Abbas Babajani-Feremi

Functional connectivity changes detected with magnetoencephalography after mild traumatic brain injury.

Mild traumatic brain injury (mTBI) may affect normal cognition and behavior by disrupting the functional connectivity networks that mediate efficient communication among brain regions. In this study, we analyzed brain connectivity profiles from resting state Magnetoencephalographic (MEG) recordings obtained from 31 mTBI patients and 55 normal controls. We used phase-locking value estimates to compute functional connectivity graphs to quantify frequency-specific couplings between sensors at various frequency bands. Overall, normal controls showed a dense network of strong local connections and a limited number of long-range connections that accounted for approximately 20% of all connections, whereas mTBI patients showed networks characterized by weak local connections and strong long-range connections that accounted for more than 60% of all connections. Comparison of the two distinct general patterns at different frequencies using a tensor representation for the connectivity graphs and tensor subspace analysis for optimal feature extraction showed that mTBI patients could be separated from normal controls with 100% classification accuracy in the alpha band. These encouraging findings support the hypothesis that MEG-based functional connectivity patterns may be used as biomarkers that can provide more accurate diagnoses, help guide treatment, and monitor effectiveness of intervention in mTBI.

Classification of patients with MCI and AD from healthy controls using directed graph measures of resting-state fMRI

HIGHLIGHTSWe used directed graph measures to identify alteration of brain network in MCI and AD.We achieved an accuracy of 93.3% for classification of AD, MCI, and control subjects.AD Patients may experience disappearing some hub regions during disease progression.Directed graph measures of rs‐fMRI data can be used to identify the early stage of AD. ABSTRACT Brain network alterations in patients with Alzheimers disease (AD) has been the subject of much investigation, but the biological mechanisms underlying these alterations remain poorly understood. Here, we aim to identify the changes in brain networks in patients with AD and mild cognitive impairment (MCI), and provide an accurate algorithm for classification of these patients from healthy control subjects (HC) by using a graph theoretical approach and advanced machine learning methods. Multivariate Granger causality analysis was performed on resting‐state functional magnetic resonance imaging (rs‐fMRI) data of 34 AD, 89 MCI, and 45 HC to calculate various directed graph measures. The graph measures were used as the original feature set for the machine learning algorithm. Filter and wrapper feature selection methods were applied to the original feature set to select an optimal subset of features. An accuracy of 93.3% was achieved for classification of AD, MCI, and HC using the optimal features and the naïve Bayes classifier. We also performed a hub node analysis and found that the number of hubs in HC, MCI, and AD were 12, 10, and 9, respectively, suggesting that patients with AD experience disturbance of critical communication areas in their brain network as AD progresses. The findings of this study provide insight into the neurophysiological mechanisms underlying MCI and AD. The proposed classification method highlights the potential of directed graph measures of rs‐fMRI data for identification of the early stage of AD.

Breathing as a Fundamental Rhythm of Brain Function

Ongoing fluctuations of neuronal activity have long been considered intrinsic noise that introduces unavoidable and unwanted variability into neuronal processing, which the brain eliminates by averaging across population activity (Georgopoulos et al., 1986; Lee et al., 1988; Shadlen and Newsome, 1994; Maynard et al., 1999). It is now understood, that the seemingly random fluctuations of cortical activity form highly structured patterns, including oscillations at various frequencies, that modulate evoked neuronal responses (Arieli et al., 1996; Poulet and Petersen, 2008; He, 2013) and affect sensory perception (Linkenkaer-Hansen et al., 2004; Boly et al., 2007; Sadaghiani et al., 2009; Vinnik et al., 2012; Palva et al., 2013). Ongoing cortical activity is driven by proprioceptive and interoceptive inputs. In addition, it is partially intrinsically generated in which case it may be related to mental processes (Fox and Raichle, 2007; Deco et al., 2011). Here we argue that respiration, via multiple sensory pathways, contributes a rhythmic component to the ongoing cortical activity. We suggest that this rhythmic activity modulates the temporal organization of cortical neurodynamics, thereby linking higher cortical functions to the process of breathing.

Predicting conversion from MCI to AD using resting-state fMRI, graph theoretical approach and SVM

BACKGROUND We investigated identifying patients with mild cognitive impairment (MCI) who progress to Alzheimers disease (AD), MCI converter (MCI-C), from those with MCI who do not progress to AD, MCI non-converter (MCI-NC), based on resting-state fMRI (rs-fMRI). NEW METHOD Graph theory and machine learning approach were utilized to predict progress of patients with MCI to AD using rs-fMRI. Eighteen MCI converts (average age 73.6 years; 11 male) and 62 age-matched MCI non-converters (average age 73.0 years, 28 male) were included in this study. We trained and tested a support vector machine (SVM) to classify MCI-C from MCI-NC using features constructed based on the local and global graph measures. A novel feature selection algorithm was developed and utilized to select an optimal subset of features. RESULTS Using subset of optimal features in SVM, we classified MCI-C from MCI-NC with an accuracy, sensitivity, specificity, and the area under the receiver operating characteristic (ROC) curve of 91.4%, 83.24%, 90.1%, and 0.95, respectively. Furthermore, results of our statistical analyses were used to identify the affected brain regions in AD. COMPARISON WITH EXISTING METHOD(S) To the best of our knowledge, this is the first study that combines the graph measures (constructed based on rs-fMRI) with machine learning approach and accurately classify MCI-C from MCI-NC. CONCLUSION Results of this study demonstrate potential of the proposed approach for early AD diagnosis and demonstrate capability of rs-fMRI to predict conversion from MCI to AD by identifying affected brain regions underlying this conversion.

Comparison of brain network models using cross-frequency coupling and attack strategies

Several neuroimaging studies have suggested that functional brain connectivity networks exhibit “small-world” characteristics, whereas recent studies based on structural data have proposed a “rich-club” organization of brain networks, whereby hubs of high connection density tend to connect among themselves compared to nodes of lower density. In this study, we adopted an “attack strategy” to compare the rich-club and small-world organizations and identify the model that describes best the topology of brain connectivity. We hypothesized that the highest reduction in global efficiency caused by a targeted attack on each models hubs would reveal the organization that better describes the topology of the underlying brain networks. We applied this approach to magnetoencephalographic data obtained at rest from neurologically intact controls and mild traumatic brain injury patients. Functional connectivity networks were computed using phase-to-amplitude cross-frequency coupling between the δ and β frequency bands. Our results suggest that resting state MEG connectivity networks follow a rich-club organization.

Identifying seizure onset zone from electrocorticographic recordings: A machine learning approach based on phase locking value

PURPOSE Using a novel technique based on phase locking value (PLV), we investigated the potential for features extracted from electrocorticographic (ECoG) recordings to serve as biomarkers to identify the seizure onset zone (SOZ). METHODS We computed the PLV between the phase of the amplitude of high gamma activity (80-150Hz) and the phase of lower frequency rhythms (4-30Hz) from ECoG recordings obtained from 10 patients with epilepsy (21 seizures). We extracted five features from the PLV and used a machine learning approach based on logistic regression to build a model that classifies electrodes as SOZ or non-SOZ. RESULTS More than 96% of electrodes identified as the SOZ by our algorithm were within the resected area in six seizure-free patients. In four non-seizure-free patients, more than 31% of the identified SOZ electrodes by our algorithm were outside the resected area. In addition, we observed that the seizure outcome in non-seizure-free patients correlated with the number of non-resected SOZ electrodes identified by our algorithm. CONCLUSION This machine learning approach, based on features extracted from the PLV, effectively identified electrodes within the SOZ. The approach has the potential to assist clinicians in surgical decision-making when pre-surgical intracranial recordings are utilized.

The Role of the Primary Sensory Cortices in Early Language Processing

The results of this magnetoencephalography study challenge two long-standing assumptions regarding the brain mechanisms of language processing: First, that linguistic processing proper follows sensory feature processing effected by bilateral activation of the primary sensory cortices that lasts about 100 msec from stimulus onset. Second, that subsequent linguistic processing is effected by left hemisphere networks outside the primary sensory areas, including Brocas and Wernickes association cortices. Here we present evidence that linguistic analysis begins almost synchronously with sensory, prelinguistic verbal input analysis and that the primary cortices are also engaged in these linguistic analyses and become, consequently, part of the left hemisphere language network during language tasks. These findings call for extensive revision of our conception of linguistic processing in the brain.

Automatic classification of Alzheimer's disease with resting-state fMRI and graph theory

Study of brain network on the basis of resting-state functional magnetic resonance imaging (fMRI) has provided promising results to investigate changes in connectivity among different brain regions because of diseases. In this study, we combine graph theoretical approaches with advanced machine learning methods to study functional brain network alteration in patients with Alzheimers disease (AD). Support vector machine (SVM) was used to explore the ability of graph measures in diagnosis of AD. We applied our method on the resting-state fMRI data of twenty patients with AD and twenty age and gender matched healthy subjects. After preprocessing of data, signals from 90 brain regions, segmented based on the automated anatomical labeling (AAL) atlas, were extracted and edges of the graph were calculated using the correlation between the signals of all pairs of the brain regions. Then a weighted undirected graph was constructed and graph measures were calculated. Fisher score feature selection algorithm were employed to choose most significant features. Finally, using the selected features, we were able to accurately classify patients with AD from healthy control ones with accuracy of 97.5%. Results of this study show that pattern recognition and graph of brain network, on the basis of the resting state fMRI data, can efficiently assist in the diagnosis of AD.

Predicting postoperative language outcome using presurgical fMRI, MEG, TMS, and high gamma ECoG

OBJECTIVE To predict the postoperative language outcome using the support vector regression (SVR) and results of multimodal presurgical language mapping. METHODS Eleven patients with epilepsy received presurgical language mapping using functional MRI (fMRI), magnetoencephalography (MEG), transcranial magnetic stimulation (TMS), and high-gamma electrocorticography (hgECoG), as well as pre- and postoperative neuropsychological evaluation of language. We constructed 15 (24-1) SVR models by considering the extent of resected language areas identified by all subsets of four modalities as input feature vector and the postoperative language outcome as output. We trained and cross-validated SVR models, and compared the cross-validation (CV) errors of all models for prediction of language outcome. RESULTS Seven patients had some level of postoperative language decline and two of them had significant postoperative decline in naming. Some parts of language areas identified by four modalities were resected in these patients. We found that an SVR model consisting of fMRI, MEG, and hgECoG provided minimum CV error, although an SVR model consisting of fMRI and MEG was the optimal model that facilitated the best trade-off between model complexity and prediction accuracy. CONCLUSIONS A multimodal SVR can be used to predict the language outcome. SIGNIFICANCE The developed multimodal SVR models in this study can be utilized to calculate the language outcomes of different resection plans prior to surgery and select the optimal surgical plan.

Predicting seizure outcome of vagus nerve stimulation using MEG-based network topology

Vagus nerve stimulation (VNS) is a low-risk surgical option for patients with drug resistant epilepsy, although it is impossible to predict which patients may respond to VNS treatment. Resting-state magnetoencephalography (rs-MEG) connectivity analysis has been increasingly utilized to investigate the impact of epilepsy on brain networks and identify alteration of these networks after different treatments; however, there is no study to date utilizing this modality to predict the efficacy of VNS treatment. We investigated whether the rs-MEG network topology before VNS implantation can be used to predict efficacy of VNS treatment. Twenty-three patients with epilepsy who had MEG before VNS implantation were included in this study. We also included 89 healthy control subjects from the Human Connectome Project. Using the phase-locking value in the theta, alpha, and beta frequency bands as a measure of rs-MEG functional connectivity, we calculated three global graph measures: modularity, transitivity, and characteristic path length (CPL). Our results revealed that the rs-MEG graph measures were significantly heritable and had an overall good test-retest reliability, and thus these measures may be used as potential biomarkers of the network topology. We found that the modularity and transitivity in VNS responders were significantly larger and smaller, respectively, than those observed in VNS non-responders. We also observed that the modularity and transitivity in three frequency bands and CPL in delta and beta bands were significantly different in controls than those found in responders or non-responders, although the values of the graph measures in controls were closer to those of responders than non-responders. We used the modularity and transitivity as input features of a naïve Bayes classifier, and achieved an accuracy of 87% in classification of non-responders, responders, and controls. The results of this study revealed that MEG-based graph measures are reliable biomarkers, and that these measures may be used to predict seizure outcome of VNS treatment