“Never trust a smiling cat!”^[*]—Accuracy and IGS in Skull Base Surgery

With considerable interest I read McLaughlin's[1] article addressing an important and critical issue in skull base surgery. Submillimetric accuracy in image-guided systems (IGS) for surgery of the paranasal sinuses is critical. A magnetic resonance imaging– (MRI-) based solution avoiding radiation appears tempting.

The authors state impressive submillimetric values for surface registration error (SRE) and target registration error (TRE). Interestingly, the preoperatively acquired images were an unspecified MRI of the sellar region and a contrast-enhanced T1-weighted MRI with a slice thickness of 2 mm (“thin slice”), which was apparently used for registration and measuring purposes according to the images provided in the article. A supporting computed tomography (CT) scan was not mentioned.

The in-plane dimensions of the MRI is not given, but judged from the coarseness of the sagittal and coronal images presented, I assume that the voxels were isometric. If they were, which would be preferable when three-dimensional reconstructions or image guidance is planned and therefore assumed for the argumentation here, the underlying voxels would have a defined edge length of 2 mm in each direction. The geometrical grid is therefore a matrix of voxels with 2 × 2 × 2 mm at best. Moreover, because “slices” are mentioned, there is probably a gap of usually 10% between the individual slices (for technical reasons in the MRI machine itself) to avoid interferences between the slices. Image processing algorithms may smooth the images visually, and interpolation can ameliorate the gap problem. Unfortunately, these techniques do not increase the accuracy because the underlying matrix is unchanged.

The SRE is a technical indicator stating the placement accuracy of the wand in respect to the predefined registration points of the underlying T1WI images; it does not show the system's accuracy in respect to the task at hand. The authors state that the registration was accepted, when “… the distance was less than 1 mm through the verification process,” indicating that one would continue the registration process until the verification process would turn out the awaited result. This is a user-dependent, error-prone process because “… care must be taken … without any pressure… erroneous interpretation of the accuracy,” as stated by the authors. The SRE therefore only states that the tedious registration process was performed meticulously; it does not indicate whether the IGS is accurate regarding the navigational task.

Using the point-to-point measuring feature of the IGS to assess TRE is methodically critical because a measurement in one coordinate space (the coordinate space used for tracking the wand) is transformed into another, much less accurate coordinate space (the coordinate space of the underlying T1-weighted images). The measurement will be quite precise in the coordinate space used for tracking by the IGS (as expressed by the two decimal digits); the underlying image matrix still remains 2 × 2 × 2 mm at best. Unfortunately, the second measured point—the MRI representation of the anatomical structure in question—is not precisely defined because it lies at an undefined point within an uncertainty space of 2 × 2 × 2 mm. Therefore, the measurement cannot be more accurate than the underlying matrix grid; thus, a claimed submillimetric accuracy in the measurement has to be corrected with an uncertainty of at least 2 mm in each direction. Thus, this measurement is unusable for the task at hand.

The main issue of the article to prove the submillimetric accuracy of the used IGS has unfortunately failed for technical reasons and should—with all due respect—be corrected.

I would like to recommend the use of a referencing tool in trying to check the accuracy of the measurement within the IGS's coordinate space. Additionally, CT of the skull base as the referencing image base and then a subsequent merge with MRI to base the measurements on another base matrix (i.e., 0.5 × 0.5 × 0.5 mm as derived from CT) would solve the underlying problem for the price of radiation exposure.

^* Davis J. Garfield, Paws Inc. 2006

• Reference

• 1 McLaughlin N, Carrau RL, Kassam AB, Kelly DF. Neuronavigation in endonasal pituitary and skull base surgery using an autoregistration mask without head fixation: an assessment of accuracy and practicality. J Neurol Surg A Cent Eur Neurosurg 2012; 73 (6) 351-357