Abstract
Background: Retrospective analysis of breast cancer imaging methods is a common tool for evaluating the effectiveness of ultrasound and mammography regarding ductal carcinoma in situ (DCIS). No large number subpopulation of pure DCIS has been reported. It is however known that mammography and ultrasound underestimate tumor dimension with increasing tumor size. We aimed to quantify this discrepancy. Materials and Methods: This retrospective analysis reviewed the ultrasound and mammography data from 173 patients with DCIS at the University of Cologne - Department of Gynecology and Obstetrics between the years 2007 and 2010. Of these 173 patients, 34 fulfilled the DCIS analysis requirements and were evaluated in this study. Results: Overall, ultrasound underestimated tumor size 79.4% of the time, while overestimating only 20.6% of the time. Mammography underestimated tumor size in 50%, overestimated in 38.2%, correctly estimating in 11.8%. Over and underestimation distributions differed drastically above and a cut-off point of ≤2 cm actual tumor size, with a significant shift toward severe underestimation by both methods above a tumor size of 2 cm. DCIS misestimation was defined as the absolute value of the difference between actual tumor size and pre-surgical measurement by an imaging method. Mean DCIS size misestimation (actual tumor size ≤2 cm) was 3 mm for ultrasound and 6.2 mm for mammography. Conclusion: We support previous findings that ultrasound and mammography lose accuracy with increasing tumor size. Nonetheless, ultrasound may be more useful in estimation of DCIS size for tumors ≤2 cm than previously expected.
Retrospective analysis of breast cancer imaging methods is a common tool for evaluating the effectiveness of ultrasound (US) and mammography regarding ductal carcinoma in situ (DCIS). A large number of publications are available on the topic of preoperative breast cancer imaging. However, detailed evaluations of the pure DCIS subpopulation remain small in number. The importance of presurgical estimation of tumor size cannot be overstated as both systemic and surgical approaches require accurate tumor size evaluation.
Recent evaluations of US analyses in direct comparison to magnetic resonance imaging (MRI) and mammography showed that US underestimates tumor size, whereas mammography and MRI tend to more accurately describe tumor size (1-5). All evaluated methods were found to become increasingly inaccurate as tumor size increased (1, 6). Of the three preoperative screening options mentioned above, MRI should, however, be considered optional, due to its significant cost. Nonetheless, when diagnosing pure DCIS, MRI has been shown to be accurate and useful (7, 8), although often times not readily available. On the other hand, literature states that the correlation of DCIS with micro-invasion evaluated by US is higher than that obtained using MRI (2, 9). Some new methods have elaborated on the use of US with automated breast volume ultrasound systems with promising results, albeit sample numbers remain small (10). Lesion description in literature is generally based on the Breast Imaging Reporting and Data System (BI-RADS) (11, 12). Altogether, sample numbers are small, prospective studies are rare and an accurate size estimation of DCIS has not been possible, especially in patients with large tumors.
This retrospective analysis, therefore, had several goals: i) Evaluation of method accuracy: US versus mammography in a pre-therapeutic setting for pure DCIS entities; ii) Evaluation of imaging inaccuracy based on actual tumor size; and iii) evaluation of over- and underestimation distributions based on tumor size.
Materials and Methods
This retrospective analysis reviewed the US and mammographic data from 173 patients with DCIS at the University of Cologne - Department of Senology between the years 2007 and 2010. Of the 173 patients, 34 fulfilled the following requirements: unilateral, pure DCIS, no invasive component, no prior breast cancer nor other malignancy, having undergone mammography as well as bilateral breast US evaluation, no prior chemotherapy or radiation therapy. Identifying patient data was only available to the reviewing physician. Surgical techniques included breast-conserving surgery (BCS), mastectomy and skin-sparing or subcutaneous mastectomy (SSM). The average patient age was 57.7 years. Both pre-and postmenopausal patients were evaluated. For the purpose of this work, it was considered to be beneficial to patients if tumor size were overestimated rather than underestimated, as this might require revision surgery due to a R1 resection.
Diagnostic equipment. Mammography was performed using a Selenia Dimensions mammography system (Hologic, Bedford, MA, USA). US was performed by personnel qualified to at least DEGUM Level 2 (Deutsche Gesellschaft für Ultraschall in der Medizin e.V.) using Voluson E8 and a Voluson 730 system (GE Healthcare, Little Chalfont, UK).
Diagnostic criteria. As this was a retrospective study, no specific diagnostic criteria were required by the physicians when diagnosing DCIS within their respective imaging method. Naturally, the absence of micro-calcification, breast compression during mammography, and high-density breast tissue [American College of Radiology (ACR) 4] (12) limits mammography. In addition, US limitations are highly dependent on investigator skill and imaging equipment. US characteristics of size, shape, location orientation, lesion boundary, and echogenicity were based on the BI-RADS system.
Patient consent and ethics. This study was conducted in accordance with institutional Review Board standard operating procedures. Ethics Committee approval was deemed unnecessary by the Ethikkommission der Aerztekammer Nordrhein; a written statement to this extent is available.
Statistical analysis. The largest tumor diameter was chosen to represent the overall diameter. Only DCIS lesions were included in the study. Qualitative analysis was performed using Bland–Altman plot. Subsequent comparison to existing data was performed using published data points. A head-to-head analysis was performed using the mean square error (MSE), as well as absolute deviations, absolute mean error and global mean error [for comparison to Gruber et al. (1)]. Mann–Whitney and chi-squared analysis was used whenever appropriate. Statistical analysis was performed using the VassarStats® (Vassar College, Poughkeepsie, NY, USA) statistics program. The level of acceptable significance was defined as a p-value of less than 0.05.
Overall results of the 34 measurements. Method over- and underestimation with respect to post-surgical pathology reports for ductal carcinoma in situ (DCIS) are shown, as well as American College of Radiology (ACR) and grading distributions.
Results
Of 173 patients, 34 patients had US DCIS estimation and mammographic size evaluation, as well as a complete pathology report of tumor size after resection. Overall, US underestimated tumor size in 79% of cases, while overestimating only 21%; accurate representation was not achieved in any case. Mammography underestimated tumor size 50% of the time, while overestimating it 38% and correctly estimating it in 12% of cases (Table I). Grading and breast density (ACR), as determined using mammography are also shown in Table I. Grading distribution was fairly even, with a grade 2 DCIS being present in 41% of all cases. ACR level 2 and 3 were the most common breast density levels, being present in 44% and 48% respectively.
The mean global preoperative/postoperative size deviation was 14.6 mm by US and 2.4 mm for mammography (underestimation by both methods). MSE analysis showed increasing deterioration of imaging accuracy above a tumor size of 2 cm (Table II). No statistically significant difference was shown between methods of size determination (p>0.05).
Bland–Altman plot. In order to adequately analyze patient data, three main parameters were evaluated. These included the estimated DCIS size prior to surgery as measured by US and mammography. Both values were then compared to the postsurgical tumor size including that of a possible re-intervention due to initial R1 status. A graphic representation of these results are found in the Bland-Altman plot in Figure 1. A positive correlation was found between increasing actual tumor size and initial tumor size estimation through US and mammography. Three cases of tumor size exceeding 10 cm were omitted from this plot as both US and mammography were entirely unable to approximate tumor size, meaning that a gross underestimation of tumor size was found – including this in the graphical representation would have been cumbersome. Data analysis of this plot is always of a qualitative nature. We found that actual tumor size of ≤2 cm, scatter varied between under- and overestimating tumor size in a similar manner. Exceeding 2 cm, however, it becomes apparent that both methods tended to underestimate the extent of the DCIS. Overestimation in tumors above 2 cm in actual size was only seen in a few cases by mammography. In data analysis and discussion, we therefore separate values using a cut-off of 2 cm actual tumor size. In addition, a regression line was derived for both datasets. Regression analysis suggests that tumor size underestimation is a problem for both methods, although US starts underestimating tumor size at approximately 1.13 cm diameter, whereas mammography begins to do so at 2.3 cm diameter. This of course is an approximation, nonetheless, a trend towards this behavior is clearly apparent.
This Bland-Altman plot depicts actual tumor size minus size by imaging method [ultrasound (US) and mammography (MG)] (Δ). A positive value of Δ indicates tumor size underestimation as depicted in the upper portion of the graph. Tumor size overestimation resulted in a value of Δ of below zero. Qualitatively, severe tumor size underestimation seemed to begin at a tumor size of approximately 2 cm. DCIS: Ductal carcinoma in situ.
Tumor size deviation by imaging method. The overall mean deviation (by imaging method versus actual) is displayed in the top portion of the table. Also listed are results taken from Gruber et al. (1). The mean square error (MSE) analysis of size by imaging method versus actual size is also listed. The MSE analysis shows a tremendous increase in tumor size underestimation above 2 cm tumor size.
Graphic representation of data for size estimation by ultrasound (US) and mammography (MG) for all ductal carcinoma in situ (DCIS) (A), those ≤2 cm in actual size (B) and those >2 cm (C). Overall, approximately 50% of tumor sizes were underestimated by both methods. Overestimation by both methods occurred approximately 21% of the time. Underestimation by ultrasound and overestimation by mammography made up the rest. In no instance was there overestimation of tumor size by ultrasound when underestimated by mammography. Tumor size underestimation increased drastically when actual tumor size exceeded 2 cm. Over- and underestimation distributions differed significantly between the two size subgroups (Chi-squared test p=0.0293).
Discussion
The focus of this work was the size-dependent evaluation of DCIS in US and mammography. Several publications are available regarding the evaluation of tumor size in US, mammography, and MRI (13-15). In particular, a recent publication by Gruber et al. described a significant underestimation of pre-surgical DCIS tumor size via US with a mean o deviation of 8 mm (1). This underestimation increased with histological tumor size. While Gruber et al. focused on several breast cancer entities, our work aimed at a sub-group analysis of DCIS alone. A data comparison is shown in Table III.
In order to maintain comparability to previous studies, no new diagnostic criteria or restrictions were introduced in this retrospective analysis. Decreasing mammography sensitivity with increasing breast density (ACR 4) is known and amply reported (16, 17). The absence of micro-calcification may also lead to a false-negative mammography result. Therefore, a combination of mammography and US is generally thought to compensate for this fact. Figure 2 shows typical images of a patient with DCIS only. It is clear that DCIS detection by US is difficult as it does not present as a typical focal lesion (i.e. displaying typical criteria for malignancy by US). Especially for larger DCIS entities, this non-typical presentation may be difficult to detect.
Although MRI support is often possible and will improve sensitivity, we may have to accept the fact that MRI diagnostics may not be readily available at all times. In our work, US underestimated overall size in 79% of the patients with DCIS, while mammography underestimation amounted to 50%. Park et al. recently reported DCIS size underestimation via US alone of 30.4% (independent of tumor size) (3), providing an underestimation range from 30-79% by US independently of tumor size. This underlines, once again, the necessity for additional diagnostic options.
Breast Imaging Reporting and Data System (BI-RADS) classifications by ultrasound and mammography in this study in direct comparison to data presented by Gruber et al. (1).
A typical under-representation of ductal carcinoma in situ, on ultrasound estimated 2.7 cm (left) and a correlating estimated 4 cm area of microcalcification on mammography (with preoperative wire marking) (right). Actual post-surgical tumor size was 4.8 cm.
Bland–Altman analysis. While an exact tumor size estimation seems realistically unachievable (here: 11.8% by mammography, 0% by US), an idea of at which point estimates become unreliable would certainly be helpful. From Figure 1 it is apparent that, qualitatively speaking, distributions of differences are evenly spread between over- and underestimation of tumor size by both methods (US and mammography) only for actual tumor sizes ≤2 cm. This changes with tumor size exceeding 2 cm. Qualitative Bland-Altman analysis should be supported using MSE analysis, as shown in Table II. Firstly, the overall mean deviation between US and pathology was 14.6 mm, which is very much comparable to data reported by Gruber et al. for this sub-group (15 mm). The same is true for the overall deviation between mammography and pathology. Again, for DCIS, mammography tends to be the more accurate method of the two. When dividing patients into two groups by actual tumor size, of 2 cm or less and more than 2 cm, results were unexpected. At an actual tumor size of 2 cm, the mean deviation was only −3 mm in the US and −6.2 mm in the mammography group (i.e. tumor size overestimation) – a trend favoring US, although this trend was not significant (p=0.2). When exceeding 2 cm tumor size, the mean deviation favors mammography. A significant increase in measurement accuracy is shown when comparing the MSE between the >2 cm and ≤2 cm groups for both methods, where MSE increased by an order of magnitude for US. Both imaging methods were thus shown to lose accuracy with increasing tumor size. This fact has previously been shown and reported (1). Our 34 cases therefore add to the subpopulation analysis reported by Gruber et al. (15 cases) (1) and further quantify and specify loss of method accuracy with increasing tumor size. In addition, this work filters out an area in which US may potentially still be used.
Distribution of method over- and underestimation. Upon distribution analysis of over- and underestimation based on DCIS size, Figure 3 shows an interesting trend. At a tumor size of 2 cm and less, mammography and US are found to both underestimate tumor size in approximately 50% of cases, with a mean deviation of −3 mm and −6.2 mm, respectively. Although combined underestimation is high, one might argue that this form of method inaccuracy may still be compensated for via surgical margins. In short: using mammography and US in combination, a DCIS will be underestimated in size about half the time, although the magnitude of error will be small.
Above 2 cm, however, overall mean deviation from actual tumor size, as calculated by MSE, increased by an order of magnitude for both methods (Table II). While for tumors of ≤2 cm we found a distribution fairly similar to that of the global distribution, with a tendency towards overestimation by US, a drastic change occurred above 2 cm tumor size. Here, 87% of DCIS size prediction was grossly underestimated by both methods (MSE: >220 mm2 US, mammography: 112 mm2; p=0.029). The general notion that US is the inferior method in DCIS imaging was therefore only true for tumor sizes above 2 cm in our patient collective and not the case for tumor size of ≤2 cm, as both mammography and US evaluated actual tumor size equally. The distribution of tumor size underestimation shifts significantly above 2 cm, where both US and mammography underestimate tumor size grossly.
Conclusion
This retrospective analysis of our patient collective showed that overall, mammography is slightly superior to US in pre-operatively estimating actual tumor size in patients with DCIS without an invasive component. The following findings should be noted: i) DCIS evaluation by US is comparable to that by mammography for tumors ≤2 cm. ii) For tumors above 2 cm, both US and mammography become exceedingly inaccurate. iii) Distributions between over- and underestimation of actual tumor size vary depending on tumor size: above 2 cm, both US and mammography tend to underestimate tumor size of DCIS. We conclude that US may be more useful in size estimation for DCIS of ≤2 cm than previously expected.
- Received May 19, 2017.
- Revision received July 1, 2017.
- Accepted July 5, 2017.
- Copyright© 2017, International Institute of Anticancer Research (Dr. George J. Delinasios), All rights reserved
