Artificial Intelligence in Orthopedics

ARTICLE SUMMARY:

Artificial intelligence has the potential to drive better outcomes and greater systemic efficiencies in orthopedics, but it is not yet ready for prime-time, according to speakers at AAOS 2024. Excerpted from our recent feature article, AAOS 2024 Takeaways.

Two symposia at the American Academy of Orthopaedic Surgeons (AAOS) 2024 meeting, held in San Francisco in February,?focused on the risks and reward of applying AI to orthopedics for clinical and operational purposes. Orthopedics is still in the early stages of adoption of AI, with the potential to drive better outcomes and greater systemic efficiencies, but it comes with risks and is not ready for prime-time clinical use, said Romil Shah, MD, in opening remarks at the program, Incorporating AI Applications into the Care Delivery Model: A Practical Guide. He pointed to vulnerabilities such as hallucinations and AI’s difficulty answering nuanced questions as near-term impediments to adoption.

The biggest challenge, however, may be convincing physicians to trust AI and integrate AI tools into their clinical workflows, given the enormous marketing and investor hype around the field, Shah noted, estimating that some 75% of MSK venture investments involve use of AI as a key technology differentiator.

While AI is not new to orthopedics, the increase in computer power and availability of data coming from health records and large data registries has accelerated its productivity, said Jacobien Oosterhoff, MD, PhD, an assistant professor of artificial intelligence at Delft University of Technology in the Netherlands. Key applications of AI currently include predictive and prescriptive analytics, wearables, administrative tools, and use case reporting, all of which could be significant for addressing provider shortages.

So, what does that mean for the development of AI applications in healthcare? Supervised machine learning uses structured data sets, which are labeled, with known patient outcomes as the basis of predictive algorithms. The labels are applied by experts, which is time-consuming, and result in simpler, easier to understand algorithms. Unsupervised machine learning does not rely on labeled data, so is less time-consuming to aggregate but the models are more complex and tend to be less accurate and less trustworthy, Oosterhoff explained. Generative AI raises concerns of its own because it is based largely on unsupervised machine learning that requires more complex models to find data within the narrative, so it tends to be less accurate because it is more difficult to understand the underlying mechanisms, she continued.

?Studies comparing the performance of statistical models using machine-learning based on structured data to traditional logistic regression algorithms for predicting the probability of a specific fracture or post-surgery adverse event in MSK trauma found that both produced similar results, but traditional statistics are simpler to use and just as powerful.?The study was published in The Journal of Bone and Joint Surgery (JBJS) in 2021.

Wearables that measure not just what the body is producing, but proactively send energy signals into the body—light, electrical energy, acoustics—are at the heart of research in the laboratory of Joseph Schwab, MD, director of spine oncology and the Center for Surgical Innovation and Engineering at Cedars Sinai. (See “What Mass General Brigham Expects to Achieve Through AI-Driven Orthopedics/ Spine Surgery,” MedTech Strategist, January 18, 2022.) Changes in energy levels can be assessed by machine-learning algorithms and help to diagnose subtle changes, he noted. “These are incredibly complex signals, and this is where machine learning excels.” As an example, he cited a patient with asymptomatic cervical stenosis due to a traffic accident. Surgeons deciding whether to operate currently may take a wait-and-see approach, but AI may be able to aid their decision-making.

Schwab is also using AI to assess and quantitate balance by collecting data points from patients who pose in different positions with wearables attached to their calves to gauge their muscle activity. The results are hard to see with the naked eye, but AI can decipher patterns of signal activity that indicate differences between healthy persons and ones with balance problems.

Continue reading AAOS 2024 Takeaways here.

Trial MyStrategist.com and?unlock 5-days of exclusive,?subscriber-only access?to the medical device industry's most trusted strategic publications: MedTech Strategist & Market Pathways.

Stay informed and get bonus content with our community newsletter by signing up here.