Multimodality imaging considerations for genicular artery embolization in knee osteoarthritis

Abstract

Genicular artery embolization (GAE) describes a minimally invasive, catheter-directed therapy that targets abnormal hypervascular synovium and periarticular neovessels in patients with symptomatic knee osteoarthritis (OA), particularly those with inflammation-predominant disease who are poor surgical candidates or wish to defer arthroplasty. As clinical adoption expands, pre-, intra-, and post-procedure imaging has become increasingly relevant to both diagnostic and interventional radiologists. In practical terms, inflammation-predominant disease refers to knees with prominent synovitis-related magnetic resonance imaging (MRI) findings despite less advanced irreversible structural destruction, whereas structure-predominant disease is characterized by extensive cartilage loss, large bone marrow lesions, meniscal maceration, and advanced radiographic degeneration. Although no prospectively validated imaging cutoff exists for routine GAE candidate selection, current studies suggest that moderate-to-large effusion-synovitis (the combined appearance of joint fluid and synovial thickening on non-contrast MRI) (grade 2-3), greater overall MRI lesion burden (≥ 4 abnormalities), and extensive full-thickness cartilage loss may help distinguish these phenotypes. Synovial enhancement and effusion-synovitis on MRI are the findings most consistently expected to improve, whereas structural hallmarks of OA often persist and may be misinterpreted as treatment failure. Early marrow or subchondral signal alterations have also been reported after embolization and require cautious follow-up interpretation. This review summarizes contemporary evidence on pre-, intra-, and post-procedure imaging, semiquantitative MRI scoring systems, and common post-GAE interpretive pitfalls.

Key Words: Genicular artery embolization; Knee osteoarthritis; Magnetic resonance imaging; Synovitis; Post-procedure imaging; Interventional radiology; Angiography; Artery; Radiography

Core Tip: Genicular artery embolization (GAE) is increasingly used for inflammation-predominant knee osteoarthritis, making accurate post-procedure imaging interpretation essential. This review summarizes expected magnetic resonance imaging (MRI) and radiographic findings after GAE, emphasizes that inflammatory features (synovial enhancement and effusion-synovitis) are the most consistent imaging response signals, and highlights pitfalls in overcalling persistent structural osteoarthritis findings (osteophytes, cartilage defects, joint space narrowing) as treatment failure. Semiquantitative MRI scoring systems used in published GAE studies are reviewed to support consistent assessment and follow-up.

INTRODUCTION

Osteoarthritis (OA) is a common degenerative joint disease and a leading cause of chronic pain and disability worldwide. Globally, it affects more than 500000000 individuals[1-5]. It represents a major contributor to healthcare utilization and mobility loss[1-5]. The current projections estimate the number of people affected by OA could approach 1000000000 by 2050, posing immense healthcare challenges[4].

Initial management of knee OA focuses on conservative therapies including physical therapy, weight reduction, activity modification, and pharmacologic interventions such as nonsteroidal anti-inflammatory drugs and intra-articular injections[6]. Although many patients experience symptomatic relief, a substantial proportion continue to report persistent pain and functional limitation despite optimized medical therapy[7]. Surgical options such as total knee arthroplasty remain effective for advanced disease but may not be appropriate for patients with moderate OA or for those wishing to delay surgery. For this reason, there is immense interest in novel, minimally invasive interventions to reduce pain and postpone surgery for knee OA[5,6]. In this setting, genicular artery embolization (GAE) has emerged as a minimally invasive endovascular therapy targeting abnormal synovial neovascularity associated with OA-related pain[8,9]. Early studies have demonstrated reductions in pain and improvements in function.

The success of GAE depends heavily on appropriate patient selection and accurate interpretation of imaging findings before and after treatment. Because OA is a heterogeneous disease, not all symptomatic knees demonstrate the inflammatory features believed to respond to embolization. As a result, imaging has become central to identifying patients whose symptoms are driven by synovitis and hypervascularity rather than advanced structural degeneration[8,9]. Radiologic assessment therefore plays a critical role not only in determining candidacy for GAE but also in establishing baseline disease patterns that influence treatment expectations.

Among available modalities, radiography and magnetic resonance imaging (MRI) form the foundation of imaging evaluation in patients with knee OA. Conventional radiographs remain the primary tool for structural assessment and disease staging, which allows characterization of joint space narrowing, osteophyte formation, and subchondral remodeling. MRI provides a more comprehensive evaluation of the joint by depicting synovitis, effusion, bone marrow signal abnormalities, cartilage damage, and other features associated with symptomatic disease activity[6,10-13]. Increasing evidence suggests that these inflammatory imaging markers correlate with symptomatic disease activity, rather than structural degeneration alone, making MRI particularly valuable when evaluating patients being considered for GAE.

Procedural angiography complements the diagnostic information obtained from radiography and MRI[14-16]. Instead, much of the radiologist’s clinical challenge lies in interpreting baseline imaging findings, recognizing inflammation-predominant disease patterns, and correctly evaluating follow-up studies after treatment. Persistent structural OA features on imaging may remain unchanged despite clinical improvement, and without proper context these findings may be misinterpreted as treatment failure. Given the expanding clinical use of GAE and the absence of standardized imaging guidance, an imaging-centered framework is needed to support clinical decision-making. At the same time, current evidence remains heterogenous. The literature is limited by small sample sizes, variable imaging protocols, inconsistent outcome definitions, and a lack of standardized imaging thresholds for candidate selection and response assessment. As such, available imaging biomarkers should currently be interpreted as supportive rather than definitive tools for clinical decision-making. For the purpose of imaging-based stratification, inflammation-predominant disease refers to knees in which synovitis-related features such as contrast enhancement, effusion-synovitis, or Hoffa’s fat pad inflammatory change are prominent despite less advanced irreversible structural destruction. In contrast, structure-predominant disease refers to knees with extensive full-thickness cartilage loss, large bone marrow lesions (BMLs), meniscal maceration, and advanced Kellgren-Lawrence (KL) grade, where symptoms may be more driven by irreversible mechanical degeneration rather than by treatable synovial hypervascularity. Although no prospectively validated cutoff exists, currently available GAE studies suggest that moderate-to-large effusion-synovitis (grade 2-3), a greater overall MRI lesion burden (≥ 4 abnormalities), and extensive full-thickness cartilage loss may influence treatment response and can be used as practical research-based markers of phenotype rather than definitive clinical thresholds[17-20]. Practically, an inflammation-predominant imaging pattern could be defined by a moderate-to-large effusion-synovitis on MRI OA knee score (MOAKS) (grade 2-3), with contrast-enhanced MRI (CE-MRI) synovial enhancement in the absence of end-stage structural destruction. Whereas, a structure predominant pattern is characterized by extensive full-thickness cartilage loss, meniscal maceration or extrusion, and this could be with substantial subchondral lesion burden that is less likely to demonstrate imaging reversibility[17,18,20-22]. Additionally, no validated CE-MRI synovitis cutoff has been prospectively established for GAE candidate selection. However, CE-MRI semiquantitative synovitis scoring systems are reliable and pain-associated in knee OA, and dynamic CE-MRI (DCE-MRI) perfusion markers such as K trans provide quantitative support for higher inflammatory burden even though no treatment-selection threshold is yet available for routine clinical practice[22-27].

Review methodology

This manuscript was conducted as a narrative review to summarize current evidence regarding multimodality imaging in GAE for knee OA. A structured literature search was performed using PubMed, MEDLINE, and Google Scholar databases from January 2010 through February 2026. Search terms included combinations of “genicular artery embolization”, “knee osteoarthritis imaging”, “MRI knee osteoarthritis”, “synovitis MRI”, “bone marrow lesions”, “digital subtraction angiography GAE”, “angiography”, “knee anatomy”, “radiography” and “interventional radiology knee osteoarthritis”.

Studies were selected based on relevance to imaging evaluation of GAE, imaging biomarkers of OA, procedural imaging guidance, and post-procedural imaging outcomes. Priority was given to prospective studies, clinical trials, systematic reviews, and high-quality observational studies. Additional references were identified through citation tracking of key articles. Given the emerging nature of GAE literature and heterogeneity of available studies, a narrative synthesis approach was used.

PATHOPHYSIOLOGY OF OA

Over the past decade, insights into OA pathophysiology have shifted attention toward the role of chronic synovitis and abnormal angiogenesis (neovascularization) in driving knee pain[8]. Figure 1 summarizes the structural and inflammatory changes that accompany progression from a healthy joint to OA. Pro-inflammatory cytokines promote proliferation of new blood vessels in the synovial membrane and adjacent structures, while accompanying nerve ingrowth increases pain sensitivity in regions that are normally minimally innervated[8]. Imaging studies support the clinical relevance of this inflammatory phenotype. For example, synovitis or effusion was identified as one of the structural features most strongly associated with knee pain severity in a recent machine-learning analysis[25]. In addition, quantitative perfusion techniques such as DCE-MRI have shown that parameters such as K trans and initial area under the curve (IAUC) correlate with synovitis severity and may better reflect active inflammatory vascularity than static non-contrast imaging alone[26,28]. In DCE-MRI studies, K trans has shown the strongest repeatability and responsiveness among perfusion parameters, supporting its use as a quantitative surrogate of synovial vascular permeability and inflammatory activity rather than relying solely on static morphologic assessment[28]. These findings support the biologic rationale for GAE, which targets synovial hypervascularity to interrupt the inflammation-angiogenesis-nerve ingrowth cycle[29,30]. This biologic framework is also supported by histopathologic and human tissue studies linking osteochondral angiogenesis with growth factor expression, including nerve growth factor, and with sensory nerve ingrowth that may amplify pain[31,32].

Figure 1 Comparison of healthy knee joint anatomy and osteoarthritis-related structural degeneration. Schematic representation comparing a healthy knee joint (left) with osteoarthritis (right). The osteoarthritic joint demonstrates cartilage thinning and degradation, osteophyte formation, synovial inflammation, and reduced joint space.

GAE AS A NOVEL TREATMENT OPTION

The recognition of pathologic neovascularization in knee OA has paved the way for GAE as a targeted treatment strategy[8]. GAE is a minimally invasive endovascular procedure that aims to occlude the small arteries supplying the inflamed synovium and other pain-generating structures in the knee[9]. GAE utilizes a catheter-directed approach to deliver embolic particles into the genicular arterial branches supplying the hypervascular synovium, thereby interrupting the inflammatory feedback loop and reducing synovial hyperplasia and pain (Figure 2)[8]. However, OA is heterogeneous, and not every painful knee is driven by active synovitis or hyperemia. For this reason, identifying the right imaging phenotype becomes essential to predict who is most likely to benefit from embolization. Therefore, this review includes pre-procedural evaluation across plain radiography and MRI, including KL staging and synovitis phenotyping using whole-organ MRI score (WORMS) and Boston-Leeds OA knee score (BLOKS), followed by intraprocedural imaging findings on fluoroscopy and cone-beam computed tomography (CBCT), the interpretation of post-procedural imaging findings, and emerging future directions in imaging-driven selection.

Figure 2 Mechanistic overview of genicular artery embolization in knee osteoarthritis. Schematic illustration demonstrating selective catheterization of a genicular artery supplying inflamed synovium in knee osteoarthritis. Embolic particles are delivered through a microcatheter to occlude abnormal hypervascular synovial neovessels while preserving flow in adjacent parent arteries. The inset magnified view illustrates embolic particles within the arterial lumen alongside circulating blood cells, representing targeted pruning of pathologic microvascular networks implicated in inflammation-driven pain.

PRE-PROCEDURAL EVALUATION AND PATIENT STRATIFICATION

Plain radiography

Radiography is the structural baseline before expanding to MRI-based phenotyping and intraprocedural angiographic targeting. Plain radiographs remain the workhorse for initial imaging assessment in OA. For decades, standard weight-bearing knee radiographs have been the first-line modality to establish a structural diagnosis and gauge disease severity[6]. In practice, radiographs are often used to confirm OA, for example, to document joint-space narrowing or osteophytes that indicate structural OA changes[6]. In short, plain radiography remains the fundamental baseline imaging for knee OA, anchoring our initial evaluation of the disease. At a practical level, radiographs are best at capturing the bony consequences of OA rather than its early inflammatory or soft-tissue drivers.

Radiographs primarily visualize the bony manifestations of OA. The classic hallmarks seen on knee radiography include joint space narrowing, signaling cartilage loss, along with osteophyte formation, subchondral sclerosis, and subchondral cysts[6]. These features are readily apparent on a good quality radiograph. Notably, radiographs only depict bone and calcified tissue directly; they do not show articular cartilage, menisci, ligaments, or synovium. As a result, radiography often underestimates the full burden of disease, particularly in patients whose symptoms are driven by synovitis, cartilage damage, or bone marrow abnormalities not visible on plain films[6,33,34].

Given the bony focus of radiographs, radiologists use standardized grading systems to describe OA severity. The most common is the KL grading, which ranks OA from 0 (no findings) up to 4 (severe OA) based on radiographic features (Table 1)[35,36]. In clinical practice and research, KL ≥ 2 (the presence of a definite osteophyte) is typically used to define established OA on radiographs[6]. KL grades 3 and 4 represent progressively severe disease, grade 3 usually denotes marked joint space narrowing with sclerosis (but the joint still has some alignment), while grade 4 implies bone-on-bone contact, large osteophytes, and often varus or valgus deformity of the knee[37]. Figure 3 highlights some structural baseline findings of the disease. Thus, the KL system remains a widely used tool for stratifying radiographic severity in both clinics and trials[35]. Although KL grading is widely used, interobserver variability remains a recognized limitation, which has motivated efforts to improve consistency through standardized atlases and artificial intelligence (AI)-assisted assessment[35].

Figure 3 Plain radiographic features of structural knee osteoarthritis. Radiograph of the knee demonstrating classic structural manifestations of osteoarthritis, including medial compartment joint space narrowing, marginal osteophyte formation, and subchondral cyst formation.

Table 1 Kellgren-Lawrence radiographic grading system for knee osteoarthritis.

Grade	Description
0	No radiographic features of osteoarthritis
1	Doubtful joint space narrowing and possible osteophytic lipping
2	Definite osteophytes and possible joint space narrowing
3	Moderate multiple osteophytes, definite joint space narrowing, some sclerosis, and possible bone deformity
4	Large osteophytes, marked joint space narrowing, severe sclerosis, and definite bone deformity

Despite its strengths, plain radiography has significant limitations in OA. Perhaps the biggest is its insensitivity to early disease. Because cartilage does not show up on radiography, definite joint space narrowing may not appear until substantial cartilage loss has already occurred. In fact, in many “pre-OA” knees, a radiograph will look normal even though the patient may already have cartilage wear or meniscal degeneration. Even in advanced OA (KL 4), plain radiographs picked up only about 50% of significant cartilage damage[38]. Even in other countries like the United Kingdom, guidelines state that routine knee radiographs are not required to diagnose OA[39]. Conversely, radiographs often show abnormalities (like osteophytes or mild narrowing) in people who have no symptoms at all. It is well documented that radiographic severity correlates only weakly with pain and function in OA[6,40]. This discordance is particularly relevant for GAE, because radiographs may not reflect the inflammatory and vascular features that drive symptoms or predict response to embolization. Accordingly, radiography should be interpreted as a structural staging tool rather than a stand-alone measure of symptom severity or procedural suitability.

Another limitation of radiography is the challenge of monitoring progression. Radiography also has limited sensitivity for short-term monitoring, because structural OA changes evolve slowly and joint space measurements are affected by positioning and technical variability[6]. For this reason, radiographs are more useful for documenting baseline structural severity than for detecting early interval changes relevant to GAE response.

Despite these limitations, plain radiographs continue to play a central role in OA care. They excel at depicting the osseous changes of OA and remain valuable for excluding alternative diagnoses such as fracture, osteonecrosis, or tumor. For instance, the presence of advanced bony changes (KL 3-4) might steer a clinician towards discussing joint replacement options, whereas milder radiograph changes might encourage more focus on conservative management. A recent series in patients under 55, for example, showed that over half of them already had definite OA changes on radiography and one-third had moderate-to-severe changes (KL 3-4)[41]. Knowing this radiographic severity can prompt consideration of interventions like GAE to manage symptoms while deferring joint replacement[41]. In the context of GAE, radiography is most useful for documenting baseline structural severity and excluding end-stage degeneration, but it should be complemented by MRI to identify the inflammatory phenotype most relevant to patient selection.

Overall, plain radiography remains the first-line modality for structural assessment of knee OA because it is accessible, reproducible, and effective for documenting joint space narrowing, osteophytes, and alignment. However, because it cannot directly evaluate synovitis, cartilage damage, or BMLs, it should be viewed as a structural baseline rather than a complete imaging assessment. For GAE candidate selection, radiography defines disease stage, but MRI is required to characterize the inflammatory phenotype and refine decision-making. In practical preprocedural assessment, radiography and MRI should be interpreted together rather than in isolation. Radiographs establish structural stage and help exclude end-stage degeneration or alternative osseous diagnoses, whereas MRI refines candidate selection by identifying whether pain is more likely driven by synovitis and hypervascularity or by advanced irreversible structural damage. A pragmatic imaging workflow is to first use radiography for baseline staging, followed by MRI to assess inflammatory burden, cartilage status, meniscal injury, and BMLs. This combined approach may also reduce inter-reader variability by anchoring MRI interpretation to an established structural baseline. Figure 4 provides an algorithmic approach for GAE candidate selection[1,10,13,15,23,30,34,40-46]. In research or multidisciplinary preprocedural assessment, this can be operationalized by using a standardized radiographic severity score (e.g., KL), a single predefined MRI semiquantitative system applied consistently across cases, and dual-reader or adjudicated review when treatment selection depends on borderline inflammatory or structural findings[47-49].

Figure 4 Proposed imaging-based algorithm for genicular artery embolization candidate selection and workflow in knee osteoarthritis. Green boxes indicate start/proceed to next step, dark blue boxes are decision boxes, purple boxes are imaging boxes, yellow box is contraindication, and orange box is uncertainty/reevaluate. Plain radiography establishes structural severity, whereas magnetic resonance imaging (MRI) helps distinguish inflammation-predominant from structure-predominant disease. Favorable candidates generally show active synovitis-related MRI findings without overwhelming irreversible structural damage. Digital subtraction angiography, with or without cone-beam computed tomography, is then used to localize hyperemic synovial blush and confirm target vessels. Post-genicular artery embolization imaging response is expected to be predominantly inflammatory rather than structural. GAE: Genicular artery embolization; MRI: Magnetic resonance imaging; DSA: Digital subtraction angiography; CT: Computed tomography; OA: Osteoarthritis; KL: Kellgren-Lawrence.

MRI

MRI-based stratification for GAE in knee OA: Conventional radiographs (e.g., KL grading) remain first-line for knee OA but have well-known limitations in depicting soft tissues and early changes[6,11]. Radiography shows bone spurs and joint space loss but cannot visualize cartilage, synovium, or marrow lesions[11]. By contrast, MRI provides a whole-joint assessment with high sensitivity and specificity for tissue damage[6,42]. Notably, MRI often reveals occult pathology even in “normal” radiographic knees, with one study stating 89% of adults > 50 with KL 0 had at least one OA-related MRI abnormality[11]. Such MRI-detected lesions (cartilage defects, bone marrow edema, synovitis, etc.) underline the discordance between pain and plain films[42,50]. In short, MRI adds diagnostic value by identifying the soft-tissue and inflammatory features that radiographs miss, which is critical for stratifying patients before GAE. For example, Gill et al[43] reported that patients with advanced radiographic OA (KL grade ≥ 3) had significantly less pain improvement after GAE[43].

MRI sensitively depicts structural OA changes, and often far earlier and in greater detail than plain radiography[51]. Fluid-sensitive sequences provide a whole-joint assessment, directly visualizing non-calcified pathology such as subchondral BMLs (edema), chondral defects, and associated joint space narrowing (Figure 5). Cartilage degeneration is visualized directly, even in early symptomatic OA (KL 0-II), MRI can show cartilage lesions in over 90% of patients[51]. Subchondral BMLs, appearing as ill-defined T2-hyperintense areas, are another hallmark that radiographs cannot show. BMLs are present in roughly 50%-60% of knees with no radiographic OA and correlate with pain flares and structural progression[11,42].

Figure 5 Magnetic resonance imaging features of inflammation and structural degeneration in knee osteoarthritis. Coronal magnetic resonance imaging demonstrating subchondral bone marrow edema-like signal, chondral defects, and medial compartment joint space narrowing.

Importantly for GAE candidacy, MRI-based studies indicate that advanced structural degeneration portends poorer outcomes. Large BMLs and meniscal maceration (complex or root tears) have been linked to reduced pain improvement after GAE[19]. MRI also helps identify end-stage changes like full-thickness cartilage loss, which has emerged as the strongest MRI predictor of poor pain relief after embolization[17]. In a 54-patient study, each increase in full-thickness cartilage defect score corresponded to significantly less Western Ontario and McMaster Universities OA Index (WOMAC) pain reduction at 6 months (β = -0.63, P < 0.001)[17]. Similarly, Badar et al[18] reported that pre-GAE lateral compartment cartilage defects were linked to non-response at 3 months (P approximately 0.04). Badar et al[18] likewise observed that lateral meniscal abnormalities were significantly associated with failure to achieve pain relief at 3 months. At 6 months, van Zadelhoff et al[17] showed that the presence of BMLs was associated with significantly less WOMAC pain improvement (β approximately -0.52 for BML presence). Consistently, the machine-learning MRI study by Dablan et al[52] found that responders had significantly lower baseline BML scores, and the MRI bone marrow score was the single strongest correlate of pain reduction (Pearson r = -0.43, P < 0.001).

Taken together, these findings suggest that patients with predominantly mechanical, end-stage disease (bone-on-bone contact, gross instability) derive less benefit from GAE. Conversely, those with moderate structural OA with intact or only partially lost cartilage, smaller BMLs, and meniscal integrity preserved or only mildly degenerated tend to respond better[17,19]. Thus, MRI allows us to stratify out structural non-responders (in whom irreversible damage dominates) from those whose pain stems more from treatable lesions. However, these associations should be interpreted cautiously, as published studies vary in cohort composition, imaging protocols, follow-up duration, and outcome definitions. As a result, MRI findings that appear predictive in one study may not consistently reproduce across other populations.

Inflammatory MRI features (synovitis and effusion): Identifying synovitis is central to an imaging-first GAE workflow, since embolization targets synovial pathology. MRI excels at detecting synovial inflammation that might be missed on exam or radiography[50,51]. Quantitatively, synovial hypertrophy is exceedingly common across OA stages. A CE-MRI study of 111 OA patients showed 89.2% had synovitis of at least moderate severity (grade 2) in ≥ 1 subregion. Furthermore, 39.6% had pockets of severe synovitis (grade 3)[44]. Crucially, these inflammatory changes are modifiable, unlike cartilage loss, synovitis can regress with targeted therapies. GAE aims to attenuate this synovial perfusion and inflammation. Thus, a preprocedural MRI that confirms an inflammation-predominant phenotype (e.g., abundant synovial pannus, effusion, Hoffa fat pad edema) helps select ideal candidates for embolization. Van Zadelhoff et al[17] reported that moderate-to-large effusion-synovitis (MRI effusion grade 2-3) at baseline corresponded to nearly 3 points less improvement in WOMAC pain scores after GAE (β approximately -2.99, P < 0.05). In this context, effusion-synovitis refers to the combined appearance of joint fluid and synovial thickening on non-contrast MRI, where the two cannot be reliably distinguished. However, findings have not been uniform. Badar et al[18] observed no significant predictive value of preprocedural synovitis in their 52-knee series (synovitis severity showed no correlation with 3-month outcome, P = 0.809). Notably, patients with an excessive total burden of disease on MRI fare worse, with Badar’s study discovering that having ≥ 4 distinct MRI lesions (e.g., concurrent cartilage, meniscal, ligamentous, and marrow abnormalities) was associated with a significantly lower likelihood of pain improvement (P = 0.004)[18]. These conflicting findings highlight that synovitis, while important mechanistically, has not yet emerged as a uniformly reliable standalone predictor of response. Differences in MRI technique, scoring methods, sample size, and timing of outcome assessment are likely contributing to this inconsistency. In current practice, these findings suggest that MRI features should be interpreted collectively and in clinical context rather than used on their own to determine a patient’s candidacy for GAE.

Contrast-enhanced vs non-contrast MRI for synovitis: A standardized knee MRI protocol for GAE evaluation typically includes fluid-sensitive sequences and, in selected cases, contrast-enhanced imaging. Common sequences are proton-density (PD) or T2-weighted fat-saturated images (axial, sagittal, coronal) to visualize effusions, cartilage morphology, and BMLs, along with T1-weighted images for anatomy. Fat-suppressed T2 or PD sequences are highly sensitive to fluid and edema, making joint effusions and synovial fluid readily apparent as hyperintense signals[10,45].

However, non-contrast MRI cannot always differentiate synovial tissue from fluid. On T2/PD fat-sat images, a thickened synovial membrane and pure fluid both appear bright, so the finding of an effusion on these sequences represents a combination of fluid and hypertrophic synovium[44]. Signs like Hoffa’s fat pad synovitis (ill-defined T2 signal or low T1 signal within the infrapatellar fat) can suggest synovial inflammation on non-contrast scans[42].

CE-MRI is considered the reference standard for visualizing synovitis[10,42]. After intravenous gadolinium, inflamed synovium shows avid enhancement and thickened, frond-like morphology, which stands out distinctly against non-enhancing joint fluid[46]. In one study, 96% of knees with a joint effusion on MRI also showed definite synovial enhancement on CE-MRI, but even 70% of knees without an effusion had areas of synovitis when contrast was used[44]. Indeed, Epelboym and Guermazi (2025) stressed that incorporating contrast MRI and dedicated synovitis scoring could refine future GAE candidate selection[23]. Current MRI-based scoring in OA without contrast often combines joint effusion and synovial thickening into an “effusion-synovitis” grade, reflecting the fact that fluid and synovium cannot always be reliably distinguished on non-contrast imaging. By contrast, CE-MRI allows a more granular synovial assessment and may better identify patients with active synovitis most likely to benefit from GAE. Moving forward, consensus on MRI-based synovitis evaluation will be key in standardizing selection criteria[11].

From a candidate-selection standpoint, CE-MRI may be most useful when non-contrast imaging shows equivocal effusion-synovitis or when inflammatory burden is central to treatment decisions. Non-contrast MRI remains useful for structural assessment and broad screening, but because it cannot reliably separate synovial thickening from joint fluid, it may overestimate or obscure true inflammatory activity. Accordingly, CE-MRI provides a more specific assessment of active synovitis because non-contrast effusion-synovitis cannot fully separate fluid from hypertrophied synovium. Therefore CE-MRI may improve phenotypic classification in patients being considered for GAE[22,27,53].

Current evidence remains limited by small cohorts, short follow-up intervals, heterogeneous inclusion criteria, and variation in MRI protocols and outcome definitions. Few imaging biomarkers have been prospectively validated across independent populations. Until larger prospective studies establish standardized thresholds, imaging biomarkers should be interpreted as supportive tools rather than definitive determinants of treatment eligibility. Accordingly, imaging findings should currently be used to support, rather than dictate, GAE decision-making.

STANDARDIZED SEMIQUANTITATIVE SCORING SYSTEMS

To standardize MRI interpretation in OA, several semiquantitative scoring systems have been developed. These scoring tools are especially useful in research and can be applied to stratify patients for interventions like GAE. Table 2 summarizes the key features of semiquantitative knee OA MRI scoring systems and their use in published GAE imaging assessments[10,17,20,52]. All three major systems, WORMS, BLOKS, and MOAKS, grade the severity of lesions in each joint subregion on MRI[20]. They encompass multiple tissue features such as cartilage integrity, BMLs, osteophytes, meniscal damage, ligament status, and also synovial abnormalities[20]. Among these systems, MOAKS may be the most practical framework for structured GAE-oriented MRI review because it separately captures effusion-synovitis, Hoffa-synovitis, cartilage loss, and BML burden while improving inter-reader consistency.

Table 2 Comparison of semiquantitative magnetic resonance imaging scoring systems for knee osteoarthritis relevant to genicular artery embolization (whole-organ magnetic resonance imaging score, Boston-Leeds osteoarthritis knee score, and magnetic resonance imaging osteoarthritis knee score).

Feature	WORMS	BLOKS	MOAKS
Primary goal	Whole-organ assessment across 14 knee subregions	Focuses on lesion detail	Unified system integrating the best aspects of WORMS and BLOKS
Synovitis assessment	Combined score for effusion and synovitis; did not separately quantify them	Separated effusion from Hoffa’s fat pad synovitis	Dual scoring: Evaluates Hoffa’s fat pad synovitis and effusion-synovitis separately (0-3 scale)
BMLs	Graded by size in various subregions	More granular scoring for BML size and number	Simplified grading (0-3) by size with a clearer, more reproducible definition
Meniscal damage	Combined tear types and extrusion into one metric	More detailed evaluation, including hypertrophy and maceration	Improved percentage-of-area scoring to enhance sensitivity to small changes. Meniscal damage is scored by tear/maceration status and extent in each meniscal subregion, with separate assessment of meniscal extrusion (0-3 scale for extrusion)
Reliability	Good but not perfect; some features overlap	Improved inter-reader reliability for specific features	Highest clinical adoption; achieves substantial-to-excellent agreement (Kappa 0.61-1.0)
GAE application	Used to summarize whole-joint MRI disease burden in GAE cohorts. Can support prognostic modeling when WORMS subscores are used as imaging inputs	Useful for separating synovial membrane thickening from joint fluid on non-contrast MRI. Supports synovitis-phenotype characterization relevant to GAE candidate selection	Supports characterization of the synovitis-phenotype. Provides a standardized framework for candidate selection. Correlates baseline synovial/cartilage health to clinical pain response

WORMS: Whole-organ magnetic resonance imaging score; BLOKS: Boston-Leeds osteoarthritis knee score; MOAKS: Magnetic resonance imaging osteoarthritis knee score; BMLs: Bone marrow lesions; GAE: Genicular artery embolization; MRI: Magnetic resonance imaging.

WORMS

WORMS, introduced in 2004, was a pioneering method that scored OA features on a scale (generally 0-4 or 0-6) in various knee subregions[20]. For example, cartilage in each compartment is graded 0 (normal) up to 6 (diffuse full-thickness loss) in WORMS. WORMS also includes scores for effusion/synovitis and for periarticular features like cysts and attrition[20]. However, some limitations became apparent. For example, the original WORMS meniscus score combined tear types and extrusion into one metric, and it did not separately quantify synovial thickening vs effusion. Reliability was good but not perfect for all features. In the context of GAE, WORMS has been used not only to describe whole-joint disease burden, but also to support prognostic stratification. Dablan et al[52] applied WORMS-based MRI inputs within a machine-learning framework and found that bone marrow and ligament-related sub-scores were among the strongest predictors of achieving ≥ 50% pain relief after GAE[52].

BLOKS

BLOKS, developed around 2008, aimed to refine certain elements of WORMS[20]. BLOKS introduced a more granular scoring for BMLs (assessing their size and number) and evaluated meniscal damage with more detail (including meniscal hypertrophy and partial maceration)[20]. Importantly, BLOKS was among the first to separately score synovitis in the infrapatellar fat pad (Hoffa’s synovitis) and synovial effusion as distinct features, rather than a single effusion score[10]. This allowed investigators to distinguish synovial membrane thickening from mere fluid accumulation on non-contrast MRI. BLOKS showed improved inter-reader reliability for certain features, but some aspects (like its BML scoring method) were seen as cumbersome or overlapping[20]. This separation is particularly relevant for GAE candidate selection, where synovitis phenotype is central and conflating effusion with synovial thickening can obscure inflammatory-driven disease.

MOAKS

MOAKS, published in 2011 by an international group, integrated the best aspects of WORMS and BLOKS into a unified system[20]. MOAKS simplified scoring where needed and added new features. Cartilage damage is scored semiquantitatively by percentage of surface area involved (0 = none up to 3 ≥ 75% area) with half-grades for depth (partial vs full thickness), thus enhancing sensitivity to small changes. BMLs are graded 0-3 by size in each subregion, with a clearer definition than prior tools. Critically, MOAKS explicitly grades synovitis on two fronts, which includes Hoffa’s fat pad synovitis (infrapatellar fat signal changes) and effusion-synovitis (joint fluid + synovium), each on a 0-3 scale (absent to severe)[10]. This dual scoring acknowledges that inflamed synovium in the fat pad can be assessed separately from synovial effusions on MRI. Examples of investigational imaging thresholds used to define a more inflammation-predominant phenotype include MOAKS Hoffa-synovitis ≥ 2 or effusion-synovitis grade 2-3 in the absence of extensive full-thickness cartilage loss, although these cutoffs have not yet been prospectively validated. Overall, MOAKS has become the most widely adopted MRI scoring system and has demonstrated very good to excellent reliability in reader studies[20]. Almost all feature grades in MOAKS achieve interobserver kappa values in the 0.61-0.80+ range (substantial agreement), with many in the approximately 0.8-1.0 range after adequate reader training. For instance, cartilage, BML, and osteophyte scores have shown high concordance between readers. A few features (such as subtle Hoffa-synovitis or small tibial osteophytes) have slightly lower agreement, but overall, the reproducibility is strong[20]. In practice, these systems may improve inter-reader consistency by converting qualitative MRI impressions into structured semiquantitative assessments, although standardized cutoffs for GAE candidacy remain to be established. AI-assisted quantitative approaches for effusion-synovitis and related inflammatory MRI features may further reduce inter-reader variability and improve discrimination beyond ordinal semiquantitative scoring, although these methods still require external validation and workflow standardization before routine clinical use[54-56].

In the context of GAE, these MRI scoring systems provide an objective framework to stratify and monitor patients based on imaging. They convert the rich qualitative data of an MRI into numeric sub-scores for inflammation and structural damage. For example, using a scoring system, one could define an inflammation-predominant knee as one with high synovitis scores (e.g., MOAKS Hoffa-synovitis ≥ 2) but only moderate cartilage loss (no MOAKS score > 2), criteria that could be used for trial inclusion. In fact, research protocols in GAE are already leveraging these tools. In practical terms, currently available studies suggest that moderate-to-large effusion-synovitis (grade 2-3), an overall MRI lesion burden of ≥ 4 abnormalities, and extensive full-thickness cartilage loss may serve as useful research-based markers when distinguishing inflammation-predominant from structure-predominant disease. However, these should not yet be interpreted as validated clinical thresholds for GAE candidacy. The pivotal study by van Zadelhoff et al[17] scored baseline MRIs with MOAKS and found that knees with higher cartilage defect scores and effusion-synovitis grades had significantly less pain improvement after embolization. Conversely, patients without full-thickness cartilage lesions (MOAKS cartilage ≤ 2) tended to have better outcomes, reinforcing the use of such scores to prospectively select candidates. Thus, semiquantitative scoring may help identify knees at risk of poorer outcome. This strategy will triage patients whose knee pain is likely mediated by reversible inflammatory processes, ideal candidates for GAE, while excluding those with irreversible structural end-stage OA less likely to benefit[17]. Such an approach may support more standardized imaging-based selection criteria in future GAE trials. Ultimately, MRI provides the granular anatomic and inflammatory detail needed to optimize GAE as a precision intervention for knee OA. Nevertheless, the current role of MRI scoring systems in GAE remains more promising than concrete. Although semiquantitative scores improve imaging consistency, values selected for the threshold for candidates have not been standardized, and no single score has been prospectively validated as a predictor for the success of the treatment. Therefore, these tools should currently be viewed as aids to structured assessment rather than as absolute decision-making criteria.

INTRAPROCEDURAL IMAGING GUIDANCE

Fluoroscopic angiography (digital subtraction angiography) in GAE

Digital subtraction angiography (DSA) remains the cornerstone for intraprocedural mapping of genicular arteries and identifying pathologic hyperemia in knee OA. Angiography from the distal superficial femoral artery with both early and delayed DSA can clarify genicular anatomy and improve visualization of hyperemic target vessels, and should ideally be performed with consistent injection technique because hyperemia appearance depends on injection parameters[57]. On angiography, the normal genicular arterial network is highly variable and richly anastomosed[58,59]. Okuno et al[14] provide a quantitative description of how frequently multiple feeding territories contribute to angiographic hypervascularity. Abnormal neovessels originated from a mean of 3.2 arteries per knee, most commonly involving the descending genicular artery (DGA) and both inferior genicular arteries, with additional contributions from the superior patellar, superior genicular, median genicular, and anterior tibial recurrent arteries[14]. This supports the practical need for systematic multi-vessel interrogation during DSA rather than assuming a single dominant feeder. They also describe a pragmatic intraprocedural localization sign, during infusion of contrast or embolic material in the pain-responsible region, patients sometimes reported pain/itching/heat at their typical symptomatic site (“evoked pain”), which the authors used to help identify the culprit artery[14].

The genicular arterial system comprises five principal vessels: The superior lateral genicular artery, superior medial genicular artery (SMGA), inferior lateral genicular artery, inferior medial genicular artery, and DGA[15,16,59]. These vessels form an anastomotic network around the knee joint, providing collateral pathways that must be considered during embolization planning. Anatomical studies have revealed considerable variation in genicular artery origins, branching patterns, and anastomotic connections. The classification system proposed by Callese et al[59], based on intraoperative CBCT analysis, categorizes these variants into types that have important implications for procedural planning. Common variants include shared origins of the SMGA and DGA from the superficial femoral artery, accessory genicular arteries arising from the popliteal artery, and anomalous contributions from the lateral circumflex femoral artery[59].

A recent multicenter angiographic analysis by Taheri Amin et al[16] characterized the origin variants, branching patterns, and anastomotic networks of the genicular arteries as encountered during GAE procedures. The study identified three distinct patterns of shared origins involving the middle genicular artery and described the angiographic appearance of anastomotic connections that may predispose to non-target embolization if not recognized (Table 3)[16,58-60]. Thus, thorough fluoroscopic angiography is critical to delineate each genicular vessel and any atypical anatomy before embolization.

Table 3 Genicular artery anatomy and common variants.

Artery	Typical origin	Common variants
Superior lateral	Popliteal artery	Early branching, accessory vessels
Superior medial	Popliteal artery	Shared origin with DGA from SFA
Inferior lateral	Popliteal artery	Origin from anterior tibial artery
Inferior medial	Popliteal artery	Multiple branches, anastomoses
Descending genicular	SFA (adductor canal)	Shared origin with SMGA

SFA: Superficial femoral artery; DGA: Descending genicular artery; SMGA: Superior medial genicular artery.

CBCT can further strengthen intraprocedural planning by improving depiction of arterial variants, collateral pathways, and the spatial relationship between target vessels and adjacent skin or soft tissues[59]. In cases with complex branching or overlapping vascular territories on DSA, CBCT may improve confidence in feeder-vessel identification and help reduce non-target embolization. This can then alter embolization while simultaneously reducing the risk to potentially embolize non-target vessels.

A key angiographic finding in osteoarthritic knees is synovial hyperemia, which appears as a region of dense, tumor-like vascular “blush” on late arterial or capillary phase imaging[61]. This hypervascular staining represents proliferative synovium and neovasculature in inflamed regions of the joint. Recognizing and localizing this blush is central to GAE, and the goal is to embolize the feeding arteries until this pathologic hyperperfusion is “pruned” while preserving flow in normal parent arteries[61].

Although angiographic blush is still interpreted semiquantitatively in most clinical practice, more objective approaches are emerging. These include standardized blush grading, documentation of early venous opacification, and use of adjunctive perfusion tools to reduce operator-dependent subjectivity[29,62]. Recent human GAE work suggests that segmentation-based blush reduction ratio may provide an objective procedural endpoint, with angiographic ‘pruning’ corresponding approximately to an 80% reduction in blush size, although this requires external validation before adoption as a routine standard[63]. At present, however, no universally accepted quantitative threshold for adequate hyperemia or embolization endpoint has been established for human GAE procedures.

Successful procedural outcomes are documented by comparing pre-treatment angiography showing dense synovial hyperemia with post-treatment images demonstrating a clear pruning of these pathologic neovessels (Figure 6A and B). Importantly, the distribution of angiographic blush often correlates with patients’ pain distribution and MRI findings of inflammation. For example, Rouzbahani et al[62] describe an arterial-phase MR angiogram that pinpointed the origins of the superior and inferior lateral genicular arteries, and a delayed-phase image showing an intense inflammatory blush in the lateral compartment that corresponded exactly to the patient’s pain locus and the subsequent DSA target during GAE. Similarly, conventional angiography frequently demonstrates hypervascular zones matching regions of bone marrow edema or synovitis seen on pre-procedure MRI[61]. This concordance allows the radiologist to map pain generators to specific arterial territories. In practice, careful correlation of clinical exam with angiographic anatomy is performed interventionalists will often mark the patient’s skin overlying the tender compartment and then identify the corresponding genicular branch on DSA[64]. Targeting such a vessel showing synovial blush can directly address a major pain generator. In summary, DSA allows real-time identification of the genicular arteries feeding pathologic synovium and facilitates a compartmental approach, treating each arterial territory corresponding to the patient’s symptoms. The primary embolization endpoint is pruning hyperemic neovessels while maintaining flow in the parent genicular artery, and emerging approaches describe objective assessment using perfusion tools to reduce subjectivity[62]. A concrete example of how angiographic hyperemia can be operationalized is provided by a controlled rabbit OA GAE model, where intra-procedural staining was graded as 0 (no abnormal staining), 1 (tumor-like blush), or 2 (tumor-like blush with early venous opacification), and technical success required that abnormal staining was barely visible on post-embolization angiography. Although preclinical, this type of explicit blush grading is directly aligned with the key intraprocedural imaging problem in human GAE, thus reducing subjective interpretation and enabling more consistent reporting of what was actually embolized[29].

Figure 6 Digital subtraction angiography before and after genicular artery embolization. A: Pre-treatment digital subtraction angiography demonstrating focal synovial hyperemia characterized by a dense periarticular vascular blush in the region corresponding to the patient’s symptomatic compartment; B: Post-treatment angiography following embolization showing marked reduction of pathologic hypervascular blush with preservation of parent arterial flow. Resolution of synovial hyperemia represents the intended intraprocedural endpoint of genicular artery embolization.

POSTPROCEDURAL IMAGING FOLLOW-UP

Role of radiography after GAE

Conventional weight-bearing knee radiographs remain the cornerstone for baseline OA evaluation, but after GAE they rarely show any acute change[62]. Because GAE works by devascularizing the synovium, post-embolization radiography predictably reveals no immediate cartilage loss or joint-space widening/narrowing[18]. In practice radiographs are thus reserved for safety follow-up as they can rapidly exclude mechanical complications such as occult fracture or rapid joint collapse. By contrast, subtle bone events (e.g., asymptomatic osteonecrosis/infarcts) have been reported on MRI after GAE, which plain films would not detect until frank collapse[62]. Over longer follow-up, serial radiographs can document structural OA progression (for example, measuring joint-space narrowing in trials), but such changes evolve very slowly. In fact, studies of GAE report essentially no change in joint structure on imaging at 2 years (aside from reduced synovitis)[65]. In the largest longer-horizon MRI follow-up cited in this space, Okuno et al[14] performed MRI before treatment and again at 2 years in a subset (35 knees), analyzing images using WORMS; they reported significant improvement in the synovitis component at 2 years (P = 0.0016) and specifically noted no osteonecrosis or other MRI evidence of aggressive degenerative progression. This highlights that the most demonstrable interval imaging signal after GAE is inflammatory rather than structural[14]. In summary, radiography is not a useful biomarker of GAE response and should not be over-interpreted post-procedure as it reinforces imaging literacy by limiting use to exclusion of complications and research endpoints, not as a measure of symptomatic improvement[18]. Table 4 summarizes expected post-GAE imaging findings by modality and follow-up interval, based on published radiographic and MRI follow-up data and interpretive guidance[12,13,18,62,65-67]. Given these limitations, advanced modalities are generally needed to image the synovial and soft tissue changes that underlie GAE’s clinical effects.

Table 4 Expected post-genicular artery embolization imaging findings by follow-up interval (0-12 months).

Modality	0-1 month (early)	1-3 months (intermediate)	6-12 months (long-term)
Radiography (X-ray)	No structural response expected. Primary role is safety to exclude fracture or rapid collapse	OA hallmarks (osteophytes, joint space narrowing) remain unchanged; not considered a response biomarker	Useful mainly for longitudinal progression tracking in research cohorts
MRI	May show decreased synovial enhancement transient subchondral marrow signal abnormalities may be seen early and warrant interval reassessment	Most consistent imaging response window: Decreased synovial enhancement/thickness; Effusion-synovitis often trends downward	Reduced synovitis generally persists; cartilage loss and osteophytes remain stable and non-reversible

MRI: Magnetic resonance imaging; OA: Osteoarthritis.

MRI as the primary post-procedural imaging modality

After GAE, MRI generally shows reduced signs of synovial inflammation. In both contrast-enhanced and standard MRI, treated knees tend to have thinner synovium and less synovial fluid than at baseline. For example, Dablan et al[12] found that the mean synovial enhancement score on CE-MRI fell from 5.1 to 2.9 by 3 months post-GAE (P < 0.001). Similarly, Hindsø et al[13] reported significant drops in synovitis scores on MRI: By 6 months, both a whole-knee synovitis score and a local synovitis thickness score were significantly lower than at baseline. Non-contrast MRI can’t distinguish effusion from synovium, but overall effusion/synovitis scores trended down on follow-up, even if not always reaching significance[13]. In practice, reduced synovial enhancement and thickness are the hallmark MRI changes after GAE[12,13]. In contrast, Hoffa-fat-pad edema or fibrosis, when present, would be expected to decrease along with synovitis, but the published studies focus on synovial measures and do not separately quantify Hoffa changes.

Bone and cartilage changes behave differently. GAE is not structure-modifying, so MRI shows no new cartilage regrowth or narrowing of OA features after treatment. In most patients, pre-existing cartilage defects, osteophytes, and joint space loss remain essentially unchanged on follow-up MRI[66]. For example, a large radiology study noted that > 75% of knees had full-thickness cartilage loss and bone marrow edema before embolization, and no study to date has shown reversal of such lesions after GAE[18]. Likewise, osteophytes and joint-space narrowing persist after GAE[66].

BMLs and subchondral changes can fluctuate early on. Many knees with moderate OA have subchondral edema at baseline, and GAE may reduce this edema[18]. In one prospective study, knees with baseline BMLs showed a significant shrinkage in lesion area and volume at 3 months post-GAE[67]. Conversely, a few patients can show transient ischemic-looking lesions shortly after embolization. For example, Hindsø et al[13] saw three cases of sharply demarcated subchondral T1-dark lesions at 1 month (in regions supplied by treated arteries) that resembled osteonecrosis; importantly, all had resolved by the 6-month scan with no permanent damage[8]. These early transient marrow or subchondral signal abnormalities likely represent reversible ischemic or ischemia-reperfusion-related change after embolization rather than irreversible osteonecrosis, particularly when they resolve on interval follow-up (Table 4)[8,13,67]. Other reports have noted occasional asymptomatic focal osteonecrosis (even patellar) on early follow-up, but these also tend to resolve within a few months[8]. In summary, any new marrow signal can appear early but typically normalizes by 6-12 months. By one year after GAE, most treated knees show only their baseline OA changes and no new bone lesions.

Effusions may also modestly diminish, though changes in fluid alone often lag behind synovial thinning[13]. Early post-procedure MRIs may occasionally show small subchondral defects or edema, but by about 6 months these have generally resolved[13]. By 6-12 months, the reduction in synovitis is sustained or even enhanced, and treated knees often remain in a lower state of synovial inflammation compared to pre-GAE[12,13]. At that point, MRI typically shows resolved or nearly normal synovial thickness in the embolized regions. In contrast, no late improvement is seen in articular cartilage or osteophytes; joint space narrowing on MRI (and radiographs) remains at baseline levels throughout the first year[18,66].

Joint fluid may be slightly reduced, but significant effusion often persists. Hoffa pad signal (when inflamed) likely lessens alongside synovium. Marrow edema lesions often shrink or disappear, though transient subchondral defects can appear at 1 month and then resolve by 6 months[13,67]. Cartilage defects, osteophytes, and joint-space loss do not improve on MRI in the months following GAE[18,66]. These imaging changes, especially the drop in synovitis, have been documented in recent studies of GAE, reflecting the procedure’s effect on the inflammatory component of knee OA[12,13]. However, interpretation of post-procedural imaging still requires caution, as the relationship between imaging improvement and clinical pain relief is not always linear, and often times structural lesions frequently remain unchanged despite the patient feeling symptomatic relief. A concise interpretation guide highlighting expected post-GAE imaging response patterns and common interpretive pitfalls is provided in Tables 5 and 6[8,12,13,18,62,65,66].

Table 5 Post-genicular artery embolization imaging interpretation guide for knee osteoarthritis showing expected response patterns.

Category	Imaging findings and interpretation
Expected response patterns	Reduced synovial enhancement on contrast-enhanced MRI. Reduced synovial thickness/volume. Lower effusion-synovitis grade when scored. Interval normalization of early marrow signal changes when present

This table highlights expected post-procedural imaging response patterns, meaning findings that generally reflect treatment-related improvement in inflammation-predominant disease on follow-up imaging. These include reduced synovial enhancement on contrast-enhanced magnetic resonance imaging, decreased synovial thickness or volume, lower effusion-synovitis grade when scored, and interval normalization of early marrow signal changes when present. MRI: Magnetic resonance imaging.

Table 6 Post-genicular artery embolization imaging interpretation guide for knee osteoarthritis showing common pitfalls.

Category	Imaging findings and interpretation
Interpretive pitfalls	Persistent osteophytes should not be interpreted as treatment failure. Static cartilage defects should not be interpreted as treatment failure. No change in joint space narrowing on radiography does not exclude clinical response. Early focal T1-hypointense subchondral signal should not be labeled osteonecrosis without follow-up

This table highlights interpretive pitfalls, meaning findings that may persist or fluctuate after genicular artery embolization and should not be automatically overcalled as treatment failure or irreversible complication without clinical correlation and interval comparison. These include persistent osteophytes, unchanged cartilage defects, no radiographic change in joint-space narrowing, and early focal subchondral T1-hypointense signal that should not be labeled osteonecrosis without follow-up assessment.

A practical interpretation point is that post-GAE MRI changes are expected to be predominantly inflammatory rather than structural. Persistent reduction in synovial enhancement or effusion-synovitis may accompany clinical improvement. Early transient marrow or subchondral signal abnormalities should therefore be interpreted cautiously, as they may represent reversible post-embolization ischemic change rather than progressive osteonecrosis[8,13,67]. Overall, the most consistent post-GAE MRI response pattern is sustained reduction in synovial inflammation, whereas structural OA lesions generally remain stable over short- and intermediate-term follow-up.

CE-MRI and quantitative imaging biomarkers

CE-MRI is now regarded as the gold-standard for knee synovitis imaging, because intravenous gadolinium makes the highly vascular inflamed synovium bright on T1-weighted images while effusions stay dark[10]. In practice this means CE-MRI can distinguish true synovial hypertrophy from simple joint fluid, a distinction non-contrast exams cannot reliably make[68]. Semiquantitative scoring systems based on CE-MRI are highly reproducible, and the summed scores show inter-reader intraclass correlation coefficient (ICC) approximately 0.94, and many centers also incorporate DCE sequences to quantify perfusion[68]. DCE-MRI generates continuous biomarkers of synovial blood flow (e.g. K trans, IAUC) that are very repeatable and sensitive. In multiple studies K trans in particular has shown excellent test-retest reliability (ICC approximately 0.84-0.90) and the best discrimination/sensitivity to change of all DCE parameters[28,62]. Importantly, DCE metrics tend to correlate more closely with pain and histologic inflammation than static scores, making them a rigorous standard. A thorough follow-up approach may combine rapid post-contrast MRI with standardized synovitis scoring and quantitative DCE analysis[10]. This approach maximizes reproducibility and enables guideline-directed assessment of post-GAE synovitis[10,68].

Advanced cartilage compositional techniques such as T2 mapping may also prove useful in future follow-up studies by assessing cartilage matrix integrity beyond routine morphologic imaging, although their role in monitoring GAE response has not yet been established. Similarly, DCE-MRI may offer a more objective way to track symptom-related synovial perfusion changes after embolization because DCE-MRI parameters correlate with synovitis severity and inflammatory vascularity. However, larger prospective studies are needed before it can be recommended as a routine postprocedural biomarker.

FUTURE DIRECTIONS

A major imaging gap in GAE is the lack of standardized preprocedural and intraprocedural criteria that can reliably identify treatable hypervascular synovitis. For example, one human series reported 2 of 38 patients demonstrated no angiographic evidence of neovascularity and therefore did not undergo embolization, underscoring that the angiographic target was not uniformly present, even among symptomatic OA referrals[30]. This highlights the need for future work focused on objective and reproducible imaging markers of synovial hyperemia, such as standardized blush grading, quantitative readouts, or validated preprocedural imaging surrogates[30]. Despite the promise of these emerging technologies, most still are investigational when it comes to GAE. Accordingly, these technologies should be considered future perspectives rather than current standard practice. Although they are conceptually promising, most studies are limited by small sample sizes, retrospective design, lack of prospective validation, and variability in how the images are acquired and then interpreted. Factors such as cost, accessibility, and standardization also remain important barriers before these approaches can be incorporated into routine clinical decision-making.

AI-enhanced MRI for synovitis phenotyping and standardized stratification

Emerging AI/MRI methods may improve quantification of synovitis and other abnormalities seen in the knee joint by automating the assessment of synovial thickness, effusion, BMLs, cartilage damage, and meniscal injury. For example, Wang et al[69] developed a deep-learning pipeline that classified synovitis with accuracy approximately 0.86 [area under the curve (AUC) = 0.83], while Astuto et al[70] reported AUC values of 0.83-0.93 for detection of cartilage defects, BMLs, meniscal tears, and anterior cruciate ligament injury. Feuerriegel et al[71] also showed that a deep-learning-accelerated noncontrast fluid-attenuated inversion recovery sequence acquired in approximately 1.5 minutes produced synovitis scores comparable to conventional gadolinium-enhanced MRI. These findings suggest that AI could support faster and more reproducible imaging-based phenotyping and follow-up in GAE. Although these data are promising, these avenues still lack prospective validation, standardized implementation, and demonstrated clinical utility that is required for routine GAE decision-making.

AI methods are also being explored to predict OA progression and treatment response. Although no prognostic AI model has yet been validated for routine clinical use in knee OA, publication volume in AI-based OA imaging increased markedly from approximately 17 to 154 articles per year between 2021 and 2022[72]. In the context of GAE, these approaches may eventually help identify patients more likely to experience pain relief by linking baseline imaging phenotypes, such as synovitis burden, with treatment response. However, their clinical role remains investigational.

Given the widespread use of radiography in knee OA, AI and radiomic approaches have also been explored to extend structural assessment beyond conventional KL grading. One OA Initiative study reported approximately 68% balanced accuracy for predicting progression to advanced OA over 8 years[73]. A 2023 convolutional neural network achieved approximately 92% test accuracy for distinguishing KL 0 from KL ≥ 2 knees[33], while a 2024 segmentation-based pipeline achieved 98.5% accuracy for separating mild (KL 0-2) from severe (KL 3-4) disease, although it was still difficult to differentiate adjacent-grades[74]. In GAE, these tools are best viewed as adjunctive structural stratification methods, whereas MRI remains necessary for inflammatory phenotyping and candidate selection.

Hybrid and spectral imaging approaches: Positron emission tomography/MRI and advanced CT techniques

Positron emission tomography (PET) combined with MRI is an emerging approach for assessing inflammatory activity in OA. Fluorodeoxyglucose-PET/MRI can visualize metabolically active synovium, and novel tracers targeting macrophages, fibroblasts, and angiogenesis are under investigation[26,74]. These methods could eventually help identify knees with high inflammatory or neovascular burden and complement anatomic MRI assessment, but their role in GAE remains investigational and has not been validated for routine candidate selection.

Advanced CT techniques may also provide additional structural and inflammatory information. Dual-energy CT with virtual non-calcium reconstruction has shown high accuracy for detecting BMLs; in knee OA, Chen et al[75] reported sensitivity of approximately 90% and specificity of approximately 99%. Photon-counting CT offers ultrahigh-resolution imaging and iodine mapping, with potential applications in depicting synovial enhancement and small periarticular vessels[72,76]. In summary, these advanced CT techniques could play a future role in procedural planning; however, these approaches remain exploratory in GAE and are currently limited by cost, availability, and lack of standardized clinical protocols.

Quantitative perfusion imaging and emerging molecular contrast platforms

DCE-MRI provides quantitative assessment of synovial perfusion and permeability and may serve as an objective biomarker of inflammatory activity[26]. Parameters such as K trans and IAUC have been associated with synovitis severity and could eventually help quantify treatment-related changes after GAE. Other non-contrast perfusion techniques, including arterial spin labeling and intravoxel incoherent motion MRI, are also being explored, although validation remains limited[26].

High-sensitivity power Doppler, microvascular ultrasound, and superb microvascular imaging can detect slow synovial flow and are already used in rheumatology to assess synovitis. Contrast-enhanced ultrasound has shown moderate correlation with MRI measures and may better highlight actively inflamed synovium, although ultrasound remains operator dependent[77]. These techniques may eventually offer a low-cost bedside method to assess GAE response by demonstrating reduced synovial perfusion[77].

Beyond conventional imaging, molecular “smart” agents are also under development. Examples include iron-oxide nanoparticles, tagged antibodies, and experimental gadolinium-chelate nanoparticles linked to matrix metalloproteinase-13 peptides, which have shown high T1 relaxivity and cartilage targeting; however, these agents remain preclinical[78].

Collectively, these innovations point to a future where knee imaging yields quantitative, cell-specific biomarkers. In practice for GAE, this means selecting patients with imaging-proven inflammatory phenotypes (e.g., high perfusion synovitis on MRI or PET uptake) who may respond best. This means using high-resolution angiographic CT or MR to map the inflamed synovium and its feeder vessels before embolization and applying quantitative perfusion or PET markers to monitor reduction in synovial inflammation after GAE. Hybrid metabolic imaging (PET/MRI) and novel tracers may play a role to improve understanding of OA-related synovitis in the future, but their role in GAE remains experimental and has not yet been established for routine clinical use[74]. Overall, these technologies should currently be viewed as research tools with future potential rather than established components of standard imaging workup for GAE.

CONCLUSION

In conclusion, imaging is integral to every phase of the GAE workflow. Prior to embolization, conventional and advanced MRI are used to confirm OA, quantify cartilage damage, and exclude alternative pain generators[8,62]. During the procedure itself, DSA is indispensable, and DSA defines the genicular arterial anatomy and identifies the blush of hypervascular synovial tissue to target for embolization[62]. After embolization, follow-up imaging (primarily MRI and increasingly ultrasound) can document outcomes and screen for complications[62]. In short, an imaging-driven strategy, from MRI-based patient selection through intra-procedural DSA guidance and post-procedure MRI/ultrasound monitoring, provides an objective framework for safely delivering GAE and evaluating its effect. Critically, this imaging framework also enables a stratified-care approach. Advanced MRI biomarkers can define an inflammatory OA phenotype that is most amenable to embolization[6,62]. Looking forward, several emerging innovations promise to deepen the role of imaging in GAE. Despite this progress, key gaps remain. Most proposed imaging biomarkers in GAE have not yet been validated prospectively as predictors of clinical response[6,79]. Development of standardized imaging guidelines for GAE is also needed. To address these issues, future research should explicitly embed advanced imaging into GAE trials and protocols. Ultimately, by validating objective imaging biomarkers and harmonizing assessment methods, future work can ensure that imaging innovation in GAE translates into more precise patient selection, safer procedures, and better monitoring of therapeutic effect. In practice, radiography establishes structural stage, MRI refines inflammatory phenotype and exclusionary structural burden, and angiographic imaging confirms treatable hypervascular targets.