Incorporation of a laser range scanner into image-guided liver surgery: surface acquisition, registration, and tracking

Abstract

As image guided surgical procedures become increasingly diverse, there will be more scenarios where point-based fiducials cannot be accurately localized for registration and rigid body assumptions no longer hold. As a result, procedures will rely more frequently on anatomical surfaces for the basis of image alignment and will require intraoperative geometric data to measure and compensate for tissue deformation in the organ. In this paper we outline methods for which a laser range scanner may be used to accomplish these tasks intraoperatively. A laser range scanner based on the optical principle of triangulation acquires a dense set of three-dimensional point data in a very rapid, noncontact fashion. Phantom studies were performed to test the ability to link range scan data with traditional modes of image-guided surgery data through localization, registration, and tracking in physical space. The experiments demonstrate that the scanner is capable of localizing point-based fiducials to within 0.2 mm and capable of achieving point and surface based registrations with target registration error of less than 2.0 mm. Tracking points in physical space with the range scanning system yields an error of 1.4 +/- 0.8 mm. Surface deformation studies were performed with the range scanner in order to determine if this device was capable of acquiring enough information for compensation algorithms. In the surface deformation studies, the range scanner was able to detect changes in surface shape due to deformation comparable to those detected by tomographic image studies. Use of the range scanner has been approved for clinical trials, and an initial intraoperative range scan experiment is presented. In all of these studies, the primary source of error in range scan data is deterministically related to the position and orientation of the surface within the scanner's field of view. However, this systematic error can be corrected, allowing the range scanner to provide a rapid, robust method of acquiring anatomical surfaces intraoperatively.

Fig. 1

The RealScan 3-D Laser range scanner mounted to an OR mechanical arm.

Fig. 2

The imaging phantom consists of a model organ surface as well as white Teflon balls used as fiducials. To eliminate unwanted data, the base of the phantom has been painted black so there will not be enough signal for the range scanner to calculate depth at that point, making the surface and fiducials easier to identify.

Fig. 3

An example of the sphere-fitting algorithm. The black dots represent points from the range scanner and the gray sphere is the sphere fitted from this data. If a point is completely visible, then it lies outside the sphere surface. Points inside the sphere are partially or completely obscured.

Fig. 4

The calibration process used to determine the transformation (T_range-star) between the range scanner (X_range) and the attached rigid body (X_star). Once this transformation is known, range scanner points can be transformed into physical space, since the OPTOTRAK is always outputting the transformation between the two emitters (T_star-opto).

Fig. 5

The calibration phantom with an optical probe and its 3 mm spherical tip placed in the divot of one of the nine white disks.

Fig. 6

(Top left) Digital photograph taken of the operating scene from the viewpoint of the range scanner; (top right) range scanner setup in the OR; and (bottom) laser range scanner output showing a textured point cloud from the liver surface.

Fig. 7

The radius (left graph) and the error residual (right graph) resulting from the sphere fit as a function of the depth in the scanner’s field of view.

Fig. 8

Segmented CT surfaces of the liver phantom and subsurface tumors. The tumors are labeled in accordance with Table IV.

Fig. 9

(Top) Nondeformed CT contour (white) and deformed CT contour (gray) of a phantom surface aligned by a point-based registration, before ICP. Notice the significant deformation on the right side of the image. (Bottom) The two CT contours after implementing the ICP registration method. Now the right surface now matches much better, at the expense of a false rotation that misaligns the left side of the surface.

Fig. 10

Range scan data registered to and overlaid on the preoperative tomographic sets. From left to right, the slices become more superior. The large primary tumor can be seen in the right image.

Fig. 11

CT to range scanner registration of clinical trial data. The dark points indicate that the range scanner points are outside the CT surface and the white points indicate that these surface points are inside the surface.

Fig. 12

Observed “bloom” in the fiducial configuration. Notice how all the scanner fiducials are further away from the centroid than the CT fiducials.