Classification of Knee X-rays That Can Be Diagnosed Radiographically Using Deep Learning and Machine Learning Methods

The aim of this study is to classify knee osteoarthritis, synovial chondromatosis, Osgood-Schlatter disease, os fabella pathologies that can be diagnosed with plain knee X-rays, and normal knee radiographs with deep learning and machine learning methods. This study was performed on 540 knee osteoarthritis, 151 Osgood_Schlatter disease, 191 knee chondromatosis, 152 os fabella and 523 normal knee X-ray images. First, classification was performed with the VGG-16 network, which is a pre-trained deep learning model. Then, the features extracted with the VGG-16 convolution layer were classified with random forest, support vector machines, logistic regression and decision tree machine learning algorithms. With VGG-16 model, 95.3% accuracy, 95.1% sensitivity, 98.7% specificity, 96.8% precision, and 95.9% F1 score results were obtained. In classifying the features extracted from the VGG- 16 convolution layer with machine learning algorithms, 98.2% accuracy, 99.0% sensitivity, 98.9% specificity, 98.2% precision and 98.5% F1 score results were obtained with the logistic regression classifier. In this study, which was conducted to classify radiographically detectable knee pathologies, successful results were obtained with the VGG-16 network. The features extracted from the convolution layer of the VGG-16 model were reclassified with machine learning algorithms, logistic regression, support vector machines and random forest classifiers, and improvements in performance metrics were obtained compared to the VGG-16 model. With this proposed method, the performance of deep learning models can be further improved.