Accurate AI-Based Characterization of Wound Size and Tissue Composition in Hard-to-Heal Wounds

Background: Detailed assessments, documentation, and evaluation of the wound characteristics in hard-to-heal wounds are essential for optimizing and individualizing wound care. However, the remaining challenge in clinical care includes the lack of high accuracy and precision tools for automated wound size (surface area and depth assessment) and a wound bed evaluation, i.e., a qualitative and quantification assessment of slough and necrosis. Objective/Methods: This study evaluates the accuracy and precision of the AI-powered technique, SeeWound© 2, compared to digital planimetry for a wound surface area and a wound bed characterization (slough and necrosis) in "in vitro" models and in patients, and a probe for depth, including diabetic foot ulcers, venous ulcers, pressure ulcers, and ischemic ulcers. Results: The data show that accuracy and precision (SeeWound© 2) for the wound surface area, the depth, and the wound bed characterization (slough and necrosis) were accuracy 96.28% and 90.00%, (CV 5.56%), respectively (wound size); 90.75% and 89.55%, (CV 3.07%), respectively (wound depth); 80.30% (slough) and 84.73% (necrosis) and 93.51% (slough) (CV 4.15%) and 82.35% (CV 8.34%) (necrosis). The precision for the digital planimetry was 88.61% (CV 7.00%) (slough) 85.74% (CV 7.54%) (necrosis). Conclusions: The overall accuracy and precision of the AI model in identifying wound size and depth were close to 90%, except for the accuracy and precision for slough and necrosis, where levels around 80% were achieved when compared to digital planimetry. The findings for the wound surface area and depth assessments, together with quantification of slough and necrosis, suggest that the SeeWound© 2 model can offer significant clinical benefits by improving documentation and supporting decision-making in wound management.