Patent No. 6594524 Adaptive method and apparatus for forecasting and controlling neurological disturbances under a multi-level control

 

Patent No. 6594524

Adaptive method and apparatus for forecasting and controlling neurological disturbances under a multi-level control (Esteller, et al., Jul 15, 2003)

Abstract

A method and apparatus for forecasting and controlling neurological abnormalities in humans such as seizures or other brain disturbances. The system is based on a multi-level control strategy. Using as inputs one or more types of physiological measures such as brain electrical, chemical or magnetic activity, heart rate, pupil dilation, eye movement, temperature, chemical concentration of certain substances, a feature set is selected off-line from a pre-programmed feature library contained in a high level controller within a supervisory control architecture. This high level controller stores the feature library within a notebook or external PC. The supervisory control also contains a knowledge base that is continuously updated at discrete steps with the feedback information coming from an implantable device where the selected feature set (feature vector) is implemented. This high level controller also establishes the initial system settings (off-line) and subsequent settings (on-line) or tunings through an outer control loop by an intelligent procedure that incorporates knowledge as it arises. The subsequent adaptive settings for the system are determined in conjunction with a low-level controller that resides within the implantable device. The device has the capabilities of forecasting brain disturbances, controlling the disturbances, or both. Forecasting is achieved by indicating the probability of an oncoming seizure within one or more time frames, which is accomplished through an inner-loop control law and a feedback necessary to prevent or control the neurological event by either electrical, chemical, cognitive, sensory, and/or magnetic stimulation.

Notes:

 

BACKGROUND OF THE INVENTION

The present invention is in the field of prediction and control of neurological disturbances, particularly in the area of electrographic and clinical seizure onset prediction based on implantable devices with the major goal of alerting and/or avoiding seizures.

Approximately 1% of the world's population has epilepsy, one third of whom have seizures not controlled by medications. Some patients, whose seizures reliably begin in one discrete region, usually in the mesial (middle) temporal lobe, may be cured by epilepsy surgery. This requires removing large volumes of brain tissue, because of the lack of a reliable method to pinpoint the location of seizure onset and the pathways through which seizures spread. The 25% of refractory patients in whom surgery is not an option must resort to inadequate treatment with high doses of intoxicating medications and experimental therapies, because of poorly localized seizure onsets, multiple brain regions independently giving rise to seizures, or because their seizures originate from vital areas of the brain that cannot be removed. For these and all other epileptic patients, the utilization of a predicting device would be of invaluable help. It could prevent accidents and allow these patients to do some activities that otherwise would be risky.

Individuals with epilepsy suffer considerable disability from seizures and resulting injuries, impairment of productivity, job loss, social isolation associated with having seizures, disabling side effects from medications and other therapies. One of the most disabling aspects of epilepsy is that seizures appear to be unpredictable. However, in this invention a seizure prediction system is disclosed. Seizure prediction is a highly complex problem that involves detecting invisible and unknown patterns, as opposed to detecting visible and known patterns involved in seizure detection. To tackle such an ambitious goal, some research groups have begun developing advanced signal processing and artificial intelligence techniques. The first natural question to ask is in what ways the preictal (i.e., the period preceding the time that a seizure takes place) intracranial EEGs (IEEGs) are different from all other IEEGs segments not immediately leading to seizures. When visual pattern recognition is insufficient, quantitative EEG analysis may help extract relevant characteristic measures called features, which can then be used to make statistical inferences or to serve as inputs in automated pattern recognition systems.

Typically, the study of an event involves the goals of diagnosing (detecting) or prognosticating (predicting) such event for corrective or preventive purposes, respectively. Particularly, in the case of brain disturbances such as epileptic seizures, these two major goals have driven the efforts in the field. On one hand, there are several groups developing seizure detection methods to implement corrective techniques to stop seizures, and on the other, there are some groups investigating seizure prediction methods to provide preventive ways to avoid seizures. Among the groups claiming seizure prediction, three categories of prediction can be distinguished, clinical onset (CO) prediction, electrographic onset (EO) prediction studies, and EO prediction systems. All these categories in conjunction with seizure detection compose most of the active research in this field.

Related art approaches have focused on nonlinear methods such as studying the behavior of the principal Lyapunov exponent (PLE) in seizure EEGs, computing a correlation dimension or nonlinear chaotic analysis or determining one major feature extracted from the ictal characteristics of an electroencephalogram (EEG) or electrocorticogram (ECoG).

IMPORTANT TERMINOLOGY DEFINITIONS

Ictal period: time when the seizure takes place and develops.

Preictal period: time preceding the ictal period.

Interictal period or baseline: period at least 1 hour away from a seizure. Note that the term baseline is generally used to denote "normal" periods of EEG activity, however, in this invention it is used interchangeably with interictal period.

Clinical onset (CO): the time when a clinical seizure is first noticeable to an observer who is watching the patient.

Unequivocal Clinical onset (UCO): the time when a clinical seizure is unequivocally noticeable to an observer who is watching the patient.

Unequivocal Electrographic Onset (UEO): also called in this work electrographic onset (EO), indicates the unequivocal beginning of a seizure as marked by the current "gold standard" of expert visual analysis of the IEEG.

Earliest Electrographic Change (EEC): the earliest change in the intracranial EEG (IEEG) preceding the UEO and possibly related to the seizure initiation mechanisms.

Focus Channel: the intracranial EEG channel where the UEO is first observed electrographically.

Focal Adjacent Channel: the intracranial EEG channels adjacent to the focus channel.

Focus Region: area of the brain from which the seizures first originate.

Feature: qualitative or quantitative measure that distills preprocessed data into relevant information for tasks such as prediction and detection.

Feature library: collection of algorithms used to determine the features.

Feature vector: set of selected features used for prediction or detection that forms the feature vector.

Aura: symptom of a brain disturbance usually preceding the seizure onset that may consist of hallucinations, visual illusions, distorted understanding, and sudden, intense emotion, such as anxiety or fear.

FIGS. 11A-11B illustrate some of the defined terms on segments of a raw IEEG signal. Comparison between the preictal segment indicated on FIG. 11A (between the EEC and the UEO times) and the interictal period in FIG. 11B demonstrates the difficulty of discerning between them. The vertical scale in both figures is in microvolts (.mu.V).

SUMMARY OF THE INVENTION

This invention is an automatic system that predicts or provides early detection of seizure onsets or other neurological events or disturbances with the objective of alerting, aborting or preventing seizures or other neurological ailments by appropriate feedback control loops within multiple layers. One of the main differences from other inventions is that the major functions of the brain implantable device is forecasting and preventing seizures or other brain disturbances rather than only detecting them. Unlike other inventions, the goal is to predict the electrographic onset of the disturbance or seizure rather than the clinical onset. Seizure UEO detection is also accomplished as a direct consequence of the prediction and as a means to assess device performance. Furthermore, the innovative presence of a supervisory control provides the apparatus with a knowledge updating capability supported by the external PC or notebook, and a self-evaluation proficiency used as part of the feedback control to tune the device parameters at all stages, also not present in the other art.

The approach disclosed in the present invention, instead of focusing on nonlinear methods, or on one particular feature, targets multiple features from different domains and combines them through intelligent tools such as neural networks and fuzzy logic. Multiple and synergistic features are selected to exploit their complementarity. Furthermore, rather than using a unique crisp output that considers one particular time frame, as the previous methods introduced, the system provides one or more probabilistic outputs of the likelihood of having a seizure within one or more time frames. Based on this, when a threshold probability is reached, an approaching seizure can be declared. The use of these multiple time frames and probabilistic outputs are other distinct aspects from previous research in the field.

The system possesses multiple levels of closed-loop control. Low-level controls are built up within the implantable device, and consist of brain stimulation actuators with their respective feedback laws. The low-level control operates in a continuous fashion as opposed to previous techniques that provide only one closed-loop control that runs only during short times when the seizure onset is detected. The high-level control is performed by a supervisory controller which is achieved through an external PC or notebook. By using sophisticated techniques, the prediction system envisioned allows the patients or observers to take appropriate precautions before the seizure onset to avoid injuries. Furthermore, the special design of the apparatus furnishes powerful techniques to prevent or avoid seizures and to obtain more insight into these phenomena, thereby revealing important clinical information. The innovative use of a supervisory control is the option that confers the apparatus its unique perspective as a warning/control/adaptive long-term device. The warning is achieved by forecasting the disturbance; the control is accomplished by an appropriate feedback law and a knowledge base update law; and the adaptive capability of the device is attained also by the knowledge base update law driven by the supervisory control. This knowledge base resides in an external personal computer (PC) or notebook that is the heart of the supervisory control, where the apparatus computes optimization routines, and self-evaluation metrics to establish its performance over time, to determine required adjustments in the system set points and produce an updating law that is fed back into the system from this higher level of control.

The control law provided in the device allows a feedback mechanism to be implemented based on electrical, chemical, cognitive, intellectual, sensory and/or magnetic brain stimulation. The main input signal to the feedback controller is the probability of having a seizure for one or more time frames. The supervisory control is based on an external control loop, operating at a higher control level, that compiles new information generated at the implantable device into the knowledge base at discrete steps and provides set point calculations based on optimizations performed either automatically, or semi-automatically by the doctor or authorized individual.

The above and other novel features, objects, and advantages of the invention will be understood by any person skilled in the art when reference is made to the following description of the preferred embodiments, taken in conjunction with the accompanying drawings.