Stage Classification Performance and Clinical Applicability of a Deep Learning–Based Mobile Application for Pressure Injury Analysis

Purpose: The purpose of this study was twofold: 1) to reduce inter‑rater variability in pressure injury staging and 2) to explore the feasibility of a supportive tool for clinical practice by evaluating the agreement between automated classification results and the clinical applicability of a deep learning‑based mobile application for pressure injury analysis.Methods: In this single‑center prospective pilot study, clinical images from 20 patients with pressure injuries were used to calculate Cohen’s kappa for inter‑expert and expertAI agreements. Stage‑specific performance analyses were conducted only for pressure injuries classified as Stages 2‑4, as no cases of Stage 1, deep‑tissue pressure injury, or unstageable pressure injury were observed. In addition, a 10‑item survey was administered to 50 nurses to assess ease of learning, interface intuitiveness, clinical usefulness, and overall satisfaction.Results: The interexpert kappa was 0.703, indicating substantial agreement, whereas the expertAI kappa ranged from 0.615 to 0.730, reflecting moderate to substantial agreement. Nurse satisfaction (N = 50) showed high positive response rates for ease of learning (94%), intuitive usability (88%), and support for clinical judgment (88%). Opportunities for further functional refinement to support broader clinical implementation were identified.Conclusion: As this was a single-center pilot study, the findings should be interpreted as preliminary. The AI-based application for determining pressure injury stage classification demonstrated moderate to substantial agreement with expert assessments, and nurse satisfaction was generally favorable, suggesting that the application may serve as an assistive decision‑support tool for pressure injury staging. Further multicenter studies are warranted to confirm its clinical applicability and performance.