Comparative analysis of machine learning and deep learning algorithms for knee arthritis detection using YOLOv8...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Knee arthritis is a widespread condition impacting global health.
• Early detection and treatment are crucial for managing knee arthritis progression and improving patient outcomes.
• Advanced computational methods are being explored for more accurate diagnostic tools.

Purpose of the Study:

• To evaluate and compare the performance of various machine learning (ML), deep learning (DL), and YOLOv8 classification models in detecting knee arthritis.
• To identify the most effective algorithm for accurate knee arthritis classification from medical images.
• To assess the utility of YOLOv8 models in a clinical context for knee arthritis diagnosis.

Main Methods:

• Utilized the "Annotated Dataset for Knee Arthritis Detection" comprising 1650 images across five classes (Normal, Doubtful, Mild, Moderate, Severe).
• Employed traditional ML models (k-NN, SVM, GBM), DL models (DenseNet, EfficientNet, InceptionV3), and YOLOv8 classification models (YOLOv8n-cls to YOLOv8x-cls).
• Data was split using the Hold-Out method (80% training, 10% validation, 10% testing).

Main Results:

• YOLOv8 models significantly outperformed ML and DL algorithms in knee arthritis detection.
• ML models achieved accuracies ranging from 63.61% (k-NN) to 67.36% (GBM).
• DL models showed varying success, with InceptionV3 reaching 79.41%, while YOLOv8x-cls achieved the highest accuracy at 86.96%.

Conclusions:

• YOLOv8 classification models, particularly YOLOv8x-cls, demonstrate superior performance for knee arthritis detection.
• These findings suggest YOLOv8 models offer a promising, highly accurate approach for automated knee arthritis diagnosis.
• The study highlights the potential of advanced AI in enhancing early detection and management of knee arthritis.