 |
|
The human spine consists of a jointed set of spinal segments.
Surgical treatment of many diseases requires insertion of spinal
screws. Screw misplacement may result in vascular or neural
injury. The surgeon cannot visualize the end position of the
screw directly. Estimation of the angle and location of screw
insertion has traditionally been based upon the surgeon's
judgment and upon intraoperatively acquired fluoroscopic images
that cannot provide 3D information. Although several commercial
companies have provided 3D image guidance based upon preoperatively
acquired CT scans, this guidance has not allowed for the fact that
the patient's position on the operating table is often different
from that during CT image acquisition. Given a jointed, "bendable"
spine, this difference in patient position can produce errors of a
centimeter or more, whereas accuracy in the order of a millimeter
is generally required.
We propose fast, segmental registration of preoperatively acquired
CT images with the intraoperative patient using intraoperative
fluoroscopy as an intermediary.
|
|
|
|
|
The general approach requires segmentation of individual spinal
segments from preoperatively acquired CT data and intraoperative
registration of individual spinal segments with the patient via
fluoroscopic images.
- SEGMENTATION OF THE SPINAL COLUMN
We can segment the spinal column as a tube, using tubular
extraction methods originally developed for
vascular imaging.
Tubular extraction program applied to the spine and to ribs, using
volume rendering
to display the selected regions.
- SEGMENTATION OF INDIVIDUAL SPINAL SEGMENTS
Given the definition of the overall spinal tube provided above,
we have two potential methods of defining individual spinal
segments from that tube.
- Segmentation by deformable shape loci
The approach is to create a generalized vertebral
segmental model which will deform to the specific
configurations of each individual patient. This approach
is under
active development
as applicable to multiple medical problems.
Deformable shape loci model of a spinal segment.
- Estimation of the volume occupied by a spinal segment and
use of intensity windowing to show the boundaries of the
segment without the requirement of perfect segmentation.
Left: Estimation of the planes defined by adjacent
vertebral endplates can be used to approximate the volume
occupied by a spinal segment.
Center: Our tubular extraction method automatically
defines segmental levels by providing "jiggles" in the
skeleton curve defining the spinal column.
Right: Given the tangent direction of the central
skeleton curve, the locations shown in the central figure
above, and the tube of the spinal column, it is possible
to automatically define confining volumes for each spinal
segment. Within each volume, details of the spinal segment
can be shown by intensity windowing and selective volume
rendering, as already used in our vascular imaging protocols.
- REGISTRATION OF INDIVIDUAL SPINAL SEGMENTS WITH A
FLUOROSCOPIC IMAGE
The approach is that of "synthetic fiducials", as described
in the publications below. In brief, pairs of digitally
reconstructed radiographs are created preoperatively from the
patient's CT scan. Associated pairs of shadows on these images
are reconstructed into 3D to create 3D tubular objects that can
be registered intraoperatively with the shadows seen on
fluoroscopic images, using the same approach we have used for
3D-2D registration of of tubular blood vessels.
(See Image-Guided Vascular
Neurosurgery and
Image-Guided Endovascular Neurosurgery)
We have tested the accuracy of this approach by implanting
metallic objects in a jointed, dry spine and performing a CT
scan of the spine. The actual 3D coordinates of the implanted
objects were determined manually as an estimate of "truth".
Synthetic fiducials were then created, and spinal segments
were registered with pairs of fluoroscopic images. The
accuracy of registration was estimated by reconstructing the
location of each metal ball by backward ray casting from the
pair of registered fluoroscopic images. All implanted objects
were reconstructed at submillimeter accuracy. As a more general
estimate of accuracy, the accuracy of reconstruction of each
point in the volume was determined by similar methods. The
figure below shows in yellow all points that could be
reconstructed at submillimeter accuracy. We believe that this
approach can provide fast, accurate registration for 3D
image-guided spine surgery.
A. Top view of a spinal segment registered with a
fluoroscopic image. All points shown in yellow could be
reconstructed at submillimeter accuracy.
B. CT view of the same segment. Implanted test objects
are present.
C. Longitiduinal view with a blue arrow pointing to the
segment in question. All points in yellow could be registered
with submillimeter accuracy. Black points were not tested.
Red points were registered with an accuracy of more than 1 but
less than 3mm.
|
|
|
|
|
Aylward SR, Pizer SM, Bullitt E, Eberly D (1996)
Intensity ridge and widths for 3D object segmentation and
description IEEE WMMBIA IEEE 96TB100056, 131-138.
Bullitt E, Liu A, Pizer SM (1997b)
Three dimensional reconstruction of curves from pairs of projection
views in the presence of error. I. Algorithms. Med Phys 24: 1671-8.
Bullitt E, Liu A, Pizer SM (1997c)
Three dimensional reconstruction of curves from projection views.
II. Analysis of error. Med Phys 24:1679-87.
Liu A, Bullitt E, Pizer SM (1997)
Surgical instrument guidance using synthesized anatomical
structures. CVRMed-MRCAS '97.
Lecture Notes in Computer Science 1205: 99-108.
Liu A, Bullitt E, Pizer SM (1998)
3D/2D registration using tubular anatomical structures as a basis.
MICCAI 1998. Lecture Notes in Computer Science 1496: 952-963.
Pizer S.M., Eberly D.H., Morse B.S., Fritsch D.S.,
Zoom invariant vision of figural shape: the mathematics of cores.
CVIU 69 (1998) 55-71.
Bullitt E, Liu A, Aylward SR, Soltys M, Boxwala A, Rosenman J,
Pizer S (1997a)
Methods for displaying intracerebral vascular anatomy. Am J
Neuroradiol 18:417-420.
Bullitt E, Aylward S, Liu A, Stone J, Mukherji S, Coffey C,
Gerig G, Pizer SM (1999)
3D graph description of the intracerebral vasculature from
segmented MRA and tests of accuracy by comparison with x-ray
angiograms IPMI 99 Lecture Notes in Computer Science 1613:308-321.
Bullitt E, Liu A, Aylward S, Coffey C, Stone J, Mukherji SK,
Muller K, Pizer SM (1999) Registration of 3D vessels with 2D
digital angiograms. Clinical evaluation.. Academic Radiology
6: 539-546.
|
|
|
