What is AI?
Artificial intelligence (AI), or machine learning, is a process by which computers use existing data or definitions in a databank to construct reality rather than executing exact instructions from human input. AI has several distinct classifications, including machine learning (ML) and large language models (LLM). ML is a subset of AI that uses mathematical algorithms which enable a machine to choose, classify, or predict without external human influence. Deep learning (DL) models contain many layers of data compilation to build a neural network architecture and are trained with a large set of labeled data. AI recognizes patterns in data, makes decisions about the data, and solves problems with the data it selects for applicability. “The predictive power and accuracy of a machine-learning algorithm is only as powerful as its training experience and volume, not unlike the expertise and judgment of a sports medicine specialist.”15 AI utilization in orthopedics and sports medicine is likely already part of your practice and will have increasing utilization in sports medicine.
Deep learning models of AI and ML are used in medical applications and protect patient personal health information, proprietary information, and other data. Because the data and definitions accessible to the AI tool are restricted to the databank built from known human intelligence, risk of inappropriate or inaccurate results is unlikely. AI learns from existing data, and ongoing data added to this databank builds the accuracy and specificity of the algorithms.
Areas in which AI is applicable to orthopedic surgery include:
- Diagnostic Assistance
Orthopedic surgery is heavily dependent on diagnostic imaging, and AI can be trained to identify and recognize patterns in images. AI algorithms can analyze and detect abnormalities including fractures, tumors, sports injuries such as ACL tears5 or meniscal tears, and rotator cuff tears11. AI has been used to assist in implant identification and evaluation of implants for positioning and complication risk, including component loosening and periprosthetic joint infection.18 While early AI models focused on a single abnormality such as presence or absence of ACL tears, newer generation algorithms can use imaging to predict treatment success, assess the surgical repair or reconstruction, and quantify reinjury risk.14 These trained AI models were found to be more accurate than radiology residents and comparable in accuracy to musculoskeletal fellowship trained radiologists.16
What’s Available Now:
AI-assisted templating for implant placement is already commonplace and evolving. If your templating software automatically generates a starting position for implants, AI is at work. Implant identification is another area in which AI excels, demonstrating up to 99% sensitivity and specificity in a fraction of the time it would take a surgeon to determine the type of implant.10 Commercially available software is available now to assist with these tasks.
- Surgical Support
Orthopedic surgery has increasingly embraced robotic assistance for accuracy in the operating room. Computer-aided techniques have demonstrated superiority over conventional techniques in both consistency and accuracy.9 Robotic systems are used to create precise bone cuts for total hip and total knee arthroplasty, and for higher accuracy of pedicle screw placement and reduced radiation exposure in spinal surgery.8
Augmented reality (AR) refers to visualization of the real world with an additional virtual information overlay; mixed reality (MR) allows the user to interact with virtual elements as if they are real. The user’s view can be augmented with a monitor or optical see-through system. Current techniques require complex registration and validation of the AR navigation system to the patient’s anatomy at the time of surgery. This application is less attractive because it can significantly slow OR efficiency and workflow. Another concern is that increasing use of AI and automation in the operating room will negatively impact surgeon aptitude in training, with less hands-on work for residents and fellows in the OR.2 However, the greater accuracy and precision, reduction in radiation, and potential for efficiency in complex cases will lead to adoption for applicability in the orthopedic OR.4,17
What’s Available Now:
Many major manufacturers offer robotic-assisted joint replacement systems, including the Stryker Mako, CORI from Smith+Nephew, and ROSA from Zimmer Biomet. Zimmer Biomet also has introduced “OmniSuite,” an intelligent operating room that collects real-time data on key surgical events and optimizes workflow. Stryker offers MR for shoulder arthroplasty, Enovis offers AR for hip and knee arthroplasty, and Zimmer Biomet has used OptiVu as an option for remote education and training. Many smaller companies also are coming online with options for augmenting the experience in the OR.
- Clinical Decision Support and Predictive Analytics
AI can be utilized to assist the treating physician in generating a differential diagnosis, selecting the most likely diagnosis, and recommending additional tests or imaging that would be helpful in evaluation or management of orthopedic patients. AI also can predict the likelihood of disease based on medical history, lifestyle data, and genetic information. In orthopedics, ML has been utilized to assist in predicting return to play after hamstring injury in elite soccer players.17 AI has predicted outcomes after surgery and assessed patient risks for readmission,12 and the development of complications. ML is being developed and trained to create online clinical prediction tools to predict the risk of retear in patients undergoing rotator cuff repair.1
ML will be utilized to provide personalized recommendations for orthopedic care. While surgeons with many years of experience in practice can identify patients who are likely to not respond well to surgery or to reach the minimum clinically important difference, ML can be utilized to identify biologic and psychological factors that indicate whether surgical intervention will or will not be helpful for a particular patient.13
What’s Available Now:
ChatGPT can be utilized to generate a differential diagnosis from inputting patient history and exam findings and suggest next tests or additional examination maneuvers.
- Research and Data Analysis
AI can quickly analyze vast amounts of healthcare data to identify trends, correlations, and insights that can inform public health strategies and medical research, correlations that would take a large amount of time and resources using traditional research methods. There has been interest in applying explainable AI (XAI) to research and clinical applications in orthopedics. Explainability refers to the ability for clinicians, researchers, and patients to understand how the AI algorithms make decisions and recommendations. This helps to characterize model accuracy, transparency, and outcomes when utilizing AI in healthcare.
What’s Available Now:
Multivariable logistic regression is a form of AI that is routinely used in research to evaluate a complex set of interactions between variables of interest. Bayesian inference (which uses prior knowledge in the form of a prior distribution to estimate posterior probabilities) also can be used in research and can form the basis for training a model in AI. AI can be utilized to assist in writing scientific manuscripts to improve readability, but should not be used to draw conclusions or provide clinical recommendations. Use of AI should be disclosed when submitting a work for publication.
- Operational Efficiency
AI has a role in repetitive administrative and clinical work, including medical record keeping and documentation. LLMs such as ChatGPT can be utilized for writing tasks such as generating letters of medical necessity, drafting responses to patient questions, or providing medical instructions. Voice-to-text software utilizes AI to transcribe, and AI-powered “scribes” can listen to (but not record) a physician-patient interaction and create appropriate medical documentation. These ambient AI scribes will filter out small talk and actually learn from their mistakes, with corrections providing valuable feedback to the AI algorithm.
AI also can be used as a tool to streamline administrative tasks such as scheduling, billing, and coding, reducing the workload on healthcare staff and minimizing errors. AI-powered chatbots and virtual assistants can provide patients with 24/7 support, answering questions, scheduling appointments, and offering reminders. AI-based coding utilizes the information in the surgical dictation, the myriad coding rules, and AI’s self-assessed confidence in its decision, sending only low confidence cases for human review.
What’s Available Now:
Many different software and practice management vendors offer dictation software, chatbot functions, scheduling, billing and coding, as well as virtual reminders. These products are HIPAA compliant. Examples include DeepScribe, Freed, and AutoNotes. Coding assistance software includes companies like CodaMetrix, PayrHealth, and XpertDox. Companies like Hyro, Veradigm, and SmartAction provide automated predictive scheduling and accommodate for complex data needed to create an optimal schedule.
- Patient Engagement and Support
AI can create images, videos, and text that can be used in patient or surgeon education (Figure 1). Both Google Bard and ChatGPT have been found to generate easy-to-read and accurate patient education for orthopedic sports medicine patients.6 Patients can interact with chatbots for 24/7 support. Many people now are wearing devices to track steps, fitness goals, sleep, and other health metrics. Wearable technology can be leveraged in the management of orthopedic sports medicine patients for rehabilitation after injury and recovery after surgery. VR, AR, gamification, and telerehabilitation have been used to provide remote, virtual techniques for orthopedic rehabilitation after surgery.3 AI is essential for analyzing and utilizing the vast amounts of data generated by wearable devices.
What’s Available Now:
Zimmer Biomet has the Mymobility platform which integrates with the Apple watch and monitors postoperative progress. Patients can receive text messages and postop instructions. “WalkAI” generates personalized daily predictions for recovery and notifies surgeons if patients are off track with lower step counts or slower gait speed. Stryker has “Recovery Coach” to engage and educate patients postoperatively. Devices are being developed for sports-specific applications, including the motus BASEBALL sleeve that measures throw counts, torque, speed, and workload.
Pitfalls
AI has been known to generate “hallucinations,” also known as confabulations or delusions, which are false or misleading information that is presented as a fact. These caveats present ethical dilemmas when utilizing AI in healthcare, and include privacy, equity and bias, transparency and trust, cybersecurity, and corporate responsibility.7 Physicians are ultimately responsible for all aspects of patient care including documentation, billing and coding, patient education and communication. Navigating these pitfalls while advancing orthopedics is crucial to ensuring that AI benefits all patients without compromising their rights or well-being and protects healthcare providers from malpractice.
Summary
AI has significant potential to improve diagnostic accuracy, assist in orthopedic sports medicine surgery and rehabilitation, improve clinical outcomes, streamline operational efficiency, and proactively engage with patients.
If you haven’t already tried AI in practice, consider where you are already integrating it into your life.
- Do you ask Siri for directions?
- Do you ask Alexa about the weather?
- Do you use Google home?
- Do you send voice-to-text to create messages?
- Do you use commercial computer programs for preoperative planning?
- Do you perform robotic surgery?
If you haven’t experimented with commercially available AI, consider trying a few simple uses.
- Try using ChatGPT to generate a draft of a letter of medical necessity.
- Browse the latest robotics and AR/MR applications available for the procedures you most often perform.
- Use DALL-E to generate a patient education image.
- Ask ChatGPT to generate a differential diagnosis.
- Evaluate patient engagement options for postoperative recovery.
References
1. Allaart LJH, Spanning S van, Lafosse L, Lafosse T, Ladermann A, Athwal GS, et al. Developing a machine learning algorithm to predict probability of retear and functional outcomes in patients undergoing rotator cuff repair surgery: protocol for a retrospective, multicentre study. BMJ Open. 2023 Feb 10;13(2):e063673. doi:10.1136/bmjopen-2022-063673
2. Beane M. How AI could keep young workers from getting the skills they need. Wall Street Journal. 2024 Jul 26
3. Berton A, Longo UG, Candela V, Fioravanti S, Giannone L, Arcangeli V, et al. Virtual reality, augmented reality, gamification, and telerehabilitation: psychological impact on orthopedic patients’ rehabilitation. J Clin Med. 2020 Aug 7;9(8):2567. doi:10.3390/jcm9082567
4. Casari FA, Navab N, Hruby LA, Kriechling P, Nakamura R, Tori R, et al. Augmented reality in orthopedic surgery is emerging from proof of concept towards clinical studies: a literature review explaining the technology and current state of the art. Curr Rev Musculoskelet Med. 2021 Feb 5;14(2):192–203. doi:10.1007/s12178-021-09699-3
5. Chang PD, Wong TT, Rasiej MJ. Deep learning for detection of complete anterior cruciate ligament tear. J Digit Imaging. 2019 Dec 11;32(6):980–986. doi:10.1007/s10278-019-00193-4
6. Giorgino R, Alessandri-Bonetti M, Del Re M, Verdoni F, Peretti GM, Mangiavini L. Google Bard and ChatGPT in orthopedics: which is the better doctor in sports medicine and pediatric orthopedics? The role of AI in patient education. Diagnostics. 2024 Jun 13;14(12):1253. doi:10.3390/diagnostics14121253
7. Jeyaraman M, Balaji S, Jeyaraman N, Yadav S. Unraveling the Ethical Enigma: Artificial intelligence in healthcare. Cureus. 2023 Aug 10;doi:10.7759/cureus.43262
8. Jia S, Weng Y, Wang K, Qi H, Yang Y, Ma C, et al. Performance evaluation of an AI-based preoperative planning software application for automatic selection of pedicle screws based on computed tomography images. Front Surg. 2023 Sep 11;10. doi:10.3389/fsurg.2023.1247527
9. Jud L, Fotouhi J, Andronic O, Aichmair A, Osgood G, Navab N, et al. Applicability of augmented reality in orthopedic surgery – A systematic review. BMC Musculoskelet Disord. 2020 Dec 15;21(1):103. doi:10.1186/s12891-020-3110-2
10. Karnuta JM, Luu BC, Roth AL, Haeberle HS, Chen AF, Iorio R, et al. Artificial intelligence to identify arthroplasty implants from radiographs of the knee. J Arthroplasty. 2021 Mar;36(3):935–940. doi:10.1016/j.arth.2020.10.021
11. Lin DJ, Schwier M, Geiger B, Raithel E, von Busch H, Fritz J, et al. Deep learning diagnosis and classification of rotator cuff tears on shoulder MRI. Invest Radiol. 2023 Jun;58(6):405–412. doi:10.1097/RLI.0000000000000951
12. Merrill RK, Ferrandino RM, Hoffman R, Shaffer GW, Ndu A. Machine learning accurately predicts short-term outcomes following open reduction and internal fixation of ankle fractures. The Journal of Foot and Ankle Surgery. 2019 May;58(3):410–416. doi:10.1053/j.jfas.2018.09.004
13. Milella F, Famiglini L, Banfi G, Cabitza F. Application of machine learning to improve appropriateness of treatment in an orthopaedic setting of personalized medicine. J Pers Med. 2022 Oct 12;12(10):1706. doi:10.3390/jpm12101706
14. Palermi S, Vittadini F, Vecchiato M, Corsini A, Demeco A, Massa B, et al. Managing lower limb muscle reinjuries in athletes: from risk factors to return-to-play strategies. J Funct Morphol Kinesiol. 2023 Nov 6;8(4):155. doi:10.3390/jfmk8040155
15. Ramkumar PN, Luu BC, Haeberle HS, Karnuta JM, Nwachukwu BU, Williams RJ. Sports medicine and artificial intelligence: a primer. Am J Sports Med. 2022 Mar 26;50(4):1166–1174. doi:10.1177/03635465211008648
16. von Schacky CE, Wilhelm NJ, Schäfer VS, Leonhardt Y, Gassert FG, Foreman SC, et al. Multitask deep learning for segmentation and classification of primary bone tumors on radiographs. Radiology. 2021 Nov;301(2):398–406. doi:10.1148/radiol.2021204531
17. Valle X, Mechó S, Alentorn-Geli E, Järvinen TAH, Lempainen L, Pruna R, et al. Return to play prediction accuracy of the MLG-R Classification System for hamstring injuries in football players: a machine learning approach. Sports Medicine. 2022 Sep 24;52(9):2271–2282. doi:10.1007/s40279-022-01672-5
18. Yi PH, Mutasa S, Fritz J. AI MSK clinical applications: orthopedic implants. Skeletal Radiol. 2022 Feb 5;51(2):305–313. doi:10.1007/s00256-021-03879-5
