Published online before print September 9, 2004, 10.1148/radiol.2332031200
(Radiology 2004;233:411.)
A more recent version of this article appeared on November 1, 2004
Computer-aided Detection in Screening Mammography: Variability in Cues1
Jay A. Baker, MD,
Joseph Y. Lo, PhD,
David M. Delong, PhD and
Carey E. Floyd, PhD
1 From the Departments of Radiology and Biomedical Engineering, Duke University Medical Center, Erwin Rd, Durham, NC 27710. From the 2003 RSNA scientific assembly. Received July 28, 2003; revision requested October 6; revision received January 24, 2004; accepted March 2. Address correspondence to J.A.B. (e-mail: jay.baker@duke.edu).
View larger version (24K):
[in a new window]
|
Figure 1. Histogram depicts the number of true-positive analyses out of 10 CAD runs for 50 breast cancers detected at screening by using case-based evaluation (ie, true-positive CAD cue in either mammographic view qualifies as true-positive for the case). Fourteen cases were marked between one and nine times but not all 10 times. Therefore, histogram illustrates that the CAD system provided inconsistent results for 14 of 50 (28%) malignancies in the present study.
|
|
View larger version (103K):
[in a new window]
|
Figure 2a. (a) Craniocaudal (left) and mediolateral oblique (right) mammograms in a 64-year-old woman with invasive ductal carcinoma detected as a 9-mm spiculated mass (arrows) at screening mammography. (b) Spot compression magnification view demonstrates the spiculated mass (arrow) in a to better advantage. (c) Photograph of computer monitor display of CAD system output. Asterisk overlying spiculated mass in the mediolateral oblique image (right) confirms accurate detection by the CAD system. Spiculated mass is noncalcified, and triangle overlying lateral right breast in craniocaudal view (left) indicates location of an artifact. (d) Photograph of computer monitor display of repeat CAD analysis. An asterisk does not overlie the mass in either projection, indicating false-negative analysis by the CAD system. This mass was detected in eight of 10 CAD runs.
|
|
View larger version (164K):
[in a new window]
|
Figure 2b. (a) Craniocaudal (left) and mediolateral oblique (right) mammograms in a 64-year-old woman with invasive ductal carcinoma detected as a 9-mm spiculated mass (arrows) at screening mammography. (b) Spot compression magnification view demonstrates the spiculated mass (arrow) in a to better advantage. (c) Photograph of computer monitor display of CAD system output. Asterisk overlying spiculated mass in the mediolateral oblique image (right) confirms accurate detection by the CAD system. Spiculated mass is noncalcified, and triangle overlying lateral right breast in craniocaudal view (left) indicates location of an artifact. (d) Photograph of computer monitor display of repeat CAD analysis. An asterisk does not overlie the mass in either projection, indicating false-negative analysis by the CAD system. This mass was detected in eight of 10 CAD runs.
|
|
View larger version (115K):
[in a new window]
|
Figure 2c. (a) Craniocaudal (left) and mediolateral oblique (right) mammograms in a 64-year-old woman with invasive ductal carcinoma detected as a 9-mm spiculated mass (arrows) at screening mammography. (b) Spot compression magnification view demonstrates the spiculated mass (arrow) in a to better advantage. (c) Photograph of computer monitor display of CAD system output. Asterisk overlying spiculated mass in the mediolateral oblique image (right) confirms accurate detection by the CAD system. Spiculated mass is noncalcified, and triangle overlying lateral right breast in craniocaudal view (left) indicates location of an artifact. (d) Photograph of computer monitor display of repeat CAD analysis. An asterisk does not overlie the mass in either projection, indicating false-negative analysis by the CAD system. This mass was detected in eight of 10 CAD runs.
|
|
View larger version (117K):
[in a new window]
|
Figure 2d. (a) Craniocaudal (left) and mediolateral oblique (right) mammograms in a 64-year-old woman with invasive ductal carcinoma detected as a 9-mm spiculated mass (arrows) at screening mammography. (b) Spot compression magnification view demonstrates the spiculated mass (arrow) in a to better advantage. (c) Photograph of computer monitor display of CAD system output. Asterisk overlying spiculated mass in the mediolateral oblique image (right) confirms accurate detection by the CAD system. Spiculated mass is noncalcified, and triangle overlying lateral right breast in craniocaudal view (left) indicates location of an artifact. (d) Photograph of computer monitor display of repeat CAD analysis. An asterisk does not overlie the mass in either projection, indicating false-negative analysis by the CAD system. This mass was detected in eight of 10 CAD runs.
|
|
View larger version (131K):
[in a new window]
|
Figure 3a. (a) Craniocaudal (left) and mediolateral oblique (right) mammograms in a 69-year-old woman with invasive ductal carcinoma detected as a 10-mm mass with an obscured margin (arrows) at screening mammography. (b) Photograph of computer monitor display of CAD system output. Asterisk (arrow) overlying breast mass in the craniocaudal projection (left) confirms accurate detection by the CAD system. Triangle overlying the inferior breast on the mediolateral oblique view (right) indicates location of an artifact. (c) Photograph of computer monitor display of repeat CAD analysis. An asterisk does not overlie the mass in either projection, indicating false-negative analysis by the CAD system. This mass was detected in seven of 10 CAD runs. LCC = left craniocaudal, LMLO = left mediolateral oblique.
|
|
View larger version (149K):
[in a new window]
|
Figure 3b. (a) Craniocaudal (left) and mediolateral oblique (right) mammograms in a 69-year-old woman with invasive ductal carcinoma detected as a 10-mm mass with an obscured margin (arrows) at screening mammography. (b) Photograph of computer monitor display of CAD system output. Asterisk (arrow) overlying breast mass in the craniocaudal projection (left) confirms accurate detection by the CAD system. Triangle overlying the inferior breast on the mediolateral oblique view (right) indicates location of an artifact. (c) Photograph of computer monitor display of repeat CAD analysis. An asterisk does not overlie the mass in either projection, indicating false-negative analysis by the CAD system. This mass was detected in seven of 10 CAD runs. LCC = left craniocaudal, LMLO = left mediolateral oblique.
|
|
View larger version (140K):
[in a new window]
|
Figure 3c. (a) Craniocaudal (left) and mediolateral oblique (right) mammograms in a 69-year-old woman with invasive ductal carcinoma detected as a 10-mm mass with an obscured margin (arrows) at screening mammography. (b) Photograph of computer monitor display of CAD system output. Asterisk (arrow) overlying breast mass in the craniocaudal projection (left) confirms accurate detection by the CAD system. Triangle overlying the inferior breast on the mediolateral oblique view (right) indicates location of an artifact. (c) Photograph of computer monitor display of repeat CAD analysis. An asterisk does not overlie the mass in either projection, indicating false-negative analysis by the CAD system. This mass was detected in seven of 10 CAD runs. LCC = left craniocaudal, LMLO = left mediolateral oblique.
|
|
View larger version (25K):
[in a new window]
|
Figure 4. Histogram depicts the number of true-positive analyses in 10 CAD runs for 100 mammographic views of 50 breast cancers detected at screening by using image-based evaluation (ie, each mammographic view considered separately). Because of inconsistency in CAD output, a malignancy was marked correctly between one and nine times but not all 10 times for 33 of the 100 mammographic views (33%).
|
|
Copyright © 2004 by the Radiological Society of North America.
