Trustworthy deep learning framework for the detection of abnormalities in X-ray shoulder images.

Musculoskeletal conditions affect an estimated 1.7 billion people worldwide, causing intense pain and disability. These conditions lead to 30 million emergency room visits yearly, and the numbers are only increasing. However, diagnosing musculoskeletal issues can be challenging, especially in emergencies where quick decisions are necessary. Deep learning (DL) has shown promise in various medical applications. However, previous methods had poor performance and a lack of transparency in detecting shoulder abnormalities on X-ray images due to a lack of training data and better representation of features. This often resulted in overfitting, poor generalisation, and potential bias in decision-making. To address these issues, a new trustworthy DL framework has been proposed to detect shoulder abnormalities (such as fractures, deformities, and arthritis) using X-ray images. The framework consists of two parts: same-domain transfer learning (TL) to mitigate imageNet mismatch and feature fusion to reduce error rates and improve trust in the final result. Same-domain TL involves training pre-trained models on a large number of labelled X-ray images from various body parts and fine-tuning them on the target dataset of shoulder X-ray images. Feature fusion combines the extracted features with seven DL models to train several ML classifiers. The proposed framework achieved an excellent accuracy rate of 99.2%, F1Score of 99.2%, and Cohen's kappa of 98.5%. Furthermore, the accuracy of the results was validated using three visualisation tools, including gradient-based class activation heat map (Grad CAM), activation visualisation, and locally interpretable model-independent explanations (LIME). The proposed framework outperformed previous DL methods and three orthopaedic surgeons invited to classify the test set, who obtained an average accuracy of 79.1%. The proposed framework has proven effective and robust, improving generalisation and increasing trust in the final results.

Domain Adaptation and Feature Fusion for the Detection of Abnormalities in X-Ray Forearm Images.

The main challenge in adopting deep learning models is limited data for training, which can lead to poor generalization and a high risk of overfitting, particularly when detecting forearm abnormalities in X-ray images. Transfer learning from ImageNet is commonly used to address these issues. However, this technique is ineffective for grayscale medical imaging because of a mismatch between the learned features. To mitigate this issue, we propose a domain adaptation deep TL approach that involves training six pre-trained ImageNet models on a large number of X-ray images from various body parts, then fine-tuning the models on a target dataset of forearm X-ray images. Furthermore, the feature fusion technique combines the extracted features with deep neural models to train machine learning classifiers. Gradient-based class activation heat map (Grad CAM) was used to verify the accuracy of our results. This method allows us to see which parts of an image the model uses to make its classification decisions. The statically results and Grad CAM have shown that the proposed TL approach is able to alleviate the domain mismatch problem and is more accurate in their decision-making compared to models that were trained using the ImageNet TL technique, achieving an accuracy of 90.7%, an F1-score of 90.6%, and a Cohen's kappa of 81.3%. These results indicate that the proposed approach effectively improved the performance of the employed models individually and with the fusion technique. It helped to reduce the domain mismatch between the source of TL and the target task.

Does forearm referencing using a retroversion guide achieve the targeted retroversion of the humeral component in reverse shoulder arthroplasty?

BACKGROUND: Component positioning affects clinical outcomes of reverse shoulder arthroplasty, which necessitates an implantation technique that is reproducible, consistent, and reliable. This study aims to assess the accuracy and precision of positioning the humeral component in planned retroversion using a forearm referencing guide. METHODS: Computed tomography scans of 54 patients (27 males and 27 females) who underwent primary reverse shoulder arthroplasty for osteoarthritis or cuff tear arthropathy were evaluated. A standardized surgical technique was used to place the humeral stem in 15° of retroversion. Version was assessed intraoperatively visualizing the retroversion guide from above and referencing the forearm axis. Metal subtraction techniques from postoperative computed tomography images allowed for the generation of 3D models of the humerus and for evaluation of the humeral component position. Anatomical humeral plane and implant planes were defined and the retroversion 3D angle between identified planes was recorded for each patient. Accuracy and precision were assessed. A subgroup analysis evaluated differences between male and female patients. RESULTS: The humeral retroversion angle ranged from 0.9° to 22.8°. The majority (81%) of the measurements were less than 15°. Mean retroversion angle (±SD) was 9.9° ± 5.8° (95% CI 8.4°-11.5°) with a mean percent error with respect to 15° of -34% ± 38 (95% CI -23 to -44). In the male subgroup (n = 27, range 3.8°-22.5°), the mean retroversion angle was 11.9° ± 5.4° (95% CI 9.8°-14.1°) with a mean percent error with respect to 15° of -21% ± 36 (95% CI -6 to -35). In the female subgroup (n = 27, range 0.9°-22.8°), mean retroversion angle was 8.0° ± 5.5° (95% CI 5.8°-10.1°) and the mean percent error with respect to 15° was -47% ± 36 (95% CI -32 to -61). The differences between the 2 gender groups were statistically significant (P = .006). CONCLUSION: Referencing the forearm using an extramedullary forearm referencing system to position the humeral stem in a desired retroversion is neither accurate nor precise. There is a nonnegligible tendency to achieve a lower retroversion than planned, and the error is more marked in females.

Restoration of glenoid joint line: a three-dimensional analysis of scapular landmarks.

Background: Restoration of the glenoid joint line in shoulder arthroplasty is important for implant positioning and function. Medialization of the glenohumeral joint line due to glenoid bone loss is commonly encountered in primary and revision of shoulder arthroplasty albeit the direction and location of bone loss varies with different pathology. Three-Dimensional (3D) planning software has assisted in preoperative planning of complex glenoid deformities. However, limited literature is available defining a reliable 3D method to evaluate the glenoid joint line preoperatively. Aims: The purpose of this study is to identify a set of reliable scapular landmarks to be used as reference points to measure the premorbid glenoid joint line using 3D segmented models of healthy scapulae. Methods: Bilateral computed tomography scans from 79 patients eligible for primary stabilization procedures were retrospectively selected from our institutional surgical database (mean age 35 ± 10 years, 58 males and 21 females). 3D models of the contralateral healthy scapulae were created via computed tomography scan segmentation using Mimics 24.0 software (Materialise, Leuven, Belgium). Anatomical landmarks were identified using 3-Matic 16.0 software (Materialise, Leuven, Belgium). The distance between identified landmarks and a sagittal plane created on the deepest point of the glenoid was recorded for each scapula and reliability of each landmark was assessed. Inter- and intra-observer reliabilities were also evaluated using intraclass correlation coefficients (ICCs). Results: Four landmarks showed statistically significant results: the scapular notch (SN), the centroid of the coracoid (CC), a point on the most medial border of the scapula in line with the scapular spine (TS), and the most lateral point of the acromion (AL). The mean (± standard deviation) joint line measured from the SN, CC, TS and AL were 28.36 ± 2.97 mm, 11.66 ± 2.07 mm, 107.52 ± 8.1 mm, and 29.72 ± 4.46 mm, respectively. Inter-observer reliability analysis for SN, TS, and AL showed excellent agreement with ICC values of 0.966, 0.997, and 0.944, respectively, and moderate agreement for CC with ICC of 0.728. Conclusion: The results from this study assist in estimating joint line medialization preoperatively and in planning its subsequent restoration. A set of reliable landmarks can be used as references to estimate the premorbid glenoid joint line preoperatively.

Single-Stage Revision Reverse Shoulder Arthroplasty: Preoperative Planning, Surgical Technique, and Mixed Reality Execution.

Revision shoulder arthroplasty is increasing with the number of primary shoulder replacements rising globally. Complex primary and revisions of shoulder arthroplasties pose specific challenges for the surgeon, which must be addressed preoperatively and intraoperatively. This article aimed to present strategies for the management of revision of shoulder arthroplasties through a single-stage approach. Preoperatively, patient factors, such as age, comorbidities, and bone quality, should be considered. The use of planning software can aid in accurately evaluating implants in situ and predict bony anatomy that will remain after explantation during the revision surgery. The planning from such software can then be executed with the help of mixed reality technology to allow accurate implant placement. Single-stage revision is performed in two steps (debridement as first step, implantation and reconstruction as the second step), guided by the following principles: adequate debridement while preserving key soft tissue attachments (i.e., rotator cuff, pectoralis major, latissimus dorsi, deltoid), restoration of glenoid joint line using bone grafting, restoration of humeral length, reconstruction and/or reattachment of soft tissues, and strict compliance with the postoperative antibiotic regimen. Preliminary results of single-stage revision shoulder arthroplasty show improvement in patient outcomes (mean 1 year), successful treatment of infection for those diagnosed with periprosthetic joint infection, and improved cost-benefit parameters for the healthcare system.

Anatomically Parameterized Statistical Shape Model: Explaining Morphometry Through Statistical Learning.

OBJECTIVE: Statistical shape models (SSMs) are a popular tool to conduct morphological analysis of anatomical structures which is a crucial step in clinical practices. However, shape representations through SSMs are based on shape coefficients and lack an explicit one-to-one relationship with anatomical measures of clinical relevance. While a shape coefficient embeds a combination of anatomical measures, a formalized approach to find the relationship between them remains elusive in the literature. This limits the use of SSMs to subjective evaluations in clinical practices. We propose a novel SSM controlled by anatomical parameters derived from morphometric analysis. METHODS: The proposed anatomically parameterized SSM (ANAT[Formula: see text]) is based on learning a linear mapping between shape coefficients (latent space) and selected anatomical parameters (anatomical space). This mapping is learned from a synthetic population generated by the standard SSM. Determining the pseudo-inverse of the mapping allows us to build the ANAT[Formula: see text]. We further impose orthogonality constraints to the anatomical parameterization (OC-ANAT[Formula: see text]) to obtain independent shape variation patterns. The proposed contribution was evaluated on two skeletal databases of femoral and scapular bone shapes using clinically relevant anatomical parameters within each (five for femoral and six for scapular bone). RESULTS: Anatomical measures of the synthetically generated shapes exhibited realistic statistics. The learned matrices corroborated well with the obtained statistical relationship, while the two SSMs achieved moderate to excellent performance in predicting anatomical parameters on unseen shapes. CONCLUSION: This study demonstrates the use of anatomical representation for creating anatomically parameterized SSMs and as a result, removes the limited clinical interpretability of standard SSMs. SIGNIFICANCE: The proposed models could help analyze differences in relevant bone morphometry between populations, and be integrated in patient-specific pre-surgery planning or in-surgery assessment.

A Musculoskeletal Model Customized for Sagittal and Frontal Knee Kinematics With Improved Knee Joint Stability.

Current lower limb musculoskeletal (MSK) models focus on sagittal plane kinematics. However, abnormal gait is typically associated with sagittal plane motions crossing into other planes, limiting the use of current MSK models. The purpose of this study was twofold, first, to extend the capability of a full-body MSK model from the literature to include frontal knee plane kinematics during healthy gait, and second, to propose and implement a realistic muscle discretization technique. Two MSK model constructs were derived-the first construct (Knee2_SM) allowed two degrees-of-freedom (sagittal and coronal) at the knee and the second construct (Knee2_MM) implemented multiline elements for all the lower limb muscles in conjunction with two knee degrees-of-freedom. Motion analysis data of normal gait cycle from 10 healthy adults were used to compare joint kinematics, muscle moment arms, muscle forces, and muscle activations, between new constructs and the original model. Knee varus-valgus trajectories were estimated with the mean peak values ranging from 9.49 deg valgus to 1.57 deg varus. Knee2_MM predicted a significant difference (p < 0.05) in moment arms and forces in those muscles responsible for medial-lateral stability of the knee. The simulated muscle activations generated by the Knee2_MM model matched more closely to the experimental electromyography (EMG) when qualitatively compared. This study enhances the capability of the sagittal plane full-body MSK model to incorporate knee varus-valgus motion while keeping the joint stability intact and improving muscle prediction.

Statistical Shape Modeling to Determine the Anterior Pelvic Plane for Total Hip Arthroplasty.

The anterior pelvic plane (APP) defined by both iliac spines and the pubic symphysis, is essential in total hip arthroplasty (THA) for the orientation of the prosthetic cup. However, the APP is nowadays still difficult to determine in computer assisted orthopedic surgery (CAOS). We propose to use a statistical shape model (SSM) of the pelvis to estimate the APP from ipsilateral anatomical landmarks, more easily accessible during surgery in computer assisted THA with the patient in lateral decubitus position. A SSM of the pelvis has been built from 40 male pelvises. Various ipsilateral anatomical landmarks have been extracted from these data and used to deform the SSM. Fitting the SSM to several combinations of these landmarks, we were able to reconstruct the pelvis with an accuracy between 2.8mm and 4.4mm, and estimate the APP inclination with an angular error between 1.3° and 2.8°, depending on the landmarks fitted. Results are promising and show that the APP could be acquired during the intervention from ipsilateral landmarks only.

Clinical relevance of augmented statistical shape model of the scapula in the glenoid region.

OBJECTIVE: To illustrate (a) whether a statistical shape model (SSM) augmented with anatomical landmark set(s) performs better fitting and provides improved clinical relevance over non-augmented SSM and (b) which anatomical landmark set provides the best augmentation strategy for predicting the glenoid region of the scapula. METHODS: Scapula SSM was built using 27 dry bone CT scans and augmented with three anatomical landmark sets (16 landmarks each) resulting in three augmented SSMs (aSSMproposed, aSSMset1, aSSMset2). The non-augmented and three augmented SSMs were then used in a non-rigid registration (regression) algorithm to fit to six external scapular shapes. The prediction error by each type of SSM was evaluated in the glenoid region for the goodness of fit (mean error, root mean square error, Hausdorff distance and Dice similarity coefficient) and for four anatomical angles (critical shoulder angle, lateral acromion angle, glenoid inclination, glenopoar angle). RESULTS: Inter- and intra-observer reliability for landmark selection was moderate to excellent (ICC>0.74). Prediction error was significantly lower for SSMnon-augmented for mean (0.9 mm) and root mean square (1.15 mm) distances. Dice coefficient was significantly higher (0.78) for aSSMproposed compared to all other SSM types. Prediction error for anatomical angles was lowest using the aSSMproposed for critical shoulder angle (3.4°), glenoid inclination (2.6°), and lateral acromion angle (3.2°). CONCLUSION AND SIGNIFICANCE: The conventional SSM robustness criteria or better goodness of fit do not guarantee improved anatomical angle accuracy which may be crucial for certain clinical applications in pre-surgical planning. This study provides insights into how SSM augmented with region-specific anatomical landmarks can provide improved clinical relevance.

Statistical Shape Modeling Approach to Predict Missing Scapular Bone.

Prediction of complete and premorbid scapular anatomy is an important aspect of successful shoulder arthroplasty surgeries to treat glenohumeral arthritis and which remains elusive in the current literature. We proposed to build a statistical shape model (SSM) of the scapula and use it to build a framework to predict a complete scapular shape from virtually created scapular bone defects. The bone defects were synthetically created to imitate bone loss in the glenoid region and missing bony part in inferior and superior scapular regions. Sixty seven dry scapulae were used to build the SSM while ten external scapular shapes (not used in SSM building) were selected to map scapular shape variability using its anatomical classification. For each external scapula, four virtual bone defects were created in the superior, inferior and glenoid regions by manually removing a part of the original mesh. Using these defective shapes as prior knowledge, original shapes were reconstructed using scapula SSM and Gaussian process regression. Robustness of the scapula SSM was excellent (generality = 0.79 mm, specificity = 1.74 mm, first 15 principal modes of variations accounted for 95% variability). The validity and quality of the reconstruction of complete scapular bone were evaluated using two methods (1) mesh distances in terms of mean and RMS values and (2) four anatomical measures (three angles: glenoid version, glenoid inclination, and critical shoulder angle, and glenoid center location). The prediction error in the angle measures ranged from 1.0° to 2.2°. For mesh distances, highest mean and RMS error was 0.97 mm and 1.30 respectively. DICE similarity coefficient between the original and predicted shapes was excellent (≥ 0.81). This framework provided high reconstruction accuracy and can be effectively embedded in the pre-surgical planning of shoulder arthroplasty or in morphology-based shoulder biomechanics modeling pipelines.

Subject-specific shoulder muscle attachment region prediction using statistical shape models: A validity study.

Subject-specific musculoskeletal models can predict accurate joint and muscle biomechanics thereby helping clinicians and surgeons. Current modeling strategies do not incorporate accurate subject-specific muscle parameters. This study reports a statistical shape model (SSM) based method to predict subject-specific muscle attachment regions on shoulder bones and illustrates the concurrent validity of the predictions. Augmented SSMs of scapula and humerus bones were built using bone meshes and five muscle attachment (origin/insertion) regions which play important role in the shoulder motion and function. Muscle attachments included Subscapularis, Supraspinatus, Infraspinatus, Teres Major and Teres Minor on both the bones. The regions were represented by subset of vertices on the bone meshes and were tracked using vertex identifiers. Subject-specific muscle attachment regions were predicted using external set of bones not used in building the SSMs. Validity of predictions was determined by visual inspection and also by using four similarity measures between predicted and manually segmented regions. Excellent concurrent validity was found indicating the higher accuracy of predictions. This method can be effectively employed in modeling pipelines or in automatic segmentation of medical images. Further validations are warranted on all the muscles of the shoulder complex.