Int J CARS (2015) 10:1127–1140
DOI 10.1007/s11548-014-1126-5
ORIGINAL ARTICLE
The role of automatic computer-aided surgical trajectory planning
in improving the expected safety of stereotactic neurosurgery
M. Trope · R. R. Shamir · L. Joskowicz ·
Z. Medress · G. Rosenthal · A. Mayer ·
N. Levin · A. Bick · Y. Shoshan
Received: 10 July 2014 / Accepted: 24 October 2014 / Published online: 20 November 2014
© CARS 2014
Abstract
Purpose Minimal invasion computer-assisted neurosurgi-
cal procedures with various tool insertions into the brain may
carry hemorrhagic risks and neurological deficits. The goal of
this study is to investigate the role of computer-based surgical
trajectory planning tools in improving the potential safety of
image-based stereotactic neurosurgery.
Methods Multi-sequence MRI studies of eight patients who
underwent image-guided neurosurgery were retrospectively
processed to extract anatomical structures—head surface,
ventricles, blood vessels, white matter fibers tractography,
A shorter and restricted version of this paper was orally presented at
the 27th International Conference on Computer Aided Radiology and
Surgery, Heidelberg, Germany, 2013.
M. Trope · L. Joskowicz (B)
School of Engineering and Computer Science, The Hebrew
University of Jerusalem, Givat Ram Campus, 91904 Jerusalem,
Israel
e-mail: josko@cs.huji.ac.il
R. R. Shamir · L. Joskowicz
The Edmond and Lily Safra Center for Brain Research (ELSC),
The Hebrew University of Jerusalem, Jerusalem, Israel
R. R. Shamir
Department of Neurosurgery, Center for Functional and
Restorative Neurosurgery, Hadassah University Hospital,
Ein-Karem, Jerusalem, Israel
Z. Medress
School of Medicine, Stanford University, Palo Alto, CA, USA
Z. Medress · G. Rosenthal · Y. Shoshan
Department of Neurosurgery, Hadassah University Hospital,
Ein-Karem, Jerusalem, Israel
A. Mayer · N. Levin · A. Bick
fMRI Unit, Department of Neuroradiology, Hadassah Hebrew
University Medical Center, Jerusalem, Israel
and fMRI data of motor, sensory, speech, and visual areas.
An experienced neurosurgeon selected one target for each
patient. Five neurosurgeons planned a surgical trajectory for
each patient using three planning methods: (1) conventional;
(2) visualization, in which scans are augmented with over-
lays of anatomical structures and functional areas; and (3)
automatic, in which three surgical trajectories with the low-
est expected risk score are automatically computed. For each
surgeon, target, and method, we recorded the entry point and
its surgical trajectory and computed its expected risk score
and its minimum distance from the key structures.
Results A total of 120 surgical trajectories were collected
(5 surgeons, 8 targets, 3 methods). The surgical trajectories
expected risk scores improved by 76 % (SD = 11.6 %; p <
0.05, two-sample student’s t test); the average distance of
a trajectory from nearby blood vessels increased by 1.6 mm
(SD = 0.5, p < 0.05) from 0.6 to 2.2 mm (243 %). The ini-
tial surgical trajectories were changed in 85 % of the cases
based on the expected risk score and the trajectory distance
from blood vessels.
Conclusions Computer-based patient-specific preoperative
planning of surgical trajectories that minimize the expected
risk of vascular and neurological damage due to incorrect
tool placement is a promising technique that yields consis-
tent improvements.
Keywords Stereotactic neurosurgery · Surgical trajectory
planning · Surgical risk reduction · Safety · Computer-aided
planning
Introduction
Precise targeting of tumors, lesions, and anatomical struc-
tures with a probe, a needle, a catheter, or an electrode inside
123
1128
the brain based on preoperative CT/MRI scans are nowadays
standard surgical tasks in stereotactic neurosurgery. Common
stereotactic neurosurgical procedures may include stereo-
tactic brain biopsies, treatment of hydrocephalus, aspiration
and evacuation of deep brain hematomas, Ommaya catheter
insertion, deep brain stimulation, depth electrode insertion
for epilepsy, endoscopic intraventricular procedures, and
minimal access craniotomies. In all cases, incorrect place-
ment of the surgical tool may result in undesired outcomes
such as non-diagnostic tissue samples, ineffective treatment,
intracranial hemorrhagic complications which despite their
low incidence may bare the highest risk of severely harmful
neurological sequels, and temporary or permanent neurolog-
ical damage [1–4]. Incorrect surgical tool placement may
also require an additional surgical procedure to achieve the
intended therapeutic effect and/or to treat the ensuing compli-
cations. Such revision surgeries, which are sometimes open
procedures, can be associated with significant additional risk,
time, and morbidity.
Planning a safe surgical trajectory that minimizes the
expected risk of incorrect surgical tool placement and its
possible undesired outcomes is of great practical importance.
Currently, preoperative trajectory planning of image-guided
stereotactic neurosurgical procedures typically involves the
use of neuronavigation systems or stereotactic frame to select
target structures on 2D axial, sagittal, and/or coronal cross-
sections of preoperative CT/MRI head scans. This manual
preoperative planning task might be complex, time consum-
ing, and user and patient dependent. It requires the surgeon
to mentally identify and reconstruct complex 3D structures,
including blood vessels, eloquent brain areas, and the ventric-
ular system and derive the spatial relationships between these
structures and the surgical trajectory. In some cases, manual
corrections and optimization of the surgical plan is empiri-
cally performed during the surgery itself in sterile conditions
and under pressure without the possibility to thoroughly and
quantitatively analyze the best possible surgical trajectory.
Indeed, the evaluation of the expected treatment risks is thus
a complex, challenging, and time-consuming task even for
experienced neurosurgeons.
Spatial visualization and segmentation of critical brain
structures has been proposed as a means of enhancing the
neurosurgeon’s spatial perception and improving the aware-
ness of structures surrounding the planned trajectory [5–8].
Recent publications describe methods for the visualization of
the expected risk and the automatic computation of safe tra-
jectories on the preoperative head image [5,8–19]. Common
to these methods is that the structures at risk are first seg-
mented and that the distance from the candidate trajectory to
each segmented structure is incorporated in a weighted sum
formula that maximizes the distances between structures and
trajectory. Previous studies can be classified into three cate-
gories based on their application domain: (1) deep brain stim-
123
Int J CARS (2015) 10:1127–1140
ulation for movement disorders where trajectories to targets
such as the subthalamic nucleus, internal Globus Pallidus,
and ventral intermediate thalamus region have been investi-
gated [9–11,16]; (2) stereo-electroencephalography, that is,
the recording of electrical signals via electrodes surgically
implanted into selected cortical targets to identify the epilep-
togenic zone of the brain using several multi-lead electrodes
in patients with epilepsy [6,7,15,17]; and (3) general appli-
cations/biopsy, that is, the sampling of brain tissue for histo-
logical testing in a variety of brain locations [7–9,12–14,18].
A variety of methods for optimal trajectory planning have
been developed in the past 5 years. Following earlier studies
[7,9,14], we have developed a method that computes safe tra-
jectories, quantifies the safety improvement, and visualizes
the risk [12,13]. Essert et al. [10] extend this approach to a
structured description of the risk factors and suggest using
a qualitative test in addition to the quantitative one consid-
ering the complex nature of the problem. Liu et al. [16] test
the automatic planning method with multiple surgeons and
various targets for DBS applications. De Momi et al. [15,20]
extend this method to multiple trajectories and new cortical
target locations for epilepsy applications. Other groups have
recently extended this approach [17,21]. The hardware/GPU
acceleration and/or software-based preprocessing of the data
to facilitate interactive selection of trajectory are described
in [13,18,19]. The expected risk that is associated with each
trajectory is color coded and rendered on the entry point to
indicate its value [12,13,19]. This allows for the intuitive
selection of the surgical trajectory entry point. Finally, aug-
mented reality visualization of the risk and optimal trajec-
tories has been proposed for a more intuitive selection of
optimal trajectory [19].
The main goal of this study is to investigate the role of
computerized preoperative surgical trajectory planning tools
in improving the expected safety of image-based stereo-
tactic neurosurgery. We hypothesize that the visualization
of anatomical structures such as blood vessels, ventricles,
function-related areas, and white matter fiber tractography
in single image data sets helps in planning safer surgical
trajectories. In particular, we hypothesize that the automatic
selection of a few surgical trajectories with the lowest esti-
mated risk systematically improves the expected safety of
candidate trajectories compared with the conventional man-
ual selection method.
This work extends previous studies by incorporating
patient-specific physiological mapping of important func-
tions such as motor, somatosensory, speech, and vision. This
facilitates the evaluation of the suggested planning method
with respect to possible direct neural-functional damage in
addition to the risk of hemorrhage that was the focus of pre-
vious studies. Moreover, the targets in this work are located
at various intracranial locations, and the planning method
was evaluated by five neurosurgeons at different level of
Int J CARS (2015) 10:1127–1140 1129

Fig. 1 Illustrative T1-Gd MR slices of eight patients and the targets
selected for each. a Patient 1: 31 y/o male with right posterior insu-
lar cavernous hemangioma; b Patient 2: 8 y/o female with recurrent
ependymoma of the left occipital ventricular wall; c Patient 3: coronal—
55 y/o male with abscess within the right cerebral peduncle and upper
midbrain; d Patient 4: coronal—5 y/o male with cystic craniopharyn-
gioma; e Patient 5: axial—65 y/o male with left temporal glioblastoma;
f Patient 6: sagittal—45 y/o male with recurrent Anaplastic Oligoastro-
cytoma close to the right pyramidal tract; g Patient 7: axial 42 y/o male
with recurrent right occipital glioma; and h Patient 8: axial—22 y/o
NF1 female with left insular and basal ganglia solid and cystic pilocytic
astrocytoma

expertise. Therefore, this work extends previous studies that

were limited to specific target zones or incorporated one/two Table 1 Summary of patients that were included in the study
neurosurgeons. Considering the limitations of a retrospective Patient # Target image Age Gender Diagnosis and
study, we believe that the distances of a planned trajectory to procedure
its surrounding structures are the most important objective 1 Fig. 1a 31 Male Right posterior insular
measures for improvement in trajectory’s safety. This study cavernous hemangioma.
adds more quantitative evidence in favor of the automation Minimal access craniotomy
of the trajectory planning and supports the results of previous 2 Fig. 1b 8 Female Recurrent ependymoma of
few studies that have reported this measure [8,11,14]. the left occipital
ventricular wall. Minimal
access craniotomy
3 Fig. 1c 55 Male Abscess within the right
Materials and methods cerebral peduncle and
upper midbrain.
We have designed a comprehensive retrospective study and Stereotactic drainage via
twist drill
implemented a software prototype of our previously pub-
4 Fig. 1d 5 Male Cystic craniopharyngioma.
lished method [12,13] to test the abovementioned hypothe-
Stereotactic drainage via
ses. twist drill
5 Fig. 1e 65 Male Left temporal glioblastoma.
Study protocol Minimal access craniotomy
6 Fig. 1f 45 Male Recurrent anaplastic
Multi-sequence anatomical, functional, and diffusion tensor oligoastrocytoma close to
MRI studies of eight patients who underwent image-guided the right pyramidal tract.
Minimal access craniotomy
stereotactic neurosurgery for intracranial biopsy, stereotac-
7 Fig. 1g 42 Male Recurrent right occipital
tic drainage, or minimal access craniotomy at our institution glioma. Minimal access
were retrospectively obtained (Fig. 1; Table 1). The patients craniotomy
were selected by a senior neurosurgeon (YS) based on their 8 Fig. 1h 22 Female Left insular and basal
tumor location to maximize target location variability. The ganglia solid and cystic
MRI studies were first preprocessed in the computational lab- pilocytic astrocytoma.
Stereotactic Biopsy and
oratory. For each patient study, the scans were first aligned
drainage via twist drill
to establish a common geometric reference frame. Next, the

123
1130
Fig. 2 Illustration of structures
segmentation and delineation on
a T1-Gd MR scan: axial,
sagittal, coronal, and 3D view
showing the blood vessels (red),
the ventricles (purple), the
function-related areas—motor
activation (orange),
somatosensory activation (blue),
speech activation (green), and
visual activation (yellow)—and
the white matter fibers
tractography motor (light
orange), sensory (light blue),
vision (light yellow), and
high-density SLF (light green)
scans were segmented to extract the head outer surface, ven-
tricles, blood vessels, high-density white matter fibers trac-
tography, and the cortical functional motor, visual, sensory,
and speech areas (Fig. 2). A senior neurosurgeon (YS) who
did not participate in the rest of the study selected one tar-
get point for each patient and defined a region on the skull
surface for candidate entry points.
Five neurosurgeons—two senior neurosurgeons (experts)
and three junior neurosurgeons (trainees)—were then asked
to plan a surgical trajectory for each predefined target using
three methods: (1) the conventional method based on 2D
axial, sagittal, and coronal views of the original MRI scans;
(2) the visualization method, in which scans are augmented
with overlays of the anatomical and functional structures seg-
mentations; and (3) the automatic method, in which the sur-
geon chooses one of the three automatically computed sur-
gical trajectories with the lowest risk score proposed by the
software.
To estimate the direct effect of the visualization and auto-
matic methods on trajectory planning with respect to the con-
ventional method, the methods were used in subsequent order
such that the trajectory selected with the conventional method
was presented to the neurosurgeon when planning with the
later methods. This may induce bias in the selected trajecto-
ries, as the neurosurgeon may prefer a trajectory nearby the
one that was originally selected. However, this bias reduces
the efficacy of our methods, and therefore, our results are
probably a conservative prediction for the actual expected
improvement. More importantly, it facilitates the direct com-
parison of the methods, and otherwise, it would be hard to
123
Int J CARS (2015) 10:1127–1140
determine to what extent the differences in trajectory plans
were followed by the presented risk information. For each
method, the surgeon was given the option to select a new
trajectory if a safer option was found or to retain the trajec-
tory from the previous selection. For each surgeon, target,
and method, we record the entry point and its corresponding
surgical trajectory and compute various estimated risk mea-
surements. Finally, we perform a detailed data analysis of the
collected data in the computational laboratory.
Patient scans
Eight patients who underwent image-guided stereotactic neu-
rosurgery at our institution were retrospectively selected.
The patients (6 males and 2 females, ages 5–65; Table 1)
were selected to represent variety of diseases and target
locations. Preoperative imaging studies of each patient con-
sisted of selected multi-sequence MR scans: (1) T1-weighted
gadolinium-enhanced magnetic resonance scan (T1-Gd)
512 × 512 × 208/216 voxels3 , 0.49 × 0.49 × 1.0 mm3 ; (2)
fluid-attenuated inversion recovery scan (FLAIR) 512 ×
512 × 208/216 voxels, 0.49 × 0.49 × 1.0 mm3 ; (3) diffu-
sion tensor imaging scan (DTI) 128 × 128 × 50/60 vox-
els, 2.0 × 2.0 × 2.0 mm3 , and; (4) functional magnetic res-
onance imaging scans (fMRI) 80 × 80 × 30/40 voxels,
2.75 × 2.75 × 3.0/3.5 mm3 . The scans were acquired with
a Siemens 3T MAGNETOM Trio scanner prior to surgery.
All scans were registered to the T1-Gd sequence as follows.
The FLAIR and T1-Gd were registered using custom soft-
ware that implements a rigid-body point-based registration
Int J CARS (2015) 10:1127–1140
method [22]. DTI and T1-Gd were registered using the statis-
tical parametric mapping software package (SPM; SPM lab,
University College London, London, UK) that computes the
rigid-body transformation by maximizing the mutual infor-
mation between the “B0” and the T1-Gd images [23]. fMRI
and T1-Gd were registered using BrainVoyager (Brain Inno-
vation, Maastricht, Netherlands) that computes the rigid-
body transformation by maximizing the mutual information
between the echo planar image (EPI) sequence with the T1-
Gd image.
Targets
Targets were chosen for each patient on the T1-Gd scan.
Patients’ characteristics and their specific clinical diagnosis
are presented in Table 1 and Fig. 1. A specific location within
each patient’s tumor was selected by a senior neurosurgeon
(YS) such that the target is a safe and effective location.
Anatomical structures
the anatomical structures include the segmentation of the
head outer surface, the ventricles, and blood vessels with
diameter >1 mm that are extracted from the volume T1-Gd
scans (Fig. 2). The outer surface of the head is extracted
as described in Shamir et al. [24]. First, the air outside
the head is automatically segmented. Then, a mesh is cre-
ated with the marching cubes algorithm [25] to model the
outer head surface. The ventricles are segmented semiau-
tomatically using the ITKSnap region-growing segmenta-
tion method [26]. First, a coarse estimate of the ventricles is
defined manually and then automatically refined. The blood
vessels are automatically or semiautomatically delineated for
each patient study with techniques described in [27]. Briefly,
the method starts with the automatic watershed-based seg-
mentation of large blood vessels and the construction of an a
priori intensity probability distribution function for smaller
ones. The blood vessels are then segmented with a graph
min-cut method based on an edge weighting function that
adaptively couples voxel intensity, intensity prior, and local
vesselness shape prior. The segmentation of all structures was
manually revised and verified by a senior neurosurgeon (YS)
by overlaying the relevant clinical images on the segmented
areas.
Candidate entry areas
The outer head surface was further edited to constraint the
optimal trajectory for clinically acceptable areas. The face
and ears zones were manually excluded. Then, the surface
above the hemisphere that is contralateral to the target was
removed to avoid cross-hemispheric trajectories. Finally, the
anterior–inferior head surface was removed to avoid the cere-
1131
bellum and the brain stem areas. The remaining surfaces were
considered candidate entry areas.
Diffusion tensor imaging (DTI)
DTI image processing and analysis was performed using
MrVista [28]. The mean of the non-diffusion weighted
images was automatically aligned to the T1-Gd scan using a
rigid registration mutual information algorithm. Fiber trac-
tography was used to delineate pathways (pyramidal tract,
optic radiation, and superior longitudinal fasciculus) for each
patient study. Pyramidal and visual pathways’ tracking was
performed using the probabilistic ConTrack algorithm [29].
Superior longitudinal fasciculus was delineated using a deter-
ministic fiber tractography algorithm [30]. A senior neuro-
surgeon (YS) verified that the resulted pathways are anatom-
ically reasonable.
Functional MRI (fMRI)
fMRI data analysis was performed using the BrainVoy-
ager Qx software package [31]. Head motion correction
and high-pass temporal filtering in the frequency domain
(3 cycles/total scan time) were applied to remove drifts and
to improve the signal-to-noise ratio. Functional images were
registered and incorporated into the 3D data sets through
trilinear interpolation. The changes in the BOLD contrast
associated with the performance were assessed on a pixel-by-
pixel basis with a multivariate linear model [32]. Language
regions were identified using a combination of three tasks:
visual verb generation, visual sentence generation, and audi-
tory verb generation. Motor regions were identified in the
images from the patient movements of their hands, feet, and
tongue. A senior neurosurgeon (YS) verified that the resulted
regions do indeed match the known anatomy.
Trajectory planning methods
We implemented the manual method, the visualization
method, and the automatic method to explore the trade-offs
involved in the trajectory planning decision-making process.
The manual method consists of manually selecting an
entry point directly on the axial, coronal, and sagittal views
of the T1-Gd scan and examining the resulting surgical tra-
jectory. This is the conventional planning method—it serves
as the control reference for the other methods.
The visualization method consists of selecting an entry
point on the head surface and evaluating the resulting tra-
jectory on the T1-Gd scan augmented with overlays of the
segmented anatomical and functional structures. The struc-
tures include the candidate entry points region, the blood ves-
sels with diameter >1 mm, and the ventricles from the T1-Gd
scans, the white matter fibers tractography from the DTI scan,
123
1132
and the functional motor, visual, sensory, and speech areas
from fMRI scans (Fig. 2). The rationale for this method is that
the surgical trajectory planning task is facilitated by having
key structures from different scans highlighted in different
colors and integrated in a single, augmented scan which was
presented to the neurosurgeon.
The automatic method: the relevant head surface was
evenly sampled with ∼1,000 points, each considered as a
candidate entry point to the predefined target. Then, a risk
Fig. 3 Illustration of the candidate insertion region as defined by the
surgeon in the automatic method (pink) and the color-coded risk val-
ues of candidate entry points on the skull (red, high; yellow, medium;
green, low; blue, lowest). Three trajectories (cylinders) are shown: the
trajectory selected with the manual method (dark purple, outside the
selected region), the trajectory selected by the visualization method
(light purple), and the trajectory selected with the automatic method
(yellow)
Fig. 4 Bar charts of the surgical trajectory expected risk score (left) and minimum distance from blood vessel (right) of each surgeon on the
average patient for each method (top) and for experts and trainees (bottom). The lines show the improvements between methods
123
Int J CARS (2015) 10:1127–1140
value was computed for each candidate trajectory (e.g., a pair
of entry and target points), and the three trajectories that are
associated with the minimal values and that are at least 2 mm
apart were selected. The surgeon then makes the final choice
by visual inspection of these three surgical trajectories. The
line between a candidate entry point and the target defines a
candidate surgical trajectory (Fig. 3). The rationale for this
method is to automate surgical trajectory selection and to
leave the final choice between the three best trajectories to
the surgeon (Figs. 4, 5).
Surgical trajectory expected risk score computation
The surgical trajectory expected risk score is an aggregate
of individual expected risk measures for all the segmented
anatomical and functional structures along the trajectory. The
score based on the proximity of the surgical trajectory to a
structure or area and on the estimated damage that may be
caused by penetrating it [12,13].
Let S = {S1 , S2 , . . . , S p } be a set of critical structures—
blood vessels, ventricles, white matter fibers tractography,
and functional areas—for which their penetration with a sur-
gical tool is undesirable or forbidden. The structures are rep-
resented by the voxels that were identified by segmentation
and delineation. Let R = {r1 , r2 , . . . , r p } be the expected
risk values associated with the penetration of these structures,
where ri = 0 indicates no expected risk and ri = 1 indicates
highest expected risk (penetration is forbidden). The sum of
the ri value is 1.
Int J CARS (2015) 10:1127–1140 1133

Fig. 5 Bar charts of the trajectory expected risk score (left) and minimum distance from blood vessels (right) of each patient on the average surgeon
for each method

The expected risk of the white matter fibers and functional of having the penetration of the surgical trajectory into one
areas is further subdivided to reflect the different risks. Since of the structure Sk is the maximum of the expected risks to
there is no consensus in the literature about the actual risk each structure. Therefore, the expected risk of trajectory tr
values for different structures, we asked the expert neurosur- is:
geon (YS) to define the values based on his experience. The
following values were predefined: blood vessels r4 = 0.4; risk(tr, riskVolume) = max (riskVolume(vi )) (1)
vi ∩tr =∅
ventricles r1 = 0.1; white matter fibers r2 = 0.2; and corti-
cal functional regions, r3 = 0.3. The risk of the white matter where vi is a voxel and riskVolume(vi ) is defined as:
fibers and cortical functional areas is further subdivided as
follows: motor fibers, f j4 = 0.4; sensory fibers, f j3 = 0.3; 
4 
4
r j · a jk
vision fibers, f j2 = 0.2; speech SLF fibers, f j1 = 0.1; riskVolume(vi ) = (2)
dist(S jk , v j ) + α
functional motor area, f a4 = 0.4; functional sensory area, j=1 k=1
f a3 = 0.2; functional vision area, f a2 = 0.1; and func-
tional speech area, f a1 = 0.3. The expected risk values where α is a nonnegative scalar constant used to avoid divi-
of the white matter fibers and the functional areas are then sion by 0 (in this study α = 1.). The weights were defined
(r2 × f i j ) and (r3 × f a j ) for j = 1, 2, 3, 4, respectively. Note by a senior neurosurgeon (YS) as follows.
⎧
that the senior neurosurgeon (YS) who determined the above ⎪
⎪ j = 1 ventricles 0.1
⎨
weights did not participate in the rest of the study. More- j = 2 white matter fibers 0.2
rj =
over, the above values were not disclosed to the participating ⎪
⎪ j = 3 functional cortical areas 0.3
⎩
neurosurgeons during the experiments to avoid bias. j = 4 blood vessels 0.4
The expected risk of having the penetration of the surgi- ⎧
⎪
⎪ j = 1 ventricles 0.25
cal trajectory into the structure is inversely proportional to ⎨
j = 2 white matter fibers fik
its minimum distance from it, weighted by its expected risk a jk =
⎪
⎪ j = 3 functional cortical areas f ak
value. The expected risk of having the penetration of the sur- ⎩
j = 4 blood vessels 0.25
gical trajectory into one of the structure is the maximum of
the expected risks to each structure. and
⎧
Let tr i = [ei , t] be a surgical trajectory with entry point ⎪ j = 1 speech fibers 0.1
⎪
⎨
ei and target t. Let V = {v1 , . . . , vn } be the set of voxels j = 2 vision fibers 0.2
fik =
intersected by the surgical trajectory. Let be the minimum ⎪
⎪ j = 3 sensory fibers 0.3
distance between the center of voxel v j intersected by the ⎩
j = 4 motor fibers 0.4
surgical trajectory tr i and structure Sk . The expected risk ⎧
⎪
⎪ j = 1 speech areas 0.3
of having the penetration of the surgical trajectory into the ⎨
j = 2 vision areas 0.1
structure Sk is inversely proportional to its minimum distance f ak =
⎪
⎪ j = 3 sensory areas 0.2
from it, weighted by its expected risk value. The expected risk ⎩
j = 4 motor areas 0.4

123
1134 Int J CARS (2015) 10:1127–1140

⎧
⎪ j = 1 ventricles (for ⎧
all k) millimeters). The expected risk score is an aggregate of the
⎪
⎪
⎪
⎪ ⎪
⎪ 1 speech expected risk of each structure and serves to compare tra-
⎪
⎪ ⎨
⎪
⎪ 2 vision jectories with a single number. The minimum distance of the
⎪
⎪ j = 2 (DTI) , k =
⎪
⎪ ⎪ 3 sensory
⎪
⎨ ⎪
⎩ surgical trajectory to the brain structures has a physical mean-
S jk = ⎧4 motor ing and is directly related to the intraoperative accuracy of the
⎪
⎪ ⎪
⎪ 1 speech needle insertion method. We differentiate between blood ves-
⎪
⎪ ⎨
⎪
⎪ 2 vision sels and other structures because of their clinical significance:
⎪
⎪ j = 3 (fMRI), k =
⎪
⎪ ⎪ 3 sensory
⎪ penetrating a blood vessel during closed surgery should be
⎪
⎪ ⎩
⎪
⎪ 4 motor avoided, as it may cause hemorrhagic complications. More-
⎩
j = 4 blood vessels (for all k) over, the blood vessels densely span the entire brain, so the
minimum distance of the surgical trajectories to them is usu-
Equation 2 reflects the hierarchical subdivision of the risk ally much smaller to that of the other structures.
factors. Index j denotes the four major risk factors: blood The relative measurements for two surgical trajectories are
vessels, cortical functional areas, functional fibers, and ven- as follows: (1) the trajectories expected risk score ratio (>0),
tricles. The functional cortical areas (fMRI) and fibers (DTI) and (2) the difference between the two trajectories minimum
are further subdivided into motor, sensory, vision and speech distance from blood vessels (in millimeters). The expected
areas that are denoted by index k. risk score ratio measures the significance of the risk improve-
The riskVolume is precomputed once on the entire seg- ment (<1) or worsening (>1), while the difference measures
mented images. Each voxel is associated with an expected the improvement (>0) or worsening (<0) of the minimum
risk value that is directly related to the estimated conse- distance to blood vessels.
quences and severity of the damage to the corresponding The measurements are organized in a 5×8×3 table (5 sur-
brain tissue/structure. The expected risk associated with geons, 8 targets, 3 methods) consisting of 120 entries. Each
each surgical trajectory is then directly computed from the entry corresponds to a surgical trajectory planned/evaluated
expected risk volume voxels that intersect the trajectory. by surgeon i on target j with method k. Each entry includes
Which risk function best represents the actual risk associ- the individual measurements (risk score, minimum distance
ated with each trajectory is a key open question. We believe to blood vessels, and minimum distance to other structures).
that any function that combines the multiple factors into a
single risk number is deemed to only partially reflect the real Data analysis methodology
risk. Thus, it should be taken as a recommendation only, and
a careful examination of the images is still necessary. The We aggregate the study measurements data and report them
function we present in this study was carefully designed to in five categories as follows.
best represent the risk associated with each trajectory. Specif-
ically, we selected α = 1 such that the maximal risk value 1. By surgeon and by patient: we analyze the performance
is 1. Setting an upper bound to the risk value is important for each patient and each surgeon for each trajectory plan-
for interpretation of the risk value and for its visualization. ning method. We report the aggregate expected risk score
Moreover, it does not lead to the misclassification of safe and the minimum distance from the blood vessels for each
or high-risk trajectories as suggested by the reviewer. For surgeon, patient, and method. We also analyze the perfor-
example, assume that a chosen trajectory is 2 mm from a mance improvement for each patient and for each type of
blood vessel (risk value of 0.4) and assume that the trajec- surgeon for the segmentation and automatic method with
tory is far from the other structures such that they can be respect to the manual method.
neglected. In this case, the trajectory has a trajectory risk 2. By surgeon only: we report the performance of each
value of 0.2 (= 0.4/2). Therefore, trajectories whose risk surgeon on the average patient for each method. We
value is 0.2 or greater may present a high risk. report the mean expected risk score and the mean min-
imum distance from the blood vessels for all eight
Surgical trajectories measurements patients. We also report these measures by the type of sur-
geon (expert/trainee) and report the expected risk score
We define individual measurements to quantify the expected improvements by surgeon for the manual/visualization
risk of a surgical trajectory and relative measurements to and manual/automatic methods.
compare two surgical trajectories. The individual measure- 3. By patient and target: we analyze the performance of the
ments for a surgical trajectory are as follows: (1) its expected average surgeon for each patient and for each trajectory
risk score (0 no risk, 1 maximal risk); (2) its minimum dis- planning method. We report the mean expected risk score
tance from blood vessels (in millimeters); and (3) its min- and the mean minimum distance from the blood vessels
imum distance from structures other than blood vessels (in for all eight patients. We also report these measures by the

123
Int J CARS (2015) 10:1127–1140 1135

Table 2 Manual method—risk scores (above) and minimum distance from blood vessels (below) for each surgeon and for each patient
Risk value (mm) P1 P2 P3 P4 P5 P6 P7 P8 Avg (SD)

Exp1 0.13 0.34 0.28 0.32 0.46 0.35 0.55 0.19 0.33 (0.14)
0.0 1.7 2.7 0.7 0.0 0.0 0.0 2.8 1.0 (1.2)
Exp2 0.09 0.47 0.17 0.21 0.60 0.37 0.36 0.02 0.29 (0.20)
0.6 6.1 0.0 3.3 0.0 0.0 0.0 0.0 1.3 (2.3)
Tra1 0.12 0.82 0.39 0.62 0.58 0.56 0.45 0.04 0.45 (0.26)
0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.5 0.2 (0.5)
Tra2 0.15 1.00 0.58 0.42 0.46 1.00 0.61 0.10 0.54 (0.34)
0.4 0.0 0.0 0.1 0.1 0.0 0.0 0.0 0.1 (0.1)
Tra3 0.15 0.60 0.53 0.26 0.48 1.00 0.10 0.18 0.41 (0.30)
0.0 4.9 0.0 0.0 0.0 0.0 1.0 0.0 0.7 (1.7)
Experts (SD) 0.11 (0.03) 0.41 (0.09) 0.22 (0.07) 0.27 (0.08) 0.53 (0.10) 0.36 (0.01) 0.46 (0.14) 0.10 (0.12) 0.31 (0.16)
0.3 (0.4) 3.9 (3.1) 1.4 (1.9) 2.0 (1.8) 0.0 (0.0) 0.0 (0.0) 0.0 (0.0) 1.4 (2.0) 1.1 (1.4)
Trainees (SD) 0.14 (0.02) 0.81 (0.20) 0.50 (0.10) 0.43 (0.18) 0.51 (0.07) 0.85 (0.26) 0.39 (0.26) 0.10 (0.07) 0.47 (0.27)
0.1 (0.2) 1.6 (2.8) 0.0 (0.0) 0.0 (0.1) 0.0 (0.1) 0.0 (0.0) 0.3 (0.6) 0.5 (0.9) 0.3 (0.6)
All surgeons (SD) 0.13 (0.03) 0.65 (0.26) 0.39 (0.17) 0.37 (0.16) 0.52 (0.07) 0.66 (0.32) 0.42 (0.20) 0.10 (0.08) 0.40 (0.21)
0.2 (0.3) 2.5 (2.8) 0.5 (1.2) 0.8 (1.4) 0.0 (0.0) 0.0 (0.00) 0.2 (0.5) 0.9 (1.3) 0.7 (0.84)
The last three rows show the scores for the experts, for the trainees, and for all surgeons

type of patient/target. We report the expected risk score
improvement by patient for the manual/visualization and
manual/automatic methods.
4. Structures other than blood vessels: we analyze the effect
that each trajectory planning method has on the minimum
distance between the planned trajectories and the struc-
tures and functional regions other than blood vessels.
5. Selection of surgical trajectories: we analyze the influ-
ence of the trajectory planning methods on the surgeons’
choice of surgical trajectory. We report how many times
the trajectory was changed as a result of using a partic-
ular planning method. We also analyze the variability of
the selected trajectories by surgeon, patient, and trajec-
tory planning method. The distance between their entry
points determines the variability between two surgical
trajectories.
Software prototype
We have developed a preoperative trajectory planning soft-
ware prototype for the purposes of the study. It is based on
the medical imaging interaction toolkit (MITK), which inte-
grates the Insight Toolkit (ITK) and the Visualization Toolkit
(VTK) [33–35] on a PC running Windows 7. The software
prototype includes the following functions: (1) visualization
of scans, segmentation, color-coded trajectories, and color-
coded candidate entry points on 2D axial, coronal, and sagit-
tal slices and in 3D; (2) interactive selection of targets and
candidate entry points regions; and (3) minimum distance
from structures and trajectory risk score computation. Par-
aview [36] was used for the candidate entry point’s region
definition. The data analysis was performed with MATLAB
[37].
Results
Tables 2 and 3 report the results for each surgeon and each
patient, the performance of the manual and the automatic
methods, respectively (first five rows). They also report the
results for the expert and trainee groups, and for all surgeons
(following three rows). We report the performance of each
surgeon and type of surgeon on the average patient/target for
each trajectory planning method (last column).
By surgeon and by patient
The expected risk score and minimum distance from blood
vessels improvements from the manual to the automatic
method are 75.7 % (SD = 11.6 %, p < 0.05, two-sample
student’s t test) and 1.6 mm (SD = 1.3 mm, p < 0.05). The
corresponding improvements from the manual to the visu-
alization method are 32.1% (SD = 23.3 %, p < 0.05) and
0.1 mm (SD = 0.46 mm, p = 0.56).
By surgeon
The expected risk score and minimum distance from blood
vessels improvements from the manual to the automatic
method for experts are 78.8 % (SD = 18.4 %, p < 0.05) and
1.2 mm (SD = 1.4 mm, p = 0.13), respectively; for trainees,

123
1136 Int J CARS (2015) 10:1127–1140

Table 3 Automatic method—risk scores (above) and minimum distance from blood vessels (below) for each surgeon and for each patient
Risk values (mm) P1 P2 P3 P4 P5 P6 P7 P8 Avg (SD)

Exp1 0.08 0.01 0.01 0.16 0.00 0.13 0.05 0.00 0.06 (0.06)
0.1 3.7 3.0 0.4 2.6 0.6 1.6 2.8 1.9 (1.4)
Exp2 0.03 0.15 0.17 0.18 0.00 0.36 0.05 0.02 0.12 (0.12)
1.3 3.6 5.0 1.3 2.6 3.0 1.6 3.6 2.8 (1.3)
Tra1 0.05 0.15 0.01 0.18 0.00 0.13 0.06 0.04 0.08 (0.07)
0.1 3.7 5.2 1.5 2.6 3.0 1.6 1.5 2.4 (1.6)
Tra2 0.00 0.65 0.10 0.19 0.00 0.24 0.07 0.05 0.16 (0.22)
1.3 0.1 3.0 1.5 2.6 1.0 1.1 2.7 1.7 (1.0)
Tra3 0.12 0.15 0.04 0.11 0.00 0.04 0.03 0.00 0.06 (0.06)
0.1 2.8 5.3 0.2 2.6 3.8 1.5 3.6 2.5 (1.8)
Experts (SD) 0.05 (0.03) 0.08 (0.10) 0.09 (0.12) 0.17 (0.01) 0.00 (0.00) 0.25 (0.16) 0.05 (0.00) 0.01 (0.01) 0.09 (0.08)
0.7 (0.9) 3.7 (0.1) 4.0 (1.4) 0.8 (0.7) 2.6 (0.0) 1.8 (1.7) 1.6 (0.0) 3.2 (0.6) 2.3 (1.3)
Trainees (SD) 0.06 (0.06) 0.32 (0.29) 0.05 (0.05) 0.16 (0.05) 0.00 (0.00) 0.14 (0.10) 0.06 (0.02) 0.03 (0.02) 0.10 (0.10)
0.5 (0.7) 2.2 (1.9) 4.5 (1.3) 1.1 (0.7) 2.6 (0.0) 2.6 (1.4) 1.4 (0.3) 2.6 (1.1) 2.2 (1.2)
All surgeons (SD) 0.06 (0.05) 0.22 (0.25) 0.07 (0.07) 0.17 (0.03) 0.00 (0.00) 0.18 (0.12) 0.05 (0.01) 0.02 (0.02) 0.10 (0.08)
0.6 (0.7) 2.8 (1.5) 4.3 (1.2) 1.0 (0.6) 2.6 (0.0) 2.3 (1.4) 1.5 (0.2) 2.8 (0.9) 2.2 (1.2)
The last three rows show the scores for the experts, for the trainees, and for all surgeons

the improvements are 87.1 % (SD = 3.7 %, p < 0.05) and
1.9 mm (SD = 1.4 mm, p < 0.05), respectively. The average
corresponding improvements from the manual to the visual-
ization method for experts are 18.3 % and 0.0 mm, respec-
tively; for trainees, the improvements are 49.8 % and 0.2 mm,
respectively.
By patient and target
The expected risk score improvements from the manual to
the automatic method was significant ( p < 0.05) and ranged
from 0.1 (SD = 0.08) to 0.02 (SD = 0.02) for patient
P8, an improvement of 400 %, and 0.66 (SD = 0.32) to
0.18 (SD = 0.12) for patient P6, an improvement of 266 %.
For minimum distance from blood vessels the improvements
were also significant ( p < 0.05) and ranged from 2.5 mm
(SD = 2.8 mm) to 2.8 mm (SD = 1.5 mm).
Distance from eloquent brain areas and the ventricular
system
For the manual method, the minimum distance from function-
related areas is 16.7 mm (SD = 4.9)—motor activation
areas 16.9 mm, somatosensory activation areas 10.5 mm,
speech-related areas 22.5 mm, and visual activation areas
16.7 mm; for the white matter fibers tractography, it is 8.2 mm
(SD = 2.8 mm)—pyramidal tract 9.7 mm, somatosensory
fibers 10.4 mm, superior longitudinal fasciculus 8.4 mm, and
optic radiation 4.1 mm. The minimum distance from the ven-
tricles is 12.6 mm (SD = 7.19 mm).
For the automatic method, the minimum distance from
function-related areas increased to 20.1 mm (SD = 4.5 mm)
—motor activation areas 23.2 mm, somatosensory activa-
tion areas 14.8 mm, speech-related areas 24.3 mm, and visual
activation areas 18.1 mm. For white matter fibers tractogra-
phy, the minimum distance increased to 10.5 mm (SD =
3.6 mm)—pyramidal tract 13.1 mm, somatosensory fibers
12.9 mm, superior longitudinal fasciculus 10.6 mm, and optic
radiation 5.4 mm. The minimum distance from the ventri-
cles increased to 10.1 mm (SD = 5.9 mm). The overall
improvement from the manual to the automatic methods is
3.4 mm ( p = 0.34) for the function-related areas and 2.3 mm
( p < 0.05) for the white matter fibers tractography.
Selection of surgical trajectories
We report on the change of the selected surgical trajec-
tory between methods: From the manual to the visualization
method, 19 out of 40 (47.5 %) trajectories were changed, and
from the visualization to the automatic method 34 out of 40
(85 %) were changed. The expected risk score improvement
increased more when trajectories were changed (77.9 %) than
when no changes were made (58.6 %). Of the trajectories
that did change, the average distance between entry points
is 11.6 mm (SD = 5.1 mm, p < 0.05) for manual to visu-
alization, 18.3 mm (SD = 9.6 mm, p = 0.28) for visual-
ization to automatic. For the experts, the mean entry point
distance changes were 7.1 and 10.5 mm for the visualization
and automatic method, respectively, and 14.6 and 12.5 mm
for the trainees, respectively.

123
Int J CARS (2015) 10:1127–1140
Discussion
Our results support our initial hypotheses and shed light on
closely related issues. We discuss them next.
Minimum distance from blood vessels
The minimum distance from blood vessels and the aggre-
gate expected risk score may be good measures to quantify
the expected risk of a surgical trajectory. The initial surgical
trajectories were changed in 85 % of the cases based on the
expected risk score and the minimum distance from blood
vessels. Since the final surgical trajectories were all vali-
dated and retained after the selection, this suggests that the
measures capture the concept of expected risk appropriately.
In addition, our study results show that the most important
structures in surgical trajectory planning are the blood ves-
sels. While functional areas and white matter fibers tractog-
raphy information can be useful in some cases, the expected
risk of most surgical trajectories is dominated by the proxim-
ity to blood vessels. Still, the automatic method decreased the
expected risk score and increased the minimum distance from
blood vessels in all cases. These observations are consistent
with recently published studies on deep brain stimulation
[5,7,9–11,21] and stereo-electroencephalography [20].
Conventional method evaluation
For the manual (conventional) method, significant differ-
ences in the selection of the entry point and its associated
surgical trajectory were found. The variability of the associ-
ated expected risk score across surgeons and targets inside
the brain suggests that there is sometimes a lack of consensus
between surgeons as to what constitutes the safest surgical
trajectory. We found significant differences ( p < 0.05, stu-
dent’s two-sample t test) in the manual (initial) expected risk
scores and the minimum distance from blood vessels between
experts (0.31 and 1.1 mm) and trainees (0.46 and 0.3 mm).
Surgical trajectory variability
We observe that the location of the surgical trajectories varies
from patient to patient. For some patients, the manual (ini-
tial) planned trajectory is nearly the same across surgeons
(patient P1, risk score SD = 0.03, minimum distance from
blood vessels SD = 0.3 mm). For others, the differences are
significant (patient P2, risk score SD = 0.3, patient P2, mini-
mum distance from blood vessels SD = 2.8 mm). The extent
to which the safety of a surgical trajectory can be improved
also varies between patients. In some cases, the improve-
ment is only modest—patient P8, risk score improvement
from 0.1 to 0.02, and for patient P6, minimum distance from
blood vessels improvement 2.5–2.8 mm, which is not signifi-
1137
cant in the clinical sense since the initial trajectory is already
very safe. However, in others, the improvement is of clinical
importance and otherwise was not identified by any of the
five surgeons. For example, the average minimum distance
from blood vessels is 0.0 mm (SD = 0.0) for patient P6 for
the manual and the visualization methods. It is increased to
2.3 mm with the automatic method—this means that in some
cases it may be difficult even for an experience neurosurgeon
to identify blood vessels along the planned trajectory, and by
using the software, potential injury to blood vessels might be
avoided.
Utility of computer-assisted trajectory planning
There is a consistent and significant reduction ( p < 0.05)
of the expected risk score and its variability and of the min-
imum distance to blood vessels and its variability for the
two advanced computer-assisted trajectory planning meth-
ods with respect to the manual method. The expected risk
scores and minimum distance from blood vessels improved
from 0.24 and 0.7 mm, for the visualization method, to 0.06
and 2.2 mm for the automatic method, for an overall improve-
ment of 75.7 % and 2.2 mm with respect to the manual selec-
tion method. There was no significant difference between the
automatic (final) expected risk score and minimum distance
from blood vessels for experts and trainees: expected final
risk score and minimum distance to blood vessels of 0.09
(SD = 0.08) and 2.3 mm (SD = 1.26) for experts, and 0.10
(SD = 0.10) and 2.2 mm (SD = 1.2) for trainees. In all
cases, the automatic computation of the three safest trajec-
tories consistently outperforms the expected risk of surgical
trajectories selected manually by all surgeons.
Experts versus trainees
The improvements were more significant for trainees than
for experts. The expected risk score and minimum distance
from blood vessels improvement between the manual and
the visualization method was modest for experts, 18.3 % and
larger for trainees, 49.8 %. However, the improvement from
the manual to the automatic method was 78.8 % and 1.2 mm
for experts, and 87.1 % and 1.9 mm for trainees.
The main difference between experts and trainees appears
to be the ability to identify and mentally reconstruct the 3D
structure of the blood vessels—this is reflected in the differ-
ence of improvement in the expected risk score from manual
to the visualization methods in each group. In some cases,
it is the delineation of the blood vessels that eliminates the
blood vessel penetration. In others, it is the quantification of
the expected risk score that makes the significant difference
(patient P2, the expected risk score decreased from 0.48 to
0.23). Finally, in other cases, the automatic method yields a
significantly different safest surgical trajectory (patient P5,
123
1138
minimum distance from blood vessels improved from 0.2 to
1.5 mm).
Risk quantification
Currently, there is no consensus among neurosurgeons as
to weather the expected risk can be quantified. Indeed, the
neurosurgical considerations involved in the choice of the
surgical trajectory are vast and complex, including the sever-
ity of a deficit, the likelihood of the permanence of a deficit,
the expected quality of life of the patient after surgery, and
many more. Thus, there is no “gold standard” to which the
results of computer-aided surgical trajectory planning can be
compared. However, there is consensus that the larger the
minimum distance from the blood vessels and other struc-
tures or areas is, the safer is the surgical trajectory. Thus,
all the published methods to date quantify risk as weighted
inversely proportional to the distances from the structures,
although there is no consensus among neurosurgeons on the
expected risk score function and weights. Clearly, different
weights will lead to somewhat different results. Therefore,
we report all our results both in terms of the expected risk
scores and the minimum distance from the blood vessels and
other structures or areas. Our assumptions should be assessed
clinically by future utilization of the automated technique in
a series of cases and compared to a control cohort prior to its
implementation.
The limitations of our study are as follows.
1. Study size: the number of patients and targets is rela-
tively small (eight), while the number of neurosurgeons
is acceptable (two experts, three trainees). A larger study
size would allow for more definitive conclusions. How-
ever, the patients and targets were carefully chosen to
represent a wide spectrum of clinical scenarios; also, the
improvements were significant enough to provide a basis
for the hypotheses.
2. Segmentations: our methods depend on the availability
and quality of segmentation corresponding to the head
surface, ventricles, blood vessels, high-density white
matter fibers tractography, and functional areas delin-
eation. Automatic methods for head surface and ven-
tricles segmentation are readily available and yield reli-
able results. However, the segmentation of blood vessels,
especially small ones, from MRI scans is more challeng-
ing. It usually requires a semiautomatic segmentation
method and surgeon validation, which may be error prone
and time consuming [21]. Moreover, it may be impossi-
ble to extract blood vessels with a diameter that is smaller
than the image resolution. Therefore, some of the risk of
hemorrhage in intracranial stereotactic procedures may
be unaccounted for. However, because of the branching
anatomy of the intracranial blood vessels, it is likely that
123
Int J CARS (2015) 10:1127–1140
most of the smaller blood vessels (<1 mm) will reside
near the larger one (>1 mm) in comparison with other
areas. Therefore, a trajectory that is as far as possible
from a large blood vessel is probably also as far as pos-
sible from brain areas with dense smaller blood vessels.
3. Brain shift: our surgical trajectory planning methods, as
well as all others [5,7,9,10,20,21], assume that the brain
structures are rigid and do not deform as the surgical
instrument is inserted. This is of course an idealization,
as the brain tissue deformation can either bring structures
closer to the instrument or further away from it.
4. Accuracy: the clinically meaningful trajectory planning
accuracy depends on the MR scan resolution, the accu-
racy of the scans co-registration, the intraoperative reg-
istration accuracy, and the accuracy of the insertion
method (stereotactic, navigated, robotic). Published stud-
ies including ours [6,20,21,30] indicate that the thresh-
old is 0.5–1.0 mm. Thus, trajectory safety improvements
below 0.5–1 mm will most likely have no practical mean-
ing in clinical practice. Still, the method serves as an addi-
tional validation to increase the surgeon’s confidence in
the planned trajectory and an automated “second opin-
ion” for the expected risk.
5. Validation: our study, as well as all those published in
the literature [20,21,30], are retrospective and do not
directly assesses the clinical benefit of computer-assisted
surgical trajectory planning. In the future, a controlled
prospective blind study should be conducted to evaluate
whether, indeed, the trajectories selected by our or similar
computer-based methods resulted in less complications
and better surgical outcomes. A retrospective postoper-
ative comparison between the actual and planned trajec-
tories may provide preliminary insights. For example, if
actual trajectories that were closer to the automatically
planned ones can be associated with better outcomes,
this would be strong evidence in favor of the automated
methods.
Conclusion
Our study indicates that computer-based surgical trajectory
planning may play an important role in improving the safety
of image-based stereotactic neurosurgery. The study indi-
cates that the computation of the minimum distance from
blood vessels and other eloquent structures may provide a
valuable tool for the quantification of the expected surgical
risk of hemorrhage and neurological damage due to incorrect
tool placement.
Our study shows that for certain patients and targets, there
are significant differences in the manual selection of the entry
point and its associated surgical trajectory. The differences
most likely stem from the surgeon’s experience and ability to
Int J CARS (2015) 10:1127–1140
identify and spatially visualize the proximity of the planned
trajectory to critical brain structures. The study shows that
there is a significant and consistent reduction of the expected
risk score and its variability and of the minimum distance
to blood vessels and eloquent brain areas and its variability
for the two computer-assisted surgical trajectory planning
methods with respect to the manual method. This may allow
novice and expert surgeons alike to consistently select the
safest surgical trajectory.
Acknowledgments Neurosurgeons Zvi Israel, Guy Rosenthal, Iddo
Paldor, Fernando Ramirez and Shweiki Moatasim participated in this
study.
Conflict of interest None of the authors has any conflict of interest.
The authors have no personal financial or institutional interest in any of
the materials, software, or devices described in this article.
References
1. Grossman R, Sadetzki S, Spiegelmann R, Ram Z (2005) Hem-
orrhagic complications and the incidence of asymptomatic bleed-
ing associated with stereotactic brain biopsy. Acta Neurochirurgica
(Wien) 147:627–631
2. Kongham P, Knifed E, Tamber M, Bernstein M (2008) Complica-
tions in 622 cases of frame-based stereotactic biopsy, a decreasing
procedure. Can J Neurol Sci 35:79–84
3. MacGirt M, Woodworth G, Coon A, Frazier J, Amundson E,
Garonzik I, Olivi A, Weingart J (2005) Independent predictors
of morbidity after image-guided stereotactic brain biopsy: a risk
assessment of 270 cases. J Neurosurg 102:897–901
4. Zrinzo L, Foltynie T, Haritz M (2012) Reducing hemorrhagic com-
plications in functional neurosurgery: a large case series and sys-
tematic literature review. J Neurosurg 116:84–94
5. Beriault S, Subaie FA, Collins DL, Sadikot AF, Pike GB (2012) A
multi-modal approach to computer-assisted deep brain stimulation
trajectory planning. Int J Comput Aid Radiol Surg 7(5):687–704
6. Bick AS, Mayer A, Levin A (2012) From research to clinical prac-
tice: Implementation of functional magnetic imaging and white
matter tractography in the clinical environment. J Neurol Sci
312:158–165
7. Lee JD, Huang CH, Lee LT (2002) Improving stereotactic surgery
using 3-D reconstruction. IEEE Eng Med Biol Mag 21:109–116
8. Navkar NV, Tsekos NV, Stafford JR, Weinberg JS, Deng Z (2010)
Visualization and planning of neurosurgical interventions with
straight access. In: Navab N, Jannin P (eds) Proceedings 1st interna-
tional conference on information processing in computer-assisted
interventions. Lecture Notes in Computer Science 6135. Springer,
Berlin, pp 1–11
9. Brunenberg EJ, Vilanova A, Visser-Vandewalle V, Temel Y, Ack-
ermans L, Platel B (2007) Automatic trajectory planning for deep
brain stimulation: a feasibility study. In: Ayache N, Ourselin S
(eds) Proceedings 10th international conference on medical image
computation and computer assisted intervention. Lecture Notes in
Computer Science 4791. Springer, Berlin, pp 584–592
10. Essert C, Haegelen C, Lalys F, Abadie A, Jannin P (2012) Auto-
matic computation of electrode trajectories for Deep Brain Stim-
ulation: a hybrid symbolic and numerical approach. Int J Comput
Aid Radiol Surg 7(4):517–532
11. Liu L, Dawant, BM, Pallavaram S, Neimat JS, Konrad PE, D’Haese
P, Datteri RD, Landman BA, Noble JH (2012) A surgeon specific
1139
automatic path planning algorithm for deep brain stimulation. Proc
SPIE Conf 8316:83161A–83161D.
12. Shamir RR, Joskowicz L, Antiga L, Foroni RI, Shoshan (2010) Tra-
jectory planning method for reduced patient risk in image-guided
neurosurgery: concept and preliminary results. Proc SPIE Conf
7625:762520I–7625824
13. Shamir RR, Joskowicz L, Tamir I, Dabool E, Pertman L, Ben-Ami
A, Shoshan Y (2012) Reduced risk trajectory planning in image-
guided keyhole neurosurgery. Med Phys 39(5):2885–2895
14. Vaillant M, Davatzikos C, Taylor RH, Bryan RN (1997) a path-
planning algorithm for image-guided neurosurgery. In: Troccaz LJ,
Mosges R (eds) Proceedings of 1st joint conference on computer
vision, virtual reality and robotics in medicine and medical robotics
and computer-assisted surgery. Lecture Notes in Computer Science
1205. Springer, Berlin, pp 467–476
15. De Momi E, Caborni C, Cardinale F, Casaceli G, Castana L, Cossu
M, Mai R, Gozzo F, Francione S, Tassi L, Lo Russo G, Antiga L,
Ferrigno G (2014) Multi-trajectories automatic planner for Stere-
oElectroEncephaloGraphy (SEEG). Int J CARS (2014) 9:1087–
1097. doi:10.1007/s11548-014-1004-1
16. Liu Y, Konrad P, Neimat J, Tatter S, Yu H, Datteri R, Landman
B, Noble J, Pallavaram S, Dawant B, DHaese PF (2014) Multi-
surgeon, multi-site validation of a trajectory planning algorithm
for deep brain stimulation procedures. IEEE Trans Biomed Eng
61(9):2479–2487. doi:10.1109/TBME.2014.2322776
17. Zelmann R, Beriault S, Mok K, Haegelen C, Hall J, Pike GB,
Olivier A, Collins DL (2014) Automatic optimization of depth
electrode trajectory planning. In: Clinical image-based procedures
translational research in medical imaging, pp 99–107
18. Rincon-Nigro M, Navkar NV, Tsekos NV, Zhigang D (2014) GPU-
accelerated interactive visualization and planning of neurosurgical
interventions. IEEE Comput Graph Appl 34(1):22–31
19. Shamir R R, Horn M, Blum T, Mehrkens J H, Shoshan Y, Joskowicz
L and Navab N (2011) IEEE International Symposium on Biomed-
ical Imaging (ISBI) Trajectory planning with Augmented Real-
ity for improved risk assessment in image-guided keyhole neuro-
surgery, pp 1873–1876
20. De Momi E, Carbonary C, Cardinale F, Castana L, Casaceli G,
Cossu M, Antiga L, Ferrigno G (2013) Automatic trajectory plan-
ner for stereo electro encephalography procedures: a retrospective
study. IEEE Trans Biomed Eng 60:986–993
21. Zombori G, Rodinov M, Nowell M, Zuluaga MA, Clarkson MJ,
Micallef C, Diehl B, Wehner T, Miserochi A, McEnvoy AW, Dun-
can JS, Ourselin, S (2014) A computer assisted planning system
for the placement of sEEG electrodes in the treatment of epilepsy.
In: Stoyanov D et al (eds) Proceedings of 5th international confer-
ence on information processing in computer-assisted interventions,
IPCAI 2014, Fukuoka, Japan. Lecture Notes in Computer Science
8498, pp 118–127.
22. Berthold KP (1987) Closed-form solution of absolute orientation
using unit quaternions. J Opt Soc Am 4(4):629–642
23. Wells WM III, Viola P, Atsumi H, Nakajima S, Kikinis R (1996)
Multi-modal volume registration by maximization of mutual infor-
mation. Med Image Anal 1(1):35–51
24. Shamir RR, Freiman M, Joskowicz L, Spektor S, Shoshan Y (2009)
Surface-based facial scan registration in neuronavigation proce-
dures: a clinical study. J Neurosurg 111(6):1201–1206
25. Lorensen WE, Cline HE (1987) Marching cubes: a high resolu-
tion 3D surface construction algorithm. Comput Graph 21(4):163–
169
26. Yushkevich PA, Piven J, Hazlett HC, Smith RG, Ho S, Gee JC,
Gerig G (2006) User-guided 3D active contour segmentation of
anatomical structures: significantly improved efficiency and relia-
bility. NeuroImage 31(3):1116–1128
27. Freiman M, Joskowicz L, Broide N, Natanzon M, Nammer E,
Shilon O, Weizman L, Sosna J (2012) Carotid vasculature modeling
123
1140 Int J CARS (2015) 10:1127–1140

from patient CT angiography studies for interventional procedures
simulation. Int J Comput Aid Radiol Surg 7(2):799–812
28. MrVista: MATLABR interface for analyzing functional and
anatomical data. http://white.stanford.edu/newlm/index.php/Mr
Vista. Accessed 18 June 2014
29. ConTrack: a probabilistic fiber-tracking algorithm. http://white.
stanford.edu/newlm/index.php/ConTrack. Accessed 18 June 2014
30. Basser PJ, Pajevic S, Pierpaoli C, Duda J, Aldroubi A (2000) In vivo
fiber tractography using DT-MRI data. Magn Reson Med 44:625–
632
31. BrainVoyager QX: analysis and visualization of functional and
structural magnetic resonance imaging data sets. http://www.
brainvoyager.com. Accessed 18 Feb 2014
32. Friston KJ, Frith C, Frackowiak RSJ, Turner J (1995) Characteriz-
ing dynamic brain responses with fMRI: a multivariate approach.
NeuroImage 2:166–172
33. ITK: The Insight Segmentation and Registration Toolkit. http://
www.itk.org. Accessed 18 June 2014
34. MITK: The Medical Imaging Interaction. http://www.mitk.org.
Accessed 18 June 2014
35. VTK: The Visualization Toolkit. http://www.vtk.org. Accessed 18
June 2014
36. ParaView: Open Source Scientific Visualization. http://www.
paraview.org. Accessed 18 June 2014
37. MATLABR : Technical Computing. http://www.mathworks.com.
Accessed 18 June 2014

123