Man 00851 rev 001 understanding image checker 9.0

Artwork consists of:

• Forty-eight (48) 8 ½ inch x 11 inch sheet(s) attached.

REV AUTHORED BY DATE Osteoporosis Assessment
LINDA BOURASSA 11/19/07 LORAD© Breast Cancer Detection
REV DRAFTED BY DATE DirectRay© Digital Imaging
LINDA BOURASSA 11/19/07
PROPRIETARY: This document contains TITLE DOCUMENT NUMBER REV
proprietary data of Hologic, Inc. No
disclosure, reproduction or use of any
part thereof may be made except by
written permission from Hologic.
REV. RELEASE DATE: 01/07/08 ARTWORK SIZE A SHEET 1 OF 1
Form ENG-0034-T03, Rev. 001

Understanding
ImageChecker 9.0
®

MAN-00851 Rev 001

Mammography

Understanding
ImageChecker 9.0 ®

PN MAN-00851 Rev 001

Technical Support
For support in North America contact:
Toll Free: +1.866.243.2533 (+1.866.CHECKED)
Email: r2support@hologic.com
Hours: Monday – Friday, 6:00 AM – 5:00 PM, PT (GMT –8:00)
Website: www.hologic.com
For support in Europe, South America, or Asia, contact your local dealer or distributor.

© 2007, Hologic, Inc. All rights reserved. Duplication or distribution without written permission
is prohibited. Hologic reserves the right to revise this manual. Issued November 2007.

Protected by one or more of the following U.S. Patents: 5133020, 5452367, 5491627, 5537485,
5622171, 5657362, 5673332, 5729620, 5732697, 5740268, 5815591, 5828774, 5832103,
5917929, 6014452, 6035056, 6075879, 6078680, 6185320, 6198838, 6263092, 6266435,
6301378, 6404908, 6434262, 6477262, 6574357, 6580818, 6640001, 6628815, 6909795,
7054473, 7072498, 7146031, 7174515

Hologic, the Hologic logo, Cenova, Citra, DigitalNow, EmphaSize, Gold Standard CAD,
ImageChecker, LesionMetrics, Malc, PeerView, Quantra, RightOn, R2, R2 Technology,
SecurViewDX, and Selenia are trademarks or registered trademarks of Hologic in the USA.
MagView is a registered trademark of MagView Corporation. MAMMOMAT is a registered
trademark of Siemens Medical Solutions in the USA. MergeCOM-3 is a trademark of
Merge Healthcare. MRS is a registered trademark of Mammography Reporting System, Inc.
PenRad is a registered trademark of PenRad Technologies, Inc.

Hologic Inc. Hologic N.V.
35 Crosby Drive Authorized Representative
Bedford, MA 01730-1401 USA Leuvensesteenweg 250A
Tel: +1.781.999.7300 1800 Vilvoorde, Belgium
Sales: +1.781.999.7453 Tel: +32.2.711.4680
Fax: +1.781.280.0668 Fax: +32.2.725.2087
Asia Pacific
Room 302, Hung Kei Building MergeCOM-3 Advanced Integrator’s
5-8 Queen Victoria Street Tool Kit is a product of Merge
Central, Hong Kong Healthcare.
Tel: +852.3102.9200

ii Understanding ImageChecker 9.0 – PN MAN-00851 Rev 001

Contents
Part 1: Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1. Intended Use. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2. Resources Available . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3. Warnings and Precautions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4. Overview of ImageChecker CAD. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.5. Benefits of ImageChecker CAD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.6. Devices Used with ImageChecker CAD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Selenia FFDM System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
R2 Cenova Server . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Diagnostic Review Workstations and PACS Devices . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.7. ImageChecker Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Input Ports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Citra Core . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
RightOn CAD Marks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
EmphaSize. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
PeerView Digital . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
LesionMetrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Operating Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

Part 2: Data Inputs and Outputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1. System Data Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Digital CAD Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Digital CAD with DICOM 6000 Overlay . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Digital and Film-Based CAD with DigitalNow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2. CAD Inputs and Supported Views . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Images with Breast Implants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3. Image and Case Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.4. CAD Outputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
DICOM 6000 Overlays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

Understanding ImageChecker 9.0 – PN MAN-00851 Rev 001 iii

Contents

Part 3: Algorithm Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.1. ImageChecker CAD Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.2. RightOn CAD Marks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Mark Capping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.3. EmphaSize Marks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.4. PeerView Digital . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.5. LesionMetrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.6. What the Algorithm Detects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
Calcifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
Masses/Architectural Distortions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.7. Detecting Calcifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Image Processing for Calcifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Case Processing for Calcifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.8. Detecting Masses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
Image Processing for Masses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
Case Processing for Masses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.9. CAD Operating Points. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

Index. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

iv Understanding ImageChecker 9.0 – PN MAN-00851 Rev 001

Part 1: Introduction
f 1.1. Intended Use
f 1.2. Resources Available
f 1.3. Warnings and Precautions
f 1.4. Overview of ImageChecker CAD
f 1.5. Benefits of ImageChecker CAD
f 1.6. Devices Used with ImageChecker CAD
f 1.7. ImageChecker Features

ImageChecker® 9.0 Gold Standard CAD™ is a proprietary software algorithm used
for analyzing mammography images. The software was originally developed by
R2 Technology, and is now available as a licensed option with Hologic’s R2 Cenova™
digital mammography server.
The R2 Cenova server provides a second application, Quantra 1.0, which is used
for volumetric assessments of breast tissue. (Quantra is not available in the United
States.) For more information, refer to the manual Understanding Quantra 1.0.
The information in this manual is intended to serve as a reference for radiologists
and clinic personnel who need to understand how computer-aided detection (CAD)
can be integrated into their practice.
1.1. Intended Use
ImageChecker 9.0 is a software application intended to identify and mark regions of
interest on routine screening and diagnostic mammograms to bring them to the
attention of the radiologist after the initial reading has been completed.
ImageChecker CAD assists the radiologist in minimizing observational oversights by
identifying areas on the original mammogram that may warrant a second review.

Understanding ImageChecker 9.0 – PN MAN-00851 Rev 001 1


1.2. Resources Available
In addition to this manual, the following resources are available to assist you when
working with ImageChecker 9.0 and the R2 Cenova server.
• R2 Member Center: This website provides easy access to electronic (pdf) versions
of Hologic | R2 manuals and training materials. You can find the R2 Member
Center by visiting the Hologic website (www.hologic.com).
• Manuals: This manual provides basic instructions for using the R2 Cenova server.
The following manuals provide additional information:
• Understanding Quantra 1.0
• R2 Cenova User Manual
• R2 Cenova Service Manual
• R2 Cenova DICOM Conformance Statement (Hologic website only)
• R2 Cenova Cybersecurity Best Practices (Hologic website only)
You can obtain additional copies of printed manuals through your Hologic
Account Manager.
• Training: The Hologic Applications team is available to train your staff, should you
feel they need additional training. Contact your Hologic Account Manager if you
want personalized instruction.
• Customer Bulletins: Hologic is committed to making our products safe and easy to
use. If an issue arises or new features become available, you will receive a
Customer Bulletin.
• Technical Support and Service: For contact information, see ‘Technical Support’
on the back of the title page of this manual.

2 Understanding ImageChecker 9.0 – PN MAN-00851 Rev 001

1.3. Warnings and Precautions

1.3. Warnings and Precautions
Note: For Warnings and Cautions related to the installation, operation, and
maintenance of the R2 Cenova server, refer to the R2 Cenova User Manual.
• The radiologist should base interpretation only on diagnostic-quality images and
not depend on ImageChecker CAD marks for interpretation.
• The device is a detection aid, not an interpretative aid. ImageChecker CAD marks
should be activated only after the first reading.
• The device does not enhance what the user sees; rather it helps to identify regions
on mammograms that should be re-examined.
• The ImageChecker software marks calcification features with triangles (Calc
marks) and mass features with asterisks (Mass marks). The software marks mass
and calcification features that occur at the same location on the image with
pointed crosses (Malc marks). These features may not represent cancer, and the
skill of the user is still required for proper interpretation of the marked areas.
• EmphaSize (variable-size) marks – Sites may choose to display prominence detail,
in which case the size of a Calc, Mass, or Malc mark is proportional to the ranking
of the feature by the algorithm. The marked features may not represent cancer,
and the skill of the user is still required for proper interpretation of areas marked
by the device.
• During routine interpretation of mammograms and results produced by
ImageChecker CAD, Hologic recommends use of a mammography workstation
that has been FDA-approved, recently calibrated, and employs a CAD marking
schema authorized by Hologic.
• ImageChecker software does not identify all areas that are suspicious for cancer.
– The software does not mark all lesions and a user should not be dissuaded from
working up a finding if the software fails to mark that site.
– The software is not designed to detect changes from prior mammograms.
– The software is not designed to detect skin thickening or nipple retractions.
– Conditions of the breast that diminish mammographic sensitivity, such as
density of normal tissue, also diminish the sensitivity of the software.
– The software is more sensitive for detection of calcifications than masses, and the
sensitivity depends on the site-specific operating points chosen. For sensitivity
values, see ‘3.9. CAD Operating Points’. In addition to not marking all masses, the
algorithm has a lower sensitivity for masses greater than 2.5 cm in diameter.
– Individual practice patterns may influence results obtained when using
ImageChecker CAD. Therefore, each facility and radiologist should carefully
monitor the results that the software has on their practice of mammography in
order to optimize its effectiveness.
• Safety and effectiveness have not been established for analyzing mammography
images from patients with breast implants. For images with breast implants, send
only Implant Displaced views to the system. Images with more than 2.5 cm (1
in) of the breast implant appearing on the image may not be processed correctly.



• The performance of the software has not been characterized for special diagnostic
views (e.g., magnified views or spot-compressed views).
• Images of partial views of the breast that are not correctly identified as such, may
be processed by the software, although the software has not been designed for that
purpose.
• On rare occasions for a lesion on or near the pectoral muscle, the system may
report a negative value for the chest wall distance.


1.4. Overview of ImageChecker CAD

1.4. Overview of ImageChecker CAD
The R2 Cenova server configured with ImageChecker software analyzes digital
mammography images produced by full-field digital mammography (FFDM)
systems. The proprietary ImageChecker algorithm identifies regions of interest,
which can include clusters of bright spots (suggestive of calcification clusters), and
dense regions with or without radiating lines (suggestive of masses or architectural
distortions). The server generates results in the form of Mammography CAD SR
objects that include CAD marks and other data identifying and characterizing the
regions of interest. At the review workstation, the results are matched with the digital
mammography images produced by the FFDM system.
After making an initial interpretation from the original diagnostic-quality images, the
radiologist displays the ImageChecker results and chooses whether or not to reinspect
the marked regions on the original images. The ImageChecker algorithm marks
visually perceptible structures that have some of the generally accepted geometric
characteristics of calcifications or masses. The marked areas may be something other
than an actual abnormality, which is generally recognized by the radiologist upon a
second review of the original images.
Note: The appearance of the images on the workstation is dependent upon the
acquisition modality and the workstation’s display capabilities, and is not affected by the
ImageChecker software.
The R2 Cenova server and ImageChecker 9.0 are designed for sites that process
only digital mammography images; the server does not scan films or support CAD
with analog (film) mammograms. For information on systems that perform film-
screen CAD, contact your Hologic Account Manager.

For Presentation
Images

Mammography
CAD SR
For Processing Images

Examination is ImageChecker CAD Radiologist
Performed Analyzes the Images Reviews Images

Workflow with ImageChecker CAD



1.5. Benefits of ImageChecker CAD
The interpretation of mammograms is challenging. Normal breast tissue varies widely
among women, even for the same woman over time or at different times of the
month. The radiologist also has to balance the need for accurate detection with the
need to limit the number of unnecessary procedures. The combination of viewing a
large number of cases, radiologist fatigue, the complex image of the breast structure,
and the subtle nature of certain observable characteristics of the disease can result in
false-negative readings. In fact, studies show that half of undetected cancers are
missed due to observational oversights. The prevalence of observational oversights is
not strongly related to experience and may be inevitable with human observers.
ImageChecker CAD functions like a spellchecker for medical images. It is designed to
help radiologists in reducing the number of false-negative readings due to
observational oversight by drawing their attention to areas that may warrant a
second review. Retrospective clinical studies using the film-based ImageChecker
algorithm have shown that use of ImageChecker CAD can result in earlier detection
of up to
23.4% of the cancers currently detected with screening mammography in those
women who had a prior screening mammogram 9–24 months earlier. Early detection
is the key to higher survival rates, better prognosis, and lower treatment costs.


1.6. Devices Used with ImageChecker CAD

1.6. Devices Used with ImageChecker CAD
All systems with ImageChecker 9.0 include:
• Selenia FFDM System
• R2 Cenova server (configured with the ImageChecker option)
• Diagnostic review workstation or PACS
The ImageChecker software can accept digital mammography images from the
Selenia FFDM System or a PACS (picture archiving and communication system).
Once received, the server performs the CAD algorithm processing and sends the
study results (CAD marks and related data) to one or more devices such as Hologic’s
SecurViewdx diagnostic workstation or a PACS for storage.
The system devices are connected by a computer network and can be in the same
room, different rooms, the same building, different buildings, or even different
cities. The server can send results to several different output devices.

Selenia FFDM System
ImageChecker 9.0 processes mammography images originating from the Selenia
FFDM System manufactured by Hologic. Support for other brands and models of
full-field mammography systems will be added in future releases.
R2 Cenova can process images originating from up to four acquisition gantry devices.
The DICOM header for each image includes an identifier (e.g., a serial number) that
indicates the device from which the image originated. Each image-acquisition device
is mapped to an input port on the server. For each R2 Cenova server, you can
purchase licenses for up to four ImageChecker input ports.

R2 Cenova Server
The ImageChecker software is supplied on the R2 Cenova server, a high-performance
computer that can be located in any non-patient environment, provided it is
connected to the mammography network. The server receives and processes digital
mammography images and supports up to four ImageChecker ports (each port
licensed to a specific detector).
Note: In addition, R2 Cenova can support up to four ports for the Quantra application.
R2 Cenova integrates easily into the clinical workflow. Following a mammography
examination, the acquisition gantry device sends digital mammography images to the
clinic’s review workstation, PACS, and the R2 Cenova server. The server accepts and
processes each For Processing image and sends the results as a Mammography CAD
SR object to the review workstation and PACS.
For more information on the server, refer to the R2 Cenova Server User Manual.



Diagnostic Review Workstations and PACS Devices
Mammography CAD SR results produced by R2 Cenova can be sent to a variety of
PACS devices and diagnostic review workstations (including SecurViewdx available
from Hologic, Inc.). For a listing of supported workstation vendors, refer to the R2
Member center at the Hologic website (www.hologic.com).
For diagnostic review workstations and PACS devices that do not support
Mammography CAD SR, but do support DICOM 6000 overlays, Hologic’s DICOM
6000 Interface server can be used to convert Mammography CAD SR objects.
For more information, contact your Hologic Account Representative.


1.7. ImageChecker Features

Each R2 Cenova server with ImageChecker software is delivered with a number of
licensed features in place. Upon installation, the Hologic | R2 field service engineer
will tailor the software to enable or disable selected features based on the
preferences of site personnel. Thereafter, your Hologic | R2 representative can help
you add new licenses and enable or disable features as needed.

Input Ports
The R2 Cenova server can process images originating from up to four acquisition
gantry devices. The DICOM header for each image includes an identifier (e.g., a
serial number) that indicates the device from which the image originated. Each
image- acquisition device is mapped to an input port on the server.
For each R2 Cenova server, you can purchase licenses for up to four
ImageChecker input ports. Similarly, you can obtain licenses for up to four
Quantra input ports.
The server can also process images sent from any number of PACS devices, as long
as the images originated from one of the licensed acquisition gantry devices.

Citra Core
The standard set of ImageChecker features is included in the Citra™ Core
license group, which includes the following features:
• RightOn™ CAD Marks
• EmphaSize™
• PeerView Digital™
• LesionMetrics™
EmphaSize, PeerView Digital, and LesionMetrics are exclusive Hologic | R2 features
supported by some diagnostic review workstations. When these features are licensed,
the ImageChecker software includes extra CAD information in the Mammography
CAD SR output. When viewed at the workstation, the results help radiologists
understand better why a region of interest received a CAD mark.
The Citra Core features are discussed in the following pages. For additional
information, see ‘Part 3: Algorithm Description’.
Important! Display of results from each ImageChecker feature depends upon several
factors, in particular:
• The feature must be licensed and enabled on the R2 Cenova server.
• The workstation must be equipped with software that can interpret the
Mammography CAD SR output produced by the server (v5.x or 9.0).
Any of the ImageChecker features can be disabled either when the server is installed or,
in most cases, at the diagnostic workstation. Consult with your workstation vendor
concerning availability and integration of ImageChecker features with your workstation.



RightOn CAD Marks
The ImageChecker software provides three types of RightOn CAD marks (Mass,
Calc, and Malc) that can appear in the results shown on the review workstation. You
can choose to display any or all of the three types of marks. Each mark identifies a
region of interest for the radiologist to review.

Calc – Marks regions suggestive of calcifications.

Mass – Marks regions suggestive of masses or architectural distortions.

Malc – Marks regions where Calc and Mass marks are within 10 mm of each
other.

EmphaSize
The server is configured by default to convert the CAD marks to EmphaSize marks.
The feature allows the review workstation to display marks of variable size that
correlate to feature significance. When the algorithm determines that a region is more
significant, the CAD mark appears larger, indicating that the region should receive
more emphasis from the radiologist. EmphaSize marks can be disabled if facility
personnel choose not to use them. For more information see ‘3.3. EmphaSize Marks’.

PeerView Digital
PeerView Digital enables the radiologist to understand better why a region of interest
was marked. On advanced review workstations such as Hologic’s SecurViewdx,
PeerView Digital highlights physical features found by the algorithm. For more
information see ‘3.4. PeerView Digital’.

LesionMetrics
LesionMetrics provides data computed by ImageChecker CAD for each region of
interest marked by the algorithm. Depending upon the type of lesion, the software
can compute metrics such as lesion size, distance to nipple, distance to chest wall,
degree of spiculation, calcification contrast, number of calcifications, and mass
density. For more information see ‘3.5. LesionMetrics’.



Operating Points
The ImageChecker algorithm offers three operating points (i.e., CAD algorithm
thresholds) to accommodate differing radiologist preferences: operating point 0,
which emphasizes specificity (a low false-mark rate), operating point 1, which is a
balanced intermediate point, and operating point 2, which emphasizes sensitivity.
The algorithm performance for each of the operating points is summarized below:

Operating Point 0 1 2
Calcification Sensitivity 95% 96% 97%
Mass Sensitivity 83% 88% 90%
False Marks per Case 1.0 1.5 2.0

When your system is installed, the server is configured to use operating point 1 for
masses and operating point 2 for calcifications. You can choose other settings, or your
service representative can change the operating points at a later time. For new
customers, Hologic | R2 recommends using the default settings for the first four to six
weeks to allow you to become familiar with the R2 marks and algorithm behavior.
Hologic | R2’s Gold Standard CAD™ algorithm provides the medical-imaging
industry’s highest sensitivity at any given false-mark rate. ImageChecker CAD was the
first computer-aided detection software approved by the FDA for full-view diagnostic
and screening mammograms. Since that first approval, the ImageChecker software
has been refined many times, resulting in an algorithm that is highly sensitive with
few false marks. For more information, see ‘3.9. CAD Operating Points’.


Part 2: Data Inputs and Outputs
f 2.1. System Data Flows
f 2.2. CAD Inputs and Supported Views
f 2.3. Image and Case Processing
f 2.4. CAD Outputs

Part 2 explains how information flows through systems with ImageChecker CAD, the
supported mammography views, and the result formats.
2.1. System Data Flows
This section describes the flow of data among the various Hologic | R2 devices.
Mammography systems may have different configurations, depending upon the
devices installed at your site:
f Digital CAD Systems
f Digital CAD with DICOM 6000 Overlay
f Digital and Film-Based CAD with DigitalNow

Digital CAD Systems
To follow the data flows, refer to the diagram on the following page.

Image Acquisition Device(s)
1, 2, 3: Selenia FFDM Systems can be configured to send Digital Mammography X-
Ray images to any device that conforms to the Digital Imaging and Communications
in Medicine (DICOM) standard. The FFDM device sends two types of images, For
Processing images to the R2 Cenova server, and For Presentation images to the
diagnostic review workstation(s) and PACS.
Note: The DICOM standard specifies two classes of archivable digital mammography
images with the following designations:
• Digital Mammography X-Ray Image Storage – For Processing images
• Digital Mammography X-Ray Image Storage – For Presentation images

R2 Cenova Server
4, 5: R2 Cenova servers send results (Mammography CAD SR objects) to one or more
diagnostic review workstation(s) and/or PACS devices simultaneously.



Diagnostic Review Workstation(s) and PACS
4, 7: The review workstation(s) are configured to receive the For Presentation images
and CAD results, which are then reviewed by the radiologist.
6, 7: The PACS can be configured to send:
• For Processing Digital Mammography X-Ray images to R2 Cenova (6), which
processes the images and distributes the CAD results according to its
configuration, and/or
• CAD results and/or For Presentation Digital Mammography X-Ray images to the
review workstations (7).

3 For Presentation 7 For Presentation Images
Images

SecurViewDX
2 For Presentation Images (or other
For Processing Images diagnostic
(if archived) PAC workstations)
S

6 For Processing Images 5 Mammography
(if archived) CAD SR Objects

Selenia FFDM
System(s) 1 For Processing
Images
4 Mammography
CAD SR Objects

R2 Cenova

Digital CAD System


Digital CAD with DICOM 6000 Overlay
Typically, diagnostic review workstations and PACS devices support the Mammo-
graphy CAD SR results format. If the review workstation or PACS does not support
Mammography CAD SR, but does support DICOM 6000 overlays, a supplementary
interface computer can be installed.
The DICOM 6000 Interface server creates a copy of the For Presentation image
with an overlay containing the CAD results that can be viewed on the diagnostic
workstation. (Refer to paths 4, 8, and 9 in the diagram below. The earlier diagram
provides information on data flows 1–7.)
For more information about the Hologic DICOM 6000 interface option, contact your
Hologic Account Representative.

7 For Presentation Images
Diagnostic
PACS Workstation(s)
(if archived)

6 For Processing Images 5 Mammography with DICOM 6000
(if archived) CAD SR Objects Overlays

1 For Processing
Selenia FFDM Images 4 Mammography
System(s) CAD SR Objects

R2 Cenova DICOM 6000
8 For Presentation
Images
Interface

Digital CAD System with DICOM 6000 Interface



Digital and Film-Based CAD with DigitalNow
Some facilities with analog (film-based) mammography systems, or those with large
film libraries may wish to acquire one of Hologic | R2’s film-scan systems (R2
DMax
or R2 LS). With these systems, the scanned films can be analyzed using ImageChecker
CAD, or they can be converted to archivable digital images using Hologic | R2’s
DigitalNow feature.
To follow the data flows, refer to the diagram on the next page. The earlier diagrams
provide information on data flows 1–7.

Films
8: Film priors are fed into the processing unit with film scanner, which scans
(digitizes) the films and processes the resulting images.

Processing Unit with Film Scanner
9: The R2 processing unit with film scanner can be configured to send
ImageChecker results (processed from digitized film images) and/or the For
Presentation Digital Mammography X-Ray images themselves to a PACS or review
workstation.
When scanning films for digital review, sites need the ability to retrieve patient
information in order to add the required identification information to the resulting
Digital Mammography X-Ray images. Sites can use the R2 Patient ID feature, an
integrated mammography reporting system (e.g., PenRad, MRS, MagView, etc.), or
retrieve the patient data from a PACS or RIS (Radiology Information System).



3 For Presentation SecurViewDX
Images (or other
diagnostic
workstations)
(if archived)

PAC
S

6 For Processing 5 Mammography
Images CAD SR Objects
(if
archived)

Selenia FFDM System(s)

4 Mammography
1 For Processing CAD SR Objects
Images

R2 Cenova

from Digitized Films

8
Films

R2 Processing Unit
with Film Scanner

Digital and Film-Scan CAD with DigitalNow


2.2. CAD Inputs and Supported Views
The ImageChecker software analyzes digital mammography images that conform to
the Digital Imaging and Communications in Medicine (DICOM) standard.
If your system has the ImageChecker license, all images are CAD-processed if
the views are supported by the software. The following table lists the views and
view modifiers supported by the ImageChecker software:

Supported DICOM ACR MQCM 1999*
Views and View View Label
Modifiers
Screening Views Cranio-Caudal CC
Medio-Lateral Oblique MLO
Equivalent Views Medio-Lateral ML
Cranio-Caudal Exaggerated XCC
Cranio-Caudal Exaggerated Laterally XCCL
Cranio-Caudal Exaggerated Medially XCCM
Reversed Equivalent Latero-Medial LM
Views
Latero-Medial Oblique LMO
Cranial-Caudal From Below FB
Superolateral to Inferomedial Oblique SIO
View Modifiers Axillary Tail (modifies MLO views only) AT
Rolled Inferior …RI
Rolled Lateral …RL
Rolled Medial …RM
Rolled Superior …RS
Implant Displaced …ID
Tangential TAN

*American College of Radiology Mammography Quality Control Manual 1999

The ImageChecker software does not support the following view modifiers:

Unsupported ACR MQCM 1999
View Modifiers View Label
Cleavage (modifies CC views only) CV
Magnification M…
Spot Compression S…

Images with Breast Implants
The ImageChecker software does attempt to process images with breast implants.
Safety and effectiveness have not been established for analyzing images from patients
with breast implants. For images with breast implants, send only Implant Displaced
views to the system. Images with more than 2.5 cm (1 in) of the breast implant
appearing on the image may not be processed correctly.


2.3. Image and Case Processing

2.3. Image and Case Processing
During processing, the ImageChecker software analyzes each received image, as long
as the view is supported by the software. The algorithm searches for patterns
suggestive of calcification clusters and masses or architectural distortions,
characterizes each suspected lesion, and finally determines the location and numbers
of CAD marks for the image. This initial analysis is known as image processing.
The R2 Cenova server reads the DICOM header of each received image and groups
successive images from a single patient into a study. There is no specified limit to the
number of images that can be included in a study. However, for many patients, a
study consists of the four screening views:
• LCC – Left Cranio-Caudal • RCC – Right Cranio-Caudal
• LMLO – Left Medio-Lateral • RMLO – Right Medio-Lateral Oblique
Oblique
As an additional step when a study includes two or more views, the ImageChecker
algorithm selects up to four views and compares the images to each other in a process
known as case processing. By checking for similarities and differences between the
images, the algorithm can further refine its findings for the four selected views. When
case processing is complete, the ImageChecker algorithm compiles the results into a
single Mammography CAD SR object that includes results for the images selected for
case processing and the remaining images.
The ImageChecker algorithm uses the following criteria to determine which images
will be case-processed:
• If a study includes one image for each of the four screening views or
their equivalents, then all images are case-processed.
• When a study includes multiple images of the same view and laterality (e.g., two
RCC views), case processing is performed on the last image produced by the
FFDM device for each of the four screening views or their equivalents. The
image- acquisition time is included in the DICOM header for each image. (Refer
to the diagram on the following page.)
• If a study does not include one or more of the four screening views, then the
equivalent views (or the reversed equivalent views) are case-processed, if they are
present. Screening views are given preference over equivalent views, which in turn
are given preference over reversed equivalent views.
• The view modifiers, if present, do not affect which images are selected for
case processing.



• When a study includes multiple equivalent (or reversed equivalent) views, the
views selected for case processing are based on the following preference
order:

Preference
Order CC Equivalent MLO Equivalent
1 CC MLO
2 FB ML
3 XCC LM
4 XCCL LMO
5 XCCM SIO

For example, if a study includes the four screening views plus two additional RMLO
images, all the images are image-processed separately. However, when selecting
images for case processing, the ImageChecker algorithm includes only the last RMLO
image acquired by the FFDM device (along with the other three screening views). As
is shown below, the algorithm would select the third RMLO image for case
processing.
Acquisition Time

RMLO 1 RMLO 2 RMLO LMLO RC LCC
3 C

Image Image plus Case
Processing Processing

Mammography
CAD SR

Image and Case Processing

As a further example, for a study with segmented breast views including multiple CC
images, you may choose to image the anterior (ductal) region of the breast last in
the sequence to ensure that that image is case-processed.
For more information about image and case processing, see ‘3.7. Detecting
Calcifications’ and ‘3.8. Detecting Masses’.

2.4. CAD Outputs

2.4. CAD Outputs
All results produced by the R2 Cenova server are compiled into DICOM-
conformant Mammography CAD SR objects. The server can be configured to send
results in either or both of the following formats:
• Mammography CAD SR 9.0 results are derived from the latest Hologic |
R2 algorithm enhancements.
• Mammography CAD SR 5.x provides backward compatibility with previous
releases of ImageChecker 5.x products.
The Mammography CAD SR format is the DICOM standard format for CAD results.
For a detailed description of the result formats, refer to the R2 Cenova DICOM
Conformance Statement.
Mammography CAD SR objects are most useful only when viewed with the images
on an advanced workstation such as Hologic’s SecurViewdx diagnostic workstation.
Most dedicated mammography workstations display the Mammography CAD SR
format as overlays on the underlying images. With certain advanced mammography
workstations, radiologists can work with CAD results interactively.
Although some PACS do not support Mammography CAD SR, there is increasing
support for storing Mammography CAD SR objects on PACS devices. In recent years,
an increasing number of PACS workstation models have been developed that can
display Mammography CAD SR content, partly in response to the Mammography
Image profile released by IHE (Integrating the Healthcare Enterprise).
If the results are not archived, they can be recreated at a later time if the For
Processing images have been archived. In most cases, the same results
(ImageChecker CAD marks, PeerView Digital data, etc.) should be generated each
time the images
are reprocessed.

DICOM 6000 Overlays
For diagnostic review workstations and PACS devices that do not support
Mammography CAD SR, but do support DICOM 6000 overlays, Hologic’s DICOM
6000 Interface server can be used to convert Mammography CAD SR objects (see
‘Digital CAD with DICOM 6000 Overlay’ on page 15). For more information, contact
your Hologic Account Representative.


Part 3: Algorithm Description
f 3.1. ImageChecker CAD Algorithm
f 3.2. RightOn CAD Marks
f 3.3. EmphaSize Marks
f 3.4. PeerView Digital
f 3.5. LesionMetrics
f 3.6. What the Algorithm Detects
f 3.7. Detecting Calcifications
f 3.8. Detecting Masses
f 3.9. CAD Operating Points

Part 3 describes the algorithm used by the ImageChecker software when analyzing
mammography images.
3.1. ImageChecker CAD Algorithm
R2 Technology’s Gold Standard CAD algorithm provides the medical-imaging
industry’s highest sensitivity at any given false-mark rate.
The ImageChecker algorithm looks for characteristics commonly associated with
cancer – specifically, calcifications and masses (including architectural distortions).
The algorithm ranks its findings by likelihood, places marks on those regions above
a fixed threshold of likelihood (operating point), and then sends the results to the
review workstation. The algorithm offers three operating points to accommodate
differing radiologist preferences. For more information, see ‘3.9. CAD Operating
Points’.
Hologic | R2 has accumulated a large database of mammography cases. This
database, which includes diverse populations and cases with proven cancers, is used
to train
and test the ImageChecker algorithm.



3.2. RightOn CAD Marks
ImageChecker 9.0 provides three types of CAD marks. Each mark indicates a
region of interest for the radiologist to review. There are two basic types of marks
plus one composite mark, as summarized below.

Calc – Marks regions suggestive of calcifications

Mass – Marks regions suggestive of densities/masses/architectural distortions

Malc – Composite mark indicates Calc and Mass marks occurring in the same
location on the image

ImageChecker CAD Marks
Note: Not all mammography workstations can display Malc marks. Consult with your
workstation vendor concerning availability and integration of ImageChecker features with
your workstation.

Mark Capping
The software limits the number of CAD marks for each image and case. The actual
number of CAD marks produced depends upon the individual case and the operating
point selected for the CAD algorithm. The specificity of the device at the most
sensitive operating point is no lower than 24.5% (the percentage of normal cases that
show no marks).

Views Per Image Limit Per Case Limit
Screening views 3 Calc marks 8 Calc marks
(RCC, LCC, RMLO, 2 Mass marks 4 Mass marks
LMLO) 2 Malc marks 4 Malc marks
Extra views 3 Calc marks Dependent upon the
2 Mass marks number of images*
2 Malc marks

*For cases with more than four views, the maximum number of marks per case
depends upon the number of images in the case.


3.3. EmphaSize Marks

3.3. EmphaSize Marks
EmphaSize is an optional feature that provides variable-size CAD marks that scale
according to feature significance. When the ImageChecker algorithm considers a
region to be more significant, it increases the size of the EmphaSize CAD mark. The
size of the mark does not correlate to the size of the lesion.
As the algorithm evaluates regions of interest, each region is given a ranking. This
ranking along with the selected algorithm operating point determines whether or not
the region of interest is marked with a CAD mark.
The system normally displays all CAD marks at the same size regardless of ranking. If
your site has enabled the EmphaSize feature and configured the review workstation to
use it, the ImageChecker algorithm adjusts the size of each mark according to its
ranking. If you do not want to use the feature, it can be disabled by a Hologic field
service engineer.

CAD without EmphaSize CAD with EmphaSize

• Calcifications: The size of a suspected lesion is of variable significance, since
small lesions may be highly significant and large lesions nonactionable. However,
there are other key characteristics. For calcifications, the algorithm looks at
characteristics such as signal intensity, number of calcifications in a cluster,
shape of the calcifications (i.e., pleomorphism), and other important features to
determine lesion significance.
• Masses: For masses, the algorithm looks at characteristics such as degree of
spiculation, lesion shape, contrast to surrounding tissue, and other important
features which help determine lesion significance.
• Masses with Calcifications (Malc Marks): Regions containing both a mass and
calcifications are suspicious and should be carefully evaluated.
Note: Not all mammography workstations can display EmphaSize marks. Consult
with your workstation vendor concerning availability and integration of ImageChecker
features with your workstation.



3.4. PeerView Digital
PeerView Digital is an optional, licensed feature designed to help radiologists better
understand why a region of interest was marked. When licensed, PeerView Digital
adds graphical information to the Mammography CAD SR output. At the review
workstation, the radiologist can use PeerView Digital to highlight regions of interest
detected by the ImageChecker algorithm.
• Calcifications: PeerView Digital highlights individual calcifications in the cluster
marked by the algorithm. PeerView Digital may not highlight all calcifications in a
cluster, and it may show features suggestive of calcifications that are not
calcifications.
• Masses: PeerView Digital defines and outlines the central density of the mass so
the radiologist can evaluate the margin, shape, and interior characteristics of
the CAD-detected mass or distortion. The outline does not generally include
spiculations associated with the mass, although the shape of the central density
may lead the eye to larger spicules.
• Masses with Calcifications: For composite Malc marks, which indicate that one or
more Mass and Calc marks occur at the same location on the image, PeerView
Digital highlights the calcifications and outlines the central density of the mass
and/or distortion found in the CAD analysis. The same region is shown with and
without a PeerView Digital Malc mark (mass with calcifications).

Note: Not all mammography workstations can display PeerView Digital highlights.
Consult with your workstation vendor concerning availability and integration of
ImageChecker features with your workstation.


3.5. LesionMetrics

3.5. LesionMetrics
LesionMetrics is an optional, licensed feature that provides data computed by the
ImageChecker algorithm for each marked region of interest marked. When licensed,
LesionMetrics are included as part of the Mammography CAD SR object for each
image processed by the algorithm. The LesionMetrics are summarized in the
following table.

LesionMetric Description Calc Mass Malc
R2 CAD Score A number from 0 to 100 that reflects the prominence of the
features detected by the algorithm for the region of interest.
The higher the number the more prominent the features. The
9 9 9
R2 CAD Score is not linearly related to probability of malignancy.
Number of Number of individual calcifications detected by the algorithm at
Calcifications the current operating point.
9 9
Size (long axis) Long axis in cm for the calcification cluster or mass. 9 9 9
Distance to Nipple The radial distance in cm from the center of the lesion to the
estimated nipple location. When the nipple is not clearly visible
(as in images with nipple ‘not in profile’), the estimate may be
9 9 9
slightly inaccurate.
Distance to The estimated perpendicular distance in cm from the center of
Chest Wall the lesion to the chest wall. For lesions in an upper quadrant,
this metric may resolve to the perpendicular distance to the
9 9 9
pectoral muscle, which can be reported as a negative value.
Calcification The average difference in brightness between the individual
Contrast calcifications and their immediate background normalized on a 9 9
(%) scale of 0 to 100.
Mass Density (%) The average brightness of a lesion relative to its immediate
background normalized on a scale of 0 to 100.
9 9
Degree of A measurement proportional to the number of pixels that lie on
Spiculation (%) lines pointing to a common origin inside the potential lesion,
reported on a normalized scale of 0–100 containing the entire
9 9
spectrum of biopsy-proven malignant masses.
Operating Point The selected operating points (algorithm thresholds) for
calcifications and masses.
9 9 9

The LesionMetrics feature is implemented on a limited number of review
workstations. On SecurViewdx (version 6-0 or later) you can enable or disable any or
all of the LesionMetrics. Refer to your workstation’s documentation to determine
whether these features are implemented and, if they are, how to use them.



3.6. What the Algorithm Detects
The ImageChecker algorithm searches an image for clusters of bright spots that are
suggestive of calcification clusters, for patterns of dense regions, and for dense
regions with radiating lines suggestive of masses or architectural distortions.

Calcifications
The algorithm marks:
• Clusters with three or more elements
• Elements that are at or within 3 mm of each other
• Where each element is at least 150 microns in size
The algorithm does not mark:
• Clusters with fewer than three elements*
• Clusters where each element is separated by more than 3 mm
• Elements it considers to be of benign morphology
• Low-contrast elements
• Lead skin markers or clips
*Note: The algorithm marks two elements if the elements have significant contrast or
shape, as defined by the cluster filter. See ‘3.7. Detecting Calcifications’.
The algorithm occasionally marks:
• Calcified arteries
• Cluster or rim benign calcifications
• Crossing linear tissues

Calcified Artery Cluster or Rim Crossing Linear
Benign Calcifications Tissues
To be considered a cluster, elements need only be at or within 3 mm of another
element within the cluster, as shown next:

≤ 3 mm square Each element is
at or within 3
mm


3.6. What the Algorithm Detects

When features in the mammogram meet the ImageChecker criteria, the algorithm
places a triangular CAD mark over the center of that region (not over a particular
element), as is shown next:

Masses/Architectural Distortions
To detect masses, the ImageChecker algorithm searches the image in a progressive
manner looking for circular, dense regions and lines radiating from a common
center. The algorithm marks:
• Regions suggestive of masses/architectural distortions
• Dense regions
• Regions with radiating lines

Less pronounced radiating lines but No central mass but pronounced
with a central mass radiating lines

The algorithm occasionally marks:
• Ducts and tissue radiating from the nipple
• Inadvertent crossing of parenchymal
tissue
• Well circumscribed masses
• Lymph nodes
• Vague opacities, skin thickening, or nipple retraction

Ducts and tissue Inadvertent crossing of
radiating from parenchymal tissue
nipple


The likelihood of a mass/architectural distortion being marked is affected by:
• Margin of the mass (the greater the spiculation, the more likely)
• Degree of density (the higher the density, the more likely)
• Presence of left/right breast asymmetry

Lobulated Ill-Defined Slightly Mostly Spiculated
Spiculated Spiculated

Increasing Likelihood

Low Density Medium High Density
Density

When the algorithm finds patterns associated with masses, it places an asterisk at the
point of maximum convergence on the image.


3.7. Detecting Calcifications

3.7. Detecting Calcifications
To detect calcifications, the ImageChecker algorithm performs a series of analyses.

DICOM For Processing Breast Area
Mammography Image Segmentation

Calcification Detection
(SIANN)

Calcification Cluster
Detection

Information From Other Images Statistical Pattern Recognition
(Similarity/Asymmetry)* (Combining/Ranking)

*For case processing the four
screening views or their
equivalents
Case Processing Image Processing
Results Results

Mark Capping

Mammography
CAD SR

Results Display
Operating Point Applied
on Workstation

Image Processing for Calcifications
For each image, the algorithm uses two filters (artificial neural networks) to
identify calcifications and their features:
• First, it runs each image through a Shift-Invariant Neural Network (SIANN), a
calcification detection filter that has been optimized using R2 Technology’s large
training database. SIANN is a patented technology developed through more than
10 years of research at the University of Chicago.
• Then, it analyzes the resulting calcifications using a cluster filter that weights over
a dozen different features, including contrast, shape, and size.
In addition, the algorithm segments the breast and identifies the location of the
cluster within the breast.



Case Processing for Calcifications
In addition to processing images individually, the ImageChecker algorithm selects up
to four images that represent the screening views – RCC, LCC, RMLO, and LMLO (or
their equivalents). The algorithm reviews the findings from each image and looks for
similarities in the findings by comparing complementary orthogonal views (for
example, LMLO and LCC):

Similarity

It also looks for asymmetries between views from each side, for example, LMLO and
RMLO:

Asymmetry

The algorithm combines the results of these various analyses, analyzes the resulting
cluster and context data, and employs statistical pattern recognition against the
training database to determine the ranking for each of the possible clusters. It then
selects the Calc marks that best meet the algorithm criteria (mark capping), and
applies the selected operating point, thereby limiting the number of marked features.
For more information see ‘2.3. Image and Case Processing’.


3.8. Detecting Masses

3.8. Detecting Masses
To detect masses, the ImageChecker algorithm performs a series of analyses.

DICOM For Breast Area
Processing Segmentation
Mammography Image

Mass Detection Radiating Line
Detection

Information From Other Images Statistical Pattern Recognition
(Similarity/Asymmetry)* (Combining/Ranking)

*For case processing the four
screening views or their
equivalents
Case Image Processing
Processing Results
Results

Mark Capping

Mammography
CAD SR

Results Display
Operating Point Applied
on Workstation

Image Processing for Masses
For each image, the algorithm uses patented technology to segment the breast. It
then identifies masses by evaluating structures based on their density, shape, and
margin characteristics. It also searches for structures appearing as radiating lines, and,
if they are present, the degree of spiculation.


Case Processing for Masses
In addition to processing images individually, the ImageChecker algorithm selects up
to four images that represent the screening views – RCC, LCC, RMLO, and LMLO (or
their equivalents). The algorithm reviews the findings from each image and looks for
similarities in the findings by comparing complementary orthogonal views (for
example, LMLO and LCC):

Similarity

It also looks for asymmetries between views from each side, for example, LMLO and
RMLO:

Asymmetry
The algorithm then runs the resulting data against the training database to determine
the ranking for each of the possible masses. Finally, the algorithm either marks or
does not mark the masses, based on the ranking and operating point.
The algorithm combines the results of these various analyses, analyzes the resulting
cluster and context data, and employs statistical pattern recognition against the
training database to determine the ranking for each of the possible clusters. It then
selects the Calc marks that best meet the algorithm criteria (mark capping), and
applies the selected operating point, thereby limiting the number of marked features.
For more information see ‘2.3. Image and Case Processing’.


3.9. CAD Operating Points

ImageChecker 9.0 allows each site to choose between three different operating points
(i.e., CAD algorithm thresholds). You can choose a different operating point for
calcifications than for masses, which provides a total of nine options for operating
point combinations. The operating points are summarized below:
• Operating point 0 trades off a lower false-mark rate for a lower overall sensitivity.
This operating point is suited for sites that prefer the system to display the fewest
false marks.
• Operating point 1 represents a balanced intermediate point. Improvements in the
algorithm have allowed points 0 and 2 to be set quite far apart, and some sites
may want balanced behavior from the CAD algorithm.
• Operating point 2 focuses on sensitivity. It gives the best performance for marking
regions of interest (i.e., CAD sensitivity), with a higher false-mark rate. It is suited
for sites that want the system to be as sensitive as possible, regardless of the higher
false-positive mark rate.
For example, if you want high sensitivity for calcifications but a moderate balance
between sensitivity and false positives for masses, choose operating point 2 for
calcifications and operating point 1 for masses.
Each system is shipped with a default set of operating points for both masses and
calcifications. You can discuss the choices that are right for your site with your
Applications Specialist or, if you would like to change configurations, contact
your Hologic | R2 Technical Service Representative.



The table below gives the sensitivity and false-mark rate values for the three operating
points for ImageChecker software version 9.0, as measured on Hologic | R2’s large
film test database of biopsy-proven malignancies and confirmed normal cases for the
four screening views (RCC, LCC, RMLO, and LMLO).

Operating Point: 0 1 2
Calcification Cases (n = 588)
Mean Sensitivity1 95% 96% 97%
95% Confidence Interval 93.1–96.7% 94.0–97.2% 95.5–98.3%
Mass Cases (n = 767)
Overall (n = 1355)
False-Positive Marks2
Calcification FP/Image 0.09 0.12 0.16
Mass FP/Image 0.17 0.25 0.35
Total False-Positive Marks/Case 1.0 1.5 2.0

Specificity2 48.3% 35.1% 24.5%

1. ‘Sensitivity’ refers only to the sensitivity of the CAD algorithm, not to the sensitivity of the
radiologist using the device.
2. ‘False-Positive’ and ‘Specificity’ data were developed from a review of 445 confirmed
normal cases (defined as those with a subsequent normal screening study). ‘FP/Image’
refers to the average number of false-positive marks per image measured on normal cases.
‘Specificity’ refers to the percentage of normal cases that, when processed, show no CAD
marks.
Note: Earlier versions of the ImageChecker algorithm (before version 8.0) produce
slightly different performance results. For further information, see the manuals provided
with those systems.
To determine false-mark rates, Hologic | R2 processes normal screening cases
(defined as those with a subsequent normal screening study) through the
ImageChecker software and measures the number of false marks per image. More
extraordinary screening cases, such as those with additional views, and or those from
diagnostic studies, may have a very different make-up of images and, as such, may
produce results that fall outside the measured normal case mark rate. Since the mark
rate is measured per image, a larger number of images in a case should correspond on
average with a higher total mark count for that case. While clinical experience
demonstrates some variation in mark rates, Hologic | R2 has not found the false-mark
rate to vary dramatically when averaged over a large number of cases.


Algorithm Sensitivity (%)
Algorithm Sensitivity (%

The following graphs shows plots of algorithm sensitivity vs. false-mark rate based on
cases with the four screening views, with data points for each of the three operating
points.

98

2
96
1
0
94

92

90

88

86
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
False Mark Rate
(False Positives/Case)

Calcification Performance

91

90
2
89

88 1
87

86
85

84

83 0
82
81
0.0 0.5 1.0 1.5 2.0
False Mark Rate
(False Positives/Case)

Mass Performance


Index

A
acquisition gantry devices DICOM 6000 interface, 8, 21
in clinical workflow, 7 in system data flow, 15
in system data flow, 13–17 digital ports. See ports, input
models supported, 7 DigitalNow, 16
algorithm, ImageChecker, 11, 23–
37 operating points, 35–37 E
what is detected by, EmphaSize, 25
28 architectural overview, 10
distortions precautions, 3
what the algorithm detects, 29
with PeerView Digital, 26 F
C false-mark rate, 11, 35–37
CAD marks, 5, 10 films, processing, 16
See also ImageChecker
archiving, 21 I
EmphaSize (variable-size), 3, image processing, 19–20
10 maximum number of, 24 for calcifications, 31
overlapping, 26 for masses, 33
types, 24 image specifications, 18
variable-size, 25 ImageChecker
Calc marks, 24 algorithm, 23–37
variable-size, 25 benefits of, 6
with EmphaSize, 25 case and image processing, 19–20
with LesionMetrics, 27 image specifications, 18
with PeerView Digital, 26 intended use, 1
calcifications, 24 overview, 5
analysis by algorithm, 31 overview of features, 9–11
case processing for, 32 performance specifications, 35–37
image processing for, 31 result formats, 21
what the algorithm detects, 28 supported views, 18
case processing, 19–20 warnings and precautions, 3–4
for calcifications, 32 implants, 18
for masses, 34 precautions, 4
Citra Core, 9
clinical workflow, 7 L
LesionMetrics
D described, 27
data flows, 13–17 overview, 10
DICOM, 18, 21
conformance statement, 21 M
image classes, Malc marks, 24
13


Index

variable-size, 25 with resources, 2
EmphaSize, 25 with results
LesionMetrics, 27 formats, 21
with PeerView Digital, 26 images selected for case processing,
Mammography CAD SR, 21 19–20
mammography reporting system, 16 review workstations, 5, 7
Mass marks, 24 features supported by, 9
variable-size, 25 in clinical workflow, 7
with EmphaSize, 25 in system data flow, 13–17
with LesionMetrics, 27 models supported, 8
with PeerView Digital, 26 RightOn CAD marks. See CAD marks
masses, 24
analysis by algorithm, 33 S
case processing for, 34 screening views, 19, 36
image processing for, 33 SecurViewdx diagnostic review
what the algorithm detects, 29 workstation, 7, 14, 17, 21,
27
Selenia FFDM System, 7
O sensitivity, 11, 35–37
operating points, 11, 35– specificity, 11, 36
37

P T
PACS, 7, 21 thresholds, algorithm. See operating
in system data flow, 13– points
17 partial view images, 4
PeerView Digital V
described, 26 variable-size marks. See EmphaSize
overview, 10 views
ports, input, 7, 9 CAD-supported, 18
precautions, 3–4 screening, 19, 36

R 40 Understanding ImageChecker 9.0
R2 CAD score, 27 – PN MAN-00851 Rev 001
R2 Cenova server, 5
clinical workflow with, 7
in system data flow, 13–17
R2 Patient ID, 16

W workstations. See acquisition gantry
warnings, 3–4 workflow, clinical, 5, 7 devices; review workstations

Man 00851 rev 001 understanding image checker 9.0

More Related Content

What's hot (17)

Similar to Man 00851 rev 001 understanding image checker 9.0 (20)

Man 00851 rev 001 understanding image checker 9.0