Arthroscopy-validated diagnostic performance of sub-5-min deep learning super-resolution 3T knee MRI in children and adolescents

Authors:
Publisher website
Google Scholar
Abstract
Objective This study aims to determine the diagnostic performance of sub-5-min combined sixfold parallel imaging (PIx3)–simultaneous multislice (SMSx2)–accelerated deep learning (DL) super-resolution 3T knee MRI in children and adolescents. Materials and methods Children with painful knee conditions who underwent PIx3–SMSx2–accelerated DL super-resolution 3T knee MRI and arthroscopy between October 2022 and December 2023 were retrospectively included. Nine fellowship-trained musculoskeletal radiologists independently scored the MRI studies for image quality and the presence of artifacts (Likert scales, range: 1 = very bad/severe, 5 = very good/absent), as well as structural abnormalities. Interreader agreements and diagnostic performance testing was performed. Results Forty-four children (mean age: 15 ± 2 years; range: 9–17 years; 24 boys) who underwent knee MRI and arthroscopic surgery within 22 days (range, 2–133) were evaluated. Overall image quality was very good (median rating: 5 [IQR: 4–5]). Motion artifacts (5 [5–5]) and image noise (5 [4–5]) were absent. Arthroscopy-verified abnormalities were detected with good or better interreader agreement (κ ≥ 0.74). Sensitivity, specificity, accuracy, and AUC values were 100%, 84%, 93%, and 0.92, respectively, for anterior cruciate ligament tears; 71%, 97%, 93%, and 0.84 for medial meniscus tears; 65%, 100%, 86%, and 0.82 for lateral meniscus tears; 100%, 100%, 100%, and 1.00 for discoid lateral menisci; 100%, 95%, 96%, and 0.98 for medial patellofemoral ligament tears; and 55%, 100%, 98%, and 0.77 for articular cartilage defects. Conclusion Clinical sub-5-min PIx3–SMSx2–accelerated DL super-resolution 3T knee MRI provides excellent image quality and high diagnostic performance for diagnosing internal derangement in children and adolescents.

This publication has 42 references indexed in Scilit: