Diagnostic value of a coronal STIR sequence in conjoined lumbar nerve root detection: an MRI accuracy study.

Abstract

To assess diagnostic accuracy for conjoined lumbosacral nerve root (CLNR) detection on MRI when adding a coronal STIR sequence to the standard lumbar spine protocol. In this retrospective study, two radiologists assessed the presence of CLNR and lumbosacral transitional vertebrae (LSTV), using a standard lumbar MRI protocol and an expanded protocol with an additional coronal STIR sequence. Prior radiologist consensus using the expanded protocol served as a reference standard for diagnosis. In 751 patients (mean age 61.2 ± 15.7 years, 435 females), CLNR was found in 38 patients (5.1%) in consensus. Without coronal STIR, 13 CLNR patients were correctly identified, CLNR was missed in 25 patients, and 3 patients were falsely detected as having one (sensitivity of 34.2%, specificity of 99.6%, positive predictive value (PPV) of 81.3%, negative predictive value (NPV) of 96.6%, and accuracy of 96.3%). With coronal STIR, 31 CLNR patients were correctly identified, CLNR were missed in 7 patients, and one patient was falsely detected as having one (sensitivity of 81.6%, specificity of 99.9%, PPV of 96.9%, NPV of 99.0%, and accuracy of 98.9%). Inter-reader agreement improved from moderate without coronal STIR (κ = 0.592; 95% CI 0.38, 0.80) to almost perfect with coronal STIR (κ = 0.915; 95% CI 0.84, 0.99). LSTV had a prevalence of 13.3% among patients without and 26.3% among patients with CLNR (p = 0.025). Coronal STIR greatly improved sensitivity and inter-reader agreement for CLNR detection on MRI while slightly improving the specificity and accuracy. A significant association of CLNR and LSTV was found.