3.0 vs 1.5 T MRI in the detection of focal cartilage pathology – ROC analysis in an experimental model

Abstract

To use receiver operator characteristics (ROC) analysis for assessing the diagnostic performance of three cartilage-specific MR sequences at 1.5 and 3 T in detecting cartilage lesions created in porcine knees. Eighty-four cartilage lesions were created in 27 porcine knee specimens at the patella, the medial and lateral femoral and the medial and lateral tibial cartilage. MR imaging was performed using a fat saturated spoiled gradient echo (SPGR) sequence (in plane spatial resolution/slice thickness: 0.20 x 0.39 mm2/1.5 mm) and two fat saturated proton density weighted (PDw) sequences (low spatial resolution: 0.31 x 0.47 mm2/3 mm and high spatial resolution: 0.20 x 0.26 mm2/2 mm). The images were independently analyzed by three radiologists concerning the absence or presence of lesions using a five-level confidence scale. Significances of the differences for the individual sequences were calculated based on comparisons of areas under ROC curves (A(Z)). The highest A(Z)-values for all three radiologists were consistently obtained for the SPGR (A(Z) = 0.84) and the high-resolution (hr) PDw (A(Z) = 0.79) sequences at 3T. The corresponding A(Z)-values at 1.5 T were 0.77 and 0.69; the differences between 1.5 and 3 T were statistically significant (P < 0.05). A(Z)-values for the low-resolution PDw sequence were lower: 0.59 at 3 T and 0.55 at 1.5 T and the differences between 1.5 and 3T were not significant. With optimized hr MR sequences diagnostic performance in detecting cartilage lesions was improved at 3 T. For a standard, lower spatial resolution PDw sequence no significant differences, however, were found.