Town Hall Details How AI Is Transforming Orthopaedics

Editor’s note: This article is the first in a two-part series covering the Artificial Intelligence (AI) Town Hall at the AAOS 2025 Annual Meeting. Part two will be published in an upcoming issue of AAOS Now.

The AAOS 2025 Annual Meeting hosted a Town Hall to a standing-room-only crowd eager to hear about “The Use of Artificial Intelligence in Orthopaedic Surgery.” Attendees heard several interesting presentations on the evolving landscape of AI and practical applications for use in orthopaedics now and in the future.

“From the Academy standpoint, we see AI as a critical piece of our practices going forward,” said AAOS Immediate Past President Paul Tornetta III, MD, PhD, FAAOS. “It is incumbent on all of us as physicians, as an Academy, and as leaders to understand where the potential positives and potential negatives are, and to think about ethical considerations as we incorporate these things into our practice.”

Dr. Tornetta, director of orthopaedic trauma at Boston Medical Center and professor and chair of the Department of Orthopaedic Surgery in the Chobanian & Avedisian School of Medicine at Boston University, explained that the AAOS Board of Directors decided to invest in understanding AI’s application and utility for AAOS members. The investment supports both the Members and Patients pillars of the new AAOS Strategic Plan.

“We want to empower the integration of this cutting-edge technology,” Dr. Tornetta said. “Everything about AI falls into our Strategic Plan.”

Beginner’s guide to AI
Peter Schilling, MD, MSc, assistant professor of orthopaedics at Geisel School of Medicine at Dartmouth and founder of the Dartmouth Center for AI Research in Orthopedics (CAIRO), views AI as a transformative change for medicine. As food for thought, Dr. Schilling compared AI to the telephone, an invention that was eventually woven into society and the infrastructure of medicine.

“Can you imagine medicine without a phone?” Dr. Schilling said. “AI is medicine’s next telephone.”

AI is still in a dynamic and exploratory phase, he said, but many foundational breakthroughs have already been made. A lot of time, money, and effort have been put into training AI models; now is the time to figure out how to use them in ways that are safe and appropriate.

One exciting example of how AI could advance medicine is AlphaFold, an AI system that predicts a protein’s 3D structure using the primary amino acid sequence.

“AlphaFold has predicted the structure of all 200 million known proteins,” Dr. Schilling explained. “If we didn’t have [AI], this would have taken centuries.”

This breakthrough is important to orthopaedics because treatment of osteoarthritis is still completely palliative, he explained. AlphaFold could play a role in gaining a better understanding of collagen structure, cartilage degradation, and more.

As AI integrates more into life and medicine, Dr. Schilling reminded attendees to keep in mind that at its core, AI is math. It is inputs and outputs. According to Dr. Schilling, these inputs and outputs are a new superpower—but before this new superpower can truly be harnessed, healthcare needs to get the data input flowing.

“Deep learning is akin to building a rocket ship. You need a huge engine and a lot of fuel. The fuel is data,” he said. “Healthcare is still on the launchpad.”

In healthcare, data remain siloed. There is not enough communication, and interoperability is lacking. Much more work needs to be done to harness data in healthcare, and there remains an ongoing discussion about how best to balance the need for data access in AI with privacy concerns.

AI has already arrived
Cody C. Wyles, MD, MS, FAAOS, director of Mayo Clinic Orthopedic Surgery AI Laboratory and a member of the AAOS Committee on Devices, Biologics, and Technology, spoke about some of the potential applications and pitfalls of AI for the practicing orthopaedic surgeon.

One of the exciting potential high-impact applications of AI relates to patient phenotyping. Dr. Wyles walked attendees through the use of AI-powered dislocation and periprosthetic proximal femur fracture risk calculators that can be used to determine patient-specific risk and the degree of risk modification by surgical decisions.

The risk calculator was created with data from 30,000 patients over a 20-year period, and it included non-modifiable risk factors such as demographics and comorbidities, modifiable risk factors such as implant choice, and “mystery” risk factors generated from an AI evaluation of preoperative radiographs. The AI model was able to predict which patients were going to dislocate versus not dislocate with an area under the curve of 0.77.

Dr. Wyles showed attendees the user interface on the calculator where a clinician enters patient variables and uploads a preoperative radiograph. From this information, the model develops a risk matrix with the different implant choices and surgical approaches, graphed with risk for dislocation. “This is highly actionable information for the patient,” Dr. Wyles said.

Another exciting practical application of AI is the use of large language models (LLMs) for automated registry curation. “Registries are the gold standard in orthopaedics,” Dr. Wyles said. “The Academy has wisely put a lot of investment in building out registries at a national level.”

However, it is very intensive to abstract data for registries. “That is where LLMs have the ability to move the needle,” Dr. Wyles said.

A third practical application of AI in orthopaedics is in generative imaging. For example, AI could help surgeons to understand anatomy from various perspectives using only one radiograph. AI enables 3D output from two-dimensional input.

Dr. Wyles showed a demonstration of the RadRotator, a model-based technology that rotates anatomical content of an input radiograph in a 3D space. “It is in the early phases of validation, but so far it is looking like it has extremely high fidelity compared with CT scan,” he said.

Challenges to adoption
Several challenges to AI applications remain though, including regulation. The performance thresholds for clinical use are still being defined in real time. Additionally, AI has a trust issue, Dr. Wyles said, including physician uncertainty and patient skepticism.

“We have an opportunity to enhance care confidence,” Dr. Wyles said. “We need research demonstrating that these models lead to secure outcomes.”

To help that effort, emphasis should be placed on models that use things like uncertainty quantification. These models not only use inputs to make predictions but also can measure how certain or uncertain those predictions are.

Despite the work that remains, Dr. Wyles shared that he is confident that AI will change orthopaedics over the next decade, and that patient phenotyping, LLMs, and generative imaging will be catalytic technologies.

“The pace of change will be too rapid to comprehend,” Dr. Wyles said. “The chance of disruptive change is 100 percent; the chance I can predict it—0 percent.”

Leah Lawrence is a freelance writer for AAOS Now.

Reference

1. Khosravi B, Rouzrokh P, Maradit Kremers H, et al: Patient-specific Hip Arthroplasty Dislocation Risk Calculator: An Explainable Multimodal Machine Learning-based Approach. Radiol Artif Intell 2022;4(6):e220067.