Individualized diagnosis of major depressive disorder via multivariate pattern analysis of thalamic sMRI features.

BACKGROUND: Magnetic resonance imaging (MRI) studies have found thalamic abnormalities in major depressive disorder (MDD). Although there are significant differences in the structure and function of the thalamus between MDD patients and healthy controls (HCs) at the group level, it is not clear whether the structural and functional features of the thalamus are suitable for use as diagnostic prediction aids at the individual level. Here, we were to test the predictive value of gray matter density (GMD), gray matter volume (GMV), amplitude of low-frequency fluctuations (ALFF), and fractional amplitude of low-frequency fluctuations (fALFF) in the thalamus using multivariate pattern analysis (MVPA). METHODS: Seventy-four MDD patients and 44 HC subjects were recruited. The Gaussian process classifier (GPC) was trained to separate MDD patients from HCs, Gaussian process regression (GPR) was trained to predict depression scores, and Multiple Kernel Learning (MKL) was applied to explore the contribution of each subregion of the thalamus. RESULTS: The primary findings were as follows: [1] The balanced accuracy of the GPC trained with thalamic GMD was 96.59% (P < 0.001). The accuracy of the GPC trained with thalamic GMV was 93.18% (P < 0.001). The correlation between Hamilton Depression Scale (HAMD) score targets and predictions in the GPR trained with GMD was 0.90 (P < 0.001, r2 = 0.82), and in the GPR trained with GMV, the correlation between HAMD score targets and predictions was 0.89 (P < 0.001, r2 = 0.79). [2] The models trained with ALFF and fALFF in the thalamus failed to discriminate MDD patients from HC participants. [3] The MKL model showed that the left lateral prefrontal thalamus, the right caudal temporal thalamus, and the right sensory thalamus contribute more to the diagnostic classification. CONCLUSIONS: The results suggested that GMD and GMV, but not functional indicators of the thalamus, have good potential for the individualized diagnosis of MDD. Furthermore, the thalamus shows the heterogeneity in the structural features of thalamic subregions for predicting MDD. To our knowledge, this is the first study to focus on the thalamus for the prediction of MDD using machine learning methods at the individual level.

Neurophysiological effects of cognitive behavioral therapy in social anxiety: An ERP study using a dot-probe task.

BACKGROUND: Social anxious individuals show attention bias towards emotional stimuli, this phenomenon is considered to be an important cause of anxiety generation and maintenance. Cognitive-behavioral therapy (CBT) is a standard psychotherapy for social anxiety disorder. CBT decreases attention biases by correcting the maladaptive beliefs of socially anxious individuals, but it is not clear whether CBT alters neurophysiological features of socially anxious individuals at early automatic and/or late cognitive strategy stage of attentional processing. METHOD: To address this knowledge gap, we collected pre-treatment event-related potential data of 22 socially anxious individuals while they performed a dot-probe task. These participants then received eight weeks of CBT, and post-treatment ERP data were collected after completion of CBT treatment. We also included 29 healthy controls and compared them with individuals with social anxiety to determine the neural mechanisms underlying the effectiveness of CBT. RESULTS: Participants' social anxiety level was significantly alleviated with CBT. ERP results revealed that (1) compared to pre-treatment phase, P1 amplitudes induced by probes significantly decreased at post-treatment phase, whereas P3 amplitudes increased at post-treatment phase; the P1 amplitudes induced by probes following happy-neutral face pairs in socially anxious individuals after treatment was significantly different with that in healthy controls; (2) amplitude of components elicited by face pairs did not change significantly between pre-treatment and post-treatment phases; (3) changes of Liebowitz Social Anxiety Scale were positively correlated with changes of P1 amplitude, and negatively correlated with changes of N1 amplitude. LIMITATIONS: Our sample was university students and lacked randomization, which limits the generalizability of the results. CONCLUSION: The present results demonstrated that CBT may adjust cognitive strategies in the later stage of attentional processing, indicating by changed ERPs appeared in probe-presenting stage for social anxiety.

Event-related alpha power in early stage of facial expression processing in social anxiety: Influence of language context.

Accurate interpretation of the emotional information conveyed by others' facial expressions is crucial for social interactions. Event-related alpha power, measured by time-frequency analysis, is a frequently used EEG index of emotional information processing. However, it is still unclear how event-related alpha power varies in emotional information processing in social anxiety groups. In the present study, we recorded event-related potentials (ERPs) while participants from the social anxiety and healthy control groups viewed facial expressions (angry, happy, neutral) preceded by contextual sentences conveying either a positive or negative evaluation of the subject. The impact of context on facial expression processing in both groups of participants was explored by assessing behavioral ratings and event-related alpha power (0-200 ms after expression presentation). In comparison to the healthy control group, the social anxiety group exhibited significantly lower occipital alpha power in response to angry facial expressions in negative contexts and neutral facial expressions in positive contexts. The influence of language context on facial expression processing in individuals with social anxiety may occur at an early stage of processing.

Social anxiety prediction based on ERP features: A deep learning approach.

BACKGROUND: Social Anxiety Disorder is traditionally diagnosed using subjective scales that may lack accuracy. Recently, EEG technology has gained importance for anxiety detection due to its ability to capture stable and objective neurophysiological activities. However, existing methods mainly focus on extracting EEG features during resting states, with limited use of psychologically features like Event-Related Potential (ERP) in task-related states for anxiety detection in deep learning frameworks. METHODS: We collected EEG data from 63 participants exposed to four facial expressions and extracted task-relevant features. Using the EEGNet model, we predicted social anxiety and evaluated its performance using metrics such as accuracy, F1 score, sensitivity, and specificity. We compared EEGNet's performance with Deep Convolutional Neural Network (DeepConvNet), ShallowConvNet, Bi-directional Long Short-Term Memory (BiLSTM), and SVM. To assess the generalizability of the results, we carried out the same procedure on our prior dataset. RESULTS: EEGNet outperformed other models, achieving 99.16 % accuracy with Late Positive Potential (LPP). ERP components yielded higher accuracy than time-domain and frequency-domain features for social anxiety recognition. Accuracy was better for neutral and negative facial stimuli. Consistency across two datasets indicates stability of findings. LIMITATIONS: Due to limited publicly available task-state datasets, only our own were used. Future studies could assess generalizability on larger datasets from different sources. CONCLUSIONS: We conducted the first test of ERP features in anxiety recognition tasks. Results show ERP features have greater potential in social anxiety recognition, with LPP exhibiting high stability and accuracy. Outcomes indicate recognizing social anxiety with negative or neutral facial stimuli is more useful.

Early Influence of Emotional Scenes on the Encoding of Fearful Expressions With Different Intensities: An Event-Related Potential Study.

Contextual affective information influences the processing of facial expressions at the relatively early stages of face processing, but the effect of the context on the processing of facial expressions with varying intensities remains unclear. In this study, we investigated the influence of emotional scenes (fearful, happy, and neutral) on the processing of fear expressions at different levels of intensity (high, medium, and low) during the early stages of facial recognition using event-related potential (ERP) technology. EEG data were collected while participants performed a fearful facial expression recognition task. The results showed that (1) the recognition of high-intensity fear expression was higher than that of medium- and low-intensity fear expressions. Facial expression recognition was the highest when faces appeared in fearful scenes. (2) Emotional scenes modulated the amplitudes of N170 for fear expressions with different intensities. Specifically, the N170 amplitude, induced by high-intensity fear expressions, was significantly higher than that induced by low-intensity fear expressions when faces appeared in both neutral and fearful scenes. No significant differences were found between the N170 amplitudes induced by high-, medium-, and low-intensity fear expressions when faces appeared in happy scenes. These results suggest that individuals may tend to allocate their attention resources to the processing of face information when the valence between emotional context and expression conflicts i.e., when the conflict is absent (fear scene and fearful faces) or is low (neutral scene and fearful faces).

Decoding Color Visual Working Memory from EEG Signals Using Graph Convolutional Neural Networks.

Color has an important role in object recognition and visual working memory (VWM). Decoding color VWM in the human brain is helpful to understand the mechanism of visual cognitive process and evaluate memory ability. Recently, several studies showed that color could be decoded from scalp electroencephalogram (EEG) signals during the encoding stage of VWM, which process visible information with strong neural coding. Whether color could be decoded from other VWM processing stages, especially the maintaining stage which processes invisible information, is still unknown. Here, we constructed an EEG color graph convolutional network model (ECo-GCN) to decode colors during different VWM stages. Based on graph convolutional networks, ECo-GCN considers the graph structure of EEG signals and may be more efficient in color decoding. We found that (1) decoding accuracies for colors during the encoding, early, and late maintaining stages were 81.58%, 79.36%, and 77.06%, respectively, exceeding those during the pre-stimuli stage (67.34%), and (2) the decoding accuracy during maintaining stage could predict participants' memory performance. The results suggest that EEG signals during the maintaining stage may be more sensitive than behavioral measurement to predict the VWM performance of human, and ECo-GCN provides an effective approach to explore human cognitive function.

Positive attention bias in high socially anxious individuals: Evidence from an ERP study.

The Bivalent Fear of Evaluation (BFEO) model posits that the fear of positive evaluation (FPE) is a core feature of social anxiety. As such, high socially anxious individuals may show attention bias when faced with positive stimuli. However, most of the previous studies focused on the negative attention bias of social anxiety, and less on the attention bias of positive stimuli. Meanwhile, the effect of stimulus presentation time on the attention bias pattern was unclear. In order to investigate this question, we used a dot-probe paradigm with facial expressions (happy, fearful, angry, neutral) presented for 100 ms and 500 ms. The ERP results showed: (1) For high socially anxious group, happy faces elicited a larger N1 for valid than for invalid cued probes, whereas for healthy control group, angry faces elicited a larger N1 for valid than for invalid cued probes. (2) When valid cues following happy faces presented for 500 ms, the N1 amplitude was larger than that of invalid cues. However, when valid cues following angry and fear faces presented for 100 ms, the N1 amplitude was larger than that of invalid cues. The results showed difficulty in attention disengagement of high socially anxious individuals from positive stimuli, as reflected by N1, illustrating the positive attention bias in social anxiety. These results prove that FPE may contribute to maintaining social anxiety.

Influence of affective verbal context on emotional facial expression perception of social anxiety.

Previous studies have shown that the perception of ambiguous facial expressions for individuals with social anxiety was influenced by the affective verbal context. However, it is still unknown how emotional facial expressions are perceived by individuals with social anxiety in the context of the verbal context. In this study, we used event-related potentials (ERPs) technology to examine how individuals with social anxiety perceive emotional facial expressions in positive and negative contexts. The results showed that: (1) Within the negative verbal contexts, the amplitude of P1 induced by facial expressions in the social anxiety group was significantly higher than that induced by the healthy control group; The N170 amplitude induced by facial expressions in social anxiety group was less negative than that in the healthy control group, and was not affected by the context. (2) The social anxiety group had significantly higher LPP in negative contexts elicited by angry expressions than by happy expressions. This study proved that the perception of emotional facial expressions was influenced by top-down information in the early and late stages of visual perception for individuals with social anxiety.

Visuospatial Bias in Children with Autism Spectrum Disorder: Evidence from Line Bisection Tasks.

Previous studies have found reduced leftward bias of facial processing in individuals with Autism Spectrum Disorder (ASD). However, it is not clear whether they manifest a leftward bias in general visual processing. To shed light on this issue, the current study used the manual line bisection task to assess children 5 to 15 years of age with ASD as well as typically developing (TD) children. Results showed that children with ASD, similar to TD children, demonstrate a leftward bias in general visual processing, especially for bisecting long lines (≧ 80 mm). In both groups, participant performance in line bisection was affected by the hand used, the length of the line, the cueing symbol, and the location of the symbol. The ASD group showed a rightward bias when bisecting short lines (30 mm) with their left hands, which slightly differed from the TD group. These results indicate that while ASD individuals and TD individuals share a similar leftward bias in general visual processing, when using their left hands to bisect short lines, ASD individuals may show an atypical bias pattern.

Treatment Response Prediction for Major Depressive Disorder Patients via Multivariate Pattern Analysis of Thalamic Features.

Antidepressant treatment, as an important method in clinical practice, is not suitable for all major depressive disorder (MDD) patients. Although magnetic resonance imaging (MRI) studies have found thalamic abnormalities in MDD patients, it is not clear whether the features of the thalamus are suitable to serve as predictive aids for treatment responses at the individual level. Here, we tested the predictive value of gray matter density (GMD), gray matter volume (GMV), amplitude of low-frequency fluctuations (ALFF), and fractional ALFF (fALFF) of the thalamus using multivariate pattern analysis (MVPA). A total of 74 MDD patients and 44 healthy control (HC) subjects were recruited. Thirty-nine MDD patients and 35 HC subjects underwent scanning twice. Between the two scanning sessions, patients in the MDD group received selective serotonin reuptake inhibitor (SSRI) treatment for 3-month, and HC group did not receive any treatment. Gaussian process regression (GPR) was trained to predict the percentage decrease in the Hamilton Depression Scale (HAMD) score after treatment. The percentage decrease in HAMD score after SSRI treatment was predicted by building GPRs trained with baseline thalamic data. The results showed significant correlations between the true percentage of HAMD score decreases and predictions (p < 0.01, r 2 = 0.11) in GPRs trained with GMD. We did not find significant correlations between the true percentage of HAMD score decreases and predictions in GMV (p = 0.16, r 2 = 0.00), ALFF (p = 0.125, r 2 = 0.00), and fALFF (p = 0.485, r 2 = 0.10). Our results suggest that GMD of the thalamus has good potential as an aid in individualized treatment response predictions of MDD patients.

Multi-Dimensional Enhanced Seizure Prediction Framework Based on Graph Convolutional Network.

In terms of seizure prediction, how to fully mine relational data information among multiple channels of epileptic EEG? This is a scientific research subject worthy of further exploration. Recently, we propose a multi-dimensional enhanced seizure prediction framework, which mainly includes information reconstruction space, graph state encoder, and space-time predictor. It takes multi-channel spatial relationship as breakthrough point. At the same time, it reconstructs data unit from frequency band level, updates graph coding representation, and explores space-time relationship. Through experiments on CHB-MIT dataset, sensitivity of the model reaches 98.61%, which proves effectiveness of the proposed model.

The influence of mouth opening and closing degrees on processing in NimStim facial expressions: An ERP study from Chinese college students.

The degree of mouth opening and closing is one of the most important attributes of expression, reflecting the intensity of facial expression and can assist people to recognize the expression more accurately. The NimStim set of facial expressions contains the open and closed expression pictures of the same actor. Although this expression set has been widely used, there is little research on the intensity effect of this set. In this study, 32 Chinese college students were recruited in to view the pictures passively in an ERP experiment, aiming to investigate the intensity effect in the NimStim set (mouth open, mouth closed) of anger, disgust, sad, happy and neutral expression in electrical physiological aspects of the reaction. Our results reported that intensity of expression early affected in VPP and mainly affected in LPP with the open mouth having a larger activity. And there was no intensity effect found in P1, N170 and EPN. Notably, culture and social environment may influence the intensity effect of different emotions. In future, researchers should use methods that ensure subjects pay more attention to the intensity effect of the NimStim facial set.

Decoding disparity categories in 3-dimensional images from fMRI data using functional connectivity patterns.

Humans use binocular disparity to extract depth information from two-dimensional retinal images in a process called stereopsis. Previous studies usually introduce the standard univariate analysis to describe the correlation between disparity level and brain activity within a given brain region based on functional magnetic resonance imaging (fMRI) data. Recently, multivariate pattern analysis has been developed to extract activity patterns across multiple voxels for deciphering categories of binocular disparity. However, the functional connectivity (FC) of patterns based on regions of interest or voxels and their mapping onto disparity category perception remain unknown. The present study extracted functional connectivity patterns for three disparity conditions (crossed disparity, uncrossed disparity, and zero disparity) at distinct spatial scales to decode the binocular disparity. Results of 27 subjects' fMRI data demonstrate that FC features are more discriminatory than traditional voxel activity features in binocular disparity classification. The average binary classification of the whole brain and visual areas are respectively 87% and 79% at single subject level, and thus above the chance level (50%). Our research highlights the importance of exploring functional connectivity patterns to achieve a novel understanding of 3D image processing.

Euler Elastica Regularized Logistic Regression for Whole-Brain Decoding of fMRI Data.

OBJECTIVE: Multivariate pattern analysis methods have been widely applied to functional magnetic resonance imaging (fMRI) data to decode brain states. Due to the "high features, low samples" in fMRI data, machine learning methods have been widely regularized using various regularizations to avoid overfitting. Both total variation (TV) using the gradients of images and Euler's elastica (EE) using the gradient and the curvature of images are the two popular regulations with spatial structures. In contrast to TV, EE regulation is able to overcome the disadvantage of TV regulation that favored piecewise constant images over piecewise smooth images. In this study, we introduced EE to fMRI-based decoding for the first time and proposed the EE regularized multinomial logistic regression (EELR) algorithm for multi-class classification. METHODS: We performed experimental tests on both simulated and real fMRI data to investigate the feasibility and robustness of EELR. The performance of EELR was compared with sparse logistic regression (SLR) and TV regularized LR (TVLR). RESULTS: The results showed that EELR was more robustness to noises and showed significantly higher classification performance than TVLR and SLR. Moreover, the forward models and weights patterns revealed that EELR detected larger brain regions that were discriminative to each task and activated by each task than TVLR. CONCLUSION: The results suggest that EELR not only performs well in brain decoding but also reveals meaningful discriminative and activation patterns. SIGNIFICANCE: This study demonstrated that EELR showed promising potential in brain decoding and discriminative/activation pattern detection.

Image categorization from functional magnetic resonance imaging using functional connectivity.

BACKGROUND: Previous studies have attempted to infer the category of objects in a stimulus image from functional magnetic resonance imaging (fMRI) data recoded during image-viewing. Most studies focus on extracting activity patterns within a given region or across multiple voxels, and utilize the relationships among voxels to decipher the category of a stimulus image. Yet, the functional connectivity (FC) patterns across regions of interest in response to image categories, and their potential contributions to category classification are largely unknown. NEW METHOD: We investigated whole-brain FC patterns in response to 4 image category stimuli (cats, faces, houses, and vehicles) using fMRI in healthy adult volunteers, and classified FC patterns using machine learning framework (Support Vector Machine [SVM] and Random Forest). We further examined the FC robustness and the influence of the window length on FC patterns for neural decoding. RESULTS: The average one-vs.-one classification accuracy of the two classification models were 74% within subjects and 80% between subjects, which are higher than the chance level (50%). The Random Forest results were better than SVM results, and the 48-s FC results were better than the 24-s FC results. COMPARISON WITH EXISTING METHOD(S): We compared the classification performance of our FC patterns with two other existing methods, inter-block and intra-block, without overlapping temporal information. CONCLUSIONS: Whole-brain FC patterns for different window lengths (24 and 48 s) can predict images categories with high accuracy. These results reveal novel mechanisms underlying the representation of categorical information in large-scale FC patterns in the human brain.

An empirical comparison of different LDA methods in fMRI-based brain states decoding.

Decoding brain states from response patterns with multivariate pattern recognition techniques is a popular method for detecting multivoxel patterns of brain activation. These patterns are informative with respect to a subject's perceptual or cognitive states. Linear discriminant analysis (LDA) cannot be directly applied to fMRI data analysis because of the "few samples and large features" nature of functional magnetic resonance imaging (fMRI) data. Although several improved LDA methods have been used in fMRI-based decoding, little is known regarding the relative performance of different LDA classifiers on fMRI data. In this study, we compared five LDA classifiers using both simulated data with varied noise levels and real fMRI data. The compared LDA classifiers include LDA combined with PCA (LDA-PCA), LDA with three types of regularizations (identity matrix, diagonal matrix and scaled identity matrix) and LDA with optimal-shrinkage covariance estimator using Ledoit and Wolf lemma (LDA-LW). The results indicated that LDA-LW was the most robust to noises. Moreover, LDA-LW and LDA with scaled identity matrix showed better stability and classification accuracy than the other methods. LDA-LW demonstrated the best overall performance.

Improved sparse decomposition based on a smoothed L0 norm using a Laplacian kernel to select features from fMRI data.

BACKGROUND: Feature selection plays an important role in improving the classification accuracy of multivariate classification techniques in the context of fMRI-based decoding due to the "few samples and large features" nature of functional magnetic resonance imaging (fMRI) data. Recently, several sparse representation methods have been applied to the voxel selection of fMRI data. Despite the low computational efficiency of the sparse representation methods, they still displayed promise for applications that select features from fMRI data. NEW METHOD: In this study, we proposed the Laplacian smoothed L0 norm (LSL0) approach for feature selection of fMRI data. Based on the fast sparse decomposition using smoothed L0 norm (SL0) (Mohimani, 2007), the LSL0 method used the Laplacian function to approximate the L0 norm of sources. RESULTS: Results of the simulated and real fMRI data demonstrated the feasibility and robustness of LSL0 for the sparse source estimation and feature selection. COMPARISON WITH EXISTING METHODS: Simulated results indicated that LSL0 produced more accurate source estimation than SL0 at high noise levels. The classification accuracy using voxels that were selected by LSL0 was higher than that by SL0 in both simulated and real fMRI experiment. Moreover, both LSL0 and SL0 showed higher classification accuracy and required less time than ICA and t-test for the fMRI decoding. CONCLUSIONS: LSL0 outperformed SL0 in sparse source estimation at high noise level and in feature selection. Moreover, LSL0 and SL0 showed better performance than ICA and t-test for feature selection.

Causal interaction following the alteration of target region activation during motor imagery training using real-time fMRI.

Motor imagery training is an effective approach for motor skill learning and motor function rehabilitation. As a novel method of motor imagery training, real-time fMRI (rtfMRI) enables individuals to acquire self-control of localized brain activation, achieving desired changes in behavior. The regulation of target region activation by rtfMRI often alters the activation of related brain regions. However, the interaction between the target region and these related regions is unclear. The Granger causality model (GCM) is a data-driven method that can explore the causal interaction between brain regions. In this study, we employed rtfMRI to train subjects to regulate the activation of the ipsilateral dorsal premotor area (dPMA) during motor imagery training, and we calculated the causal interaction of the dPMA with other motor-related regions based on the GCM. The results demonstrated that as the activity of the dPMA changed during rtfMRI training, the interaction of the target region with other related regions became significantly altered, and behavioral performance was improved after training. The altered interaction primarily exhibited as an increased unidirectional interaction from the dPMA to the other regions. These findings support the dominant role of the dPMA in motor skill learning via rtfMRI training and may indicate how activation of the target region interacts with the activation of other related regions.

Bayesian reconstruction of multiscale local contrast images from brain activity.

BACKGROUND: Recent advances in functional magnetic resonance imaging (fMRI) techniques make it possible to reconstruct contrast-defined visual images from brain activity. In this manner, the stimulus images are represented as the weighted sum of a set of element images with different scales. The contrast weight of local images were decoded using fMRI activity recorded when the subject was viewing the stimulus images. Multivariate methods, such as the sparse multinomial logistic regression model (SMLR), have been proven effective for learning the mapping between fMRI patterns of primary visual cortex voxels and contrast of stimulus images. However, the SMLR method is highly time-consuming in practical application. NEW METHOD: The Naive Bayesian classifier based on independent component analysis (NB-ICA) is proposed to efficiently decode the contrast of multi-scale local images. First, temporal independent components of fMRI data which were treated as new features for NB classifier were acquired by ICA decomposition. Second, the contrast for each local element image was computed based on NB estimation theory. RESULTS: NB-ICA method can be used to reconstruct novel visual images. The average spatial correlation between the represented and reconstructed images was 0.41 ± 0.13 (p<0.001). COMPARISON WITH EXISTING METHOD(S): At the expense of reconstruction accuracy, NB-ICA is more efficient than SMLR which reduces the computation time from hours to seconds. CONCLUSIONS: A new method, termed NB-ICA, is proposed and can efficiently reconstruct contrast-defined visual images from fMRI data. This study provides theoretical support for brain-computer interface research and also provides ideas for the study of real-time fMRI data.

Comparative study of SVM methods combined with voxel selection for object category classification on fMRI data.

BACKGROUND: Support vector machine (SVM) has been widely used as accurate and reliable method to decipher brain patterns from functional MRI (fMRI) data. Previous studies have not found a clear benefit for non-linear (polynomial kernel) SVM versus linear one. Here, a more effective non-linear SVM using radial basis function (RBF) kernel is compared with linear SVM. Different from traditional studies which focused either merely on the evaluation of different types of SVM or the voxel selection methods, we aimed to investigate the overall performance of linear and RBF SVM for fMRI classification together with voxel selection schemes on classification accuracy and time-consuming. METHODOLOGY/PRINCIPAL FINDINGS: Six different voxel selection methods were employed to decide which voxels of fMRI data would be included in SVM classifiers with linear and RBF kernels in classifying 4-category objects. Then the overall performances of voxel selection and classification methods were compared. Results showed that: (1) Voxel selection had an important impact on the classification accuracy of the classifiers: in a relative low dimensional feature space, RBF SVM outperformed linear SVM significantly; in a relative high dimensional space, linear SVM performed better than its counterpart; (2) Considering the classification accuracy and time-consuming holistically, linear SVM with relative more voxels as features and RBF SVM with small set of voxels (after PCA) could achieve the better accuracy and cost shorter time. CONCLUSIONS/SIGNIFICANCE: The present work provides the first empirical result of linear and RBF SVM in classification of fMRI data, combined with voxel selection methods. Based on the findings, if only classification accuracy was concerned, RBF SVM with appropriate small voxels and linear SVM with relative more voxels were two suggested solutions; if users concerned more about the computational time, RBF SVM with relative small set of voxels when part of the principal components were kept as features was a better choice.