Research & Development Projects

The stuff we do

Research

Localization of epileptogenic tissue in patients suffering from pharmacoresistant epilepsy still remains challengins. Within this project we develop algorithms and methods to by analysing interictal intracranial EEG. We use multiple signal processing algorithms to extract iEEG features using the library which are subsequently stored in the database. The extracted iEEG features known for their localisation potential (high frequency oscillations, spikes, relative entropy) are presented to the clinicians through web interface that is connected to the database. The features are further used in statistical analyses and machine learning algorithms to provide further information to the clinicians about the epileptogenic zone and to predict surgical outcome. Apart from feature extraction approach we also develop deep learning algorithms to perform classification of signals as normal or pathologic. About 40 % of epileptic patients do not respond to anti seizure drugs and the most effective way of treatment is surgical removal of the epileptogenic zone, the area of cortex that is indispensable for the generation of epileptic seizures. Localization of epileptogenic tissue in patients suffering from pharmacoresistant epilepsy still remains challenging. Within this project we develop advanced signal processing tools and AI assisted diagnostics algorithms to localize epileptogenic tissue by intracranial electro-encephalography. We use multiple signal processing algorithms to extract iEEG features using the EPYCOM library which are subsequently stored in the MINED database. The extracted iEEG features known for their localisation potential (high frequency oscillations, spikes, relative entropy) are presented to the clinicians through a web interface that is connected to the database. The features are further used in statistical analyses and machine learning algorithms to provide further information to the clinicians about the epileptogenic zone and to predict surgical outcome. Apart from the feature extraction approach we also develop deep learning algorithms to perform classification of signals as normal or pathologic.

Collaboration:
Mayo Clinic (Greg Worrell), Duke University (Birgit Frauscher)

Supported by:
Ministry of Health of the Czech Republic, project NU22-08-00278, Czech Science Foundation, project 22-28784S

Formation and retrival of memories is a subject of active research. In this project we aim to elucidate the memory formation and retrival process in the brain. We ask the patients who undergo implantation of intracranial electrodes to perform memory tasks. We extract features (high frequency oscillations, functional connectivity) from the memory tasks and evaluate how the brain reacts when the patient successfully remembers the stimulus or not. The features that correlate with the memory formation are used in machine learning models to predict wherther the stimulus will be remembered. We also activelly develop deep learning models to predict memory encoding.

Collaboration:
Gdansk Technical University (Michal Kucewicz)

Supported by:
Czech Science Foundation, Lead Agency projects 21-44843L and 22-28594K

Development

EPYCOM is a python module with algorithms for processing intracranial EEG. The main motivation behind creation of EPYCOM was to have standardized input/output of various iEEG processing algorithms. This has a benefit of easy incorporation of these algorithms into computational pipelines and promotes automated processing of iEEG signals. The EPYCOM library contains various algorithms for event detections (high frequency oscillations, spikes) and computation of univariate (entropy, frequency-amplitude coupling, ...) and bivariate (linear correlation, relative entropy, ...) signal features.

Documentation:
Source:

MINED is a PostgreSQL based database for storage of recording metadata and computed signal features. MINED database was created to create a unified structured storage of recording metadata and features computed from iEEG signals. The centralized and structured data are important for data reusability, sharing between researchers and data analysis/mining. Added benefit of the structured data is the possibility to create web-based applications for data input and analysis.

Demo:
Documentation:
Source: