Diagnostic deep learning algorithms that use resting EEG to distinguish major depressive disorder, bipolar disorder, and schizophrenia from each other and from healthy volunteers.

BACKGROUND: Mood disorders and schizophrenia affect millions worldwide. Currently, diagnosis is primarily determined by reported symptomatology. As symptoms may overlap, misdiagnosis is common, potentially leading to ineffective or destabilizing treatment. Diagnostic biomarkers could significantly improve clinical care by reducing dependence on symptomatic presentation. METHODS: We used deep learning analysis (DLA) of resting electroencephalograph (EEG) to differentiate healthy control (HC) subjects (N = 239), from those with major depressive disorder (MDD) (N = 105), MDD-atypical (MDD-A) (N = 27), MDD-psychotic (MDD-P) (N = 35), bipolar disorder-depressed episode (BD-DE) (N = 71), BD-manic episode (BD-ME) (N = 49), and schizophrenia (SCZ) (N = 122) and also differentiate subjects with mental disorders on a pair-wise basis. DSM-III-R diagnoses were determined and supplemented by computerized Quick Diagnostic Interview Schedule. After EEG preprocessing, robust exact low-resolution electromagnetic tomography (ReLORETA) computed EEG sources for 82 brain regions. 20 % of all subjects were then set aside for independent testing. Feature selection methods were then used for the remaining subjects to identify brain source regions that are discriminating between diagnostic categories. RESULTS: Pair-wise classification accuracies between 90 % and 100 % were obtained using independent test subjects whose data were not used for training purposes. The most frequently selected features across various pairs are in the postcentral, supramarginal, and fusiform gyri, the hypothalamus, and the left cuneus. Brain sites discriminating SCZ from HC were mainly in the left hemisphere while those separating BD-ME from HC were on the right. LIMITATIONS: The use of superseded DSM-III-R diagnostic system and relatively small sample size in some disorder categories that may increase the risk of overestimation. CONCLUSIONS: DLA of EEG could be trained to autonomously classify psychiatric disorders with over 90 % accuracy compared to an expert clinical team using standardized operational methods.

Integrity of central nervous function in diabetes mellitus assessed by resting state EEG frequency analysis and source localization.

Diabetes mellitus (DM) is associated with structural and functional changes of the central nervous system. We used electroencephalography (EEG) to assess resting state cortical activity and explored associations to relevant clinical features. Multichannel resting state EEG was recorded in 27 healthy controls and 24 patients with longstanding DM and signs of autonomic dysfunction. The power distribution based on wavelet analysis was summarized into frequency bands with corresponding topographic mapping. Source localization analysis was applied to explore the electrical cortical sources underlying the EEG. Compared to controls, DM patients had an overall decreased EEG power in the delta (1-4Hz) and gamma (30-45Hz) bands. Topographic analysis revealed that these changes were confined to the frontal region for the delta band and to central cortical areas for the gamma band. Source localization analysis identified sources with reduced activity in the left postcentral gyrus for the gamma band and in right superior parietal lobule for the alpha1 (8-10Hz) band. DM patients with clinical signs of autonomic dysfunction and gastrointestinal symptoms had evidence of altered resting state cortical processing. This may reflect metabolic, vascular or neuronal changes associated with diabetes.

A machine learning approach using auditory odd-ball responses to investigate the effect of Clozapine therapy.

OBJECTIVE: To develop a machine learning (ML) methodology based on features extracted from odd-ball auditory evoked potentials to identify neurophysiologic changes induced by Clozapine (CLZ) treatment in responding schizophrenic (SCZ) subjects. This objective is of particular interest because CLZ, though a potentially dangerous drug, can be uniquely effective for otherwise medication-resistant SCZ subjects. We wish to determine whether ML methods can be used to identify a set of EEG-based discriminating features that can simultaneously (1) distinguish all the SCZ subjects before treatment (BT) from healthy volunteer (HV) subjects, (2) distinguish EEGs collected before CLZ treatment (BT) vs. those collected after treatment (AT) for those subjects most responsive to CLZ, (3) discriminate least responsive subjects from HV AT, and (4) no longer discriminate most responsive subjects from HVs AT. If a set of EEG-derived features satisfy these four conditions, then it may be concluded that these features normalize in responsive subjects as a result of CLZ treatment, and therefore potentially provide insight into the functioning of the drug on the SCZ brain. METHODS: Odd-ball auditory evoked potentials of 66 HVs and 47 SCZ adults both BT and AT with CLZ were derived from EEG recordings. Treatment outcome, after at least one year follow-up, was assessed through clinical rating scores assigned by an experienced clinician, blind to EEG results. Using a criterion of at least 35% improvement after CLZ treatment, subjects were divided into "most-responsive" (MR) and "least-responsive" (LR) groups. As a first step, a brain source localization (BSL) procedure was employed on the EEG signals to extract source waveforms from specified brain regions. ML methods were then applied to these source waveform signals to determine whether a set of features satisfying the four conditions outlined above could be discovered. RESULTS: A set of cross-power spectral density (CPSD) features meeting these criteria was identified. These CPSD features, consisting of a combination of brain regional source activity and connectivity measures, significantly overlap with the default mode network (DMN). All decrease with CLZ treatment in responding SCZs. CONCLUSIONS: A set of EEG-derived discriminating features which normalize as a result of CLZ treatment was identified. These discriminating features define a network that shares significant commonality with the DMN. Our findings are consistent with those of previous literature, which suggest that regions of the DMN are hyperactive and hyperconnected in SCZ subjects. Our study shows that these discriminating features decrease after treatment, consistent with portions of the DMN normalizing with CLZ therapy in responsive subjects. SIGNIFICANCE: Machine learning is proposed as a potentially powerful tool for analysis of the effect of medication on psychiatric illness. If replicated, the proposed approach could be used to gain some improved understanding of the effect of neuroleptic medications in treating psychotic illness. These results may also be useful in the development of new pharmaceuticals, since a new drug which induces changes in brain electrophysiology similar to those seen after CLZ could also have powerful antipsychotic properties.

A cortical source localization analysis of resting EEG data after remifentanil infusion.

OBJECTIVE: To explore changes in current source density locations after remifentanil infusion in healthy volunteers using source localization of the electroencephalography (EEG). METHODS: EEG data was collected from 21 males using a 62-electrode system. Additionally, cognitive performance was evaluated by a continuous reaction time paradigm, and pain scores were obtained for experimental bone and heat stimuli. Data were recorded before and during treatment with remifentanil and placebo. Source localization was performed by sLORETA at delta (1-3.9 Hz), theta (4-7.9 Hz), alpha (8-12 Hz), beta1 (12.1-18 Hz), and beta2 (18.1-30 Hz) frequency bands. RESULTS: Pre-treatment recordings demonstrated reproducible source characteristics. The alterations (i.e., pre- versus post-treatment) due to remifentanil were significantly and robustly different from placebo infusions. The results indicated that neurons in several brain areas including inferior frontal gyrus and insula at frontal lobe oscillated more strongly after remifentanil infusion compared to placebo. Furthermore, the source activity at delta band was correlated with continuous reaction time index. CONCLUSIONS: These results indicate that alterations in brain oscillations during remifentanil are mostly localized to frontal, fronto-temporal and fronto-central lobes and related to cognitive function. SIGNIFICANCE: The approach offers the potential to be used for understanding the underlying mechanism of action of remifentanil on brain activity.

Disruption of cortical connectivity during remifentanil administration is associated with cognitive impairment but not with analgesia.

BACKGROUND: The authors investigated the effect of remifentanil administration on resting electroencephalography functional connectivity and its relationship to cognitive function and analgesia in healthy volunteers. METHODS: Twenty-one healthy male adult subjects were enrolled in this placebo-controlled double-blind cross-over study. For each subject, 2.5 min of multichannel electroencephalography recording, a cognitive test of sustained attention (continuous reaction time), and experimental pain scores to bone-pressure and heat stimuli were collected before and after infusion of remifentanil or placebo. A coherence matrix was calculated from the electroencephalogram, and three graph-theoretical measures (characteristic path-length, mean clustering coefficient, and relative small-worldness) were extracted to characterize the overall cortical network properties. RESULTS: Compared to placebo, most graph-theoretical measures were significantly altered by remifentanil at the alpha and low beta range (8 to 18 Hz; all P < 0.001). Taken together, these alterations were characterized by an increase in the characteristic path-length (alpha 17% and low beta range 24%) and corresponding decrements in mean clustering coefficient (low beta range -25%) and relative small-worldness (alpha -17% and low beta range -42%). Changes in characteristic path-lengths after remifentanil infusion were correlated to the continuous reaction time index (r = -0.57; P = 0.009), while no significant correlations between graph-theoretical measures and experimental pain tests were seen. CONCLUSIONS: Remifentanil disrupts the functional connectivity network properties of the electroencephalogram. The findings give new insight into how opioids interfere with the normal brain functions and have the potential to be biomarkers for the sedative effects of opioids in different clinical settings.

Altered cortical causality after remifentanil administration in healthy volunteers: a novel approach for pharmaco-EEG.

Alterations in cortical causality information flow induced by remifentanil infusion in healthy volunteers was investigated in a placebo-controlled double-blind cross-over study. For each of the 21 enrolled male subjects, 2.5 minutes of resting electroencephalography (EEG) data were collected before and after infusion of remifentanil and placebo. Additionally, to assess cognitive function and analgesic effect, continuous reaction time (CRT) and bone pressure and heat pain were assessed, respectively. The causality information was extracted from the EEG by phase slope index (PSI). Among the features being reproducible between the two baseline recordings, several PSI features were altered by remifentanil administration in comparison to placebo. Furthermore, several of the PSI features altered by remifentanil were correlated to changes in both CRT and pain scores. The results indicate that remifentanil administration influence the information flow between several brain areas. Hence, the EEG causality approach offers the potential to assist in deciphering the cortical effects of remifentanil administration.

A machine learning approach using EEG data to predict response to SSRI treatment for major depressive disorder.

OBJECTIVE: The problem of identifying, in advance, the most effective treatment agent for various psychiatric conditions remains an elusive goal. To address this challenge, we investigate the performance of the proposed machine learning (ML) methodology (based on the pre-treatment electroencephalogram (EEG)) for prediction of response to treatment with a selective serotonin reuptake inhibitor (SSRI) medication in subjects suffering from major depressive disorder (MDD). METHODS: A relatively small number of most discriminating features are selected from a large group of candidate features extracted from the subject's pre-treatment EEG, using a machine learning procedure for feature selection. The selected features are fed into a classifier, which was realized as a mixture of factor analysis (MFA) model, whose output is the predicted response in the form of a likelihood value. This likelihood indicates the extent to which the subject belongs to the responder vs. non-responder classes. The overall method was evaluated using a "leave-n-out" randomized permutation cross-validation procedure. RESULTS: A list of discriminating EEG biomarkers (features) was found. The specificity of the proposed method is 80.9% while sensitivity is 94.9%, for an overall prediction accuracy of 87.9%. There is a 98.76% confidence that the estimated prediction rate is within the interval [75%, 100%]. CONCLUSIONS: These results indicate that the proposed ML method holds considerable promise in predicting the efficacy of SSRI antidepressant therapy for MDD, based on a simple and cost-effective pre-treatment EEG. SIGNIFICANCE: The proposed approach offers the potential to improve the treatment of major depression and to reduce health care costs.

A machine learning approach using P300 responses to investigate effect of clozapine therapy.

Clozapine (CLZ) is uniquely effective as a treatment for medication resistant schizophrenia. Information regarding its mechanism of action may offer clues to the pathophysiology of the disease and to improved treatment. In this study we employ a machine learning (ML) analysis of P300 evoked potentials obtained from quantitative electroencephalography (QEEG) data to identify changes in the brain induced by CLZ treatment. We employ brain source localization (BSL) on the EEG signals to extract source waveforms from specified regions of the brain. A subset of 8 features is selected from a large set of candidate features (consisting of spectral coherences between all identified source waveforms at multiple frequencies) that discriminate (by means of a classifier) between the pre- and post-treatment data for the schizophrenics (SCZ) most responsive to CLZ. We show these same selected features also discriminate between pre-treatment most responsive SCZ and healthy volunteers (HV), but not after treatment. Of note, these same features discriminate the least responsive SCZ from HV both pre- and post-treatment. This analysis suggests that the net beneficial effects of CLZ in SCZ are reflected in a normalization of P300 brain-source generators.

A machine learning approach for distinguishing age of infants using auditory evoked potentials.

OBJECTIVE: To develop a high performance machine learning (ML) approach for predicting the age and consequently the state of brain development of infants, based on their event related potentials (ERPs) in response to an auditory stimulus. METHODS: The ERP responses of twenty-nine 6-month-olds, nineteen 12-month-olds and 10 adults to an auditory stimulus were derived from electroencephalogram (EEG) recordings. The most relevant wavelet coefficients corresponding to the first- and second-order moment sequences of the ERP signals were then identified using a feature selection scheme that made no a priori assumptions about the features of interest. These features are then fed into a classifier for determination of age group. RESULTS: We verified that ERP data could yield features that discriminate the age group of individual subjects with high reliability. A low dimensional representation of the selected feature vectors show significant clustering behavior corresponding to the subject age group. The performance of the proposed age group prediction scheme was evaluated using the leave-one-out cross validation method and found to exceed 90% accuracy. CONCLUSIONS: This study indicates that ERP responses to an acoustic stimulus can be used to predict the age and consequently the state of brain development of infants. SIGNIFICANCE: This study is of fundamental scientific significance in demonstrating that a machine classification algorithm with no a priori assumptions can classify ERP responses according to age and with further work, potentially provide useful clues in the understanding of the development of the human brain. A potential clinical use for the proposed methodology is the identification of developmental delay: an abnormal condition may be suspected if the age estimated by the proposed technique is significantly less than the chronological age of the subject.

Using pre-treatment electroencephalography data to predict response to transcranial magnetic stimulation therapy for major depression.

We investigate the use of machine learning methods based on the pre-treatment electroencephalograph (EEG) to predict response to repetitive transcranial magnetic stimulation (rTMS), which is a non-pharmacological form of therapy for treating major depressive disorder (MDD). The learning procedure involves the extraction of a large number of candidate features from EEG data, from which a very small subset of most statistically relevant features is selected for further processing. A statistical prediction model based on mixture of factor analysis (MFA) model is constructed from a training set that classifies the respective subject into responder and non-responder classes. A leave-2-out (L2O) cross-validation procedure is used to evaluate the prediction performance. This pilot study involves 27 subjects who received either left high-frequency (HF) active rTMS therapy or simultaneous left HF and right low-frequency active rTMS therapy. Our results indicate that it is possible to predict rTMS treatment efficacy of either treatment modality with a specificity of 83% and a sensitivity of 78%, for a combined accuracy of 80%.

A pilot study to determine whether machine learning methodologies using pre-treatment electroencephalography can predict the symptomatic response to clozapine therapy.

OBJECTIVE: To investigate whether applying advanced machine learning (ML) methodologies to pre-treatment electroencephalography (EEG) data can predict the response to clozapine therapy in adult subjects suffering from chronic schizophrenia. METHODS: Pre-treatment EEG data are collected in 23+14 schizophrenic adults. Treatment outcome, after at least one year follow-up, is determined using clinical ratings by a trained clinician blind to EEG results. First, a feature selection scheme is employed to select a reduced subset of features extracted from the subjects' EEG that is most statistically relevant to our treatment-response prediction. These features are then entered into a classifier, which is realized in the form of a kernel partial least squares regression method that performs response prediction. Various scales, including the positive and negative syndrome scale (PANSS) are used as treatment-response indicators. RESULTS: We determined that a set of discriminating EEG features do exist. A low-dimensional representation of the feature space showed significant clustering into clozapine responder and non-responder groups. The minimum level of performance of the proposed prediction methodology, tested over a range of conditions using the leave-one-out cross-validation method using the original 23 subjects, with further testing in an independent sample of 14 subjects, was 85%. CONCLUSIONS: These findings indicate that analysis of pre-treatment EEG data can predict the clinical response to clozapine in treatment resistant schizophrenia. SIGNIFICANCE: If replicated in a larger population, this novel approach to EEG analysis may assist the clinician in determining treatment-efficacy.

Using pre-treatment EEG data to predict response to SSRI treatment for MDD.

The problem of identifying in advance the most effective treatment agent for various psychiatric conditions remains an elusive goal. To address this challenge, we propose a machine learning (ML) methodology to predict the response to a selective serotonin reuptake inhibitor (SSRI) medication in subjects suffering from major depressive disorder (MDD), using pre-treatment electroencephalograph (EEG) measurements. The proposed feature selection technique is a modification of the method of Peng et al [10] that is based on a Kullback-Leibler (KL) distance measure. The classifier was realized as a kernelized partial least squares regression procedure, whose output is the predicted response. A low-dimensional kernelized principal component representation of the feature space was used for the purposes of visualization and clustering analysis. The overall method was evaluated using an 11-fold nested cross-validation procedure for which over 85% average prediction performance is obtained. The results indicate that ML methods hold considerable promise in predicting the efficacy of SSRI antidepressant therapy for major depression.

Diagnosis of psychiatric disorders using EEG data and employing a statistical decision model.

An automated diagnosis procedure based on a statistical machine learning methodology using electroencephalograph (EEG) data is proposed for diagnosis of psychiatric illness. First, a large collection of candidate features, mostly consisting of various statistical quantities, are calculated from the subject's EEG. This large set of candidate features is then reduced into a much smaller set of most relevant features using a feature selection procedure. The selected features are then used to evaluate the class likelihoods, through the use of a mixture of factor analysis (MFA) statistical model [7]. In a training set of 207 subjects, including 64 subjects with major depressive disorder (MDD), 40 subjects with chronic schizophrenia, 12 subjects with bipolar depression and 91 normal or healthy subjects, the average correct diagnosis rate attained using the proposed method is over 85%, as determined by various cross-validation experiments. The promise is that, with further development, the proposed methodology could serve as a valuable adjunctive tool for the medical practitioner.