Page 108 - Read Online

Page 54 Ambati et al. Art Int Surg. 2025;5:53-64 https://dx.doi.org/10.20517/ais.2024.45

parallel, recent advances in artificial intelligence (AI) and machine learning (ML) have already touched
nearly every facet of modern life, from industry and transportation to the arts and music. AI and ML rely on
large datasets to recognize patterns in data and, when properly deployed, can perform specialized tasks
more quickly and accurately than humans. Due to these promises, spine surgeons are looking to AI/ML to
usher in the next generation of technical advancements for their patients.

We are currently at an inflection point for the impact of AI/ML in spine surgery. The number of studies on
healthcare AI applications continues to grow exponentially , a trend that is reflected in spine surgery as
[4]
[6]
[5]
well . Furthermore, national funding agencies have established priorities in healthcare AI , and the Food
and Drug Administration has gained experience in regulating these tools . Simultaneously, private venture
[7]
[8]
funding has grown substantially in healthcare AI , but rates of progress have not been equal across
applications . In particular, AI diagnostic tools in radiology and pathology have grown faster than other
[4]
areas of medicine , such as the surgical specialties. A better understanding of the landscape of AI/ML in
[4]
spine surgery could help bridge the gap between research and commercialization of such tools.
Currently, disorders of the spine are among the major contributors to both healthcare costs and disability
both in the United States and worldwide [9,10] . Spine surgery is a major contributor to healthcare
spending [11,12] , and while safety has improved significantly year over year , when complications do occur,
[13]
[14]
they cause a substantial impact on patient quality of life and healthcare spending . Due to the wide range
of variables that influence patient selection, preoperative planning, intraoperative technique and decision
making, opportunities for the potential impact of AI and the possible challenges are high. AI-based tools
that could reliably make advances in efficiency, technical proficiency, or complication minimization would
have immense clinical and economic impact.

This article provides a targeted primer on AI/ML algorithms and critically reviews select applications of AI/
ML to spine surgery. We highlight those that aim to assist in pre-surgical planning, intraoperative
assistance, and prediction of postoperative course. These tools span a spectrum of development and
commercial deployment, employ a variety of data sources, and interface with clinicians and patients in a
number of ways. Through this narrative review, we identify a set of shared challenges facing the field,
namely the substantial heterogeneity in patients with spinal disorders, the uncertainty and subjectivity in
outcome measures, and the quality and quantity of data available for algorithm development. Finally, we
propose solutions to these challenges, which we hope can forecast a path toward incorporating robust AI/
ML tools in spine clinics and operating rooms (ORs), thereby achieving the best outcomes for future patients.

AN AI/ML PRIMER FOR SPINE SURGEONS
AI is a group of computational approaches that aim to provide human-level expertise and decision making
and predominantly rely on ML, a class of powerful statistical models that recapitulate in silico different
facets of human sensory processing and cognition, ranging from vision and language to estimation and
prediction tasks [15,16] . While the technical aspects of AI and ML are beyond the scope of this review, we
discuss several key concepts that all spine surgeons should familiarize themselves with, as these technologies
continue to play an increasing role in our field.

Traditional ML algorithms adapt structured formulas to relate input and output variables and generate
future predictions. Common types of these algorithms include logistic regression, decision trees and
[15]
random forest, and support vector machines . They differ based on the types of input and output variables
they can handle, as well as their ability to process noisy and non-linear data, which are prevalent in
healthcare applications. The other major class of models is known as ‘deep learning’, which accounts for the