Artificial intelligence-powered brain screening may save the lives of newborns
An interdisciplinary team from UCC led by researchers from the School of Engineering and Architecture, has discovered an ingenious method for analysing neonatal brain waves through artificial intelligence-assisted sonification. The findings have been published in the prestigious international journal Nature Scientific Reports.
The study proposes a revolutionary method which will empower healthcare professionals to leverage Artificial Intelligence (AI) decision-support using the auditory perception of brainwaves, leading to faster response times and the potential to save many vulnerable newborn babies.
Analysing brainwaves or Electroencephalograms (EEG) is the gold standard in detecting anomalies in brain activity such as seizures. This ground-breaking research has the potential to make Electroencephalogram (EEG) monitoring more prevalent in medical settings (including those in disadvantaged communities). It also reduces the burden of analysing EEG data, allowing two hours of EEG to be screened in just three seconds.
The new method of analysis extends the concept of the stethoscope used by doctors to listen to heart, lung or other sounds from a patient’s body to also listen to brainwaves. While neurophysiologists use EEG recordings to identify seizures visually, this is a slow and cumbersome process for the medic, which involves scrolling through thousands of images. This expertise also requires a significant amount of training that is not readily available on a continuous basis in all hospitals.
This new AI-driven mechanism ingeniously converts brainwaves to sound. Human ears are more sensitive to changes in frequency and evolution of morphology in time – the signature of many seizures. By focusing the listener’s attention to interesting segments in the recording, EEG seizures can be distinctly heard.
Feedback from the end-users (medical professionals with no training in interpreting raw neonatal EEG) indicates that the accuracy of detecting the presence of a seizure in a long recording using this method is on par with experienced neurophysiologists trained to visually interpret neonatal EEG.
Dr. Emanuel Popovici, co-supervisor of the study, School of Engineering and Architecture, UCC, commented:
“This potentially high-impact research further demonstrates the importance of interdisciplinary research and the power of openly available EEG data. It is another great example of the type of projects which can better humanity. The study opens up many possibilities in the future, from battery-operated edge devices to bringing this technology closer to the patient through commercialisation - ultimately contributing to improved care in disadvantaged communities settings”.
Professor Andriy Temko, co-supervisor of the study, School of Engineering and Architecture, UCC, added:
“A lack of interpretation expertise has always been a bottleneck in the widespread usage of EEG monitoring in newborns. To address that, previous research has focused on developing black box AI models to analyse EEG signals. While outstanding performances were obtained, the practical applicability was limited. The AI was uninterpretable by most and useful for those who already know how to analyse neonatal EEG. We have developed a method where AI augments human senses in an explainable manner to keep a healthcare professional in the decision-making loop. It is a potential game-changer in the EEG monitoring industry as the new method brings both accuracy and speed while requiring no training to be adopted in clinical settings".
The work was carried out by researchers in the multi-award-winning Embedded.Systems@UCC lab, School of Engineering and Architecture, UCC which is financially supported by the SFI CRT-AI and Insight Centres. The study involved contributions from researchers from the Department of Anatomy and Neuroscience, UCC and Computer Science, Munster Technological University (MTU).
The paper entitled “A method for AI-assisted human interpretation of neonatal EEG” was co-authored by Sergi Gomez Quintana (Lead author, Embedded.Systems@UCC and CRT-AI Centre), Alison O’Shea (Computer Science at MTU, Andreea Factor (Anatomy and Neuroscience, UCC), Emanuel Popovici (Embedded.Systems@UCC/CRT-AI/Insight Centre), Andriy Temko (Embedded.Systems@UCC).