Improving the Accuracy of Diagnostic Imaging using Artificial Intelligence: A Method for Assessing Necrotic Tissue in Pressure Injury.

Abstract

Background: Accurate assessment of pressure injuries is critical in clinical settings, especially when evaluating necrotic tissue using the DESIGN-R® scale widely adopted in Japan. This study aimed to integrate artificial intelligence (AI) into the evaluation process to enhance diagnostic consistency and accuracy. By leveraging deep learning and convolutional neural networks, we explored the potential of AI models in classifying necrotic tissue from wound images.

Methods: A retrospective observational study was conducted using electronic medical records and wound photographs from patients treated at Tottori University Hospital between 2014 and 2022. Two supervised learning models were developed: a Categorical Classification Model (CCM) for multi-class prediction, and a Binary Classification Model (BCM) implementing a two-step binary classification. Necrotic tissue was categorized based on the DESIGN-R® scale into three classes: n0 (no necrosis), N3 (soft necrosis), and N6 (hard, adherent necrosis). The models' performance was evaluated using standard classification metrics.

Results: The CCM showed recall rates of 0.7824 for n0, 0.6620 for N3, and 1.0000 for N6. In contrast, the BCM achieved higher recall rates: 0.9074 for n0, 0.9884 for N3, and 1.0000 for N6. Overall metrics for CCM were: accuracy 0.8148, precision 0.8166, and F-1 score 0.8089. The BCM surpassed these with an accuracy of 0.8711, precision 0.8418, and F-1 score 0.8508. Across all performance indicators, the BCM demonstrated superior classification capability.

Conclusion: The study demonstrated that AI, particularly the binary classification approach, can enhance necrotic tissue assessment in pressure injury evaluation. The BCM consistently outperformed the CCM, supporting its potential as a reliable tool to assist clinicians in objective and standardized pressure injury evaluation using the DESIGN-R® framework.