Hasso-Plattner-InstitutSDG am HPI
Hasso-Plattner-InstitutDSG am HPI

Prediction of Epileptic Seizures by Affect Assessment and Analyzing Medication Dosages

Sidratul Moontaha

Chair for Digital Health - Connected Healthcare
Hasso Plattner Institute

Office: Campus III Building G2, Room: G-2.1.21
Email: sidratul.moontaha(at)hpi.de
Links: Homepage  

Supervisor: Prof. Dr. Bert Arnrich


In my doctoral thesis, I perform affective computing with multimodal sensors: Electroencephalography (EEG), Photoplethysmography (PPG), and Electrodermal Activity (EDA)  sensors. The goal is to predict human affects in terms of mental workload, stress, and emotion in a controlled setup and a daily life environment. The motivation is to provide objective biofeedback to epilepsy patients of an upcoming seizure by observing their affectivity. Moreover, within my research work, I analyze the medication dosages given to epilepsy patients to predict seizures. The following section lists all projects that have been completed or ongoing to date during my doctoral studies.


Quantifying Mental Workload and Stress in Everyday Situations

Mental workload is a term from cognitive psychology which refers to the amount of working memory used in the brain. The ratio of the occupied processing capability of the working memory and the amount required by the task can be referred to as mental workload. Therefore, identifying a potential mental overload is essential, especially for drivers, pilots, medical professionals, emergency workers, and air traffic control professionals. Moreover, complex cognitive tasks alone or combined with other factors like time or social pressure can release cortisol resulting in psychological stress, a primary premonitory symptom of an epileptic seizure. Therefore, this project aims to predict the mental workload to provide pre-emptive therapy for epilepsy patients.

This project was started by curating EEG data from 11 healthy participants while performing mental workload and relaxation tasks in a controlled environment. The recent development of wearable EEG devices made data collection convenient through fewer electrodes. Frequency domain features were extracted from raw EEG data after pre-processing using outlier rejection based on a movement filter, spectral filtering, common average referencing, and normalization. An exploratory feature analysis shows that brain asymmetry features are the most important to predict mental workload. After that, the data collection is extended to include multimodality i.e., PPG and EDA sensors and daily life data collection along with the controlled environment data. The data collection and recruitment is ongoing with a successful data collection from 20 participants. The research question of how reliably mental workload can be predicted  mainly in daily life with providing information on the important modalities is yet to be answered from this project. 


Data analysis pipeline for mental workload classification

Online Learning for Wearable EEG-based Emotion Classification

Giving emotional intelligence to the machines could, for instance, facilitate earlydetection and prediction of (mental) diseases and symptoms. Therefore, in this project, objective prediction of emotional states in real-time has been provided by recording Electroencephalography (EEG) data. The real-time emotion classification pipeline trains different binary classifiers for the dimensions of Valence and Arousal from an incoming EEG data stream coming from the state-of-art AMIGOS dataset and the curated dataset within this project. The pipeline outperformed the related work with the immediate label setting and tested for live scenario where the labels arrive after a certain delay. This prototype on the healthy participants can be used to assess the emotional bias of the patient cohort in realtime.The results of this project were submitted at the MDPI (Sensors) journal. As potential future works, we plan to include multimodal sensors, incorporate bio-feedback, and increase the cohort size of the curated dataset. 

Emotion prediction from EEG data stream

State Space Modelling of Event Count Time Series

This project develops an algorithm for analysing event count time series by application of non-linear state space modelling and Kalman filtering. The algorithm is applied to time series of the daily number of seizures of drug-resistant epilepsy patients. This number may depend on dosages of simultaneously administered anti-epileptic drugs, their superposition effects, delay effects, and unknown factors, making the objective analysis of seizure count time series arduous. In order to estimate the states of the non-linear state space model, an iterative extended Kalman filter is employed. Positive definiteness of covariance matrices is preserved by a square-root filtering approach based on Singular Value Decomposition. Non-negativity of the count data is ensured, either by an exponential observation function, or by a newly introduced “affinely distorted hyperbolic” observation function. The resulting algorithm has to be validated on simulated data by deciding whether a particular anti-epileptic drug is increasing or reducing the seizure rate. The algorithm has also to be applied to clinical data from patients suffering from Myoclonic Astatic Epilepsy. The decision on the increase or decrease of seizure counts is then validated by statistical testing and by visual assessment by experienced pediatric epileptologists.

Teaching Activities


  • Moontaha, S., Schumann, F., Arnrich, B. Online Learning for Wearable EEG-based Emotion Classification. MDPI Sensors 2022, 1. (Submitted)
  • Moontaha, S., Kappattanavar, A., Hecker, P., Arnrich, B. (2023, February). Wearable EEG-based Cognitive Load Classification by Personalized and Generalized Model using Brain Asymmetry. 16th International Joint Conference on Biomedical Engineering Systems and Technologies, HEALTHINF 2023. (Submitted)
  • Hecker, P., Kappattanavar, A., Schmitt, M., Moontaha, S., Wagner, J., Eyben, F., Schuller, B., Arnrich, B.(2022, December). Quantifying Cognitive Load from Voice using Transformer-Based Models and a Cross-Dataset Evaluation. IEEE International Conference on Machine Learning and Applications (IEEE ICMLA). (In proceedings)
  • Anders, C., Moontaha, S., Arnrich, B.(2022, October). Towards Multi-Modal Recordings in Daily Life: A Baseline Assessment of an Experimental Framework. Pervasive Health and Smart Sensing at the Information Society. (In proceedings)
  • Moontaha, S., Steckhan, N., Kappattanavar, A., Surges, R., & Arnrich, B. (2020, May). Self-prediction of seizures in drug resistance epilepsy using digital phenotyping: a concept study. In Proceedings of the 14th EAI International Conference on Pervasive Computing Technologies for Healthcare (pp. 384-387).
  • Galka, A., Moontaha, S., & Siniatchkin, M. (2020). Constrained expectation maximisation algorithm for estimating ARMA models in state space representation. EURASIP Journal on Advances in Signal Processing2020(1), 1-37.
  • Moontaha, S., Galka, A., Siniatchkin, M., Scharlach, S., von Spiczak, S., Stephani, U., ... & Meurer, T. (2019, July). SVD square-root iterated extended Kalman filter for modeling of epileptic seizure count time series with external inputs. In 2019 41st Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) (pp. 616-619). IEEE.
  • Moontaha, S., Galka, A., Meurer, T., & Siniatchkin, M. (2018, July). Analysis of the effects of medication for the treatment of epilepsy by ensemble Iterative Extended Kalman filtering. In 2018 40th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) (pp. 187-190). IEEE.