Automated knee osteoarthritis severity classification using three-stage preprocessing method and VGG16 architecture

As a nutshell, an early diagnosis of Knee Osteoarthritis (KOA) enhances the likelihood of an individual receiving treatment at its onset and prevents joint replacement surgery. Computer-Aided-Diagnosis (CAD) from the radiograph images have gained a wide-spread attention in automated grading of KOA severity. But radiograph images have certain limitations, such as presence of too much noise and uneven contrast distribution across the image that impacts the accuracy and reliability of CAD systems. Therefore, a novel three-stage pre-processing method has been proposed using a combination of different techniques applied at sequential stages such as noise-reduction using gaussian-filter, normalization using pixel-centering method, and balanced contrast enhancement technique. A transfer-learning based VGG16 architecture has been used for severity classification. Our classification framework outperforms the existing state-of-the-art methods achieving an excellent accuracy of 89.95\%. Therefore, our system can be used by the radiologists and physicians as a decision-support tool in their daily diagnosis procedure.