DOI: 10.1148/radiol.2451060760
(Radiology 2007;245:88-94.)
© RSNA, 2007
Evaluation of Computer-aided Detection Systems in the Detection of Small Invasive Breast Carcinoma1
Richard L. Ellis, MD,
Andrew A. Meade, MD,
Michelle A. Mathiason, MS,
Kathy M. Willison, RT, and
Wende Logan-Young, MD
1 From the Norma J. Vinger Center for Breast Care, Gundersen Lutheran Health System, 1900 South Ave, La Crosse, WI 54601 (R.L.E., A.A.M.); Gundersen Lutheran Medical Foundation, La Crosse, Wis (M.A.M.); and The Elizabeth Wende Breast Clinic, Rochester, NY (K.M.W., W.L.). From the 2005 RSNA Annual Meeting. Received May 1, 2006; revision requested June 27; revision received December 14; accepted December 20; final version accepted March 1, 2007.
Address correspondence to R.L.E. (e-mail: rlellis@gundluth.org).
 |
ABSTRACT
|
|---|
Purpose: To retrospectively compare two CAD systems for detecting invasive breast cancers manifesting as noncalcified masses smaller than 16 mm.
Materials and Methods: Waiver of informed consent was granted by the Institutional Review Board that approved this HIPAA-compliant study. Mammograms obtained from two institutions providing consecutive invasive carcinomas manifesting as noncalcified masses smaller than 16 mm were evaluated by using two commercially available CAD systems (R2 ImageChecker M1000, version 5.0A and iCAD Second Look, version 6.0 mid operating point). To provide statistical power to test for a possible 10% difference in the sensitivity performance between the systems, 192 consecutive mammographic studies (182 unifocal, six multifocal, and four bilateral cancers) were collected. Masses were characterized using the Breast Imaging Reporting and Data System (BI-RADS). Per study specificity and mass false marker rate were determined by using 51 normal four-view studies, while scoring only the mass false-positive marks for noncalcified masses. Associations between mass characteristics and supplying institution were compared by using 
2 tests. A P value of .05 was considered to indicate a significant difference.
Results: The respective per study sensitivity, per image (ie, per view) sensitivity, per study specificity, and mass false-positive marker rates were 81.8%, 64.7%, 39.2%, and 1.08 for the R2 ImageChecker M1000 system, and 60.9%, 42.6%, 31.4%, and 1.41 for the iCAD Second Look system. The overall per study and per image sensitivities were significantly better for R2 than for iCAD (McNemar test, all P < .001), with a nonsignificant higher per study specificity and lower mass false marker rate on normal studies. CAD results demonstrated at least a 20% variation between BI-RADS categories 4a and 5 for per study and per image sensitivity.
Conclusion: A statistically significant difference was observed in per study and per image sensitivity in our mammography data set with small (<16 mm), noncalcified invasive breast malignancies between two CAD systems. Differences in per study specificity and mass false marker rate were noted but were not statistically significant.
© RSNA, 2007
|
INTRODUCTION
|
|---|
Studies have reported a benefit from having two independent radiologist assessments (a double reading) for screening mammography to increase the sensitivity of breast cancer detection (1–5). However, due to staffing limitations and screening mammography reimbursement rates, most institutions do not use double reading. Computer systems to aid in detection of breast cancer (computer-aided detection [CAD] systems) are now available to radiologists, primarily through commercial vendors. As with double reading, prospective clinical articles (6–10) have reported that, by serving as a surrogate second reader, CAD assists radiologists in increasing the breast cancer detection rate.
In the first reported prospective screening mammography study aided with CAD, Freer and Ulissey (6) reported a 19.5% increase in the number of cancers detected in 12 860 patients. Further review of their report notes that only two additional invasive cancers were detected with CAD, along with an additional six cases of ductal carcinoma in situ. Likewise, Birdwell et al (7) reported a 7.4% increase in cancer detection, with two additional invasive cancers identified from 8682 patients, while in a similar study, Morton et al (8) reported a 7.62% increase in cancer detection, with eight additional cancers (three invasive carcinoma and five ductal carcinoma in situ) identified from the total of 18 096 patients.
However, Gur et al (9) reported no associated statistically significant change in recall or breast cancer detection rates for the 24 participating radiologists reading 59 139 screening mammograms aided with CAD compared with 56 432 mammograms read prior to the introduction of CAD into their clinical practice. For the subset of radiologists reading a lower volume of screening mammograms, there was a substantial (19.7%), but not statistically significant, increase in cancer detection rate (11). Further, they did not report data on invasive tumor size and stage before and after CAD implementation. Finally, Cupples et al (10) reported a 16.1% increase in cancer detection rate, with a 164% increase in the detection rate of invasive cancers 1.0 cm or smaller in size and an 8.1% increase in recall rate with the use of CAD.
Successful randomized controlled screening mammography trials have demonstrated that to reduce breast cancer mortality, it is critical to detect invasive breast carcinoma at an early stage, thus interrupting the disease prior to development of regional or systemic metastatic disease (12–16). Although important, detection of microcalcifications suggestive of ductal carcinoma in situ, masses 16 mm or larger, or masses of any size with associated microcalcifications were not a focus of our study. Detection of ductal carcinoma in situ in a screening population has been reported to contribute to an approximately 12% reduction in mortality (17), while the screening detection of larger invasive tumors has been reported to provide no benefit over clinical breast examination (18,19).
Therefore, CAD will be most helpful in reducing breast cancer mortality for patients participating in screening mammography only if it aids the radiologist in detecting small invasive tumors and ductal carcinoma in situ with the propensity for differentiation into invasive carcinoma. While there have been previous retrospective studies (20–22) of the ability of CAD to detect breast cancers missed by radiologists, ours is the first study, to our knowledge, to retrospectively compare two CAD systems (R2 ImageChecker M1000, version 5.0A; R2 Technology, Sunnyvale, Calif and iCAD Second Look, version 6.0 mid operating point; iCAD, Nashua, NH) for detecting invasive breast cancers manifesting as noncalcified masses smaller than 16 mm.
|
MATERIALS AND METHODS
|
|---|
Patients and Masses
In our report, the term masses includes masses, architectural distortions, and asymmetries as defined by Breast Imaging Reporting and Data System (BI-RADS, 4th edition). Our study was a collaborative effort between two breast care centers, both of which obtained waiver of informed consent from the Institutional Review Board that approved our Health Insurance Portability and Accountability Act–compliant study. Both R2 Technology and iCAD supported this study by providing the CAD systems and an operator to perform the CAD studies. The authors had control over the data and the information submitted for publication.
Consecutive noncalcified masses seen at mammography that were biopsy-proved invasive carcinoma smaller than 16 mm were included in the study. Tumor size of less than 16 mm was an arbitrary size used to achieve a population large enough to provide statistical power to demonstrate a 10% sensitivity difference in the two CAD systems. A total of 192 women (age range, 40–97 years; median, 67 years) imaged between June 15, 1999, and June 5, 2004, were included. Of these, 182 had unifocal, six had multifocal, and four had bilateral disease. Median tumor size was 10.0 mm. One institution included 127 patients and the other included 65.
Fifty-one normal four-view mammograms (51 women; age range, 38–81 years) were evaluated to determine per study specificity and the mass false-positive marker rate for the CAD systems. A four-view mammogram was declared normal provided that prior to, and for at least 1 year after mammography, screening mammogram evaluations were negative for malignancy (BI-RADS 1–2). The normal mammograms selected had no masses of any type in an effort to have a pure population of normal studies, and were thus neither randomly nor consecutively selected. To approximate the breast tissue density percentages for 51 screening mammography studies performed at the principal investigator's institution, 13 studies depicted fatty replaced tissue, 17 depicted scattered glandular tissue, 17 depicted dense glandular tissue, and four depicted extremely dense glandular tissue.
Mass Evaluation
Prior to CAD evaluation, the precise location and BI-RADS characteristics for each cancer had been determined by the three study radiologists (R.L.E., A.A.M., and W.L., with 12, 22, and 35 years breast imaging experience, respectively). The radiologists had full knowledge of the mass location from imaging and biopsy results. To ensure uniformity of mass characterization, all mammographic studies were reviewed by the principal investigator (R.L.E.), a fellowship subspecialty-trained radiologist in breast imaging (2003 audit performance: 10 808 screening mammograms, 4.5% screening recall rate, 11-mm median size screening-detected invasive cancer, 48.6% biopsy positive predictive value for BI-RADS 4–5 lesions).
Each study was evaluated for several characteristics: (a) mass location (clock position), focality (unifocal, multifocal, and multicentric), and laterality (right, left, and bilateral); (b) BI-RADS category and mass characterization (mass shape, margins, density, and level of suspicion); (c) visibility of the mass on one or both views (craniocaudal [CC] and/or mediolateral oblique [MLO]); and (d) histologic information (tumor size, type, and grade).
CAD Analysis
Routine mammographic studies (CC and MLO views) were evaluated by using the latest United States Food and Drug Administration-approved software (as of June 2004) for the two commercially available CAD systems. Mammograms were digitally scanned by each CAD system (R2 ImageChecker at 50-µm resolution; iCAD Second Look at 43.5-µm resolution) and analyzed with proprietary software. CAD results were provided on a paper printout for review, and final scoring for each system was performed.
R2 ImageChecker uses an asterisk to mark masses, and iCAD Second Look uses an oval. In our study, only CAD markings for masses were evaluated. Each vendor supplied its own CAD system and operator to scan the mammograms and provide the paper printouts at the principal investigator's institution. The vendor operators were blinded to the findings of mass evaluation acquired by the principal investigator. Mammograms were divided between the two operators. When the operators finished scanning their half of the mammograms, they traded with each other to complete the entire set.
CAD Scoring
By using the original mammographic studies and the CAD printout results, each study was scored by the principal investigator (R.L.E.) for detection with the CAD systems. A single image (view) containing the cancer was considered correctly marked if the CAD mark was within the boundaries of the previously defined cancer location. In mammographic studies where the mass is seen on both CC and MLO views, each view was scored separately to determine per image sensitivity. For studies analyzed with the R2 ImageChecker, the center of the asterisk must lie over any portion of the mass to be scored as a true-positive result. For mammograms analyzed by the iCAD Second Look, the center of the oval must lie over any portion of the mass to be scored a true-positive result. None of the studies presented a challenge to determine CAD mass scoring by using the above definitions.
CAD per study sensitivity was calculated as the number of correctly marked cancers on either the CC or MLO mammogram from the total number of studies containing biopsy-proved cancers. CAD per image sensitivity was calculated as the number of correctly marked cancers from the total number of cancers identified from the CC and MLO views on each study. In addition, subgroup analysis was provided for each CAD system to determine if any recorded mass variables were significant for studies where the masses were not detected. The mass false marker rate was calculated from the 51 normal mammographic studies on the basis of a four-view mammogram. The total number of mass marks per study was tallied, and the average number of mass marks over all the normal studies was calculated to provide the mass false-positive marker rate per nromal study. CAD per study specificity was calculated as the percentage of 51 normal studies without false-positive marks.
Statistical Analysis
Analysis was performed with a statistical software package (SAS, version 9.1; SAS Institute, Cary, NC) and recorded in a database (Access; Microsoft, Redmond, Wash). Generalized estimating equations analysis was performed and showed independence of bilateral breast cancer masses. Results are presented per mass unless otherwise stated. The McNemar test was used to determine statistical significance for matched group sensitivity. Associations between mass characteristics and supplying institution were compared by using 2 tests. A P value of .05 was considered to indicate a significant difference.
|
RESULTS
|
|---|
Given the selection criterion of consecutive invasive breast cancers smaller than 16 mm, our study population (Table 1) displayed a wide variation in BI-RADS assessment categories (Table 2), BI-RADS characteristics (Table 3), and histologic tumor types (Table 4). In 192 patients, 202 cancers with 380 visible (24 seen on only one view and 178 seen on both CC and MLO views) noncalcified masses were identified (Table 1).
Two CAD Systems
The R2 ImageChecker system correctly identified at least one mass from the two-view mammogram in 157 of 192 studies, providing a per study sensitivity of 81.8%. Of 380 masses visible on the CC and MLO views, R2 ImageChecker correctly identified 246, providing an image sensitivity of 64.7%. Per study specificity was 39.2% (20 of 51), while the average mass false-positive marker rate was 1.08 (55 of 51). The iCAD Second Look system correctly identified at least one mass from the two-view mammogram in 117 of 192 studies, providing a per study sensitivity of 60.9%. Of 380 masses visible on the CC and MLO views, iCAD Second Look system correctly identified 162, providing an image sensitivity of 42.6%. Per study specificity was 31.4% (16 of 51), while the average mass false-positive marker rate was 1.41 (72 of 51) (Table 5). R2 correctly marked 51 masses missed by iCAD, and iCAD correctly marked 11 masses missed by R2. Likewise, both systems correctly marked 106 masses and missed 24 masses. Of the 24 masses missed by both systems, 12.6% (16 of 127) were from institution A and 12.3% (eight of 65) were from institution B.
Masses missed by both systems were BI-RADS 4a or 4b and categorized as lesions of equal or lower density to surrounding tissue (46%, 11 of 24), embedded in or at the margin of glandular tissue of similar density to surrounding tissue (29%, seven of 24), near the nipple (13%, three of 24), overlying the pectoralis muscle (4%, one of 24), very small lesion in a fatty replaced breast (4%, one of 24), or lying along the posterior margin of the image (4%, one of 24). Seventy-one percent (17 of 24) of the lesions measured smaller than 11 mm.
CAD Systems and Institutions
Statistical analysis was performed for BI-RADS assessment categories and mass shape, margins, and density for the cancers from both institutions. When comparing the studies provided by the participating institutions (Tables 6, 7), the CAD systems' performance varied. Although a statistical difference was found for the BI-RADS mass characteristics between the two institutions, the major difference between R2 and iCAD performance was mass density (Table 8). Specifically, the sensitivity of iCAD was statistically inferior to that of R2 in masses of density equal to that of the surrounding glandular tissue, a difference more evident in mammographic studies from institution A than in those from institution B. Further, a study and imaging sensitivity difference greater than 20% exists when comparing the results of BI-RADS category 4a to BI-RADS category 5 masses for both CAD systems (Table 9).
Normal Mammographic Studies
Of 51 normal mammographic studies, both CAD systems placed a mass mark in the same region on 16 of them. These 16 studies have been reviewed, and all but two of the normal study patients have had 2 or more years of subsequent normal mammograms.
|
DISCUSSION
|
|---|
The primary purpose of screening mammography is to reduce breast cancer mortality. Prognostic indicators for breast cancer survival include tumor size, type, and grade; mammographic presentation; regional lymph node status; and metastasis. However, tumor size acts as a governor, directly influencing the other prognostic indicators. To reduce breast cancer mortality, invasive breast carcinoma must be detected at an early phase, prior to the development of regional or systemic metastatic disease. Therefore, CAD systems are most beneficial only if they aid the radiologist in detecting those breast cancers that have the greatest potential for metastatic spread while they remain confined to the breast.
Given the finite number of breast cancers in any particular population at any given time, the potential benefit for CAD systems is indirectly proportional to the performance of the radiologist interpreting screening mammograms (23). Benefits from CAD could be reflected in either an additional number of breast cancers detected and/or detection of the same number of breast cancers but with improved primary prognostic indicators (ie, smaller size), as noted in Cupples et al (10).
Our results suggest that for the CAD systems evaluated, there is a need for continued improvement in detection of invasive breast carcinomas smaller than 16 mm, especially for those in BI-RADS category 4a or 4b, to achieve the authors' desired goals for a study sensitivity 90% or higher, image sensitivity 70% or higher, study specificity 50% or higher, and a false marker rate 1 or lower. Vyborny et al (24) noted similar findings where the CAD system used in their study identified 45% (40 of 88) of the cancers defined as having subtle spiculated masses.
Other studies comparing R2 with iCAD have resulted in similar suggestions for the evaluation of masses and for architectural distortion (25,26), indicating that improvements are needed. Although theirs was not a comparative study, Soo et al (27) concluded that radiologists should not become overly reliant on CAD because of low sensitivity for helping identify possible amorphous calcifications (ImageChecker M1000, version 3.2; R2 Technology) (27). The same message is advised for noncalcified masses smaller than 16 mm in BI-RADS categories 4a or 4b, where study sensitivity was lower than 81% and image sensitivity was lower than 61%.
Sensitivity and false-positive marker rate are measurements commonly used in assessing the performance of a CAD algorithm. However, results are dependent on the composition of the mammographic study set being analyzed by the CAD system. For instance, two articles have reported iCAD Second Look study sensitivity for lesions manifesting mammographically as masses and have stratified performance on the basis of mass size. Malich et al (28) measured mass size from mammography, while Brem et al (29) used pathologic mass size, as used in our study. Both articles report CAD performance with the original Food and Drug Administration-approved version of iCAD Second Look version 3.4 in the United States (29) and version 3.5 in Europe (28), while we report performance with an updated version (version 6.0 mid operating point).
Malich et al also stratifies masses on the basis of histopathologic findings, so a set of mammograms with selection criteria similar to those of our study can be assessed, although they included masses with associated calcifications, while we excluded such masses. Combining all invasive cancers manifesting mammographically as masses smaller than 10 mm, their CAD study sensitivity was 81% (21 of 26), and for masses 20 mm or less in size, CAD study sensitivity was 92% (59 of 64). Brem et al (29) did not stratify masses on the basis of histopathologic findings, so the set most comparable to ours would be masses smaller than 16 mm. Brem et al excluded masses with associated calcifications, as we did. Combining all masses smaller than 16 mm in that study, CAD study sensitivity was 85% (84 of 99).
Therefore, it might be argued that the study sensitivity we report as 60.9% for iCAD Second Look could be compared with 81%–92% from Malich et al and with 85% from Brem et al. However, the differences we discussed (ie, method of assessing mass size, masses with associated calcifications, and CAD version) may contribute in part to these differences in reported performance. Further, the differences may also be an example of a more challenging study set in our study for CAD detection and subsequent performance. Nishikawa et al (30) reviewed the effect of study selection on CAD performance and note that a 20% change in the studies comprising the data set can affect the measured sensitivity by 15%–25%. Variability in study set composition and performance is confirmed in our own study when we compare the results for either CAD system on the consecutive studies collected at the participating centers (n = 127 and n = 65).
Regardless of any level of performance by CAD systems, the action taken by radiologists for CAD marks for both true- and false-positive results provided on the mammogram remains a vital component. The reported combined prevalence and incident cancer rate per 1000 women over age 40 is three to five. Given a false-positive marker rate of one per study, and five cancers per 1000 screening patients, a radiologist would need to correctly respond with a recall request for the five cancer patients (assuming these studies were correctly marked) and dismiss the vast majority of the 995 false marks to perform successfully by using a CAD system. The number of false-positive marks can distract the radiologist and reduce the validity of those marks that may indeed help detect a cancer.
Taylor et al (31) reported that 60% of false-negative screening mammograms are due to detection errors (masses suggestive of cancer overlooked), while 40% are due to decision errors (masses detected but reader decides not to recall the patient for further evaluation). Therefore, appropriate orientation and training for a new CAD system into a screening mammography program is important to ensure that the systems are used properly to maximize their potential benefit. It should be incumbent on each institution that uses CAD to determine the value of its system for cancer rate detection, primary prognostic variables for the detected cancers, recall rates, and financial implications both before and after the installation of CAD. Given the variable performance among radiologists in the interpretation of a screening mammogram, the value of CAD will also vary.
In conclusion, a statistically significant difference was observed in study and image sensitivity in our mammography dataset with small (<16 mm), noncalcified invasive breast malignancies between two commercially available CAD systems (R2 ImageChecker M1000 and iCAD Second Look). Differences in both study specificity and mass false marker rate were noted but were not statistically different.
|
ADVANCES IN KNOWLEDGE
|
|---|
- We demonstrated a significant difference (P = .001) in sensitivity between two computer-aided detection (CAD) systems tested for tumors smaller than 16 mm.
- There was a broad range in the CAD sensitivities among the Breast Imaging Reporting and Data System categories for masses smaller than 16 mm.
- The CAD sensitivities we report are lower than those in prior reports with similar size criteria, indicating the fundamental importance that a particular mammographic study set has in CAD performance.
|
ACKNOWLEDGMENTS
|
|---|
The authors gratefully acknowledge Cathy Mikkelson Fischer, MA, for editing the manuscript.
|
FOOTNOTES
|
|---|
Abbreviations: BI-RADS = Breast Imaging Reporting and Data System • CAD = computer-aided detection • CC = craniocaudal • MLO = mediolateral oblique
See Materials and Methods for pertinent disclosures.
Author contributions: Guarantor of integrity of entire study, R.L.E.; study concepts/study design or data acquisition or data analysis/interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; manuscript final version approval, all authors; literature research, R.L.E.; clinical studies, R.L.E., K.M.W., W.L.; experimental studies, R.L.E.; statistical analysis, R.L.E., M.A.M.; and manuscript editing, all authors
|
References
|
|---|
- Thurfjell EL, Lernevall KA, Taube AA. Benefit of independent double reading in a population-based mammography screening program. Radiology 1994;191:241–244. [Abstract/Free Full Text]
- Anderson ED, Muir BB, Walsh JS, Kirkpatrick AE. The efficacy of double reading mammograms in breast screening. Clin Radiol 1994;49:248–251. [CrossRef][Medline]
- Thurfjell E. Mammography screening: one versus two views and independent double reading. Acta Radiol 1994;35:345–350.
- Warren RM, Duffy SW. Comparison of single reading with double reading of mammograms, and change in effectiveness with experience. Br J Radiol 1995;68:958–962. [CrossRef][Medline]
- Ciatto S, Del Turco MR, Morrone D, et al. Independent double reading for screening mammograms. J Med Screen 1995;2:99–101. [Medline]
- Freer TW, Ulissey MJ. Screening mammography with computer-aided detection: prospective study of 12 860 patients in a community breast center. Radiology 2001;220:781–786. [Abstract/Free Full Text]
- Birdwell RL, Bandodkar P, Ikeda DM. Computer-aided detection with screening mammography in a university hospital setting. Radiology 2005;236:451–457. [Abstract/Free Full Text]
- Morton MJ, Whaley DH, Brandt KR, Amrami KK. Screening mammograms: interpretation with computer-aided detection—prospective evaluation. Radiology 2006;239:375–383. [Abstract/Free Full Text]
- Gur D, Sumkin JH, Rockette HE, et al. Changes in breast cancer detection and mammography recall rates after the introduction of a computer-aided detection system. J Natl Cancer Inst 2004;96:185–190. [Abstract/Free Full Text]
- Cupples TE, Cunningham JE, Reynolds JC. Impact of computer-aided detection in a regional screening mammography program. AJR Am J Roentgenol 2005;185:944–950. [Abstract/Free Full Text]
- Feig SA, Sickles EA, Evans WP, Linver MN. Re: Changes in breast cancer detection and mammography recall rates after the introduction of a computer-aided detection system [letter]. J Natl Cancer Inst 2004;96:1260–1261; author reply 1261. [Free Full Text]
- Michaelson JS, Satija S, Kopans D, et al. Gauging the impact of breast carcinoma screening in terms of tumor size and death rate. Cancer 2003;98:2114–2124. [CrossRef][Medline]
- Michaelson JS, Silverstein M, Wyatt J, et al. Predicting the survival of patients with breast carcinoma using tumor size. Cancer 2002;95:713–723. [CrossRef][Medline]
- Tabar L, Duffy SW, Vitak B, Chen HH, Prevost TC. The natural history of breast carcinoma: what have we learned from screening? Cancer 1999;86:449–462. [CrossRef][Medline]
- Tabar L, Vitak B, Chen HH, Yen MF, Duffy SW, Smith RA. Beyond randomized controlled trials: organized mammographic screening substantially reduces breast carcinoma mortality. Cancer 2001;91:1724–1731. [CrossRef][Medline]
- Tabar L, Dean PB. Mammography and breast cancer: the new era. Int J Gynaecol Obstet 2003;82:319–326. [CrossRef][Medline]
- Duffy SW, Tabar L, Vitak B, et al. The relative contributions of screen-detected in situ and invasive breast carcinomas in reducing mortality from the disease. Eur J Cancer 2003;39:1755–1760. [CrossRef][Medline]
- Miller AB, To T, Baines CJ, Wall C. The Canadian National Breast Screening Study-1: breast cancer mortality after 11 to 16 years of follow-up—a randomized screening trial of mammography in women age 40 to 49 years. Ann Intern Med 2002;137:305–312. [Abstract/Free Full Text]
- Miller AB, To T, Baines CJ, Wall C. Canadian National Breast Screening Study-2: 13-year results of a randomized trial in women aged 50–59 years. J Natl Cancer Inst 2000;92:1490–1499. [Abstract/Free Full Text]
- Destounis SV, DiNitto P, Logan-Young W, Bonaccio E, Zuley ML, Willison KM. Can computer-aided detection with double reading of screening mammograms help decrease the false-negative rate? initial experience. Radiology 2004;232:578–584. [Abstract/Free Full Text]
- Warren Burhenne LJ, Wood SA, D'Orsi CJ, et al. Potential contribution of computer-aided detection to the sensitivity of screening mammography. Radiology 2000;215:554–562. [Abstract/Free Full Text]
- Birdwell RL, Ikeda DM, O'Shaughnessy KF, Sickles EA. Mammographic characteristics of 115 missed cancers later detected with screening mammography and the potential utility of computer-aided detection. Radiology 2001;219:192–202. [Abstract/Free Full Text]
- Balleyguier C, Kinkel K, Fermanian J, et al. Computer-aided detection (CAD) in mammography: does it help the junior or the senior radiologist? Eur J Radiol 2005;54:90–96.
- Vyborny CJ, Doi T, O'Shaughnessy KF, Romsdahl HM, Schneider AC, Stein AA. Breast cancer: importance of spiculation in computer-aided detection. Radiology 2000;215:703–707. [Abstract/Free Full Text]
- Gur D, Stalder JS, Hardesty LA, et al. Computer-aided detection performance in mammographic examination of masses: assessment. Radiology 2004;233:418–423. [Abstract/Free Full Text]
- Baker JA, Rosen EL, Lo JY, Gimenez EI, Walsh R, Soo MS. Computer-aided detection (CAD) in screening mammography: sensitivity of commercial CAD systems for detecting architectural distortion. AJR Am J Roentgenol 2003;181:1083–1088. [Abstract/Free Full Text]
- Soo MS, Rosen EL, Xia JQ, Ghate S, Baker JA. Computer-aided detection of amorphous calcifications. AJR Am J Roentgenol 2005;184:887–992. [Abstract/Free Full Text]
- Malich A, Sauner D, Marx C, et al. Influence of breast lesion size and histologic findings on tumor detection rate of a computer-aided detection system. Radiology 2003;228:851–856. [Abstract/Free Full Text]
- Brem RF, Hoffmeister JW, Zisman G, DeSimio MP, Rogers SK. A computer-aided detection system for the evaluation of breast cancer by mammographic appearance and lesion size. AJR Am J Roentgenol 2005;184:893–896. [Abstract/Free Full Text]
- Nishikawa RM, Giger ML, Doi K, et al. Effect of case selection on the performance of computer-aided detection schemes. Med Phys 1994;21:265–269. [CrossRef][Medline]
- Taylor P, Given-Wilson R, Champness J, Potts HW, Johnston K. Assessing the impact of CAD on the sensitivity and specificity of film readers. Clin Radiol 2004;59:1099–1105. [CrossRef][Medline]
TOP
ABSTRACT
INTRODUCTION
