• ABSTRACT
    • Numerous studies have presented fully automated techniques for assessing structural osteoarthritis (OA) progression, with recent work increasingly relying on deep learning (DL)-based methods. The objective of this narrative review was to summarize findings from studies comparing the validity of fully automated methods for assessing progression in (peri-) articular joint tissues with reference measures (e.g., manual segmentation) in clinical OA models. A literature search in PubMed and arXiv.org identified 873 studies. Of these, nine evaluated the clinical validity of fully automated longitudinal measures for assessing progression. Five met the inclusion criteria by comparing sensitivity to differences in change in clinically defined cohorts between fully automated vs. reference assessments, and four reported at least the sensitivity to change for both methods. One of the studies evaluated longitudinal change in radiographic joint space width, five change in MRI-based cartilage thickness, two change in cartilage composition, and one change in thigh muscle and adipose tissue cross-sectional areas. Most of the studies were based on DL methods and relied on data from the Osteoarthritis Initiative (OAI). The included studies reported similar or greater sensitivity to change and similar discriminative power for detecting differences in change between clinically defined groups compared with reference measurements. Therefore, the techniques validated in these studies appear suitable for assessing structural progression provided that key requirements are met, including consistent imaging protocols, scanner settings, and data quality.