EEG Signal Processing and Machine Learning. Saeid Sanei
of intellectual loss [43].
As for the sleep EEG pattern, older adults enter into drowsiness with a more gradual decrease in EEG amplitude. Over the age of 60, the frontocentral waves become slower, the frequency of the temporal rhythms also decreases, and frequency lowering with slow eye movements become more prominent, and spindles appear in the wave pattern after the dropout of the alpha rhythm. The amplitudes of both phasic and tonic NREM sleep EEG [43] reduce with age. There is also significant change in REM sleep organization with age; the REM duration decreases during the night and there is significant increase in the sleep disruption [43].
Dementia is the most frequent mental disorder that occurs predominantly in the elderly. Therefore, the prevalence of dementia increases dramatically with ageing of the society. Generally, EEGs are a valuable diagnostic tool in differentiation between organic brain syndromes (OBSs) and functional psychiatric disorders [43], and together with EPs play an important role in the assessment of normal and pathological ageing. Ageing is expected to change most neurophysiological parameters. However, the variability of these parameters must exceed the normal degree of spontaneous variability to become a diagnostic factor in acute and chronic disease conditions. Automatic analysis of the EEG during sleep and wakefulness may provide a better contrast in the data and enable a robust diagnostic tool. We next describe particular and very common mental disorders whose early onset may be diagnosed with EEG measurements.
2.9 Mental Disorders
2.9.1 Dementia
Dementia is a syndrome that consists of a decline in intellectual and cognitive abilities. This consequently affects the normal social activities, mode, and the relationship and interaction with other people [44]. EEG is often used to study the effect of dementia. In most cases such as in primary degenerative dementia, e.g. Alzheimer's, and psychiatric disorder, e.g. depression with cognitive impairment, the EEG can be used to detect the abnormality [45].
In [45] dementia is classified into cortical and subcortical forms. The most important cortical dementia is Alzheimer's disease (AD), which accounts for approximately 50% of the cases. Other known cortical abnormalities are Pick's disease and CJD. They are characterized clinically by findings such as aphasia, apraxia, and agnosia. CJD can often be diagnosed using EEG signals. Figure 2.13 shows a set of EEG signals from a patient with CJD. Conversely, the most common subcortical diseases are Parkinson's disease, Huntington's disease, lacunar state, normal pressure hydrocephalus, and progressive supranuclear palsy. These diseases are characterized by forgetfulness, slowing of thought processes, apathy, and depression. Generally, subcortical dementias introduce less abnormality to the EEG patterns than the cortical ones.
In AD, the EEG posterior rhythm (alpha rhythm) slows down and the delta and theta wave activities increase. Conversely, beta wave activity may decrease. In the severe cases epileptiform discharges and triphasic waves can appear. In such cases, cognitive impairment often results. The spectral power also changes; the power increases in delta and theta bands and decreases in beta and alpha bands and also in mean frequency.
The EEG wave morphology is almost the same for AD and Pick's disease. Pick's disease involves the frontal and temporal lobes. An accurate analysis followed by an efficient classification of the cases may discriminate these two diseases. CJD is a mixed cortical and subcortical dementia. This causes slowing of the delta and theta wave activities and, after approximately three months of the onset of the disease, periodic sharp wave complexes are generated which occur almost every second, together with decrease in the background activity [45]. Parkinson's disease is a subcortical dementia, which causes slowing down of the background activity and an increase of the theta and delta wave activities. Some works have been undertaken using spectral analysis to confirm the above changes [46]. Some other disorders such as depression have lesser effect on the EEGs and more accurate analysis of the EEGs has to be performed to detect the signal abnormalities for these brain disorders.
Generally, EEG is usually used in the diagnosis and evaluation of many cortical and subcortical dementias. Often it can help to differentiate between a degenerative disorder such as AD, and pseudodementia due to psychiatric illness [45]. The EEG may also show whether the process is focal or diffuse (i.e. involves the background delta and theta wave activities). The EEG may also reveal the early CJD‐related abnormalities. However, more advanced signal processing and quantitative techniques may be implemented to achieve robust diagnostic and monitoring performance.
Figure 2.13 A set of multichannel EEG signals from a patient suffering from CJD.
2.9.2 Epileptic Seizure and Nonepileptic Attacks
Often the onset of a clinical seizure is characterized by a sudden change of frequency in the EEG measurement. It is normally within the alpha wave frequency band with slow decrease in frequency (but increase in amplitude) during the seizure period. It may or may not be spiky in shape. Sudden desynchronization of electrical activity is found in electrodecremental seizures. The transition from preictal to ictal state, for a focal epileptic seizure, consists of gradual change from chaotic to ordered waveforms. The amplitude of the spikes does not necessarily represent the severity of the seizure. Rolandic spikes in a child of 4–10 years old for example, are very prominent, however the seizure disorder is usually quite benign or there may not be clinical seizure [47].
In terms of spatial distribution, in childhood the occipital spikes are very common. Rolandic central–midtemporal–parietal spikes are normally benign, whereas frontal spikes or multifocal spikes are more epileptogenic. The morphology of the spikes varies significantly with age. However, the spikes may occur in any level of awareness including wakefulness and deep sleep.
Distinction of seizure from common artefacts is not difficult. Seizure artefacts within an EEG measurement have prominent spiky but repetitive (rhythmical) nature, whereas the majority of other artefacts are transients or noise‐like in shape. For the case of the ECG, the frequency of occurrence of the QRS waveforms is approximately 1 Hz. These waveforms have a certain shape which is very different from that of seizure signals.
The morphology of an epileptic seizure signal slightly changes from one type to another. The seizure may appear in different frequency ranges. For example, a petit mal discharge often has a slow spike around 3 Hz, lasting for approximately 70 ms, and normally has its maximum amplitude around the frontal midline. Conversely, higher frequency spike wave complexes occur for patients over 15 years old. Complexes at 4 and 6 Hz may appear in the frontal region of the brain of epileptic patients. As for the 6 Hz complex (also called benign EEG variants and patterns), patients with anterior 6 Hz spike waves are more likely to have epileptic seizures and those with posterior discharges tend to have neuro‐autonomic disturbances [48]. The experiments do not always result in the same conclusion [47]. It was also found that the occipital 6 Hz spikes can be seen and are often drug related (due to hypoanalgetics or barbiturates) and withdrawal [49].
Among nonepileptics, the discharges may occur in patients with cerebrovascular disorder, syncopal attacks, and psychiatric problems [47]. Fast and needle‐like spike discharges may be seen over the occipital region in most congenitally blind children. These spikes are unrelated to epilepsy and normally disappear in older age patients.
Bursts of 13–16 Hz or 5–7 Hz, as shown in Figure 2.14, (also called 14 and 6 Hz waves) with amplitudes less than 75 μV and arch shapes may be seen over the posterior temporal and the nearby regions of the head during sleep. These waves are positive with respect to the background waves. The 6 and 14 Hz waves may appear independently and be found respectively in younger and older children. These waves may be confined to the regions lying beneath a skull defect. Despite the 6 Hz wave, there are rhythmical theta bursts of wave activities relating