Two-View and Single-View Tomosynthesis versus Full-Field Digital Mammography: High- Resolution X-Ray Imaging Observer Study 1 - PDF

Please download to get full document.

View again

of 9
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Information Report
Category:

Presentations

Published:

Views: 13 | Pages: 9

Extension: PDF | Download: 0

Share
Related documents
Description
Note: This copy is for your personal non-commercial use only. To order presentation-ready copies for distribution to your colleagues or clients, contact us at Original Research
Transcript
Note: This copy is for your personal non-commercial use only. To order presentation-ready copies for distribution to your colleagues or clients, contact us at Original Research n Breast Imaging Two-View and Single-View Tomosynthesis versus Full-Field Digital Mammography: High- Resolution X-Ray Imaging Observer Study 1 Matthew G. Wallis, MB, ChB, FRCR Elin Moa, MSc Federica Zanca, MSc, PhD Karin Leifland, MD, PhD Mats Danielsson, MSc, PhD 1 From the Cambridge Breast Unit, Cambridge University Hospitals NHS Foundation Trust, Box 97, Hills Rd, Cambridge CB2 0QQ, England (M.G.W.); Philips Women s Imaging Healthcare, Solna, Sweden (E.M.); Leuven University Center of Medical Physics in Radiology, University Hospitals Leuven, Leuven, Belgium (F.Z.); Mammography Department, Unilabs AB, Capio St Göran s Hospital, Stockholm, Sweden (K.L.); and Department of Physics, Royal Institute of Technology, Stockholm, Sweden (M.D.). From the 2010 RSNA Aual Meeting. Received January 10, 2011; revision requested February 23; revision received August 16; accepted September 2; final version accepted October 6. Supported by the Commission of the European Communities Research Directorate General Sixth Framework research program. Address correspondence to M.G.W. ( q RSNA, 2012 Purpose: Materials and Methods: Results: Conclusion: To compare the diagnostic accuracy of two-dimensional (2D) full-field digital mammography with that of two-view (mediolateral and craniocaudal) and single-view (mediolateral oblique) tomosynthesis in an observer study involving two institutions. Ethical committee approval was obtained. All participating women gave informed consent. Two hundred twenty women (mean age, 56.3; range, years) with breast density of 2 4 according to American College of Radiology criteria were recruited between November 2008 and September 2009 and underwent standard treatment plus tomosynthesis with a prototype photon-counting machine. After exclusion criteria were met, this resulted in a final test set of 130 women. Ten accredited readers classified the 130 cases (40 cancers, 24 benign lesions, and 66 normal images) using 2D mammography and two-view tomosynthesis. Another 10 readers reviewed the same cases using 2D mammography but single-view tomosynthesis. The multireader, multicase receiver operating characteristic (ROC) method was applied. The significance of the observed difference in accuracy between 2D mammography and tomosynthesis was calculated. For diagnostic accuracy, 2D mammography performed significantly worse than two-view tomosynthesis (average area under ROC curve [AUC] = for 2D, AUC = for tomosynthesis, P =.021). Significant differences were found for both masses and microcalcification (P =.037 and.049). The difference in AUC between the two modalities of was significant (P =.03) only for the five readers with the least experience (,10 years of reading); with AUC of for the five readers with 10 years or more experience (P =.25). No significant difference (P =.79) in reader performance was seen when 2D mammography (average AUC = 0.774) was compared with single-view tomosynthesis (average AUC = 0.775). Two-view tomosynthesis outperforms 2D mammography but only for readers with the least experience. The benefits were seen for both masses and microcalcification. No differences in classification accuracy was seen between and 2D mammography and single-view tomosynthesis. q RSNA, radiology.rsna.org n Radiology: Volume 262: Number 3 March 2012 Mammography is a proven technique for both diagnosis and screening of breast cancer, with randomized controlled trials of screenfilm mammography showing reduction in mortality (1). The more recent development of full-field digital mammography has been shown to be at least equivalent to screen-film mammography in both randomized control trials (2 4) and in-service screening (5). It is superior in patients with dense breasts and for the detection of ductal carcinoma in situ (4). Screening mammography is far from perfect, with up to one-third of the cancers in a screening population Advances in Knowledge In an enriched test set study, two-view tomosynthesis performed better than 2D digital mammography in terms of diagnostic accuracy (P =.021). The analysis of reader experience showed that this effect is only evident for readers with less than 10 years of experience (P =.03), as there was no significant difference for readers with more than 10 years of experience (P =.25). Overall, two-view tomosynthesis performed better for both masses (P =.037) and microcalcification (P =.049). No differences in overall performance were seen between 2D digital mammography and singleview tomosynthesis; subanalysis by lesion type and experience had no significant effect. For comparison of 2D mammography with tomosynthesis, the recall rates showed a reduction of about 11% for two-view tomosynthesis (recall rate, 0.53 for 2D mammography and 0.47 for two-view tomosynthesis) and of 9.5% for single-view tomosynthesis (recall rate, 0.63 for 2D mammography and 0.57 for single-view tomosynthesis). Reading time of tomosynthesis images was almost double that of 2D mammographic images. manifesting between screenings. It also lacks specificity, with recall rates for assessment between 5% in the northern European population screening programs and 15% for opportunistic screening in the United States (6). One of the reasons for this lack of sensitivity and specificity is the difficulty of finding and discerning a subtle change against the complex background of the glandular tissue, a particular issue for younger premenopausal women with a so-called dense glandular background pattern. Considerable effort has been invested in the development of tomosynthesis systems in which a series of low-dose images are obtained over a limited arc through the breast. These projections are mathematically reconstructed into a series of sections, which the radiologist can scroll through (7). By imaging the breast in this way the hope is that improvements in both sensitivity and specificity may be made. Initial work using tomosynthesis as an addon to conventional two-dimensional (2D) mammography suggested the potential for substantial improvements in specificity (8,9). Findings of two more recent studies directly comparing tomosynthesis with conventional 2D mammography (10,11) suggest that although images are easier to evaluate, enriched receiver operating characteristic (ROC) studies show little benefit. Dose constraints have encouraged researchers to explore the use of single-view tomosynthesis as opposed to the conventional two views (10,12,13). The aim of this study was to compare the diagnostic accuracy of 2D digital mammography with that of twoview (mediolateral and craniocaudal) and single-view (mediolateral oblique) Implication for Patient Care These results suggest that twoview tomosynthesis will improve diagnostic accuracy for the less experienced reader and reduce recall rates, but this needs to be tested in real-life studies with economic analysis before widespread introduction to population screening can be contemplated. tomosynthesis in an observer study involving two institutions. Materials and Methods This study and detector development were funded by a European Union grant LSHC-CT Sectra Mamea provided the mammography equipment. M.D. and E.M. are employees of Sectra Mamea. The authors not employed by Sectra Mamea (M.W., F.Z., K.L.) had full control of inclusion of all data, the information submitted for publication, and information that might present a conflict of interest. Patient Population Approval for this study was obtained from the ethics committees of both participating centers, and written informed consent was obtained from all participating women. Women older than 40 years with background density 2, (scattered fibroglandular densities, approximately 25% 50% glandular), 3 (breast tissue is heterogeneously dense, approximately 51% 75% glandular), or 4 (breast tissue is extremely dense, approximately.75% glandular) according to the American College of Radiology criteria (14) were recruited from those attending symptomatic referral clinics or who were recalled from routine screening Published online before print /radiol Content codes: Radiology 2012; 262: Abbreviations: AUC = area under the ROC curve CI = confidence interval ROC = receiver operating characteristic 2D = two-dimensional Author contributions: Guarantor of integrity of entire study, M.G.W.; 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, M.G.W., F.Z., K.L., M.D.; clinical studies, M.G.W., E.M., K.L.; experimental studies, M.D.; statistical analysis, F.Z.; and manuscript editing, all authors Potential conflicts of interest are listed at the end of this article. Radiology: Volume 262: Number 3 March 2012 n radiology.rsna.org 789 to one of two participating centers, depending on the availability of a research radiographer, between November 2008 and September Some asymptomatic women were enrolled and some women with masses detected with ultrasonography (US) or clinically or those with dense breasts at mammography were enrolled. The reader studies were completed in March Women with previous breast cancer, pregnant women, or those unable to give consent were excluded. After their routine 2D mammographic assessment, eligible women were invited to participate. Consent was obtained to obtain two-view tomosynthesis images of both breasts. Women then returned to the clinic for US examination and needle biopsy, as appropriate. Ground truth was determined from cytology or pathology reports in those in whom lesions were believed to be present (99 cases) and from clinical follow-up findings in those in whom mammograms were interpreted as normal (97 cases). Clinical follow-up entailed that no woman had presented to either of their respective host unit or had a record on the respective cancer registries at 1 year after the index study. None of the women with benign findings of needle biopsy or who were discharged from the clinic have subsequently presented with breast cancer within the first 12 months of follow-up. Cysts were defined as normal cases because biopsy was not performed on them. It was not possible to register location information directly on the picture archiving and communication system workstations for either 2D mammography or tomosynthesis. To be able to record the location, the plug-in software divided the breast into nine regions of interest, each of which was subsequently subdivided into four subregions, and section information was noted. Conventional 2D mammography was predominantly performed with a MicroDose D40 system (Sectra Mamea, Solna, Sweden) at the Swedish site and with either a Senograph DS or Senograph Essential system (GE Medical Systems, Buc, France) in England. The tomosynthesis systems were based on the MicroDose Mammography system (Sectra Mamea), using a slit-scaing photon-counting detector (100-µm pixel size) (15,16) The slit-scaing technology removes nearly all scattered radiation (17), and all electronic noise can be eliminated by setting a threshold that is high enough to discriminate the noise while still counting all the x-ray pulses; this leads to a more efficient use of the x-ray signal, resulting in reduced dose and a modified gantry. By using a calculated tomosynthesis angle of 11, 21 projections were acquired and reconstructed into sections with a 3-mm thickness. An iterative reconstruction method was used to create stacks in a Digital Imaging and Communications in Medicine, or DICOM, format (18). The average glandular doses for the 2D mammographic data sets were calculated from the DICOM headers of the images, and the tomosynthesis dose was calculated according to Dance et al (19). The average tomosynthesis radiation dose per exposure was 0.70 mgy (range, mgy) for the participating women in Sweden and 0.82 mgy (range, mgy) for those participating in England, with automatic exposure settings. The average dose per exposure was 0.6 mgy (range, mgy) for the Swedish mammography cases and 1.2 mgy (range, mgy) for the English mammography cases; these differences are due to different 2D mammography equipment. Two hundred twenty women (mean age, 56.3 years; range, years) were imaged with both modalities. Exclusions from this initial set were as follows: women with incomplete mammographic or tomosynthesis images, women whose 2D mammographic images were on computerized radiography, women in whom in retrospect the principal investigator (M.G.W., 22 years of experience in breast imaging) considered the American College of Radiology background density to be 1, and women with prior surgery (Figure). This exclusion resulted in 66 women in whom needle biopsy had been performed and 73 healthy women (discharged from clinic). An additional nine randomly selected women were excluded, for a final test set of 130 cases. That 12/22/2011 number was estimated to be appropriate according to the study by Obuchowski et al (20) with the following assumptions: 10 observers, small to moderate variability among the observers, moderate level of accuracy of the test (area under the ROC curve [AUC], ~0.75), small difference in accuracy between imaging techniques (effect size, 0.05), and patients without to patients with the condition ratio around 1:1. The final test set consisted of 64 women with abnormal images, 40 with cancer and 24 with benign lesions (mean age, 57.3 years; range, years), and 66 women with normal images (mean age, 55.4 years; range, years). The abnormal cases had a total of 70 lesions; four women had multiple lesions and two had bilateral lesions. Table 1 shows the distribution of women between screening and referral cases, and Table 2 shows the lesions according to the radiologic signs. Observer Performance Experiment Two sets of 10 accredited screen-film readers (18 radiologists and two radiographers) were recruited from the national breast screening programs in England, Ireland, and Sweden. All are formally trained and work according to national guidelines. The readers had between 2.5 and 25 years of mammography experience (average, 10.2 years), of which between 0.5 and 8 years (average, 3.4 years) was in digital mammography. They read between 3000 and mammograms per year (average, 12350). The first set of readers read two-view tomosynthesis images versus 2D mammograms; five readers had less than 10 years of experience (average, 5.1 years) and five had 10 years or more (average, 15 years). The second set of readers read single-view tomosynthesis images versus 2D mammogams; five readers had less than 10 years of experience (average, 5.4 years) and five had 10 years or more (average, 18.6 years). All images were displayed in standard picture archiving and communication system (Sectra, versions IDS5 and IDS7). The system was configured to show section numbers but was 790 radiology.rsna.org n Radiology: Volume 262: Number 3 March 2012 Patient flowchart. Abnormal cases include all women with a radiologic abnormality that warranted core biopsy. Normal cases include all women with normal images, including cysts. All normal and benign cases have been followed clinically for a minimum of 12 months. ACR = American College of Radiology, CR = computerized radiography. not modified otherwise. The readers were free to use tools such as zooming and window and/or level, and all reading occurred in a darkened room. The study was divided into two parts. The first half aimed to investigate two-view tomosynthesis versus 2D mammography, and the second half looked at single-view tomosynthesis versus 2D mammography. The first 10 readers performed the two-view tomosynthesis study and the next 10 readers performed the single-view tomosynthesis study. The reader studies were completed in Each reader undertook two reading sessions with 65 tomosynthesis cases and 65 2D mammographic cases per session so that they saw each case once per session using either modality, giving a fully-crossed study design. The reading was fully blinded so that no clinical information was shown to the readers, and the two sessions were at least 4 weeks apart with cases in a different order to reduce memory effects (21). A software package (Sara Softare Systems, Olathe, Kan) was used for reporting (22). The readers were asked to record suspicion of malignancy on a scale of 0 5 and recall decision for the whole case. The software recorded the reading time. Training A selection of the rejected cases was used to develop a training set. This was divided into two parts. An initial collection of seven abnormal cases consisting of both mammographic and tomosynthesis stacks with a full radiologic description and pathologic findings were used to acclimatize readers to the appearances of lesions on tomosynthesis images and to train them on the use of the work station. Three further sets of multifocal or bilateral cases were used to train in the use of the data recording system. The readers were then presented with a further 20 cases (malignant and benign cases and false-positive recalls) where they were first shown the tomosynthesis stacks before having access to detailed pathologic and other imaging findings. All readers spent a minimum of 2 hours on the training sets. Statistical Analysis Although the Sara software collected the score for each subregion, a true location-specific analysis was not possible: The subregions of interest were too large to be considered clinically relevant in terms of the acceptance radius, which is the distance from the center of a lesion; any mark within the acceptance radius is considered a correct localization. For this reason, a standard ROC analysis was performed by using the lesion with the highest rating (per case) assumption (23). The data were analyzed by using DBM-MRMC, version 2.2, software (http://perception.radiology.uiowa.edu). The software is based on analysis of variance approach (24,25) and is designed to determine the statistical significance of the difference between ROC indexes, when the performance of Radiology: Volume 262: Number 3 March 2012 n radiology.rsna.org 791 an imaging modality is affected both by the cases analyzed (eg, patient) and by the observer (eg, technician or physician). This allows the results to be generalized to the population of readers and cases from which the used cases and readers are sampled. The final outcome of the statistical comparison is a 95% confidence interval (CI) of the difference between the ROC index values, along with a two-tailed P value. CIs that do not include zero (or, equivalently, P,.05) are interpreted as the two modalities being significantly different. A software package (DBM-MRMC, version 2) was used to generate the ROC index, fitted by using the proper binormal model. The significance of the difference between the ROC indexes of 2D mammography compared with tomosynthesis was calculated for the diagnostic accuracy. The analysis was performed for both two-view and single-view tomosynthesis study. Subanalyses were also performed for masses and microcalcification to see the effect of diagnostic accuracy on lesion type. We also investigated the effect of the readers experience by grouping the readers into experienced ( 10 years of reading) and less experienced (,10 years of reading) groups. The recall rate was also calculated as the number of cases recalled versus the total number of cases for both twoview and single-view tomosynthesis study. Average reading time was also calculated for each study (2D mammography and two-view and single-view tomosynthesis). Results For diagnostic accuracy of all readers, 2D mammography performed significantly worse (AUC = 0.772) than twoview tomosynthesis
Recommended
View more...
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks