Recognition of hand disinfection by an alcohol-containing gel using 2D imaging in a clinical setting.

BACKGROUND: Hand hygiene compliance is important for the prevention of healthcare-associated infections. The conventional method of measuring hand disinfection guidelines involves an external observer watching the staff personnel, which introduces bias, and observations are only made for a set period of time. An unbiased, non-invasive automated system for assessing hand sanitisation actions can provide a better estimate of compliance.

AIM: To develop an automated detector to assess hand hygiene compliance in hospitals, without bias from an external observer, capable of making observations at different times of the day, as non-invasive as possible by using only one camera, and collecting as much information as possible from 2D video footage.

METHODS: Video footage with annotations from various sources was collected to determine when staff performed hand disinfection with gel-based alcohol. The frequency response of wrists movement was used to train a Support Vector Machine to identify hand sanitisation events.

FINDINGS: Our system detected sanitisation events with an accuracy of 75.18 %, precision of 72.89 %, and recall of 80.91 %. These metrics provide an overall estimate of hand sanitisation compliance without bias due to the presence of an external observer while collecting data over time.

CONCLUSION: Investigation of these systems is important because they are not constrained by time-limited observations, are non-invasive, and eliminate observer bias. Although there is room for improvement, the proposed system provides a fair assessment of compliance that the hospital can use as a reference to take appropriate action.