Current accuracy of surface matching compared to adhesive markers in patient-to-image registration

Since its introduction in the 1980s, frameless stereotaxy, or neuronavigation, has become increasingly popular and has now become an integral part of intracranial surgery. In principle, neuronavigation determines and demonstrates the relationship between preoperative imaging data, typically computed tomographic (CT) and/or magnetic resonance (MR) images, and the intraoperative position of a surgical instrument. The neuronavigation computer finds this relationship with a patient-to-image registration procedure based on a rigid body transformation. A rigid body transformation finds the best “fit” between two coordinate systems, in this case, image coordinates and operating room (OR) coordinates, without distorting either coordinate system. Typically, this procedure uses either point-pair matching or surface matching to find this “fit.”

Although anatomical landmarks may be used to perform point-pair matching, they are relatively difficult to digitize accurately, not abundantly available on a human head, and those that are available are not widely distributed (ears, eyes, and nose). To overcome these limitations, adhesive markers can be applied to the scalp, prior to image acquisition, which remain in place until they are digitized in OR space. As a consequence, their use requires additional preoperative image acquisition. To circumvent the costs of adhesive markers and the necessity for additional imaging, without reverting to anatomical landmarks, one may also use surface matching. In surface matching, a cloud of randomly digitized points on the scalp is fitted to a volume rendering of the imaging data.

The accuracy with which the image and OR coordinate systems are coregistered impacts directly on the application accuracy during surgery. Unfortunately, in previous comparisons of patient-to-image registration techniques, surface matching was found to be significantly less accurate than point-pair matching with either anatomical landmarks or adhesive markers [3, 13, 14]. Despite these reports, surface matching is still available on modern neuronavigation systems and appears to gain popularity. Partly, this may be due to its advantages mentioned above, i.e., ease of use when compared to anatomical landmarks and cost-effectiveness when compared to adhesive markers, but unawareness about the inaccuracy of surface matching may also contribute. Furthermore, a belief may settle in, that software advances have led to improved surface matching algorithms, which no longer suffer the inaccuracies reported earlier.

We aimed to determine the current application accuracy of neuronavigation, using patient-to-image registration based on either surface matching or point-pair matching with adhesive markers, using three different commercially available navigation systems.

Eleven patients were included as detailed above, between November 2017 and July 2018. Most of these had either a frontal (5 patients) or temporal (3 patients) location for the target marker. There were two occipital locations and one parietal location.

TRE measurements are offered in Fig. 2. In all cases, surface matching was significantly less accurate than point-based registration (p < 0.001). On average (±SD), TRE after point-based registration was 2.49 ± 0.86 mm and TRE after surface matching was 5.35 ± 1.64 mm. The mean difference in accuracy was 2.85 (± 1.20) mm. There was no obvious difference between navigation systems, regarding these results (Fig. 3). Moreover, in all cases with the Curve, the second surface matching attempt worsened accuracy, rather than improved it. This further degradation was, on average, 1.00 ± 0.68 mm (p = 0.06).

For the StealthStation cases, Fig. 4 demonstrates lack of correlation between ERA values and TRE values (R² = 0.04, p = 0.67).

In this paper, we present a clinical evaluation of the application accuracy of current neuronavigation systems, comparing the use of adhesive markers with that of surface matching. Our results clearly demonstrate that the development of neuronavigation systems over the last decade has not led to increased accuracy of either technique. More importantly, surface matching is still significantly inferior to point-pair matching with regard to accuracy, regardless of which navigation system is used. Furthermore, we demonstrate that the estimation of accuracy offered by the navigation system, following patient-to-image registration (ERA), shows no correlation with true application accuracy (TRE).

The current results are similar to those from our earlier publication (2.49 ± 1.07 mm for fiducial markers and 5.03 ± 2.30 mm for surface matching) [14]. Although, since then, a number of publications have reported relatively high accuracy of surface matching systems [1, 5, 8, 9, 11]; most of these do not compare to the use of fiducial markers, and none of these report on actual measurements in a true neurosurgical setting. Instead, they use cadaver measurements, phantom measurements, calculated estimations of accuracy or measurements in non-surgical cases such as Gamma Knife treatment. Such calculations or measurements should be interpreted with caution as inaccuracy may easily be underestimated. In surgery, correct surface digitization suffers from lack of line-of-sight for the navigation system, the presence of hair and facial distortion due to loss of muscle tone, intubation, nasogastric tube placement, patient positioning, and skin movement due to use of the navigation pointer.

Part of the confusion regarding the application accuracy of neuronavigation may be due to the fact that it is not a trivial task to assess accuracy in an individual case. Some systems show a numerical value (ERA) following point-pair matching, but our current analysis demonstrates no correlation between this ERA and actual target registration accuracy, as has been demonstrated earlier for both the ERA [1, 7, 10] and root mean squared error (RMSE) [2, 4, 14]. Following surface matching, some systems show a “distance to skin surface” when the pointer is placed on the scalp. This is also of little value, as this distance only shows the error in one direction. It may be worse in another. The same is true when the surgeon runs the pointer across the scalp to check, visually, whether the navigation system indeed shows proximity to the scalp surface on the MRI images. The only sensible assessment consists of pointing at multiple targets, either adhesive markers or anatomical landmarks, and to evaluate whether the distance to these targets in the MRI images is acceptable for the type of surgery at hand. And then, still, one is not sure whether this is representative for the accuracy at deeper structures, as accuracy is not homogenous throughout the operative volume.

Nevertheless, surface matching has such strong appeal that even procedures requiring the highest possible accuracy, such as needle biopsies, are being performed with this registration technique [12], and some even regard the use of adhesive markers unacceptable [6], as it requires additional preoperative imaging. Based on our results, such claims are still not warranted.

Our study has a number of limitations. The most important one is the fact that the accuracy measured at a single adhesive marker does not represent the accuracy of the entire surgical volume. At another point in the patient’s head, the accuracy may be different; better or worse. Furthermore, the use of an adhesive marker as a target introduces inaccuracies. The image coordinates may be inaccurate because of the size of the marker, and the real-life coordinates may be inaccurate since the skin is slightly mobile. The marker is merely used because it allows a pairwise comparison between registration strategies. Finally, we omitted a comparison with point-pair matching based on anatomical landmarks. Our previous publication demonstrated this to yield accuracy comparable to that of surface matching (4.97 ± 2.29 mm) [14]. In the light of current results, this is also likely unchanged. Anatomical landmarks could, therefore, provide an adequate back-up option if adhesive markers are unavailable and surface matching fails.

Not every neurosurgical procedure requires the same level of accuracy from the neuronavigation system. Delineating the location and extent of a craniotomy for the removal of a convexity meningioma, for instance, will tolerate larger inaccuracies than a needle biopsy of a small deep-seated lesion. Thus, the level of accuracy offered by surface matching may be tolerable for one procedure but not for another. It is important that the surgeon is aware of the difference in accuracy between different patient-to-image registration accuracies and makes an informed decision to use one or the other, depending on the type of surgery at hand.

The use of surface matching with current neuronavigation systems is still consistently associated with application inaccuracies that are twice as large as those offered by the use of adhesive markers. Furthermore, these inaccuracies are not reliably reflected by the accuracy estimation provided by the neuronavigation system. Although the lower accuracy offered by surface matching may be sufficient in selected cases, it is important for a neurosurgeon to make an informed choice between these patient-to-image registration strategies, depending on the type of surgery at hand.