University of Twente Student Theses

Login

Seizure onset detection for responsive stimulation in the sensorimotor cortex

Smits, Paul L. (2019) Seizure onset detection for responsive stimulation in the sensorimotor cortex.

[img] PDF
3MB
Abstract:Responsive cortical electrical stimulation (CES) may be a promising alternative to surgery for the treatment of medically refractory focal epilepsy in the sensorimotor cortex. Improved approaches for highly specific and fast seizure detection, validated on central lobe epilepsy (CLE) patients, are required for the development of a new generation of intelligent implantable responsive CES devices. Based on a literature review, a 138-dimensional feature space, consisting of cross-correlation features and a set of per-channel time and frequency domain features was chosen to be used in a patient-specific machine learning algorithm. Six bipolar electorcorticogram (ECoG) channels were selected for each of ten CLE patients to represent cortical areas inside and outside the seizure onset zone. Features were extracted from 1s epochs of ictal and interictal data. A Random Forest classifier was trained for each patient and early detection (<10s) sensitivity was obtained from seizure-level leave-one-out cross-validation. The false detection rate (FDR) was determined using a 24h interictal test set from the corresponding patients. The algorithm demonstrates an improvement of early detection sensitivity (98%) and FDR (1.53/h) as compared to the reference algorithm, with short detection delay (3.9s). The patient-specific approach may be employed to achieve closed-loop seizure suppression in responsive CES implants.
Item Type:Essay (Master)
Clients:
University Medical Center Utrecht, Utrecht, Netherlands
Faculty:TNW: Science and Technology
Subject:44 medicine, 50 technical science in general, 54 computer science
Programme:Technical Medicine MSc (60033)
Link to this item:http://purl.utwente.nl/essays/79743
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page