Imaging and Pathology of Ductal Carcinoma in Situ of the Breast: The Forest and the Trees

See also the article by Lee and Kim et al in this issue.

Dr Habib Rahbar is an associate professor and vice chair of clinical operations in the Department of Radiology at the University of Washington. His primary research focus is on the use of breast MRI in detecting and characterizing ductal carcinoma in situ. He is a National Institutes of Health–funded principal investigator, a fellow of the Society of Breast Imaging, and cochair of the ECOG-ACRIN Radiomics Working Group.

Ductal carcinoma in situ (DCIS) of the breast is considered to be the earliest form of breast cancer and is usually identified in women who are asymptomatic and present for screening. Although early detection of breast cancer remains a vital pillar of the medical success story of decreasing breast cancer mortality over the past 5 decades, DCIS also accounts for the most instances of breast cancer overdiagnosis. Breast cancer overdiagnosis is problematic because it leads to overtreatment. Overtreatments include the use of surgery and adjuvant therapies more appropriate for full-fledged invasive cancer than a nonlethal precursor (1).

Part of the challenge in addressing DCIS overtreatment is it is generally treated as a single disease rather than what it is, which is a range of intraductal proliferations reflecting a spectrum of biologic potential and aggressiveness. To this end, individual DCIS lesions demonstrate substantial intralesional heterogeneity, with half composed of multiple nuclear grades within the same lesion (2). The inherent heterogeneity of individual DCIS lesions combined with the subjective and variable outcomes in nuclear grade characterization (3) and discrimination from related atypical lesions (4) by using limited sections of tissue highlights the daunting task our colleagues in the field of pathology face when diagnosing and characterizing DCIS. To make matters worse, up to a quarter of DCIS lesions diagnosed at core-needle biopsy (CNB) upstage to invasive cancer at surgical excision (5). Even surgical pathologic analysis is prone to sampling error: It is estimated that less than 0.2% of excised breast tissue in a lumpectomy is evaluated with standard-of-care pathologic analysis by using slides that consist of ultrathin tissue slices (6).

The imprecise pathologic characterization of DCIS is problematic because nuclear grade, comedonecrosis, and hormone receptor positivity are the principal markers of aggressiveness and risk of recurrence that clinicians rely on to make treatment decisions. For example, an early report from the Low Risk DCIS, or LORIS, active surveillance trial of DCIS indicated that almost half (45 of 100) of potentially eligible patients had differing assessments of comedonecrosis, nuclear grade, and the presence of invasive disease between site and central levels (7), highlighting the challenge in identifying appropriate populations for whom DCIS treatment may be de-escalated (8). Whereas the problem of invasive breast cancer understaging in the setting of CNB-diagnosed DCIS is acknowledged and studied (5,9), the issue of imprecise DCIS nuclear grade assessment remains underappreciated. Each DCIS lesion is heterogeneous. Thus, a CNB-diagnosed DCIS lesion classified as low nuclear grade and considered an appropriate candidate for active surveillance may in fact consist of higher risk areas filled with intermediate to high nuclear grade DCIS because of sampling error.

In this issue of Radiology (10), Lee and Kim et al aim to develop a predictive model by using conventional (mammography and US) imaging features to help increase the confidence of DCIS nuclear grade assessments at CNB. By using a single institution data set of 470 women with 477 pure DCIS lesions (ie, no presence of invasive disease at final excision), an expert breast imaging radiologist who was blinded to outcomes retrospectively characterized all lesions into one of five categories on the basis of the qualitative US and mammographic features. The authors then divided the patients into training and validation sets and evaluated whether these imaging categories, alone or in combination with clinical and pathologic features, could be used to confirm the presence of purely low nuclear grade DCIS and to exclude upgrade to higher nuclear grade assessment at surgical excision.

Remarkably, the authors found that almost four in 10 low nuclear grade DCIS lesions diagnosed at CNB were upgraded to a higher nuclear grade lesion at surgical excision, underscoring the challenge of relying on CNB pathologic features alone to stratify DCIS risk. They also found that lesions manifesting as a mass at US without mammographic calcifications and no comedonecrosis at CNB were much more likely to represent low nuclear grade DCIS, and the validation set demonstrated high diagnostic performance (area under receiver operating characteristic curve [AUC], 0.97). This is exciting because it could provide patients who are considering active surveillance approaches more reassurance that their DCIS lesions are low risk and may be monitored rather than excised. These features could also be integrated into active surveillance trials to increase certainty about appropriate patient selection.

Toward the ultimate goal of reducing overdiagnosis of low-risk DCIS, the authors also assessed the value of imaging features alone without CNB information to identify low nuclear grade DCIS lesions that might avoid biopsy altogether, which resulted in a lower but still impressive performance (AUC, 0.87). Finally, they explored the ability of imaging and pathologic features to help predict which DCIS lesions diagnosed at CNB will upgrade to higher nuclear grade lesions at surgery. They found that lesion size, presence of comedonecrosis, and Ki-67 positivity were associated with nuclear grade upstaging at surgery.

There are some important limitations worth emphasizing. The study is from a single institution that includes both mammographic and US information. As the authors acknowledge, many sites do not routinely use US in evaluation of DCIS lesions, which limits clinical translation. In addition, the study does not leverage the value of DCIS characterization at breast MRI. Breast MRI has the potential to further refine DCIS risk assessments and help identify a distinct subset of biologically active DCIS independent from nuclear grade and other pathologic analysis–based assessments because of its ability to reflect angiogenesis (1). Qualitative imaging feature categorization was performed by a single expert radiologist familiar with appearance of DCIS at US and mammography; it thus remains unclear if these categorizations are reproducible. Finally, the study relies on an imperfect reference standard of surgical excision assessment of nuclear grade.

This study also underscores a truth about the pathologic evaluation of DCIS and the supportive role imaging can play that remains underrecognized. Current standard-of-care pathologic evaluation of breast lesions is limited because it reflects small samples of otherwise heterogeneous lesions. This point is illustrated by the fact that so many low nuclear grade DCIS lesions diagnosed at CNB were upgraded to a higher nuclear grade at excision. The study also suggests that imaging features can provide large-scale information of the entirety of a DCIS lesion complementary to the microscopic pathologic assessment to allow for improved radiology-pathology concordance. To adapt a common expression, imaging can provide the perspective of the whole forest whereas pathologic analysis provides the details of a few trees. Both perspectives will likely prove essential for optimal DCIS risk stratification, yet the imaging perspective remains underappreciated by clinicians and radiologists alike. To this end, reduction of DCIS overdiagnosis and overtreatment will indeed hinge on the ability of imaging to provide greater certainty that the pathologic features of a few DCIS "trees" provide an accurate representation of the whole "forest."

As breast imaging radiologists look to help our oncology specialists optimize DCIS management (8), it will become ever more important to recognize that within a given radiologic image, there is information beyond a Breast Imaging Reporting and Data System assessment and recommendation indicating whether or not a biopsy should be performed. Indeed, there is also prognostic information embedded across various breast imaging modalities that will likely only be fully realized when automated quantitative approaches leveraging advanced radiomics and artificial intelligence are used. The breast oncology community would be well served to consider this untapped potential as clinical trials are designed to address this controversial disease. In particular, the development of well-annotated multimodality imaging repositories linked to pathologic and clinical outcomes should be a priority.