Titulo Estágio
EEGen: A user-centric framework for the generation of physiologically plausible synthetic EEG time-series.
Áreas de especialidade
Sistemas Inteligentes
Engenharia de Software
Local do Estágio
CISUC
Enquadramento
Data acquired from electroencephalography (EEG), which captures the brain’s dynamic activity, are important across many different clinical research domains such as epilepsy, Parkinson’s disease, anaesthesiology, depression, and the development of brain–computer interfaces.
EEG time-series data are typically multichannel recordings that may involve over a hundred electrodes, each varying in placement, channel montage, sampling rate, recording duration, invasiveness, and file format. Although a wide array of machine-learning algorithms has been devised for diagnosis and analysis, their clinical applicability has been limited due to a lack of access to high-quality long-term EEG datasets derived from patients. Due to several reasons such as privacy concerns, cost and complexity in data acquisition.
To meet the demand for training data, many researchers opt to use generative models from computational neurobiology to create simulated data.
On one hand many of the models used across disciplines are in fact identical and may only differ slightly in their parametrization.
On the other hand, their implementations tend to be one-off tools tailored to specific disorders or research groups. Thus, many tools are not easily accessible, have many dependencies themselves and often require additional user knowledge in fields such as computational modelling and simulation and data management.
Objetivo
The aim of this thesis is the development of an intuitive and accessible tool allowing to generate realistic EEG datasets for different research purposes and exporting them in common formats without requiring specialized expertise.
The plan is to assert the requirements of different research communities for synthetic EEG data, covering commonly used models, standard EEG configurations, and prevailing data standards (EDF, BIDS, etc.).
After survey of current brain dynamics simulation packages, comparing scope, usability, and interoperability, the student might build upon a preexisting Julia module including both general and epilepsy-specific neural mass models, flexible model initialization and integration routines, and data export capabilities to create a unified, user-friendly platform.
Plano de Trabalhos - Semestre 1
1. Review of the state of the art related to the theme in question (1 month)
The current methodologies for simulation of EEG data, as well as file and database standards for real EEG timeseries.
2. Analysis of the existing model and proposed improvements (3 months)
The existing model implementations should be analysed and understood and improvements proposed.
The expected improvements may involve:
- implementation of additional models
- creation of useful scenario presets
- design of a prototype GUI, web interface or command line tool (CLI)
- creation of static compiled binaries
Plano de Trabalhos - Semestre 2
3. Efficient software implementation (3 months).
Either as webservice, well documented CLI, or executable GUI.
4. Test and comparison
Generation of synthetic dataset examples for different applications; comparison with real world data; performance comparison of machine-learning models trained with and without data augmentation.
5. Writing of thesis and possible articles in Conferences, Magazines (2 months)
Depending on the work quality, publications in international conferences may occur, as well in journals.
Condições
Existense of an extensive EEG database for comparison.
Experiênce of the supervisors on both EEG and machine-learning.
Orientador
César Teixeira
cteixei@dei.uc.pt 📩