Saeid Sanei - EEG Signal Processing and Machine Learning

Given the diversity in electric and dielectric properties of the head layers, the distribution of attenuation of brain discharges including cortical, subcortical, and hippocampal activities is not uniform over the scalp and is subject to nonlinearity. Therefore, to model the neuronal pathways or localization of brain activity sources an accurate head electrical model should be available.

From an anatomical point of view the brain may be divided into three parts: the cerebrum, cerebellum, and brain stem ( Figure 1.8). The cerebrum consists of both left and right lobes of the brain with highly convoluted surface layers called the cerebral cortex.

The cerebrum includes the regions for movement initiation, conscious awareness of sensation, complex analysis, and expression of emotions and behaviour. The cerebellum coordinates voluntary movements of muscles and balance maintaining.

The brain stem controls involuntary functions such as respiration, heart regulation, biorhythms, neurohormones, and hormone secretion [26].

The study of EEG paves the way for diagnosis of many neurological disorders and other abnormalities in the human body. The acquired EEG signals from a human (and also from animals) may for example be used for investigation of the following clinical problems [26, 27]:

Monitoring alertness, coma, and brain death.

Locating areas of damage following head injury, stroke, and tumour.

Testing afferent pathways (by EPs).

Monitoring cognitive engagement (alpha rhythm).

Producing biofeedback situations.

Controlling anaesthesia depth (servo anaesthesia).

Investigating epilepsy and locating seizure origin.

Testing epilepsy drug effects.

Assisting in experimental cortical excision of epileptic focus.

Monitoring the brain development.

Testing drugs for convulsive effects.

Investigating sleep disorders and physiology.

Investigating mental disorders. Figure 1.8 Diagrammatic representation of the major parts of the brain.

Recognition of emotions for autistics.

Monitoring mental fatigue for pilots and drivers.

Providing a hybrid data recording system together with other imaging modalities.

This list confirms the rich potential for EEG analysis and motivates the need for advanced signal processing techniques to aid the clinician in their interpretation. We next proceed to describe the brain rhythms, which are expected to be measured within EEG signals.

Some of the mechanisms that generate the EEG signals are known at the cellular level and rest on a balance of excitatory and inhibitory interactions within and between populations of neurons. Although it is well established that the EEG (or MEG) signals result mainly from extracellular current flow, associated with summed post‐synaptic potentials in synchronously activated and vertically oriented neurons, the exact neurophysiological mechanisms resulting in such synchronization to a given frequency band, remain obscure.

Networks are already an integral part of human daily social, business, and intellectual life. Networking science is appealing to the field of neuroscience as the brain function stems from the communication and signalling between the neurons. The measured EEG amplitudes probed at each electrode have been the main parameter in evaluating brain function. Synchrony between left and right brain lobes gives further insight into detection of abnormalities such as mental fatigue and dementia and is associated with many other brain states such as emotions, as stated in the next chapter. The synchrony is often measured in the frequency domain where the variations in frequency and phase, corresponding to the time delay between the lobes, can be easily measured. More generally, recent developments in network science, however, have created a new direction in the study of brain normal and abnormal functions.

Although the fundamental concepts in network science originated from mathematics [28] and are used mostly in communications, a number of well established approaches, such as autoregressive modelling, have been used in characterizing the brain functional connectivity from the multichannel EEG. In addition, graph theory has become popular in designing effective classifiers which can segment the EEGs in time–space into the regions each encompassing a separate functionally connected brain region. In [29], a review of recent advances in neuroscience research in the specific area of brain connectivity as a potential biomarker of Alzheimer's disease with a focus on the application of graph theory can be studied.

Later in this book, we derive equations for the graphs applied to EEG in a similar way to those of brain connectivity estimators. We also observe that machine learning techniques such as deep neural networks can be directly applied to graphs for recognition of the brain state.

Following some details on EEG history, this chapter overviews the neuronal level analysis of the brain function. It also provides some information about the head anatomy. Generation of EEG signals as the result of signalling at the dendrite‐dendrite or axon‐dendrite synapses and production of APs, is an important and fundamental concept covered in this chapter. It is also highlighted that normal brain rhythms, brain evoked responses, and brain connectivity are the outcomes of neuronal activities and should be treated differently to recognize the brain normal and abnormal states.

1 1 Caton, R. (1875). The electric currents of the brain. British Medical Journal 2: 278.

2 2 Walter, W.G. (1964). Slow potential waves in the human brain associated with expectancy, attention and decision. Archiv für Psychiatrie und Nervenkrankheiten 206: 309–322.

3 3 Cobb, M. (2002). Exorcizing the animal spirits: Jan Swammerdam on nerve function. Neuroscience 3: 395–400.

4 4 Danilevsky, V.D. (1877). Investigation into the physiology of the brain. Doctoral thesis. University of Charkov, Quoted after Brazier.

5 5 Brazier, M.A.B. (1961). A history of the electrical activity of the brain; the first half‐century. New York: Macmillan.

6 6 Massimo, A. (July 2004). In memoriam Pierre Gloor (1923–2003): an appreciation. Epilepsia 45 (7): 882.

7 7 Grass, A.M. and Gibbs, F.A. (1938). A Fourier transform of the electroencephalogram. Journal of Neurophysiology 1: 521–526.

8 8 Haas, L.F. (2003). Hans Berger (1873–1941), Richard Caton (1842–1926), and electroencephalography. Journal of Neurology, Neurosurgery, and Psychiatry 74: 9.

9 9 Spear, J.H. (2004). Cumulative change in scientific production: research technologies and the structuring of new knowledge. Perspectives on Science 12 (1): 55–85.

10 10 Kornmüller, A.E. (1935). Der Mechanismus des epileptischen anfalles auf grund bioelektrischer untersuchungen am zentralnervensystem. Fortschritte der Neurologie‐Psychiatrie 7: 391–400; 414–432.

11 11 Fischer, M.H. (1933). Elektrobiologische auswirkungen von krampfgiften am zentralnervensystem. Medizinische Klinik 29: 15–19.

12 12 Fischer, M.H. and Lowenbach, H. (1934). Aktionsstrome des zentralnervensystems unter der einwirkung von krampfgiften, 1. Mitteilung Strychnin und Pikrotoxin. Naunyn‐Schmiedebergs Archiv für experimentelle Pathologie und Pharmakologie 174: 357–382.

13 13 Bremer, F. (1935). Cerveau isole’ et physiologie du sommeil. Compte Rendu de la Sociéte de Biologie (Paris) 118: 1235–1241.

14 14 Niedermeyer, E. (1999). Chapter 1, Electroencephalography, basic principles, clinical applications, and related fields. In: Historical Aspects, 4e (eds. E. Niedermeyer and F.L. da Silva), 1–14. Lippincott Williams & Wilkins.

15 15 Berger, H. (1929). Über das Elektrenkephalogramm des Menschen. 7th report. Archiv für Psychiatrie, Nervenkr, 100: 301–320.