Women's Imaging
Original Research
Automated Breast Ultrasound in Breast Cancer Screening of Women With Dense Breasts: Reader Study of Mammography-Negative and Mammography-Positive Cancers
Audio Available | 
Share
OBJECTIVE. The objective of our study was to assess and compare, in a reader study, radiologists' performance in the detection of breast cancer using full-field digital mammography (FFDM) alone and using FFDM with 3D automated breast ultrasound (ABUS).
MATERIALS AND METHODS. In this multireader, multicase, sequential-design reader study, 17 Mammography Quality Standards Act–qualified radiologists interpreted a cancer-enriched set of FFDM and ABUS examinations. All imaging studies were of asymptomatic women with BI-RADS C or D breast density. Readers first interpreted FFDM alone and subsequently interpreted FFDM combined with ABUS. The analysis included 185 cases: 133 noncancers and 52 biopsy-proven cancers. Of the 52 cancer cases, the screening FFDM images were interpreted as showing BI-RADS 1 or 2 findings in 31 cases and BI-RADS 0 findings in 21 cases. For the cases interpreted as BI-RADS 0, a forced BI-RADS score was also given. Reader performance was compared in terms of AUC under the ROC curve, sensitivity, and specificity.
RESULTS. The AUC was 0.72 for FFDM alone and 0.82 for FFDM combined with ABUS, yielding a statistically significant 14% relative improvement in AUC (i.e., change in AUC = 0.10 [95% CI, 0.07–0.14]; p < 0.001). When a cutpoint of BI-RADS 3 was used, the sensitivity across all readers was 57.5% for FFDM alone and 74.1% for FFDM with ABUS, yielding a statistically significant increase in sensitivity (p < 0.001) (relative increase = 29%). Overall specificity was 78.1% for FFDM alone and 76.1% for FFDM with ABUS (p = 0.496). For only the mammography-negative cancers, the average AUC was 0.60 for FFDM alone and 0.75 for FFDM with ABUS, yielding a statistically significant 25% relative improvement in AUC with the addition of ABUS (p < 0.001).
CONCLUSION. Combining mammography with ABUS, compared with mammography alone, significantly improved readers' detection of breast cancers in women with dense breast tissue without substantially affecting specificity.
Keywords: breast imaging, screening, ultrasound, whole-breast ultrasound
Mammography allows the early detection of nonpalpable breast cancers and has an overall sensitivity of approximately 85% [1, 2]. However, 40% of women in the United States have dense breast tissue, which decreases the sensitivity of mammography and results in nearly one third of breast cancers not being mammographically visible [3, 4]. Additionally, breast cancers diagnosed in women with dense breast tissue are often larger and are more frequently node-positive than cancers diagnosed in women who have nondense breast tissue [5–7].
Not only does dense breast tissue result in a lower sensitivity of mammography, but it is also an independent risk factor for developing breast cancer. Women with dense breast tissue have up to a four- to sixfold increased risk of developing breast cancer when compared with women with fatty breast tissue [8–10]. For women with dense breast tissue, the limitations of mammography combined with the increased risk of developing breast cancer suggest the need for additional imaging modalities to improve breast cancer detection.
The recent enactment of legislation in many states requires that women be informed of their breast density and, in some cases, be informed that additional imaging modalities are available to detect mammographically occult breast cancer in dense breast tissue [11].
Handheld screening ultrasound has been shown to improve breast cancer detection in women with dense breast tissue [12–16]. A study of women at increased risk for breast cancer showed that handheld screening ultrasound results in the detection of an additional 4.2 cancers per 1000 screening examinations [16]. However, that study also reported that handheld screening ultrasound requires 20 minutes of scanning time and yields more false-positives. Additional studies in subsequent years have reported that handheld screening ultrasound in women with increased cancer risk and dense breast tissue continues to show similar detection of mammographically occult breast cancer [17]. In Connecticut, the first state to enact the breast density inform legislation, postimplementation studies have likewise shown that screening breast ultrasound detects an additional 2–4 cancers per 1000 women [18]. However, the same limitations of handheld screening breast ultrasound—the time required to perform the examination and the greater number of false-positives—were reported again. Because of a broad range of skill levels in performing handheld ultrasound, there is operator variability and dependence, which result in a lack of standardization and an increased number of false-positives. Another study has shown that a reduction in operator dependence can be obtained through standardization of the imaging technique and operator training [19]. However, without standardization and improvements in workflow, widespread integration into the screening environment is not feasible.
The development of screening automated breast ultrasound (ABUS) allows improved detection of breast cancer, standardization, and efficient workflow integration [20]. ABUS covers the entire breast, automates the ultrasound scanning process, reduces the problems of operator subjectivity and variation, and provides technique standardization [21]. Another advantage of ABUS is that the images acquired in the transverse plane are reconstructed in 3D to allow interpretation in the coronal plane, which has been shown to improve detection of breast cancer [21] (Fig. 1).
 View larger version (283K) | Fig. 1A —Example case of 70-year-old woman who underwent screening full-field digital mammography (FFDM) and automated breast ultrasound (ABUS). A, Stable findings on screening FFDM images were assessed clinically as BI-RADS density C and BI-RADS assessment category 2 (i.e., routine annual follow-up recommended). |
 View larger version (364K) | Fig. 1B —Example case of 70-year-old woman who underwent screening full-field digital mammography (FFDM) and automated breast ultrasound (ABUS). B, Stable findings on screening FFDM images were assessed clinically as BI-RADS density C and BI-RADS assessment category 2 (i.e., routine annual follow-up recommended). |
 View larger version (109K) | Fig. 1C —Example case of 70-year-old woman who underwent screening full-field digital mammography (FFDM) and automated breast ultrasound (ABUS). C, Abnormal findings (arrow, D; circles, D) on 3D ABUS images yielded BI-RADS assessment category 0 (i.e., immediate management recommended). Ultrasound-guided core needle biopsy was performed. Pathology revealed invasive ductal carcinoma. |
 View larger version (108K) | Fig. 1D —Example case of 70-year-old woman who underwent screening full-field digital mammography (FFDM) and automated breast ultrasound (ABUS). D, Abnormal findings (arrow, D; circles, D) on 3D ABUS images yielded BI-RADS assessment category 0 (i.e., immediate management recommended). Ultrasound-guided core needle biopsy was performed. Pathology revealed invasive ductal carcinoma. |
Recently, a multiinstitutional study of more than 15,000 women showed that the addition of ABUS to digital mammography resulted in an increased detection of 2 cancers per 1000 women screened [20]. This study was an observational study in the clinical setting that included women with dense breast tissue as the criterion for inclusion. This study is in distinction to previous studies investigating screening breast ultrasound that included women with additional risk factors. In this study [20], most of the cancers detected were small, invasive, node-negative cancers. Additionally, a recent study showed that the use of ABUS not only increases the sensitivity of cancer detection, but also decreases the variability of the readers in the study in interpretation, suggesting that the use of ABUS results in improved consistency of interpretation [22].
The purpose of our study was to assess and compare radiologist performance in breast cancer detection in women with dense breast tissue using screening full-field digital mammography (FFDM) alone and using screening FFDM with ABUS.
| Materials and Methods |   |
|---|
An institutional review board–approved, HIPAA-compliant, sequential-design, multireader, multicase ROC reader study was conducted using a cancer-enriched set of FFDM and 3D ABUS images of asymptomatic women with breast density classified as BI-RADS C (heterogeneously dense) or D (extremely dense).
All women older than 18 years old with BI-RADS C or D breast tissue who were asymptomatic and presented for screening mammography from 2009 to 2010 at 13 institutions were invited to participate in our study. All mammograms were obtained with U.S. Food and Drug Administration–approved FFDM mammography systems. The case set for the reader study was obtained from the 13-site clinical registry study [23] (Fig. 2).
 View larger version (60K) | Fig. 2 —Schematic shows retrospective case selection from prior multicenter clinical case-collection study. “Both conditions” refers to full-field digital mammography (FFDM) alone and FFDM with automated breast ultrasound. |
All ABUS images were obtained using an ABUS system (somo•v Automated Breast Ultrasound System, GE Healthcare) that is composed of the two following major components: an Automated Breast Scan Station (GE Healthcare) and a somo•VIEWer Workstation (U-Systems). This ABUS system images a breast in the axial plane with a 15.4-cm transducer (768 elements and frequency range of 6–14 MHz). The ultrasound image data are obtained from a 15.4 × 17 × 5 cm volume of breast parenchyma in 1 minute. Spatial compounding, tissue equalization, and gain correction technologies were used during the acquisition. After the acquisition was completed, proprietary postprocessing algorithms were applied including coronal reconstruction.
The reader study used a multireader multicase ROC study design [24–26] and was conducted in 2011. A sequential reading design was used to simulate the proposed clinical method of case interpretation in a screening setting in which ABUS would be interpreted after interpretation of the screening mammograms. Here, a “case” is defined as four-view screening mammograms and, for most cases, the ABUS volumes of each breast. Typically, ABUS images are obtained to cover an entire breast and are acquired from three projections that differ in the location of the transducer during the axial sweep (anteroposterior [AP], lateral, and medial).
Screening FFDM images were displayed on a viewer (SecurView DX Viewer, Hologic), and ABUS images were displayed on a workstation (somo•VIEWer, U-Systems). This workstation also served for the electronic case report input during the reader study. The entire dataset of the sonographic images was used, and the images were viewed on the workstation.
Each reader interpreted all cases in a randomized order. During the study, no prior images were available for comparison, no clinical information or patient history was provided to the readers, and no handheld ultrasound data were provided. For each case, each reader interpreted and initially scored the screening mammograms. After the results of the mammography (FFDM alone) interpretations were recorded, the ABUS images were made available (FFDM combined with ABUS). An electronic case report form was used by the readers to input their rating data for direct collection and subsequent statistical analyses (Fig. 3).
 View larger version (121K) | Fig. 3 —Screen shot shows electronic case report form used by radiologists in reader study for automated breast ultrasound studies. |
During each interpretation condition (i.e., FFDM alone or FFDM with ABUS), the reader identified any potential lesion and chose a description of the lesion using standardized BI-RADS terms. Each reader then assigned a likelihood of malignancy (0–100% scale) to the lesion and a BI-RADS assessment category of 0 (recall), 1 (negative), or 2 (benign). Then, if the reader gave a BI-RADS category of 0, an additional forced diagnostic BI-RADS assessment was given using BI-RADS assessments of 3, 4A, 4B, 4C, or 5. Finally, the reader gave the case a likelihood that the woman had cancer using a 0–100% scale (i.e., the reader's estimate of the likelihood of malignancy using the interpretation) (Fig. 3).
The case set for the reader study came from the 13-site clinical registry study [20]. Cases were selected to yield a cancer-enriched case set, which allowed a manageable number of cases for the reader study. All cases were obtained from asymptomatic women with breast density assessed as BI-RADS C or D by the registry clinical site radiologist during the initial mammographic screening interpretation (Fig. 2).
The total reader study case set included 200 cases: 145 noncancer cases and 55 biopsy-proven malignancies. Of the 200 cases, 185 were included in the statistical analyses. The remaining 15 cases with only mammograms were included in the reader study sessions and were excluded from the statistical analyses to maintain reader vigilance and reduce bias caused by reader anticipation that ABUS images would follow. That is, for these 15 cases (12 BI-RADS 1 or 2 cases and three BI-RADS 0 cases), no ABUS images were presented in the reader study.
Note that the status of each case in the reader study was determined on the basis of its status at the clinical site—both in terms of radiologist detectability and pathology. For example, a mammography-negative cancer case means that it was not detected on mammography at the clinical site and it was a biopsy-proven cancer. Of the 52 cancer cases in the reader study analyses, 31 cases had FFDM studies assigned BI-RADS 1 or 2 at the clinical site (i.e., mammography-negative but detected on ABUS at the clinical site) [20] and 21 had been FFDM studies assigned BI-RADS 0 at the clinical site (i.e., mammography-positive without the use of ABUS at the clinical site). The noncancer cases were randomly selected from 6506 BI-RADS 1 or 2 cases.
Regarding case selection, all mammography-negative, ABUS-detected cancers from the clinical registry study [20] were included in the reader study. However, the mammographically visible cancers representing masses, calcifications, and architectural distortions were randomly selected for the reader study from the mammographically visible cancers in the clinical registry study [20].
Several case selection criteria and quality control mechanisms were applied. Cancers were included if assessable bilateral craniocaudal and mediolateral oblique views were available for the initial mammographic screening interpretation and if the clinical site radiologist at the time of initial reading assigned a BI-RADS assessment category of 1 or 2 for the mammography-negative cases or a BI-RADS assessment category of 0 for the mammography-positive cases. In addition, cancers were included if assessable bilateral AP, lateral, and medial ABUS views were available; if there were no significant protocol deviations that could be expected to bias reader interpretation; if source records were available for clinical status verification purposes; and if complete electronic data capture records were available.
Cases were excluded if cases were assessed as a BI-RADS composition or density score of A or B, if cases were declared noncancer cases and had prior breast interventional procedures, or if there were administrative or technical errors. Examples of the latter errors are incomplete or missing views from the FFDM or ABUS examination and inadequate FFDM or ABUS image quality due to technologist error in labeling or positioning or in acquisition technique as determined by an adjudicating radiologist (Fig. 4).
 View larger version (200K) | Fig. 4A —Examples of acceptable automated breast ultrasound (ABUS) images and of rejected ABUS images based on poor positioning of transducer. A, Right anteroposterior ABUS images with properly positioned transducer. |
 View larger version (228K) | Fig. 4B —Examples of acceptable automated breast ultrasound (ABUS) images and of rejected ABUS images based on poor positioning of transducer. B, Right anteroposterior ABUS images with transducer positioned too low, excluding superior tissue. |
The pathology, size, and stage of the 31 mammographically occult and the 21 mammographically visible cancers were determined (Table 1).
All BI-RADS 1 or 2 mammograms from the registry study that were not excluded due to clinical history were considered noncancer cases (n = 4008 cases). A random sampling from this group, after exclusion due to technical reasons and that had both FFDM and ABUS images, was included in the analysis (n = 133). These cases included noncancer cases confirmed by 1-year follow-up mammography (at least 365 days from original screening mammography study), cancer cases confirmed by follow-up in which a biopsy-proven breast cancer was identified more than 365 days after the original screening mammography study (where no cancer was confirmed before 1-year follow-up), and cases with no follow-up (i.e., unverified cases for which no follow-up confirmation regarding the presence or absence of breast cancer could be obtained). This latter group was included so that no category of patient was excluded from the study, thus making it more similar to clinical practice.
Two board-certified expert “truthing” radiologists—who each read more than 2000 mammography examinations annually, are expert trainers in ABUS interpretation, and were not readers in this study—independently reviewed all cancer cases and confirmed the location and size of each malignancy on FFDM and ABUS. The truthing radiologists had complete access to source records for each cancer case. The source records included the clinical site radiologist's initial image reading results for FFDM and ABUS, the workup reports, biopsy reports with malignancy location identification, and final surgical pathology reports.
The readers were 17 Mammography Quality Standards Act–qualified radiologists who were fellowship-trained in breast imaging or had 10 years or more of experience in breast imaging in a practice that was at least 70% breast imaging (or both). In addition, to participate in the reader study, each reader had interpreted at least 1000 mammography examinations and 500 (handheld) breast ultrasound examinations in the prior year. The readers consisted of physicians from varying clinical settings: Seven came from academic radiology practices, six from private practice, and four from community clinics.
All readers participated in ABUS training via online webinars and an in-person 1-day training session before the reader study. Each reader passed a skill test consisting of 25 cases including 10 biopsy-confirmed cancers, 10 biopsy-confirmed benign lesions, and five negative cases. Training was performed using the same electronic case report form used to record the readers' interpretations during the actual reader study.
Reader performance in detecting cancer was assessed using the AUC under the ROC curve for each reader [25]. The reader-indicated likelihood of malignancy (0–100% scale) was used, and an ROC analysis was conducted on the likelihood of malignancy ratings from FFDM alone and from FFDM with ABUS. The AUC values were computed using the trapezoidal method for statistical analyses, and the proper binormal ROC model for curve plotting (DBM MRMC [version 2.31, Berbaum KS, Metz CE, Pesce LL] for Microsoft Windows OS). CIs for the average AUCs and the differences in AUCs were computed using the ANOVA-after-jackknife method following the approach of Dorfman, Berbaum, and Metz [25–27].
Sensitivity and specificity analyses were based on a BI-RADS cutpoint of 3—that is, cases were considered as recalled for BI-RADS ratings of 3 or higher. Sensitivity and specificity for each reader were calculated for FFDM alone and for FFDM with ABUS. Differences in sensitivity and specificity corresponding to the sequential effect [(FFDM with ABUS) – (FFDM alone)] were computed. CIs for overall reader sensitivities and specificities were calculated from 1000 bootstrap samples (bootstrapping both cases and readers) with cases and readers included or excluded from each sample in their entirety.
Results are given for all 52 cancers and separately for the 31 mammography-negative cancers and for the 21 mammography-positive cancers.
| Results | |
|---|
From the reader study including all 52 cancers comparing FFDM with ABUS to FFDM alone, both the improvement in average AUC and sensitivity were statistically significant (p < 0.001), with an increase in average AUC and sensitivity of 0.10 and 16.6% (relative increase of 29%), respectively (Table 2). The modest decline in specificity of 1.9% (relative decrease of 2.4%) failed to reach statistical significance (p = 0.496), and the noninferiority of the specificity for the FFDM-with-ABUS condition could not be established with respect to the FFDM-alone condition (Table 2).
When we compared FFDM with ABUS to FFDM alone for the mammography-negative cancers (n = 31), the improvements in both average AUC and sensitivity were statistically significant. The increase in average AUC was 0.15 (p < 0.001), and the increase in sensitivity was 23.9% (p = 0.004) (relative increase or 62%) (Table 2).
When we compared FFDM with ABUS to FFDM alone for the mammography-negative cancers without prior breast interventions (n = 16), the improvements were even more pronounced. The increase in average AUC was 0.21 (p < 0.001), and the increase in sensitivity was 35.7% (p < 0.001) (relative increase of 110%) (Table 2). Note that this separation of the 31 cancers was done for clinical interest as a tertiary analysis.
When we compared FFDM with ABUS to FFDM alone for the mammography-positive cancers (n = 21) (i.e., those FFDM-detected cancers at the 13 contributing clinical sites), the improvement in AUC of 0.05 was also statistically significant (p = 0.002) (Table 2 and Fig. 5); however, the improvement in sensitivity of 5.9% failed to reach statistical significance (p = 0.234) (Table 2).
 View larger version (63K) | Fig. 5 —Performance of 17 readers in terms of ROC curves in detecting mammography-negative and mammography-positive breast cancers for two conditions of full-field digital mammography (FFDM) alone and FFDM with automated breast ultrasound (ABUS). Operating points indicated with triangles were obtained from sensitivity and specificity analyses (Table 2), and both horizontal and vertical lines are error bars showing SDs obtained through bootstrapping. |
For both the FFDM-alone and FFDM-with-ABUS conditions, there was a considerable range of performance for each of the 17 individual readers (Fig. 6). Note that the sequential effect on AUC was positive for all readers: That is, all readers obtained a higher AUC for the FFDM-with-ABUS condition than for the FFDM-alone condition. Although the performance of all individual readers seemed to improve after ABUS interpretation in terms of AUC (Fig. 7A), the effects on individual sensitivity (Fig. 7B) and specificity (Fig. 7C) were less clear. Note that none of the readers detected all cancers deemed mammography-positive when interpreting FFDM alone and that only three of the readers detected all mammography-positive cancers after ABUS interpretation (Fig. 7B).
 View larger version (88K) | Fig. 6A —Performance of 17 readers for entire dataset. A, Bar graphs show individual performance of 17 readers for entire dataset in terms of AUC (A), sensitivity (B), and specificity (C) values for two conditions of reader study: full-field digital mammography (FFDM) alone and FFDM with automated breast ultrasound (ABUS). Error bars show SD. |
 View larger version (86K) | Fig. 6B —Performance of 17 readers for entire dataset. B, Bar graphs show individual performance of 17 readers for entire dataset in terms of AUC (A), sensitivity (B), and specificity (C) values for two conditions of reader study: full-field digital mammography (FFDM) alone and FFDM with automated breast ultrasound (ABUS). Error bars show SD. |
 View larger version (98K) | Fig. 6C —Performance of 17 readers for entire dataset. C, Bar graphs show individual performance of 17 readers for entire dataset in terms of AUC (A), sensitivity (B), and specificity (C) values for two conditions of reader study: full-field digital mammography (FFDM) alone and FFDM with automated breast ultrasound (ABUS). Error bars show SD. |
 View larger version (70K) | Fig. 7A —Case-based performance of the 17 individual readers in terms of AUC, sensitivity, and specificity. A, Relative reader performance of individual 17 reader case-based performances in terms of AUC (A), sensitivity (B), and specificity (C) values shown separately for mammography-negative and mammography-positive breast cancers for two conditions of full-field digital mammography (FFDM) alone and FFDM with automated breast ultrasound (ABUS). Note that data points above dashed line indicate improvement due to addition of ABUS images in reading interpretations. For specificity, larger symbols indicate that two readers obtained same performance. |
 View larger version (71K) | Fig. 7B —Case-based performance of the 17 individual readers in terms of AUC, sensitivity, and specificity. B, Relative reader performance of individual 17 reader case-based performances in terms of AUC (A), sensitivity (B), and specificity (C) values shown separately for mammography-negative and mammography-positive breast cancers for two conditions of full-field digital mammography (FFDM) alone and FFDM with automated breast ultrasound (ABUS). Note that data points above dashed line indicate improvement due to addition of ABUS images in reading interpretations. For specificity, larger symbols indicate that two readers obtained same performance. |
 View larger version (50K) | Fig. 7C —Case-based performance of the 17 individual readers in terms of AUC, sensitivity, and specificity. C, Relative reader performance of individual 17 reader case-based performances in terms of AUC (A), sensitivity (B), and specificity (C) values shown separately for mammography-negative and mammography-positive breast cancers for two conditions of full-field digital mammography (FFDM) alone and FFDM with automated breast ultrasound (ABUS). Note that data points above dashed line indicate improvement due to addition of ABUS images in reading interpretations. For specificity, larger symbols indicate that two readers obtained same performance. |
| Discussion | |
|---|
Women with dense breast tissue constitute the largest population of intermediate-risk women with a 15–20% lifetime risk of developing breast cancer. The current recommendation is to screen this population of women with mammography alone and forego any additional supplemental imaging that could potentially detect the 35% of mammographically occult breast cancers that are currently underdetected with screening mammography [28]. Ultrasound has been shown to substantially increase the detection of breast cancers not visible on mammography in both women with dense breasts who are at higher risk for breast cancer [16] and women who have dense breast tissue alone [12, 18, 20]. The purpose of this reader study was to compare screening FFDM with screening FFDM plus ABUS in a population of women with dense breasts. Limiting the population to include only those at higher risk for breast cancer would not have reflected routine clinical practice especially because 85% of women who develop breast cancer have no family history of breast cancer [28].
In this reader study, compared with screening FFDM alone, screening FFDM combined with ABUS showed a statistically significant improvement in readers' ability to detect mammography-negative breast cancers and mammography-positive breast cancers in women with dense breasts without a significant reduction in specificity. In some cases, the ABUS findings obviated recall of women for evaluation of mammographic findings because the mammographically depicted mass was shown to be a simple cyst with ABUS. We find the 110% relative increase in sensitivity for the 16 mammographic-negative cancers that did not have any prior breast interventions to be of special interest. This finding shows that the confidence and performance of screening interpretations are improved with the use of supplemental screening ABUS.
The ABUS technique allows standardization and reduces operator dependence, which results in only a minimal impact on specificity. The specificity results in this study differ from those reported in a prospective multiinstitutional observational clinical registry study [20]. An explanation for this difference is likely secondary to varying training procedures provided to the interpreting physicians due to resource constraints with the early technology. A robust training program was instituted with this study, unlike the clinical registry study, that highlighted the importance of quality assurance and quality control efforts and multiplanar interpretation to help minimize false-positives. Therefore, with subsequent implementation of ABUS technology, a strong emphasis should be placed on the importance of training. Although further economic modeling is required, the encouraging nonsubstantial increase in false-positives suggests that broad implementation could be financially feasible.
The use of ABUS instead of bilateral handheld ultrasound to screen women may result in a more efficient integration of ultrasound into the screening workflow environment [20]. Acquisition of each ABUS image requires approximately 1.5 minutes; with the usual three images per breast, the ABUS examination time is far less than the 20 minutes reported for scanning with hand-held ultrasound [29]. The scanning times for handheld ultrasound make it difficult, if not unfeasible, to easily integrate handheld screening ultrasound into the screening environment. Also, if handheld scanning is performed by an ultrasound technologist, the significant findings are those detected and saved by the scanning technologist, not by the interpreting radiologist. In contrast, the automated image acquisition uncoupled from interpretation, as occurs with ABUS, allows the interpreting radiologist to analyze the entire dataset. The time to image the patient and the time for interpretation and reporting were not part of this study. However, in another related study, the physician interpretation time has been found to be 3 minutes with ABUS [30].
The previous studies evaluating screening breast ultrasound, whether automated [20] or handheld [12, 18], were performed in a clinical setting with a single radiologist interpreting the studies. In this reader study, 17 radiologists interpreted the FFDM studies versus the FFDM plus ABUS studies to allow determination of the AUC, an approach that is well established for rigorous evaluation of the impact of a new technology on incremental cancer detection imaging modalities.
A limitation of this study is that the study was a reader study and not a clinical trial; thus, the radiologists' performance may not match the actual clinical performance [31]. However, the performance of a reader study neutralizes the impact of a single radiologist's interpretation because 17 radiologists interpreted the same examinations, albeit in different order. In addition, many studies are performed only in academic intuitions that are not representative of varying clinical environments. Given the broad range of physicians from differing clinical settings included in this study, the possibility of similar real-world results is plausible. Furthermore, a preliminary study on reader concordance, which used a part of the reported dataset, focused on individual readers and showed acceptable interreader agreement with the use of ABUS [22]. The interreader concordance again shows the feasibility of implementing ABUS into routine clinical practice.
Our study confirms the findings of observational studies [12, 18, 20] that the use of supplemental screening breast ultrasound results in the detection of additional cancers that were not detected with mammography alone. These same studies have shown that most of these sonographically detected cancers are small, node-negative, invasive cancers [12, 18, 20] that are clinically important to detect in their earliest stages for improved prognosis.
| Acknowledgments | |
|---|
We thank Dave P. Miller, Ingrid Reiser, and Robert M. Nishikawa for their help with this study.
| References | |
|---|
