Figure 1. 3-Dimensional images. A,
Rendering overlaid on computed tomographic images. B, Virtual neck
exploration highlighting a left inferior parathyroid adenoma.
Figure 2. 3-Dimensional rendering applied
externally for initial operative localization. Patient in the standard
operative position (A), with external augmented reality overlaid (B).
C, Augmented reality images delivered to monitors in the operating
Figure 3. 3-Dimensional rendering applied
through the videoscope for precise operative localization (A and B).
D’Agostino J, Wall J, Soler L, Vix M, Duh Q, Marescaux J. Virtual Neck Exploration for Parathyroid AdenomasA First Step Toward Minimally Invasive Image-Guided Surgery. JAMA Surg. 2013;148(3):232-238. doi:10.1001/jamasurg.2013.739
Authors Affiliations: IRCAD/Institut
Hospitalo Universitaire, Strasbourg, France (Drs D’Agostino,
Wall, Soler, Vix, and Marescaux); University of California, and VA
Medical Center, San Francisco, California (Dr Duh).
Objective To evaluate the performance of 3-dimensional (3D) virtual neck
exploration (VNE) as a modality for preoperative localization of parathyroid
adenomas in primary hyperparathyroidism and assess the feasibility
of using augmented reality to guide parathyroidectomy as a step toward
minimally invasive image-guided surgery.
Design Enhanced 3D rendering methods can be used to transform computed
tomographic scan images into a model for 3D VNE. In addition to a
standard imaging modality, 3D VNE was performed in all patients and
used to preoperatively plan minimally invasive parathyroidectomy.
All preoperative localization studies were analyzed for their sensitivity,
specificity, positive predictive value, and negative predictive value
for the correct side of the adenoma(s) (lateralization) and the correct
quadrant of the neck (localization). The 3D VNE model was used to
generate intraoperative augmented reality in 3 cases.
Setting Tertiary care center.
Patients A total of 114 consecutive patients with primary hyperparathyroidism
were included from January 8, 2008, through July 26, 2011.
Results The accuracy of 3D VNE in lateralization and localization was
77.2% and 64.9%, respectively. Virtual neck exploration had superior
sensitivity to ultrasonography (P < .001),
sestamibi scanning (P = .07), and
standard computed tomography (P < .001).
Use of the 3D model for intraoperative augmented reality was feasible.
Conclusions 3-Dimensional VNE is an excellent tool in preoperative localization
of parathyroid adenomas with sensitivity, specificity, and diagnostic
accuracy commensurate with accepted first-line imaging modalities.
The added value of 3D VNE includes enhanced preoperative planning
and intraoperative augmented reality to enable less-invasive image-guided
Primary hyperparathyroidism (PHPT) is the third most common
endocrine disorder globally. Symptomatic disease is routinely treated
with parathyroidectomy, and the revised National Institutes of Health
consensus guidelines from 2002 have further expanded the surgical
indications for patients with asymptomatic disease.1 The improved sensitivity of preoperative
localizing studies and the advent of intraoperative parathyroid hormone
(IOPTH) assays have enabled focused approaches. The treatment of localized
single adenomas can now be reliably performed with minimally invasive
approaches, avoiding a bilateral neck exploration when preoperative
studies are concordant.
The first minimally invasive parathyroidectomy (MIP) was performed
in 1996.2 Since then, most studies
have confirmed that MIP offers less pain, less cervical dissection,
decreased hospitalization, decreased cost, and improved patient satisfaction.3,4 Cure rates equal to classic
bilateral neck exploration have been achieved with various combinations
of preoperative localization studies and IOPTH assay.5
Multiple imaging modalities have been reported for the localization
of parathyroid adenomas including ultrasonography, Tc 99m sestamibi
scans, computed tomographic (CT) scans, and magnetic resonance imaging
(MRI). Ultrasonography and sestamibi scans are the most sensitive,
with reported values ranging from 57% to 89% and 54% to 84%, respectively.6
In the last decade, many high-volume centers have adopted the
combination of ultrasonography and sestamibi for initial localization
studies.7 Two concordant preoperative
imaging studies can lead to successful focused exploration in 79%
to 89% of cases, and overall cure rates with MIP are 95% to 97%.8- 10 Some studies
have found value in the use of IOPTH in all cases of focused exploration.11 Others have found minimal benefit in cases
of preoperative concordance, and they reserve the role of IOPTH for
discordant or negative preoperative imaging.12 The cost to benefit ratio of the IOPTH
assay remains questionable.13
Ultrasonography is limited in lower-volume centers by the need
for parathyroid-specific radiologic expertise owing to the high interoperator
variability inherent to the modality.14 Sestamibi is limited to centers that have the nuclear medicine
expertise, and results vary widely among institutions.15
Computed tomographic scanners are becoming ubiquitous in the
developed world owing to their versatility in both diagnostic and
interventional radiology, as well as the modality's relatively low
interoperator variability. Computed tomographic scan parathyroid localization
was first reported in the 1970s.16,17 Initial results were disappointing
for parathyroid adenoma localization; however, improved sensitivity
has been reported with thin-cut contrast-enhanced studies.18 Recently, several studies have examined
the use of 4-dimensional (4D) CT scanning.19- 22 This modality combines 3-dimensional (3D) reconstruction of CT
images with the added dimension of contrast diffusion over time. When
used in patients with PHPT, 4D CT scanning was found to have better
lateralization (88%) and localization (70%) than either ultrasonography
or sestamibi, which offered lateralization in 57% and 65% of cases,
respectively, and localization in 29% and 33% of cases, respectively.22 In patients with negative or discordant
standard preoperative localization studies, 4D CT was recently found
to accurately lateralize 73% and localize 60% of abnormal glands;
however, the accuracy dropped considerably for when multiple lesions
were seen.21 In localization for
reoperative parathyroidectomy, 4D CT was found to be more sensitive
than sestamibi.19 Concerns exist
surrounding radiation exposure based on linear extrapolations from
the known risk of high-dose radiation; however, no threshold for lower
doses of radiation has been established.23 At our institution, neck CT and sestamibi carry a similar radiation
exposure in the range of 2 to 6 mSv.
Our group has an ongoing interest in the use of enhanced 3D
rendering methods to create preoperative virtual models for surgical
planning. 3-Dimensional models provide excellent anatomic detail for
the purpose of surgical planning. The models can further be used during
the procedure to create an augmented reality with real-time image
overlay on both the exterior of the patient and on the laparoscopic
images. The technology has proven beneficial in the preoperative planning
of both adrenal and hepatic resections.24,25
In 2006, our group began to apply VR-Render–enhanced 3D
models to parathyroid surgery, enabling interactive preoperative virtual
neck explorations (VNEs) and intraoperative augmented reality.26 Given our experience in the rendering of
CT images for hepatic and adrenal surgical procedures, we began with
3D models generated from neck CT scans. The models provide the surgeon
with excellent color contrast to differentiate the critical structures
of the neck and highlight candidate parathyroid lesions (Figure 1). Individual layers of tissues,
including bone, muscle, arteries, and veins, can be viewed in any
combination to allow greater focus on the relevant anatomy. Using
a 3D model of the patient, the endocrine surgical team can perform
a 3D VNE to decide whether the anatomic location of the lesion is
consistent with a typical adenoma and choose the target lesion(s)
for the subsequent intervention. The model can further be used in
the operating room to provide augmented reality by overlaying images
on the exterior of the patient (Figure
2) or through the videoscopic image (Figure 3). The combination of preoperative
3D VNE with intraoperative augmented reality has the potential to
improve preoperative localization and provide intraoperative guidance.
The aim of this study was to evaluate the performance of 3D
VNE as a modality for preoperative localization of parathyroid adenomas
in PHPT and report the feasibility of augmented reality to guide parathyroidectomy
as a step toward minimally invasive image-guided surgery.
The study population consisted of all consecutive adult patients
with biochemically confirmed PHPT treated at an academic tertiary
referral center. From January 8, 2008, through July 26, 2011, 114
patients with PHPT underwent parathyroidectomy. This study covers
a period in which our institution transitioned to a minimally invasive–focused
surgical approach as described by Miccoli et al2,27,28 with IOPTH.
The study was performed and reported in accordance with the ethical
guidelines of the University of Strasbourg Hospital.
All patients underwent at least 1 preoperative imaging modality
before referral to our institution. Seventy-one patients underwent
ultrasonography, 64 had a sestamibi scan, and 36 had both. Every patient
subsequently had a single standard CT scan of the neck with intravenous
iodine–based contrast and 0.75-mm slices in cervical hyperextension
to simulate the operative positioning. The standard CT images were
first interpreted by a general radiologist, without specific expertise
in parathyroid imaging.
VR-Render software was then used on the same CT images to create
an enhanced 3D virtual model of the neck. The rendering process included
the visualization of multiplanar CT images (axial, sagittal, and coronal)
by a specialized radiology technician. The technician applied color
contrast to each layer of tissue in a semi-automatic fashion with
manual corrections for obvious rendering errors. Candidate lesions
were chosen and manually highlighted based on the criteria listed
in Table 1. All 3D renderings
were reviewed by a PhD expert with specialization in medical imaging
software in collaboration with the team of endocrine surgeons. This
team was blinded to the interpretation of the radiologist. The team
performed a preoperative VNE with the rendering and decided on the
likely location of the adenoma(s).
In cases of augmented reality, the 3D rendering was manually
registered to visible rigid landmarks in the neck in real time by
a medical imaging technician during the case. The surgeon was in direct
communication with the medical imaging team to fine-tune registration
and control the image contrast on the operating monitor.
Ultrasonography, sestamibi, standard CT, and 3D VNE were analyzed
for their sensitivity, specificity, positive predictive value, and
negative predictive value for the correct side of the adenoma(s) (lateralization)
and the correct quadrant of the neck (localization). For localization,
the neck was divided into 4 quadrants and each was analyzed independently
true positive, true negative, false positive, or false negative using
the imaging and operative pathology data. For lateralization, the
same analysis was conducted for each side of the neck. No cases of
ectopic glands were predicted by imaging or found at surgery, thus
ectopic position was not included in our analysis. Glands were considered
negative if they were either explored and deemed normal by the surgeon
or not explored with drop in IOPTH that met the Miami criteria.29
To characterize the clinical use of the studies, we performed
an analysis of diagnostic accuracy compared with operative findings
by side and quadrant for each patient. For this analysis of accuracy,
a true positive was a study that exactly localized or lateralized
the adenoma in a given patient. A study that missed any gland or called
any gland nonpathologic was false negative or false positive, respectively,
for that given patient.
Statistical analysis of contingency data was performed using
the epiR package for R (www.r-project.org). Fisher exact
test was used to compare the operating characteristics of different
The median age of the population was 60.8 years (range, 21-87
years), with a female to male ratio of 3.7:1.0. Fifty patients were
symptomatic. The preoperative mean serum calcium level was 11.5 mg/dL
(to convert to millimoles per liter, multiply by 0.25) and the preoperative
PTH level was 162.4 ng/L. A total of 112 patients underwent a primary
procedure and 2 patients underwent reoperation for recurrent PHPT
after previous bilateral neck explorations.
Among the 114 patients, 69 underwent planned MIP technique using
the gasless method through a median 3.0-cm skin incision. All MIP
patients had IOPTH drawn in the operating room and the Miami criteria
were used to determine the need for further exploration. Of these
cases, 20 were converted to bilateral neck exploration owing to either
failure to meet the Miami criteria or surgeon choice. Forty-six patients
underwent planned bilateral neck exploration for reexplorations (n = 2),
coexisting thyroid disease (n = 10), large adenomas (>3.5
cm; n = 3), and early in our experience with MIP to verify
the technique (n = 31). Among the 114 procedures, 316 parathyroid
glands were visualized, 132 glands were resected, and 129 glands were
pathologically confirmed adenomas. In 11 cases (9.6%), multiglandular
disease was discovered.
3-Dimensional VNE correctly localized single adenomas in 82
patients. In 5 cases, VNE correctly identified a double adenoma. In
11 cases, a second enlarged gland was suspected but only 1 adenoma
was found concordant at surgery. The mean size of the detected adenomas
was 18.8 mm (95% CI, 8.8-27.4) and for missed glands was 13.5 mm (95%
CI, 7.9-18.1). 3-Dimensional VNE missed a total of 29 abnormal glands
that were found in the operating room. Nine of the missed glands were
in the superior position and 20 were in the inferior position. There
were no missed glands found in an ectopic location.
The test characteristics with 95% confidence intervals of all
modalities for both lateralization and localization are shown in Table 2. For lateralization, VNE had
superior sensitivity to ultrasonography (P < .001), sestamibi (P < .001),
and standard CT (P < .001).
Specificity of VNE was similar to ultrasonography (P = .10) but less than sestamibi (P < .001) and standard CT (P < .001). The diagnostic accuracy of 3D VNE
in lateralization and localization was 77.2% and 64.9%, respectively
(Table 3). Diagnostic accuracy
of 3D VNE was equivalent to sestamibi (P = .16),
but significantly higher than both ultrasonography (P < .001) and standard CT (P = .003).
In 3 cases, augmented reality was applied during the operation.
A single adenoma was successfully identified using the interactive
real-time image overlay in all 3 patients.
The average follow-up was 15.6 months (range, 6-44 months).
Two patients were lost to follow-up. Of the remaining patients, the
cure rate was 98.8%, with only 1 case of persistent disease from the
2 patients who had undergone a reoperation for recurrent PHPT.
Surgery for PHPT has shifted from routine bilateral neck exploration
to focused, minimally invasive techniques enabled by preoperative
localization of parathyroid adenomas and IOPTH.30 In theory, our center could have performed
much fewer bilateral neck explorations with the preoperative localization
that 3D VNE provided. However, in practice, owing to a lack of rapid
IOPTH and initial surgeon experience with focused neck explorations,
our overall rate of bilateral neck exploration was high. Our rate
of bilateral neck exploration has decreased with increased experience.
The first-line studies for localization are routinely ultrasonography
and sestamibi, with MRI and CT scans usually reserved for cases of
negative or discordant studies. The advantage of planar imaging of
both CT and MRI is the visualization of all cervical structures, making
them particularly good for cases of ectopic glands. Recent developments
in 4D CT scans have shown promising results as a localizing study
with excellent sensitivity and specificity.
Computed tomography holds much appeal as it is becoming ubiquitously
available, is familiar to surgeons, and has little interoperator variability.
Using VR-Render software to generate a virtual neck model from CT
images, we performed 3D preoperative neck explorations prior to surgery.
3-Dimensional VNE significantly improved the sensitivity of standard
CT scans and resulted in significantly higher sensitivity than all
modalities from our referring institutions. Specificity of 3D VNE
was lower than CT and sestamibi. It would stand to reason that preoperative
planning by surgeons committed to operating owing to the biochemical
diagnosis might target a suspicious lesion more readily than a radiologist,
increasing the sensitivity and driving down the specificity. The overall
diagnostic accuracy of 3D VNE was better than referral center ultrasonography
and standard CT, making it a clinically useful tool.
Centers of excellence in endocrine surgery achieve high sensitivity
and specificity for preoperative ultrasonography and sestamibi. The
sensitivity and diagnostic accuracy of 3D VNE in this study is comparable
to the best results of ultrasonography, sestamibi, and 4D CT. In our
practice, the quality of ultrasonography and sestamibi from referral
centers is variable, making 3D VNE an attractive practical option.
One downside of CT and sestamibi is radiation exposure, although
an unsafe threshold dose is yet to be established. While currently
using CT images to generate the 3D model, a great advantage of 3D
VNE is that it has also been applied to MRI images, which eliminates
The limitations of 3D VNE are similar to other modalities in
that smaller adenomas are harder to detect accurately as well as adenomas
in the setting of concurrent thyroid nodules. Limitations specific
to 3D VNE are that it is based on the morphologic data from a single
contrast phase that does not offer functional information. Additionally,
the system is only semi-automatic, requiring manual rendering correction
and candidate lesion determination by a radiology technician. This
is currently labor intensive, demanding approximately 30 to 60 minutes
per study. Automated rendering of vascular structures has been very
consistent to date, and work is ongoing to reliably automate the entire
rendering process. Our institution plans to begin offering 3D VNE
to referring institutions through a web-based interface that would
allow DICOM images to be submitted for rendering.
The limitations of this study include uncontrolled initial preoperative
imaging modalities performed and interpreted at referring institutions.
Both ultrasonography and sestamibi have been shown to be user dependent.
Reinterpretation or even repetition of these studies at our institution
may have been ideal. However, the reality of our practice within a
nationalized health system did not allow for repeat studies as are
commonly performed elsewhere. Cases referred to a tertiary center
always present the potential for being more difficult than average;
the generally low sensitivity of the outside studies could reflect
such a bias. Furthermore, costs are difficult to compare within the
nationalized health system. There is no retail cost in our public
institution, as seen in the United States. The costs of imaging studies
are fixed by the Social Security system based primarily on the price
of disposable materials used. The charge for CT and ultrasonography
is approximately 100 euros, while the charge for sestamibi is approximately
Beyond preoperative localization, VR-Render was feasible for
real-time intraoperative augmented reality. Our experience of using
augmented reality is limited, but feedback from users is positive
and the technique has proven particularly useful in cases of aberrant
anatomy. While only used to date in conjunction with the MIP technique
of Miccoli et al, the overlay of anatomic information as seen in Figure 2 has potential application to
all surgical techniques. Limitations of the current augmented reality
system are that it requires manual registration to the intraoperative
image and has no compensation for patient movement (ie, respiration)
or tissue deformation (ie, retraction). Both automatic registration
and real-time compensation are being developed at our institution.
With such improvement, augmented reality could play a major role in
precision image-guided parathyroidectomy.
In conclusion, accurate preoperative localization of parathyroid
adenomas is critical in enabling the surgeon to perform MIP. VR-Render
is a novel software system that can generate an enhanced 3D virtual
neck model from multiple imaging modalities including CT and MRI.
3-Dimensional VNE with VR-Render has operating characteristics commensurate
with accepted first-line preoperative localizing study at our center.
The ubiquity and low interoperator variability of CT scanning makes
3D VNE attractive to centers that do not have the expertise or capability
in parathyroid ultrasonography, sestamibi, and single photon-emission
CT. The VR-Render platform offers the flexibility of incorporating
additional functional information through perfusion or spectroscopy
that could increase the specificity of 3D VNE. 3-Dimensional VNE can
also be performed with MRI imaging in place of CT scanning that would
eliminate radiation exposure. 3-Dimensional VNE is an excellent tool
in preoperative localization of parathyroid adenomas with the added
value of enabling preoperative planning and intraoperative augmented
reality. Precise image guidance is key to less-invasive surgical approaches
to PHPT and the potential to approach the treatment via endovascular
or extracorporeal methods.
Correspondence: Jacques Marescaux,
MD, FRCS(Hon), FJSES(Hon), IRCAD/Institut Hospitalo Universitaire,
1 Place de l’Hopital, 67091 Strasbourg, France (firstname.lastname@example.org).
Accepted for Publication: August 22,
Published Online: November 19, 2012.
Author Contributions: Drs D’Agostino
and Wall share first authorship of this study. Study
concept and design: D’Agostino, Wall, Soler, Vix, and
Marescaux. Acquisition of data: D’Agostino,
Soler, and Vix. Analysis and interpretation of data: D’Agostino, Wall, Vix, Duh, and Marescaux. Drafting of the manuscript: D’Agostino and Wall. Critical revision of the manuscript for important intellectual
content: D’Agostino, Wall, Soler, Vix, Duh, and Marescaux. Statistical analysis: D’Agostino and Wall. Administrative, technical, and material support: Soler and Vix. Study supervision: D’Agostino,
Soler, Vix, and Marescaux.
Conflict of Interest Disclosures: None