Dr Yujiang Wang (yujiang.wang@ncl.ac.uk)
Dr Yu Guan (yu.guan@ncl.ac.uk)
Dr Rhys Thomas (rhys.thomas@ncl.ac.uk)
Background
The symptoms and severity of epileptic seizures can change from one seizure to the next within the same patient. Our recent publication (Schroeder et al. 2020, PNAS) showed that these changes are not random, but rather change according to circadian or longer-term rhythms. In parallel, recent work has also shown that seizure likelihood often follows daily, monthly, and/or seasonal rhythms. These rhythms are also possibly connected with other physiological fluctuations such as respiration and heart rate and influenced by environmental factors like air temperature and pressure. Together, the current evidence suggests that we should be able to predict the periods of time when seizures are likely to be more severe.
As wearable technology and subcutaneous implants become more accessible, physiological signals and EEG can be continuously recorded over a long period of time. We propose to find and track markers of seizure modulation using wearable devices, combined with long-term EEG and environmental sensor data.
Hypothesis
Predictive markers of seizure modulation can be extracted from (ultra-)long-term EEG and wearable sensors ahead of time.
Approach
We will initially use retrospectively collected data (in hand) from epilepsy monitoring units, where epilepsy patients are continuously monitored for several days to weeks. This data typically contains EEG as well as heart rate and blood oxygen saturation levels. We will also combine this with records of environmental factors such as weather.
During the PhD project, we will also setup a study to collect long-term data from patients, through both standard video telemetry and also a subcutaneous chronic EEG system in collaboration with UNEEG. We will additionally capture wearable sensor data at the same time and the student will extract features from the raw data related to movement level, sleep duration/quality, body temperature.
To analyse these long-term multimodal datasets, we propose to use and develop new unsupervised machine learning approaches (such as dimensionality reduction and fluctuation analysis) on different time scales. We will then relate those results to observed seizure features (timing, severity, type) by applying supervised learning techniques to extract clinically-relevant features.
Skills and training
The student will be embedded in the CNNP Lab (www.cnnp-lab.com), which offers a vibrant environment for interdisciplinary neurological research. After getting familiar with existing data and analyses, the student will have the choice to focus on analytical (data science) aspects of the project or establish the prospective collection of wearable data in the epilepsy monitoring unit. This student will be invited to spend 3+ months at UNEEG with their data analytics lab and work alongside experts in wearable technology. The student should complete their PhD with skills in both advanced data analytics, wearable data acquisition, and have an insight into an exciting med-tech domain.
Wider impact
The goal is to create forecasting models that can predict the type and severity of upcoming seizures from non-invasive or minimally invasive devices. Such predictive models can help in designing a seizure advisory system, enabling an entirely new way of treatment.