346
K. Lehnertz
seizures are believed to instantaneously involve almost the entire brain [40], whereas
focal-onset seizures appear to originate from a circumscribed region of the brain
(epileptic focus [61, 134]).
Epilepsy affects more than 50 million people worldwide [111] and is associated
with severe comorbidity, including depression, anxiety and increased mortality [127,
153]. Despite more than 30 licensed epilepsy therapies (anti-epileptic drugs [35, 93],
neuromodulation devices [89, 142]), seizures remain poorly controlled in approximately 25% of people with epilepsy. Epilepsy surgery offers the chance of seizure
remission for about 50% of these people, though there is still some hesitation because
of the associated side effects [41, 58]. For the person with epilepsy, it is the apparent
unpredictability of seizures that is central to the burden of an uncontrollable epilepsy.
If it were possible to reliably identify seizure precursors, a novel approach to control previously uncontrollable seizures would consist of a precursor-identificationtriggered, on-demand delivering a suitable counteracting influence (e.g., neuromodulation, local cooling, local drug perfusion, or behavioral intervention [101, 116,
123]) to prevent the generation of the extreme event [67, 85]. Research over the last
three decades provides strong evidence for seizures in many people with epilepsy to
be preceded by a measurable changes in brain dynamics, which constitute a precursor
of sufficient duration [77]. This review summarizes the advances made in studies on
seizure prediction during this time and highlights open issues that would need to be
addressed in future studies.
22.2 Seizure Prediction and Time Series Analysis
Seizure prediction is an interdisciplinary research field that targets at predicting (or
forecasting or anticipating; these terms are often used interchangeably) the occurrence of epileptic seizures with the aim to enable timely modifications of brain
dynamics that prevent seizure generation [77, 91, 106]. Lacking detailed knowledge
about the exact mechanisms underlying seizure generation in humans (and the same
holds true for seizure spreading and seizure termination, despite many decades of
intense research efforts worldwide), there are currently two types of scenarios of
how a seizure could occur [92]. The first scenario considers a sudden and abrupt
transition between two brain states (normal and seizure) and would be conceivable
for the initiation of generalized-onset seizures. The second scenario considers the
transition to the seizure state as a gradual change (or a cascade of changes) in dynamics which could in theory be detected. This type of transition could be more common
for focal-onset seizures.
Lacking an adequate model of the human brain that is capable of mimicking the
immense functionality, and—as in the case of brain pathologies—coexisting normal
and abnormal functions [27, 97, 163], seizure prediction mostly relies on time series
analysis of accessible observables to identify a preseizure state of sufficient duration,
well enough in advance and with high sensitivity and specificity. Most often used
are time series of brain dynamics (e.g., invasively and/or non-invasively recorded
K. Lehnertz
seizures are believed to instantaneously involve almost the entire brain [40], whereas
focal-onset seizures appear to originate from a circumscribed region of the brain
(epileptic focus [61, 134]).
Epilepsy affects more than 50 million people worldwide [111] and is associated
with severe comorbidity, including depression, anxiety and increased mortality [127,
153]. Despite more than 30 licensed epilepsy therapies (anti-epileptic drugs [35, 93],
neuromodulation devices [89, 142]), seizures remain poorly controlled in approximately 25% of people with epilepsy. Epilepsy surgery offers the chance of seizure
remission for about 50% of these people, though there is still some hesitation because
of the associated side effects [41, 58]. For the person with epilepsy, it is the apparent
unpredictability of seizures that is central to the burden of an uncontrollable epilepsy.
If it were possible to reliably identify seizure precursors, a novel approach to control previously uncontrollable seizures would consist of a precursor-identificationtriggered, on-demand delivering a suitable counteracting influence (e.g., neuromodulation, local cooling, local drug perfusion, or behavioral intervention [101, 116,
123]) to prevent the generation of the extreme event [67, 85]. Research over the last
three decades provides strong evidence for seizures in many people with epilepsy to
be preceded by a measurable changes in brain dynamics, which constitute a precursor
of sufficient duration [77]. This review summarizes the advances made in studies on
seizure prediction during this time and highlights open issues that would need to be
addressed in future studies.
22.2 Seizure Prediction and Time Series Analysis
Seizure prediction is an interdisciplinary research field that targets at predicting (or
forecasting or anticipating; these terms are often used interchangeably) the occurrence of epileptic seizures with the aim to enable timely modifications of brain
dynamics that prevent seizure generation [77, 91, 106]. Lacking detailed knowledge
about the exact mechanisms underlying seizure generation in humans (and the same
holds true for seizure spreading and seizure termination, despite many decades of
intense research efforts worldwide), there are currently two types of scenarios of
how a seizure could occur [92]. The first scenario considers a sudden and abrupt
transition between two brain states (normal and seizure) and would be conceivable
for the initiation of generalized-onset seizures. The second scenario considers the
transition to the seizure state as a gradual change (or a cascade of changes) in dynamics which could in theory be detected. This type of transition could be more common
for focal-onset seizures.
Lacking an adequate model of the human brain that is capable of mimicking the
immense functionality, and—as in the case of brain pathologies—coexisting normal
and abnormal functions [27, 97, 163], seizure prediction mostly relies on time series
analysis of accessible observables to identify a preseizure state of sufficient duration,
well enough in advance and with high sensitivity and specificity. Most often used
are time series of brain dynamics (e.g., invasively and/or non-invasively recorded