Brain connectivity biomarker characterization by using EEG and TMS techniques

Abstract

Mental health costs the global economy $16 trillion USD in lost economic output. Elaborative investigation for accurate mental health diagnosis at clinics becomes critical. Among neuroimaging modality, electroencephalography (EEG) has advantages of high temporal resolution, low cost, and high portability. Averaging with event-locked based potentials are applied to EEG for achieving higher signal-to-noise ratio also called event-related potentials (ERP). However, 2D EEG is limited at the scalp. Hence, we apply source localization with the sLORETA algorithm which gives cortical source estimates, 3D EEG, and hence links EEG measurement with neural activations thereby adding physiological meaning to the EEG electrode recordings. High performance computers enabled signal data analysis with machine learning techniques, and statistical analysis by which more meaningful biomarkers can be derived for aiding clinicians with mental health diagnosis and treatment. We characterize biomarkers to distinguish schizophrenia patients from healthy controls by using machine learning analysis on localized EEG response to two types of events: (1) auditory stimulus for characterizing biomarkers signifying sensory gating deficits, and (2) Transcranial Magnetic Stimulation (TMS) for brain network characterization that can aid treatment. Schizophrenia is a serious mental illness due to which patients interpret reality abnormally, experience a combination of hallucinations, delusions, and exhibit extremely disordered thinking and behavior that impairs daily functioning. People with schizophrenia require lifelong treatment thereby eliciting huge costs. We achieve meaningful physiological results that can aid diagnosis and treatment. In response to auditory stimulus, we uncover insights about the brain's activity in higher temporal resolution that matched with earlier studies that reported similar findings with fMRI. In response to TMS, analysis and found temporal signatures that were relatively unstable in patients along with spatial signatures that indicated sensory gating deficits as compared to controls. Energy landscape analysis defined selectively 26 biomarker network states that significantly distinguished patients from healthy controls (p<0.05/n). For brain connectivity characterization and analysis, our methods include Localized EEG Dynamics Analysis (LEDA), Localized Source Activation to Duration ratio metric, nonnegative matrix factorization, and data-driven approach, based on maximum entropy and Boltzmann distribution, called energy landscape analysis. Our novel methodologies and biomarkers in high resolution can aid clinical applications.