Epilepsy Research

Bringing Precision Medicine to Neurological Conditions:
Deep Learning for Individualized Disease Management

Despite many new advances in drug therapy and disease understanding, our capabilities in treating and managing neurological diseases are extremely limited.

For example, over 1% of the world’s population, 65 million people, suffer from epilepsy. For one third of these patients, no medical treatment options exist. These patients need to find ways to live with their condition and manage their daily lives around it. For the remaining two thirds of the patient population, medical treatment options are available but have vastly differing and constantly changing results, and quality of treatment. These shortcomings in diagnosis and treatment options are caused by the fact that epilepsy – like most neurological diseases – is a highly individualized condition, i.e it does not look the same in all patients and even for an individual patient, disease expression changes over time. As a result, until recently, the lack of data and measurements made the correct matching of patients and drugs into an unnecessarily long process of trial and error. Measuring drug sensitivity was like targeting a moving goal post. Manual diaries are the basic data source, but these have been proven to be inaccurate and often unreliable.

With the advent of mobile devices that allow to collect patient information in real-time, continuously and at the point of sensing, whilst leveraging miniaturization and IoT data collection platforms, new efforts are being directed towards building individualized patient management systems. Data that is more accurate and more extensive can be used to gain a patient specific understanding of the disease and provide support for decision-making in managing it.

Deep learning has been successfully used to address a large variety of problems in the biomedical field, ranging from image classification in cancer diagnosis to the automatic interpretation of electronic health records. We leverage deep learning to bring precision medicine to epilepsy: the Epilepsy Research Team of IBM Research is at the forefront of developing custom deep learning models for analyzing EEG signals, video data, data from a variety of wearable sensors and Electronic Health Records. Our algorithms learn from individual patient data, synthetic data and generalize across patient populations and monitoring environments. Working closely with our colleagues from IBM Watson Health, our global IBM Research Labs and leading clinical, academic and industry partners our most recent work has shown the feasibility of using specialized neural networks to automatically predict and detect epileptic seizures in near real-time on a mobile device, to classify EEG data into normal/abnormal EEG and to group different types of seizures automatically. This technology is the foundation of automated digital seizure diaries which can be used to improve diagnosis, treatment and management of epilepsy. Digital disease diaries also play a key role in designing clinical trials to determine the efficacy and suitability of epilepsy drug treatments more efficiently.



MRI scan



EEG Signals

Electroencephalography (EEG) data is a physical measure for brain activity, i.e. the electrical pulses originating from neurons firing as the brain processes information. This information is encoded in EEG signals in the form of characteristic patterns which for example correlate with specific epileptic seizure types. We develop advanced deep learning models which can automatically and in near real-time detect, predict and classify EEG patterns that correlate with epileptic seizures.

Video Analysis

Video data -together with EEG tracking - is the gold standard method for monitoring patients in epilepsy monitoring units in clinical settings and increasingly also in non-clinical home environments. We develop novel software suites for de-identifying patient video data and we build new privacy preserving deep learning algorithms to automatically detect and describe epileptic seizure activity in video streams.

Wearable Sensors

Wearable sensors can measure a variety of bodily functions from heart rate to body movements. This sensor data can enhance and augment EEG and Video data allowing a more diversified characterization of epileptic seizure episodes. We build novel deep learning models to analyse wearable sensor data and integrate them with EEG and video data systems for automatic near real-time, point-of-sensing detection, prediction and classification of epileptic seizures.

Electronic Health Records

Electronic Medical Health Records (EHR) hold patient information on for example administered medication, comorbidities and other details of disease expression and progression. We develop novel machine learning and deep learning models to analyze EHR data in an effort to build prognostic models towards faster and more accurate diagnosis as well as individualized and efficient treatment of epilepsy.

Mechanistic Seizure Modeling

Seizures occur in the brain, EEG signals originate from the brain, and neurological conditions affect the brain. Understanding neural mechanisms that generate those brain activities is fundamental to find new treatments. We simulate the brain in a principled way to provide insight about which neuronal mechanisms underly complex cognitive processes, both in healthy and neurological conditions such as epilepsy.