Translate this page into:
Femorotibial degeneration equilibrium analysis for objective and clinically interpretable knee osteoarthritis stratification
⁎Corresponding author: I. Govindharaj. govindharaji@veltech.edu.in
-
Received: ,
Accepted: ,
This article was originally published by Reed Elsevier India Pvt. Ltd. and was migrated to Scientific Scholar after the change of Publisher.
Abstract
Abstract
Radiographic evaluation of knee osteoarthritis (KOA) is subject to inter-observer variability and limited objectivity, particularly in early disease stages. This study aimed to develop and validate a clinically interpretable framework for objective KOA identification and severity stratification by quantifying anatomical and functional degenerative imbalance from routine knee radiographs. We propose a novel Femorotibial Degeneration Equilibrium Analysis (FDEA) framework that models KOA as a progressive biomechanical imbalance process rather than an appearance-based classification task. Patient-specific anatomical stability anchors were first extracted to establish a femorotibial reference framework. A virtual mechanical load axis was then projected to quantify medial–lateral load drift, while joint space integrity was assessed using continuity tracing without pixel-wise segmentation. Directional degeneration gradients were further estimated to capture compartment-specific disease progression. These clinically meaningful indicators were integrated using orthopaedic reasoning rules to derive an objective equilibrium loss score for KOA severity stratification. The proposed method was evaluated using knee radiographs from the Osteoarthritis Initiative (OAI) dataset with expert Kellgren–Lawrence (KL) grading as the reference standard and was compared with joint space width measurement, texture-based analysis, shape deformation analysis, and radiologist rule-based scoring. External validation was performed on an independent subset from the Multicenter Osteoarthritis Study (MOST). FDEA demonstrated superior agreement with expert KL grading (Cohen's κ = 0.86 ± 0.04; p < 0.01) and achieved higher sensitivity for early-stage KOA (92.4% ± 3.1%) and specificity for advanced disease (94.1% ± 2.8%) than all comparative methods. Assessment time did not differ significantly from conventional radiographic evaluation. The proposed FDEA framework offers a transparent, clinically interpretable decision-support tool that complements conventional grading and supports early diagnosis and treatment planning in knee osteoarthritis.
Keywords
Knee osteoarthritis
Tibiofemoral degeneration
Radiographic assessment
Disease severity stratification
Clinically interpretable algorithm
Femorotibial equilibrium analysis
1 Introduction
Knee osteoarthritis (KOA) is among the most common musculoskeletal diseases in the global community and a major cause of pain, disability, and poor quality of life among the geriatric society. Proper radiographic evaluation of KOA is essential in the early diagnosis, disease follow-up and planning of treatment, especially in assisting in conservative treatment and surgical decision making. Traditional assessment of KOA severity depends mostly on plain radiographs, because of their extensive availability, low cost and accepted place in normal orthopaedic practice.
The Kellgren-Lawrence (KL) grading system is the most popular radiographic criterion of KOA severity measurement. Although it has clinical acceptability, KL grading is always subjective and can be subject to inter and intra-observer variability particularly at an earlier stage of the disease when radiographic changes are not very noticeable. In addition, KL grading gives a discrete categorical score that might not be sufficient to document progressive structural degradation or compartment-specific degenerative patterns. Such constraints have led to continued attempts to come up with more objective and re-producible radiographic assessment strategies.
In orthopaedic perspective, knee osteoarthritis is a progressive load transmission during the transmission of loads of the femur to the tibia, not the isolated structural degeneration. Compartment overload (malalignment) is usually present before even the radiographic alterations can be detected, especially in the initial stages of the disease.
To enhance the level of objectivity, quantitative methods like Joint Space Width (JSW) measurement has been introduced to quantify the loss of cartilage by estimating it indirectly by reducing the radiographic joint space. Although the JSW measurement has better reproducibility than visual grading, it is also sensitive to positioning of imaging, does not entirely reveal redistribution of compartmental loads and often does not identify early degenerative imbalance. Analysis of the texture of trabecular bone and the periarticular areas has also been determined as a complementary method that has shown the relationships between the patterns of radiographical texture and disease progression. Nevertheless, texture properties cannot be directly interpreted clinically, which is limited by their explicit anatomical or biomechanical interpretation.
In more recent studies, shape deformation measurements of femoral condyles and tibial plateaus have been suggested as a way of measuring morphological changes with KOA development. These techniques give useful structural data, yet normally describe degeneration strictly as a type of geometric change and not a dynamic process of imbalance involving both the distribution of load and compartmental interaction. Radiologist rule-based scoring systems, which seek to formalise expert decision-making through the combination of various radiographic signs, are at the same time limited to predefined rules and scalable to a limited degree.
Recent developments in machine learning and deep learning have resulted in many automated KOA assessment frameworks, with many of them showing good classification on large public datasets. These developments notwithstanding, most of the data-driven models are black box models, which provide little insight into the way anatomical or functional changes play into the severity decision. This interpretability is a major impediment to clinical adoption especially in orthopaedic practice where explainability and anatomical reasoning are fundamental.
Clinically, isolated changes in structure are not alone features of knee osteoarthritis, but through a series of progression of equilibrium between the femur and tibia components, which are manifested by changes in load distribution, compartmental degeneration and directional disease progression. Nevertheless, this notion of tibiofemoral degenerative imbalance has not been clearly represented in current computational paradigms of KOA evaluation.
In the paper, we are presenting a clinically derived computational methodology that unravels tibiofemoral degeneration through quantification of equilibrium loss instead of appearance-based classification. The suggestions Femorotibial Degeneration Equilibrium Analysis (FDEA) models KOA as a progressive imbalance process by means of patient-specific anatomical reference frame, load drift estimation, joint space continuity analysis, and medial-lateral degeneration gradients. Severity stratification is done based on orthopaedic reasoning rule, which is transient and clinically readable.
The suggested technique is assessed using large, publicly available radiographic cohorts of the Osteoarthritis Initiative (OAI) and externally validated on the Multicenter Osteoarthritis Study (MOST) by utilizing expert KL grading as the standard of reference. The performance is measured by comparing it to the established radiographic assessment methods, such as joint space width measure, texture-based analysis, shape deformation analysis, and radiologist rule-based scoring. To achieve an objective, interpretable, and clinically relevant decision support tool in knee osteoarthritis assessment as part of routine practice, the current study will be based on the alignment of computational analysis with orthopaedic reasoning.
This article presents a completely new, clinically based paradigm of radiographic knee osteoarthritis evaluation, and the main contributions are as follows:
An original Femorotibial Degeneration Equilibrium Analysis (FDEA) algorithm that represents knee osteoarthritis as a process of tibiofemoral equilibrium loss and not appearance-based classification.1.The suggested FDEA model makes it possible to have clinically interpretable radiographic evaluation incorporating anatomical alignment, load drift, joint space continuity, and compartment-wise degeneration based on orthopaedic reasoning.2.The present work presents a balance-based radiographic biomarkers which are objective indicators of functional imbalance and directional degeneration in excess of traditional measures of the width and morphology of the joint space.3.The proposed methodology is tested on knee radiographs of the Osteoarthritis Initiative (OAI) and externally validated on Multicenter Osteoarthritis Study (MOST), which proves strength and generalizability.4.FDEA has better compatibility with expert Kellgren-Lawrence grading as well as greater sensitivity to early-stage knee osteoarthritis without raising the assessment time.5.The focus of the research is to offer objective, clear, and clinically-based alternative to KOA grading systems based on appearances.
In contrast to the literature using radiographic and deep learning-based methods, which largely use the appearance of knee osteoarthritis as a classification problem, the Femorotibial Degeneration Equilibrium Analysis (FDEA) directly presents KOA as a loss in femorotibial equipment. FDEA represents a clinically interpretable, medially-laterally load redistribution, and joint space continuity into a single biomechanical valid system to offer a functional and biomechanical based alternative to black-box severity grading techniques. Despite promising accuracy, most automated KOA assessment methods treat radiographs as appearance-based classification problems, offering limited biomechanical insight and restricted clinical interpretability.
2 Related works
Deep learning-based automatic severity grading has been one of the recent knee osteoarthritis (KOA) assessment studies based on radiographic imaging. It was demonstrated that convolutional neural networks and modular deep learning models can result in a high Kellgren-Lawrence (KL) grade classification accuracy,1,2 and also Sharma et al..3 According to Panwar et al.,4 Pan et al.,5 Maqsood et al.,6 and Jahan et al.,7 transformer-based and hierarchical models have also enhanced the performance of grading. The diagnostics value of the deep learning-based analysis of the X-ray was supported by systematic reviews and meta-analyses conducted by Zhao et al.8 and Teoh et al.9 especially with moderate-to-severe KOA. Nevertheless, despite these developments, many deep learning methods demand big, annotated datasets and are non-transparent and non-clinically interpretable.
Together with deep learning, there is several studies that have concentrated on automated radiographic measurements and handcrafted feature-based solutions to enhance interpretability. Rayegan et al.10 explored automated radiographic measurements in the joint space width and parameters associated with alignment. Mehdi et al.11 and Tariq et al.12 investigated the texture-based and shape-based radiographic characterization, and Li et al.13 and Almhdie-Imjabbar et al.14 used radiomics-based grading with the help of single - and multi-view radiographs. Despite the clinical meaning of these approaches in structural measurements, they tend to use single features and might fail to provide a comprehensive indication of the intricate biomechanical interactions and compartmental degeneration of KOA progression, especially in the initial stages of the disease.
Combining deep learning with classical machine learning or optimization approaches to create hybrid and ensemble methods to improve robustness and grading accuracy have also been suggested. Hybrid and GA-enhanced ensemble models were proposed to evaluate the severity of KOA by Ahmed and Mstafa15 and Nguyen-Tat and Nguyen-Duong,16 whereas Chaugule et al.17 examined feature fusion tactics to enhance classification. To support the issue of interpretability, explainable artificial intelligence methods have been explored, with Teoh et al.9 and Shahid et al.18 showing the relevance of transparent decision-support systems. The majority, however, of describeability approaches are still feature-attribution approaches and fail to directly correspond with the biomedically intuitive reasoning of biomechanism.
Multimodal and longitudinal, in addition to radiographic imaging, have also been considered in the characterization of KOA and prediction of its progression. Yao et al.19 constructed cartilage morphmetric analysis frameworks automatically, Wang et al.20 and Chen et al.21 combined longitudinal MRI radiomics and neural networks to forecast disease progress. Raza et al.22 and Guo et al.23 also investigated structural biomarkers and multi-joint predictors. These studies do not have a limited clinical application since they require advanced imaging modalities and longitudinal data, which is not always possible in routine clinical scenarios, as plain radiographs are the major diagnostic method.
One of the recent knee osteoarthritis (KOA) assessment investigations, which is built on radiographic images, is that of deep learning-based automatic severity grading. Convolutional neural networks and modular deep learning models turned out to lead to a high KL grade of classification accuracy,1,2 and Sharma et al.3 as well. Transformer-based and hierarchical models have also contributed to the improvement of grading performance as postulated by Panwar et al.,4 Pan et al.,5 Maqsood et al.,6 and Jahan et al..7 The systematic reviews and meta-analyses by Zhao et al.8 and Teoh et al.9 in particular with moderate-to-severe KOA supported the diagnostics value of the deep learning-based analysis of the X-ray. However, even though this is taking place, most deep learning approaches require large annotated datasets and are non-transparent and unable to be clinically explained.
Large-scale cohort studies such as the Osteoarthritis Initiative and the Multicenter Osteoarthritis Study have played a central role in advancing KOA research. Liu et al.24 reviewed the evolution and significance of the OAI cohort, while phenotype-based refinement of KL grading and progression risk assessment using OAI and MOST data were proposed by Nurmirinta et al.25 and Ko et al..26 These works underscore the importance of population-level validation but continue to frame KOA severity primarily as a categorical or appearance-based classification problem.
In contrast to prior studies, the proposed Femorotibial Degeneration Equilibrium Analysis (FDEA) framework models KOA progression as a loss of femorotibial biomechanical equilibrium. By explicitly integrating anatomical alignment, load redistribution, and joint space continuity into an interpretable assessment paradigm, the proposed approach addresses limitations of appearance-driven and black-box models while remaining compatible with routine radiographic imaging and large-scale cohort data.
In addition, recent studies have explored continuous severity grading, hybrid deep learning architectures, and multimodal perspectives for knee osteoarthritis assessment. Joseph et al.27 reviewed recent machine learning models for clinical and structural KOA prediction, while lateral-view radiographic analysis was investigated by Abdullah et al..28 Feature-fusion and deep ensemble strategies were further examined by Ren et al.,29 Belton et al.,30 Rani et al.,31 and Alavanthar et al.,32 demonstrating improved grading robustness. Broader perspectives on artificial intelligence applications in osteoarthritis research across imaging and omics domains were discussed by Ou et al..33 Large-scale cohort resources, including the Osteoarthritis Initiative34 and the Multicenter Osteoarthritis Study,35 continue to provide essential population-level data for KOA research and external validation.
Table 1 highlights that most existing approaches rely on appearance-driven or data-intensive models, whereas the proposed FDEA framework emphasizes biomechanical equilibrium and clinical interpretability using routine radiographs.
| Study | Data Modality | Methodological Approach | Key Strengths | Major Limitations |
| Lee et al.,1 Kinger,2 Sharma et al.3 | X-ray (AP) | CNN-based deep learning | High grading accuracy, automation | Black-box models, limited interpretability |
| Zhao et al.,8 Teoh et al.9 | X-ray | Systematic reviews/meta-analyses | Strong evidence for DL effectiveness | Limited insight into clinical reasoning |
| Panwar et al.,4 Pan et al.,5 Jahan et al.7 | X-ray | Transformer/hierarchical DL | Improved feature representation | High data and computational demand |
| Rayegan et al.,10 Li et al.,13 Almhdie-Imjabbar et al.14 | X-ray | JSW, radiomics, texture/shape analysis | Interpretable structural metrics | Focus on isolated features |
| Mehdi et al.,11 Tariq et al.12 | X-ray | Shape + texture learning | Captures structural patterns | Limited biomechanical modeling |
| Ahmed & Mstafa,15 Nguyen-Tat & Nguyen-Duong16 | X-ray | Hybrid ML/DL, ensemble methods | Improved robustness | Still appearance-driven |
| Belton et al.,30 Rani et al.31 | X-ray | Continuous/CNN-based grading | Handles severity continuity | Limited clinical transparency |
| Yao et al.,19 Wang et al.,20 Chen et al.21 | MRI/multimodal | Radiomics + ML | Progression modeling | Not routine clinical imaging |
| Raza et al.,22 Nurmirinta et al.25 | X-ray | Structural biomarkers | Prognostic insight | Requires longitudinal data |
| Ko et al.26 | X-ray | KL phenotype refinement | Improved clinical association | Still categorical grading |
| OAI,34 MOST35 | X-ray/clinical | Large cohort datasets | Population-level validation | No inherent grading method |
Unlike prior radiographic and learning-based methods, the proposed FDEA framework models knee osteoarthritis as a biomechanical imbalance process grounded in orthopaedic principles of alignment, load redistribution, and compartmental degeneration, rather than as an image appearance classification task.
3 Materials and methods
3.1 Overview of the proposed framework
The proposed methodology is based on the sequential and clinically grounded workflow of objective knee osteoarthritis (KOA) evaluation based on regular radiographs. As shown in Fig. 1, the workflow, the method starts with the radiographic input and moves on to anatomical reference extraction and equilibrium-based analysis leading to the compartment-wise degeneration mapping and eventual stratification of the KOA severity into mild, moderate, and severe. The entire framework is implemented as the Femorotibial Degeneration Equilibrium Analysis (FDEA) algorithm.
The pipeline shown below illustrates various stages of transformation of an input knee radiograph to anatomical reference, equilibrium features, degeneration equilibrium, compartment-based differentiation, and objective knee osteoarthritis severity stratification in accordance with clinical grading.1.Radiographic Input Acquisition: The input was standardized anteroposterior knee radiographs. Images were retrieved in open cohorts and left in their raw form without image enhancement or learning-based image transformation to allow the images to maintain the radiographic features of the original image as they would otherwise be in the orthopaedic examination.2.Anatomical Reference Extraction: Significant femorotibial anatomical landmarks were determined to determine a patient specific framework of reference. They consisted of the femur condyles, the tibial plateaus, which are stable anatomical locations in further analysis. This process will allow the consistent spatial positioning and interpretation of radiographs, which is independent of the level of disease.3.Equilibrium Feature Construction: Equilibrium related features were created using the previously established anatomical reference framework to describe degenerative imbalance. These characteristics are the alignment of the femur and tibia and computation of the medial-lateral load drift using a virtual mechanical axis and testing of the continuity in the joint space around the tibiofemoral interface. Instead of using pixel-wise segmentation, the continuity tracing was used to measure joint space integrity, and this technique enabled the detection of local compression and asymmetry indicative of osteoarthritic development.4.Femorotibial Degeneration Equilibrium Analysis (FDEA): The removed equilibrium features were incorporated in FDEA model to measure functional and structural imbalance between the femur and tibia. Patterns of directional degeneration and redistribution of loads are used to distinguish between balanced and imbalanced compartmental states. This modeling is based on the equilibrium approach that indicates orthopaedic knowledge of KOA as a gradual disproportion process and not a look-like disorder.5.Compartment-wise Degeneration Mapping: wherein, Degenerative activity at the medial and lateral compartments was mapped using characteristics of loss of equilibrium. It was found that compartment dominance and asymmetry were useful to describe clinically significant progressions, and that they assist in distinguishing between medial-dominant, lateral-dominant, and mixed degeneration phenotypes.6.KOA Severity Stratification: Knees were stratified to three clinically significant severity groups using the magnitude of the loss of equilibrium and patterns of degeneration of the compartments: mild, moderate, and severe osteoarthritis. This stratification is consistent with concepts of existing clinical grading but gives a continuous objective interpretation based on anatomical and functional imbalance.
Table 2 shows the Definition of variables and symbols used in the proposed Femorotibial Degeneration Equilibrium Analysis (FDEA) framework and associated mathematical formulations.
| Symbol | Variable Name | Description/Clinical Meaning |
| I | Input radiograph | Standardized weight-bearing anteroposterior (AP) knee radiograph |
| In | Normalized radiograph | Intensity-normalized knee radiograph used for analysis |
| μ | Mean intensity | Mean pixel intensity within the tibiofemoral joint region |
| σ | Standard deviation | Standard deviation of pixel intensities in the joint region |
| Fc | Femoral condyle points | Set of points defining inferior contours of femoral condyles |
| Tp | Tibial plateau points | Set of points defining superior surface of tibial plateaus |
| AFT | Femorotibial reference axis | Patient-specific anatomical reference axis |
| Wt | Tibial plateau width | Total medial–lateral width of the tibial plateau |
| Mc | Medial compartment | Medial half of tibial plateau defined by midpoint |
| Lc | Lateral compartment | Lateral half of tibial plateau |
| θ | Measured alignment angle | Observed femorotibial alignment angle from radiograph |
| θ0 | Neutral alignment | Reference mechanical alignment angle |
| Δθ | Alignment deviation | Degree of varus/valgus malalignment (Eq. (5)) |
| xload | Load axis intersection | Intersection point of projected mechanical load axis |
| Dload | Load drift index | Medial–lateral deviation of load transmission (Eq. (6)) |
| JSM | Medial joint space width | Joint space width measured in medial compartment |
| JSL | Lateral joint space width | Joint space width measured in lateral compartment |
| ΔJS | Joint space disruption | Absolute medial–lateral joint space asymmetry (Eq. (7)) |
| JSA | Joint Space Asymmetry Index | Directional indicator of compartment-dominant narrowing (Eq. (10)) |
| GM | Medial degeneration gradient | Degeneration magnitude in medial compartment |
| GL | Lateral degeneration gradient | Degeneration magnitude in lateral compartment |
| Dc | Compartment dominance | Medial, lateral, or balanced degeneration classification |
| α,β,γ | Weighting coefficients | Clinically calibrated weights derived from KL grading |
| ELS | Equilibrium Loss Score | Integrated biomechanical degeneration score (Eq. (8) and (13)) |
| T1,T2 | Severity thresholds | Clinically calibrated boundaries for severity stratification |
| S | KOA severity label | Final classification: Mild/Moderate/Severe |
| ε | Neutral tolerance | Threshold for near-neutral load distribution |
The proposed Femorotibial Degeneration Equilibrium Analysis (FDEA) framework of objective knee osteoarthritis evaluation presented in Algorithm 1. The algorithm combines radiographic input image, anatomy reference image, balance feature computation, and clinically trained model stage with an expert based Kellgren-Lawrence grading to calculate an equilibrium loss score and categorize the disease severity as mild, moderate and severe.Algorithm 1Femorotibial Degeneration Equilibrium Analysis (FDEA)Input: Weight-bearing anteroposterior (AP) knee radiograph I; expert Kellgren–Lawrence (KL) grades (for calibration).Output: Knee osteoarthritis severity level S∈{Mild,Moderate,Severe}1.Radiographic Input Acquisition:Acquire standardized AP weight-bearing knee radiograph I from the OAI or MOST dataset.2.Preprocessing:Resize I to a uniform spatial resolution and perform intensity normalization to obtain In.3.Anatomical Reference Extraction:Identify femoral condyles and tibial plateaus from In.Construct the femorotibial reference axis and determine the tibial plateau midpoint xm and width Wt.4.Compartment Delineation:Partition the tibial plateau into medial and lateral compartments based on xm.5.Equilibrium Feature Computation:Compute equilibrium-based indicators as follows:•Alignment deviationAd=∣θ−θ0∣where θ is the measured femorotibial alignment angle and θ0 is the neutral alignment.•Load drift indexLd=∣xc−xm∣Wtwhere xc is the projected load axis intersection point.•Joint space continuity disruptionJc=∣JSWm−JSWl∣JSWm+JSWlwhere JSWm and JSWl are medial and lateral joint space widths.•Compartmental degeneration gradientDg=∣Dm−Dl∣Dm+Dlwhere Dm and Dl denote medial and lateral degeneration indicators.6.Model Training (Clinical Calibration):Using OAI radiographs with expert KL grading, calibrate weighting coefficients.α,β,γ,δ such thatα+β+γ+δ=1and determine severity thresholds τ1 and τ2.7.Equilibrium Loss Score Estimation:Compute the femorotibial equilibrium loss score:ELS=αAd+βLd+γJc+δDg8.Severity Stratification:Assign knee osteoarthritis severity:S={Mild,ELS<τ1Moderate,τ1≤ELS<τ2Severe,ELS≥τ29.External Validation:Apply the calibrated FDEA framework to the independent MOST dataset to assess robustness and generalizability.10.Return Output: Output the final severity label S.
3.1.1 Clinical motivation for equilibrium-based modeling
From an orthopaedic perspective, knee osteoarthritis progression is driven by imbalanced load transmission across the femorotibial compartments rather than isolated structural abnormalities. Malalignment-induced load redistribution often precedes visible joint space narrowing, particularly in early disease stages. The proposed FDEA framework is designed to capture this biomechanical imbalance by modeling alignment deviation, load drift, and joint space continuity as equilibrium-driven indicators of degeneration.
3.2 Anatomical reference extraction
Accurate identification of the anatomical reference structures of the femur and tibia is the key to the equilibrium-based analysis of knee osteoarthritis. Within the proposed Femorotibial Degeneration Equilibrium Analysis (FDEA) model, extraction of anatomical reference produces a patient-specific coordinate system with which the alignment, joint space continuity and compartment-wise degeneration can be assessed with consistent results using standard radiographs.
Using the normalized radiograph In(x,y), the femoral condyles and tibial plateaus were identified as stable anatomical anchors. Let F={(xfi,yfi)}i=1Nf denote the set of points corresponding to the inferior contours of the femoral condyles, and T={(xtj,ytj)}j=1Nt denote the set of points representing the superior surface of the tibial plateau.
A femorotibial reference axis LFT is determined as the central point of those anatomical structures in below Equations (2) and (3).(2)cf=1Nf∑i=1Nf(xfi,yfi),ct=1Nt∑j=1Nt(xtj,ytj)(3)LFT=cfct‾
This reference axis provides a consistent anatomical baseline for evaluating medial–lateral alignment and load distribution. To enable compartment-wise analysis, the tibial plateau was partitioned into medial and lateral compartments using the midpoint of the plateau width in below Equation (4):(4)xm=xt,min+xt,max2where xt,min and xt,max denote the lateral ranges of the tibial plateau. This partitioning allows independent assessment of compartment-specific degeneration patterns.
The resulting anatomical reference frame is the basis of the further computation of the equilibrium features, such as load drift estimation and joint space continuity analysis. Notably, this measure is based on only clear anatomic geometry, which makes it transparent and compatible with orthopaedic diagnostic logic.
Fig. 3(a) provides the Anteroposterior (AP) view of a knee radiograph which depicts some important landmarks in the radiographic knee osteoarthritis assessment. The photo identifies the structures that are of the reference to anatomical alignment, joint space assessment, and compartment-by-compart analysis; they are the femoral condyles, tibial condyles, medial and lateral joint spaces, intercondylar eminence and tibial spines, patellar shadow, fibular head, and proximal tibia and fibula.
The proposed framework does not need full-length radiographs as input because, in Fig. 3(b) Illustrative clinical concept only; full-length radiographs are not required to participate in the proposed framework. Panel (A) of the figure illustrates the non-alignment of the mechanical axis with the knee joint center, whereas Panel (B) illustrates the alignment of the mechanical axis with the knee joint center, which is much closer to the tibial plateau center. These pictures demonstrate that mechanical axis measurement has clinical importance in assessing the load distribution and degeneration of the knee joint associated with its alignment.
Fig. 3(c) displays the Radiographic demonstration of joint space measure and the assessment of the femur tibial alignment. Calibration-based medial and lateral joint space measurement relative to tibial reference axis displayed in Panel (A), which demonstrates the axis of projection of the load and joint space limits. In panel (B), compartment-wise joint space profiling is shown at the tibial plateau; the medial-lateral positions of the knee joints were normalized and the normalization was done to quantify the continuity and asymmetry of joint space which is used to evaluate equilibrium-based knee osteoarthritis.
In Fig. 3(d). Comparison of traditional Kellgren-Lawrence (KL) grading with the proposed FDEA on either knee radiograph.(A)The Medial (M) and lateral (L) compartments of representative anteroposterior knee radiograph.(B)KL grading uses a global categorical severity score that is determined by the radiographic characteristics of osteophyte formation and reduction of joint space.(C)With a compartment-based interpretation, the proposed FDEA offers the benefits of an equilibrium-based interpretation which measures the deviation in alignment, continuity of the joint space, and redistribution of loads, allowing objective stratification of severity, and greater sensitivity to early degenerative imbalance.
The qualitative application of the proposed equilibrium-based analysis to complete the KL grading is shown in Fig. 3(d) where it is possible to observe patterns of compartment-specific degeneration that are not reflected in categorical grading.
3.3 Radiographic input acquisition
The basic step in the proposed FDEA framework is radiographic input acquisition. The aim of this phase is to make sure that knee radiographs give anatomically consistent and clinically meaningful images of the tibiofemoral joint that would be used in the analysis of degeneration based on equilibrium.
Weight-bearing (AP) knee radiographs were standardized as these x-rays indicate transmission of physiological loads on the femur-tibia compartments and the x-rays are actively used in clinical orthopaedic assessment. The Osteoarthritis Initiative (OAI) data were used to derive radiographs to be analyzed and to provide external validation by a different independent subset of the Multicenter Osteoarthritis Study (MOST). The images that were selected to ensure anatomical integrity only included those that had a detailed visualization of the tibial plateaus and the femoral condyles.
To minimize acquisition-related variability while preserving native radiographic characteristics, a conservative intensity normalization procedure was applied. Given an input radiograph I(x,y), the normalized radiograph In(x,y) was computed as in Equation (1):(1)In(x,y)=I(x,y)−μIσIwhere μI and σI denote the mean and standard deviation of pixel intensities within the tibiofemoral joint region. This normalization reduces scanner-dependent intensity variation without altering structural information relevant to joint space continuity, alignment, or compartmental asymmetry.
Radiographs that displayed motion artifacts, gross exposure variation or partial coverage of the joint were filtered out in a quality screening step. No contrast additions, learning-based preprocessing, or artificial enhancement was used to make sure that the suggested framework is used in the conditions that are similar to the real orthopaedic practice with regard to radiographic interpretation.
The resultant quality and normalized radiographs form the direct input of the following stages of the FDEA framework namely anatomical reference extraction and equilibrium feature calculation.
Fig. 2 (a) below illustrates the Radiographic input acquisition process of the proposed Femorotibial Degeneration Equilibrium Analysis (FDEA). Fig. 2 (a) shows normalized weight-bearing anteroposterior knee radiographs of the Osteoarthritis Initiative (OAI) and Multicenter Osteoarthritis Study (MOST) databases, with quality screening and little preprocessing to get anatomically consistent inputs to assess equilibrium-based knee Osteoarthritis.
In Fig. 2(a). Representative knee radiographs depicting the radiographic progression of knee osteoarthritis among the Kellgren Lawrence grades, i.e., normal knee joint (Grade 0) to severe knee osteoarthritis (Grade 4). The pictures show progressive joint space thinning, osteophyte, and structure deformity, which is clinical reference standards of stratifying severity.
3.4 Femorotibial degeneration equilibrium analysis (FDEA)
Femorotibial Degeneration Equilibrium Analysis (FDEA) is a scale that is used to measure knee osteoarthritis as a progression towards imbalance between the tibial and femoral components when subjected to physiological loading conditions. Based on the anatomical reference structure of Section 2.3, FDEA incorporates alignment deviation, redistribution of the medial-lateral loads, and continuity to the femur tibia joint space to depict the clinical interpretable loss of femorotibial equilibrium.
In the first stage, using stable anatomical reference points based on the femoral condyles and tibial plateaus, a patient-specific reference optimal in terms of femoral tibia equilibrium are determined. This reference is the completely idealized value of balanced transmission of load through the medial and lateral compartments. Directional load drift is then measured using a virtual mechanical load axis projected which then measures the deviation of the axis as compared to the equilibrium reference.
Integrity in the joint space is then considered in the tibiofemoral interface without using pixel-wise segmentation. Comparing medial and lateral joint space profiles, localized compression and asymmetry is detected, which allows one to identify early compartment-dominant degeneration even without obvious radiographic changes on a global basis.
A combination of these indicators associated with equilibrium is used to calculate a Femorotibial Equilibrium Loss Score (ELS), which is the degree and direction of degenerative imbalance. The equilibrium loss score is used to categorize knees as mild, moderate and severe osteoarthritis based on clinically calibrated thresholds based on expert Kellgren-Lawrence (KL) grading. Through preservation of anatomical and biomechanical interpretation articulately, FDEA offers a transparent and reproducible decision-support model and augments traditional radiographic grading in the routine orthopaedic practice.
Mathematical Formulation (Consistent withAlgorithm 1).
Let θ denote the measured femorotibial alignment angle and θ0 the neutral alignment reference.
Alignment Deviation.
Equation (5) shows the Quantification of femorotibial alignment deviation relative to the neutral mechanical alignment, representing varus or valgus malalignment contributing to compartmental load imbalance.(5)Δθ=∣θ−θ0∣where Δθ represents varus or valgus deviation contributing to compartmental overload.
Load Drift Index.
Equation (6) represents the Medial–lateral load drift index derived from the projected mechanical load axis relative to the tibial plateau center, indicating compartment-dominant load redistribution.
Let xL denote the intersection of the projected mechanical load axis on the tibial plateau, and W the tibial plateau width.(6)DL=xL−W2Wwhere negative and positive values indicate medial- and lateral-dominant load redistribution, respectively.
Joint Space Continuity Disruption.
Equation (7) represents the Measurement of compartmental joint space continuity disruption based on the absolute difference between medial and lateral joint space widths, reflecting asymmetric cartilage degeneration.
Let JSm and JSl represent medial and lateral joint space widths.(7)JC=∣JSm−JSl∣which quantifies asymmetry in compartmental joint space preservation.
Femorotibial Equilibrium Loss Score.
Equation (8) shows the Computation of the femorotibial equilibrium loss score by integrating alignment deviation, load drift, and joint space continuity using clinically calibrated weighting coefficients.(8)ELS=αΔθ+βDL+γJCwhere α,β,γ are clinically calibrated weighting coefficients derived using expert KL grading.
Severity Stratification.
Equation (9) shows the Rule-based stratification of knee osteoarthritis severity into mild, moderate, and severe categories based on clinically calibrated equilibrium loss thresholds.(9)KOASeverity={Mild,ELS<τ1Moderate,τ1≤ELS<τ2Severe,ELS≥τ2where τ1 and τ2 denote severity thresholds determined during the calibration phase.
3.4.1 Clinical Interpretation of equilibrium components
Deviation of alignment is an indicator of valgus or varus malalignment, the major cause of compartment-specific overload. The index of load drift measures the direction and extent of the redistribution of the medial-lateral load in the tibial plateau. Joint space continuity disruption represents the asymmetric cartilage degeneration, which might not be noticeable by the global joint space width measures. The fact that they are incorporated into the femorotibial equilibrium loss score allows objective and interpretable stratification of the severity.
3.5 Compartment-wise degeneration mapping
Compartment-wise degeneration mapping is used to convert the indicators of equilibrium into patterns of localized osteoarthritic activity in the medial and lateral tibiofemoral compartments. This step rather than the global severity grading identifies the predominant compartment that degeneration occurs in under physiological conditions described as an asymmetrical load transmission and cartilage wear.
The tibial plateau according to the anatomy reference framework used in Section 2.3 and 2.4 is divided at the middle to provide medial and lateral compartments. The continuity of the joint space is evaluated separately in each compartment and therefore individual asymmetrical narrowing is detected. The mechanical load drift direction as calculated using the Femorotibial Degeneration Equilibrium Analysis (FDEA) may give functional help in determining whether degenerative stress is found to be predominantly medial or lateral.
By integrating joint space asymmetry with load drift direction, knees are categorized into medial-dominant, lateral-dominant, or balanced degeneration phenotypes. This compartment-level characterization enhances clinical interpretability and supports treatment-relevant decision-making, such as unloading strategies, alignment correction, or compartment-focused interventions.
Fig. 5 show the Architecture-level illustration of compartment-wise degeneration mapping in knee osteoarthritis. The tibiofemoral joint is partitioned into medial and lateral compartments using anatomical reference landmarks. Independent evaluation of joint space continuity and integration with load drift direction enable classification into medial-dominant, lateral-dominant, or balanced degeneration phenotypes, providing spatially interpretable characterization of disease progression.
Joint Space Asymmetry Index:
Let JSm and JSl denote the medial and lateral joint space widths, respectively.
In Equation (10) shows the Joint Space Asymmetry Index is given by(10)AJS=JSm−JSlwhere negative values indicate medial-dominant narrowing and positive values indicate lateral-dominant narrowing.
Compartment Dominance Classification:
Let DL represent the load drift index obtained from FDEA.
In Equation (2) represents the Compartment Dominance Classification CD is given by(11)CD={Medial‐dominant,DL<0Lateral‐dominant,DL>0Balanced,∣DL∣≤εwhere ε is a small tolerance defining near-neutral load distribution.
Compartment-wise Degeneration Score:
Equation (3) shows the Compartment-wise Degeneration Score CDS is given by,(12)CDS={∣JSm∣,CD=Medial‐dominant∣JSl∣,CD=Lateral‐dominant∣JSm∣+∣JSl∣2,CD=Balancedwhich reflects degeneration severity within the dominant compartment.
Clinical Interpretation.•Medial-dominant degeneration: Associated with varus alignment and increased medial compartment loading, commonly observed in early-to-moderate knee osteoarthritis.•Lateral-dominant degeneration: Linked to valgus alignment and lateral compartment overload.•Balanced degeneration: Indicates symmetric joint space reduction, typically observed in advanced or diffuse disease stages.
3.6 KOA severity stratification
KOA severity stratification converts equilibrium-derived degeneration indicators into clinically meaningful disease stages. Based on the outputs of the Femorotibial Degeneration Equilibrium Analysis (FDEA) and compartment-wise degeneration mapping, knee osteoarthritis is categorized into mild, moderate, and severe stages according to the magnitude of femorotibial equilibrium loss.
The stratification is motivated by three explainable indicators, namely, alignment deviation, medial-lateral load drift and joint space continuity disruption. These signs are the indicators of the development of biomechanical imbalance in the initial stages to structural failure of the tibiofemoral joint in advanced stages. The proposed method instead of utilizing appearance-based pattern recognition focuses on functional degeneration when the physiological load bearing conditions prevail.
An equilibrium loss rating is calculated by summing the three indicators with an integration of weighting coefficients which are clinically set. Professional Kellgren-Lawrence grading is then used in determining the severity of KOA. This strategy that is rule-based allows transparency, reproducibility and direct interpretation by clinical.
In Fig. 6, shows the knee osteoarthritis (KOA) severity stratification based on femorotibial degeneration equilibrium analysis. Clinically interpretable indicators reflecting alignment deviation, medial–lateral load redistribution, and joint space continuity loss are integrated to determine the degree of femorotibial equilibrium loss and to classify knees into mild, moderate, or severe osteoarthritis categories, consistent with established orthopaedic grading practice.
Equilibrium Loss Score.
Let.•Δθ denote femorotibial alignment deviation,•DL denote the medial–lateral load drift index,•JC denote joint space continuity disruption.(13)ELS=αΔθ+βDL+γJCwhere α, β, and γ are weighting coefficients calibrated using expert Kellgren–Lawrence grades.
In Equation (13), Computation of the femorotibial equilibrium loss score by integrating alignment deviation, medial–lateral load redistribution, and joint space continuity disruption to quantify the overall biomechanical severity of knee osteoarthritis.
KOA Severity Assignment.
In Equation (14). Rule-based assignment of knee osteoarthritis severity into mild, moderate, and severe categories based on clinically calibrated equilibrium loss thresholds derived from expert Kellgren–Lawrence grading.(14)KOASeverity={Mild,ELS<τ1Moderate,τ1≤ELS<τ2Severe,ELS≥τ2where τ1 and τ2 represent clinically determined severity thresholds.
Algorithm 2. Rule-based stratification of knee osteoarthritis severity using equilibrium-derived indicators. The algorithm integrates alignment deviation, medial–lateral load redistribution, and joint space continuity disruption to compute an equilibrium loss score and assigns mild, moderate, or severe disease categories based on clinically calibrated thresholds.Algorithm 2KOA Severity StratificationInput:•Alignment deviation Δθ•Load drift index DL•Joint space continuity disruption JCOutput:•KOA severity category (Mild/Moderate/Severe)Steps:1.Compute equilibrium loss score using Equation (13).2.Compare the computed score with calibrated thresholds τ1 and τ2.3.Assign KOA severity according to Equation (14).4.Report severity along with dominant compartment information.Clinical Interpretation.•Mild KOA: Low equilibrium loss with minimal malalignment and preserved joint space.•Moderate KOA: Intermediate equilibrium loss with measurable compartment dominance and joint space narrowing.•Severe KOA: High equilibrium loss characterized by pronounced malalignment, dominant load redistribution, and substantial joint space collapse.
4 Results
4.1 Dataset and evaluation protocol
The test of the proposed Femorotibial Degeneration Equilibrium Analysis (FDEA) framework was done by using knee radiographs of the Osteoarthritis Initiative (OAI) data set, and expert Kellgren-Lawrence (KL) grading was used as the reference standard. External validation was done using an independent subset of Multicenter Osteoarthritis Study (MOST) to determine the robustness and generalizability of the study across cohorts.
The measurements of the performance concerned the agreement with expert grading measured using a Cohen's κ statistic, sensitivity to the early stage of knee osteoarthritis (KL grades 1-2), sensitivity to the late disease (KL grades 3-4), and the total classification consistency. It was compared to the other known radiographic assessment methods like joint space width (JSW) measurement, texture-based analysis, shape deformation analysis, and radiologist rule-based scoring.
The values of agreement, sensitivity and specificity have been presented as mean ± standard deviation and the statistical significance as the paired comparison with a significance level of p < 0.05.
4.2 Agreement with expert Kellgren–Lawrence grading
The proposed FDEA model showed significantly better congruence with professional KL grading than with all reference strategies. FDEA had a Cohen k mean score of 0.86 ± 0.04 on the OAI data, which shows almost perfect agreement.
In contrast, joint space width measurement achieved a κ of 0.71 ± 0.06, texture-based analysis 0.68 ± 0.07, shape deformation analysis 0.73 ± 0.05, and radiologist rule-based scoring 0.77 ± 0.05.
Statistical analysis proved that the effect of FDEA was big in comparison to all comparative methods (p < 0.01). These results give strong intersections between the suggested equilibrium-based model and specialist radiographical grading.
4.3 Quantitative performance comparison with existing methods
The Quantitative effectiveness of the suggested Femorotibial Degeneration Equilibrium Analysis (FDEA) framework was evaluated against well-known radiographic knee osteoarthritis evaluation methods, such as, joint space width (JSW) measurement, texture-based analysis, shape deformation analysis, and radiologist rule-based scoring. Evaluation metrics evaluated were agreement with expert Kellgren-Lawrence (KL) grading, sensitivity to disease at early stage (KL grades 1-2), and specificity to disease at advanced stage (KL grades 3-4). Table 3 and Figs. 7–9 provide complementary numerical and visual summaries of the comparative performance of all the methods considered.
| Method | Agreement with KL (Cohen's κ) | Early KOA Sensitivity (%) | Advanced KOA Specificity (%) | Clinical Interpretability |
| Joint Space Width (JSW) Measurement | 0.71 ± 0.06 | 78.6 ± 6.4 | 88.2 ± 4.1 | Moderate |
| Texture-based Analysis | 0.68 ± 0.07 | 75.9 ± 7.1 | 86.7 ± 4.5 | Low |
| Shape Deformation Analysis | 0.73 ± 0.05 | 82.3 ± 5.2 | 89.6 ± 3.8 | Moderate |
| Radiologist Rule-based Scoring | 0.77 ± 0.05 | 85.1 ± 4.6 | 91.3 ± 3.2 | High |
| Proposed FDEA (OAI) | 0.86 ± 0.04 | 92.4 ± 3.1 | 94.1 ± 2.8 | High |
| Proposed FDEA (MOST – External Validation) | 0.84 ± 0.05 | 91.2 ± 3.6 | 93.5 ± 3.1 | High |
As summarized in Table 3, the proposed FDEA framework achieved the highest agreement with expert grading, with a mean Cohen's κ of 0.86 ± 0.04, indicating near-perfect concordance. This represents a notable improvement over JSW measurement (0.71 ± 0.06), texture-based analysis (0.68 ± 0.07), shape deformation analysis (0.73 ± 0.05), and radiologist rule-based scoring (0.77 ± 0.05).
In terms of stage-specific metrics, FDEA was found to be better sensitive to identify early-stage knee osteoarthritis (92.4% ± 3.1%), but more specific to advanced disease (94.1% ± 2.8%), than all reference measures. These findings suggest that assessment of severity based on equilibrium is more consistent and reliable in characterizing severity at various stages of the disease.
External validation on the independent Multicenter Osteoarthritis Study (MOST) dataset further confirmed the generalizability of the proposed approach. FDEA maintained high agreement (κ = 0.84 ± 0.05) with comparable sensitivity and specificity values. Performance improvements were statistically significant compared with reference methods (p < 0.01), with no significant degradation observed between datasets (p > 0.05).
Table 3. Quantitative performance comparison of the proposed Femorotibial Degeneration Equilibrium Analysis (FDEA) framework with established radiographic knee osteoarthritis assessment approaches. Performance is reported in terms of agreement with expert Kellgren–Lawrence grading (Cohen's κ), sensitivity for early-stage disease (KL grades 1–2), and specificity for advanced disease (KL grades 3–4).
In Fig. 7 shows the Overall quantitative comparison of knee osteoarthritis assessment methods in terms of agreement with expert Kellgren–Lawrence grading, measured using Cohen's κ. The proposed FDEA framework demonstrates superior agreement across both the OAI cohort and the independent MOST validation dataset.
In Fig. 8, Comparison of early-stage knee osteoarthritis (KOA) detection performance in terms of sensitivity across different radiographic assessment methods. The proposed Femorotibial Degeneration Equilibrium Analysis (FDEA) framework demonstrates substantially higher sensitivity for early KOA identification compared with conventional approaches, with consistent performance observed on both the Osteoarthritis Initiative (OAI) cohort and the independent Multicenter Osteoarthritis Study (MOST) validation dataset.
In Fig. 9, Comparison of advanced-stage knee osteoarthritis (KOA) detection performance in terms of specificity across radiographic assessment methods. The proposed Femorotibial Degeneration Equilibrium Analysis (FDEA) framework demonstrates higher specificity than conventional approaches, indicating robust discrimination of advanced disease, with consistent performance observed across the Osteoarthritis Initiative (OAI) cohort and the independent Multicenter Osteoarthritis Study (MOST) validation dataset.
4.4 Early- and advanced-stage knee osteoarthritis performance
Knee osteoarthritis (KOA) is characterized by different diagnostic features in the initial and developed stages, which are especially difficult to observe in the framework of regular radiography. To compare the performance of stages, the FDEA was examined in terms of early (Kellgren-Lawrence [KL] grades 1-2) and advanced (KL grades 3-4) disease.
For early-stage KOA, FDEA achieved a sensitivity of 92.4% ± 3.1%, significantly outperforming joint space width (JSW) measurement (78.6% ± 6.4%) and texture-based analysis (75.9% ± 7.1%, p < 0.01). This improvement indicates that equilibrium-based indicators—particularly alignment deviation and joint space continuity disruption—are effective in capturing functional biomechanical imbalance before pronounced structural joint space collapse becomes radiographically apparent.
In case of advanced-stage KOA, the specificity of the proposed framework was 94.1% ± 2.8% that is higher than any comparative approach. The high specificity indicates stringent selection of severe structural and biomechanical impairment, hence a decrease in false-positive classification and an assurance of a solid clinical decision-making in the later stages of the disease.
Table 4 presents the performance comparison of KOA at early and advanced stage. The proposed FDEA framework has better sensitivity to early disease and specificity an advanced disease in relation to traditional radiographic assessment methods. This has clinical implications as early biomechanical instability is associated with irreversible cartilage loss and early sensitivity can help prevent disease-modifying interventions and approaches.
| KOA Stage | Metric | FDEA | JSW Measurement | Texture-based Analysis |
| Early-stage (KL 1–2) | Sensitivity (%) | 92.4 ± 3.1 | 78.6 ± 6.4 | 75.9 ± 7.1 |
| Advanced-stage (KL 3–4) | Specificity (%) | 94.1 ± 2.8 | 88.2 ± 4.1 | 86.7 ± 4.5 |
4.5 Component contribution (ablation) analysis
To determine the contribution of each component of the equilibrium, a component contribution (ablation) analysis was performed through the integration of the equilibrium indicators in the FDEA model. The deviation in alignment alone offered the minimal agreement with expert grading and decreased sensitivity in the early disease diagnosis. This continuity of the joint space was added and showed the most important enhancement of agreement and sensitivity to detect the asymmetric cartilage degeneration. It was further improved with the incorporation of the load drift index to consider a redistribution of the medial-lateral loads.
The FDEA framework with full implementation of alignment deviation, load drift, and joint space continuity gave the best agreement with the expert grading and the best performance at the early stage of detection. These findings confirm that every component of the equilibrium adds complementary and clinically significant data and that their combination is crucial in achieving strong and interpretable knee osteoarthritis severity stratification. Notably, the ablation analysis has verified that none of the indicators work independently, which supports the clinical justification of the integration of radiographics based on equilibrium and not on a single radiographic parameter.
Table 5 presents the Component contribution (ablation) analysis which reveals the incremental value of incorporating the equilibrium-based indicators into the FDEA model. The complete model has maximum congruence with professional scoring and sensitivity of early detection. Notably, this review confirms that none of the radiographic indicators alone is adequate, which supports the clinical rationale of integrating in an equilibrium mode. These findings validate the fact that to have sound and dependable severity stratification, a combination of complementary equilibrium indicators is inevitable.
| Configuration | Cohen's κ | Early KOA Sensitivity (%) |
| Alignment deviation only | 0.72 ± 0.06 | 80.1 ± 5.8 |
| Alignment + Joint space continuity | 0.80 ± 0.05 | 87.6 ± 4.2 |
| Full FDEA (Alignment + Load drift + Continuity) | 0.86 ± 0.04 | 92.4 ± 3.1 |
4.6 External validation on the MOST dataset
A test of robustness and generalizability was done by external validation in an independent subset of Multicenter Osteoarthritis Study (MOST). The FDEA structure suggested gained the Cohen's κ of 0.84 ± 0.05 and sensitivity of the framework at the initial stages and specificity at the final stages similar like that of the OAI dataset. There was no statistically significant decrease in performance between datasets (p > 0.05), and this suggests that the algorithm is resistant to heterogeneity in the population and variability in imaging across multi-center cohorts.
Table 6 present the External validation of the proposed FDEA framework on the MOST dataset, which reveals constant performance and generalizability across independent cohorts as reflecting strong performance to the heterogeneity of the population and the variability of centric imaging.
| Dataset | Cohen's κ | Early KOA Sensitivity (%) | Advanced KOA Specificity (%) |
| OAI (Primary cohort) | 0.86 ± 0.04 | 92.4 ± 3.1 | 94.1 ± 2.8 |
| MOST (External validation) | 0.84 ± 0.05 | 91.2 ± 3.6 | 93.5 ± 3.1 |
4.7 Assessment time and practical feasibility
There was no statistically significant difference between the time taken to assess using FDEA and using a conventional radiographic assessment (p > 0.05). FDEA unlike deep learning-based methods does not need to train any model, or perform pixel-wise segmentation, or run an implementation on a GPU, making it efficient to deploy it and maintain full clinical interpretability.
No statistically significant difference in assessment time was observed compared to routine radiographic evaluation, supporting clinical feasibility.
5 Discussion
The assessment of the degree of knee osteoarthritis (KOA) in radiography has not been accurately determined especially in the early stages of the disease where the structural alterations are too subtle and non-homogenous. This paper suggested a Femorotibial Degeneration Equilibrium Analysis (FDEA) model representing the KOA evolution as the process of biomechanical imbalance and not merely as an occurrence of appearance. The findings indicate that assessment based on equilibrium enhances agreement with expert assessment, sensitivity at an early stage and gives reliable discrimination of advanced disease as compared with traditional radiographic methods.
The recommended framework was much more consistent with expert Kellgren-Lawrence (KL) grading, as compared to joint space width measurement, texture-based analysis, shape deformation analysis, and rule-based scoring. This enhanced concordance demonstrates the capability of FDEA to combine anatomical correspondence, redistribution of loads, and joint space continuity these factors are consistent with orthopaedices logic and recognized biomechanical factors that contribute to KOA progression.
A clinically significant outcome is an improved sensitivity on early KOA. Biomechanical imbalance can commonly be followed by irreversible cartilage loss and overt cartilage joint space, which restrict the usefulness of radiographic-only measures in early disease. The increased detection at an early stage of FDEA indicates that equity-based indicators may address functional degeneration prior to the development of radiographically advanced structural deterioration that may support early intervention and disease-modifying measures.
In case of further-progressed KOA, the framework provided had high specificity, which means that the deterioration of structures and biomechanical conditions of severe structural and biomechanical deterioration is strongly identified without any false-positive classification. Right diagnosis of advanced disease is critical in making the right clinical decisions, such as making a referral to surgical care.
The contribution analysis of components established that none of the equilibrium indicators could be applied as a solo construct. Rather, alignment deviation, load drift, and joint space continuity integration presented complementary data that allowed reliable stratification of severity, which supported the notion of multifactoriality of KOA progression.
The IRS was tested on the independent Multicenter Osteoarthritis Study (MOST) dataset using external validation mechanisms and revealed similar results which allowed concluding that the proposed framework can be generalized to other cohorts with different population structure and imaging conditions. Practically, FDEA does not need model training, pixel-wise separation, or exclusive processing tools, and assessment time was equal to standard radiographic examination, which allows it to be utilized in standard orthopaedic practices.
This study has limitations. The analysis of anteroposterior radiographs was carried out and the calibration of severity was performed on the basis of expert KL grading, which is subjective. The future research will aim at longitudinal progression modeling, integration of other imaging views, and expansion of the equilibrium model to multimodal imaging.
From an orthopaedic perspective, knee osteoarthritis progression is fundamentally driven by altered load transmission across the femorotibial compartments. The proposed equilibrium-based framework directly captures this biomechanical process, enabling clinicians to identify functional imbalance before irreversible structural collapse occurs. This capability is particularly relevant for early intervention planning, alignment correction strategies, and compartment-specific management.
6 Conclusion
This study presented a Femorotibial Degeneration Equilibrium Analysis (FDEA) framework for objective assessment of knee osteoarthritis severity from routine radiographs. By modeling disease progression as a biomechanical equilibrium loss process, the proposed approach integrates anatomical alignment, load redistribution, and joint space continuity into a clinically interpretable assessment paradigm.
Quantitative analysis revealed that FDEA is more in agreement with expert Kellgren-Lawrence grading compared to traditional radiographic assessment techniques with better sensitivity of early stage disease and high specificity of advanced stage osteoarthritis. The high consistency in the performance of the independent cohorts also contributes to the strength and applicability of the proposed framework.
Notably, FDEA is also fast to deploy, and as a result of the absence of information-driven model training or computationally expensive inference, it retains a high level of transparency in its decision-making. These attributes place the suggested equilibrium-based model as a viable and explainable decision-support instrument of the regular orthopaedic evaluation, screening, and long-term care of knee osteoarthritis.
Guardian/patient's consent
Not applicable. This study was conducted using de-identified knee radiograph images obtained from publicly available datasets (the Osteoarthritis Initiative (OAI) and the Multicenter Osteoarthritis Study (MOST)). No direct patient involvement or identifiable personal information was included, and informed consent was obtained by the original data providers.
Ethical statement
This study was conducted using de-identified knee radiograph images obtained from publicly available datasets, namely the Osteoarthritis Initiative (OAI) and the Multicenter Osteoarthritis Study (MOST). The datasets were collected and made publicly accessible by the original study investigators following approval from their respective Institutional Review Boards (IRBs) and in accordance with relevant ethical guidelines and regulations. As the present study involved secondary analysis of anonymized data with no direct patient involvement, additional ethical approval and informed consent were not required.
Credit author statement
Govindharaj I: Conceptualization, Project administration, Resources, Software, Data curation, Writing – original draft, Writing – review & editing. Gnanajeyaraman Rajaram: Methodology, Investigation, Formal analysis, Data curation, Funding acquisition. K. K. Ezhilarasan: Validation, Writing – original draft. Viswanath J: Supervision, Formal analysis, Visualization.
Funding statement
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.
References
- Accurate, automated classification of radiographic knee osteoarthritis severity using a novel method of deep learning: plug-in modules. Knee Surgery and Related Research. 2024;36(1):24.
- [Google Scholar]
- Deep learning for automatic knee osteoarthritis severity grading and classification. Indian J Orthop. 2024;58(10):1458-1473.
- [Google Scholar]
- Comparative analysis of deep learning models for early detection of osteoarthritis using knee radiographs: a retrospective study. Journal of Clinical Orthopaedics and Trauma. 2025;70
- [Google Scholar]
- Early diagnosis of knee osteoarthritis severity using vision transformer. BMC Muscoskelet Disord. 2025;26(1):884.
- [Google Scholar]
- Automatic knee osteoarthritis severity grading based on X-ray images using a hierarchical classification method. Arthritis Res Ther. 2024;26(1):203.
- [Google Scholar]
- Knee osteoarthritis network: a hybrid transformer-based approach for enhanced detection and grading of knee osteoarthritis. Eng Appl Artif Intell. 2025;159
- [Google Scholar]
- KOA-CCTNET: an enhanced knee osteoarthritis grade assessment framework using modified compact convolutional transformer model. IEEE Access. 2024;12:107719-107741.
- [Google Scholar]
- The value of deep learning-based X-ray techniques in detecting and classifying K-L grades of knee osteoarthritis: a systematic review and meta-analysis. Eur Radiol. 2024;35(1):327-340.
- [Google Scholar]
- Deciphering knee osteoarthritis diagnostic features with explainable artificial intelligence: a systematic review. IEEE Access. 2024;12:109080-109108.
- [Google Scholar]
- Automated radiographic measurements of knee osteoarthritis. Cartilage. 2023;14(4):413-423.
- [Google Scholar]
- Deep learning based approach combining shape and texture features for knee osteoarthritis prediction from X-ray images. Biomed Signal Process Control. 2024;101
- [Google Scholar]
- Knee osteoarthritis detection and classification using X-Rays. IEEE Access. 2023;11:48292-48303.
- [Google Scholar]
- Automatic grading of knee osteoarthritis with a plain radiograph radiomics model: combining anteroposterior and lateral images. Insights Imaging. 2024;15(1):143.
- [Google Scholar]
- Short-term variations in trabecular bone texture parameters associated to radio-clinical biomarkers improve the prediction of radiographic knee osteoarthritis progression. Sci Rep. 2023;13(1)
- [Google Scholar]
- Identifying severity grading of knee osteoarthritis from X-ray images using an efficient mixture of deep learning and machine learning models. Diagnostics. 2022;12(12):2939.
- [Google Scholar]
- Optimizing knee osteoarthritis severity diagnostics: a GA-enhanced deep ensemble approach in medical imaging. Ain Shams Eng J. 2025;16(9)
- [Google Scholar]
- Analysis of feature fusion in deep learning for knee osteoarthritis severity grade classification. 2025:3-14.
- [Google Scholar]
- Potential of AI-based diagnostic grading system for knee osteoarthritis. Front Med. 2025;12
- [Google Scholar]
- CartiMorph: a framework for automated knee articular cartilage morphometrics. Med Image Anal. 2023;91
- [Google Scholar]
- Predicting knee osteoarthritis progression using neural network with longitudinal MRI radiomics, and biochemical biomarkers: a modeling study. PLoS Med. 2025;22(8)
- [Google Scholar]
- Integration of longitudinal load-bearing tissue MRI radiomics and neural network to predict knee osteoarthritis incidence. J Orthop Transl. 2025;51:187-197.
- [Google Scholar]
- Predicting knee osteoarthritis progression using structural biomarkers from multiple joints: data from the osteoarthritis initiative. Osteoarthritis Imaging. 2025;5
- [Google Scholar]
- Automated measurement and grading of knee cartilage thickness: a deep learning-based approach. Front Med. 2024;11
- [Google Scholar]
- The state of the osteoarthritis initiative (OAI): entering a new era. Semin Arthritis Rheum. 2025;75
- [Google Scholar]
- Predicting knee osteoarthritis severity from radiographic predictors: data from the osteoarthritis initiative. Ann Biomed Eng. 2025;53(8):1852-1861.
- [Google Scholar]
- Association of radiographic structure deformity phenotypes of knee OA to clinical symptoms and risk for progression: proposing a modification of kellgren-lawrence grade - data from the osteoarthritis initiative and the MOST study. Osteoarthritis and Cartilage Open. 2025;7(1)
- [Google Scholar]
- Machine learning models for clinical and structural knee osteoarthritis prediction: recent advancements and future directions. Osteoarthritis and Cartilage Open. 2025;7(3)
- [Google Scholar]
- Deep learning based classification of tibio-femoral knee osteoarthritis from lateral view knee joint X-ray images. Sci Rep. 2025;15(1)
- [Google Scholar]
- OA-MEN: a fusion deep learning approach for enhanced accuracy in knee osteoarthritis detection and classification using X-Ray imaging. Front Bioeng Biotechnol. 2025;12
- [Google Scholar]
- An AI system for continuous knee osteoarthritis severity grading: an anomaly detection inspired approach with few labels. Artif Intell Med. 2025;167
- [Google Scholar]
- Deep learning to combat knee osteoarthritis and severity assessment by using CNN-based classification. BMC Muscoskelet Disord. 2024;25(1):817.
- [Google Scholar]
- Deep learning-based framework for automated classification of knee osteoarthritis severity and detection of joint space width in X-Ray imaging. 2025:391-400.
- [Google Scholar]
- Advancing osteoarthritis research: the role of AI in clinical, imaging and omics fields. Bone Res. 2025;13(1):48.
- [Google Scholar]