Deep Learning-Based Application for Pressure Injury Management: Samsung Medical Center

Pressure injuries (PIs) are localized damage to the skin and underlying soft tissue, often occurring over bony prominences or in association with medical devices. PIs are classified into six stages (Stages 1-4, Unstageable, and Deep Tissue Injury) based on the depth and type of affected tissue.

PIs can lead to negative outcomes such as prolonged hospital stays, increased medical costs, and delayed recovery, thus worsening patient prognosis. Therefore, the incidence and prevalence of PIs are critical indicators used to evaluate the quality of healthcare institutions. Various scales are used to assess the prevention and treatment of PIs, with the Pressure Ulcers Scale for Healing (PUSH) Tool and the Bates-Jensen Wound Assessment Tool (BWAT) being the most widely used. (Gunes, U. Y., 2009. A Prospective Study Evaluating the Pressure Ulcers Scale for Healing [PUSH Tool] to Assess Stage II, Stage III, and Stage IV PIs. Ostomy/Wound Management, 55(5), 48). Utilizing these tools is crucial for evaluating the effectiveness of PI-prevention interventions implemented by nurses. At Samsung Medical Center (SMC), the PUSH score is used to assess PI status, progression, and deterioration.

SMC has developed a deep-learning based application that supports nurses in assessing PI stages and recommends appropriate dressing materials for each stage. This app, integrated into a BPOC (Barcode-enabled Point of Care, a handheld patient identification barcode system), transmits the recorded data in real-time to the hospital's electronic medical records (EMR) system.

To evaluate the app’s effectiveness, we compared the PUSH scores between app users and non-users. PI wound healing was determined by the improvement in PUSH scores from the initial assessment to a follow-up at seven days. Since the PUSH score decreases as healing occurs, we defined negative changes as an improvement. A chi-square test was employed to determine differences between the groups, with statistical significance set at P<0.05.

The study found that for Pls at Stage 2 and above, the app user group showed 37% higher healing rate than the non-user group.