Page 67 - Read Online

Ganesan et al. Art Int Surg 2024;4:364-75 https://dx.doi.org/10.20517/ais.2024.68 Page 370

[40]
that might require input from clinicians . Similarly, Cai et al. developed a machine-learning model that
used factors including age, surgical procedure, weight, and disease category to predict a patient’s risk of
[41]
pressure ulcers after cardiovascular surgery . Lee et al. created an algorithm to predict nursing home
[42]
patients at risk for pressure ulcers with 81% accuracy . Lustig et al. developed a machine-learning
algorithm for early detection of deep tissue injuries in the heel . They trained the model with a database of
[43]
six consecutive daily measurements of sub-epidermal moisture, which is an established biophysical marker
that can detect pressure ulcer formation . The algorithm resulted in a strong power to predict deep tissue
[43]
injury in the heel the next day, with a sensitivity and specificity of 77% and 80%, respectively . With the
[43]
sensitivity and specificity being moderately high, clinicians may be able to use this tool as a diagnostic guide;
however, clinical decision making must still be applied. Several studies explored an algorithm to identify
certain risk factors associated with diabetic foot ulcer development [44-46] . These models can be applied to
various wound types to predict which patients are most at risk of developing hard-to-heal wounds.
Healthcare providers can allocate resources accordingly and work to prevent those injuries from happening.

Once those injuries do occur, complications are possible, and AI-assisted technologies can predict outcomes
based on an array of inputted data. For example, poorly managed diabetic foot ulcers may result in
amputation. Manual scoring systems help providers determine which ulcers are most at risk, but they do
not capture as much data as AI can, leading to less accurate predictions. Schäfer et al. used machine learning
with certain socioeconomic risk factors, such as household income, ethnic background, and changes in
family status, to predict incidence and amputation risk for diabetic foot ulcers . Xie et al. used
[47]
demographic features, medical and medication history, clinical and laboratory data, and various ulcer
classifications in a machine-learning model to predict which hospitalized patients with diabetic foot ulcers
would undergo amputation . The researchers demonstrated a 0.90, 0.85, and 0.86 predictive ability for
[48]
non-amputation, minor amputation, and major amputation outcomes, respectively . These predictions
[48]
can help providers with wound management and resource allocation.

Challenges with AI in wound care
[49]
Although AI has the potential to significantly impact wound care, it raises several challenges [Figure 2] .
The World Health Organization outlines six challenging yet crucial regulatory areas for thoughtful
implementation of AI in health: transparency and documentation, risk management, data validation with
clear indications of intended use, ensuring unbiased and quality data, safeguarding privacy and data
security, and fostering collaboration among regulatory bodies to secure safe usage of AI .
[50]

Transparency regarding how patient data will be used is crucial for building societal trust. Mitigating risk by
safely integrating AI into clinical practice, training algorithms without bias, and ensuring data quality
control will contribute to safer and more accurate AI. Integrating AI into clinical practice and syncing
patients’ EMRs with data gathered from AI-assisted technology demands time, effort, and appropriate
safeguards to protect patient data. Considering that AI is usually trained with patient data, and AI can better
achieve its goals if trained with a large quantity of high-quality and diverse data, care must also be taken to
protect databases of patient information. Compromised data could deter patients from sharing their
information in the future.

Ensuring that patients feel comfortable sharing data is important for creating diverse databases. Resource-
limited populations are more likely to be excluded from databases, leading to biased outputs. As AI becomes
more capable of improving wound management, care should be taken to implement it in an equitable
manner [Figure 3]. AI-assisted technologies may be costly, creating barriers to resource-constrained
practices. To address these issues, being mindful of regulatory guidelines and collaborating between