Prospective, double-blind evaluation of knee MR imaging versus arthroscopy

The significance of subtle changes in signal intensity on MR images of the meniscus is not clear, as the authors have a high rate of false-negative diagnoses using previously reported criteria. Unselected patients scheduled for knee arthroscopy were imaged in sagittal and coronal planes (repetition time/echo time = 800-1,500/20, 70 msec) with 5-mm and 3-mm section thickness, respectively. A total of 140 menisci were evaluated. The images and interpretations were withheld from the arthroscopist until the arthroscopic diagnosis was recorded so that the MR imaging results did not influence which cases received arthroscopic confirmation. Before the arthroscope was withdrawn the MR images and interpretations were unsealed so that additional arthroscopic evaluation could be obtained of areas suspicious on MR imaging. Review of the first 60 menisci showed that significant tears of the meniscus were often represented by subtle signal intensity that did not extend to the surface of the meniscus. When previously reported criteria were used, approximately 50% of those menisci with increased internal signal not extending to the surface were interpreted as false-negative diagnoses. An initial classification system based on ten categories of signal morphology was developed which reduced the false-negative rate to 10% but led to a 30% rate of false-positive diagnoses in those cases with equivocal signal. Retrospective, blinded categorization of signal morphology was optimized based on ROC curves, and this is being tested prospectively in an ongoing investigation