University of Twente Student Theses

Login

Enhancing interictal discharge detection via a deep learning approach

Meulenbrugge, Erik-Jan (2020) Enhancing interictal discharge detection via a deep learning approach.

[img] PDF
14MB
Abstract:Introduction: The usage of EEG analysis, a widely used diagnostic tool, predicted to increase manifolds in the near future, due to the aging world population. Creating an automated EEG analysis will reduce the increasing burden. Our contribution is twofold, 1) enhancing an interictal epileptiform discharge (IED) detector (SpikeNet). 2) Validating and enhancing automated slowing (tBSI) and asymmetry (BSI) detection. Methods: 1) We used 15 rounds of hard example mining and we enlarged the training set by generating IED’s using a Generative Adversarial Network. 2) We generated a reference matrix via deep learning to bypass the need for healthy reference EEG for the tBSI. We, furthermore, enhanced performance by optimizing bandwidths where calculations are applied. Results: 1) The hard example mining increased the AUCROC to 0.9985 and reduces the false positive per hour (FP/h) up to 70% to 15. Incorporating generated EEG increases the FP/h to 18.3. 2) The AUCROC of the BSI and tBSI increased to 0.95 and 0.88 respectively. Conclusion: 1) Hard example mining does statistically improves the performance of SpikeNet. The GAN showed promising results, but contributions need to be made before this method is applicable. 2) We did overcome the need for reference EEG dependency using our reference matrix.
Item Type:Essay (Master)
Faculty:TNW: Science and Technology
Subject:44 medicine, 50 technical science in general
Programme:Technical Medicine MSc (60033)
Link to this item:http://purl.utwente.nl/essays/85121
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page