Optimal landmarks selection for minimal target registration error in
image-guided neurosurgery
Reuben R Shamir1*, Leo Joskowicz1, Sergey Spector2, and Yigal Shoshan2
School of Engineering and Computer Science, The Hebrew University, Jerusalem, Israel.
2
Department of Neurosurgery, Hadassah University Hospital, Ein-Karem, Jerusalem, Israel.
1
ABSTRACT
We describe a new framework and method for the optimal selection and placement of anatomical landmarks and
skin/skull-mounted markers in image-guided neurosurgery. The method allows the surgeon to optimally plan marker
locations on the diagnostic images before preoperative imaging, and to intraoperatively select the fiducial markers and
the anatomical landmarks that minimize the Target Registration Error (TRE). The optimal marker configuration selection
is done on the routine diagnostic image by the surgeon following the target selection based on a computed visual
Estimated TRE (E-TRE) map that indicates the expected TRE. The E-TRE map is automatically updated when the
surgeon interactively adds/deletes candidate landmarks and targets to satisfy additional clinical requirements, such as the
patient pose and the preferred entry point. The method takes the guesswork out of the registration process, provides a
reliable localization uncertainty error for navigation, and can reduce the localization error without additional imaging and
hardware. Our clinical experiments on five patients who underwent brain surgery with a commercial navigation system
show that optimizing one marker location and the anatomical landmarks configuration reduces the average TRE from
4.7mm to 3.2mm, with a maximum improvement of 4mm. The reduction of the target registration error supports safer
and more accurate minimally invasive neurosurgical procedures.
Keywords: Registration, Neurosurgical Procedure, Therapy Planning, Data Integration for the Clinic/OR, Localization
and Tracking Technologies
This work is not published or considered for publication elsewhere.
1. PURPOSE
A key step in Image Guided Surgery (IGS) is the accurate intraoperative alignment, called registration, between
preoperative MRI/CT images and the intraoperative physical anatomy. Landmark-based registration is the most common
approach in existing IGS. In it, predefined landmarks (e.g. anatomical landmarks and skin/skull mounted markers) are
localized on the MRI/CT images and correlated with their counterparts on the physical anatomy. Most commercial
systems report the registration error at the landmarks, called the Fiducial Registration Error (FRE), although the Target
Registration Error (TRE), which is the distance between the predefined target locations after registration, is the most
clinically relevant error.
The placement of fiducial markers and the selection of the anatomical landmarks have a direct influence on the TRE and
on the navigation error at various locations of interest. Poorly placed fiducial markers, or a selection of anatomical
landmarks with large localization errors can yield clinically unacceptable navigation errors [1, 2]. The accuracy can be
improved by the judicious placement of the fiducial markers before imaging, and by the selection of the subset of
anatomical landmarks that minimizes the localization error. Moreover, the placement and selection error should be
estimated at the target, and not at the landmarks.
Previous work shows that the optimal placement of markers before point-based registration can significantly reduce the
predicted TRE [2-5]. West et al. [2] propose guidelines for landmarks placement for a given target point. Liu et al. [5]
describe a computerized module to approximate optimal landmarks that minimizes the expected TRE with a genetic
algorithm for abdominal and thoracic photogrammetry positioning system. They measured the TRE on abdomen
phantom and achieved an average TRE improvement of 0.5 mm. Riboldi et al. [4] developed algorithm for optimal
* Email: [email protected]; phone +972-77-200-5991; http://www.cs.huji.ac.il/~caslab/site/
Figure 1. (a) 3D image of the patient head surface extracted from a preoperative MRI image showing the selected anatomical
landmarks and target; (b) Estimated TRE map showing the expected error for adding markers at various locations on the head’s
surface. Examples of a ‘good’ and a ‘bad’ markers placements, associated with a small and a large expected TRE, respectively, are
shown. The expected TRE values (in mm) are color-coded as shown on the rightmost scale.
markers placement that combines genetic algorithms and Taboo Search (TS). They tested it in a simulation study on ten
images of prostate patients and report an improvement in the TRE/FRE ratio of 0.18, without directly addressing the
observed TRE, which is the clinically relevant measure. Atuegwu et al. [3] developed an improved patient-specific
module that approximates optimal landmarks for a target region instead of a single target location and incorporated a
model of the possible deviation caused by skin motion. An average improvement of 1.5mm is observed in their phantom
and simulation study.
The existing solutions have the following limitations. 1) the methods were tested on a phantom or a simulation study,
and thus do not capture all the contributing error factors; 2) no user control or error visualization is provided; 3) the
solutions are a locally optimal and depend on the initialization setup; 4) the methods are applicable only to fiducial
markers and not for anatomical landmarks, since they assume identical Fiducial Localization Error (FLE) distribution,
which as we have shown previously, does not hold for anatomical landmarks [1].
2. METHODS
In this paper, we describe and test in a clinical setup a new framework and method that overcomes the limitations of
current solutions for the optimal selection and placement of landmarks in image-guided neurosurgery. The method
computes a globally optimal solution and provides the surgeon with visual and easy to understand feedback of the
expected TRE. The surgeon can then interactively change the optimal landmarks setup based on clinical considerations,
and evaluate the results. The TRE estimation module incorporates varying FLE distributions that are recorded in the
same clinical setup. We have validated the method in a clinical setup.
The full paper methods section consists of three subsections: 1) optimal landmarks computation; 2) error visualization
and user interaction, and; 3) experimental setup. We briefly summarize each next.
2.1 Optimal landmarks computation
We define formally two problems of interest: 1) the optimal landmarks selection and; (2) the optimal landmarks
placement.
Optimal landmarks selection is the task of selecting a subset of landmarks from a given initial set of landmarks such
that the TRE is minimized. It is applicable for pre-imaging anatomical landmarks selection, or for intra-operative
refinement selection of the previously chosen landmarks – both anatomical landmarks and skin/skull mounted markers.
Since these sets are usually small, an exhaustive search is a feasible solution.
Optimal landmarks placement is the task of defining landmarks locations on a surface such that the TRE is minimized.
It is applicable for pre-imaging planning of skin/skull mounted markers. Since the surface may include hundreds of
thousands of points, an exhaustive search is computationally infeasible. Instead, we propose an approximation solution
to obtain a globally optimal solution. First, several thousands of points are uniformly sampled from the surface and are
stored in a hierarchical spatial data-structure. Each sampled point is considered a potential landmark that may be added
to the selected landmarks set. Next, a small (5-20) representative group of potential landmarks is selected from top of the
spatial data structure. An exhaustive search is performed on the selected set to find the subset that minimizes the TRE.
The points in the selected subgroup are recursively refined by selecting points in the lower levels of the hierarchical data
structure. The expected TRE is computed with a custom module that incorporates empirically measured, non-uniform
FLE distributions.
2.2 Error visualization and user interaction
For practical and clinical reasons, the surgeon may be interested in changing the computed optimal landmarks. We have
developed a module that allows the surgeon to add and/or remove the selected landmarks with a visual error feedback.
The expected TRE map is superimposed on the patient’s head surface and indicates the predicted TRE of adding or
removing each landmark (Figure 1). With this module, the surgeon can incorporate clinical information into the
landmarks planning and avoid hard locating landmarks or areas with expected large skin deformations.
2.3 Experimental setup
To quantify the impact of optimal selection and placement of landmarks on the targeting accuracy, we conducted the
following clinical experiment. Five patients whom underwent a brain surgery with a commercial navigation system were
randomly selected. On the diagnostic MRI image, the surgeon defined one target location and 5–9 anatomical landmarks.
The outer head surface was modeled with a set of points where each point is a possible marker location. In addition, nine
anatomical landmarks were defined on the patient’s image. The E-TRE was computed for each possible marker location
with isotropic, identically distributed FLEs (note that this is a valid assumption for fiducials, not for anatomical
landmarks). Then, an E-TRE map was generated. The surgeon selected two landmarks locations on the map, one with a
low E-TRE value (good) and the other with a high E-TRE value (bad). Before the pre-operative imaging, three fiducials
were attached to the patient’s head: two at the good and bad locations, and one at the target location. After MRI imaging
of the patient’s head with the standard navigation protocol, the surgeon localized the corresponding anatomical
landmarks and fiducials on the patient’s preoperative image. Then, in the operating room, two surgeons correlated the
predefined landmarks with the physical landmarks locations by touching them with a tracked pointer. Each landmark
was repeatedly selected 3–6 times on both the preoperative image and on the physical anatomy.
In the laboratory, our program automatically selected the optimal anatomical landmarks. The E-TRE values of all the
possible subsets of the anatomical landmarks were computed and the subset with the smallest E-TRE value was taken as
optimal. The anatomical landmarks were modeled with different FLE distributions that were obtained empirically in the
same setup and reported in our previous work [1]. E-TRE values were computed by Monte-Carlo statistical simulation
with these FLEs distributions.
Three landmarks configurations were generated and compared to their actual TRE: 1) original predefined anatomical
landmarks and a bad fiducial placement; 2) original predefined anatomical landmarks and a good fiducial placement,
and; 3) selected optimal anatomical landmarks and a good fiducial placement. Overall, 324 different landmarks pairings
were built and analyzed from the three different landmarks configurations, the 3-6 repetitive selections of the same
landmarks on 5 patients’ MRI images and physical anatomy by the two surgeons.
3. RESULTS
Table 1 shows the mean, STD, and maximum observed TRE values for each patient. The mean observed TRE with the
‘good’ marker and the original predefined anatomical landmarks was 4.7±1.3mm, with the 95% confidence interval of
6.7mm, and the maximum error of 7.6mm. The mean TRE with the ‘good’ marker and the original predefined
anatomical landmarks was 3.5±1.4mm, with the 95% confidence interval at 5.7mm, and maximum error of 6.0mm. The
mean observed TRE with the optimal marker and selected anatomical landmarks was 3.2±1.1mm, the 95% confidence
interval was 4.8mm, and the maximum error of 5.2mm. The mean difference between the TRE of the ‘bad’ and ‘good’
markers was 1.2±1.0mm with a maximum improvement of 4.1mm. The mean difference between the TREs of the ‘bad’
and ‘good’ marker with optimal anatomical landmarks selection was 1.5±1.0mm with maximum improvement of 4.1mm.
These results show that optimizing a single fiducial location while keeping the others fixed reduces the mean TRE from
4.7mm to 3.5mm. Optimal landmarks selection reduced the mean TRE even further, to an average of 3.2mm.
Patient 1 Patient 2 Patient 3 Patient 4 Patient 5
Landmarks setup
1. Bad marker +
mean
6.2
5.2
4.3
3.4
6.0
all anatomical
std
0.4
0.8
0.8
0.8
0.7
landmarks
max
6.9
6.8
5.5
5.4
7.6
2. Good marker +
mean
5.3
4.0
4.0
1.8
4.2
all anatomical
std
0.5
0.5
0.7
0.4
0.5
landmarks
max
6.0
5.12
5.1
2.8
5.2
3. Good marker +
mean
3.6
4.0
4.0
1.8
4.2
selected anatomical
std
0.6
0.5
0.7
0.4
0.5
landmarks
max
4.7
5.12
5.1
2.8
5.2
Table 1. Three landmarks configurations are compared with respect to their measured TRE on 5 patients: 1) all
predefined anatomical landmarks and the ‘bad’ marker; 2) all predefined anatomical landmarks and the ‘good’ marker,
and; 3) selected optimal anatomical landmarks and ‘good’ marker. A total of 324 different landmarks pairings were
analyzed from the three different landmarks configurations, the 3-6 repetitive selections of the same landmarks on 5
patients’ MRI images, and physical anatomy selection by two surgeons. All measured TRE values are reported in mm.
4. NOVELTY
The key novel features of our method are: 1) the optimal landmarks selection method was tested with clinical data to
incorporate all the intraoperative error factors; 2) our method produces a globally optimal solution that does not depend
on the initial landmarks setup and selection; 3) we provide a user control with an error visualization feedback that allows
the surgeon to incorporate other clinical and practical factors, and; 4) the TRE estimation module incorporates different
landmark-specific FLE distributions.
5. CONCLUSIONS
We have presented new methods that allow surgeons to optimally plan landmarks locations prior to preoperative
imaging, and to select the landmarks in the operating room that minimize the Target Registration Error. The
improvement in accuracy is expected to reduce complications and may allow new procedures that could not be done
because of the suboptimal targeting accuracy. It is expected to provide a new, principled pre-imaging method for
landmarks placement, save preoperative and intraoperative time, and improve the reliability and accuracy of
intraoperative target registration error.
The proposed method has several key advantages. Preoperatively, before preoperative image acquisition, it allows to
automatically compute the optimal landmarks placement based on the diagnostic MRI/CT. Since the diagnostic images
are acquired routinely no additional scanning of the patient is required. The E-TRE map provides a visual and easy to
understand feedback of the targeting error, and enables the surgeon to relocate landmarks and evaluate the expected
targeting errors. It also provides a more realistic target localization error based on empirical localization error models.
Intraoperatively, it can achieve the most accurate targeting localization in neurosurgical navigation without additional
hardware. Since the landmarks selection process is fully automatic, it takes the guesswork out of the registration process
and can thus shorten the intra-operative setup time. In a clinical experiment optimizing one marker placement and the
anatomical landmarks selection we have observed an average improvement of 1.5mm in the TRE. The TRE
improvement can allow safer and more accurate minimally invasive neurosurgical procedures.
References
1.
Shamir, R. R., et al., Localization and registration accuracy in image guided neurosurgery: a clinical study.
International Journal of Computer Assisted Radiology and Surgery, submitted, 2008.
2.
West, J.B., et al., Fiducial point placement and the accuracy of point-based, rigid body registration.
Neurosurgery, 2001. 48(4): p. 810-6; discussion 816-7.
3.
Atuegwu, N.C. and R.L. Galloway, Sensitivity analysis of fiducial placement on transorbital target registration
error. International Journal of Computer Assisted Radiology and Surgery, accepted, 2008. 2(6): p. 397-404.
4.
Riboldi, M., et al., Genetic evolutionary taboo search for optimal marker placement in infrared patient setup.
Phys Med Biol, 2007. 52(19): p. 5815-30.
5.
Liu, H., et al., Optimal marker placement in photogrammetry patient positioning system. Med Phys, 2003.
30(2): p. 103-10.