Quickly track epilepsy diagnosis with machine learning technology

Researchers at Monash University are applying advanced machine learning techniques to provide epilepsy patients with more accurate and rapid diagnoses.

The study monitored more than 400 electroencephalogram (EEG) records in patients with and without epilepsy at hospitals in Alfred Health and Royal Melbourne. The strength of this study is that it analyzes data from two different hospital groups and uses two different patient groups to evaluate and compare the results.

There are an estimated 142,000 people living with epilepsy in Australia, and an additional 3-4% of the population will develop symptoms at some stage in their lives.

Currently, the epilepsy diagnostic process is very long and may require multiple EEG records to detect abnormal activity (“spike”). By using machine learning technology to automate epilepsy diagnosis, healthcare professionals can free up time to improve patient care.

These two different hospital patient information datasets show how this particular machine learning model can be applied across different datasets to recognize signs of epilepsy and facilitate automatic labeling of EEG data. I have.

MNHS Faculty Epilepsy Research Fellow and PhD Candidate Muben Janmohamed is responsible for identifying and manually labeling the appropriate EEG, and IT Faculty Project Researcher and PhD Candidate Duong Nhu Guarantee Machine Learning Models I was responsible for doing it. I was able to interpret and train two different datasets.

“The collaboration between Alfred Health and Royal Melbourne Hospital allowed our team to collect two large datasets of routine EEG recordings and use them to model machine learning algorithms,” Nhu said. He said.

“The ability to apply machine learning models to different datasets demonstrates the ability to create more reliable, adaptable, and intelligent algorithms than existing models, such as diagnosing patients in clinics. The model becomes more useful when applied to real-world scenarios. “

Dr. Levin Kuhlmann, a senior lecturer in data science and AI in the IT department, said the study was conducted in two stages.

“The purpose of the first stage is to evaluate an existing pattern associated with the detection of abnormal electrical recordings between neurons in the brain, called epilepsy-like activity. These abnormalities are often the rhythm of a patient’s EEG scan. It’s a sharp spike that stands out from the pattern of the brain, “he said.

“At this stage, we will develop new techniques based on the existing methods of the algorithm and generate more accurate results. By adding long-term data, the second phase of the project will be the detection of new seizures and Focus on how to predict. “

Professor Patrick Kwan of the Department of Neuroscience, School of Medicine, said the future potential of the study gave hope to epilepsy patients.

“Our plan for this study is to continue to improve the current model and further train on additional datasets from other hospitals. We are reliable in multiple hospital environments and In the early stages of epilepsy diagnosis, we aim to develop accurate algorithms that can be used from both routine and sleep-deprived EEG recordings. ”

“Such algorithms are also valuable in low-GDP countries and remote areas with limited expertise in interpreting EEG scans. This is a task that usually requires years of training,” Kwan said. The professor said.

Future applications of this technology could also be used as a training tool for graduate neurologists, providing a set of baseline diagnostics that can be used to compare with epilepsy patient records.

/ Public release.