Ever-advancing technology has already brought clinical knowledge and reasoning skills closer to the level of ophthalmologists, according to scientists at the University of Cambridge.
Artificial intelligence is better than doctors at accurately assessing eye problems, according to a new study.
Scientists at the University of Cambridge say that ever-advancing technology has already brought clinical knowledge and reasoning skills closer to that of an eye specialist. His GPT-4, a “large-scale AI language model,” was tested on doctors at various stages of their careers, including junior doctors, residents, and specialist ophthalmologists.
Each subject was presented with a series of 87 patient scenarios regarding a specific eye problem and asked to choose from four options to make a diagnosis or advise treatment. GPT-4 scored “significantly better” on the test than non-specialist junior doctors who are comparable to general practitioners (GPs) in their level of eye expertise.
The findings, published in the journal PLOS Digital Health, also showed that residents and specialist ophthalmologists scored similarly on the GPT-4, although top-performing doctors had higher scores. The Cambridge research team says that while large-scale language models are unlikely to replace medical professionals, they have the potential to improve medicine as part of clinical workflows.
Researchers say that state-of-the-art large-scale language models such as GPT-4 can be used in “well-controlled situations” such as patient triage and access to healthcare facilities to provide information on eye-related advice, diagnosis, and management. We believe it may be helpful to provide suggestions. The number of specialized medical professionals is limited. Dr Arun Thirunavukarasu, lead author of the study, said, “We will need to examine these patients and be able to determine which cases are not.'' Treatment is required.
“The model can follow well-defined algorithms already in use, making GPT-4 as good as an expert clinician at processing eye symptoms and signs to answer more complex questions. With further development, the large-scale language model could also advise GPs who are struggling to get prompt advice from their eye doctor. Masu.
“Large amounts of clinical text are required to fine-tune and develop these models, and work is underway around the world to facilitate this.” It said the study was “better” than previous studies because it was compared with practicing doctors instead of patients.
Dr Thirunavukarasu, currently an academic foundation doctor at Oxford University Hospitals NHS Foundation Trust, said: 'Doctors don't review exams throughout their careers. I wanted to know how to deal with it.” The ability of practicing physicians to provide a fair comparison. ”
He added: “Patients may already be using commercial models rather than the internet for advice, so we need to characterize the capabilities and limitations of commercial models.”
This test includes several eye health problems, including extreme photosensitivity, poor vision, lesions, and eye itching and pain, taken from a textbook used to test resident ophthalmologists. It included questions about. This textbook is not freely available on the Internet, so its contents are unlikely to be included in his GPT-4 training dataset.
Dr. Thirunavukkarasu said, “Even with the future use of AI, I think physicians will continue to be in charge of patient care. Most importantly, patients will decide whether they want a computer system involved or not.” “It's about making it possible.” That will be an individual decision for each patient. ” GPT-4 and GPT-3.5, or “Generative Pre-Trained Transformers” – trained on datasets containing hundreds of billions of words from articles, books, and other internet sources.
GPT-4 powers the online chatbot ChatGPT to provide “tailor-made” responses to human queries. ChatGPT has recently gained a lot of attention in the medical world because it performs well on medical school exams and provides more accurate and empathetic messages in response to patient questions than human doctors.
The researchers noted that the field of large-scale language models for artificial intelligence is progressing “very rapidly,” with more advanced models being released since the study was conducted, and further reaching the level of expert ophthalmologists. He pointed out that it may be getting closer.
