Automated Classification of Knee Osteoarthritis Using Radiographic Analysis

Abstract

Thisprojectaimstodevelopanautomateddeeplearningmodelforclassifyingkneeosteoarthritis (KOA)severity into five stages based onthe Kellgren-Lawrence(KL)gradingsystem, usingX-ray images. KOAisadegenerativejoint diseasethataffects millionsworldwide,andaccurategrading isessential for proper diagnosis andtreatment. However,manualassessmentofX-rayimagescan be subjective and time-consuming, making automation crucial for improving diagnostic efficiency and consistency. The model will utilize the ResNet-50 architecture, a powerful convolutional neural network(CNN) knownforit sability toextractcomplexfeaturesandtraindeepnetworkseffectively. ResNet-50 will process knee X-ray images and classify them into KLgrades 0 to 4, ranging from healthytosevereosteoarthritis.Byapplyingtransferlearning, themodelwillbepre-trainedon large datasets and fine-tuned using KOA-specific data. Additionally, data augmentation techniques such as rotation, flipping, and zooming will be used to enhance training data diversity.The model’s performancewillbeevaluatedusingaccuracy,precision,recall,andF1-scoremetrics.Thegoalis to provide radiologists with an automated, objective tool that improves the speed and consistency of KOA diagnosis, ultimately contributing to better patient care and more efficient clinical decision- making

Introduction

Overview

Knee osteoarthritis (KOA) is a common degenerative joint disease, especially among the elderly, marked by cartilage degradation leading to pain, stiffness, and mobility issues. Diagnosing KOA early is crucial for effective treatment. The Kellgren-Lawrence (KL) scale is the standard grading system, ranging from 0 (no OA) to 4 (severe OA), typically based on visual assessment of X-rays—a process that is often subjective and inconsistent.

AI in KOA Diagnosis

To address subjectivity and improve consistency, deep learning models, particularly ResNet-50, are being used to automate KOA grading from X-rays. ResNet-50's residual connections help capture complex visual features, overcoming challenges like vanishing gradients. Using this model allows for objective, fast, and reliable KOA severity classification.

Proposed Model

The proposed approach involves:

• ResNet-50 for classification into KL grades 0 to 4.

• Transfer learning from ImageNet to handle limited medical data.

• Advanced image preprocessing: resizing, contrast enhancement, edge preservation, and normalization.

• Data augmentation: flipping, rotation, brightness/contrast adjustments, and zoom to correct class imbalance.

• Class-weighted loss function to improve prediction on underrepresented KL-3 and KL-4 grades.

• Evaluation metrics: accuracy, precision, recall, mean absolute error, and quadratic weighted kappa.

Related Work Highlights

Several studies have contributed to the field:

1. Abedin et al.: Combined clinical data and CNNs for KOA severity prediction.

2. Yong et al.: Used ordinal regression instead of standard classification to better capture KL grade order.

3. Rehman et al.: Developed the CRK model combining CNN with Random Forest and k-NN, achieving 90% accuracy.

4. Jain et al.: Introduced attentive multi-scale CNNs to detect subtle joint changes.

5. Yeoh et al.: Created multi-task learning models using MRIs for segmentation and classification.

6. Tariq et al.: Used ensemble deep learning for KL classification with 87% accuracy.

7. Jahan et al.: Proposed KOA-CCTNet, a transformer-based model with 84.6% accuracy using a large augmented dataset.

8. Rehman et al.: Achieved >93% accuracy using a VGG16+CNN hybrid.

9. Naguib et al.: Classified uni- and bicompartmental KOA types using CNNs.

10. Chen et al.: Developed a bi-modal model combining thermal imaging and clinical data with 89% accuracy.

Methodology

• Data Collection: KOA X-rays labeled by KL grades from hospitals and public datasets.

• Preprocessing: Standardization, resizing, and enhancement to improve model accuracy.

• Model Architecture: ResNet-50 with fine-tuning and a custom classification head.

• Training: Incorporates image normalization and loss balancing to improve learning from imbalanced datasets.

• Quality Control: Ensures only high-quality, properly labeled images are used in training and testing.

Conclusion

In conclusion, this project has the potential to transform KOA diagnosis and treatment planning, making AI a vital tool in orthopedic care.By combining deep learning advancements, real-world deployment strategies, and predictive healthcare analytics, the future of AI-powered KOA management looks promis- ing.TheuseoftheseAI-poweredmodelswillhelpthemedicalcommunityaddresstherisingincidence of osteoarthritis globally, improve patient care, and increase diagnostic accuracy.

References

[1] Abedin, J., Antony, J., McGuinness, K. et al. ”Predicting knee osteoarthritis severity:com- parative modeling based on patient’s data and plain X-ray images”. Sci Rep 9, 5761 (2021). https://doi.org/10.1038/s41598-019-42215-9 [2] Yong,C.W.,Teo,K.,Murphy,B.P.etal.”Kneeosteoarthritisseverity classificationwithordinal regressionmodule”.MultimedToolsAppl81, 41497–41509(2022). https://doi.org/10.1007/s11042-021-10557-0 [3] Antony,J.,McGuinness,K.,O’Connor,N.,Moran,K.,Smeaton,A.(2023),”AutomaticDetection of Knee Osteoarthritis Using Convolutional Neural Networks,” in Proceedings of the International Conference on Pattern Recognition, 2016 [4] S. U. Rehman and V. Gruhn, ”A Sequential VGG16+CNN-Based Automated Approach With Adap- tiveInputforEfficientDetectionofKneeOsteoarthritisStages,”inIEEEAccess,vol.12,pp.62407- 62415,2024,doi:10.1109/ACCESS.2024.3395062. [5] P. S. Q. Yeoh, S. L. Goh, K. Hasikin, X. Wu and K. W. Lai, ”3D Efficient Multi-Task Neural Network for Knee Osteoarthritis Diagnosis Using MRI Scans:Data From the Osteoarthritis Initiative,” in IEEE Access, vol. 11, pp. 135323-135333, 2023, doi:10.1109/ACCESS.2023.3338379 [6] Jain,R.K.,Sharma,P.K.,Gaj,S.etal.”Kneeosteoarthritisseveritypredictionusinganatten- tive multi-scale deep convolutional neural network”. Multimed Tools Appl 83, 6925–6942 (2024). https://doi.org/10.1007/s11042-023-15484-w [7] M. Jahan et al., ”KOA-CCTNet: An Enhanced Knee Osteoarthritis Grade Assessment Framework UsingModifiedCompactConvolutionalTransformerModel,”inIEEEAccess,vol.12,pp.107719- 107741,2024,doi:10.1109/ACCESS.2024.3435572 [8] A. Rehman, A. Raza, F. S. Alamri, B. Alghofaily and T. Saba, ”Transfer Learning-Based Smart Features Engineering for Osteoarthritis Diagnosis From Knee X-Ray Images,” in IEEE Access, vol. 11,pp.71326-71338,2023,doi:10.1109/ACCESS.2023.3294542 [9] Soaad M. Naguiba, Mohamed A. Kassemb, Hanaa M. Hamzac, Mostafa M. Foudad, Mohammed K. Salehe, Khalid M. Hosnyc ”Automated system for classifying uni-bicompartmental knee osteoarthri- tis by using redefined residual learning with convolutional neural network” in Heliyon, Volume 10, Issue 10, e31017, doi:10.1109/ACCESS.2024.3395062. [10] J.Chen,B.Ma,M.Hu,G.Zhai,W.Q.SunandS.X.Yang,”ObjectiveBi-ModalAssessmentofKnee OsteoarthritisSeverityGrades: ModelandMechanism,”in IEEETransactionson Instrumentation andMeasurement,vol.73,pp.1-11,2024,Artno.4508611,doi:10.1109/TIM.2024.3462998 [11] Antony,J.,McGuinness,K.,O’Connor,N.,Moran,K.,Smeaton,A.(2023),”AutomaticDetection of Knee Osteoarthritis Using Convolutional Neural Networks,” in Proceedings of the International Conference on Pattern Recognition, 2016. [12] Norman,B.,Pedoia,V.,Majumdar,S.(2022),”UsingDeepLearningforFullyAutomatedMR Imaging-based Cartilage Loss Assessment in Knee Osteoarthritis,” Radiology, 290(2):398-407. [13] Tiulpin, A., Thevenot, J., Rahtu, E., Lehenkari, P., Saarakkala, S. (2023), ”Automatic Knee Os- teoarthritis Diagnosis from Plain Radiographs:A Deep Learning-based Approach,” Scientific Re- ports, 8(1), 1727 [14] Zeng,G.,Li,Y.,Huang,W.,Wei,H.,Cheng,L.(2023),”ADeepLearningFrameworkforIdentifying Osteoarthritis from Radiographs,” Computers in Biology and Medicine, 111, 103558. [15] Shamir,L.,Wolkowicz,R.,Rahimi,S.,Ferrucci,L.(2024),”AutomaticPredictionofKneeOs- teoarthritis Progression using Radiographic Data,” PLOS ONE, 14(4):e0213791. [16] Teramoto,A.,Fujita,H.,Yamashita,Y.(2024),”KneeOsteoarthritisSeverityClassificationUsingaDeepConvolutionalNeuralNetworkwithaComparisonofTwoDatasets,”BMCMedicalImaging, 20(1), 67. [17] Choi,J.A.,Byun,S.H.,Park,Y.,Kim,D.,Shin,Y.(2023),”ArtificialIntelligencefortheDetection of Knee Osteoarthritis Using Digital Radiographs,” Clinical Radiology, 75(9), 692-699. [18] Murphy, L., Cibere, J., Roemer, F. W. (2023), ”AI-Powered Assessment of Knee Osteoarthritis Severity in Population Studies,” Osteoarthritis and Cartilage Open, 3(2), 100148. [19] Liu,F.,Dong,J.,Liu,L.,Cao,L.,Xu,H.(2024),”DeepTransferLearningforKneeOsteoarthritis Classification Based on X-ray Images,” Journal of Healthcare Engineering, 2022, 9842914. [20] Hong, Y., Li, S., Lu, M., Qian, Y. (2023), ”An Explainable Deep Learning Framework for Knee Osteoarthritis Severity Assessment,” IEEE Journal of Biomedical and Health Informatics, 27(4), 1124-1132.

Copyright

Copyright © 2025 Dhanu Hasi Prem, David Joseph, Joby George, Rotney Roy Meckamalil. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.