Advances in computer-aided detection (CAD) took center stage at
the Opening Session of the 87th Scientific Assembly and Annual
Meeting of the Radiological Society of North America (RSNA
2001).
Robert Schmidt, MD, of New York University School of Medicine,
told attendees at the Chicago meeting that he foresees a great
future for CAD as its ability to "improve performance and decrease
observer variability" becomes clearer. Noting that screening
mammography programs are threatened by economic difficulties and a
lack of trained personnel, he suggested that such programs are
ideally suited for automation through CAD, which should "function
as a second reader," similar to a spell checker in word-processing
programs.
Calling CAD "one of the most rapidly expanding areas of
radiology," Heber MacMahon, MD, of the University of Chicago
Hospitals focused on its usefulness in chest imaging. He stated
that the ability to enhance images through subtraction techniques
could lead to more effective detection and quantification.
Specifically, he explained how dual-energy chest radiography and
subtraction could be used to amplify abnormalities and produce more
accurate images. He noted that CAD is especially useful in the
identification of previously missed information and that it can
help reduce the number of false-positive findings. He concluded
that physician acceptance of these "powerful tools" should increase
when computer assistance clearly allows radiologists to evaluate
images more quickly.
CAD clinical studies
Several scientific studies presented at the meeting held
November 25-30, 2001, focused on the role of CAD in chest
radiography. Two of these are highlighted here: one that found that
the use of CAD software detected nearly 80% of previously missed
lung cancers on computed tomography (CT) scans and another that
tested an algorithm designed to reduce the number of false-positive
findings.
University of Chicago Study
Samuel G. Armato, III, PhD, of the University of Chicago presented
"Performance of Automated CT Lung Nodule Detection on Missed
Cancers." In this study, the researchers developed a computerized
method for the automatic detection of lung nodules on CT scans. The
program first divided the lungs into segmented lung volumes within
which the analysis was performed. Three-dimensional structures were
identified and, based on structure size, were considered as
potential nodules. A rule-based method was then used to eliminate
false-positive nodules and an automated classifier distinguished
the remaining structures as being either true nodules or normal
anatomy.
This program was applied to 38 low-dose CT scans with a total of
50 lung nodules. Of those, 38 were biopsy-confirmed lung cancers
that had not been found during the initial clinical screening. The
CAD system located 41 of the 50 nodules and 30 of the 38 missed
cancers, with a detection rate of 79% and a false-positive rate of
1.6 per section.
The researchers concluded that "such a computerized method may
aid radiologists in the identification of lung nodules and is
expected to play a key role in CT programs for lung cancer
screening."
Stanford University Study
A second study, conducted at Stanford University, Palo Alta, CA,
tested the effectiveness of a two-stage algorithm designed to
provide more accurate analysis. The algorithm is used to first
detect nodules based on a spheroidal or hemispheroidal shape model.
It then rejects most of the false-positive finding using a more
sophisticated shape analysis.
In this study, "Computer-aided Detection of Lung Nodules in CT:
Preliminary Results," presented by David S. Paik, MS, the
researchers tested the new algorithms on 304 axial CT images,
reconstructed at 1-mm intervals, of a 54-year-old man with
metastatic renal cell carcinoma. The images tested included 60
actual nodules, as confirmed by 3 radiologists, and 60
software-simulated nodules of similar size and morphology. The
nodules ranged in size from 1.3 to 13.3 mm, with 24% having pleural
contact and 30% having large vessel contact.
Overall, the CAD system found 90.8% of all nodules, with the
best results found for the largest nodules. The computer system
located all 71 actual nodules 6 mm and larger with only 1
false-positive finding. For nodules 3 to 6 mm in size, the CAD
algorithm achieved 92.5% sensitivity with 10 false-positive
findings. The researchers reported that "the second stage of the
algorithm led to a 30-fold reduction in false positives without
loss of sensitivity." They hypothesized that the remaining false
positives occurred due to pulmonary vessels, especially
bifurcations. They concluded that these findings show that "our CAD
algorithm can reliably detect pulmonary nodules, including those
with pleural or vascular contact, with a small number of false
positives. Algorithms such as these are a necessary step for
efficient mass screening for lung cancer using CT."
CAD technical exhibits
Several of the nearly 600 technical exhibits demonstrated recent
advances in CAD technology.
R2 highlights CAD for chest and mammography
R2 Technology (Los Altos, CA) previewed its LungCheck CAD software,
designed for use in chest CT studies, and announced that it has
entered the initial clinical investigation phase with a workstation
recently installed at Palo Alto Medical Clinic in Palo Alto, CA.
The system includes CAD software for the automatic detection and
analysis of potential lung cancers; a patented user interface for
navigation between 2D and 3D information; reporting tools,
including nodule measurement, temporal comparison between current
and prior scans; and automatic report generation.
The system was shown as a work-in-progress in conjunction with
TeraRecon, Inc.
(San Mateo, CA) as part of a joint demonstration of advances in
volumetric CAD for chest using TeraRecon's Aquarius Workstation, a
real-time 2D review and 3D post-processing workstation.
Noting that the volumetric capability of the current imaging
modalities present both opportunities and challenges for
radiologists, Robert Taylor, PhD, executive vice president of
TeraRecon said, "The increased quantity and quality of information
will enable radiologists to offer more precise and effective
diagnoses, but only when tools are available that provide practical
and effective means to manipulate and manage these large data
sets."
R2 also previewed enhancements to their ImageChecker CAD system
for mammography, the CheckMate Ultra. This version, shown as a
work-in-progress, will offer SmartView, which will outline masses
and highlight calcifications and allow the radiologist to magnify
regions of interest for greater visualization. The CheckMate Ultra
system is expected to be available in early 2002.
CADx receives approvable letter
Also during RSNA 2001, CADx Medical Systems (Laval, Canada)
announced that it had received an approvable letter from the Food
and Drug Administration (FDA) for its CAD system for mammography,
Second Look.
"We're pleased that the PMA process for Second Look is coming to
a close and we look forward to the approval of our product in the
United States," said Greg Arnsdorff, president of CADx. "We are
excited about moving one step closer to providing more women access
to this important tool in the fight against breast cancer."
Virtual colonography clinical studies
Another area of computer-assisted radiography that generated
great interest at RSNA 2001 was CT colonography. The so-called
"virtual colonoscopy" is a non-invasive technique that has shown
promise as a screening method for potentially cancerous polyps.
University College of London Study
In one of the largest studies to date, "Is CT Colonography a
Reliable Method for Detecting Colorectal Cancer in Symptomatic
Patients?" presented by William Lees, MBBS, of the University
College of London and Middlesex Hospital, researchers found that
virtual colonoscopy provided a high degree of sensitivity and
specificity in cancer detection among all symptomatic patients,
except those with flat lesions.
Over a 5-year period of offering a routine CT colonography
service, Dr. Lees and co-workers studied 1,400 symptomatic
patients. CT diagnosis was tested against colonoscopy, laparotomy,
or a minimum of 6 months' clinical follow-up. All patients
underwent standard bowel preparation, with fecal marking used as an
alternative for some elderly or weak patients. Using a
single-detector spiral scanner for the first 1,000 patients, and a
Somatom Plus 4 Volume Zoom (
Siemens
, Erlangen, Germany) multidetector-row scanner for the last 400,
each patient underwent an abdominal CT imaging protocol, most in
both supine and prone positions.
To date, the researchers have identified 259 cancers, 255 of
which were seen on CT colonography. Three lesions were missed in
the first 200 cases; 2 were Duke's A flat cancers found at
colonoscopy, and 1 was a malignant polyp hidden in residue that was
later discovered in a subsequent CT colonography. There were 16
false positives, 14 of which were inflammatory lesions.
Approximately 7% of the studies were considered suboptimal, due to
the inability to obtain prone and supine views, lack of IV access,
or poor bowel prep.
Overall, Dr. Lees concluded, the technique was 98.5% sensitive
and 98.2% specific for the diagnosis of colon cancer.
CAD reimbursement increased
The Centers for Medicare and Medicaid Services (formerly the
Healthcare Finance Administration [HCFA]) recently announced that
the reimbursement for the use of CAD for screening mammography will
be increased to $17.74 per exam effective in January 2002. Of that
amount, $14.48 is allocated to the technical component and $3.26
for the professional fee. This is an increase of 18% overall and an
increase of 45% within the technical component.
"Medicare reimbursement not only validates CAD's clinical
efficacy, but it also allows this valuable technology to be made
accessible for women across the nation," said Jerry Kolb,
president, Breast Health Management, Inc., Bend, OR.
Use of CAD in virtual colonoscopy
A second study, "Evaluation of Computer-aided Detection in CT
Colonography: Potential Applicability to a Screening Population,"
presented by David S. Paik, MS, of Stanford University, looked at
the use of CAD in CT colonography.
Although the exam itself is relatively quick for the patient,
the review of hundreds of images for each study can be quite time
consuming. This study looked at the effectiveness of CAD as a tool
for "prescreening" the images to highlight potential polyps. "CAD
can provide radiologists with much-needed assistance as they review
hundreds and hundreds of scans in a screening environment," said
Mr. Paik.
For this study, his research group scanned 51 patients with
single- and multidetector CT in both supine and prone positions
with varying colonic preparations. Of those studied, 9 patients had
a total of 14 significant polyps (defined as >8.5 mm) and the
remaining 42 patients had no lesions larger than 8.5 mm. Fiber
optic colonoscopy and prospective expert correlation to the CT
colonography were used to establish the gold standard against which
CAD as an adjunct to virtual colonoscopy was assessed.
Overall, the CAD CT colonography achieved 92.9% sensitivity with
7.9 false positives per colon. There were no statistically
significant differences in accuracy between studies performed in
the supine or prone position.
The researchers concluded that CAD reliably enables the
detection of clinically significant polyps with a low
false-positive rate with supine and prone scans, varying
preparation, distension and imaging quality, and a low polyp
prevalence.
Paik suggested that the CAD method's greatest utility might be
as a "first reader" of CT exams. "Given the performance of this
study, we feel confident that there really aren't polyps there if
the computer doesn't indicate any suspicious areas," he concluded.
"The scans that the computer flags as suspicious can then be
reviewed by a radiologist to make a definite determination."
Virtual colonoscopy technical exhibits
Viatronix unveils V3D-Colon enhancements
Viatronix Inc. (Stony Brook, NY) displayed the latest enhancements
in their V3D-Colon virtual colonoscopy software.
In addition to the existing supine and prone correlated axial
views, the new Supine and Prone 3D Registration correlates
endoscopic views for diagnostic confirmation of abnormalities.
These linked views facilitate handling of real-world clinical
situations such as peristalsis, colon collapse, and retained fluid.
The Automatic 2D Flight component was developed to complement the
existing automatic 3D flight capabilities and the Automatic Patient
MPEG CD creation capabilities will allow patient viewing of their
virtual colonoscopy.
Voxar unveils virtual colonoscopy software
Voxar Limited (Edinburgh, Scotland) launched its virtual
colonoscopy software, Colonscreen, at this year's meeting. A
complete virtual colonoscopy workflow and reporting application,
Colonscreen allows radiologists to interpret and generate a report
from CT or magnetic resonance (MR) studies.
Colonscreen provides 2D and 3D views of the insufflated colon,
and offers real-time manipulation of image data and a reporting
system designed to convey patient information to referring
clinicians. The system is designed to allow the radiologist to
fully interpret and generate a report from CT or MR image studies
within 10 minutes.
Prone and supine datasets can be loaded from the patient list
and a manual cine function can be used to track the colon lumen in
transverse CT slices. Suspected lesions and polyps can be marked
and confirmed with 2D and 3D viewing. Once such lesions are marked,
the image is captured and the reporting function allows the user to
select details and images for inclusion in the report. The report
is then generated and can be printed or e-mailed immediately.
E-Z-EM Debuts Two New Colonoscopy Products
Contrast agent developer E-Z-EM (Westbury, NY) highlighted two
colonoscopy products at RSNA 2001: the NutraPrep meal kit and
Tagitol, a tagging agent designed to improve visualization of the
bowel during virtual colonoscopy exams.
NutraPrep is the first meal kit designed to replace the clear
liquid diet required prior to examination of the colon. The kit
contains a full day's supply of shakes, soup, energy bars, and
chips specifically designed to provide essential nutrition while
significantly reducing the amount of residue remaining in the bowel
after digestion.
"Until now, preparing for a colon exam forced people to fast for
an entire day, leaving them weak, hungry, and irritable," said Mark
DeLegge, MD, assistant professor of medicine at the Medical
University of South Carolina in Charleston. "But with specially
prepared foods, patients will be able to eat the day before the
exam and improve compliance with the preparation process."
The second product featured, Tagitol, is a radiopaque contrast
medium designed to improve visualization of the bowel during
virtual colonoscopy. It works by marking any stool retained in the
colon. An electronic subtraction system automatically distinguishes
the stool from pathology, allowing the radiologist to eliminate the
stool from the virtual image electronically.
"In the past, a great deal of time was spent in the analysis of
images to assure that retained stool was not mistaken for polyps,"
said Anthony Stauffer, MD, medical director of the Mission Regional
Imaging Center, Mission Viejo, CA. "But with this new tagging
agent, we can better overcome the problem of residual stool and
complete the examination with an accurate diagnosis."
The low-density suspension of lemon-lime-flavored barium sulfate
comes in 250-mL bottles for administration with meals prior to the
procedure.
16-slice CT
Another technology that generated a lot of interest at RSNA 2001
was 16-slice CT. Currently available from one vendor, several
companies displayed works-in-progress for CT capable of 16 or more
slices.
Siemens Medical Solutions
(Iselin, NJ) featured its new, commercially available 16-slice CT
scanner, Somatom Sensation16. According to the company, this
scanner provides up to 12 times the acquisition speed of existing
4-slice scanners. It has a 0.4-second rotation time and produces
thin slices and isotropic resolution, with a voxel size of 0.5 x
0.5 x 0.6 mm, according to Markus Lusser, segment manager for
CT.
The FDA recently cleared the system for marketing and the first
American site is scheduled to be installed by January 2002, with
standard commercial deliveries beginning in July.
Philips Medical Systems
(Bothell, WA) previewed its Infinite Detector Technology (IDT),
deployed on its Mx8000 multislice CT scanner, as a
work-in-progress. With the addition of IDT, the Mx8000 can collect
16 slices of data simultaneously, and 38 slices of data per second
due to its 0.42-second rotation time. According to company vice
president for CT, Jim Green, the Mx8000 with IDT can cover more
than 4 cm of patient anatomy in 1 second with submillimeter
isotropic accuracy.
Philips believes that this technology also offers the future
possibility of 32-slice and 64-slice acquisition. The company
expects to begin shipping the IDT-equipped Mx8000 scanner in
mid-2002. Previously installed Mx8000 scanners can be upgraded to
IDT.
GE Medical Systems
(Milwaukee, WI), which highlighted the LightSpeed Ultra, a new
8-slice CT, also announced that it is developing a 16-slice scanner
that they hope to introduce next year.
Other RSNA news
Wuestec, Inc.
(Mobile, AL) introduced its film scanning and conversion services:
Film Scanning, to digitize large volumes of data; and Digital
Conversion, to convert digital images to DICOM-3 format.
The Film Scanning service digitizes medical images and
incorporates all films, patient records, and jackets into the
client's existing database system. The service includes all the
necessary equipment and labor to process more than 500,000 films
per month.
The Digital Data Conversion process converts the client's
digital image archive to DICOM 3.0 format, then reviews and stores
the data.
Fees for the digitizing service are waived for institutions that
register with the company's off-site storage offering.