Seizure Onset Detection based on Space Time Frequency Energy Distribution map
Translated title
Seizure Onset Detection based on
Author
Wang, Lin
Term
4. term
Publication year
2015
Submitted on
2015-06-03
Pages
80
Abstract
Formålet med dette projekt er at undersøge, om en visualiseringsmetode kan hjælpe med at finde epileptiske anfald i EEG og se, hvordan de spreder sig. Metoden, et Space‑Time‑Frequency Energy Distribution‑kort (STFED‑kort), viser hvordan signalenergien fordeler sig på rum (EEG‑kanaler), tid og frekvens; energiniveauet vises som intensitet, så man får et firedimensionelt overblik. I disse kort fremstår anfaldets start og spredning som en kegleform. Ved hjælp af enkle billedbehandlingstrin—thresholding (grænseværdier) og matematisk morfologi (formbaseret filtrering)—klassificerer vi anfald og skelner dem fra normale EEG‑signaler og artefakter (støj, bevægelse m.m.). Vi analyserede EEG fra 10 patienter med i alt 53 registrerede anfald og sammenlignede vores tilgang med en Support Vector Machine (SVM). STFED‑metoden klarede sig en smule dårligere, men den er ikke patientspecifik og kræver derfor ikke tilpasning til den enkelte patient.
This project investigates whether a visualization method can help detect epileptic seizures in EEG and track how they spread. The method, called the Space‑Time‑Frequency Energy Distribution (STFED) map, shows how signal energy is distributed across space (EEG channels), time, and frequency; energy level is displayed as intensity, creating a four‑dimensional view. In these maps, the onset and spread of a seizure appear in a cone‑like shape. Using simple image‑processing steps—thresholding (setting cutoffs) and mathematical morphology (shape‑based filtering)—we classify seizure segments and separate them from normal EEG signals and artifacts (noise, movement, etc.). We analyzed EEG from 10 patients with 53 recorded seizures and compared our approach with a Support Vector Machine (SVM). The STFED‑based method performed slightly worse, but it is patient‑non‑specific and does not require tailoring to individual patients.
[This abstract was generated with the help of AI]
Keywords
Documents