Dual-energy three-compartment breast imaging for compositional biomarkers to improve detection of malignant lesions

We use compositional mammographic imaging to improve breast cancer detection. With our 3CB technique, we were able to derive unique compositional signatures for different breast lesion pathologies.

Like Comment
Read the paper

Breast cancer screening using mammography has drastically improved early detection and reduced the cancer mortality (1, 2). However, mammography is only able to describe the radiopacity (brightness) and morphometry (shape characteristics) of suspicious lesions. Radiologists subjectively use these visual properties when deciding if a suspicious lesion is likely to be cancerous, and it is common for cancerous lesions to be masked by overlaying texturally complex or dense normal breast tissue. Masking reduces mammography accuracy and leads to not only missed cancer findings but also uncertain diagnoses of detected lesions resulting in a tendency to biopsy lesions at perceived probabilities of malignancy as little as 2%.

Providing Computer-Aided Diagnostic (CAD) to radiologists reading 2D and 3D mammograms objectifies the interpretation of radiopacity and morphometry, and improves the diagnostic accuracy but still uses and was developed to use powerful computer vision algorithms to aid radiologist and evaluate 2D and 3D images beyond human visual abilities. However, CAD doesn’t really have novel information to bring to the reading, it just standardizes and objectifies what the radiologist already knows. 

We sought to add the composition of the breast and lesion, in terms of its lipid, water and protein content, to the interpretation of 2D mammograms. They call this three-compartment breast (3CB) imaging (3, 4). 3CB adds this information by acquiring a high-energy mammogram (higher X-ray tube voltage for more penetration into the breast tissue) and an accurate measure of breast thickness, along with the standard mammogram, to solve for composition. It can be simply thought of as the algebra of solving for three unknowns (lipid, water, and protein thicknesses) using three equation created using the additional information. Previous work using the 3CB imaging technique was able to determine the compositional characteristics of malignancies versus other common biopsies lesions such as fibroadenomas and benign breast disease. (5, 6). In this work, we demonstrate improvements in the accuracy of detecting breast cancer when adding 3CB imaging to CAD. Artificial intelligence (AI) was used to create a 3CB/CAD diagnostic model that was superior to CAD alone. To follow up this work, we will repeat their studies in 3D mammography and ask a panel of radiologist to score the diagnostic improvement of 3CB/CAD versus CAD alone versus radiologist alone.

Standard mammography view and compositional imaging with radiologist delineations of breast leisions
3CB provides standard mammographic imaging as well as compositional heatmaps of the breast. The first column contains the standard mammogram presentation, the second, third, and fourth columns contain the corresponding 3CB lipid water and protein thickness map. Each row consists of a lesions with a different pathology. Colorbars adjacent to each 3CB map indicate thickness in centimetres where red indicates areas of high thickness and thickness decreases towards the color violet. Thickness ranges are normalized across each column. Yellow lines are radiologist delineations of where biopsies were taken from which lesion pathology was determined.

We envision that the radiologist and CAD would make an initial read on the standard views for  2D or 3D mammograms, inspect the 3CB visual representations of lipid/water/fat, select a feature or lesion to be evaluated by the AI program, and lastly review quantitative probabilities of malignancy generated by the AI program using all the available morphometry, CAD modeling, and compositional information. All digital mammography systems are equipped with the ability to take images at higher energies and costly hardware upgrade are not necessarily needed. Our on-going work involves integrating 3CB with the existing acquisition protocols for contrast-enhanced mammography systems (CEM), and integration with digital breast tomosynthesis (DBT, 3D mammography). CEM systems intrinsically acquire the needed dual-energy mammograms.


  1. Helvie MA, Chang JT, Hendrick RE, Banerjee M. Reduction in late-stage breast cancer incidence in the mammography era: Implications for overdiagnosis of invasive cancer. Cancer. 2014;120(17):2649-56. Epub 2014/05/21. doi: 10.1002/cncr.28784. PubMed PMID: 24840597.
  2. Myers ER, Moorman P, Gierisch JM, Havrilesky LJ, Grimm LJ, Ghate S, Davidson B, Mongtomery RC, Crowley MJ, McCrory DC, Kendrick A, Sanders GD. Benefits and Harms of Breast Cancer Screening: A Systematic Review. JAMA. 2015;314(15):1615-34. Epub 2015/10/27. doi: 10.1001/jama.2015.13183. PubMed PMID: 26501537.
  3. Laidevant AD, Malkov S, Flowers CI, Kerlikowske K, Shepherd JA. Compositional breast imaging using a dual‐energy mammography protocol. Medical physics. 2010;37(1):164-74.
  4. Avila J, Malkov S, Giger M, Drukker K, Shepherd JA, editors. Energy Dependence of Water and Lipid Calibration Materials for Three-Compartment Breast Imaging. International Workshop on Digital Mammography; 2016: Springer.
  5. Drukker K, Giger M, Joe B, Kerlikowske KK, Greenwood H, Drukteinis J, Niell B, Fan B, Malkov S, JA A, Kazemi L, Shepherd JA. Combined Benefit of Quantitative Three-Compartment Breast Image Analysis and Mammography Radiomics in the Classification of Breast Masses in a Clinical Data Set. Radiology. 2018;Online First.
  6. Drukker K, Duewer F, Giger ML, Malkov S, Flowers CI, Joe B, Kerlikowske K, Drukteinis JS, Li H, Shepherd JA. Mammographic quantitative image analysis and biologic image composition for breast lesion characterization and classification. Medical physics. 2014;41(3):031915.

Lambert Leong

Researcher, University of Hawaii

I am a medical and cancer researcher with a background in both biology and computer science.  My interest include utilizing high throughput data processing and artificial intelligence/machine learning within the biological and human health domain. My current projects and areas of focus include breast cancer imaging, dual energy X-ray absorptiometry, and 3D scanning.