Published online before print October 19, 2006, 10.1148/radiol.2413051366
(Radiology 2006;241:689-694.)
© RSNA, 2006
Ductal Carcinoma in Situ: Computer-aided Detection in Screening Mammography1
Vidya R. Pai, MD,
Nancy E. Gregory, MD,
Ann E. Swinford, MD, MS and
Murray Rebner, MD
1 From the Department of Diagnostic Radiology, William Beaumont Hospital, 3601 W Thirteen Mile Rd, Royal Oak, MI 48073. From the 2004 RSNA Annual Meeting. Received August 15, 2005; revision requested November 2; revision received December 15; accepted January 11, 2006; final version accepted March 17.
Address correspondence to M.R. (e-mail: mrebner{at}beaumont.edu).
 |
ABSTRACT
|
|---|
Purpose: To retrospectively evaluate the sensitivity of computer-aided detection (CAD) in depicting ductal carcinoma in situ (DCIS) on screening mammograms by using biopsy proved lesion location as the reference standard.
Materials and Methods: Institutional review board approval was obtained, with a waiver of patient informed consent for this HIPAA-compliant study. Findings of all image-guided biopsies with a pathologic diagnosis of DCIS during a 1-year period were reviewed. Fifty-eight lesions in 55 women (average age, 61.41 years ± 12.89 [standard deviation]) were available for review. The screening mammogram of the affected breast and, if available, the prior screening mammogram were digitized by the CAD system. An assessment was then made as to whether the CAD system marked the area of DCIS on the current and prior mammograms. Patient age, location and mammographic size of the lesion, type of lesion, and breast density were recorded and were analyzed by using 
2, Fisher exact, or Cochran-Mantel-Haenzel tests, where applicable.
Results: CAD identified DCIS in 53 (91%) of 58 lesions on craniocaudal (CC) and mediolateral oblique (MLO) views of screening mammograms obtained in the year of the diagnosis. On screening mammograms obtained prior to the year of the diagnosis (34 patients), no radiologically or CAD-detected lesion was present on 11 (32%) of 34 mammograms. CAD identified DCIS in 16 (70%) of 23 lesions on one of the two views. Seven (30%) of 23 lesions had mammographic findings at retrospective review that were not identified with CAD.
Conclusion: CAD had a high sensitivity in the depiction of DCIS.
© RSNA, 2006
|
INTRODUCTION
|
|---|
Mammography remains the reference standard for screening for breast cancer, although the sensitivity and specificity are not ideal. Findings of retrospective studies on interval cancers have shown a substantial false-negative rate, with prior mammograms showing some abnormality in up to half of the cases (1). Double reading has been shown to increase cancer detection rates by 15%–20% (1–3). Although a second radiologist would be ideal to act as the second reader, it is more efficient and economical to have this double reading performed by using computer-aided detection (CAD). Findings of studies have shown increased cancer detection when CAD and radiologists are combined (1,4). The ImageChecker CAD system (version 3.2; R2 Technology, Los Altos, Calif) has a reported sensitivity of 86% for masses and 98% for calcifications, with a false-positive rate of 2.06 per case (a four-view screening mammogram) (5).
To our knowledge, no study has been performed with regard to the detection of DCIS by using CAD. Thus, the purpose of our study was to retrospectively evaluate the sensitivity of CAD in depicting ductal carcinoma in situ (DCIS) on screening mammograms by using biopsy proved lesion location as the reference standard.
|
MATERIALS AND METHODS
|
|---|
Patients
Institutional review board approval was obtained, with a waiver of patient informed consent for this retrospective study. The study was compliant with Health Insurance Portability and Accountability Act. In 2001, a diagnosis of DCIS was made in 113 patients who had undergone image-guided core-needle biopsy. Fifty-eight patients were excluded (outside mammograms were no longer available for 34 patients or mammograms were either missing or released to 24 patients). This left 55 patients for review, three of whom had two lesions each, for a total of 58 lesions. Ultrasonographically (US) guided core-needle biopsy was performed for two of 58 lesions, and a stereotactic core-needle biopsy was performed for the remaining lesions. The average age of the 55 women was 61.41 years ± 12.89 (standard deviation). Of the 55 women, 34 had prior screening mammograms for review. At surgery, the diagnosis was upgraded to invasive ductal carcinoma in seven (13%) of 55 women. Three (5%) of 55 women had no surgical follow-up at our institution.
Mammogram Digitization
The screen-film mammogram (craniocaudal [CC] and mediolateral oblique [MLO] views) of the breast with a diagnosis of cancer in 2001 (prior to biopsy will be referred to as the index study) and the prior screening mammogram of the affected side were digitized by the CAD system (ImageChecker; R2 Technology). The CAD system digitizes each film to a spatial resolution of 50 µm with a 12-bit gray-scale range. The digital image is then analyzed by the computer, which marks areas of concern such as calcification, clusters, masses, and architectural distortion.
Review of Images
These mammograms were then reviewed by a single radiologist (V.R.P., with 9 years of post–breast imaging fellowship experience), who recorded the mammographic lesion size (in millimeters), type (mass, calcification, architectural distortion), and location (quadrant, subareolar); the density of the breast (dense, heterogeneously dense, scattered fibroglandular tissue, or fatty); and the type of intervention performed (stereotactic or US-guided core-needle biopsy). For the index study (2001), a judgment was also made if CAD marked the finding on the CC or the MLO view within 1 cm of the known DCIS. The radiologist knew the location of the diagnosed cancer while reviewing the mammograms, as this area had been marked for diagnostic evaluation and subsequent biopsy. The radiologist only reviewed the original mammographic reports if a discrepancy was present or if the mammograms were not marked. For the prior screening mammogram, a judgment was first made whether any retrospective finding was present, and if present, whether CAD marked it as being within 1 cm on either the CC or the MLO view.
Statistical Analysis
All variables were first analyzed to help determine if they met the distributional assumptions of the statistical tests that were being used to analyze them. On the basis of these preliminary numeric and graphic techniques, parametric, nonparametric, or exact statistical tests were used to analyze the data. Categoric variables were described by using frequencies and row and column percentages and were analyzed in contingency tables by using 2, Fisher exact, or Cochran-Mantel-Haenszel tests, wherever appropriate. Continuous variables were described by using means, standard deviations, medians, minimums, and maximums. Statistics with 95% confidence intervals (CIs) based on asymptotic standard errors were computed to assess the agreement rate between the CAD reading of the two views, and the McNemar test was used to assess the significance of this agreement rate. Sensitivity, 1 – sensitivity, standard error, and the corresponding 95% CIs were computed for each view.
The potential predictor variables of the side (left or right), patient age (in years), location of the lesion, mammographic lesion size (in millimeters), lesion type, type of core-needle biopsy (stereotactic or US guided), and density of the breast (dense, heterogeneously dense, scattered fibroglandular, or fatty) were compared with the findings in each view by using stepwise multiple logistic regression based on binary models that use logits to help determine the most parsimonious subset of variables that best explained the likelihood of a positive finding with CAD. The corresponding odds ratios and Wald test 95% CIs were also computed for each of these predictor variables. The analysis was also repeated after the results from the two views for each patient were combined. P values less than .05 (probability of type I error) were considered to indicate a significant difference throughout this study.
Statistical analysis was performed by using software (SAS for Windows, version 8.02; SAS Institute, Cary, NC). Dense and heterogeneously dense breasts were combined and scattered fibroglandular and fatty breasts were combined for statistical analysis in order to have sufficient data in each category. Any lesion with a mass (pure mass or mass with calcification) was placed in one category and was compared with a combined category of pure calcifications and calcifications and architectural distortion. Again, this was performed to obtain sufficient data in each category.
|
RESULTS
|
|---|
Fifty-eight lesions in 55 patients were identified with a pathologic diagnosis of DCIS at image-guided core-needle biopsy.
Diagnostic Accuracy of CAD
Table 1 presents the pertinent details of the computations of the CIs. The diagnostic accuracy of CAD in marking the DCIS on either view was 91% (53/[53 + 5]) (95% CI: 0.84, 0.99). The diagnostic accuracy of CAD in marking the DCIS on the CC view was 79% (46/[46 + 12]) (95% CI: 0.69, 0.90), and that on the MLO view was 83% (48/[48 + 10]) (95% CI: 0.73, 0.93).
Lesion Size and Location and Breast Density
The mean mammographic lesion size was 13.81 mm ± 17.82 along its longest axis. Of a total of 58 lesions, 21 (36%) were in the right breast and 37 (64%) were in the left. Twenty-one (36%) of 58 cancers were in the upper outer quadrant, five (9%) were in the upper inner quadrant, four (14%) were in the lower outer quadrant, four (7%) were in the lower inner quadrant, nine (15%) were in the 12-o'clock position, four (7%) were in the 9-o'clock position, one (2%) was in the 6-o'clock position, eight (14%) were in the 3-o'clock position, and two (3%) were in the central or subareolar position. Forty (73%) of the 55 breasts were of scattered fibroglandular tissue, 13 (23%) were heterogeneously dense, one (2%) was fatty, and one (2%) was dense.
Calcifications and Masses
Fifty-two (90%) of 58 lesions had calcifications only, one had a mass only, four had masses and calcifications, and one had architectural distortion and calcification (Table 2). The CAD system did not identify five cancers on either view; including a 7-mm mass with calcifications (Fig 1), a 3-mm cluster of calcifications, a 4-mm cluster of calcifications (Fig 2), a 5-mm cluster of calcifications, and a 7-mm mass.
View larger version (84K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 1: Mammograms in a 72-year-old woman with partly calcified 7-mm mass in central medial part of right breast. Magnification MLO (left) and CC (right) views demonstrate mostly circumscribed mass with internal granular-type calcifications. Neither the mass nor calcifications in the DCIS lesion were identified with CAD on either view.
|
|
View larger version (69K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 2: Mammograms in a 68-year-old woman with 4-mm cluster of calcifications in the 12-o'clock position of left breast, which were not identified with CAD. Magnification MLO (left) and CC (right) views show faint pleomorphic granular-type calcifications in the DCIS lesion.
|
|
At multiple logistic regression analysis (Table 3), the only variable that was statistically significant as a predictor of the likelihood of a positive finding of CAD was the lesion type on the CC view and on combined CC and MLO views. The data showed that calcifications were more readily identified than were masses. None of the variables were a significant predictor of positive findings on the basis of the MLO view.
Review of Prior Mammograms
Among the 34 prior mammograms, the mean number of months when the prior mammogram was obtained before the index study was 23.35 months ± 19.56 (median, 16.5 months; minimum, 1 months; maximum, 90 months). Eleven of the prior mammograms did not have a mammographic finding or a CAD finding at retrospective review. Among the 23 mammograms with findings, CAD marked 12 findings on both the CC and MLO views, three on the CC views only, and one on the MLO view only. The sensitivity of CAD was 70% (16 of 23) on either view, 65% (15 of 23) on the CC view, and 56% (13 of 23) on the MLO view. Seven (30%) of the 23 findings were missed at CAD, although they were identified on mammograms at retrospective review. Specificities could not be computed since there were no patients with either true-negative or false-positive findings. The agreement rate (
value) between the two views was weak, with a magnitude of 0.3282 (exact P = .774, McNemar test; 95% CI for value: 0.0302, 0.6262). For the 23 mammograms with findings on retrospective review, the agreement rate ( value) between the two views was moderately strong, with a magnitude of 0.7580 (exact P = .625, McNemar test; 95% CI for value: 0.5370, 0.9790). For this subgroup of patients, the sensitivity in detecting DCIS on either view was 70% (16/[16 + 7]). The corresponding agreement rate ( value) between the two views was fairly strong, with a magnitude of 0.627 (exact P = .625, McNemar test; 95% CI for value: 0.3029, 0.9513).
|
DISCUSSION
|
|---|
The sensitivity of mammography ranges from 70% to 90% (1). Search, detection, and interpretation errors can all account for this decreased sensitivity. CAD can be used to decrease the search and detection errors without a substantial increase in the recall rate (6). Since these errors account for half of the missed breast cancers, CAD can be beneficial. Previous study findings have shown double interpretation to increase cancer detection rates (1,2,4–15). Although two radiologists can perform double reading, CAD allows double reading to be performed more quickly and economically. However, CAD is limited because it does not give a double interpretation that would be achieved by using two radiologists. CAD simply allows the reader to review a specific area that CAD has deemed as possibly abnormal. In our study, CAD correctly identified 91% of DCIS lesions on at least one of the two views.
The sensitivity of CAD was shown to be high in our study, but the specificity was not assessed. Our purpose was to look solely at the sensitivity of CAD specifically to the diagnosis of DCIS. We knew the previous study of Freer and Ulissey (6) showed the overall specificity of CAD to be low; however, the majority of false-positive marks were quickly dismissed by the reader. The purpose of the CAD system is to detect an abnormality; it is then up to the radiologist to asses whether that abnormality is actionable by determining if it is real and whether it is present on prior studies (3,16). Quek et al (4) showed that there was no mammographic abnormality in half the areas that the CAD system had marked.
Our study findings did not show the sensitivity of CAD to be significant for lesion location, lesion mammographic size, breast density, or age of the patient. However, findings did show that CAD was more sensitive in identifying calcifications than masses, which had been previously noted by others (3,5,8,10,12,17–19). However, only one lesion in our study had calcifications associated with it, thus potentially causing a bias. Ho and Lam (8) found the sensitivity of CAD to decrease as the density of the breast increased, but this was not found in our study, although the majority of breasts in our study either had scattered fibroglandular tissue or were heterogeneously dense. Other study findings (10,20) have concurred with our findings and have not shown an association with breast density. Malich et al (2) did show an association with size, demonstrating the highest detection rates for lesions 10–30 mm.
Retrospectively, breast cancers can be seen in 30%–70% of patients (1,5,10,13,21). In our study, an abnormality was seen on the prior mammogram in 68% of the patients. It is difficult to assess whether all of these findings were actionable since these studies were being viewed with the knowledge of where the cancer was subsequently diagnosed and also without the benefit of prior studies, which may have dissuaded the callback. CAD identified 70% of these abnormalities on the prior mammogram in our study. This finding is similar to those of other studies, which showed CAD identifying 71%–81% of the cancers on the prior mammogram (5,6,13). Attention to the region of abnormality provided by CAD would eliminate the detection error, which could have caused this false-negative finding but does not address the interpretation error.
Another potential limitation of CAD is the inconsistency in reproducing the prompts, which has been shown by Taylor et al (9). Zheng et al (19) showed that CAD systems demonstrated identical results in only 42% of all images when they were digitized three times. Although the CAD system has been shown to have high sensitivity, it is not perfect and did not identify five cancers on either view. Therefore, vigilance still needs to be maintained when evaluating a mammogram.
Our study, although limited to cases with a diagnosis of DCIS, showed that CAD is just as sensitive in identifying DCIS as previous studies have shown with mixed cancer cases. We realize that not all DCIS lesions evolve into invasive cancers. Currently, however, almost all patients with DCIS are treated; therefore, identifying DCIS at mammography is important. In the majority of our patients, at least 82% (45 of 55), the diagnosis was not upgraded to invasive ductal carcinoma at definitive surgical therapy. Five percent of patients had no surgical follow-up at our institution. In 13% of patients, the diagnosis was upgraded to invasive ductal carcinoma, which does place a limitation on our study since all of the cases were not pure DCIS.
Limitations of our study were that it was a retrospective rather than a prospective study. Despite acquisition of data from a 1-year period, only 58 lesions were included in the study. Future prospective studies with a greater number of subjects will add to these results. The limited number of lesions also caused our sample size for our variables to be small. Specifically, we had only five masses among the 58 lesions, which is appropriate since 10% of DCIS can manifest as masses. However, the statistical analysis of such a small subset is limited. A potential bias in our study was that only a single nonblinded reader evaluated all the images; however, the major factor that was evaluated was whether the area of a known cancer was marked by CAD, which is not subjective. The CAD unit we used operated on software version 3.2. Since our study, there have been further upgrades of the software.
In conclusion, CAD had a high sensitivity in the depiction of DCIS in our relatively small study. It ultimately may prove helpful in decreasing the detection errors involved in screening mammography and, by drawing attention to the area of abnormality, could aid in earlier detection of cancer. More studies with larger patient populations will be needed to validate this conclusion for DCIS.
|
ADVANCE IN KNOWLEDGE
|
|---|
- Computer-aided detection has a high sensitivity (91%) for the depiction of ductal carcinoma in situ.
|
ACKNOWLEDGMENTS
|
|---|
We thank Judith Kreiselmeier and Mamtha Balasubramaniam, MS, for all their efforts placed into this study.
|
FOOTNOTES
|
|---|
Abbreviations: CAD = computer-aided detection • CC = craniocaudal • CI = confidence interval • DCIS = ductal carcinoma in situ • MLO = mediolateral oblique
Authors stated no financial relationship to disclose.
Author contributions: Guarantor of integrity of entire study, V.R.P.; study concepts/study design or data acquisition or data analysis/interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; approval of final version of submitted manuscript, all authors; literature research, all authors; clinical studies, V.R.P., M.R.; statistical analysis, V.R.P., M.R.; and manuscript editing, all authors
|
References
|
|---|
- Brem RF, Baum J, Lechner M, et al. Improvement in sensitivity of screening mammography with computer-aided detection: a multiinstitutional trial. AJR Am J Roentgenol 2003;181:687–693.[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]
- Astley SM, Gilbert FJ. Computer-aided detection in mammography. Clin Radiol 2004;59:390–399.[CrossRef][Medline]
- Quek ST, Thng CH, Khoo JBK, Koh WL. Radiologists' detection of mammographic abnormalities with and without a computer-aided detection system. Australas Radiol 2003;47:257–260.[CrossRef][Medline]
- Warren Burhenne LJ, Wood SA, D'Orsi JD, et al. Potential contribution of computer-aided detection to the sensitivity of screening mammography. Radiology 2000;215:554–562.[Abstract/Free Full Text]
- 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]
- Ciatto S, Rosselli Del Turco M, Burke P, Visioli C, Paci E, Zappa M. Comparison of standard and double reading and computer-aided detection (CAD) of interval cancers at prior negative screening mammograms: blind review. Br J Cancer 2003;89:1645–1649.[CrossRef][Medline]
- Ho WT, Lam PW. Clinical performance of computer-assisted detection (CAD) system in detecting carcinoma in breasts of different densities. Clin Radiol 2003;58:133–136.[CrossRef][Medline]
- Taylor CG, Champness J, Reddy M, Taylor P, Potts HW, Given-Wilson R. Reproducibility of prompts in computer-aided detection (CAD) of breast cancer. Clin Radiol 2003;58:733–738.[CrossRef][Medline]
- 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]
- Zheng B, Ganott A, Britton CA, et al. Soft-copy mammographic readings with different computer-assisted detection cuing environments: preliminary findings. Radiology 2001;221:633–640.[Abstract/Free Full Text]
- Baker JA, Rosen EL, Lo JY, Gimenez El, Walsh R, Scott Soo M. 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]
- 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]
- Brem RF, Schoonjans JM. Radiologist detection of microcalcifications with and without computer-aided detection: a comparative study. Clin Radiol 2001;56:150–154.[CrossRef][Medline]
- Karssemeijer N, Otten JDM, Verbeek AL, et al. Computer-aided detection versus independent double reading of masses on mammograms. Radiology 2003;227:192–200.[Abstract/Free Full Text]
- Ikeda DM, Birdwell RL, O'Shaughnessy KF, Sickles EA, Brenner RJ. Computer-aided detection output on 172 subtle findings on normal mammograms previously obtained in women with breast cancer detected at follow-up screening mammography. Radiology 2004;230:811–819.[Abstract/Free Full Text]
- D'Orsi CJ. Computer-aided detection: there is no free lunch [editorial]. Radiology 2001;221:585–586.[Free Full Text]
- Petrick N, Sahiner B, Chan HP, Helvie MA, Paquerault S, Hadjiiski LM. Breast cancer detection: evaluation of a mass-detection algorithm for computer-aided diagnosis—experience in 263 patients. Radiology 2002;224:217–224.[Abstract/Free Full Text]
- Zheng B, Hardesty LA, Poller WR, Sumkin JH, Golla S. Mammography with computer-aided detection: reproducibility assessment—initial experience. Radiology 2003;228:58–62.[Abstract/Free Full Text]
- 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]
- Evans WP, Warren Burhenne LJ, Laurie L, O'Shaughnessy KF, Castellino RA. Invasive lobular carcinoma of the breast: mammographic characteristics and computer-aided detection. Radiology 2002;225:182–189.[Abstract/Free Full Text]
TOP
ABSTRACT
INTRODUCTION
