Clinical Comparison of Full-Field Digital Mammography and Screen- Film Mammography for Detection of Breast Cancer

Transcription

1 John M. Lewin 1 Carl J. D Orsi 2 R. Edward Hendrick 1,3 Lawrence J. Moss 2 Pamela K. Isaacs 1 ndrew Karellas 2 Gary R. Cutter 4 Received July 6, 2001; accepted after revision February 19, Supported by contract DMD17-96-C-6104 from the United States rmy Medical Research and Materiel Command reast Cancer Research Program. 1 University of Colorado Health Sciences Center, 4200 E. 9th ve., Mail Stop F724, Denver, CO ddress correspondence to J. Lewin, University of Colorado Health Sciences Center, Mail Stop F724, 1635 N. Ursula St., urora, CO University of Massachusetts Medical Center, 55 Lake ve. N., Worcester, M Northwestern University Medical School, 357 E. Chicago ve., Chicago, IL MC Cancer Research Center, 1600 Pierce St., Lakewood, CO JR 2002;179: X/02/ merican Roentgen Ray Society Clinical Comparison of Full-Field Digital Mammography and Screen- Film Mammography for Detection of reast Cancer OJECTIVE. The purpose of this work is to compare full-field digital mammography and screen-film mammography for the detection of breast cancer in a screening population. SUJECTS ND METHODS. Full-field digital mammography was performed in addition to screen-film mammography in 6736 examinations of women 40 years old and older presenting for screening mammography at either of two institutions. Two views of each breast were acquired with each technique. The digital and screen-film mammograms were each interpreted independently. In addition to a clinical assessment, each finding was assigned a probability of malignancy for use in receiver operating characteristic analysis. In cases in which the digital and screen-film interpretations differed, a side-by-side analysis was performed to determine the reasons for the discrepancy. With few exceptions, findings detected on either technique were evaluated with additional imaging and, if warranted, biopsy. RESULTS. dditional evaluation was recommended on at least one technique in 1467 cases. These additional evaluations led to 181 biopsies and the detection of 42 cancers. Nine cancers were detected only on digital mammography, 15 were detected only on screen-film mammography, and 18 were detected on both. The difference in cancer detection is not statistically significant ( p > 0.1). Digital mammography resulted in fewer recalls than did screenfilm mammography (799 vs 1007, p < 0.001). The difference between the receiver operating characteristic curve area for digital (0.74) and screen-film (0.80) mammography was not significant ( p > 0.1). s for discrepant interpretations of cancer were approximately equally distributed among those relating to lesion conspicuity, lesion appearance, and interpretation. CONCLUSION. No significant difference in cancer detection was observed between digital mammography and screen-film mammography. Digital mammography resulted in fewer recalls than did screen-film mammography. nalysis of contemporary mammographic screening programs has revealed sensitivities to breast cancer of only about 75% [1 3]. ecause mammography fails to detect a large percentage of cancers, new technologies for detection are being pursued. One such technology, full-field digital mammography, was introduced in prototype form in 1996, after years of anticipation [4]; it was approved for clinical use by the Food and Drug dministration (FD) in January The basis for this technology is the replacement of the fluorescent screen and film used in standard screen-film mammography with a digital detector. The digital detector records X rays detected as electrical signals that are converted to digital data. Compared with film, the digital detector has greater contrast resolution and wider latitude [5]. The hope is that these properties will allow digital mammography to detect more cancers that might be hidden on screen-film mammography by dense normal tissue. However, the digital detector does not have as high a spatial resolution for high-contrast objects as film does, and the effect on cancer detection of this trade-off between spatial resolution and contrast resolution cannot be predicted, because both play a role in revealing the features of breast cancers. dditionally, the clinical performance of digital mammography will depend on the performance of the radiologist in interpreting the images, and the effect of the new technology on radiologists performance cannot be reliably modeled. This fact is especially true because digital mammograms will likely be interpreted from a com- JR:179, September

2 Lewin et al. puter workstation rather than from film. For these reasons, clinical trials comparing digital mammography and screen-film mammography are needed to determine whether the new technology will be better than, equal to, or worse than the existing technology in screening for breast cancer. This article describes the final results of one such trial. Interim results on the first 4945 examinations of this trial have been previously reported [6]. These final results do not differ in statistically significant outcomes from those in the previous report, but the additional cases add power to the study and enable the analysis of trends over long periods. dditionally, this article for the first time reports the reasons for discrepant digital and screen-film mammographic interpretations. Subjects and Methods ll women 40 years old and older who presented for screening mammography at either of two centers were offered enrollment, provided that each breast was no larger than could be fitted on large, standard size (24 30 cm) mammography film. total of 6736 paired examinations were performed on 4489 subjects; 1665 subjects enrolled twice in the study and 291 subjects enrolled three times. Subjects enrolling more than once had at least 11 months between enrollments (average, 14.3 months). The average age of the subjects was 55.6 years. Thirty-two symptomatic patients presenting for bilateral diagnostic mammography were originally imaged but are excluded from this analysis. The data presented here are thus from asymptomatic subjects presenting for screening mammography. Each subject signed a consent form approved by the institutional review board of the enrolling institution and the United States rmy Medical Research and Materiel Command. Screen-film mammography and full-field digital mammography were performed on each subject, usually by the same technologist at the same visit. Screen-film mammography was performed on a commercial mammography unit (DMR; General Electric Medical Systems, Milwaukee, WI). Full-field digital mammography was performed on a prototype unit using an amorphous silicon solid-state detector with CsI crystal. The technical details of this digital prototype have been reported previously [7]. slightly modified version of this device, using the same detector technology, is now available commercially (Senographe 2000D; General Electric Medical Systems). The major technical improvements in this commercial unit are the addition of a system for automatically selecting the technique factors, analogous to the utomatic Optimization of Parameters mode available on the DMR screen-film unit, and an improved postprocessing algorithm for equalizing regional signal values as the breast thickness decreases near the skin. Technique factors for each screen-film examination were selected by the screen-film unit in utomatic Optimization of Parameters mode. ecause no method automatically determines technique factors on the digital prototype, technique factors for each digital image were matched for X-ray beam target material, filtration material, peak kilovoltage, and (as closely as possible) beam current to that used for the corresponding screen-film image. The screen-film and digital images were each interpreted independently by a board-certified radiologist with experience in mammography. Years of interpretation experience for each reviewer ranged from 3 to more than 20 years. Digital mammography was interpreted on a prototype two-monitor workstation supplied with the unit. Details of this workstation have been given previously [6]. This workstation was notably less powerful than the workstation used with the FD-approved commercial unit. For example, the monitor resolution of the prototype workstation was only pixels versus pixels for the commercial system. This difference is significant in that the entire image could not be displayed in full resolution on the prototype system, whereas it can be on the commercial system. dditionally, the software on the commercial system enables far easier manipulation of the images than the software on the prototype system. On the prototype system, image interpretation typically included viewing the image at a magnification of 2. For the first approximately cases, magnification was accomplished with a moving square ( mag glass ) showing a portion of the image magnified. Subsequently, each quadrant of each image was magnified in its entirety and examined. The digital workstation was located in a darkened room away from film alternators. Minimizing ambient light was important because the monitors used on the prototype system operated at a maximum luminance of about 40 foot-lamberts versus about 70 foot-lamberts for those on the prototype system. Comparison mammograms were viewed on film using a standard viewbox placed next to the digital workstation. The viewbox was turned off during the detailed evaluation of digital images to avoid glare. The reviewers were trained on the use of the workstation with a small number of sample images before the study. For each finding on the mammogram, the interpreting radiologist was instructed to provide its location, an assessment and recommendation using the merican College of Radiology reast Imaging Reporting and Data System (I-RDS) nomenclature [8], and a probability of malignancy on a scale of 0 100, for use in receiver operating characteristic (ROC) curve analysis. Each reviewer in the study interpreted an approximately equal number of screen-film and digital examinations. Prior mammograms and patient history were equally available for the screen-film and digital interpretations. Findings detected on either screen-film or digital mammography were further evaluated, in a manner consistent with routine clinical practice, using any combination of additional mammography, sonography, and biopsy. The additional mammographic images were generally obtained using the technique on which the finding was better seen, with the exception that all magnification images were obtained on screen-film mammography because the grid on the prototype was not easily removed. Each subject was followed up for 1 year after her enrollment to determine whether she developed breast cancer. Cancer was considered to be present at the time of imaging if it was proven by biopsy within 1 year. Discrepancy nalysis For a given finding, if the interpretation of the screen-film examination differed from that of the digital examination, the case was set aside for discrepancy evaluation. The interpretations were considered different if one resulted in a recommendation of immediate additional imaging or biopsy and the other resulted in a recommendation of short-interval or routine follow-up. These recommendations generally corresponded to I-RDS assessments of 0, 4, or 5 in the former instance, and to I-RDS assessments of 1, 2, or 3 in the latter case. Discrepancy evaluation consisted of a side-byside comparison of the screen-film and digital images by two radiologists using soft-copy display for digital and hard-copy for screen-film. Ninety percent of discrepancy evaluations were performed by the two radiologists who had originally interpreted the case; in 10% of cases, one of the two radiologists was different. The purpose of the evaluation was to determine the causes of different interpretations for each finding. discrepancy evaluation form was used that consisted of an 11-point scale for measuring the relative visibility and conspicuity of a finding on screen-film versus digital mammography; the form gave a choice of 39 reasons for the discrepancy. The reviewers were instructed to choose, by consensus, a single major reason and, optionally, a single minor reason for the discrepant interpretations between the two techniques. t discrepancy evaluation, reviewers had the option of deciding not to work up a finding, but they were instructed to use this option only if the finding detected on one technique could be dismissed with a definite explanation based on the other technique or if the discrepancy was due to an error in interpretation. This option was rarely used: only 68 (13%) of 510 digital-only findings and 14 (2.0%) of 751 screen-film-only findings were dismissed. Two invasive lobular carcinomas in the regions of digital-only findings that were dismissed at discrepancy evaluation became palpable within 1 and 5 months of the screening examination. ecause additional imaging was not performed after the cancers became palpable, it was not possible to exactly correlate the imaging finding and the palpable abnormality, but in both cases the cancers involved the quadrant of the finding. In one case the finding was a density that, on the basis of evaluation of the postlumpectomy 672 JR:179, September 2002

3 Full-Field Digital Mammography Versus Screen-Film Mammography mammogram, was removed at lumpectomy. The other finding was calcifications, and evaluation of the mastectomy specimen showed calcifications in the tumor. On the basis of the available information, both of these cases were considered true-positive for digital mammography. n additional digital-only finding that was dismissed became palpable 10 months after screening and was diagnosed as a fibroadenoma at surgical biopsy. nalysis of ROC Curve nalysis of the ROC curve was performed using the alternative free-response ROC method [9]. The area under each curve was calculated using the trapezoidal method. reas were compared using the Wilcoxon s statistic, which accounts for the correlation gained by the use of the same cases for each curve [10]. Statistical analysis on agreement tables was performed using the McNemar chi-square test for paired data [11]. Results Two thousand thirty-one findings on 1467 examinations led to patients being recalled for TLE 1 Note. FFDM = full-field digital mammography. additional workup. More patients were recalled on the basis of screen-film mammographic findings than on the basis of digital mammographic findings (1345 vs 979). These numbers included the 293 findings found on both techniques. Similarly, more examinations had at least one patient recalled on screen-film mammography than had at least one patient recalled on digital (1001 vs 793), leading to a higher recall rate for screen-film mammography (14.9%) than for digital mammography (11.8%). These numbers include the 327 examinations considered positive on both techniques. The differences in the rates of both positive findings and positive mammographic examinations were statistically significant (p < 0.001). Tables 1 and 2 list the most common major and minor reasons given at discrepancy analysis for the findings seen only on screen-film and those seen only on digital mammography, respectively. In both cases, the two most common reasons, by far, were fortuitous positioning and minor difference in opinion. Figure 1 gives Most Common s for Findings Seen Only on Screen-Film Mammography (SFM) Major Minor an example of fortuitous positioning resulting in a discrepant finding detected only on screen-film mammography. Minor difference of opinion was defined as an interpretation difference between the reviewers resulting in a recall by one reviewer even though both reviewers had a low suspicion of malignancy. minor reason was given for only about half the discrepant interpretations. The most common minor reason given for both screen-film-only and digital-only findings was compression difference. One hundred seventy-nine biopsies were performed on findings detected in the study. Eighty-seven biopsies were of findings originally detected only on screen-film mammography, 38 biopsies were of findings originally detected only on digital mammography, and 56 biopsies were on findings originally detected on both techniques. The difference between the number of biopsies resulting from screen-film findings and the number resulting from digital findings is statistically significant (p < 0.001). Forty-two cancers were diagnosed. Screen-film mammography detected 33 of the cancers and Fortuitous positioning Visibility 309 None 572 Minor difference of opinion Interpretation 277 Compression difference Visibility 104 Visibility or contrast difference (unsure of cause) Visibility 59 Minor difference of opinion Interpretation 73 Technique difference Visibility 49 Fortuitous positioning Visibility 54 Mass margin or shape more suspicious on SFM ppearance 40 Technique difference Visibility 37 More calcifications visible on SFM ppearance 38 Workstation suboptimal 21 Error in detection on FFDM Interpretation 32 Error in detection on FFDM Interpretation 20 Positioning Visibility 31 Mass margin or shape more suspicious on SFM ppearance 17 Sharpness greater on SFM Visibility 28 Visibility or contrast difference (unsure of cause) Visibility 16 Major difference of opinion Interpretation 20 Calcification forms more suspicious on SFM ppearance 14 TLE 2 Most Common s for Findings Seen Only on Full-Field Digital Mammography Major Minor Fortuitous positioning Visibility 238 None 343 Minor difference of opinion Interpretation 138 Compression difference Visibility 105 Mass margin or shape more suspicious on digital ppearance 62 Fortuitous positioning Visibility 58 Visibility or contrast difference (unsure of cause) Visibility 42 Minor difference of opinion Interpretation 47 Technique difference Visibility 35 Technique difference Visibility 30 Sharpness greater on digital Visibility 34 Mass margin or shape more suspicious on digital ppearance 19 Error in detection on film Interpretation 19 Sharpness greater on digital Visibility 10 More calcifications visible on digital ppearance 16 bility to magnify on digital workstation Visibility 10 Error in interpretation on digital Interpretation 14 Visibility or contrast difference (unsure of cause) Visibility 9 Positioning Visibility 10 Error in detection on film Interpretation 8 JR:179, September

4 Lewin et al. digital mammography detected 25. These numbers include the 18 cancers detected on both techniques. Eight additional cancers were detected by palpation and were biopsied within 1 year of negative interpretations on both techniques. The difference in cancer detection was not statistically significant (p > 0.1). TLE 3 Mammographically Detected Cancers by Lesion Type Note. SFM = screen-film mammography, FFDM = full-field digital mammography. Table 3 gives the detected cancers by mammographic type. rchitectural distortion and calcifications accounted for most of the differences in cancer detection. Table 4 gives the major reasons at discrepancy analysis for cancers being found only on one technique. No reason dominated, and Detected on Lesion Type SFM FFDM Calcification Mass rchitectural distortion symmetric density Total TLE 4 Major s for Cancer Seen on Only One Technique Detected on Screen-Film Mammography (SFM) Detected on Full-Field Digital Mammography (FFDM) Fig. 1. Fortuitous positioning causing recall on screen-film mammography but not on full-field digital mammography. Spot compression images performed for further evaluation of screen-film finding in 45-yearold woman showed no abnormality., Screen-film mediolateral oblique image shows apparent density (arrow)., Digital mediolateral oblique image shows that density is due to overlapping tissue. the overall categories were fairly evenly divided among visibility, appearance, and interpretation, with no discernible trends between the techniques. Figures 2 4 are examples of cancers missed on one technique because of appearance, visibility, and interpretation, respectively. Five of the 15 cancers detected only on screen-film mammography were judged to be equally conspicuous on both techniques. The other 19 discrepant cancers were determined to be more conspicuous on the technique on which they were found. The positive predictive value of screening, defined as the percentage of screening examinations with positive findings that led to a diagnosis of cancer, was slightly lower for screen-film mammography (33/1001, 3.3%) than for digital mammography (27/793, 3.4%). Minor difference of opinion Interpretation 4 Fortuitous positioning Visibility 2 Visibility or contrast difference (unsure of cause) Visibility 2 Sharpness greater on FFDM Visibility 1 Fortuitous positioning Visibility 2 bility to magnify on digital workstation Visibility 1 Mass margin or shape more suspicious on SFM ppearance 2 More calcifications visible on FFDM ppearance 1 Sharpness greater on SFM Visibility 1 Mass margin or shape more suspicious on FFDM ppearance 1 Compression difference Visibility 1 Error in interpretation, digital a Interpretation 1 Calcification forms more suspicious on SFM ppearance 1 Major difference of opinion Interpretation 1 Film assessed as more changed from comparison b Interpretation 1 Minor difference of opinion Interpretation 1 Workstation suboptimal Interpretation 1 a Evaluation before obtaining diagnosis of cancer. b Observer interpreting SFM judged the finding to be more different from the comparison study that did the observer interpreting the FFDM. 674 JR:179, September 2002

5 Full-Field Digital Mammography Versus Screen-Film Mammography Free-response ROC curves for screen-film and digital mammography are shown in Figure 5. The area under the curve was 0.80 for screen-film mammography and 0.74 for digital mammography. This difference was not statistically significant ( p = 0.18). Fig. 2. Invasive ductal carcinoma in 57-year-old woman detected on fullfield digital mammography but not on screen-film mammography., Full-field digital mammogram shows irregular mass (arrow ) with sharp, angular margins., On screen-film mammogram, cancer (arrow ) is visible but appearance is similar to that of surrounding foci of normal tissue. Difference in appearance was thought to be primarily caused by difference in projection of cancer after breast compression. Fig. 3. Radial scar containing invasive ductal carcinoma and ductal carcinoma in situ detected in 55- year-old woman on screen-film mammography but not on full-field digital mammography., On screen-film mammogram, lesion (arrow ) fortuitously projects over pectoral muscle, causing lesion to appear dense and making it more conspicuous., On full-field digital mammogram, lesion (arrow ) straddles pectoral muscle, resulting in low-density center and less conspicuity. Fine spiculations characteristic of radial scar are well shown by both techniques. Discussion The hope that the technical advantages of digital mammography would immediately translate into the detection of more cancers was not borne out by our study. The trend, in fact, was for screen-film mammography to detect more cancers, although the difference was not statistically significant. On the other hand, the concern that digital mammography might lead to more false-positive screening examinations was also not borne out by our study. Digital mammography used alone would have led to significantly fewer recalls than would have screen-film mammography used alone. The difference in recall rates may have been from an inherent difference in the imaging techniques; however, the possibility exists that the difference was a result of the experimental situation on the behavior of the radiologists participating in the study. lthough they knew that both the screen-film and digital examinations were being used for clinical treatment, the reviewers may have used a lower threshold for recall on the standard examination, screen-film, than on the experimental examination, digital. Now that full-field digital mammography has been disseminated for routine clinical use, the recall rate obtained with that technique in clinical practice can be measured and compared with that obtained with screen-film mammography. Given that digital mammography resulted in both fewer false-positive examinations and fewer true-positive examinations (the latter not reaching statistical significance), other measures are needed to compare the overall diagnostic performance of the two techniques. The key question is whether the greater number of cancers detected on screen-film mammography is simply the result of a lower threshold for determining a finding. One measure for comparison is the positive predictive value. The fact that this value was nearly equal for the two techniques suggests that the difference in cancer detection was more than simply a shift in threshold, because such a shift would be expected to result in a lower positive predictive value. In other words, the increase in sensitivity from such a shift would be accompanied by a decrease in specificity. The trade-off between sensitivity (true-positive rate) and specificity (1 false-positive rate) is directly illustrated by plotting an ROC curve. The area under this curve can then be used to compare two tests independent of the threshold used for positivity. This measurement also favored screen-film mammography (Fig. 5), although the difference was not statistically significant. lthough each technique detected cancers at or above the expected rate for our population [12], each missed a large fraction of those detected by the other. No dominant cause accounted for the misses. In some cases, the appearance of the cancer was significantly different on the two techniques, possibly because of a slight variation in the JR:179, September

6 Lewin et al. True-Positive Fraction projection of the lesion during imaging. Figure 2 is an example of one such cancer, which was detected only on digital mammography. In other cases, the relationship of nearby normal tissue affected detection. In some of these cases, fibroglandular tissue superimposed over the cancer on the image of one technique obscured the lesion from view, whereas on the other technique fibroglandular tissue projected next to the cancer, allowing it to be detected. This difference in superposition of structures occurred even though both techniques used identically designed compression devices and the same technologist positioned the patient for both examinations. Generally, the breast appeared to be positioned identically, on the basis of standard landmarks such as the pectoral False-Positive Fraction Fig. 4. Ductal carcinoma in situ in 53-year-old woman detected on screen-film mammography but not on full-field digital mammography because of interpretation difference between reviewers. Cancer was occult on mediolateral oblique images from both techniques., On screen-film mammogram, density at lateral aspect of craniocaudal image was unchanged from multiple prior studies. New indentation in contour of this density, which was of concern as possible architectural distortion, led to recall., On digital craniocaudal mammogram, appearance of density and indentation (arrow) is essentially identical to appearance on screen-film muscle, inframammary fold, and nipple. In still other cases, interpretation differences resulted in cancers that looked similar on both techniques being detected on only one, as illustrated in Figure 4. There are limitations to applying the results of this study to clinical practice. First, digital mammography technology is still evolving. The workstation now being sold for clinical use is technically superior to those used in this study; it incorporates brighter, higher resolution monitors and has more ergonomically designed controls. breast thickness equalization algorithm is now applied to each image to reduce the amount of image adjustment needed for interpretation. Such adjustments are fatiguing and possibly distracting to the interpreting radiologist in a screening environment when examinations are interpreted in a batch. n improved workstation would possibly have resulted in more cancers being detected on digital mammography; for six of the cancers missed on digital mammography, the major reason was related to interpretation, including one cancer miss thought to be directly caused by the workstation. On the other hand, for three of the cancers missed on screen-film mammography, the misses were determined to be interpretation-related. It might be valuable to use the images acquired Fig. 5. Free-response receiver operating characteristic (ROC) curves for screen-film mammography (, dotted line) and fullfield digital mammography (, solid line) based on rating scale of Scale for x-axis is probability of false-positive finding occurring on two screening images of given breast and is analogous to false-positive rate in standard ROC experiment. rea under screen-film curve is 0.80; area under digital curve is Difference is not statistically significant ( p = 0.18). 676 JR:179, September 2002

7 Full-Field Digital Mammography Versus Screen-Film Mammography in this study in a multireviewer performance study that could use the latest digital mammography workstation and postprocessing to see whether the results would change. Unfortunately, such a study can never exactly reproduce the screening environment in which, on average, approximately 250 examinations must be interpreted to find a single cancer. Digital mammography is still in its infancy compared with screen-film mammography. The latter has had the benefit of more than three decades of clinical use and technologic improvement. For this reason, perhaps we should not be surprised that a digital prototype cannot outperform screen-film mammography clinically. ecause full-field digital mammography is a new technology, technical improvements can be expected to occur at a faster rate than for screen-film mammography, and future studies may show that the technical advantages of full-field digital mammography have translated into a clinical advantage. References 1. Poplack SP, Tosteson N, Grove MR, Wells W, Carney P. Mammography in 53,803 women from the New Hampshire Mammography Network. Radiology 2000;217: Rosenberg RD, Hunt WC, Williamson MR, et al. Effects of age, breast density, ethnicity, and estrogen replacement therapy on screening mammographic sensitivity and cancer stage at diagnosis: review of 183,134 screening mammograms in lbuquerque, New Mexico. Radiology 1998;209: Mandelson MT, Oestricher N, Porter PL, et al. reast density as a predictor of mammographic detection: comparison of interval- and screendetected cancers. J Natl Cancer Inst 2000;92: Shtern F. Digital mammography and related technologies: a perspective from the National Cancer Institute. Radiology 1992;183: Feig S, Yaffe MJ. Digital mammography, computer-aided diagnosis, and telemammography. Radiol Clin North m 1995;33: Lewin JM, Hendrick RE, D Orsi CJ, et al. Comparison of full-field digital mammography to screen-film mammography for cancer detection: results of 4945 paired examinations. Radiology 2001;218: Vedantham S, Karellas, Suryanarayanan S, et al. Full breast digital mammography with an amorphous silicon-based flat panel detector: physical characteristics of a clinical prototype. Med Phys 2000;27: merican College of Radiology. reast imaging reporting and data system (I-RDS), 2nd ed. Reston, V: merican College of Radiology, Chakraborty DP, Winter LH. Free-response methodology: alternate analysis and a new observer-performance experiment. Radiology 1990; 174: Hanley J, McNeil J. method of comparing the areas under receiver operating characteristic curves derived from the same cases. Radiology 1983;148: Siegel S. Non-parametric statistics for the behavioral sciences. New York: McGraw-Hill, 1956: assett LW, Hendrick RE, assford TL, et al. Quality determinants of mammography: clinical practice guideline no. 13. HCPR publication no Rockville, MD: gency for Health Care Policy and Research, Public Health Service, United States Department of Health and Human Services, 1994:81 83 JR:179, September