a ﬂat-panel detector (FPD; PaxScan 4030CB,
Varian, Palo Alto, CA), a motorized orbital drive,
geometric calibration, and a computer control sys-
tem for 3D reconstruction. CBCT acquisition
involved collection of projections across a rota-
tional arc of about 178
, nominally 200 projections
acquired in about 60 s (3.3 fps). The 3D ﬁeld of
view (FOV, 20 cm 20 cm 15 cm) is sufﬁcient
to encompass the skull base.
ated with CBCT imaging on the C-arm has been
: approximately 2.9 mGy
sufﬁcient for visualization of bone and soft tis-
sues, respectively. Such dose levels represent a
fraction (1/10–1/3) that of diagnostic CT.
Specimen Preparation and CBCT Imaging. Five
cadaver heads were acquired from the Division of
Anatomy, University of Toronto, in accord with
the Anatomy Act of Ontario. All heads were ini-
tially scanned on a 16-slice diagnostic CT scanner
(Discovery PET-CT; General Electric, Milwaukee,
WI) to review any anatomic variations and
excluded any anatomic abnormality or disease.
During the head preparation, each head under-
went a complete surgical clearance of the eth-
moid, maxillary, and sphenoidal sinuses. Air
space along the skull base (attributed to postﬁxa-
tion brain shrinkage) was ﬁlled with water-equiv-
alent gel via craniotomy. Subsequently, the
cranial bone ﬂap was reopposed and secured.
After preparation, a baseline CBCT scan was
taken. Each head was immobilized in a carbon-
ﬁber frame using 6 to 8 cranial pins, and was
supported at the C-arm isocenter for each scan.
Eight locations on the anterior skull base were
selected for introduction of skull base breach as
illustrated in Figure 1B: the (left and right) cri-
briform plate, fovea ethmoidalis, lateral lamella,
and planum. After the baseline scan, an experi-
enced skull base surgeon used a 1-mm-diameter
drill to introduce skull base defects into each
site, and a second CBCT scan was acquired.
Then each of the 1-mm defects was widened
using a 2-mm drill (2-mm defects are shown in
Figure 2), followed by a third CBCT scan.
Finally, all sites were drilled to a 4-mm diame-
ter, and a ﬁnal CBCT scan was acquired.
Visualization of Skull Base Defects: Observer Study
and Data Analysis.
The images were viewed in 3D
visualization software (3D Slicer v3.2; Brigham &
Women’s Hospital, Boston, MA, and Massachu-
setts Institute of Technology, Cambridge, MA).
The 3D coordinates of each defect were iden-
tiﬁed by an independent observer. For smaller
defects that were difﬁcult to identify (eg, 1 mm),
the coordinates of the large defects (4 mm) were
translated to the earlier scans, using anatomic
landmarks to account for possible displacement
of the specimen between scans. In all, 120
defects (5 heads 8 locations 3 diameters)
were thus localized.
For the observer studies described in the fol-
lowing text, image slices through each defect
were extracted from CBCT volume reconstruc-
tions using the ‘‘3D Slicer’’ software. These
included ‘‘orthogonal’’ slices that correspond to
conventional triplanar views (coronal, sagittal,
and axial) as well as ‘‘oblique’’ slices. The
oblique slices included both a coronal and a lon-
gitudinal slice (ie, a quasi-sagittal slice in the
plane of the drill) and a tangential slice (ie, a
quasi-axial slice in the plane of the defect).
For each defect, an area of interest was
cropped from t he image such that the defect
was presented in the center of the image. Slice
triples (either orthogonal or oblique) were dis-
played in comparison with the corresponding
‘‘baseline’’ area of interest (ie, the identical area
of skull base before introduction of a defect).
Example images are shown in Figures 2 and 3
for each anatomic location.
Observer studies were conducted to assess
visibility of skull base defects in each location as
a function of defect size. Five expert observers
participated, including 1 radiologist, 2 skull
base surgeons, and 2 head and neck surgeons.
The study was conducted under controlled view-
ing conditions: a darkened reading room on a
diagnostic workstation (Dell Precision 380, 3-
GHz Pentium 4 with dual-head Barco displays,
1536 2048 resolution, 8-bit grayscale; Dell
Inc., Round Rock, TX). A ﬁxed viewing distance
of 50 cm was recommended but not enforc ed.
Observers received identical instructions and
training. Reading order across 120 cases was in-
dependently randomized for each observer. For
each case, observers were shown (1) the full
FOV (nominal resolution) coronal and sagittal
images for purposes of orientation; (2) the base-
line images (orthogonal or oblique slices with no
defect); and (3) the defect images (orthogonal or
oblique slices with a defect). Observers were
then asked to rate their satisfaction in detect-
ing, localizing, and characterizing the defect on
a scale of 1 to 5 (1 ¼ Unable to identify any
breach, 2 ¼ The breach could be overlooked, 3 ¼
506 Intraoperative Cone-Beam CT HEAD & NECK—DOI 10.1002/hed April 2010