Effectiveness of Automatic Detection of Osteoarthritis using Convolutional Neural Network (CNN) Method with DenseNet201 on Digital Images of Knee Joint Radiography

Abstract

The manual detection of osteoarthritis using Kellgren Lawrence system depends on experience and agreement between doctors. The study was conducted to develop DenseNet201 to assist doctors in making a diagnosis of osteoarthritis grading. This study analyzes the accuracy; sensitivity; specificity; positive predictive value (PPV) and negative predictive value (NPV) of DenseNet201 in grading osteoarthritis and compares the classification results between DenseNet201 and radiologists in detecting osteoarthritis on knee joint images. This study is an applied experiment that compares the classification results of DenseNet201 and radiology specialists. Firstly, DenseNet201 is built with the MATLAB R2021a. Tests are carried out by measuring accuracy, sensitivity, specificity, PPV and NPV of 75 images of knee joint. Lastly, the data is analyzed using the Wilcoxon statistical test. The study has shown that the performance of DenseNet201 was good in detecting osteoarthritis, with accuracy value 91.84%; sensitivity value 76.61%; specificity value 94.32%; PPV 82.60% and NPV 94.32%. There was no significant difference between classification results using DenseNet201 and radiologist with a value (p>0.05) of 0.119. DenseNet201 can be considered as an alternative diagnostic tool for osteoarthritis with the condition that verification of the diagnostic decision still refers to the confirmation and justification of the radiologist.