The technique of prospective stereotaxy [1] was developed for aiming a class of frameless stereotaxy trajectory guides. Localizing commercial devices in MR-guided neurosurgery with prospective stereotaxy is achieved through iteratively imaging and manipulating the device until it is deemed on target, as in the ClearPoint system (MRI Interventions, Irvine, CA). This process may take several minutes to complete. Compressed sensing improves imaging performance by exploiting sparse image representations, sparse sampling patterns, and novel reconstruction. We use these concepts to compute the orientation of a trajectory guide with a frame rate 25x faster than our previous real-time, Cartesian imaging based approach. We demonstrate our method’s utility in a survival experiment of gene delivery designed to alter behavioral expression of anxiety in 8 non-human primates over several months.

Brain cannulas are inserted with sub-mm accuracy through a trajectory guide fixed to the skull, bypassing the blood-brain-barrier. Preoperative MRI is used to determine a suitable mounting point for the base of the guide such that the target falls within the guide’s range of motion (a 30° cone, Fig. 1).

Intraoperative MRI is used during the procedure to visually identify the target and pivot. Our previous real-time approach [2] then imaged a plane above the brain and perpendicular to the planned trajectory. The operator viewed those cross-sectional images of the stem while moving the guide into alignment with the planned trajectory. Though much faster than iterative approaches, performance is hindered by 1) slow frame rates to generate high resolution images of the guide’s cross section and 2) difficulties visually selecting the true geometric pivot with sub-mm accuracy. Instead of imaging, our new approach acquires sets of 1D projections through the alignment stem at various angles and offsets, as shown in Fig. 2.

Each 1D projection is correlated with the known theoretical projection of the circularly symmetric alignment stem to find its center location along the projection. These computed locations can be configured into an overdetermined system of equations to find the orientation of the guide (blue lines in Fig. 2) and indicate it to the operator. Lines representing device orientation are recorded as the operator moves the guide to three different alignments. All trajectories are constrained to pass through the pivot point, so the 3D intersection of those lines is computed to yield improved pivot point coordinates. The operator then aligns the guide with the target by simply moving the computed point onto the green aim point (Fig. 3). If the trajectory guide is aimed based on an incorrect pivot, the tip of the device may not hit the target, as illustrated in Fig. 4.

We demonstrate the method using a Navigus trajectory guide (Medtronic, Minneapolis, MN) consisting of a pivoting ball joint within a base that can be fixed to the skull over a burr hole [3]. Viral vector containing a genetic payload was precisely delivered to the amygdala using 2 infusions, one anterior and one posterior, in a study of gene therapy to alter expression of anxiety. The technique has been used in 8 primate invivo surgeries to accurately treat two locations within the central amygdala, an obliquely oriented cylinder only 5 mm in length and 2 mm in diameter.

Each 1D projection was acquired with 0.4 mm resolution. Sampling 18 angles at 2 offsets was found to produce accurate, repeatable calculation of the device orientations. As each 1D MR experiment required only 5 ms, the device orientation could be calculated at 5 frames/s, a considerable improvement over our 0.2 frames/s performance using Cartesian imaging. The clear visualizaiton of Gd-doped infusate is shown in Fig. 5.

Over 8 subjects, in the N=22 instances of computing a pivot point, the total 3D separation between the manually selected and computed points was 1.62 ± 0.90 mm, and the radial (perpendicular to trajectory) separation was 0.67 ± 0.36 mm.

Our new approach eliminates error-prone manual selection of the pivot. Solving for location with an overdetermined set of equations allows for alignment with theoretical precision greater than the image resolution while reducing the chances for operator error to affect alignment.

We have implemented a system for rapidly and computationally identifying the true pivot point and orientation within a trajectory guide. This improves the targeting process by reducing errors that may arise from alignment stem tip uniformity and human interpretation of MR images. The 25x improvement in frame rate eases the aiming process and may be further increased as we experiment with using fewer projections.