Intelligent wearable system for preventing medication errors in anesthesiology - Project Summary/Abstract Adverse events associated with injectable medications are estimated to impact millions of hospitalizations annually with billions of dollars in associated costs and thousands of deaths. Drug administration errors are the most frequently reported critical incidents in anesthesia with error rates for drug delivery events as high as 5-10% of all drugs given. Accidentally selecting the wrong syringe or vial, or giving an incorrect dose, are common drug errors. Multiple solutions have been suggested to decrease these errors, however, most solutions require additional steps prior to drug administration and provider compliance can be problematic. A real-time auditory or visual feedback system that detects syringe handling by an anesthesia provider could provide a virtual second set of eyes to check medications before mistakes occur. This proposal seeks to obtain head-mounted high resolution video footage of anesthesiology providers. Computer vision algorithms will be developed to identify syringes and vials in real-time. The central hypothesis of this work is that machine learning algorithms operating from images obtained from a power efficient camera can accurately identify medications before they are given to patients in the operating room and generate an electronic medical record which is at least as accurate as current, manually generated tools. (1) Determine reliability of drug label identification using machine learning tools. Build a database of 7,500- 15,000 syringe handling events obtained from head mounted cameras on 30 providers at two hospitals and build real time machine learning algorithms to identify syringes, vials, and labels of commonly used anesthesia medications optimized for timely, accurate identification. (2) Assess accuracy of automated drug delivery dose calculations. Build volume analysis software to determine the volume of a drug administered as well as identify the drug concentration to calculate dose. Compare timing and dosages from the manually produced clinical record to those determined from the computer vision algorithm using the video footage as the true value to determine the accuracy of each modality. (3) Design and test the feasibility of a power efficient real-time wearable camera system. A low-cost camera system optimized for all day use with real-time analysis capabilities will be constructed including two separate cameras, one low and one high resolution to minimize battery consumption and thus overall device size and weight, recording in high resolution only when the low-resolution camera detects syringe handling. Through this K08 award the candidate will learn skills needed to be a leading expert in computer vision, specifically, as it can be applied to patient safety in anesthesiology.
