Automatic knee osteoarthritis severity assessment from plain radiographs with limited data

Abstract

Purpose: Recent studies have proved the potential of Deep Learning (DL) in automatic assessment of knee Osteoarthritis (OA) severity from plain radiographs. However, current state-of-the-art methods require using large amounts of annotated data that are typically available only in the research setting. Large datasets are costly to collect in real-life industrial or clinical applications. Developing methods that would allow to overcome the need of large annotated cohorts is crucial for bringing DL-based automatic severity assessment to clinical practice.