Accuracy Analysis And Error Correction For Anatomical Landmarks Based Registration In Image-Guided Neurosurgery

Accurate patient to image registration is the core for successful and safe image-guided neuro-navigation. Point-matching is the most common technique in practice to achieve this registration. While Skin Adhesive Markers (SMs) are widely used in point-matching registration, a proper implementation of Anatomical Landmarks (ALs) may overcome the inconvenience brought by the use of SMs. However, the accuracy and applicability of ALs registration in neuro-navigation have remained a controversial issue due to the great variability of the reported results and conclusions. Therefore, we are providing an analysis for the accuracy when using different sets of ALs as well as the applicability of the corresponding surgical fields. In addition, one of the significant factors that influence the registration accuracy at the target point is the distribution of the fiducial points. The optimal distribution may be difficult to achieve either due to the limited number of distinct anatomical features on head surface or even due to the poor planning of skin adhesive markers. Therefore we have also tried to overcome this problem and correct the quality of the registration in the areas that normally suffer low accuracy when using ALs-based registration.In the accuracy analysis study, we propose a set of three configurations using nine ALs. These configurations are defined according to the required positioning of the patient’s head during surgery and the resulting distribution of the expected Target Registration Error (TRE). We first evaluated these configurations by simulation experiments using real clinical data of20patients from two hospitals, and then tested the applicability of them in eight real clinical surgeries of neuronavigation.In the correction study, the proposed method is based on an initial point-matching registration using six distinct ALs as fiducial points, followed by selection of some surface points on the patient’s head at locations where natural ALs are not available to improve the distribution of the fiducial points. The projection of the surface points from patient space into image space reflects the errors introduced during the initial registration process. These errors can be identified and inspected in the image space by calculating the distance between the projected surface point positions transferred with the standard spatial relation and also by calculating their nearest point positions on the head surface. This information is then used to improve registration accuracy by adopting the calculated nearest points (the more accurate points) in the image space, instead of the projected positions, along with their corresponding actual surface points in the patient space, as additional pairs of registration points. The simulation experiment of the accuracy analysis showed that, by incorporating a Fiducial Registration Error (FRE) of3.5mm measured in the clinical setting, the expected TRE in the whole skull was less than2.5mm, and the expected TRE in the whole brain was less than1.75mm when using the configuration with all the nine ALs. A small TRE could also be achieved in the corresponding surgical field, by using the other three configurations with less ALs. In the clinical experiment, the Fiducial Localization Error (FLE) ranges in the image and the patient space were1.4-3.6mm and1.6-5.5mm, respectively. The measured TRE and FRE were3.1±0.75mm and3.5±0.17mm, respectively.For the correction trials, experiments with real clinical data showed that when using the surface points to correct the initial registration transformation, the TRE decreased in the whole-brain area by approximately20%, and this improvement is more dominant in the posterior and superior parts of the brain.The ALs configurations proposed in the accuracy analysis provide sufficient registration accuracy and can help to avoid SMs disadvantages if used clinically. The correction technique helps as well to overcome the naturally impaired distribution of the ALS, which is the most significant factor that prevents their wide use in registration. The method also allows more precise selection of corresponding fiducial points than traditional ALs and without the need for tagging adhesive markers. Results showed an improvement in registration quality in the targeted area in all cases by this kind of correction.