Dr Yvette Pyne, an Academic Clinical Fellow at the Centre for Academic Primary Care of Bristol University, and Dr Edwin Simpson, Lecturer in the School of Computer Science and Electrical and Electronic Engineering at the University of Bristol, have developed a proof-of-concept machine learning tool demonstrating how an ‘AI assistant’ could monitor GP consultations and determine what medical issue is being discussed, which may have widespread ramifications in the future.
Healthcare AI
Dr Pyne explains: “Artificial Intelligence is increasingly being considered as a valuable tool in healthcare. We’ve developed a proof-of concept Natural Language Processing machine learning tool which can analyse the conversation between a doctor and their patient during a consultation. Eventually such tools might be able to generate patient records automatically as well as integrating the consultation information with existing information held on the Electronic Health Records.”
Using a variety of funding, including a grant from the Elizabeth Blackwell Institute’s Health Data Science research strand, Dr Pyne and Dr Simpson hope that the project might yield a variety of benefits beyond answering AI ‘best practice’ research questions for GPs in primary care.
Beyond best practice
“As well as fostering a valuable and productive collaboration between Bristol University departments – the Bristol Medical School and the Department for Computer Science – we hope that this project will pave the way for further research in computing technology for medical applications across both departments”, said Dr Pyne.
“Eventually, we hope that tools such as this will be able to create concise and complete consultation notes – which would be able to free the doctor to concentrate more fully on the patient, and to deliver better care in the short time slots they are scheduled,” she continued.
NLP – a quick primer
But what is Natural Language Processing (NLP)? Simply put, it’s the application of computational technology to the processing and interpretation of human language. But that rather glib phrase overlooks the work that’s required to develop code sufficient to analyse a given dataset.
Dr Edwin Simpson said: “The challenge is to find enough data of high quality that can be used to train the NLP system. We use examples of consultations that have been assigned medical codes by GPs, but we don’t have enough of this kind of data, so we are developing new ways for NLP to make use of background information about different conditions, such as descriptions of symptoms.”
“The funding from the Elizabeth Blackwell Institute was extremely valuable,” added Dr Pyne. “We could recruit someone with the exact skillset we required to complete the work, which sped the process up considerably, and enabled the project outputs to increase in both quality and quantity.”
The future looks bright
Following the publication of a paper in British Medical Journal Health and Care Informatics, both lead researchers are working on related projects: Dr Pyne is focussed on research into diagnosis of the perimenopause, which she hopes to combine with this project into a doctoral research fellowship application in 2025. Dr Simpson is further investigating the clinical applications of interactive conversational agents.