Deep Ensemble Architecture for Knee Osteoarthritis Severity Prediction and Report Generation*

Abstract

Knee osteoarthritis is a condition in which the knee's articular cartilage, which is a slippery material that normally protects bones from joint friction, degenerates and changes to the underneath of the cartilage. If detected early the degeneration can be slowed down. The severity is relied on for detection on the expertise of Physicians. In this paper, to automatically measure OA severity we discuss the usage of deep CNN as a tool to successively develop a system, that is based on a grading system known as Kallgren-Lawrence (KL-grading). In this approach the OA severity is predicted using the radiographic Images. The method of automatic prediction of knee OA severity comprises three steps. a) Automatic localization of the knee joints. b) Classification of the localized knee joints and c) Create the report summary for identified symptoms The CNN is trained from scratch on the X-ray images. Along with the development of severity prediction through localization and classification, we will be developing the method to automatic report generation that consists of the description of the finding from the radiographs.