Liu employed two language models: the first is called transformer architecture, and was introduced in a study published last year in the Advances in Neural Information Processing Systems journal. Thus, the focus of this paper is in building language models for clinical notes.”įigure 2: Schematic showing how raw data is transformed to model training data. “The stronger the model, the more effective such features would likely be. “Assistive-writing features for notes, such as auto-completion or error-checking, benefit from language models,” Liu writes in his paper. Artificial intelligence tools could help to tackle this issue, reducing costs spent on additional staff and resources. In his study, pre-published on arXiv, he trained generative models using the MIMIC-III (Medical Information Mart for Intensive Care) EHR dataset, and then compared the notes generated by the models with real notes from the dataset.Ĭommonly adopted methods to reduce the time that clinicians spend on note-taking include the use of dictation services and the employment of assistants who can write up notes for them. Peter Liu, a researcher at Google Brain, has recently developed a new language modeling task that can predict the content of new notes by analyzing patient medical records, which include data such as demographics, laboratory measurements, medications and past notes. Thanks to cutting-edge artificial intelligence tools, this note-writing process could soon become automated, helping doctors to better manage their shifts and relieving them from this tedious task. According to a 2016 study, doctors spend approximately two hours on administrative work for every hour spent with a patient. Physicians currently spend a lot of time writing notes about patients and inserting them into electronic health record (EHR) systems.
0 kommentar(er)
