Pragmatic and automated pressure injury detection across a heterogeneous patient population

The incidence rates of pressure injuries (PrIs) are on the rise, intensifying the workload for nurses, leading to bad outcomes for patients, and increased healthcare costs. Individuals whose pressure injuries are not detected early go on to face longer hospital stays, more severe infection, reduced wellbeing, and, in some cases, premature death. Overburdened nurses are dealing with sicker patients, more patient care, and documentation requirements on admission. This leaves less time for skin checks, making the already difficult challenge of detecting pressure injuries even more challenging. Clinical practice guidelines recommend technological supplementation of visual and manual skin inspections, but few facilities have adopted these methods. Technologies to help identify early signs of tissue damage that are easily implemented at the bedside are needed to reduce pressure injuries across all patients, especially in the context of overburdened nursing workflows. Thermal imaging (thermography) shows potential for the early detection of PrIs but there is a critical need to thoroughly evaluate its performance across all patients. Streamlined implementation of thermography is also critical to its adoption in clinical practice. The requirement for frequent image capture and asynchronous review adds to burden and creates a fragmented workflow. Without an automated bedside interpretation, there is no clear feedback loop to motivate further skin inspection or guide interventions. Our project’s overall goal is to increase early detection of PrIs using an automated, nursing-informed solution to improve the use of thermography in the usual clinical workflow. We will accomplish this by collecting and validating a robust and balanced dataset of thermal and optical images containing more than 600 PrI- susceptible patients, more than 100 of whom have PrIs. We will then build and assess an automated, balanced, and robust PrI detection solution using deep learning models to create balanced PrI detection across all patients. A Nursing Advisory Board will participate in the entire study, guiding all decisions and ensuring clinical relevance and external validity of our work.