머신러닝 기반 무릎 골 관절염 심각도 분류를 위한 특징 중요도 분석

Abstract

In this study, the OsteoArthritis Initiative(OAI) dataset was used to evaluate four statistical methods to determine the importance of 6 clinical information and 18 X-ray image features for knee osteoarthritis severity classification. The Kellgren-Lawrence Grade, knee osteoarthritis seveirity classification scoring system was classified using a random forest method which used features of high importance. Joint space narrowing, osteophytes, and osteoosclerosis were ranked in the top five most important features in all methods. The model using 24 features had a balanced classification accuracy of 93.7%, while the model using 12 features including clinical information, joint space narrowing, and osteophytes, had a balanced classification accuracy of 93.4%. Both models had similar levels of accuracy, but calculation cost to obtain image features of the second model was lower than the first model. It used one-thirds of features of the first model and required the use of only 6 image features compared to the first methods required use of 18 image features.