Deep learning-based pressure injury staging: a multicentre study involving 59 hospitals

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Deep learning for medical diagnostics

Background:

• Accurate pressure injury staging is crucial for effective patient care and reducing healthcare burdens.
• Developing reliable AI tools can significantly improve the accuracy and efficiency of pressure injury assessment.
• This study focuses on creating a deep learning model for pressure injury recognition and its translation into a practical application.

Purpose of the Study:

• To develop and evaluate a deep learning-based model for accurate pressure injury recognition and staging.
• To translate the best-performing AI model into a preliminary smartphone application for clinical use.

Main Methods:

• A multicentre retrospective study involving 1903 pressure injury images from 59 hospitals.
• Evaluation of three AI models: Mask R-CNN with ResNet-18, Mask R-CNN with Swin Transformer, and Segmenting Objects by Locations v2.
• Performance assessment using metrics like mean average precision (mAP), average precision (AP), and average recall (AR).

Main Results:

• The Mask R-CNN model with Swin Transformer achieved the highest performance (mAP = 0.894).
• This model outperformed other evaluated AI models in pressure injury staging accuracy.
• The best-performing model was successfully integrated into a smartphone application.

Conclusions:

• The developed deep learning system demonstrates promising performance for pressure injury staging.
• This AI tool can potentially support clinical decision-making in pressure injury management.
• Further validation with larger, diverse datasets is recommended to enhance clinical applicability.