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 LIMITATIONS: The use of superseded DSM-III-R diagnostic system and relatively small sample size in some disorder categories that may increase the risk of overestimation.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.

Using Machine Learning Algorithms to Enhance the Management of Suicide Ideation.

Combat veterans; especially those with mental health conditions are an at risk group for suicidal ideation and behaviour. This study attempts to use machine learning algorithm to predict suicidal ideation (SI) in a treatment seeking veteran population. Questionnaire data from 738 patients consisting of veterans, still serving members of the Canadian Forces (CF) and Royal Canadian Mountain Police (RCMP) were examined to determine the likelihood of suicide ideation and to identify key variables for tracking the risk of suicide. Unlike conventional approaches we use pattern recognition methods, known collectively as machine learning (ML), to examine multivariate data and identify patterns associate with suicidal ideation. Our findings show that accurate prediction of SI of over 84.4% can be obtained with 25 variables, and 81% using as little as 10 variables primarily obtained from the patient health questionnaire (PHQ). Surprisingly the best identifiers for SI did not come from occupational experiences but rather the patient quality of health, signifying that these findings could be applied to the general population. Our results suggest that ML could assist clinicians to develop a better screening aid for suicidal ideation and behaviour.

Electrographic and pharmacological characterization of a progressive epilepsy phenotype in female MeCP2-deficient mice.

Rett Syndrome is a neurodevelopmental disorder caused primarily by mutations in the gene encoding Methyl-CpG-binding protein 2 (MECP2). Spontaneous epileptiform activity is a common co-morbidity present in Rett syndrome, and hyper-excitable neural networks are present in MeCP2-deficient mouse models of Rett syndrome. In this study we conducted a longitudinal assessment of spontaneous cortical electrographic discharges in female MeCP2-deficient mice and defined the pharmacological responsiveness of these discharges to anti-convulsant drugs. Our data show that cortical discharge activity in female MeCP2-deficient mice progressively increases in severity as the mice age, with discharges being more frequent and of longer durations at 19-24 months of age compared to 3 months of age. Semiologically and pharmacologically, this basal discharge activity in female MeCP2-deficient mice displayed electroclinical properties consistent with absence epilepsy. Only rarely were convulsive seizures observed in these mice at any age. Since absence epilepsy is infrequently observed in Rett syndrome patients, these results indicate that the predominant spontaneous electroclinical phenotype of MeCP2-deficient mice we examined does not faithfully recapitulate the most prevalent seizure types observed in affected patients.

Prediction of antiepileptic drug treatment outcomes using machine learning.

OBJECTIVE: Antiepileptic drug (AED) treatments produce inconsistent outcomes, often necessitating patients to go through several drug trials until a successful treatment can be found. This study proposes the use of machine learning techniques to predict epilepsy treatment outcomes of commonly used AEDs. APPROACH: Machine learning algorithms were trained and evaluated using features obtained from intracranial electroencephalogram (iEEG) recordings of the epileptiform discharges observed in Mecp2-deficient mouse model of the Rett Syndrome. Previous work have linked the presence of cross-frequency coupling (I CFC) of the delta (2-5 Hz) rhythm with the fast ripple (400-600 Hz) rhythm in epileptiform discharges. Using the I CFC to label post-treatment outcomes we compared support vector machines (SVMs) and random forest (RF) machine learning classifiers for providing likelihood scores of successful treatment outcomes. MAIN RESULTS: (a) There was heterogeneity in AED treatment outcomes, (b) machine learning techniques could be used to rank the efficacy of AEDs by estimating likelihood scores for successful treatment outcome, (c) I CFC features yielded the most effective a priori identification of appropriate AED treatment, and (d) both classifiers performed comparably. SIGNIFICANCE: Machine learning approaches yielded predictions of successful drug treatment outcomes which in turn could reduce the burdens of drug trials and lead to substantial improvements in patient quality of life.

Support vector machines using EEG features of cross-frequency coupling can predict treatment outcome in Mecp2-deficient mice.

Anti-convulsive drug treatments of epilepsy typically produce varied outcomes from one patient to the next, often necessitating patients to go through several anticonvulsive drug trials until an appropriate treatment is found. The focus of this study is to predict treatment outcome using a priori electroencephalogram (EEG) features for a rare genetic model of epilepsy seen in patients with Rett Syndrome. Previous work on Mecp2-deficient mice, exhibiting the symptoms of Rett syndrome, have revealed EEG-based biomarkers that track the pathology well. Specifically the presence of cross-frequency coupling of the delta-like (3-6 Hz) frequency range phase with the fast ripple (400 - 600 Hz) frequency range amplitude in long duration discharges was found to track seizure pathology. Support Vector Machines (SVM) were trained with features generated from phase-amplitude comodulograms and tested on (n=6) Mecp2-deficient mice to predict treatment outcome to Midazolam, a commonly used anti-convulsive drug. Using SVMs it was shown that it is possible to generate a likelihood score to predict treatment outcomes on all of the animal subjects. Identifying the most appropriate treatment a priori would potentially lead to improved treatment outcomes.

Identification of brain regions of interest for epilepsy surgery planning using support vector machines.

In patients with intractable epilepsy, surgical resection is a promising treatment; however, post surgical seizure freedom is contingent upon accurate identification of the seizure onset zone (SOZ). Identification of the SOZ in extratemporal epilepsy requires invasive intracranial EEG (iEEG) recordings as well as resource intensive and subjective analysis by epileptologists. Expert inspection yields inconsistent localization of the SOZ which leads to comparatively poor post surgical outcomes for patients. This study employs recordings from 6 patients undergoing resection surgery in order to develop an automated and scalable system for identifying regions of interest (ROIs). Leveraging machine learning techniques and features used for seizure detection, a classification system was trained and tested on patients with Engel class I to class IV outcomes, demonstrating superior performance in the class I patients. Further, classification using features based upon both high frequency and low frequency oscillations was best able to identify channels suited for resection. This study demonstrates a novel approach to ROI identification and provides a path for developing tools to improve outcomes in epilepsy surgery.

Rescue of behavioral and EEG deficits in male and female Mecp2-deficient mice by delayed Mecp2 gene reactivation.

Mutations of the X-linked gene encoding methyl CpG binding protein type 2 (MECP2) are the predominant cause of Rett syndrome, a severe neurodevelopmental condition that affects primarily females. Previous studies have shown that major phenotypic deficits arising from MeCP2-deficiency may be reversible, as the delayed reactivation of the Mecp2 gene in Mecp2-deficient mice improved aspects of their Rett-like phenotype. While encouraging for prospective gene replacement treatments, it remains unclear whether additional Rett syndrome co-morbidities recapitulated in Mecp2-deficient mice will be similarly responsive to the delayed reintroduction of functional Mecp2. Here, we show that the delayed reactivation of Mecp2 in both male and female Mecp2-deficient mice rescues established deficits in motor and anxiety-like behavior, epileptiform activity, cortical and hippocampal electroencephalogram patterning and thermoregulation. These findings indicate that neural circuitry deficits arising from the deficiency in Mecp2 are not engrained, and provide further evidence that delayed restoration of Mecp2 function can improve a wide spectrum of the Rett-like deficits recapitulated by Mecp2-deficient mice.

Characterization of HFOs in short and long duration discharges recorded from in-vivo MeCP2-deficient mice.

Mutations in the X-linked gene encoding methyl CpG-binding protein 2 (MeCP2) have been linked to a neurodevelopmental disorder known as Rett syndrome. The disorder is associated with a number of symptoms, of which epileptic seizures are common. In this study we examined the presence of high frequency oscillations (HFOs) and their interactions with low frequency oscillations (LFOs) during epileptiform-like discharges using intracranial electroencephalogram (iEEG) recordings from male and female Mecp2-deficient mice. The study compared differences in mean HFO power levels normalized to baseline along with LFO-HFO modulation observed in short and long duration discharges. Short duration discharges, common to both male and female Mecp2-deficient mice, showed a decrease in mean HFO power levels compared to baseline levels. During the short duration discharges the theta (7-9 Hz) LFOs were found to modulate fast ripple (350-500 Hz) HFOs predominantly in the female Mecp2-deficient mice. Long duration discharges, predominantly observed in male Mecp2-deficient mice, were found to have elevated mean power levels in the ripple (80-200 Hz) and fast ripple (350-500 Hz) frequency ranges when compared to baseline. During the long duration discharges a lower frequency range theta LFO (4-6 Hz) modulated both the ripple (80-200 Hz) and fast ripple (350-500 Hz) HFOs. These findings suggest that the long duration discharges observed in male Mecp2-deficient mice share biomarkers indicative of seizure-like activity.

Low frequency-modulated high frequency oscillations in seizure-like events recorded from in-vivo MeCP2-deficient mice.

Rett syndrome is a neurodevelopmental condition caused by mutations in the gene encoding methyl CpG-binding protein 2 (MeCP2). Seizures are often associated with Rett syndrome and can be observed in intracranial electroencephalogram (iEEG) recordings. To date most studies have focused on the low frequencies oscillations (LFOs), however recent findings in epilepsy studies link high frequency oscillations (HFOs) with epileptogenesis. In this study, we examine the presence of HFOs in the male and female MeCP2-deficient mouse models of Rett syndrome and their interaction with the LFOs present during seizure-like events (SLEs). Our findings indicate that HFOs (200-600 Hz) are present during the SLEs and in addition, we reveal strong phase-amplitude coupling between LFOs (6-10 Hz) and HFOs (200-600 Hz) during female SLEs in the MeCP2-deficient mouse model.

Characterization of seizure-like events recorded in vivo in a mouse model of Rett syndrome.

Rett syndrome is a neurodevelopmental disorder caused by mutations in the X-linked gene encoding methyl-CpG-binding protein 2 (MECP2). Spontaneous recurrent discharge episodes are displayed in Rett-related seizures as in other types of epilepsies. The aim of this paper is to investigate the seizure-like event (SLE) and inter-SLE states in a female MeCP2-deficient mouse model of Rett syndrome and compare them to those found in other spontaneous recurrent epilepsy models. The study was performed on a small population of female MeCP2-deficient mice using telemetric local field potential (LFP) recordings over a 24 h period. Durations of SLEs and inter-SLEs were extracted using a rule-based automated SLE detection system for both daytime and nighttime, as well as high and low power levels of the delta frequency range (0.5-4 Hz) of the recorded LFPs. The results suggest SLE occurrences are not influenced by circadian rhythms, but had a significantly greater association with delta power. Investigating inter-SLE and SLE states by fitting duration histograms to the gamma distribution showed that SLE initiation and termination were associated with random and deterministic mechanisms, respectively. These findings when compared to reported studies on epilepsy suggest that Rett-related seizures share many similarities with absence epilepsy.

Daily rhythmic behaviors and thermoregulatory patterns are disrupted in adult female MeCP2-deficient mice.

Mutations in the X-linked gene encoding Methyl-CpG-binding protein 2 (MECP2) have been associated with neurodevelopmental and neuropsychiatric disorders including Rett Syndrome, X-linked mental retardation syndrome, severe neonatal encephalopathy, and Angelman syndrome. Although alterations in the performance of MeCP2-deficient mice in specific behavioral tasks have been documented, it remains unclear whether or not MeCP2 dysfunction affects patterns of periodic behavioral and electroencephalographic (EEG) activity. The aim of the current study was therefore to determine whether a deficiency in MeCP2 is sufficient to alter the normal daily rhythmic patterns of core body temperature, gross motor activity and cortical delta power. To address this, we monitored individual wild-type and MeCP2-deficient mice in their home cage environment via telemetric recording over 24 hour cycles. Our results show that the normal daily rhythmic behavioral patterning of cortical delta wave activity, core body temperature and mobility are disrupted in one-year old female MeCP2-deficient mice. Moreover, female MeCP2-deficient mice display diminished overall motor activity, lower average core body temperature, and significantly greater body temperature fluctuation than wild-type mice in their home-cage environment. Finally, we show that the epileptiform discharge activity in female MeCP2-deficient mice is more predominant during times of behavioral activity compared to inactivity. Collectively, these results indicate that MeCP2 deficiency is sufficient to disrupt the normal patterning of daily biological rhythmic activities.

Responsive neuromodulators based on artificial neural networks used to control seizure-like events in a computational model of epilepsy.

Deep brain stimulation (DBS) has been noted for its potential to suppress epileptic seizures. To date, DBS has achieved mixed results as a therapeutic approach to seizure control. Using a computational model, we demonstrate that high-complexity, biologically-inspired responsive neuromodulation is superior to periodic forms of neuromodulation (responsive and non-responsive) such as those implemented in DBS, as well as neuromodulation using random and random repetitive-interval stimulation. We configured radial basis function (RBF) networks to generate outputs modeling interictal time series recorded from rodent hippocampal slices that were perfused with low Mg²âº/high K⁺ solution. We then compared the performance of RBF-based interictal modulation, periodic biphasic-pulse modulation, random modulation and random repetitive modulation on a cognitive rhythm generator (CRG) model of spontaneous seizure-like events (SLEs), testing efficacy of SLE control. A statistically significant improvement in SLE mitigation for the RBF interictal modulation case versus the periodic and random cases was observed, suggesting that the use of biologically-inspired neuromodulators may achieve better results for the purpose of electrical control of seizures in a clinical setting.

EEG analysis for estimation of duration and inter-event intervals of seizure-like events recorded in vivo from mice.

Rett syndrome is a neurodevelopmental disorder of the brain that affects females more often than males. Its cause is linked to the mutations within the gene encoding methyl CpG-binding protein 2 (MeCP2). Presently, there is little information regarding how the loss of MeCP2 affects brain activity. It has been documented that during awake but immobile state, the MeCP2 deficient mice exhibit spontaneous, rhythmic electroencephalogram (EEG) seizure-like events (SLEs) in the range of 6-9 Hz. In this study, we analyze the cortical EEG activity in female MeCP2-deficient mice over 24 hour recordings. Characterizing the SLE and inter-SLE durations by fitting to a gamma distribution we show similarity to previous in vivo epilepsy studies. These results suggest that the SLE and inter-SLE dynamics differ. More precisely, the SLE terminations appear to be a result of time-dependent mechanisms, whereas the inter-SLEs are a result of a random process.