Normalize the radiograph
Each knee X-ray is prepared and resized into several model-specific resolutions.
AI-assisted musculoskeletal imaging
Objective knee OA grading.
From one X-ray.
OrthoGrade AI combines eight complementary deep-learning networks to estimate Kellgren–Lawrence severity from knee radiographs—then reveals the joint regions that influenced the decision.
Powered with ❤️ by Solicitous ®
0Knee X-rays
0Model families
0KL severity grades
0Cross-validation folds
Predicted KL grade
0
Ensemble confidence
91%
THE CLINICAL SCALE
Five stages. One continuous severity story.
The Kellgren–Lawrence scale progresses from a radiographically normal joint to severe osteoarthritis. Select a grade to inspect the characteristic pattern and watch the dataset spectrum respond.
Representative radiograph
Joint space
Marginal bone
KL GRADE0
No radiographic osteoarthritis
Normal-appearing joint space without definite osteophytes or structural narrowing.
Joint-space narrowingNone
Osteophyte evidenceNone
Structural deformityNone
Important The grade is a radiographic severity category, not a standalone diagnosis. Symptoms and clinical context still matter.
ANIMATED DATASET VIEW
Severity spectrum scanner
Grade 0 active
WHAT IS HAPPENING
Eight specialists vote on one answer.
Instead of trusting a single neural network, the system trains a portfolio of architectures at image sizes that suit each model. During inference, original and horizontally flipped views are combined through mixed hard-and-soft voting.
Run diverse CNNs
Eight pretrained model families learn complementary texture, shape and joint-space cues.
Test-time augmentation
The original image and a horizontal flip are evaluated to make predictions more stable.
Mixed ensemble vote
Hard class votes and soft probabilities are fused into the final grade estimate.
Explain the focus
Grad-CAM highlights the image regions contributing most strongly to the prediction.
MODEL ARCHITECTURE
Model-specific image sizing
Different network families receive images at resolutions selected for their own representational strengths, rather than forcing every model to use one common size.
EXPLAINABLE AI
See where the ensemble is looking.
Grad-CAM heatmaps provide a visual check that the network is attending to clinically meaningful regions—especially the tibiofemoral joint space and areas associated with osteophyte formation.
Joint-space narrowing
The gap between femur and tibia becomes progressively reduced as cartilage loss advances.
Osteophyte formation
Marginal bony growths around the joint are a defining radiographic cue in KL grading.
Subchondral change
Increased density and structural remodeling may appear with more advanced disease.
Bone-end deformity
Severe grades can show marked contour change and loss of normal joint architecture.
REPORTED RESEARCH PERFORMANCE
Consistency through diversity.
The ensemble was evaluated on a held-out test set after stratified five-fold cross-validation. Results below reflect the reported experimental study and are not a claim of clinical deployment performance.
Overall accuracy76.93%
Ensemble
F1 score0.7665
Balance between precision and recall across five ordered classes.
Study images8,260
Posterior–anterior fixed-flexion knee radiographs across KL grades 0–4.
Independent test set1,656
Held apart from cross-validation and model training.
GENERALIZATION STRATEGY
Stratified five-fold validation
The 6,604 training-and-validation images are divided into five class-balanced folds. Four folds train the models while one validates them, rotating until every fold has served as validation.
Training foldsValidation fold
Run 1
Run 2
Run 3
Run 4
Run 5
AI
INSTRUMENT-READY STORY
From acquisition to structured severity insight.
For a demonstration environment, the workflow can sit beside a digital radiography system: receive a knee image, estimate the KL grade, expose the attention map, and produce a standardized summary for clinician review.
Automatic image intakeFive-class probability outputAttention-map reviewPDF-ready summary
Research presentation notice
This website demonstrates a published experimental AI workflow. It is not a medical device and must not be used as the sole basis for diagnosis, treatment or patient management.
ORTHOPAEDIC IMAGING INTELLIGENCE
Return to presentation start ↑
Make every knee X-ray easier to interpret, compare and communicate.
Objective grading. Visible evidence. A presentation built for clinical, laboratory and instrument stakeholders.
Powered with ❤️ by Solicitous ®