The accuracy of computed tomography in assessing cervical pedicle screw placement

A blinded, prospective comparison of computed tomography scan accuracy for determining the location of cervical pedicle screw position in human cadavers.

To establish recommended computed tomography technique guidelines for assessing location of cervical pedicle screws.

A small number of studies have described the accuracy of roentgenography regarding the assessment of pedicle screw position. However, a few studies have investigated the accuracy of computed tomography in this respect. Ebraheim et al evaluated the relation of lateral mass screws to the nerve roots within the intervertebral foramen on oblique radiographs. No study has been undertaken, to our knowledge, to specifically define the reliability and validity of computed tomography scans in the case of cervical pedicle screw placement.

As a pilot study, 10 cadaveric cervical spines from another study with bilateral 3.5 mm titanium pedicle screws were scanned with 1.0 mm axial slices. After the scans were interpreted by three blinded readers, each panel member was "trained" with regard to individual accuracy. Ten more cadaveric cervical spines were instrumented with 3.5 mm titanium screws in each pedicle (C2-C7). The specimens were then scanned with a variety of computed tomography techniques, including spiral acquisitions at 1.0 mm, 1.0 mm + reconstruction, 2.5 mm, 5.0 mm slices, and the three-dimensional Stealth Station recipes. The specimens were dissected, and malpositioned screws were recorded and photographed by independent raters. The same three readers from the pilot study then read each new scan in random order.

Reader accuracies in the pretraining pilot study were 74%, 68%, and 52%, with kappa coefficients of 0.49, 0.37, and 0.07, respectively, and significant intrarater variances (P = 0.014). After training, the accuracy rate improved significantly to 89%, 88%, and 85% in posttraining study, and the kappa coefficients were 0.81, 0.78, and 0.73, respectively. Kappa statistical analysis showed negligible interreader variance on the entire pivotal study except by the three-dimensional Stealth Station format. The overall mean kappa coefficients were 0.77, 0.75, and 0.73. Assessment of pedicle screw position was statistically inferior with 5.0 mm axial slices, in contrast to slices <3.0 mm.

We demonstrated that reliance on computed tomography scan data in determining the misplacement of a pedicle screw is usually accurate given proper scan acquisition, presentation windows, and adequate reader training, but a clinically significant error rate remains. A conventional computed tomography scan should not be treated as a gold standard, particularly without regard to the readers' training.