Deep Learning of Digital Breast Tomosynthesis and Mammography

PI: Xiaoqin Wang, MD; Co-I: Nathan Jacobs, PhD

Xiaoqin (Jennifer) Wang, MD, Assistant Profressor in Radiology is the Principal Investigator of this multidisciplinary project, which is funded by the American Cancer Society IRG grant. The digital breast tomosyntesis (DBT) or 3D mammography has emgerged as an important screening and diagnostic tool for breast cancer detection. DBT, when combined with a 2 D mammogram, has been proven to increase breast cancer detection rate while reducing recall rate.  However, interpreting a DBT is more time consuming. Therefore, it is very important to develop an automated interpretation tool which can aid radiologists in reading DBT and mammogram more accurate and efficiently. 

Recent advancement in deep learning has shown great promise in computer vision. However, the performance of deep learning models depends on the training data. The comprehensive breast imaging divsion of the UK Radiology Department and Markey Cancer Center is one of the earliest adopters of the DBT and has accumulated a large amount of DBT and mammogram data with ground truth. In recognition of the need and resource, Dr. Wang and her team propose to develop deep learning tools to classify the breast cancer on both 2D and 3D mammograms. Co-I, Nathan Jacobs, PhD, is an Associate Professor of Computer Science at the Unversity of Kentucky and an expert on the computer vision. Their work on this project has been presented at multiple national and international conferences and published in peer-review journals. 

Selected Publication:

  • Wang X., Liang G., Zhang Y., Blanton H., Bessinger Z., Jacobs N. Inconsistent performance of deep learning models on  mammogram classification, Journals of American College of Radiology, 2020, Feb 14. S1546-1440(2030028-4. PMID: 32068005 
  • Zhang Y, Wang X, Blanton H, Liang G., Xing X, and Jacobs N. 2D Convolutional Neural Networks for 3D Digital Breast Tomosynthesis Classification. Published in: 2019 IEEE International Conference on Bioinformatics (BIBM), pp. 1013-1017. DOI: 10.1109/BIBM47256.2019.8983097
  • Liang G, Wang X, Zhang Y, Xing X, Blanton H. Salem T, and Jacobs N. Joint 2D-3D Breast Cancer Classification. Published in: 2019 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), pp. 692-696.

Clincal Evaluation of a Novel 3D Imaging Tool for Intraoperative Specimen Assessment During Breast Cancer Lumpectomy Resection

PI: Xiaoqin Wang, MD; Co-I's Patrick McGrath, MD; Emily Marcinkowski, MD; Erin Burke, MD; and Heidi Weiss, PhD

The goal of this project is to evaluate the clinical performance of a new intraoperative 3D imaging system in assessment of specimen margins and tumor excision. Each year >200,000 breast cancer patients undergo breast conserving surgery in the United States. 2D radiography is the current method used by surgeons and radiologists to assess the margin status of the specimen in the operating room, but the intrinsic limitations of a 2D analysis of a 3D specimen inevitbably reduces accuracy. As a result, approximately one in four patients who pursue this treatment under the current specimen assesssment system require re-excision, which causes additional health care costs, a worse cosmetic outcome, and significant stress on the patient. A prototype of a 3D specimen CT device was developed by a spinout company from the University of Chicago. This new intraoperative 3D imaging system allows for a cross-sectional and direct visualization of the entire specimen via 3D reconstruction. Threfore, it has great potential to improve the specimen margin assessment and reduce the re-excision rate. This proposed study will be the first to investigate the clinical performance of this novel tool in the OR, which is an important translational step toward clinical application. In the future, building on the results of this project, we will advance the integration and implementation of this technology to surgical suites to enable a full-scale clinical trial with this intraoperative imaging system. 

This project is funded by the MCC CCSG grant. Xiaoqin (Jennifer) Wang, MD, Assistant Professor in Radiology, is the Principal Investigator of this pilot clinical study. In addition, the research team includes a clinical resarch coordinator (Jennifer Isaacs, MS), multiple breast surgeons (Patrick McGrath, MD; Emily Marcinkowski, MD; Erin Burke, MD) at University of Kentucky as well as Heidi Weiss, PhD, Professor and Director of Biostatistics and Bioinformatics Shared Resource Facility at Markey Cancer Center Biostatistics, University of Kentucky and Markey Cancer Center.