CREED-N: Clinical Reporting of Electroencephalograms Enhanced through Deep Learning and NLP

Lead Supervisor
Dr James Teo
Consultant Neurologist & Senior Lecturer
Institute of Psychology, Psychiatry and Neuroscience (IoPPN), King’s College London & King’s College Hospital
jamesteo@nhs.net

Co-supervisors
Professor Mark Richardson (IoPPN, KCL); Dr Jorge Cardoso (BEIS)

Project Details

The brain operates by passing electrical signals around its extraordinarily complex network of brain cells. Only one investigation, electroencephalography (EEG), directly measure the electrical activity of the brain. Each EEG examination produces a large amount of digital data describing the function of the brain. These data are extremely complex and difficult to understand. To obtain information from EEG about diagnosis and potential treatment outcome, the current conventional approach is for a trained physician to review the EEG. Usually, the EEG data are presented to the physician on a computer screen as a set of fluctuating lines, similar to the way the variability of stock prices is displayed to a stock trader. The trained EEG physician attempts to identify “by eye” familiar patterns in the EEG data that may reflect specific disease states. 

The PhD candidate will a project to combine signal processing  approach of evaluating EEG waveforms with deep-learning based Natural Language Processing (NLP) to evaluate neurophysiological and clinical features from electronic health records. Combining these two techniques will combine the best of machine learning and human interpretation.

The candidate will have the opportunity to acquire:

  • Natural Language Processing (NLP) skills including working with neural networks like Transformers
  • Signal processing skills ranging from simple basic Fourier transforms to analysis of non-stationary waveforms, and machine-learning-based approaches
  • Clinical Neuroscience, working with neurophysiologists, neurologists and neuroscientists in real-world clinical scenarios

The project will be based between multiple collaborating departments: (1) Clinical Neuroscience in IOPPN, (2) Biomedical Engineering and Imaging Sciences, and (3) Precision Health Informatics. The individual will also work closely and be embedded with the Kings College Hospital Neuroscience department.

Datasets

  1. Neurophysiology signal data and reports from KCH (>10k)
  2. Seizure and epilepsy clinic reports (>10k)

NLP prototypes are already currently being tested on epilepsy clinic reports. Approvals for wide-scale work for (1) and (2) will require approvals from the KCH data governance committee (KERRI) which has research ethics to proceed on opt-out basis on real-world routinely collected data. The primary supervisor will lead this process at KCH.

Keywords

Epilepsy, Seizure, Electroencephalography, Signal Processing