Events

Public defence in Biomedical Engineering, M.Sc. (Tech) Karoliina Tapani

Automated method in detecting seizures from newborn EEG
 Midjourney AI art generator's interpretation of automated neonatal seizure detection by Peter Hundlinger.

Title of the doctoral thesis: Automated seizure detection for neonatal EEG

Opponent: Associate Professor Maarten De Vos, KU Leuven, Belgium
Custos: Professor Lauri Parkkonen, Aalto University School of Science, Department of Neuroscience and Biomedical Engineering

The doctoral thesis will be publicly displayed 10 days before the defence in the publication archive of Aalto University.

Electronic doctoral thesis

Public defence announcement:

Epileptic seizures, occurring in 1-5 out of 1000 live births, damage the newborn brain. The gold standard of seizure detection is to measure the electrical activity of the brain with electroencephalography (EEG), which is then manually analyzed by clinical experts. Seizure detection of critically ill newborns must be prompt and reliable to optimize treatment and predict outcome. However, as clinical expertise is often not available on demand, there is a pressing need for automated seizure detection methods.

In this thesis, a machine-learning based seizure detection algorithm (SDA) was developed. The aims of the thesis were threefold: 1) find measures that differentiate seizure EEG from background brain activity, and combine them into an SDA, 2) determine measures to assess the goodness and clinical relevance of the SDA, and 3) to publish both the algorithm, and the data set. The developed measures were highly discriminatory for seizures, our SDA outperformed other well-known SDAs and provided clinically relevant interpretation of the EEG. Our results suggest that labor-intensive development of seizure detection measures is still a feasible approach in the era of deep learning, which has been criticized for its low interpretability. Even though our results were encouraging, the question remains: when implemented to clinical workflow, would the algorithm improve clinical practice and patient outcomes? To finally determine the clinical value, clinical trials should be performed.

Besides developing tools for newborn seizure detection, open science was an essential part of this thesis. Openly accessible data sets and software facilitate the entire scientific community to reach their goals ¬¬¬– in our field, improve newborn patient care and outcomes. Both the algorithm and our open access data have been utilized in multiple studies in the field, and our algorithm has been integrated into a commercial software for automated seizure detection by a clinical neurodiagnostics company (ANT Neuro, Hengelo, The Netherlands).

Contact details of the doctoral student: [email protected]

  • Published:
  • Updated: