LLM
A neurologist's startup has developed a foundation model, based on 60,000 hours of brain wave records, which reportedly achieved 95% accuracy in identifying epilepsy biomarkers.
The LLM-based tool, developed by researchers from Yonsei University College of Medicine, automatically generates legally required discharge notes in the emergency department.
Trained on local medical laws and treatment guidelines, the model scored 96.4 on the 2025 Korean Medical Licensing Examination.
The model can also make zero-shot predictions.
It is now being tested at Princess Alexandra Hospital in Brisbane.
Claimed to be a world's first, the LLM-driven model has demonstrated over 90% accuracy.
Developed using a corpus of 38 million clinical texts, the open-source model scored 86% in the Korean Medical Licensing Examination.
The EMR-linked tool has been implemented across 16 departments.
Developers at the Chinese University of Hong Kong claim it "communicates like a real doctor."
Researchers claim their invention offers a cost-effective and faster way to do single-cell analysis.