Aim: To develop a deep learning-based smart assessment model for pressure injury surface.
Design: Exploratory analysis study.
Methods: Pressure injury images from four Guangzhou hospitals were labelled and used to train a neural network model. Evaluation metrics included mean intersection over union (MIoU), pixel accuracy (PA), and accuracy. Model performance was tested by comparing wound number, maximum dimensions and area extent.
Results: From 1063 images, the model achieved 74% IoU, 88% PA and 83% accuracy for wound bed segmentation. Cohen's kappa coefficient for wound number was 0.810. Correlation coefficients were 0.900 for maximum length (mean difference 0.068 cm), 0.814 for maximum width (mean difference 0.108 cm) and 0.930 for regional extent (mean difference 0.527 cm2).
Conclusion: The model demonstrated exceptional automated estimation capabilities, potentially serving as a crucial tool for informed decision-making in wound assessment.
Implications and impact: This study promotes precision nursing and equitable resource use. The AI-based assessment model serves clinical work by assisting healthcare professionals in decision-making and facilitating wound assessment resource sharing.
Reporting method: The STROBE checklist guided study reporting.
Patient or public contribution: Patients provided image resources for model training.
Keywords: assessment; deep learning; image segmentation; neural network; pressure injury.
© 2025 John Wiley & Sons Ltd.