Development of a Method for Clinical Evaluation of Artificial Intelligence-Based Digital Wound Assessment Tools

Area of Science:

• Medical technology
• Digital health
• Wound care informatics

Background:

• Accurate wound assessment, including wound area and percentage of granulation tissue (PGT), is crucial for effective wound management and healing.
• Artificial intelligence (AI) offers potential for enhanced accuracy, consistency, and efficiency in wound assessment workflows.

Purpose of the Study:

• To develop and apply a quantitative and qualitative methodology for evaluating AI-based wound assessment tools against expert human assessments.
• To compare the performance of AI in measuring wound area and PGT with that of experienced wound specialists.

Main Methods:

• A diagnostic study was conducted across two wound centers using 199 deidentified wound photographs.
• AI and human specialists independently traced wound area and PGT; quantitative performance was assessed via error metrics (FNA, FPA, ARE).
• Qualitative assessment involved physician reviewers evaluating AI and human tracings for agreement with standard definitions and PGT estimation variability.

Main Results:

• AI vs. human comparisons showed no statistically significant differences in false-positive area (FPA) and absolute relative error (ARE).
• AI exhibited slightly elevated false-negative area (FNA) compared to human vs. human comparisons, indicating a tendency to underestimate wound boundaries.
• AI-based PGT measurements were within the range of inter-reviewer variability, though human PGT estimates showed considerable variation.

Conclusions:

• The study provides a framework for evaluating AI wound assessment tools, applicable to other wound features and AI diagnostic tools.
• Rigorous validation of AI performance is essential as these tools become more integrated into clinical practice.
• AI tools demonstrate potential for accurate wound assessment, aiding clinicians in guiding patient care.