AI model grades nursing procedures with near-human precision

Editor's Note

A new study shows video-language models (VLMs) can accurately evaluate nursing skills and generate meaningful feedback, potentially transforming how future nurses are trained and assessed, Cornell University October 6 reports. The study describes the first framework to apply VLMs to automated nursing competency evaluation.

According to the article, the system mimics how humans learn, progressing from recognizing broad procedural actions to detecting fine-grained errors and reasoning about step order. Trained on a large, annotated video dataset of more than 50 nursing procedures, the model identifies when key subactions are missing or incorrect, explains its assessments in natural language, and standardizes evaluation across different levels of skill.

In testing, the artificial intelligence (AI) demonstrated strong performance: a 31.4% accuracy rate in identifying nursing procedures (surpassing leading video models), a 51.7% improvement in segmenting procedural steps, and a 55.4% gain in detecting missing actions compared with baseline models. The model also more than doubled accuracy in identifying misordered steps. These results confirm the feasibility of using VLMs to capture the complexity of real clinical tasks, which require both technical precision and situational reasoning.

Per the study, the framework’s design could ease the heavy workload on nurse educators by delivering consistent, objective feedback. It also provides students with interpretable explanations of their performance, offering a scalable solution to the long-standing problem of subjective and resource-intensive skill evaluation. The researchers note, while temporal precision remains a challenge, the system could initially supplement rather than replace instructor-led assessments. Beyond nursing, the study highlights applications in other procedural training domains such as medical technician certification and emergency response.