Page 52 - Read Online

Page 201 Jabbari et al. Art Int Surg. 2025;5:200-9 https://dx.doi.org/10.20517/ais.2024.77

[2,3]
mellitus . Chronic limb-threatening ischemia (CLTI), the most severe form of PAD, carries an estimated
[4]
1-year major limb amputation rate of 22% , and patients with concomitant PAD and diabetes have a four
times higher risk of amputation compared to the national average . In addition, various forms of
[3,5]
oncologic management, severe trauma, and battlefield injuries are affected by LLA. The 5-year mortality
rate for patients with index LLA is reported to be as high as 77% , especially with comorbid diseases such as
[6]
diabetes mellitus. When compared to amputation above the ankle, limb free-flap reconstruction has been
shown to significantly increase the 5-year survival rate (86.8% vs. 41.4%, P < 0.001) .
[7]

The conventional socket attachment of a prosthetic limb presents inherent functional limitations, and many
of these limitations may remain chronic or present despite numerous treatments. Mechanical imbalance can
contribute to difficulty with gait or even increased wear and osteoarthritis on the spine and contralateral
lower extremity . Even with an optimal soft-tissue envelope, changes in strength, tactile feedback, and
[8]
range of motion can be limited. Relative motion between the residual limb and socket may also cause
chronic pain, ulceration, and breakdown .
[9]

Another limiting factor following LLA contributing to decreased prosthetic use, increased rate of surgical
revision or proximal amputation can be the various forms of neurogenic pain following amputation.
Chronic post-amputation pain, including residual limb pain and/or phantom limb pain (PLP), limits
function by interfering with the use of lower limb prosthesis [10-12] . Surgical methods such as targeted muscle
reinnervation and regenerative peripheral nerve interfaces have led to improvements in both amputation-
related pain symptoms and myoelectric prosthetic control [13-17] . Current autonomous lower limb prostheses
can assist in cyclic gait; however, they lack versatility and anticipatory adjustment based on user input [18,19] .
In the last decade, research on myoelectric lower limb prostheses has started to emerge [20,21] , yet the literature
[18]
lacks consensus on the methodology for electromyographic control of lower limb prostheses .

Novel strategies and technologies such as AI and ML are emerging to overcome the distinct challenges faced
by patients with LLA. Herein, we present a scoping review describing how AI and ML can optimize
diagnosis, treatment, and postoperative outcomes among patients with LLA. Further, we aimed to describe
how AI and ML applications can improve peripheral nerve injury outcomes in this population.

EVALUATING THE ROLE OF AI
AI refers to the ability of computer systems to resemble human cognition in learning, synthesis, and
perception of information. ML is a subset of AI in which algorithms can learn from data. ML is driven by
mathematical models that are trained to yield optimized predictions based on a training dataset. There are
two primary methods by which these models are trained. In supervised training, the algorithm learns from
pre-labeled data known as the “ground truth”. In unsupervised training, the input data are not labeled, and
the algorithm autonomously derives meaningful organization from the dataset. A subfield of ML known as
“deep learning” (DL) employs multiple layers of artificial neural networks. This method allows for an
increased level of abstraction and performance via convolutional neural networks (CNN) [Figure 1].

While basic science and translational research is well established in lower extremity amputation care, there
are limited studies elucidating the direct application of AI in the field of LLA. Until recently, most AI
[22]
extremity research had focused on hand and upper extremity amputations , though an understanding of
prior applications can guide efforts to improve LLA outcomes. AI-assisted analysis of medical images is well
established in the literature, including the interpretation of radiographs, electrocardiograms, magnetic
resonance imaging (MRI) slices, and histopathological images [23-26] .