Page 66 - Read Online

Page 369 Ganesan et al. Art Int Surg 2024;4:364-75 https://dx.doi.org/10.20517/ais.2024.68

patient needs repositioning, enhancing effective healing.

Several AI-assisted technologies already aim to reposition patients automatically. Ni et al. developed an AI
mattress that utilized three-dimensional InterSoft technology to detect bony prominences and redistribute
[34]
pressure . They studied this mattress on a 79-year-old male with left-sided hemiplegia, a need for
positional changes, a sacral ulcer measuring 2.5 × 2 × 3 cm at five months of standard treatment, and an
3
[34]
overall Braden score of 12, indicating a high risk of pressure injuries . The AI mattress had an active
pressure sensory array that ensured correct positioning and calculated pressure every second . It used
[34]
results from a first-time scan as a baseline to locate areas of bony prominences and employed a color-coded
scheme to indicate areas at the highest risk for pressure injury . As a result, the mattress redistributed
[34]
pressure off those highest-pressure areas . After four weeks of AI mattress usage, the wound measured
[34]
shrunk in size, the patient reported more comfort, and he had healthier tissue types .
[34]
Recent advances in AI have also been employed through other methods to facilitate wound healing. One
example includes the AI bandage. Kalasin et al. created a smart bandage with a flexible sensor and deep
neural network algorithm . The bandage has MXene, a new class of graphene-like two-dimensional
[35]
[35]
transitional metal carbon, which enhances its conductivity and sensory capabilities . It also has a wound
dressing made of poly(vinyl acrylic) gel combined with polyaniline that can react to the wound’s pH
[35]
level . The wound dressing generates a voltage that responds to changes in pH, indicating different stages
of healing . The deep learning network processes the voltage to classify the wound’s healing stage with 95%
[35]
[35]
accuracy . Healthcare professionals can make informed decisions about wound treatment based on the
data.
AI in wound prognosis
Chronic wounds are hard-to-heal due to a complex web of contributing factors including
immunocompromise, poor blood circulation, and chronic inflammation, resulting in often bleak prognoses.
Accurate prognosis requires comprehensive data collection, and researchers have applied AI to this
challenge. Topaz et al. developed a natural language processing application that selected detailed wound
information from free text clinical notes, gathering comprehensive data on wound comorbidities, risk
factors, and underlying contributors . With a strong ability to extract relevant data, AI can be leveraged to
[36]
predict outcomes and prognoses. Robnik-Sikonja et al. used machine learning to analyze the effects of
wound, patient, and treatment attributes on wound healing rates . Ngo et al. studied how machine
[37]
learning could use textural features from thermal images of venous leg ulcers to predict delayed healing
[38]
outcomes . They achieved a 79% sensitivity and 60% specificity with a Bayesian neural network . With
[38]
moderate sensitivity but mild specificity, clinicians will still need to rule out false positives to avoid
unnecessary treatment.

Chen et al. similarly used AI to assess images of pressure ulcers for tissue changes, wound stages, and
healing conditions . These researchers aimed to provide clinicians with valuable information to guide
[39]
treatment decisions and resource allocation.

With the creation of AI-assisted technologies like mattresses and bandages, appropriate resource allocation
becomes crucial. Following the ethical principles of justice, resource allocation typically prioritizes those
with the most dire conditions or those who stand to benefit the most and avoid the worst prognoses. Studies
have examined wound healing prediction rates, but AI can also predict wound incidence. Alberden et al.
created a machine-learning model to predict the development of pressure ulcers among surgical critical care
[40]
patients . The predictions are made from data in the patient’s EMR, differentiating it from other models