Breast Cancer Risk Assessment


To create reliable standard for breast cancer risk based on multivariable information




Breast Imaging
Define-AI ID 19060003


Syam Reddy
Syam Reddy

Panel Chair

Beth Burnside

Panel Reviewers

Breast Imaging


Creative Commons 4.0 
Status Public Comment
RadElement Set RDES120 

Clinical Implementation

Value Proposition

Risk stratification for patients allows radiologists to provide focused additional screening and prevention efforts where necessary to help reduce risk and less surveillance in those patients with lower risk to allow for better cost effective healthcare. Currently, there are numerous variables that need to be assessed for modeling a risk assessment tool. Current models have varying strengths that rely on patient’s own personal history or some that rely more on family history of breast or ovarian cancer. To date, no clinical risk assessment models incorporate breast density, known 4 to 6 fold increased risk factor for developing breast cancer. AI can help incorporate breast density and currently available risk assessment tools (inclusive of patient’s personal and family history) to improve overall risk assessment in individualized patients. Additionally, deep learning applied to breast cancer patient data may lead to new risk factors to make the model more robust. In light recent breast density legislation, patients are now more so than ever before, wanting to better understand their individualized breast cancer risk and optimal screening strategy.


A 50 year old patient decides to have her first baseline mammogram and presents with extremely dense breasts. She has a family history of mother and sister with diagnosed with breast cancer at 40 and 50 years, respectively. AI tools would incorporate this information, as well as other variables from EMR, including breast density from the mammogram or radiology report, and patient in-take questionnaire. The information would run through varying pre-existing risk assessment calculators as well as use deep learning to create a more robust and accurate breast cancer risk assessment tool. Information could then be used for tailoring individualized surveillance plans and risk reduction strategies where appropriate.

Workflow Description

Standard mammography views to assess quantitative breast density, historical data from EMR (ie: age, menopausal status) and answers to questions from the patient smart tablet (ie: smoker, parity etc) would be sent to AI engine which would use deep learning and various risk calculators. DICOM header output information from AI engine to PACS would be given for documentation and incorporated into the radiology report interpretation for added value. Based on risk assessment, recommendation for additional screening methods or risk reduction strategies would be performed. Additionally, tailoring personalized screening algorithms including interval of recommended screening, and/or best supplemental screening methods will be studied.

Considerations for Dataset Development



Screening mammography


Breast cancer screening and early detection

Breast Anatomy

{No prior surgery, no implants}


40 years and older


{Standard craniocaudal, standard mediolateral}

Prior radiation exposure



Historical data from EMR; parity, estrogen use, menopausal status, BMI/obesity, age


{family cancer history, menarche, personal cancer history, etc.}

Risk assessment calculators

{Gail, Tyrer Cuzick, BRCA pro, etc.}

Technical Specifications




Screening mammography


Standard Craniocaudal and mediolateral

Data Type




Body Region


Anatomic Focus


Primary Outputs

Risk Assessment for Breast Cancer

RadElement ID



Personal lifetime risk assessment for breast cancer

Data Type


Value Set



Probability of developing breast cancer

Future Development Ideas

  • Include risk assessment score for each common type of breast cancer.
  • Analyze cases over time to better learn the correlations with breast cancer such as density to cancer occurrence, types of cancer with various risk factors, and demographics to cancer occurrence among others.

Related Datasets