Distribution of interscan measurement error in AI-based 3D MRI analysis of knee cartilage thickness in osteoarthritis

• Published in: PloS one Vol. 20; no. 8; p. e0329610
• Main Authors: Katano, Hisako, Sasaki, Eiji, Nagai, Kanto, Hashiguchi, Naofumi, Kaneko, Haruka, Ishibashi, Yasuyuki, Kuroda, Ryosuke, Adachi, Nobuo, Ishijima, Muneaki, Tomita, Makoto, Masumoto, Jun, Sekiya, Ichiro
• Format: Journal Article
• Language: English
• Published: United States Public Library of Science 07.08.2025; Public Library of Science (PLoS)
• Subjects: Aged; Algorithms; Artificial Intelligence; Biology and Life Sciences; Cartilage; Cartilage diseases; Cartilage, Articular - diagnostic imaging; Cartilage, Articular - pathology; Computer and Information Sciences; Deep learning; Development and progression; Diagnosis; Diagnostic errors; Dimensional analysis; Engineering and Technology; Female; Forecasts and trends; Health aspects; Humans; Imaging, Three-Dimensional - methods; Knee; Knee joint; Knee Joint - diagnostic imaging; Knee Joint - pathology; Magnetic resonance imaging; Magnetic Resonance Imaging - methods; Male; Measurement; Medicine and Health Sciences; Middle Aged; Osteoarthritis; Osteoarthritis, Knee - diagnostic imaging; Osteoarthritis, Knee - pathology; Patients; Physical Sciences; Research and Analysis Methods; Social Sciences; Software; Thickness measurement; Japan; United States > US
• ISSN: 1932-6203; 1932-6203
• DOI: 10.1371/journal.pone.0329610

A novel AI-based 3D analysis system was developed to automatically extract bone and cartilage from MRI data and provide average cartilage thickness. This study aimed to analyze the interscan measurement error of knee cartilage thickness in osteoarthritis patients. Fifty knee osteoarthritis patients underwent two scans using MRI systems from five different vendors. Each model included five Kellgren-Lawrence grade (KL) 1-2 and five KL3-4 patients. Cartilage thickness was automatically quantified for seven regions, and interscan measurement error was analyzed. In the KL1-2 group, measurements with errors ≤0.05 mm, ≤ 0.10 mm, and ≤0.20 mm were 42%, 75%, and 97%, respectively. For the KL3-4 group, these proportions were 31%, 59%, and 90%. The entire cohort (KL1-4) showed errors ≤0.05 mm, ≤ 0.10 mm, and ≤0.20 mm in 37%, 67%, and 93% of measurements. Differences between KL1-2 and KL3-4 groups were significant for all thresholds. Overall, 93% of interscan measurement errors were within 0.20 mm when using fully automatic MRI 3D analysis software to assess knee cartilage thickness in osteoarthritis patients. This study provides valuable insights into the reliability of automated cartilage thickness measurements across different disease severities and MRI systems.