Artificial intelligence in epilepsy phenotyping

Abstract

Artificial intelligence (AI) allows data analysis and integration at an unprecedented granularity and scale. Here we review the technological advances, challenges, and future perspectives of using AI for electro-clinical phenotyping of animal models and patients with epilepsy. In translational research, AI models accurately identify behavioral states in animal models of epilepsy, allowing identification of correlations between neural activity and interictal and ictal behavior. Clinical applications of AI-based automated and semi-automated analysis of audio and video recordings of people with epilepsy, allow significant data reduction and reliable detection and classification of major motor seizures. AI models can accurately identify electrographic biomarkers of epilepsy, such as spikes, high-frequency oscillations, and seizure patterns. Integrating AI analysis of electroencephalographic, clinical, and behavioral data will contribute to optimizing therapy for patients with epilepsy.

Keywords: EEG; artificial intelligence; seizure.

Figure 1.

Artificial intelligence-guided analysis of behavior in animal models of epilepsies. A) Data of freely behaving animals is acquired using various image sensors (e.g., 3D data from a depth camera), either alone or in combination. B) Raw image data (e.g., pixel values) or selected features (e.g., 2D or 3D keypoints) are typically used to capture an animal’s behavior at high granularity. C-D) Different analysis approaches (e.g., probabilistic graphical models) can be deployed to infer the underlying dynamics of behavior and discretize them into stereotypic, reoccurring behavioral states. E) The quantification of behavioral states in mice with acquired and genetic epilepsies reveals distinct, previously hidden behavioral phenotypes that can be tracked during epileptogenesis and in response to anti-seizure medication (adapted from Gschwind; shaded area indicates methods used in the latter paper).

Figure 2.

Various “high-level” feature sources may lead to more complete multi-modal seizure classification models, including 1: deep motion features/motions of interest/action recognition, 2: skeletal keypoint/pose analysis, 3: facial landmarks/emotion recognition, 4: vital signs inference (respiratory/heart rate), 5: sound identification.

Figure 3:. Bioelectronics & Neuromodulation Platform.

The patient interface application (PIA) is an edge-AI device using wireless, bi-directional communication to integrate smartphone, wearables, and implantable devices with a cloud computing environment. Cloud Environment: Database and high-performance computing enabling data storage and the most complex artificial intelligence (AI) analytics. The platform enables tracking and treating ambulatory patients living in their home environments and make near real-time data review and analytics available to remote care-teams and for closed-loop adaptive therapy.