights over the anterior part of lateral and medial temporal lobe, posterior insula, and lateral orbital frontal gyrus, and contralateral positive weights over the anterior frontal lobe, temporal lobe, and lingual gyrus. Through region-of-interest analyses, we verify that in patients with MTLE, the negatively weighted regions are hypometabolic, and the positively weighted regions are hypermetabolic, compared to controls. Interestingly, despite that both hypo- and hypermetabolism have mutually contributed to our model, they may reflect different pathological and/or compensative responses. For instance, patients with earlier age at epilepsy onset present greater hypometabolism in the ipsilateral inferior temporal gyrus, while we find no evidence of such association with hypermetabolism. In summary, quantitative models utilizing multidimensional brain metabolic information may provide additional assistance to presurgical workups in TLE.

Four methods of brain pattern analyses of fMRI signals associated with wrist extension versus wrist flexion studied for potential use in future motor learning BCI

OBJECTIVE

In stroke survivors, a treatment-resistant problem is inability to volitionally differentiate upper limb wrist extension versus flexion. When one intends to extend the wrist, the opposite occurs, wrist flexion, rendering the limb non-functional. Conventional therapeutic approaches have had limited success in achieving functional recovery of patients with chronic and severe upper extremity impairments. Functional magnetic resonance imaging (fMRI) neurofeedback is an emerging strategy that has shown potential for stroke rehabilitation. There is a lack of information regarding unique blood-oxygenation-level dependent (BOLD) cortical activations uniquely controlling execution of wrist extension versus uniquely controlling wrist flexion. Therefore, a first step in providing accurate neural feedback and training to the stroke survivor is to determine the feasibility of classifying (or differentiating) brain activity uniquely associated with wrist extension from that of wrist flexion, first in healthy adults.

APPROACH

We studied brain signal of 10 healthy adults, who performed wrist extension and wrist flexion during fMRI data acquisition. We selected four types of analyses to study the feasibility of differentiating brain signal driving wrist extension versus wrist flexion, as follows: 1) general linear model (GLM) analysis; 2) support vector machine (SVM) classification; 3) 'Winner Take All'; and 4) Relative Dominance.

RESULTS

With these four methods and our data, we found that few voxels were uniquely active during either wrist extension or wrist flexion. SVM resulted in only minimal classification accuracies. There was no significant difference in activation magnitude between wrist extension versus flexion; however, clusters of voxels showed extension signal > flexion signal and other clusters vice versa. Spatial patterns of activation differed among subjects.

SIGNIFICANCE

We encountered a number of obstacles to obtaining clear group results in healthy adults. These obstacles included the following: high variability across healthy adults in all measures studied; close proximity of uniquely active voxels to voxels that were common to both the extension and flexion movements; in general, higher magnitude of signal for the voxels common to both movements versus the magnitude of any given uniquely active voxel for one type of movement. Our results indicate that greater precision in imaging will be required to develop a truly effective method for differentiating wrist extension versus wrist flexion from fMRI data.

Combined Support Vector Machine Classifier and Brain Structural Network Features for the Individual Classification of Amnestic Mild Cognitive Impairment and Subjective Cognitive Decline Patients

OBJECTIVE

Individuals with subjective cognitive decline (SCD) or amnestic mild cognitive impairment (aMCI) represent important targets for the early detection and intervention of Alzheimer's disease (AD). In this study, we employed a multi-kernel support vector machine (SVM) to examine whether white matter (WM) structural networks can be used for screening SCD and aMCI.

METHODS

A total of 138 right-handed participants [51 normal controls (NC), 36 SCD, 51 aMCI] underwent MRI brain scans. For each participant, three types of WM networks with different edge weights were constructed with diffusion MRI data: fiber number-weighted networks, mean fractional anisotropy-weighted networks, and mean diffusivity (MD)-weighted networks. By employing a multiple-kernel SVM, we seek to integrate information from three weighted networks to improve classification performance. The accuracy of classification between each pair of groups was evaluated via leave-one-out cross-validation.

RESULTS

For the discrimination between SCD and NC, an area under the curve (AUC) value of 0.89 was obtained, with an accuracy of 83.9%. Further analysis revealed that the methods using three types of WM networks outperformed other methods using single WM network. Moreover, we found that most of discriminative features were from MD-weighted networks, which distributed among frontal lobes. Similar classification performance was also reported in the differentiation between subjects with aMCI and NCs (accuracy = 83.3%). Between SCD and aMCI, an AUC value of 0.72 was obtained, with an accuracy of 72.4%, sensitivity of 74.5% and specificity of 69.4%. The highest accuracy was achieved with features only selected from MD-weighted networks.

CONCLUSION

White matter structural network features help machine learning algorithms accurately identify individuals with SCD and aMCI from NCs. Our findings have significant implications for the development of potential brain imaging markers for the early detection of AD.

Altered Functional Brain Network in Systemic Lupus Erythematosus Patients Without Overt Neuropsychiatric Symptoms Based on Resting-State Functional Magnetic Resonance Imaging and Multivariate Pattern Analysis

Objective: This study aims to investigate the alterations in functional brain network in systemic lupus erythematosus patients without overt neuropsychiatric symptoms [neuropsychiatric systemic lupus erythematosus (non-NPSLE)] from the perspective of degree centrality (DC) and functional connectivity (FC) using resting-state functional magnetic resonance imaging (MRI) and multivariate pattern analysis (MVPA) approach.

Methods: DC analysis was performed based on the resting-state functional MRI data derived from 47 non-NPSLE patients and 47 healthy controls (HCs). Nodes with abnormal DC were utilized as seeds for further FC analysis. The correlation between MRI variables and clinical or neuropsychological data was analyzed using Pearson correlation analysis. Finally, MVPA classification based on DC was performed.

Results: When compared with the HCs, the non-NPSLE patients exhibited remarkably higher DC in the bilateral hippocampus (HIP), right insula (INS), and lower DC in the left superior parietal gyrus. Furthermore, the patients displayed significantly higher FC between the left HIP and the left INS/left dorsolateral middle frontal gyrus/left supramarginal gyrus and higher FC between the right HIP and the right middle temporal gyrus/right dorsolateral middle frontal gyrus/right dorsolateral inferior frontal gyrus/right supramarginal gyrus (all imaging variables mentioned earlier underwent cluster-level false discovery rate corrections, the voxel threshold was p < 0.001, cluster threshold was p < 0.05). Correlation analysis revealed significantly negative correlations between DC values of the right INS and disease activity and the DC values of the right HIP and the Montreal Cognitive Assessment scores. The accuracy, sensitivity, and specificity of MVPA classification based on DC were 72.34, 63.83, and 80.85%, respectively. The most discriminative power brain regions were chiefly located within the temporal, parietal, and frontal regions.

Conclusion: Patients with non-NPSLE exhibited abnormal DC and FC in the brain network. MVPA based on DC possessed commendable classification ability. Our study may provide a novel perspective on the neuropathological mechanisms underlying subclinical brain damage in non-NPSLE.

Cross-network interaction for diagnosis of major depressive disorder based on resting state functional connectivity

Previous studies have suggested that resting-state functional connectivity plays a central role in the physiopathology of major depressive disorder (MDD). However, the individualized diagnosis of MDD based on resting-state functional connectivity is still unclear, especially in first episode drug-naive patients with MDD. Resting state functional magnetic resonance imaging was enrolled from 30 first episode drug-naive patients with MDD and age- and gender-matched 31 healthy controls. Whole brain functional connectivity was computed and viewed as classification features. Multivariate pattern analysis (MVPA) was performed to discriminate patients with MDD from controls. The experimental results exhibited a correct classification rate of 82.25% (p < 0.001) with sensitivity of 83.87% and specificity of 80.64%. Almost all of the consensus connections (125/128) were cross-network interaction among default mode network (DMN), salience network (SN), central executive network (CEN), visual cortex network (VN), Cerebellum and Other. Moreover, the supramarginal gyrus exhibited high discriminative power in classification. Our findings suggested cross-network interaction can be used as an effective biomarker for MDD clinical diagnosis, which may reveal the potential pathological mechanism for major depression. The current study further confirmed reliable application of MVPA in discriminating MDD patients from healthy controls.

Progressive Brain Degeneration From Subjective Cognitive Decline to Amnestic Mild Cognitive Impairment: Evidence From Large-Scale Anatomical Connection Classification Analysis

People with subjective cognitive decline (SCD) and amnestic mild cognitive impairment (aMCI) are both at high risk for Alzheimer’s disease (AD). Behaviorally, both SCD and aMCI have subjective reports of cognitive decline, but the latter suffers a more severe objective cognitive impairment than the former. However, it remains unclear how the brain develops from SCD to aMCI. In the current study, we aimed to investigate the topological characteristics of the white matter (WM) network that can successfully identify individuals with SCD or aMCI from healthy control (HC) and to describe the relationship of pathological changes between these two stages. To this end, three groups were recruited, including 22 SCD, 22 aMCI, and 22 healthy control (HC) subjects. We constructed WM network for each subject and compared large-scale topological organization between groups at both network and nodal levels. At the network level, the combined network indexes had the best performance in discriminating aMCI from HC. However, no indexes at the network level can significantly identify SCD from HC. These results suggested that aMCI but not SCD was associated with anatomical impairments at the network level. At the nodal level, we found that the short-path length can best differentiate between aMCI and HC subjects, whereas the global efficiency has the best performance in differentiating between SCD and HC subjects, suggesting that both SCD and aMCI had significant functional integration alteration compared to HC subjects. These results converged on the idea that the neural degeneration from SCD to aMCI follows a gradual process, from abnormalities at the nodal level to those at both nodal and network levels.

Supervised Multidimensional Scaling and its Application in MRI-Based Individual Age Predictions

It has been a popular trend to decode individuals' demographic and cognitive variables based on MRI. Features extracted from MRI data are usually of high dimensionality, and dimensionality reduction (DR) is an effective way to deal with these high-dimensional features. Despite many supervised DR techniques for classification purposes, there is still a lack of supervised DR techniques for regression purposes. In this study, we advanced a novel supervised DR technique for regression purposes, namely, supervised multidimensional scaling (SMDS). The implementation of SMDS includes two steps: (1) evaluating pairwise distances among entities based on their labels and constructing a new space through a distance-preserving projection; (2) establishing an explicit linear relationship between the feature space and the new space. Based on this linear relationship, DR for test entities can be performed. We evaluated the performance of SMDS first on a synthetic dataset, and the results indicate that (1) SMDS is relatively robust to Gaussian noise existing in the features and labels; (2) the dimensionality of the new space exerts negligible influences upon SMDS; and (3) when the sample size is small, the performance of SMDS deteriorates with the increase of feature dimension. When applied to features extracted from resting state fMRI data for individual age predictions, SMDS was observed to outperform classic DR techniques, including principal component analysis, locally linear embedding and multidimensional scaling (MDS). Hopefully, SMDS can be widely used in studies on MRI-based predictions. Furthermore, novel supervised DR techniques for regression purposes can easily be developed by replacing MDS with other nonlinear DR techniques.

Classification-Biased Apparent Brain Age for the Prediction of Alzheimer's Disease

Machine Learning methods are often adopted to infer useful biomarkers for the early diagnosis of many neurodegenerative diseases and, in general, of neuroanatomical ageing. Some of these methods estimate the subject age from morphological brain data, which is then indicated as “brain age”. The difference between such a predicted brain age and the actual chronological age of a subject can be used as an indication of a pathological deviation from normal brain ageing. An important use of the brain age model as biomarker is the prediction of Alzheimer's disease (AD) from structural Magnetic Resonance Imaging (MRI). Many different machine learning approaches have been applied to this specific predictive task, some of which have achieved high accuracy at the expense of the descriptiveness of the model. This work investigates an appropriate combination of data science techniques and linear models to provide, at the same time, high accuracy and good descriptiveness. The proposed method is based on a data workflow that include typical data science methods, such as outliers detection, feature selection, linear regression, and logistic regression. In particular, a novel inductive bias is introduced in the regression model, which is aimed at improving the accuracy and the specificity of the classification task. The method is compared to other machine learning approaches for AD classification based on morphological brain data with and without the use of the brain age, including Support Vector Machines and Deep Neural Networks. This study adopts brain MRI scans of 1, 901 subjects which have been acquired from three repositories (ADNI, AIBL, and IXI). A predictive model based only on the proposed apparent brain age and the chronological age has an accuracy of 88% and 92%, respectively, for male and female subjects, in a repeated cross-validation analysis, thus achieving a comparable or superior performance than state of the art machine learning methods. The advantage of the proposed method is that it maintains the morphological semantics of the input space throughout the regression and classification tasks. The accurate predictive model is also highly descriptive and can be used to generate potentially useful insights on the predictions.

Early Detection of Alzheimer's Disease Based on Clinical Trials, Three-Dimensional Imaging Data, and Personal Information Using Autoencoders

Background:

A timely diagnosis of Alzheimer's disease (AD) is crucial to obtain more practical treatments. In this article, a novel approach using Auto-Encoder Neural Networks (AENN) for early detection of AD was proposed.

Method:

The proposed method mainly deals with the classification of multimodal data and the imputation of missing data. The data under study involve the MiniMental State Examination, magnetic resonance imaging, positron emission tomography, cerebrospinal fluid data, and personal information. Natural logarithm was used for normalizing the data. The Auto-Encoder Neural Networks was used for imputing missing data. Principal component analysis algorithm was used for reducing dimensionality of data. Support Vector Machine (SVM) was used as classifier. The proposed method was evaluated using Alzheimer's Disease Neuroimaging Initiative (ADNI) database. Then, 10fold crossvalidation was used to audit the detection accuracy of the method.

Results:

The effectiveness of the proposed approach was studied under several scenarios considering 705 cases of ADNI database. In three binary classification problems, that is AD vs. normal controls (NCs), mild cognitive impairment (MCI) vs. NC, and MCI vs. AD, we obtained the accuracies of 95.57%, 83.01%, and 78.67%, respectively.

Conclusion:

Experimental results revealed that the proposed method significantly outperformed most of the stateoftheart methods.

An interictal measurement of cerebral oxygen extraction fraction in MRI-negative refractory epilepsy using quantitative susceptibility mapping

PURPOSE

Oxygen extraction fraction (OEF) can be a factor to identify brain tissue's disability in epileptic patients. This study aimed to assess the OEF's level measurement in refractory epileptic patients (REPs) using a quantitative susceptibility mapping (QSM) method and to determine whether the OEF parameters change.

METHODS

QSM-OEF maps of 26 REPs and 16 healthy subjects were acquired using 3T MRI with a 64-channel coil. Eighteen regions-of-interest (ROIs) were chosen around the cortex in one appropriate slice of the brain and the mean QSM-OEF for each ROI was obtained. The correlations of QSM-OEF among different clinical characteristics of the disease, as well as between the patients and normal subjects, were also investigated.

RESULTS

QSM-OEF was shown to be significantly higher in REPs (44.9 ± 5.8) than that in HS (41.9 ± 6.2) (p < 0.05). Mean QSM-OEF was statistically lower in the ipsilateral side (44.5 ± 6.6) compared to the contralateral side (46.4 ± 6.8) (P < 0.01). QSM-OEF was illustrated to have a strong positive correlation with the attack duration (r = 0.6), and a moderate negative correlation with the attack frequency (r = -0.3). Using an optimized support vector machine algorithm, we could predict the disease in subjects having abnormal OEF values in the brain-selected-ROIs with sensitivity, specificity, AUC, and the precision of 0.96, 1, 0.98, and 1, respectively.

CONCLUSIONS

The results of this study revealed that QSM-OEF of the REPs' brain is higher than that of HS, which indicates that QSM-OEF is associated with disease activity.

The Microstructure of Attentional Control in the Dorsal Attention Network

The top-down control of attention involves command signals arising chiefly in the dorsal attention network (DAN) in frontal and parietal cortex and propagating to sensory cortex to enable the selective processing of incoming stimuli based on their behavioral relevance. Consistent with this view, the DAN is active during preparatory (anticipatory) attention for relevant events and objects, which, in vision, may be defined by different stimulus attributes including their spatial location, color, motion, or form. How this network is organized to support different forms of preparatory attention to different stimulus attributes remains unclear. We propose that, within the DAN, there exist functional microstructures (patterns of activity) specific for controlling attention based on the specific information to be attended. To test this, we contrasted preparatory attention to stimulus location (spatial attention) and to stimulus color (feature attention), and used multivoxel pattern analysis to characterize the corresponding patterns of activity within the DAN. We observed different multivoxel patterns of BOLD activation within the DAN for the control of spatial attention (attending left vs. right) and feature attention (attending red vs. green). These patterns of activity for spatial and feature attentional control showed limited overlap with each other within the DAN. Our findings thus support a model in which the DAN has different functional microstructures for distinctive forms of top-down control of visual attention.