Mammography topics

Future of Breast Cancer Screening with Digital Breast Tomosynthesis

Breast screening programs today generally offer a mammographic examination every two years to women in a specified age range. Ongoing research is exploring the added value of personalized risk-based screening (i.e., screening strategies tailored to a woman’s individual risk of breast cancer). A vital prerequisite for personalized risk-based screening is a comprehensive breast cancer risk prediction model. The next step is to develop screening regimens for women at varying levels of risk, which may be addressed in several ways: changing starting and stopping ages, changing screening intervals, and offering further imaging techniques in addition to digital mammography. Using modeling techniques, the harm-benefit ratios and cost-effectiveness can then be estimated to find the optimal screening strategies. In parallel, it is essential to understand the acceptability of personalized approaches, especially for women at low risk, and the challenges in communication and implementation.

Breast density is an important factor in this research area because it can contribute to risk prediction, but it will also play a role in finding the optimal screening strategy. Automated measures of breast density on digital mammography are being updated rapidly, which will facilitate getting this data for all women in the target population. However, the fact that DBT may replace digital mammography in the near future raises the following questions: How does DBT perform across breast density categories? What is the relationship between automated breast density measures on DBT and breast cancer risk? What is the role of additional imaging modalities when DBT is the screening test?

According to some of the important principles of screening established by WHO in 1968, a screening test should be fast, safe, efficient, and acceptable for the target population. Screening with 2D mammography fulfills many of these prerequisites, but how will digital breast tomosynthesis, DBT, fit in? The organization and workflow in mammography screening is long-standing and well-established in many countries, and changes may have a great impact on these issues.

This talk will focus on the facts about digital breast tomosynthesis regarding screening performance, effectiveness, and possible measures to reduce reading time. When planning the Malmö Breast Tomosynthesis Screening Trial we took the principles of screening into consideration and used 1-view DBT and reduced compression. Furthermore, artificial intelligence in combination with the radiologist opens up possibilities for further future improvements.

The superiority of digital breast tomosynthesis (DBT) over digital mammography (DM) for cancer detection in screening has been proven in several population-based prospective screening trials. However, most of these results have been achieved with considerably longer reading times than with DM. In some cases, higher radiation doses have also been used. Is this acceptable? What are the most important features and characteristics on our wish list for the ideal breast cancer screening program with DBT in the future?

In this symposium, the latest results of clinical trials will be presented, analyzed in terms of effectiveness when considering population screening, and the remaining knowledge gaps identified. We will discuss whether we should just copy the standard approach with DM for screening with DBT or whether we should go in new directions. Finally, we will review the challenges faced in the widespread implementation of DBT for population screening.

Aim: The study compares breast cancer screening performance using Full Field Digital Mammography images (2D) and three-dimensional Digital Breast Tomosynthesis (DBT) images, in MLO projection only (3DMLO) and as a complete DBT exam in MLO and CC projections (3D).

Introduction: Radiation dose at 3D screening can be reduced by omitting separate 2D acquisition and abandoning CC projection in DBT, if 3DMLO could demonstrate comparable diagnostic performance.

Material and methods: Women aged 40 and older who had screening exams using a Siemens DBT unit in the period from Aug. 1, 2016, to Sep. 30, 2017, were included. 21 biopsy-proven cancer patients with 24 cancers were identified. Based on age and Breast Imaging Reporting and Data System (BI-RADS) density assessment (DS), they were matched with 24 biopsy-proven benign patients and 25 patients who were negative or benign on screening. Age, race, and DS were collected. Lesion location, type, and size, axillary lymph nodes status, and pathological diagnosis were also collected for biopsy patients.

Screening exams of those 70 patients, consisting of 2D and 3D mammography images, were included in the image set. Three fellowship-trained breast radiologists and two breast imaging fellows were the readers.

We will compare the diagnostic performance of 3DMLO, 3D, and 2D readings, separately for each reader and for all readers, by looking at recall rate, cancer detection rate, positive predictive value, and area under curve (AUC).

So far, CAD has not routinely been applicable for screen reading in organized screening programs, mostly due to excessive false positive rates (of 1 to 1/3 hit per mammogram).

Tomosynthesis, which is considered the most promising method for breast cancer screening, is associated with a 10-20-fold increase in the number of images. Insufficient data exists concerning the capability of maintaining the reader’s concentration with this work load, so the need for CAD support increases. Also, new artificial intelligence (AI) algorithms promise improved sensitivity and specificity.

We present a software solution that allows integration of an AI-CAD system with the screen-reading software. Mammograms are automatically assigned a degree of suspicion (1–10), based on a large number of screening mammograms. Filtering cases with high suspicion (9 or 10), either a selective third reading may be considered, or a second reader might decide to spend more time on these cases or read them first. Due to the high sensitivity for microcalcifications, less reading time can be used for cases rated 1–4.

In a different study using AI-CAD for tomosynthesis an excellent sensitivity of 90–92%, combined with a very low false positive rate of 9–15%, was achieved in a demanding case series of 400 proven cases. These results indicate important progress compared to prior CAD systems on 2D mammograms.

A synthetic mammography (SM) is a two-dimensional image, reconstructed from digital breast tomosynthesis (DBT) data. Simulation studies (work of Michielsen et al.) using insertion of simulated microcalcifications and mass lesions in raw data of tomosynthesis showed an impact of reconstruction algorithm on the lesion detection fraction. The algorithm of SM impacts on the presentation of line structures and high-density structures like microcalcifications resulted in increased conspicuity of lesions in the SM. In the Oslo trial (Per Skaane et al.) SM was included in the screening protocol, and in this study it was concluded that improvement of the reconstruction algorithm had an impact on the lesion detection. Further work on the reconstruction algorithm of SM can improve the presentation of suspicious lesions with careful attention to the creation of false positive lesions: How far is this CAD-related enhancement allowed in clinical practice? Physical-technical evaluation of the quality of SM is ongoing and will be complicated by the complex structure of the applied processing algorithms.

Breast screening programs today generally offer a mammographic examination every two years to women in a specified age range. Ongoing research is exploring the added value of personalized risk-based screening (i.e., screening strategies tailored to a woman’s individual risk of breast cancer). A vital prerequisite for personalized risk-based screening is a comprehensive breast cancer risk prediction model. The next step is to develop screening regimens for women at varying levels of risk, which may be addressed in several ways: changing starting and stopping ages, changing screening intervals, and offering further imaging techniques in addition to digital mammography. Using modeling techniques, the harm-benefit ratios and cost-effectiveness can then be estimated to find the optimal screening strategies. In parallel, it is essential to understand the acceptability of personalized approaches, especially for women at low risk, and the challenges in communication and implementation.

Breast density is an important factor in this research area because it can contribute to risk prediction, but it will also play a role in finding the optimal screening strategy. Automated measures of breast density on digital mammography are being updated rapidly, which will facilitate getting this data for all women in the target population. However, the fact that DBT may replace digital mammography in the near future raises the following questions: How does DBT perform across breast density categories? What is the relationship between automated breast density measures on DBT and breast cancer risk? What is the role of additional imaging modalities when DBT is the screening test?