Diagnostic test accuracy, inter- and intrarater reliability of a knee osteoarthritis machine learning decision-aid algorithm in a near-clinical setting

Abstract

Purpose: Overall increase in medical imaging, scarcity of radiologist and concerns on interrater variability have sparked efforts to develop time saving and robust auto-reporting. Plain radiography remains mainstay in the diagnostic workup of knee osteoarthritis (KOA). We used a machine learning (ML)-tool (RBknee™, Radiobotics ApS) which is a CE-marked deep learning-based decision-aid software for automated interpretation of radiographic KOA including Kellgren-Lawrence (KL) grading on images from the daily clinical routline.