Limited and frequently overlooked radiological evidence of knee osteoarthritis

Abstract

Knee osteoarthritis (KOA) is a complex degenerative condition involving the entire knee joint structure, with a pathological foundation that extends far beyond the fibrosis, wear, and denudation of articular cartilage. The complete pathological process also includes subchondral bone sclerosis, osteophyte formation, synovitis, and alterations in periarticular soft tissues. However, in routine radiological diagnosis, the evaluation of KOA is often confined to the narrowing of the tibiofemoral joint and overly reliant on the Kellgren-Lawrence grading system as a single standard. This has led to the systematic neglect of assessments for critical compartments such as the patellofemoral joint. In fact, evaluating patellofemoral alignment (e.g., congruence angle) via axial X-rays and observing early cartilage damage and bone marrow lesions via magnetic resonance imaging are essential for a comprehensive diagnosis of KOA. Currently, the limited sensitivity of Kellgren-Lawrence grading for early KOA changes is increasingly recognized. This necessitates a shift in perspective, moving beyond traditional grading to integrate multimodal and multicompartment radiological evidence, including patellofemoral joint assessments and novel functional magnetic resonance imaging techniques. Building on this foundation, future research directions - such as developing deep learning models that incorporate these comprehensive parameters - can be grounded in a solid and accurate pathophysiological understanding, ultimately advancing KOA diagnosis and treatment toward greater precision and personalization.

Key Words: Knee osteoarthritis; Artificial intelligence; Kellgren-Lawrence; Magnetic resonance imaging; Deep learning models

Core Tip: The prevalence of knee osteoarthritis is rising annually, causing chronic pain, impaired joint function, and reduced quality of life. Accurate diagnosis has thus become crucial. With evolving understanding, clinicians must assess all knee compartments comprehensively. The limitations of single diagnostic tools have driven the development of multimodal approaches. These are enhanced by intelligent network technologies that analyze large clinical datasets, enabling deep learning models to improve diagnostic precision.

TO THE EDITOR

As global aging continues to deepen, the incidence of knee osteoarthritis (KOA) has risen steadily each year. KOA is characterized by a range of symptoms and signs, such as joint pain, stiffness, restricted mobility, knee swelling, deformity, and muscle atrophy. As the condition advances, activities that were once manageable may gradually trigger limited joint movement and worsening pain; persistent discomfort, rest-related pain, and nighttime soreness often signal a more advanced stage of the disease[1]. Data from the China Health and Retirement Longitudinal Study (CHARLS) database reveals that the prevalence of symptomatic KOA - defined by a Kellgren-Lawrence (KL) score of ≥ 2 accompanied by knee pain - stands at 8.1% in China[2]. Given China’s large population, this translates to over 100 million symptomatic KOA patients. When accounting for asymptomatic individuals with potential KOA, the massive patient pool places a significant economic burden on both society and the healthcare system.

From a pathological standpoint, KOA typically begins with subtle changes. The disease may develop due to factors like obesity, a history of joint injuries, or elevated bone mineral density - all of which can disrupt the normal anatomical alignment of joint surfaces or subject them to prolonged compressive stress, leading to excessive wear and damage[3,4]. Routine daily use further accelerates the degeneration of articular cartilage. Under abnormal joint stress, subchondral bone undergoes structural changes, including ivory-like degeneration and the formation of osteophytes (OST). These alterations not only reduce knee joint stability and increase mechanical strain but also worsen pain and mobility issues.

The 1986 classification and reporting of KOA by the American College of Rheumatology marked the first systematic definition of KOA diagnostic criteria, covering three key dimensions: Clinical symptoms, imaging findings, and laboratory tests (Table 1)[5]. This 1986 American College of Rheumatology guideline was undoubtedly a milestone in standardizing osteoarthritis (OA) diagnosis. It provided a reliable framework for clinical practice and research, exerted a far-reaching impact, served as a critical reference for subsequent guideline development, and greatly promoted the continuous refinement of OA diagnostic criteria. Nevertheless, the guideline had limitations: It devoted relatively little attention to early lesions, soft tissue damage, and inflammatory characteristics, and some indicators lacked sufficient applicability - factors that somewhat compromised its comprehensiveness and accuracy in OA diagnosis.

Table 1 The 1986 American College of Rheumatology’s definition of knee osteoarthritis diagnosis.

Clinical and laboratory	Clinical and radiographic	Clinical+
Knee pain + at least 5 of 9: Age > 50 years, stiffness < 30 minutes, crepitus, bony tenderness, bony enlargement, no palpable warmth, erythrocyte sedimentation rate (Westergren) < 40 mm/hour, rheumatoid factor < 1:40, synovial fluid signs of osteoarthritis (clear, viscous, or white blood cell count < 2000/mm3)	Knee pain + at least 1 of 3: Age > 50 years, stiffness < 30 minutes, crepitus, osteophytes	Knee pain + at least 3 of 6: Age > 50 years, stiffness < 30 minutes, crepitus, bony tenderness, bony enlargement, no palpable warmth
92% sensitive	91% sensitive	95% sensitive
75% specific	86% specific	69% specific

Alternative for the clinical category would be 4 of 6, which is 84% sensitive and 89% specific.

With technological advancements and deeper research into KOA, diagnostic criteria for the disease have been updated repeatedly. Two years ago, the China Association of Chinese Medicine issued the latest guideline for the integrated traditional Chinese and Western medicine diagnosis and treatment of KOA. This document offers more detailed descriptions of KOA symptoms at each stage, imaging indicators [including X-ray KL grading and magnetic resonance imaging (MRI) Recht grading], and relevant laboratory test results (such as changes in C-reactive protein and erythrocyte sedimentation rate during the acute phase).

In terms of KOA treatment, conservative approaches usually involve personalized combinations of medications, including non-steroidal anti-inflammatory drugs, acetaminophen, opioid analgesics, slow-acting anti-OA drugs, and cartilage protectants. If the disease progresses to a point where these medications or external treatments no longer improve the patient’s quality of life, surgical intervention - such as total knee arthroplasty - remains a last-resort option.

However, since the goal of treating KOA is to relieve clinical symptoms and enhance patients’ quality of life, it is essential to avoid improper medication use and excessive medical intervention. Therefore, accurate diagnosis plays a particularly crucial role in guiding effective KOA treatment.

MULTICOMPARTMENT RADIOLOGICAL EVIDENCE

The knee joint, the largest weight-bearing joint in the human body, consists of the femur, tibia, and patella, and encompasses three distinct cavities: The patellofemoral joint, along with the medial and lateral tibiofemoral joints. In clinical practice, when diagnosing and assessing KOA, it is equally vital to fully account for the actual lesion status of these three joint cavities. Only by adopting a holistic perspective of the entire bilateral knee joints can clinicians make a comprehensive and accurate disease judgment, thereby ensuring that the selected treatment yields more effective therapeutic outcomes.

For conventional X-ray diagnosis and evaluation of KOA, the semi-quantitative KL grading system is commonly employed. Standing weight-bearing anteroposterior and lateral radiographs of the knee joint prevent the patella from obscuring the tibiofemoral joint, enabling clearer visualization of the tibiofemoral joint space width and density changes in the bones above and below. In cases of patellofemoral arthritis, supplementing with patellar skyline views and lateral flexion axial images can better capture joint space narrowing (JSN), subchondral sclerosis, osteophyte formation, and patellar alignment alterations when the patellofemoral joint is under load[6]. Studies indicate that lateral patellar displacement and changes in the lateral patellar tilt angle are not only contributing factors to patellofemoral joint OA but also key clinicopathological manifestations of the condition[7,8]. Thus, when patellofemoral arthritis is suspected, in addition to the standard X-ray KL grading assessment, measuring the Q angle and congruence angle to quantify patellar displacement can further assist in diagnosing patellofemoral arthritis.

While X-ray imaging can reveal the skeletal characteristics of OA, it cannot directly visualize other joint structures or periarticular tissues - such as cartilage and synovium. Instead, it can only indirectly estimate cartilage thickness and meniscal integrity based on joint space width[9]. To directly observe the condition of joint-related soft tissues, MRI plays an irreplaceable and critical role in KOA diagnosis, given its status as a key examination technique for detecting soft tissue injuries. Since the initial pathological feature of KOA is articular cartilage damage or degeneration, the cartilage may still retain its external shape in the early disease stage (KL = 0-1), making it impossible to accurately detect JSN on X-rays. However, the recently identified meniscal spatial specific signal index in tibiofemoral MRI images can accurately predict the presence of KOA on imaging[10]. Similarly, the Rapid Osteoarthritis MRI Qualifying Score, developed by Roemer et al[11], fully considers the severity of cartilage lesions, bone marrow lesions (BMLs), OST, meniscal abnormalities, and inflammatory reactions. It comprehensively and meticulously classifies tibiofemoral joint OA into five structural phenotypes and incorporates exclusion diagnoses to facilitate precise patient screening. Although the clinical validity of Rapid Osteoarthritis MRI Qualifying Score for assessing patellofemoral joint OA has not yet been confirmed, relevant research is still in progress.

MULTIMODAL DIAGNOSIS

Building on the breakthroughs of multimodal diagnosis in overcoming traditional limitations, as more KOA-related imaging indicators and biomolecules are identified, and with the growing maturity of artificial intelligence (AI) applications, AI - particularly deep learning methods - has generated considerable interest in the medical diagnosis and assessment of KOA over the past decade. As Alshahrani et al[12] showed in their observational study, there is a strong correlation between the severity of imaging findings (based on the KL grading system) and patient-reported symptoms. While this method can detect structural changes in the joint, it struggles to identify early-stage lesions and account for individual variations. By contrast, multimodal diagnosis overcomes the limitations of single-dimensional assessment: It integrates clinical manifestations (e.g., pain, functional impairment), risk factors (e.g., obesity, occupational factors), auxiliary examinations (e.g., MRI, biomarkers), and physical examination results. Through the fusion of multiple technologies and intelligent algorithms, it enables a leap from “structural diagnosis” to “precision prediction”, laying a scientific foundation for personalized treatment and prognosis evaluation while driving OA diagnosis and management toward earlier intervention, comprehensiveness, and dynamic monitoring.

In a study by Paz-González et al[13], researchers integrated key non-imaging risk factors for KOA development - including female sex, advanced age, high body mass index, and Western Ontario and McMaster Universities Arthritis Index pain scores. They also included relevant protein biomarkers (APOA1, APOA4, A2AP, and ZA2G) from participants with no radiological signs of KOA (KL = 0 in both knees). Subsequently, they developed a KOA predictive model that combined these risk factors with specific protein biomarkers (such as A2AP). Compared to existing single clinical models, this new model showed a significant improvement in predictive performance, with an area under the curve (AUC) of 0.831 (95%CI: 0.750-0.913), as well as enhanced sensitivity and specificity.

Beyond incorporating protein biomarkers – a type of biological data - predictive models that integrate multiple direct digital radiography (DR) indicators have also emerged; these models focus on the precise interpretation of imaging data, offering a new approach to KOA severity assessment. Studies have found a 66.1% overall discrepancy between pre-operative radiological assessments (KL grading) and actual intraoperative findings (cartilage damage) in patients undergoing total knee arthroplasty[14]. To reduce excessive or inappropriate medical intervention caused by inaccurate disease severity evaluation, more precise auxiliary diagnostic indicators or tests should be widely adopted in assessing patient disease severity. As the most cost-effective, reliable, low-risk, and accessible imaging method in most clinical settings, DR has become a practical choice for large-scale screening and diagnostic assessment - making it particularly important to improve the accuracy of DR-based diagnosis[9]. In a model developed by Sun et al[15], two integrated models - the combined RKOA model (AUC = 0.969) and combined KL model (AUC = 0.969) - which incorporated DR-related radiomic features and clinical data, both demonstrated high sensitivity in evaluating KOA severity. Additionally, clinicians can use the model scores to objectively quantify patients’ surgical needs.

Research indicates that primary care general practitioners have limited knowledge of KOA diagnosis and basic management[16]. However, the rise of multimodal comprehensive diagnostic models is expected to address this gap, enabling patients to receive more accurate and actionable diagnoses and thereby reducing misdiagnosis and delayed treatment. At the same time, since multimodal diagnosis requires integrating results from multiple auxiliary examinations, the implementation of such models in primary care settings must consider the equipment conditions and personnel technical capabilities of grassroots medical institutions. In China, the penetration rate of X-ray machines in grassroots medical institutions is as high as 98%, and DR penetration reaches 76%. With the advancement of relevant policies, the testing capabilities of these institutions are continuously improving - providing a solid foundation for the promotion and popularization of multimodal comprehensive diagnostic models.

AI: DEEP LEARNING MODEL

As more KOA-related imaging indicators and biomolecules are identified, and with the growing maturity of AI applications, AI - particularly deep learning methods - has generated considerable interest in the medical diagnosis and assessment of KOA over the past decade. Predictive models and deep learning models that integrate multiple types of relevant data have emerged continuously, further driving KOA diagnosis and treatment toward greater precision and personalization.

While the automated tibiofemoral OA diagnostic model developed by Thomas et al[17] only utilized KL grading and OARSI classification from single X-ray images, the high-resolution network it employed significantly reduced the impact of physicians’ experience levels and subjective judgment. In the test set, the high-resolution network model achieved an AUC of 0.86-0.99 for diagnosing tibiofemoral arthritis and its associated imaging features (including OST and JSN), as well as for the precision-recall AUC. This breakthrough standardized X-ray-based KOA diagnosis. From an imaging perspective, moving beyond AI-driven automated X-ray diagnosis, new progress has also been made in research using AI deep learning to assess KOA severity via MRI - a modality better suited for soft tissue visualization.

From an imaging perspective, beyond AI-driven automated X-ray diagnosis, new progress has been made in research using AI deep learning to assess KOA severity via MRI. In a study by Moradi et al[18], researchers built an automated deep learning model for segmenting and quantifying BMLs. The model was used to perform knee MRI scans on subjects without radiological evidence of OA, quantify tibiofemoral BML volume, and analyze the correlation between longitudinal BML changes and the onset of KOA. The model’s predictive results showed high consistency with actual verification (dice similarity coefficient: 0.86-0.88). These findings confirm that AI can improve the detection of BMLs through MRI, which supports the potential for early KOA diagnosis.

Compared with auxiliary examination methods like X-rays and MRI - which require image interpreters to have certain experience or image processing skills - serum-related protein biomarkers (such as APOA1) typically only need to be read from test reports. For this reason, there remains substantial room for advancing AI applications in KOA-related biological diagnosis.

While AI has demonstrated significant potential for precision and automation in KOA diagnosis, greatly advancing diagnostic efficiency and accuracy, it still faces multiple challenges in practical clinical implementation. First, privacy and security issues pose major obstacles to the application of AI technology. AI systems process large volumes of patient health data, including sensitive information such as demographic data, imaging data, and biomarkers. How to ensure the security and privacy of this data while complying with national laws and regulations has become an urgent issue. Thus, establishing more robust data protection mechanisms and formulating corresponding ethical guidelines are critical.

Additionally, AI model training primarily relies on retrospective clinical datasets. Overfitting to historical data can lead to a sharp drop in the model’s generalized predictive performance for new datasets, creating an overfitting risk. A key bottleneck in current research is the lack of independent external datasets (not used in model development) to validate AI systems - this directly limits the clinical generalization and reliability of AI models.

SUMMARY

KOA has a vast patient base worldwide, and the development of its diagnosis and treatment is constantly being updated and even overturned and rebuilt. The understanding and diagnosis of KOA not only require radiological evidence from multiple chambers in the physical space of the knee joint, but also can rely on a comprehensive diagnostic model that integrates other relevant indicators such as the patient’s risk factors, symptoms, signs, imaging and laboratory to assist clinical workers in conducting more precise, specific and personalized treatment. At the same time, AI summarizes experience through deep learning of a large amount of patient data, which to a certain extent makes the diagnosis of KOA have more simplified operations and more standardized and accurate results. Although there is still a lack of a large number of clinical trials for related diagnostic evaluation models, the emergence of such research already represents the development of KOA diagnosis towards standardization, comprehensiveness and automation. It is believed that in the near future, AI deep learning models with higher accuracy and specificity will emerge, ultimately promoting the diagnosis and treatment of KOA towards higher precision and personalization.