University of Twente Student Theses
The electrophysiological signature of epilepsy studied by artificial intelligence and explainable artificial intelligence
Hoogteijling, S. (2022) The electrophysiological signature of epilepsy studied by artificial intelligence and explainable artificial intelligence.
Full text not available from this repository.
| Full Text Status: | Access to this publication is restricted |
| Embargo date: | 1 December 2026 |
| Abstract: | Purpose: Neurosurgeons and epileptologist use intra-operative electrocorticography (ioECoG) to delineate epileptogenic tissue. This delineation is challenging due to the paucity of accurate epileptic ioECoG biomarkers. Computer-aided pattern recognition algorithms can aid the delineation process but need to be validated. Our aim was to develop, train and validate AI models for binary classification of ioECoG channels as epileptic or non-epileptic. Methods: We retrospectively included patients who had an Engel 1A outcome from the RESPect database – a database with intracranial electroencephalography data from patients who underwent epilepsy surgery from 2008 on at University Medical Center Utrecht. Patients undergoing amygdala-hippocampectomy were excluded. The ioECoG was measured at 2048Hz and the channels were labeled as being inside or outside the resected area based on pre- and postresection photos. All resected channels were assumed epileptic, given that all patients had Engel 1A outcome. We split the patients into an 80% training and 20% test set after stratification by pathology. We trained and validated an extra tree classifier (ETC) to dichotomously classify a single ioECoG channel based on spectral band powers. We trained and validated multiple deep learning models on the raw ioECoG. We used explainable AI to identify the features learned by the models. Results: In total 93 patients (median age: 14 [range: 0-68] years) were included where 44 had frontal lobe epilepsy, 26 temporal lobe epilepsy, and 23 another anatomical location; all showed MRI abnormalities; 35 patients had tumor tissue, 35 cortical development malformation, 18 other pathology types, and 5 no abnormalities confirmed by pathology. Per patient, we acquired 1-6 pre-resection ioECoG recordings with 6-36 channels where at least 1 channel covered the resection area. The ETC delineated epileptogenic tissue in 5 of the 21 patients in the test set correctly. Explainable AI found that relative fast ripple power was the most specific individual feature for epileptogenicity. The deep learning models did not find any predictive patterns and failed to identify epileptogenic tissue. Significance: This work is the first to build a large ioECoG dataset to train and validate AI models for epileptogenic tissue delineation. We did not find a suitable deep learning model that can identify epileptogenic tissue from the ioECoG. The ETC performances show the potential of AI in epilepsy surgery. |
| Item Type: | Essay (Master) |
| Clients: | UMC Utrecht, Utrecht, The Netherlands |
| Faculty: | TNW: Science and Technology |
| Subject: | 44 medicine, 54 computer science |
| Programme: | Technical Medicine MSc (60033) |
| Link to this item: | https://purl.utwente.nl/essays/93832 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page