Page 55 - Read Online

Jabbari et al. Art Int Surg. 2025;5:200-9 https://dx.doi.org/10.20517/ais.2024.77 Page 204

improve predictive models, as evidenced by the balanced bagging classifier with a F1 score of 0.80 and an
AUC of 0.94. This model identified several important features, such as postoperative day 1 opioid use, body
mass index, age, preoperative opioid use, prescribed opioids at discharge, and hospital length of stay. The
identification of high-risk patients may guide clinical decisions and interventions.

Another postoperative challenge specific to amputation patients is the development of pain or sensation
that originates from the absent, amputated limb, known as PLP. Ortiz-Catalan et al. showed that motor
execution of the phantom limb via ML, augmented and virtual reality, and gaming may hold potential as a
[42]
treatment for PLP . Their cohort included fourteen patients with upper limb amputation and chronic
intractable PLP. After the 12-session study period, a comparison of pre- and post-treatment PLP
demonstrated significant decreases by 47% (P = 0.001) for weighted pain distribution, 32% (P = 0.007) for
the numeric rating scale, and 51% (P = 0.0001) for the pain rating index. These findings further exemplify
the potential role of AI applications in the evolution of treatment options for LLA patients.

AI in lower extremity nerve injuries
There is a need for innovative peripheral nerve injury strategies among LLA patients, as neurogenic pain
secondary to hyperactive terminal neuroma formation is largely responsible for postoperative morbidity. In
this effort, AI technologies can be used to understand the pathology of PNI and to better explore
therapeutic approaches.

Such an approach has led to the development of new research methods and strategies for nerve
regeneration. Romeo-Guitart et al. showed the power of therapeutic performance mapping system (TPMS)
technology for the design of drug therapies promoting nerve regeneration and functional recovery after
[43]
PNI . TPMS develops mathematical models that simulate human physiology in silico, a process that is
based on AI and pattern recognition models that source all available biological, medical, and
pharmacological knowledge. A total of 5,400 drugs were screened, generating approximately 15 million
binary drug combinations. After further screening, the team selected the top 3 binary combinations with
more than 75% of potential regenerative capabilities. The neuroprotective effects of these drug combinations
were then validated in in vitro and in vivo models. This strategy elucidated the therapeutic actions of
combinatorial drug therapy with acamprosate plus ribavirin. Most importantly, the authors demonstrated
the discovery of repurposed drug therapies with a network-centric approach, which uses ML tools to
validate both efficacy and mechanism of action with preclinical in vivo models.

Additionally, large image datasets can be utilized by AI systems for rapid biomedical research. Daeschler
et al. validated a DL model of automated segmentation and histomorphometry of myelinated peripheral
nerves via light microscopic images . A CNN was trained for automated axon and myelin segmentation
[44]
using a dataset of light-microscopic cross-sectional images of rat nerves at various stages of axonal
regeneration. Their CNN model demonstrated high pixel-wise accuracy for nerve fiber segmentation with
ground truth overlap (mean ± standard deviation) of 0.93 ± 0.03 and 0.99 ± 0.01 for axons and myelin
sheaths, respectively. Nerve fibers were identified with high sensitivity (0.99) and precision (0.97), with
automated histomorphometry reducing analysis time to less than 2.5% of that for manual morphometry.
Neural network-powered biomedical image analysis can significantly increase the rate of experimental nerve
research via performance, time, and resource efficiency.

Beyond its role in drug therapy and image processing, AI has potential applications in the direct repair of
PNI using 3D printing and biomaterials. Nerve guidance conduits (NGCs) have been widely explored for
the treatment of PNI. Current research on functional NGCs attempts to create microenvironments that