Innovation That Matters

| Photo source Paul G. Allen School of Computer Science & Engineering

Can an AI wearable prevent medication errors?

Health & Wellbeing

This deep learning model could help correct human error and protect patients

Spotted: Emergency clinical settings such as operating rooms and intensive-care units are often fast-paced, stressful environments where human errors sometimes occur. Globally, medication errors, whether that’s mislabelling a syringe or selecting the wrong vial, account for over half of the overall preventable harm in medical care. Could AI help to limit these oversights?

Although there are already safeguard systems in place to prevent medical errors such as scannable barcode systems, in high-stress situations these measures can be forgotten. Instead, a team of researchers at the University of Washington have developed an AI detection model using wearable GoPro style cameras, which can alert clinicians to errors with a high degree of accuracy.

The system doesn’t function by reading words on labels, it scans for other clues including the vial or syringe shape, the cap colour, or other visual features. The researchers trained the model to recognise containers and their contents by using data from over 400 drug draws, with videos collected from 13 different anaesthesiology providers, who were filmed managing vials and syringes in operating rooms.

During tests, the AI video system achieved 99.6 per cent sensitivity and 98.8 per cent specificity when it came to detecting vial swap errors, successfully differentiating between vials being held and those in the background.

Written By: Jessica Wallis

Email: kellyem@uw.edu

Website: washington.edu

Contact: washington.edu/contact