How Can Advanced Imaging Be Used to Mitigate Potential Breast Cancer Overdiagnosis?

Abstract

Radiologists, as administrators and interpreters of screening mammography, are considered by some to be major contributors to the potential harms of screening, including overdiagnosis and overtreatment. In this article, we outline current efforts within the breast imaging community towards mitigating screening harms, including the widespread adoption of tomosynthesis and potentially adjusting screening frequency and thresholds for image-guided breast biopsy. However, the emerging field of breast radiomics may offer the greatest promise for reducing overdiagnosis by identifying imaging-based biomarkers strongly associated with tumor biology and, therefore, helping prevent the harms of unnecessary treatment for indolent cancers.

Introduction

The recently revised breast cancer screening recommendations from the U.S. Preventive Services Task Force (USPSTF) and the American Cancer Society (ACS) have renewed the controversy around the potential benefits and harms of mammography among advocates and detractors of breast cancer screening.^1,2 While all authorities reiterate the mortality benefits of routine screening for the general population, they also now consider overdiagnosis and overtreatment among the potential harms of mammography. By definition, overdiagnosis is screen-detected cancer that would not have become clinically apparent during a patient’s lifetime.³ While it is now fairly widely accepted in the medical community as a legitimate potential risk of screening, it is important to note that overdiagnosis is an event that cannot be directly observed.

Accordingly, precise measurement of overdiagnosis is a challenge that requires not only understanding the effects of screening but also knowledge of alternative causes of death among women prior to development of breast cancer symptoms.³ There is no consensus on the appropriate methods for estimating overdiagnosis in breast cancer. A recent systematic review and meta-analysis of the medical literature on the harms of mammography screening that accompanied the 2016 USPSTF recommendations found that methodologies used in overdiagnosis studies are highly variable, with approaches adjusting for lead time falling in the lower range of estimates.⁴ Regardless of the true magnitude, both the USPSTF and ACS now acknowledge overdiagnosis from mammography screening and the eventual downstream diagnostic and treatment cascades that follow the detection of indolent cancers as potential harms that should be communicated to patients during shared decision-making.^1,2

While some have previously pointed to the breast imaging community as major contributors to the problem of overdiagnosis,⁵ detection of a malignancy at screening would have limited impact on a patient’s health without subsequent intervention and treatment, or overtreatment. Nevertheless, abnormal screening does launch a series of events as part of an integrated care pathway, where multiple disciplines contribute to diagnosis and treatment planning. After identification of abnormalities at screening and image-guided biopsy, pathologists assist in diagnosing breast malignancy. After the diagnosis of malignancy is made, including ductal carcinoma in situ (DCIS), treatment decisions are determined by a group of subspecialists, including surgeons, oncologists, and radiation oncologists.

As first line physicians in a cascade of medical care that is well intentioned, many breast imagers aim to balance the known benefits with the potential harms when making a decision to recall patients from screening. The most effective approach by which breast imagers can mitigate overdiagnosis is, perhaps, the most exciting aspect of this controversial issue. Eliminating screening mammography is not a realistic or ethical option since it would lead to later stage breast cancer diagnoses and increased mortality, even in this era of improved therapies.⁶

As members of multidisciplinary breast care teams with imaging expertise, it is imperative that radiologists engage in this issue by examining how current or emerging advanced breast imaging technologies can lessen the potential harms of overdiagnosis. In this article, we highlight recent advances and areas that warrant further investigation with the hopes that breast imagers will take an active and leading role in a collaborative effort to decrease breast cancer overdiagnosis and overtreatment.

Adjusting Imaging Frequency and Thresholds

Some experts have argued that revised imaging interpretation strategies can be used to curb diagnosis by raising the threshold for defining disease.^7,8 Current standards of practice as defined by the American College of Radiology (ACR) Breast Imaging-Reporting and Data System (BI-RADS) guidelines, any finding with >2% likelihood of malignancy require tissue diagnosis to exclude breast cancer.⁹ As a result, microcalcifications that may represent low-grade DCIS are biopsied rather than followed with serial non-invasive imaging. Currently, such findings are designated as BI-RADS category 4, which indicates a 2% to 95% likelihood of malignancy.⁹ This extremely broad range of suspicion and resulting low threshold for intervention, particularly in cases of calcifications, may contribute to overdiagnosis. This raises the question of whether breast imagers could safely follow low-risk lesions (i.e. BI-RADS category 4A) with serial imaging rather than proceeding to image-guided biopsy. Such adjustments could include different biopsy thresholds based on individualized risk, aided by imaging features. For example, biopsy thresholds for equivocal calcifications may be lower for women with a high lifetime risk of developing breast cancer than for women at average lifetime risk of breast cancer.

Others have suggested that less frequent screening could mitigate potential harms, including overdiagnosis, at the population level. Both the 2015 ACS and 2016 USPSTF recommendations suggest that starting routine screening at an older age (45 or 50 years rather than 40 years) and undergoing biennial rather than annual screening may lead to reductions in screening harms, including overdiagnosis, based on projections from simulation models.^1,2 However, the actual reduction in overdiagnosis from less frequent screening has not been rigorously assessed in prospective studies. While some indolent cancers may go undetected with less screening, other more aggressive cancers may be detected at later stages and lead to greater morbidity and mortality. Overdiagnosis likely increases with age, attributable primarily to increasing competing mortality; 14–36% of screen-detected cancers at age 80 have been projected to represent overdiagnosis.¹⁰

While worthy pursuits, it is important to note that both adjusting the threshold for biopsy based on imaging features and adjusting the frequency of screening mammography do not directly address the issue of overdiagnosis. The major limitations with these two approaches remains the reliance on standard imaging features for identifying breast cancer, as well as standard of care dictating a rigid treatment cascade once malignancy is confirmed after image-guided breast biopsy.

New Image-Based Screening Technologies

There is a great deal of excitement around digital breast tomosynthesis (DBT) and its ability to decrease screening-related harms, especially false positive findings. A recent review suggests that adding DBT to digital mammography screening can decrease the frequency of false positive results by 15–30%.¹¹ In addition, DBT may improve cancer detection. Friedewald et al found that adding DBT to standard digital mammography screening increased the overall cancer detection rate by 29% (from 4.2 to 5.4 per 1,000 screens). Moreover, there was a 41% increase in invasive cancer detection and no significant change in DCIS detection.¹² While detection of a greater number of early invasive cancers without increasing the number of pre-invasive malignancies may very well lead to greater lives saved, ongoing clinical trials will require many additional years of follow up to definitively establish this long term outcome benefit. Thus, it remains uncertain what DBT’s effect will be on overdiagnosis.

Further studies are needed to determine if the additional cancers detected by DBT over routine mammography are aggressive, more lethal cancers or less aggressive, indolent cancers based on tumor biology.^2,13–16 Initial data from cancer rates in a population undergoing repeated DBT screening suggests a decrease in interval cancers.¹⁷ If this trend continues, it may indicate that DBT screening is detecting more clinically significant cancers. Currently, more longitudinal and multi-institutional data is needed to substantiate these results. Even if DBT is identifying clinically significant cancers, it is still unlikely to mitigate overdiagnosis in and of itself since DBT is based on digital mammography technology, and is limited to identifying morphologic features rather than providing more biological or physiological insights regarding tumors.

Supplemental screening ultrasound has also been shown to increase cancer detection among women with dense breasts and other additional risk factors.¹⁸ These cancers tend to be small, invasive, and node-negative. However, similar to DBT, screening ultrasound evaluation is heavily reliant on morphology of masses without providing additional biological or functional information. The low positive predictive value of ultrasound screen-detected masses (less than 10%) continues to be a concern, as supplemental screening leads to a relatively high number of benign biopsies.¹⁹ Recently, abbreviated breast MRI, with an acquisition time of 3 minutes and expert radiologist maximum-intensity-projection image interpretation time of 3 seconds, has been cited as a potential screening tool for women at average risk with a negative predictive value of 99.8%.²⁰ For all of these technologies, however, the standard of care remains surgical treatment with or without adjuvant therapies for all breast malignancies. Thus, as long as all small cancers detected by screening continue to undergo aggressive treatments aimed at curing the patient, the potential contribution of these newer technologies for directly mitigating overdiagnosis, based on currently available information, is low.

The Potential of Radiomics

An emerging area that may help breast imagers address overdiagnosis and overtreatment is the field of radiomics. Radiomics refers to the high throughput extraction and analysis of quantitative features from medical images in order to build predictive and prognostic models relating image features to phenotypes or genotypes.²¹ These quantitative imaging techniques have the potential to facilitate personalized screening by allowing construction of individualized risk assessment tools incorporating a woman’s phenotype, such as breast parenchymal complexity, into mammography screening decision-making aids. Furthermore, radiomics may be incorporated into personalized treatment algorithms by fully characterizing a tumor according to its functional and morphological properties and linking those quantitative characteristics to discrete clinical outcomes. Thus, radiomic feature sets have the potential to serve as both predictive and prognostic biomarkers of disease.

For breast imaging to directly impact the potential harm of overdiagnosis, the field will have to develop better treatment decision-making tools with the aid of advanced breast imaging technologies. This effort will likely require a combination of advanced imaging and molecular markers to help physicians differentiate life-threatening tumors that need treatment from those less aggressive cancers that may be followed or treated less aggressively.²² One area of active investigation is the combined use of multiparametric MRI and tumor molecular biomarkers to improve management of low-grade DCIS with the goal of reducing overtreatment of low-risk lesions.

Imaging Biomarkers for Malignancy Progression

DCIS lesions, which comprise the earliest form of breast cancer, currently account for one in every five new cancer diagnoses in the United States.²² It is estimated that by 2020, more than one million U.S. women will be living with a diagnosis of DCIS.²³ The pathologic diagnosis of DCIS remains an area of controversy, with interpretive challenges and substantial clinical implications upon diagnosis.^24–29 For instance, in a recent study of breast pathologists across 8 states, there were relatively low levels of concordance in differentiating between atypia and DCIS compared to concordance in identifying invasive cancer during breast slide interpretation, suggesting that some women may be overdiagnosed as having malignancy in the setting of atypia.³⁰ Once the diagnosis of DCIS is made, however, the current standard of care for treatment remains surgical excision and radiation (if breast conservation is chosen), possibly including endocrine therapy for hormone receptor positive disease.³¹

One reason DCIS overdiagnosis has been difficult to address to date is that the natural history of untreated DCIS is poorly understood. Page et al described the natural course of low nuclear grade DCIS lesions diagnosed by biopsy, demonstrating that low grade DCIS lesions can in fact progress to invasive breast cancer and metastatic disease when left untreated.³² Subsequent studies confirmed that DCIS can progress to invasive cancer regardless of pathologic subtype, but also noted that the time to progression is longest for low nuclear grade lesions and the frequency of progression was greatest for high nuclear grade lesions.^33,34

Recent research has demonstrated that the progression of DCIS to invasive breast cancer relies on multiple influences, including both tumor genetic heterogeneity and the presence of a vascular-rich and chronically inflamed tumor microenvironment.^35–37 Unfortunately, no pathology feature has yet been shown to provide strong enough prognostic value to be relied upon for treatment decision-making alone or even when combined with clinical features (e.g. Van Nuys Prognostic Index and Memorial Sloan Kettering Cancer Center DCIS Nomogram).^38–41

A recent tissue-based multigene assay (Oncotype DX DCIS score) has been developed and may aid in determining risk of DCIS recurrence,^42,43 but has only been validated in a cohort that contained a high proportion of lesions that were small and/or estrogen positive. Furthermore, because DCIS is closely related to noncancerous intraductal lesions, such as atypical ductal hyperplasia, pathological characterization is challenging and prone to inter-observer variability and poor reproducibility.⁴⁴ For example, it has been estimated that 50% of DCIS lesions exhibit more than one nuclear grade⁴⁵ while another 15–20% will upgrade to invasive disease on surgical excision.⁴⁶

Compared to pathologic assays, the use of advanced imaging to further assess DCIS risk provides a key advantage: high spatiotemporal resolution MRI can assess an entire lesion and is thus less prone to sampling error. MRI can assess lesions in vivo and non-invasively, which allows for physiologic features to be assessed in the lesions’ natural, non-formalin-fixed state. Of the available advanced imaging tools, MRI has demonstrated the greatest promise to reflect the biologic features of breast pathology, such as vascularity and capillary permeability, cell membrane integrity, and cellularity.

Through the use of dynamic contrast enhancement (DCE) techniques, radiologists can measure the differential enhancement of DCIS relative to normal breast tissue, theoretically assessing the rates by which the abnormal vessels that support these pre-invasive malignancies’ growth leak nutrients into a tumor bed. This can be achieved semi-quantitatively by measuring basic kinetic enhancement curves or quantitatively for high temporal resolution acquisitions (e.g. K^trans).⁴⁷

The use of diffusion weighted imaging (DWI) in breast imaging can provide complementary biological information by indirectly measuring cellular density and membrane integrity of tumors.⁴⁸ DWI utilizes motion-sensitizing gradients during MR image acquisition to probe local water diffusion characteristics, and the level by which water diffusion is restricted by cell membranes or other obstacles can be quantified by measuring apparent diffusion coefficients (ADCs).⁴⁹ Many malignancies demonstrate reduced water diffusivity when compared to normal tissue or benign pathologies, which has been hypothesized to be reflective of high cellular density.

Initial studies using multiparametric MRI to characterize DCIS biology have been encouraging. A large prospective imaging trial for the detection of DCIS performed by Kuhl and colleagues established that MRI is the most sensitive imaging tool for the detection of DCIS. Interestingly, the authors also noticed that the absence of suspicious enhancement on MRI was a potential biomarker of less aggressive DCIS phenotypes, since a greater fraction of high nuclear grade lesions were identified on MRI when compared to mammography.⁵⁰ Similarly, Strobel et al found that using MRI to evaluate pure calcifications (no associated mass) prior to biopsy, MRI was able to identify all non-low grade DCIS lesions.⁵¹

Multiple authors have found that specific MRI features can reflect DCIS biology. Lesions that demonstrate small, focal enhancement on DCE with high signal (i.e. contrast-to-noise ratio) without corresponding low ADC values on DWI are likely to reflect lower grade DCIS.⁵² Utilizing more quantitative MRI markers, Jansen and colleagues found that solid forms of DCIS exhibit unique early contrast features when compared to cribriform subtypes,⁵³ while Li et al subsequently found that K^trans and ADC can discriminate between DCIS and invasive cancer and correlate with markers of proliferation (Ki-67) and angiogenesis (CD105).⁵⁴ A recent study using higher spatial resolution imaging at 3 tesla further suggested that low risk DCIS lesions can be identified on the basis of DWI biomarkers.⁵⁵ Kim et al further found that higher amounts of parenchymal enhancement surrounding DCIS lesions correlate with increased recurrence risk,⁵⁶ suggesting that MRI features may serve as biomarkers of the tumor microenvironment, which is increasingly recognized as being an important factor in breast cancer tumorigenesis.

Given these promising preliminary findings, further work assessing the ability of advanced imaging markers to predict outcomes and guide individualized therapies of DCIS is currently underway. An ongoing multi-center trial (Eastern Cooperative Oncology Group-American College of Radiology Imaging Network 4112 trial) may provide valuable insight into the practical application of MRI to address overtreatment of DCIS. This study will examine whether MRI, in conjunction with Oncotype DX-DCIS scores, can safely allow patients with DCIS lesions identified to be at low risk of recurrence to avoid radiation therapy.

Shifting Treatment Paradigms Based on Imaging Features

In addition to concerns about overdiagnosis of DCIS, there are small invasive cancers that might never become clinically significant, contributing to overdiagnosis and overtreatment. Addressing this issue requires shifting the current paradigm of disease specific treatment for breast cancer to one of tumor specific treatment, personalized to each woman. Radiologists can take an active role in developing shared diagnostic decision-making tools that will characterize the biologic aggressiveness of newly identified cancers, and provide both predictive (response to treatment) and prognostic (risk of recurrence/disease specific mortality) information. Encouraging results have been obtained in preliminary studies utilizing multiparametric tumor phenotyping on breast MRI (incorporating computer-extracted kinetic, morphologic, and spatial heterogeneity features) to predict clinical outcome from breast cancer based on Oncotype Dx recurrence score.^57–59 With these promising preliminary data, investigation of new models that directly relate MRI imaging phenotype to biological-based outcomes are needed to determine whether imaging can provide clinical information that improves upon that provided by tumor genomics alone.

Conclusion

In summary, the field of breast imaging can, in many ways, lead the charge towards decreasing harms associated with screening. Newer technologies such as DBT are facilitating improved cancer detection and lower false-positives. Many radiologists are also engaging in the dialogue of adjusting screening frequencies and biopsy thresholds. With regards to mitigating the harms of overdiagnosis and overtreatment, further development of novel radiomic approaches to identify and quantify multiparametric advanced imaging features strongly associated with tumor physiology and biology holds the greatest potential for lasting impact. Current and planned trials investigating the use of multiparametric breast MRI in combination with multi-gene assays are one such promising approach. Through these efforts, breast imagers can have a sustained impact by tipping the balance of screening towards greater benefits and fewer harms.