[Knee Osteoarthritis Detection using CNN]

[Ramachandra Udupa]
[USN: 4MW21AD043, 7th Sem Dept. Of Artificial
Intelligence and Data Science, SMVITM]

Co-Authors: [Shravya S Devadiga, Anusha R, Omkar Shettigar]

[7th Sem, Dept. Of Artificial Intelligence and Data Science, SMVITM]

12/20/2024 CISC-2024
 Contents

Introduction
Motivation
Objective and Contributions
Literature survey
Block diagram/System designed and Working
Results and Discussion
Conclusion
Future Scope
References

12/20/2024 CISC-2024
 Introduction
• Knee osteoarthritis (KOA) is a common joint condition where the cartilage
in the knee gradually wears down.
• Cartilage is the smooth tissue that covers the ends of bones and allows
them to move smoothly.
• When it wears out, bones begin to rub against each other, causing pain,
swelling, stiffness, and difficulty moving the knee.
• KOA mostly affects older adults, but it can happen to anyone, especially
those with previous knee injuries, excess weight, or a family history of the
condition.
• There is no complete cure, but treatments like physical therapy,
medication, or surgery can help manage symptoms and improve knee
function.
 Motivation
Diagnosing knee osteoarthritis (KOA) using X-rays and MRIs is
time-consuming and can vary between doctors. Our machine learning
model automates this process, providing quick, accurate, and
consistent results, reducing the burden on doctors and improving
patient care.
 Objective and Contribution
Objectives:
1. To design an automated knee osteoarthritis detection system using CNNs for improved
diagnostic accuracy.
2. To evaluate CNN Model Performance on the Dataset
3. To Integrate a chatbot and create a scalable, cost-effective solution for early OA detection.
Contribution:
• Custom CNN: Achieved 86% accuracy with balanced precision, recall, and F1-score.
• Pre-Trained Models: Utilized ResNet101, MobileNetV2, and AlexNet for enhanced detection
through transfer learning.
• Advanced Data Processing: Addressed challenges like class imbalance and complex X-ray
images with robust preprocessing and augmentation.
• Comprehensive Evaluation: Validated effectiveness using precision, recall, and F1-score
metrics.
• AI Impact: Demonstrated AI’s role in improving diagnostics and streamlining healthcare
workflows.
 Literature Review
Author(s) Year Title Approach Key Findings

Chen et al. 2019 Fine-Tuned Versions of ResNet and Fine-tuning of ResNet and Enhanced model accuracy for classification
InceptionV3 InceptionV3 models tasks

Lim et al. 2019 Deep Neural Network with Scaled Applied PCA with deep Improved model performance by reducing
Principal Component Analysis neural networks dimensionality

Turkiewicz et al. 2014 Current and Future Impact of Population-based Predicted increase in osteoarthritis cases
Osteoarthritis on Health Care: A projection study by 2032
Population-Based Study

Saleem et al. 2020 X-Ray Image Analysis for Automated Automated analysis of Achieved automated detection accuracy
Knee Osteoarthritis Detection X-ray images for KOA diagnosis

Kurmis et al. 2023 A Role for Artificial Intelligence Review of AI in total knee Discussed AI’s role in surgical and
Applications Inside and Outside the arthroplasty non-surgical KOA applications
Operating Theatre
Author(s) Year Title Approach Key Findings

Ahmed, N., Saeed, M., Aftab, 2023 Knee Osteoarthritis Transfer learning on Enhanced classification
M., Mehmood, A., & Ilyas, Q. Detection and Classification pre-trained models for OA accuracy for OA, providing
M. Using Transfer Learning detection efficient stage differentiation
for clinical use.

B. C. Dharmani and K. Khatri 2023 Deep Learning for Knee Transfer learning with Improved severity stage
Osteoarthritis Severity Stage EfficientNetB1 on X-rays detection of OA, leveraging
Detection using X-Ray EfficientNetB1 for accurate
Images classification.

7
Block Diagram
 System Design
1. Data Acquisition: We utilized the Osteoarthritis Initiative dataset, which contains essential
knee joint images for diagnosing knee osteoarthritis (KOA).

2. Data Preprocessing: Segmentation and equalization techniques were applied to improve

image clarity, specifically highlighting the knee joint areas critical for diagnosis.

3. Model Training: The dataset was split into training, testing, and validation sets with a 7:2:1
ratio. We experimented with 4 deep learning models, including pre trained CNNs like
ResNet101, MobileNetV2, AlexNet and a Custom CNN.

4. Model Selection: Each model was evaluated based on accuracy, efficiency, inference speed,
and memory footprint, with the best performer selected for reliable KOA classification.
Working Principle
Results and Discussion

Fig: Confusion matrix of ResNet101 & MobileNetV2

Results and Discussion

Fig: Confusion matrix of AlexNet & Custom CNN

Results and Discussion
 Results and Discussion

1. ResNet101: Performed well for minority classes like Class 4 but struggled with
majority classes (0, 1, 2, 3), leading to misclassifications.
2. MobileNetV2: Excelled in Class 2 but struggled to distinguish between similar
classes, such as Class 1 and Class 2.
3. AlexNet: Struggled across all classes due to its simpler architecture, especially
failing to extract features for Classes 3 and 4.
4. Custom CNN: Performed well for minority classes (0, 1, 2) but sometimes
misclassified majority classes like Class 3 as Class 4, likely due to overfitting or
limited feature extraction.
 Conclusion
• Machine learning and deep learning were successfully applied to medical imaging
for automated detection and grading of knee osteoarthritis (OA).

• Pre-trained models like ResNet101, MobileNetV2, and AlexNet were used along
with a custom-designed CNN specific to the knee OA dataset.

• The custom CNN achieved the best performance with 86% accuracy, precision,
recall, and F1-score.

• The project highlights the effectiveness of combining data preprocessing,

augmentation, and tailored model design.

• The results validate the potential of AI to improve diagnostic workflows in medical

imaging
 Future Work
• Multimodal Imaging: Add MRI and CT scans to improve detection of
small changes in joint health.

• Model Interpretability: Use techniques like Grad-CAM or SHAP to help

clinicians visually understand AI predictions and build trust.

• Lightweight Architectures: Optimize models like MobileNetV2 for

real-time use on mobile and edge devices, especially in resource-limited
areas.

• Track OA Progression: Develop systems to analyze OA progression

over time for personalized care and better management.

• Clinical Integration: Link AI systems with electronic health records

(EHR) to streamline workflows and make diagnostics more accessible
and scalable in healthcare.
 References
• Mohammed, A. S., Hasanaath, A. A., Latif, G., & Bashar, A. (2023). Knee osteoarthritis detection and
severity classification using residual neural networks on preprocessed x-ray images. Diagnostics,
13(8), 1380.
• Chen, P., Gao, L., Shi, X., Allen, K., & Yang, L. (2019). Fully automatic knee osteoarthritis severity
grading using deep neural networks with a novel ordinal loss. Computerized Medical Imaging and
Graphics, 75, 84-92.
• Saleem, M., Farid, M. S., Saleem, S., & Khan, M. H. (2020). X-ray image analysis for automated knee
osteoarthritis detection. Signal, Image and Video Processing, 14(6), 1079-1087.
• Kurmis, A. P.: A role for artificial intelligence applications inside and outside of the operating theatre:
a review of contemporary use associated with total knee arthroplasty. In: Arthroplasty, vol. 5, no. 1,
pp. 40. Publisher, Location (2023). 17

• Tan, M., & Le, Q.: EfficientNet: Rethinking model scaling for convolutional neural networks. In:
Proceedings of the 36th International Conference on Machine Learning, vol. 97, pp. 6105–6114.
Publisher, Location (2019).