2.1 Imaging of elbow overuse injuries in athletic populations

In the recent past, medical imaging and computational analysis has advanced the role of radiographs and other advanced modalities in the determination of sports related musculoskeletal injuries. Preliminary models were involved in automated fracture and injury detection based on the imaging model, proving that it was possible to extract clinically meaningful features of regular radiographs.¹ In addition to acute injuries, there is an increasing amount of evidence that repetitive mechanical loading is a key factor in the pathology of overuse musculoskeletal in sportspeople.²

Elbow overuse arthritis is a documented injury in the athletic group especially overhead and throwing athletes who are susceptible to repetitive valgus and axial loading and high-cycle joint articulation.³,⁴,⁵ Various types of clinical and radiological studies have characterized a continuum of elbow pathologies such as valgus extension overload, osteochondritis dissecans, Panners lesion, microinstability and lateral elbow disorders.⁶,⁷,⁸ Reviews of the imaging focus on how radiographs and magnetic resonance imaging are used in the detection of known structural abnormalities including subchondral sclerosis, cartilage defects, joint incongruity, and ligamentous injury.⁶,⁹ But such studies deal mainly with symptomatic or advanced disease and give little information on how cumulative stress changes can be detected at an early stage.

MRI has proven to be very sensitive in the detection of soft tissue as well as cartilage abnormalities in the elbow and other athletic joints.⁹,¹⁰ Stress-related MRI investigations have demonstrated a strong association between widening of the joint space and exposure to workload on the part of professional pitchers supporting the association of repetitive mechanical load and structural change.¹⁰ Radiographic stress images have also been investigated to identify small amounts of joint instability under loads, which can be viewed as the promise of stress-based imaging in the identification of early mechanical changes.¹¹ Although these benefits exist, the limitation of MRI-based methods is still in the cost, availability and the lack of capability of the method to conduct longitudinal monitoring regularly to support orthopaedic practice.

2.2 Quantitative and radiomics-based musculoskeletal imaging

The application of quantitative imaging and radiomics to assess musculoskeletal injury has gained more importance to enhance diagnostic performance and prognostic assessment. Radiomics and deep learning-based methods have been applied with success in fracture, ligament injury, and degenerative joint disease grading in various anatomical areas.¹,¹²,¹³,¹⁴,¹⁵ These techniques derive high dimensional features on medical images to improve classification and prediction.

Artificial intelligence-based models have also been suggested in sports medicine in injury prevention and diagnosis, such as multimodal injury prediction through the combination of imaging, biomechanical, and temporal data.¹⁶,¹⁷,¹⁸,¹⁹,²⁰ Although these methods show a promising predictive performance, most of them are based on pixel-wise segmentation, complicated architecture, or massive annotated datasets, which reduces interpretability and clinical integration. It is stressed in reviews that integration into the orthopaedic practice must have transparent anatomically based measures that are based on clinical reasoning in addition to the use of black-box prediction.¹⁷,¹⁹

2.3 Biomechanics, fatigue, and overuse injury mechanisms

Biomechanical studies have established that repetitive submaximal loading changes joint kinematics, coordination and distribution of tissue stress in overhead athletes.²¹,²² Experimental and translational research continues to show that cumulative mechanical fatigue is related to overuse injury, which leads to a progressive microstructural damage and long-term functional loss in musculoskeletal tissues.² Other putatively similar adaptations associated with fatigue have been reported in other joints of the body such as the shoulder in adolescent and overhead sportsmen, indicating the generalizability of cumulative mechanical loading in the development of sports-related injury.²³,²¹

Wearable sensor-based technologies and motion analysis models have become a growing use to measure exposure of external workloads, variability of movements, and increase risk of fatigue-stressful injury.²⁴,²⁵,²⁶ These methods have the benefit of offering important information on biomechanical loading patterns, but they focus mostly on external loads and the nature of motion and are not a direct measure of internal structural adaptation in the joint as seen on routine imaging.

2.4 Data resources and imaging benchmarks

Publicly available musculoskeletal imaging datasets have facilitated the development and validation of quantitative imaging methods. Reproducible assessment of imaging-based algorithms has been facilitated by standardized collections of elbow radiographs with classified fractures and degenerative conditions.²⁷ Besides, open-source pediatric elbow radiography databases have allowed a wider exploration of morphological variability and methods of radiographic analysis.²⁸ These sources describe the increasing attention to objective radiographic testing and indicate the necessity to have clinically interpretable paradigms that can be applied to the athletic populations.

2.5 Identified research gap and study positioning

Taken altogether, the current body of literature offers solid grounds in terms of clinical characterization, sophisticated imaging, biomechanics, and computerized analysis of sports-related musculoskeletal injuries.³,⁴,¹⁷,⁵,²⁹ Nevertheless, much imaging research concentrates on well-known pathology, instead of pre-symptomatic risk, and advanced quantitative approaches to imaging often lack clinical interpretability, and biomechanical monitoring models are not explicitly interested in internal structural adaptation. More so, limited methods use the ubiquitous nature of routine elbow radiographs to objectively measure cumulative mechanical stress with anatomically reasonable measures consistent with orthopaedic arguments.

Alternatively, the current work presents a structural fatigue modeling framework into the transparent objective evaluation of cumulative mechanical stress in the athletic elbow through the standard radiographs. The proposed method has surpassed a major orthopaedic imaging gaps in the sports medicine field, enabling the early risk stratification and preventive decision-making by combining patient-specific anatomical normalization with fatigue-sensitive structural markers, without the use of pixel-based segmentation and opaque computational models.

Table 1 outlines possible current imaging, biomechanical, and computational methods and compares them to the suggested framework of structural fatigue model.

3.1 Study design and datasets

This study employed a retrospective observational design to develop and validate an imaging-based structural fatigue modeling framework for assessment of cumulative mechanical stress in the athletic elbow. Robustness and generalizability were tested with the help of two independent datasets. The main data was comprised of regular elbow radiographs obtained on sportspeople playing overhead and throwing games and exposed to different degrees of repetitive elbow loading. Radiographs were not only provided with recorded workload history and long-term clinical follow-up that includes further diagnosis of elbow overuse injury or the long-term asymptomatic state. Only normal anteroposterior and lateral radiographs of the elbow taken as part of the routine clinical care were considered.

Testing validation External validation came by way of an independent athletic cohort with no overlapping subjects and imaging acquisition environment. This dataset was applied only to independent testing and performance evaluation to analyze whether the results are generalizable to populations and imaging conditions. All radiographs were coded before analysis. The research was based on de-identified retrospective data and did not affect clinical decision-making.

The primary dataset included 128 athletes, comprising 52 individuals who developed elbow overuse injury during follow-up and 76 who remained asymptomatic. The independent external validation dataset included 64 athletes (26 injured, 38 asymptomatic), acquired from a separate cohort with no subject overlap. All subjects had documented workload exposure and longitudinal clinical follow-up.

The summary of the baseline demographic and clinical characteristics of the study cohorts are summarized in Table 2. The major dataset was athletes who had undergone repetitive loading on the elbow, among whom some of them developed the elbow overuse injury in future and others did not. The external validation data was obtained separately and showed similar demographic and workload features, which proves the generalizability of the proposed framework of structural fatigue modeling. There was no significant difference in the demographic distribution among datasets in any respect.

3.2 Inclusion and exclusion criteria

Radiographs were included in the study if they met the following criteria:1)acquisition using standard clinical elbow radiographic protocols (anteroposterior and lateral views);2)adequate image quality to permit reliable identification of anatomical landmarks; and3)availability of associated workload exposure information and clinical follow-up data.

Radiographs were eliminated when they revealed acute fracture, previous surgery, developmental anomaly, or severe degenerative alteration that may complicate measurement of cumulative mechanical stress. Mages with severe motion artefacts, improper positioning, or incomplete visualization of the elbow joint were also excluded.

These criteria were used throughout the primary and external validation datasets to guarantee methodological coherence and compatibility.

3.3 Overview of the proposed structural fatigue modeling framework

Fig. 1 shows the system structure of the proposed radiograph-based structural fatigue model of cumulative mechanical stress evaluation on the athletic elbow. Yet, the framework works based on regular anteroposterior and lateral elbow radiographs and is aimed at measuring fatigue-induced structural changes without any pixel-by-pixel segmentation or complex imaging options.

Close

Fig. 1 Proposed structural fatigue modeling framework for the athletic elbow.

Export to PPT

Fig. 1 illustrates that the input to the system is standard elbow radiographs, which after preprocessing (standardization of image appearance and orientation) are then transmitted to a processing module. It involves grayscale normalization, orientation correction and image standardization to achieve consistency between subjects and imaging conditions.

The preprocessed radiographs are then evaluated in the anatomic analysis module where patient specific peaks of the bones can be recognized throughout the distal humerus, proximal ulna and radial head. These landmarks give a consistent anatomical point of reference to be used later. Normalization of geometry is then carried out to get patient specific alignment and scale normalization, which can compare meaningful structuring indicators among different individuals.

The structural fatigue analysis module after normalization filters fatigue-sensitive structural indicators of cumulative mechanical loading. Such markers are imbalanced density in the region, deviation in articulation continuity, directional gradients of micro-deformation between the humeroulnar and radiocapitellar joints. The mined indicators are incorporated as an orthopaedic reasoning-based integration module to calculate cumulative mechanical stress index.

The final output of the framework is an injury risk stratification, categorizing athletes into low-, moderate-, or high-risk groups based on the computed cumulative mechanical stress index. This is a structured workflow that allows clinically interpretable evaluation of cumulative assessments of mechanical stress through routine elbow radiographs, which will assist in identifying those at high risk of elbow overuse injury earlier on in athletes.

3.4 Radiographic preprocessing

All the radiographs of the elbow underwent standardized preprocessing to reduce the variability caused by the conditions of image acquisition and provide consistency of the datasets. The preprocessing step represents the preprocessing module as shown in Fig. 1 and was applied in a consistent way to all images before any further analysis.

Preprocessing was performed with grayscale normalization to minimize intensity differences across radiographs, orientation remediation to guarantee equal anatomical positioning, and standardization of the images to have equal spatial representation. These procedures were done to enable normalization of geometry and anatomical landmarks which would be identified in subsequent stages.

There was no manual and automated pixel-based segmentation in preprocessing. The suggested framework uses the processed radiographs directly in the proposed framework, which ensures preservation of anatomical integrity and does not have variability based on segmentation or operator dependence.

3.5 Anatomical landmark identification

After radiographic preprocessing, anatomical landmarks of the patient were established to provide a standard reference point to be used in structural analysis of the elbow joint. This step is associated with the anatomical analysis component, as demonstrated in Fig. 1 and gives the anatomy of structure normalization and the further structural fatigue evaluation.

The criteria used in choosing landmarks were reproducibility, anatomical relevance and biomechanical importance in the transmission of loads over the elbow joint. The most important anteroposterior and lateral radiographic landmarks were recognized, which are the medial and lateral epicondyles of the distal humerus, radial head, capitellum, trochlear tip, and olecranon. These landmarks are consistent points of anatomy, which are related to the joints of the elbow joints in the distribution of force and articulation during recurring movements.

The anteroposterior and lateral radiographs of the elbow with the identified landmarks marked are shown in Fig. 2. As demonstrated, landmark placement maintains native anatomical associations and allows a consistent definition of joint alignment and articulation geometry among subjects.

Close

Fig. 2 Anatomical landmark identification on routine elbow radiographs.

Export to PPT

The identification of the landmarks of the anatomy was also done on the processed radiographs without pixel-by-pixel segmentation. This non-segmentation method is less sensitive to image noise and operator bias and still clinically interpretable. The identified landmarks were used as the reference points to establish spatial reference axes, as well as normalize the geometry of elbows at the subsequent stages of the proposed framework.

3.6 Geometry normalization

After the landmark identification of the anatomical landmarks, the geometry of the elbows was normalized to consider the inter-individual differences in the size, orientation, and radiographic projection of the joint. This step is synonymous with the module of geometry normalization as shown in Fig. 1 and is founded on the anatomical landmarks of the patient determined in Fig. 2 to allow the structural fatigue indicators to be compared across participants in a consistent manner.

Based on the set anatomical landmarks, a subject-specific coordinate system was determined in relation to each elbow radiograph. To normalize the orientation of the joints, landmark-based alignment was used and to reduce the impact of variations in the size of bones and in imaging magnification, scale normalization was used. This procedure conserved the anatomical relationships of the natives as it allowed standardized analysis of space.

Fig. 3 shows sample before and after geometry normalization radiographs of the elbow. As indicated, normalization helps to eliminate rotational differences and scaling differences without affecting the anatomical consistency of the humeroulnar and radiocapitellar joints. The reason to take this step is to make sure the subsequent indicators of structural fatigue are related to cumulative mechanical loading, not variability that has been added by radiographic acquisition, or subject positioning.

Close

Fig. 3 Elbow radiographs before and after geometry normalization.

Export to PPT

Normalized elbow geometry was used as the reference frame in extracting fatigue sensitive structural indicators as described in the next section.

3.7 Structural fatigue indicator extraction

After geometry normalization, we extracted fatigue sensitive structural clues that were obtained using the normalized elbow radiographs to describe cumulative mechanical stress related to repetitive loading. This step is the same as the structural fatigue analysis module in Fig. 1 and was implemented to detect minor remodelling processes in the structure over time as opposed to the acute pathological processes.

There were three types of fatigue indicators structures generated. The density imbalance was measured regionally to determine the bone density distribution asymmetry in the loading areas of the humeroulnar and radiocapitellar joints as a sign of localized adaptive remodelling in response to repetitive forces. Continuity of articulation was measured to quantify fine changes in joint congruity and fit which might occur due to micro-deformation accruing in the joint of repetitive loading. Directional micro-deformation gradients were measured to describe spatial patterns of adaptation of the structure in harmony with the well-known biomechanical loading directions throughout the elbow joint.

The extraction of indicators was done in the normalized radiographs without using the pixel-wise segmentation. This image noise independent and operator independent segmentation method reduced sensitivity to image noise and operator variability and maintained clinically interpretable anatomical relationships. All indicators have been calculated using the normalized coordinate system to enable parallel subject and data results.

The obtained structural fatigue indicators were complementary data on the cumulative mechanical loading, and they were the inputs in calculating cumulative mechanical stress index stated in the next section.

3.8 Cumulative mechanical stress index

The structural fatigue indicators were extracted and combined to give a Cumulative Mechanical Stress Index (CMSI) of the total burden of repetitive mechanical loading on the elbow joint. This index was intended to give one clinically interpretable cumulative structural fatigue measure using routine radiographs.

Before combining measures, the individual structural fatigue scales were brought to a normalized level so that similar markers were contributed to across different scales. It introduced orthopaedic reasoning-based integration structure of indicators, which made use of the complementary quality of density imbalance, articulation continuity deviation, and directional micro-deformation gradients. This method focused on cumulative structural adjustment and not individual radiographic alterations.

The resultant CMSI is the summation of the recurrent mechanical loading on the humeroulnar and radiocapitellar joints. An increase in CMSI values signifies the increased cumulative mechanical stress, and a higher probability of the development of the elbow overuse injury, whereas a decrease in CMSI value denotes the limited structural adaptation to fatigue.

The CMSI was the main quantitative deliverable of the proposed framework and it was utilized in the further risk stratification and performance assessment, as illustrated in the next section.

3.9 Validation strategy and model evaluation

The performance and robustness of the proposed structural fatigue modeling framework were evaluated using both internal validation and independent external testing. Model development and parameter tuning were performed exclusively on the primary dataset, while the external validation dataset was reserved for independent performance assessment to evaluate generalizability across populations and imaging conditions.

A cross-validation approach was used to check internal validation using primary datasets. The subjects were divided into non-overlapping samples to balance the sample between the athletes who obtained an elbow overuse injury and those who did not. This method reduced overfitting and allowed measuring the stability of models with varying data partitions.

The trained model was then used on the external validation dataset, and no retraining or parameter modification was done. The generalizability of the offered approach to unseen data and independent imaging settings was evaluated based on the performance on the external dataset.

The accuracy, sensitivity and specificity were used in the classification of high risk and low risk cases to quantify model performance. Moreover, correlation examination was conducted to measure the association between the cumulative stress index on the mechanical system and the ensuing injury. The correlation analysis was done using Pearson correlation test and a preset significance level of p < 0.05.

4.1 Performance on the primary dataset

The proposed radiograph-based structural fatigue modeling framework demonstrated strong performance on the primary dataset comprising athletes exposed to varying levels of repetitive elbow loading. The framework had a total classification accuracy of 91.6% in high-risk and low-risk cases with the use of cross-validation.

The sensitivity to detect the presence of athletes who later developed elbow overuse injury was 90.8%, and thus it can be concluded that it is very good in detecting high-risk individuals before they develop symptoms. Specificity with respect to the accurate identification of low-risk athletes who were not identified in the follow-up was 93.2%, which represents a strong non-injury case discrimination. These results suggest that cumulative mechanical loading based on standard elbow radiographs may be an effective way of describing the fatigue-based structural adaptation in the athletic elbow.

4.2 External validation results

Analysis of external validation dataset showed consistent performance under independent evaluation, which can be used to support the generalizability of the proposed framework. The model in this dataset had a total of 89.4% accuracy with the sensitivity of 88.1% and specificity of 91.7%.

The similarity of the performance of the primary and external datasets suggests that the framework is resilient to changes in the subject population and conditions of imaging and is not based on the characteristics of the databases.

4.3 Association between cumulative mechanical stress and injury occurrence

The results of the correlation analysis revealed that the cumulative mechanical stress index is strongly positively correlated with the occurrence of the subsequent elbow overuse injury. On the primary dataset, significant correlation was found (r = 0.71, p < 0.01) and thus the higher the values of cumulative mechanical stress index, the higher was the probability of developing an injury in the follow-up period. The same was also found in the external validation dataset (r = 0.69, p < 0.01).

These findings support the clinical relevance of the proposed index as a quantitative marker of cumulative mechanical stress and its potential utility for early risk stratification prior to symptom onset.

4.4 Illustrative comparative performance analysis

Table 3 shows a demonstrative performance analysis between the proposed framework and typical existing imaging-based methods under the conditions of normalized experimentation. Previous studies simulate performance values to contextually compare and do not reflect reported outcomes of the respective publications. The comparison, which follows, is done to give the contextual interpretation of relative methodological performance as opposed to direct benchmarks across studies.

As illustrated in Table 3, the proposed framework exhibits desirable illustrative results using accuracy, sensitivity, specificity, and correlation measures when only normal elbow radiographs are used. It is important to note that the suggested approach delivers stable performance on the internal and external evaluation without the use of pixel-wise segmentation and high-imaging modalities. These results demonstrate that structural fatigue modeling can have a clinical benefit in terms of radiograph evaluation of accumulated mechanical stress in athletic groups.

Fig. 4 gives a visual comparison illustration of classification performance of representative imaging-based methods and the proposed structural fatigue modeling framework. The figure shows the values of accuracy, sensitivity and specificity of both methods, and it can be intuitively assessed in terms of relative performance trends under normalized assumptions.

Close

Fig. 4 Illustrative comparative performance of imaging-based methods.

Export to PPT

As demonstrated, the traditional imaging-based and multimodal techniques displayed moderate results on the considered metrics. Conversely, the suggested framework has a greater level of illustrative accuracy, sensitivity and specificity with the primary and external validation data and uses only routine elbow radiographs. The high level of stability in the provided strategy is underscored by the consistent results of the internal and external analyses.

Notably, Fig. 4 is supposed to be used in the contextual sense of interpretation but not benchmarking. The performance values are simulated under some normalized assumptions of previous studies do not reflect reported values of results in the relevant publications. The figure with Table 3 completes the visualization of relative methodological performance and supports the possibility of clinical usefulness of structural fatigue modeling in early risk stratification.

Overall, the proposed framework achieved high classification performance on both primary and external datasets and demonstrated a strong association between imaging-derived structural fatigue and subsequent injury occurrence. These findings argue in favor of the capability of making objective determination of cumulative mechanical stress and earlier detection of athletes at high risk of elbow overuse injury by the use of routine elbow radiographs.

5.1 Clinical implications

The suggested structural fatigue modeling allows orthopaedic surgeons and sports physicians to derive objective and fatigue-sensitive data out of routine elbow radiographs. In this way, by helping to detect athletes at higher risk of injury in their early stages, this method might enable proactive workload adjustment, specific monitoring, and preventive intervention as part of an ordinary clinical practice.

5.2 Limitations

This study has several limitations. The proposed framework relies on radiographic assessment and does not directly evaluate soft-tissue or ligamentous pathology, which may also contribute to elbow overuse injuries. The exposure to workloads was sport discipline and competition specific and might differ among sporting disciplines and levels of competition. Even though the external validation was conducted, future multi-central studies of larger groups would better support the generalizability. The clinical utility of structural fatigue modeling could be improved in the future by multimodal imaging and longitudinal workload monitoring.