Comparison of three methods of estimating confidence intervals for stereotactic error

When planning a stereotactic procedure, it is clinically valuable to know the spatial confidence intervals of a particular stereotactic technique. To do this, the aggregate error distribution of the stereotactic technique must first be estimated. In a frame-based stereotactic procedure, there is an imaging step and a treatment delivery step. If error is introduced independently at these steps, then the error of a stereotactic procedure may be computed from the error distributions of the component steps. Three computational methods of doing this were compared: parametric, convolution, and Monte Carlo. To test these methods, the error distributions of an imaging technique, a delivery technique, and the corresponding stereotactic imaging-plus-delivery system were measured empirically using a phantom, computed tomography, and the CRW stereotactic system. The three methods gave concordant estimates of mean aggregate error (respectively 2.71 +/- 1.52, 2.45 +/- 2.30, 2.51 +/- 2.34, and 2.47 +/- 2.31 mm for the empiric, convolution, Monte Carlo, and parametric methods). However, the estimates of the confidence intervals differed between the parametric and the nonparametric methods. In particular, the parametric method gave significantly higher estimates of the 99% spatial confidence interval (6.40 mm versus 5.41 mm and 5.38 mm for the convolution and Monte Carlo methods). Knowledge of the confidence intervals allows a neurosurgeon to determine a priori whether a particular stereotactic technique is likely to satisfy a clinically defined error budget, and thereby achieve clinical success.