£75,203 over 36 months
Awarded in 2019
Improving diagnostic procedures for epilepsy through automated recording and analysis of patient’s history
Professor Markus Reuber
Professor Dimitri Kullmann, Institute of Neurology, University College London
University of Sheffield
The successful treatment of epilepsy depends on a correct diagnosis. However, it often takes much too long to make the initial diagnoses and too many turn out to be wrong. Despite technological advances the diagnosis of epilepsy still depends on an expert interpretation of a patient’s history. The time has come to use advances in computer technology and automatic speech analysis to support the decision-making process with artificial intelligence systems.
Professor Markus Reuber
Transient loss of consciousness (TLOC) is one of the most common reasons why people access emergency care services. Three causes account for over 90% of presentations with TLOC: epilepsy, fainting, or dissociation (also known as psychogenic nonepileptic seizures). These disorders require different treatment, and successful prevention of recurrence depends on a correct diagnosis, so appropriate referral from emergency or primary care to the correct area of specialty (neurology, cardiology or psychiatry) is of great importance. Currently initial diagnostic services are not performing well: about 20% of individuals receive an incorrect diagnosis.
Despite technological advances, the most important tool for the differentiation between the common causes of TLOC is the history from patients, and witnesses if available. However, the process of taking and interpreting the history requires expertise and time which may not be available in emergency and primary care settings. Professor Reuber and colleagues want to improve this process by programming an artificial intelligence system to ask patients about their TLOC symptoms and analyse their answers.
This project is a PhD programme based on previous work in which the researchers demonstrated the diagnostic contributions made by particular TLOC symptom combinations as well as by how patients speak about their TLOC experiences, rather than what they describe. The project will benefit from the research team’s experience with analysing conversations between patients and a ‘digital doctor’ using automatic speech recognition and decision-making software. During the course of this PhD programme the researchers will validate a computer-presented and analysed symptom questionnaire, and explore the additional diagnostic contribution of a fully automated process for capturing and analysing patients’ conversations with a digital doctor.
An automated history-taking machine could increase the diagnostic value of information from patients and witnesses through its more methodical collection and analysis. This should improve the initial triaging process of TLOC presentations in emergency or primary care settings and mean that more patients are referred to the most appropriate medical speciality for the management of their problems. Ultimately the system should enable more patients to get the best possible treatment for their problem more rapidly.