Computer-assisted diagnosis of epilepsy is one step closer with Monash project


By Michael Woodhead

28 Sep 2020

Computer assisted diagnosis of epilepsy  – and possibly even seizure prediction – may be possible by applying artificial intelligence (AI) techniques to interpretation of EEG signals, according to Monash University researchers.

Professor Patrick Kwan and colleagues are involved in a project that aims to automate and simplify the epilepsy diagnosis process by using AI and machine learning to replace the time-consuming and repetitive analysis of EEGs recordings over days and weeks by neurologists.

Working in collaboration with epilepsy clinics at Alfred Health and The Royal Melbourne hospitals, researchers from the university’s Faculty of Information Technology (IT) have obtained over 400 EEG recordings for comparison from patients and healthy individuals.

Senior Lecturer in the Faculty of IT Department of Data Science and AI, Dr Levin Kuhlmann, says the objective of the first stage is to evaluate existing patterns involved in the detection of abnormal epileptiform activity.

“In this stage we will develop new techniques based on the foundation of existing methods for the algorithm to produce more accurate results. With the addition of long-term data, the second phase of the project will focus on novel seizure detection and prediction methods,” he said

Professor Patrick Kwan from the Faculty of Medicine’s Department of Neuroscience, said the future possibilities of this research provide hope of improved diagnostic methods for people with epilepsy.

“Our plans for this research will be to continue to improve the current models and further train it against additional datasets from other hospitals. We aim to develop an accurate algorithm which will be reliable across multiple hospital settings and usable in the early stages of epilepsy diagnosis, from both routine and sleep-deprived EEG recordings.”

“An algorithm like this would also be of value in low GDP countries or remote regions where there is limited expertise in interpreting EEG scans, a task which generally requires many years of training,” said Professor Kwan.

A recent review article co-authored by Professor Kwan explored the progress towards a longer term Holy Grail’ of seizure prediction through the use of artificial intelligence and ‘deep learning’ to recognise predictive patterns in pre-ictal EEG recordings.

It noted that while there has been rapid progress in this field with the development of machine learning algorithms, many barriers and challenges remain to developing a workable system for clinical application.

One of the main barriers to the development of seizure prediction if the lack of open  access to large EEG datasets

Other challenges include the poor reliability and reproducibility of seizure predictions within individuals in different settings – such as sleep cycles – and also across the multiple different seizure types.

“Despite decades of research, seizure prediction remains an unsolved problem. This is likely to remain at least partly because of the inadequate amount of data to resolve the problem,” it concludes.

Already a member?

Login to keep reading.

Email me a login link