Source: Department of Orthopedics, PLA General Hospital
Author: Li Haifeng, Chai Wei
Generative AI is an artificial intelligence technology that can autonomously generate text, images, audio, and other content. The evolution of generative AI is a process of continuous iteration and innovation. In 1972, IBM's Watson Lab built the first human language model, which was upgraded to the second generation in 1993. In 2010, Jeff Dean et al. of Google implemented the first truly practical deep learning system, Google Brain, and used it to build the third generation of language models. In 2018, OpenAI released the GPT (Generative Pre-trained Transformer) series of models, which have achieved great success in natural language understanding and generation. In 2020, the release of GPT-3 further advanced the use of generative AI in the field of natural language processing, which is capable of generating coherent, logical text and excels at multilingual tasks. With the improvement of machine computing power and the optimization of algorithms, generative AI has gradually made breakthroughs. As soon as ChatGPT-4 was released in 2023, it immediately demonstrated its potential to understand, generate, and process multiple data types such as images, text, and audio [1,2], and was rapidly introduced into various fields such as art creation, design, education, and medical care [3]. In the past year or so, generative AI has also begun to be applied to all aspects of the field of orthopedic surgery, and has gradually become a research hotspot. This article attempts to summarize the application status of generative AI, especially GPT-4, in orthopedic education, clinical decision support, patient communication guidance, disease diagnosis and treatment, etc., and discusses its challenges and possible future development directions.
1. The application of generative AI in orthopedic education
The application of generative AI in orthopedic education is gradually changing the way medical students and orthopedic surgeons learn, further improving the quality and efficiency of education. Generative AI can help produce teaching materials, such as creating virtual cases and simulating real-world clinical scenarios, to assist medical students and residents in improving their clinical thinking and decision-making skills, and medical students can practice clinical decision-making, diagnosis, and treatment plan development by processing these virtual cases. Medical students can also practice surgical operations in these simulated environments, improving surgical skills without the need for real patient involvement. Generative AI can also help students and physicians improve the efficiency of their writing and exams, and can even be used as an interactive learning tool in orthopedic residency training.
Lum examines ChatGPT's performance in response to orthopaedic In-Training Examination (OITE) questions. To avoid bias, 193 (48% of questions) were excluded, and the remaining 212 text-based questions were answered by ChatGPT with orthopedic residents of different levels, and it was found that ChatGPT could answer 47% of the questions accurately, and, as the complexity of the questions increased, ChatGPT also performed better. The authors believe that ChatGPT performs on par with orthopedic residents and that ChatGPT can be used as an additional tool to assist in orthopedic learning education [4].
Kung et al. used ChatGPT 3.5 and GPT-4 versions to test OITE questions (questions that do not contain images) in 2020, 2021, and 2022 without using any prompts. The results of the study showed that ChatGPT answered 3.5 out of 360 questions correctly (196%) (54.3%), which is equivalent to the level of the first year of residency (PGY-1). ChatGPT 3.5 cites verifiable sources, including journal articles, books, or websites, when answering questions, with an average journal impact factor of 5.4. GPT-4 answered 265 questions correctly (73.6%), which is equivalent to the average of the fifth year of residency (PGY-5) and exceeds the passing score of 67% on the American Board of Orthopaedic Surgery Part I exam. GPT-4 cites verifiable sources when answering questions, with an average journal impact factor of 5.2. According to the authors, ChatGPT outperformed the average PGY-1 level on standardized tests, while GPT-4 outperformed the average PGY-5 level, making significant improvements. As a result, generative AI designs can be tested and evaluated in a standardized manner to ensure their impartiality and consistency. AI can even be used to analyze students' learning progress and comprehension to generate personalized learning plans and resources, and AI can adjust the teaching content and difficulty based on student feedback and test results to ensure that each student can learn at their own pace. Students' test results can also be analyzed to provide personalized feedback and suggestions for improvement [5].
2. Application of generative AI in orthopedic clinical decision support
Orthopedic diseases include a wide range of pathological types, including fractures, joint diseases, spinal deformities, and sports injuries, and their diagnosis and treatment decisions often face complex challenges. Generative AI has the potential to be used as a clinical decision support tool to provide real-time decision support to doctors and help doctors make more accurate judgments in complex clinical situations. For example, generative AI can identify lesions such as fractures, deformities, bone diseases, etc., and help clinicians make decisions in terms of diagnosis, differential diagnosis, and treatment selection by rapidly analyzing relevant information such as patient symptoms, medical history, and imaging results, and can also recommend diagnostic tests or appropriate imaging modalities for further evaluation. Generative AI can also use machine learning algorithms to personalize treatment based on a patient's specific circumstances, such as age, gender, health status, and lifestyle. This can improve the effectiveness of treatment and reduce unnecessary interventions.
In addition, treatment recommendations for many orthopaedic conditions are often based on evidence-based guidelines and clinical experience. Generative AI can be optimized by assisting clinicians in conducting literature searches, summarizing research articles, and identifying key findings, and providing updated treatment recommendations based on the specific characteristics of the patient and their condition. This process helps optimize treatment options, promotes adherence to evidence-based practices, and reduces variability in clinical decision-making. Generative AI can also provide clinical decision support by analyzing patients' electronic health records (EHRs). For example, by summarizing a patient's chief complaint history, laboratory test results, and imaging reports, AI can assist doctors in quickly obtaining key information, thereby improving the efficiency of diagnosis and treatment. The treatment of many orthopedic conditions involves decisions about surgical and non-surgical treatments. By analyzing large amounts of clinical data and literature, generative AI is able to provide evidence-based surgical or non-surgical treatment recommendations to physicians. For example, generative AI can evaluate the effects and risks of different treatment options based on the patient's specific situation, and assist doctors in formulating personalized treatment plans.
Rajjoub et al. evaluated generative AI's response to the 2011 North American Spine Society (NASS) clinical guidelines for clinical issues such as the diagnosis and treatment of degenerative lumbar spinal stenosis (LSS). The authors used ChatGPT to answer 14 questions about LSS in the NASS clinical guidelines and compared the decisions it generated with the recommendations provided by the guidelines. At the same time, 40 articles on the diagnosis and treatment of lumbar spinal stenosis were reviewed to confirm or refute, covering the period from January 2012 to April 2023. The results found that ChatGPT's response was in line with the current literature on LSS, which not only accurately answered questions, but also provided evidence for the diagnosis and treatment of LSS, which is consistent with the current literature evidence. The authors believe that ChatGPT can be applied to the clinical decision-making work of spine surgeons, and that ChatGPT has great potential as a means to support the decision-making process for the diagnosis and treatment of LSS [6].
Jayakumar et al. evaluated the effects of AI-assisted decision-making tools on patient decision-making quality, patient experience, and functional outcomes, and included 129 patients who were considered TKA surgery for knee osteoarthritis and assigned to the intervention group that received decision-making aids or the control group that only received educational materials. The decision-making aids used in the intervention group consisted of three modules: education, preference assessment, and individualized outcome prediction. The results showed that the intervention group performed better in terms of decision-making quality, collaborative decision-making, satisfaction, and functional outcomes after 4 to 6 months. There were no significant differences between the intervention and control groups in terms of consultation time, TKR rates, or consistency of treatment. This randomized clinical trial demonstrated that the use of AI-assisted decision aids can significantly improve the quality of decision-making, satisfaction, and physical limitations of patients with knee osteoarthritis considering TKR surgery, without significantly affecting consultation time, TKR surgery rates, or treatment consistency [7].
3. Application of generative AI in orthopedic patient communication
Generative artificial intelligence (AI) is making progress in orthopedic patient communication, opening up new possibilities to improve patient education, enhance communication efficiency, and optimize the patient experience. Generative AI can provide personalized patient education materials that explain complex medical conditions, treatment options, and surgical procedures. Through natural language processing (NLP) technology, generative AI is able to understand patients' questions and provide clear and accurate answers. Generative AI can also serve as a bridge between doctors and patients, helping to explain medical terms and concepts and ensuring that patients fully understand their condition and treatment options. This is especially beneficial for those who do not speak English or who have cognitive impairments. Generative AI chatbots can also provide emotional support to help patients cope with stress and anxiety during diagnosis and treatment. They can provide coping strategies, mental health resources, or simply being a listener. Generative AI can also provide personalized post-operative care and rehabilitation guidance to help patients understand the expectations and potential challenges of recovery. Generative AI can also track the progress of a patient's post-operative recovery and remind them to review or adjust their treatment plan if necessary. Through mobile apps and online platforms, patients can communicate with AI in real-time and get instant feedback and recommendations. This immediacy helps patients better participate in the management of their own health and get quick answers to questions when they have them. Generative AI can support patients from multiple languages and cultural backgrounds, ensuring a high-quality communication experience for patients around the world. This is especially important for orthopedic surgeons, who may need to communicate with patients from different geographical and linguistic backgrounds.
Mika et al. investigated whether generative AI could accurately answer common questions about total hip arthroplasty. The researchers asked ChatGPT ten frequently asked questions about THA and evaluated them. Each response was analysed for accuracy by an evidence-based approach and rated according to the quality of the response. The results of the study found that only 1 of the responses given by ChatGPT was rated as "unsatisfactory", 2 did not need correction, and most required minimal (4 out of 10) or moderate (3 out of 10) further clarification. Although several responses require subtle clarification, ChatGPT's responses are generally unbiased and evidence-based, even for controversial topics. According to the authors, generative AI is able to accurately answer questions asked by patients before THA surgery, and in addition, these responses are not only based on clinical evidence, but also present information in a way that most patients can understand, making it more "approachable." Generative AI is likely to become a valuable clinical tool for patient education and understanding in the future [8].
4. Application of generative AI in the diagnosis and treatment of orthopedic clinical diseases
Generative artificial intelligence (AI) has shown its potential and advantages at multiple levels in the diagnosis and treatment of orthopedic clinical diseases, including a-disease auxiliary diagnosis, which assists in the diagnosis of orthopedic diseases such as fractures, osteoarthritis, and tumors by analyzing medical imaging data, such as X-rays, CT scans, and MRI images. Generative AI uses algorithms to identify and quantify lesion features, providing more accurate measurements and analysis to assist physicians in making more accurate diagnoses. bSurgical planning and simulation, AI technology can help doctors plan complex surgical processes and optimize surgical plans by simulating surgical steps. This planning technique can improve the success rate of surgery and reduce accidents and complications during surgery.
4.1 Orthopedic image recognition and analysis
The application of AI in orthopedic imaging is mainly focused on image recognition and analysis. AI can help identify lesions such as fractures, bone tumors, and joint diseases, improving the accuracy and efficiency of diagnosis. (1) Automatic image recognition and analysis: AI algorithms can automatically identify and analyze medical images such as X-ray, CT, and MRI, and quickly detect and mark abnormal areas, such as fractures, osteoarthritis, tumors and other lesions. (2) Fracture detection and classification: AI technology excels in fracture detection, accurately identifying different types of fractures, such as simple fractures, complex fractures or open fractures, and assisting doctors in assessing the severity of fractures and treatment options. (3) Osteoarthritis assessment: AI can quantify the severity of osteoarthritis and help doctors assess disease progression and treatment effect by analyzing parameters such as joint space, bone destruction, and cartilage thickness. (4) Bone tumor recognition: AI has also shown potential in the identification and classification of bone tumors, which can distinguish between benign and malignant tumors, and even predict the biological behavior of tumors, providing important information for clinical treatment. (5) Multi-modal image fusion: AI technology can integrate different sources and types of medical images, such as CT and MRI images, to provide more comprehensive bone and soft tissue information and enhance the accuracy of diagnosis.
Guermazi et al. evaluated the feasibility of AI-assisted technology to detect fractures on radiological images. The imaging results of 480 patients were included, and fracture diagnosis was performed by different doctors with and without the assistance of AI technology. The diagnostic sensitivity was found to be 64.8% (3732 of 5760) without AI assistance. When using AI assistance, the sensitivity was 95.6% (5504 of 5760). AI-assisted technology has reduced the time it takes to read a video by 6.3 seconds. The authors suggest that AI-assisted technology improves the sensitivity of radiologists and non-radiologists to detect fractures without prolonging the reading time [9].
4.2 Surgical planning and simulation
Generative AI can be used for pre-operative planning and simulation, analyzing patient imaging data to predict surgical procedures and outcomes. This helps doctors develop a more precise surgical plan and identify possible complications in advance. (1) Personalized surgical plan: AI technology can generate a personalized surgical plan based on the patient's specific anatomy and condition data. By analyzing CT scans, MRIs and other imaging data, AI can help doctors design surgical plans, determine the location, angle, and size of surgical incisions, as well as the size and placement of implants. It can also assist in postoperative evaluation, monitor the position and fixation of implants, and ensure the effectiveness of the surgery. (2) Surgical simulation: The use of AI for surgical simulation can predict the problems and challenges that may be encountered during the operation in advance, so that doctors have the opportunity to practice and optimize the surgical steps before the actual operation. This simulation is especially important for complex or rare surgeries. (3) Surgical navigation system: AI technology is being integrated into surgical navigation system to provide real-time, image-based guidance to help doctors locate and operate more accurately during surgery. This system can reduce the uncertainty in the operation and improve the accuracy and safety of the operation. (4) Robot-assisted surgery: The combination of AI and robotics is changing the face of orthopedic surgery. Robot-assisted surgical systems can perform complex surgical tasks under the guidance of AI, improving the stability and repeatability of surgeries. (5) Risk assessment: AI can assess the risk of surgery by analyzing the patient's clinical data and imaging data, including possible complications, postoperative recovery time, and failure rate. This helps doctors communicate better with their patients and develop risk management strategies accordingly. (6) Postoperative recovery prediction: AI models can predict the postoperative recovery process and outcomes of patients, including pain management, functional recovery, and quality of life improvement. This information is essential for developing a personalized rehabilitation plan and adjusting treatment strategies.
Velasquez et al. evaluated the application of artificial intelligence in three dimensional (3D) templates for preoperative planning of total joint replacement. The authors included nine studies, including CT or MRI-based AI templates. The results suggest that the AI-based 3D template system reduces the surgical planning time and improves the accuracy of the prosthesis size/position. The authors believe that AI-based 3D template systems have the potential to improve preoperative planning for joint replacement. This technique provides more accurate and personalized preoperative planning, with the potential to improve patient functional outcomes [10].
In conclusion, the application prospects of generative AI in the field of orthopedics are multifaceted and have potentially transformative impacts. Generative AI can assist doctors in making more accurate diagnoses by analyzing large amounts of medical imaging data. Generative AI can also provide personalized treatment recommendations based on the patient's specific situation and historical data to optimize treatment options. In healthcare education and training, generative AI can create simulated clinical scenarios to help doctors and medical students practice in a virtual environment to improve surgical skills and clinical decision-making. Generative AI can serve as a bridge between patients and physicians, providing clearer and easier to understand medical information, enhancing patient education and engagement. Generative AI technology can help doctors conduct detailed planning and simulation before surgery, improving the success rate and safety of surgery. Generative AI can be used for postoperative monitoring and rehabilitation guidance for patients, providing personalized rehabilitation plans and early warning of potential complications by analyzing patients' recovery data. The application of generative AI can facilitate collaboration between orthopedics and other medical fields, driving the development of integrated treatment options by integrating multidisciplinary data and knowledge.
In summary, the application of generative AI in the field of orthopedics is promising, and it is expected to greatly improve the quality and efficiency of medical services. However, as technology evolves, there is also a need to address the challenges and constraints that come with it to ensure the responsible and safe use of technology. First, the accuracy and reliability of generative AI needs to be further regulated, and the recommendations it provides must be clinically validated to ensure that they are accurate and applicable to clinical realities. There is also a need to ensure the privacy and security of physician-patient data, and it is important to ensure compliance with data protection regulations and patient privacy when using clinical data for generative AI decision-making. The use of generative AI in medical decision-making may also raise ethical and legal questions about accountability, transparency, and patient autonomy, which require further research. AI cannot completely replace the empathy and intuition of human doctors, so AI communication tools should be implemented as an enhancement rather than a replacement for traditional patient communication. Generative AI may also disseminate erroneous or biased information, which may negatively impact disease diagnosis and treatment, and may also pose risks such as the risk that over-reliance on technology among students in medical education may lead to decreased critical thinking and clinical decision-making [11].
About the Author
Chai Wei
Director of the Department of Joint Surgery, Department of Orthopedic Medicine, Chinese People's Liberation Army General Hospital, Chief Physician and Professor
Academic positions: Member of the International Hip Society, member of the Joint Surgery Working Committee and leader of the Youth Group of the Orthopaedic Branch of the Chinese Medical Doctor Association, member of the Joint Surgery Group of the Youth Committee of the Orthopaedic Branch of the Chinese Medical Association, vice president of the Bone and Joint Branch of the Chinese Geriatric Health Care Association, vice chairman of the Youth Committee of the Orthopaedic Branch of the Beijing Medical Association and leader of the Joint Surgery Group.
Li Haifeng
Deputy Chief Physician of the Department of Joint Surgery, Department of Orthopedics, Chinese General Hospital of the People's Liberation Army
He is committed to the basic research and clinical treatment of adult shoulder, elbow, knee and ankle joint reconstruction surgery. He is good at a series of stepwise whole-process treatments such as osteotomy around the knee joint, minimally invasive unicompartmental replacement, total knee replacement, navigation or robot and other digital and intelligent technologies to assist knee preservation, osteotomy and orthopedic treatment of congenital deformity of the knee joint and sequelae of traumatic fractures, joint reconstruction after knee bone tumor resection, robot-assisted knee joint partial and total knee replacement, etc.
He is a member of the Youth Committee of the Joint Surgery Group of the Orthopedic Branch of the Beijing Medical Association and a member of the Bone Infection Group. He has won 1 second prize and 2 third prizes of the Military Science and Technology Progress Award, 16 national patents, and published more than 60 papers.
bibliography
1 Lee P, Bubeck S, Petro J. Benefits, limits,and risks of GPT-4 as an AI Chatbot formedicine. New Engl. J. Med.2023，388, 1233–1239. doi: 10.1056/NEJMsr2214184
2 Sanderson, K. GPT-4 is here: what scientiststhink. Nature ，2023，615, 773. doi:10.1038/d41586-023-00816-5
3 Kaul V, Enslin S, Gross SA. History ofartificial intelligence in medicine. Gastrointest Endosc. (2020) 92:807–12. doi:10.1016/j.gie.2020.06.04
4 Lum ZC. Can artifcial intelligence pass theAmerican board of orthopaedic surgery examination? Orthopaedic residents versusChatGPT. Clin Orthop Relat Res,2023.481:1623–1630
5 Kung JE, Marshall C, Gauthier C,at al. Evaluating ChatGPT Performanceon the Orthopaedic In-Training Examination.. JB JS Open Access. 2023 Sep8; 8(3):e23.00056.
6 Rajjoub R, Arroyave JS, Zaidat B, at al. ChatGPT and its Role in the Decision-Making for the Diagnosis and Treatment of Lumbar Spinal Stenosis: A Comparative Analysis and Narrative Review.. Global Spine J. 2024Apr; 14(3):998-1017.
7 Prakash Jayakumar.Meredith G,Moore、Kenneth A,etal. Comparison of an Artificial Intelligence–EnabledPatient Decision Aid vs Educational Material on Decision Quality, SharedDecision-Making, Patient Experience, and Functional Outcomes in Adults WithKnee Osteoarthritis: A Randomized Clinical Trial.JAMA Network Open.2021; 4(2):e2037107. doi:10.1001/jamanetworkopen.2020.37107
8 Aleksander P. Mika、J. Ryan Martin、Stephen M.et al. AssessingChatGPT Responses to Common Patient Questions Regarding Total Hip Arthroplasty.JBone Joint Surg Am.2023 Oct 4; 105(19):1519-1526. doi:10.2106/JBJS.23.00209. Epub2023 Jul 17.
9 Ali Guermazi、Chadi Tannoury、AndrewJ. et al. Improving Radiographic Fracture Recognition Performance and EfficiencyUsing Artificial IntelligenceRadiology. 2022 Mar; 302(3):627-636. doi:10.1148/radiol.210937. Epub 2021 Dec 21.
10 Ausberto Velasquez Garcia、Lainey G. Bukowiec、at el . Wyles.Artificial intelligence–basedthree‑dimensionaltemplating for total joint arthroplasty planning: a scoping review. IntOrthop. 2024 Apr; 48(4):997-1010.doi: 10.1007/s00264-024-06088-6. Epub2024 Jan 15.
11 Madhan Jeyaraman, Shanmuga Priya K, Naveen Jeyaraman,et al. ChatGPT in MedicalEducation and Research: A Boon or a Bane? Cureus. 2023 Aug 29; 15(8):e44316
Disclaimer: The content and pictures of this article are provided by the contributor and are only for learning and exchange, and do not represent the views of Orthopedics Online.