Detection of Knee Osteoarthritis Severity using a Fusion of Machine and Deep Learning models

Knee Osteoarthritis accounts for more than 80% cases of arthritis impacting life quality of individuals. It is an irreversible disease that the only cure is the replacement of the knee, being important to diagnose it at early stages to prevent its progression. This study aims to improve the detection of Knee Osteoarthritis at all stages based on Kellgren-Lawrence scale using machine learning models such as Random Forest, Gradient Boosting and Xtreme Gradient Boosting trained with patient’s information and deep learning models including DenseNet201 and InceptionResNetV2 trained with knee x-ray images., Their individual predictive capabilities are combined using late fusion strategy to select the final class. Machine learning models showed similar overall prediction performance between them although Deep learning models had higher efficiency showed in ROC curves compared to them, however, both altogether achieved better performance evaluated through Precision, Recall and F1-score. Moreover, using patient’s data in machine learning models were identified the main features that influence the disease