A machine learning pipeline for predicting joint space narrowing in knee osteoarthritis patients

Abstract

Osteoarthritis is the common form of arthritis in the knee (KOA). It is identified as one of the main causes of pain leading even to disability. To exploit the continuous increase in medical data concerning KOA, various studies employ big data and Artificial Intelligence analytics for KOA prognosis or treatment. However, most of the studies are limited to either specific groups of patients or specific groups of features, such as MRI, X-ray images or questionnaires. In this study, a machine learning pipeline is proposed to predict knee joint space narrowing (JSN) in KOA patients. The proposed methodology, that is based on multidisciplinary data from the osteoarthritis initiative (OAI) database, employs: (i) a clustering process to identify groups of people with progressing and non-progressing JSN; (ii) a robust feature selection process consisting of filter, wrapper and embedded techniques that identifies the most informative risk factors that contribute to JSN prediction; and (iii) a decision making process based on the evaluation and comparison of various classification algorithms towards the selection and development of the final prediction model for JSN. The evaluation was conducted with respect to model's overall performance, robustness and highest achieved accuracy. A 78.3% and 77.7% accuracy were achieved in left and right leg by Logistic Regression on the group of the 164 risk factors and SVM on the group of the 88 and 90 risk factors, respectively. © 2020 IEEE.