Outlook of Future AI-Supported All-Around Rheumatic Disease Management: An Example of Knee Osteoarthritis

Abstract

Artificial Intelligence (AI) has emerged as a transformative force across various industries, and health care is no exception. In rheumatic disease management, AI holds tremendous potential to revolutionize how we diagnose, predict, and treat this debilitating condition [1, 2]. Published research has shed light on the transformative potential of AI applications in addressing key aspects of patient care, specifically focusing on pain management. By leveraging AI algorithms and machine learning techniques, healthcare professionals can tap into a wealth of patient-specific data, enabling accurate diagnosis, personalized treatment plans, and enhanced rehabilitation strategies [3]. Furthermore, the development of intelligent assistive devices and wearables propelled by AI promises to empower individual rheumatic patients to take control of their condition [4].

Knee osteoarthritis (KOA) is considered the most common form of arthritis that causes disability, which is associated with morphological changes in the subchondral bone, functional restrictions, articular cartilage degeneration, and damage to the surrounding soft tissue [5-7]. KOA has significant impacts on quality of life, which is also associated with gender, mental health, body weight, and other factors of patients [6]. This Editorial article takes KOA as an example of remarkable ways AI can contribute to rheumatology disease management.

One of the key challenges in KOA management is the timely and accurate diagnosis of the condition. AI algorithms excel in analyzing medical images, such as X-ray and magnetic resonance imaging (MRI), to detect and assess the severity of KOA. Today, many methods have been developed to diagnose KOA, but problems exist in automatic analysis and transparent diagnostic methods. Farooq et al. [8] developed a semi-supervised, multitask deep learning model that enhances KL-grade classification (KL-grade refers to Kellgren-Lawrence grade, a common method of classifying the severity of osteoarthritis) of KOA by integrating leg side identification as an auxiliary task, effectively leveraging additional unlabelled data to boost feature learning and achieving high accuracy (75.53%) and robustness across two major public datasets. Gornale et al. [9] compared the effectiveness of using k-nearest neighbors (KNN) and decision trees in detecting KOA from X-ray images and found that KNN performed better with a high accuracy of 99.80%, showing strong clinical relevance as a reliable computer-aided tool for early diagnosis and grading of osteoarthritis. Tiulpin et al. [10] introduced a Siamese convolutional neural network (CNN) to automatically diagnose and classify KOA from X-ray images, achieving high accuracy and strong agreement with expert evaluations (agreement with expert annotations on test dataset: 0.83). By incorporating attention maps and probabilistic outputs, the model enhances diagnostic transparency and supports clinical decision-making, particularly in early-stage disease detection. Pagano et al. [11] explored the ability of ChatGPT-4 to diagnose knee disorders, and results showed that diagnoses obtained from experts and the model were the same in all cases, highlighting its clinical potential as a decision support tool to be integrated into clinical workflows. These AI systems showed promise in identifying early signs of KOA, enabling timely intervention and improved patient outcomes.

AI models can leverage machine learning to analyze various factors, including clinical data, lifestyle, and imaging information, to predict disease progression. In a study, researchers utilized AI techniques to predict the progression of KOA. By analyzing clinical data, including patient demographics, symptom severity, and imaging results, the AI model accurately identified individuals at a high risk of developing severe symptoms and joint deterioration. Abdullah et al. [12] used a series of AI models, such as R-CNN, ResNet-50, and AlexNet, to locate and grade KOA from digital X-ray images based on the KL-grading system, and achieved up to 98.90% accuracy, demonstrating the potential of artificial intelligence to deliver highly precise, automated OA severity predictions and support more efficient, objective clinical assessments.

The era of one-size-fits-all treatment approaches is gradually giving way to personalized medicine, and AI is at the forefront of this shift. Understanding how KOA may progress in individual patients is crucial in tailoring effective treatment plans. In the UK, the AI to Revolutionize the Patient Care Pathway in Hip and Knee Arthroplasty (ARCHERY) project has been conducted to support the delivery of an automated solution for arthroplasty selection [13]. In terms of individual KOA patients, it is expected that AI algorithms take into account variables such as medical history, genetic information, and imaging results to develop personalized treatment plans. This tailored approach accounts for disease severity, patient preferences, and response to previous treatments, maximizing the likelihood of successful outcomes and reducing unnecessary interventions. Jayakumar et al. [14] conducted a clinical trial to evaluate an AI-enabled patient decision aid. They used a national patient-reported outcome measure (PROM) database and machine learning algorithms to generate personalized outcome predictions for patients with advanced KOA considering total knee replacement (TKR). Compared to education-only materials, the AI-driven tool significantly improved decision quality (mean difference: 20%), shared decision-making, patient satisfaction, and functional outcomes, demonstrating its clinical potential to support more informed, personalized, and effective treatment decisions.

Effective rehabilitation and physical therapy play a crucial role in managing KOA. AI-based systems can revolutionize these domains by designing personalized exercise programs that cater to an individual's specific needs. Zhu et al. [15] analyzed 36 studies on digital behavioral therapy applications for KOA and found that integrating AI enhanced rehabilitation by enabling personalized goal setting, precise monitoring, and data-driven treatment adjustments. AI-powered tools, such as mobile apps, wearables, and messaging platforms, improve patient engagement, physical function, pain reduction, and adherence by offering real-time feedback and tailored interventions, transforming KOA care into a more accessible, adaptive, and patient-centered process. Farías et al. [16] developed a chatbot using advanced artificial intelligence techniques to improve adherence to exercise-based treatment in patients with KOA. By integrating high-quality clinical guidelines into a Retrieval-Augmented Generation (RAG) system and deploying it via Telegram, the chatbot delivered accurate, personalized, and empathetic responses, significantly enhancing patient engagement and satisfaction, and demonstrating AI's transformative role in supporting consistent and evidence-based KOA treatment.

Recently, wearable sensors and assistive devices have been extensively studied for gait analysis and remote body condition monitoring [17, 18]. An artificial intelligence-based body sensor network framework (AIBSNF) [17] was proposed to strategize the body sensor networks (BSNs), which optimized real-time location system (RTLS) and wearable biosensors to gather multivariate, low-noise, and high-fidelity data. By analyzing those data, the potential OA-related changes could be recognized. Besides, the quantification of varus thrust in KOA patients could be done with the placement of inertial sensor [19]. Those findings reveal the potential of wearable sensors or assistive devices as an evaluation tool for rehabilitation performances and therapeutic effects. Although the findings are exciting and inspirational, the outcome domain for data collection approach has not been established and validated with clinical presentation.

Despite the advancement of AI that could bring into rheumatic disease management, it also faces several challenges [2, 4]. Data fragmentation and lack of standardization could limit the performance of AI models, requiring standardized data frameworks and interoperability solutions. Algorithmic bias and poor generalizability could lead to disparities in care, necessitating diverse training datasets and fairness-aware AI models. The application of AI could also be hindered by patient trust and ethical concerns, such as data privacy and transparency, which require educating patients and healthcare professionals, implementing robust data governance, and other measures. The limitations of AI in clinical settings include their limited external validation, which hampers both accuracy and generalizability, particularly across diverse institutions where factors like local policies and resource availability can affect outcomes. Additionally, these models require large, complex datasets for training and testing, face challenges due to their opaque decision-making processes (the “black box” problem), and are vulnerable to data bias, continual change, and regulatory concerns such as quality assurance, security, and resistance to adversarial threats [20]. The challenges should be addressed through cross-domain collaborations, continuous innovation and validation, and regulatory oversight to unlock AI's full potential, transforming rheumatic disease management into a more precise, proactive, and patient-centered approach.

These recent research findings present the transformative impact of technology in rheumatic disease management. AI has been applied to KOA primarily through machine learning models that enable automatic grading of radiographic severity, prediction of the need for total knee arthroplasty (TKA), and estimation of postoperative outcomes such as patient satisfaction and complications, which could enhance diagnostic accuracy and clinical decision-making but require further validation and integration of clinical variables to improve their real-world utility. As we stand on the cusp of a new era in health care, it is imperative that we embrace the power of AI. The potential of AI in revolutionizing rheumatic disease management is both promising and inspiring, from accurate diagnosis and prediction of disease progression to personalized treatment plans, rehabilitation optimization, and intelligent assistive devices. However, ongoing research, collaboration between healthcare professionals and AI experts, and the integration of these technologies into clinical practice are essential to address the challenges that hinder the application of AI and fully harness the power of AI in improving patient outcomes. As we embrace the potential of AI, we embark on a journey toward a future where rheumatic disease is managed with unprecedented precision, compassion, and efficacy.

C.-W.C. contributed to the conception, writing, review, and editing of this paper. J.L.-H.S. contributed to the conception and writing of the original draft of this paper. Y.-H.L. and J.C.-C.W. contributed to reviewing and editing of this paper. All authors were involved in the study design and writing of the paper, and had final responsibility for the decision to submit for publication.

The authors declare no conflicts of interest.