Screening mammography has limited sensitivity and specificity. Digital Breast Tomosynthesis (DBT) is an emerging modality that has been shown to significantly improve the detection and characterization of soft- tissue lesions. However, initial studies have shown that subtle microcalcification (MC) clusters, which are often the only sign of early breast cancer, can be difficult to visualize in DBT. Some have suggested that DBT be used in parallel with FFDM in screening, (i.e., adding one- or two-view DBT to the two-view FFDMs so that FFDM could be used for MC detection while DBT could be used for mass detection). This approach would increase imaging costs, reading time, and patient dose, which are all major concerns with regards to introducing DBT into clinical practice. The main goal of the proposed Partnership between the University of Michigan Computer-Aided Diagnosis Research Laboratory (UM) and GE Global Research (GE) is to develop an integrated practical approach to resolving the MC visualization and detection problems in DBT without increasing patient dose, thereby facilitating the eventual replacement of FFDM by DBT. To achieve this goal, we propose two Specific Aims: (SA1) to develop specially designed MC enhancing methods to improve human and machine visualization of MCs in DBT and develop a computer-aided detection (CAD) system to highlight significant MC clusters, and (SA2) to implement the developed MC-enhancing and CAD reading tools in a DBT workstation and conduct observer performance studies to compare MC detection in DBT with that in FFDM. The following tasks will be conducted to accomplish the specific aims: (1) perform phantom studies to determine the best set of image acquisition parameters for data collection, (2) collect a database of human subject DBTs for development of algorithms and observer study, (3) develop lesion-specific reconstruction and MC enhancing methods to improve the visibility of MCs in DBT for radiologist's reading and computerized detection, (4) develop computer-vision methods to detect MC candidates, (5) develop MC analysis method to reduce false positives (FPs) and insignificant CAD marks, (6) design two-view analysis to further reduce FPs, (7) study dependence of MC detection on reconstruction methods and tomosynthesis acquisition parameters, and (8) design a DBT workstation implemented with the MC-enhancing and CAD- assisted tools to highlight significant MCs for radiologist's reading. We hypothesize that the specially designed DBT display system can assist radiologists in detection of MCs in DBT with accuracy at least comparable to that in FFDM. To test this hypothesis, we will (9) conduct observer ROC studies to compare the detection accuracy of MCs under three conditions: (a) two-view DBT without CAD vs. two-view FFDM without CAD, (b) two-view DBT with CAD vs. two-view FFDM with CAD, and (c) a special protocol of CC-view FFDM plus MLO-view DBT with CAD vs. two-view FFDM with CAD.

Public Health Relevance

DBT is a promising modality for improving breast cancer detection. It is widely regarded that DBT is superior to FFDM for detecting soft-tissue lesions. This Partnership between UM and GE aims at finding an integrated, effective solution to address the critical remaining issue of MC visualization and detection in DBT. The partners bring unique and complementary capabilities to the proposed program. GE has expertise in the design of DBT systems, image analysis, workstation implementation, and most importantly, limited-angle tomosynthesis and full-angle CT reconstruction in practical commercial imaging systems. UM has extensive experience in development of CAD methods and DBT imaging, medical physicists with expertise in the evaluation of x-ray systems, and strong clinical support from the Breast Imaging Division within one of the top academic radiology departments in the country. Together, UM and GE have the full complement of skills, experience, and resources required for success on this important public health project. If we are successful in achieving our aims, we will have a practical solution to the MC detection problems in DBT. We will produce an MC-enhanced and CAD-assisted reading protocol that will serve as a model for DBT system and workstation design. This will pave the way for the acceptance of DBT for clinical use, thereby improving the sensitivity and specificity of breast cancer screening which will benefit all women.

Agency
National Institute of Health (NIH)
Institute
National Cancer Institute (NCI)
Type
Research Project (R01)
Project #
5R01CA151443-03
Application #
8514397
Study Section
Special Emphasis Panel (ZRG1-SBIB-U (55))
Program Officer
Baker, Houston
Project Start
2011-09-02
Project End
2016-07-31
Budget Start
2013-08-01
Budget End
2014-07-31
Support Year
3
Fiscal Year
2013
Total Cost
$589,550
Indirect Cost
$139,554
Name
University of Michigan Ann Arbor
Department
Radiation-Diagnostic/Oncology
Type
Schools of Medicine
DUNS #
073133571
City
Ann Arbor
State
MI
Country
United States
Zip Code
48109
Zheng, Jiabei; Fessler, Jeffrey A; Chan, Heang-Ping (2018) Detector Blur and Correlated Noise Modeling for Digital Breast Tomosynthesis Reconstruction. IEEE Trans Med Imaging 37:116-127
Samala, Ravi K; Chan, Heang-Ping; Hadjiiski, Lubomir M et al. (2018) Evolutionary pruning of transfer learned deep convolutional neural network for breast cancer diagnosis in digital breast tomosynthesis. Phys Med Biol 63:095005
Lu, Yao; Chan, Heang-Ping; Wei, Jun et al. (2017) Improving image quality for digital breast tomosynthesis: an automated detection and diffusion-based method for metal artifact reduction. Phys Med Biol 62:7765-7783
Zheng, Jiabei; Fessler, Jeffrey A; Chan, Heang-Ping (2017) Segmented separable footprint projector for digital breast tomosynthesis and its application for subpixel reconstruction. Med Phys 44:986-1001
Samala, Ravi K; Chan, Heang-Ping; Hadjiiski, Lubomir M et al. (2017) Multi-task transfer learning deep convolutional neural network: application to computer-aided diagnosis of breast cancer on mammograms. Phys Med Biol 62:8894-8908
Chan, Heang-Ping; Helvie, Mark A; Hadjiiski, Lubomir et al. (2017) Characterization of Breast Masses in Digital Breast Tomosynthesis and Digital Mammograms: An Observer Performance Study. Acad Radiol 24:1372-1379
Samala, Ravi K; Chan, Heang-Ping; Hadjiiski, Lubomir et al. (2016) Mass detection in digital breast tomosynthesis: Deep convolutional neural network with transfer learning from mammography. Med Phys 43:6654
Chan, Heang-Ping (2016) Comment on ""Large area CMOS active pixel sensor x-ray imager for digital breast tomosynthesis: Analysis, modeling, and characterization"" [Med. Phys. 42, 6294-6308 (2015)]. Med Phys 43:1578
Samala, Ravi K; Chan, Heang-Ping; Hadjiiski, Lubomir M et al. (2016) Analysis of computer-aided detection techniques and signal characteristics for clustered microcalcifications on digital mammography and digital breast tomosynthesis. Phys Med Biol 61:7092-7112
Lu, Yao; Chan, Heang-Ping; Wei, Jun et al. (2015) Multiscale bilateral filtering for improving image quality in digital breast tomosynthesis. Med Phys 42:182-95

Showing the most recent 10 out of 21 publications