Breast Density Quantification

Purpose

To automate the assessment of breast density on digital mammography and digital breast tomosynthesis by developing an AI tool that: (1) assigns a BI-RADS® breast composition category, (2) assigns a composite numeric breast density score, and (3) provides regional breast density numeric sub-scores.

Tag(s)

 

Panel

Breast Imaging

Define-AI ID

19060012

Originator

Sarah Eskreis-Winkler, Todd Hertzberg

Panel Chair

Elizabeth Burnside

Panel Reviewers

Breast Imaging Panel

License

Creative Commons 4.0 
Status Public Comment
RadElement Set RDES75 
                               

Clinical Implementation


Value Proposition

Recent federal legislation requires that breast density information be included in all mammography reports. Breast density is an independent risk factor for breast cancer and can decrease the sensitivity of mammography.  Reliable breast density assessment is needed to identify women who may benefit from additional breast cancer screening.  However, qualitative mammographic breast density assessment is subjective and has high inter-reader variability. Automated categorization of breast density on mammography and tomosynthesis could reduce this subjective variability.  Auto-populating breast density assessment in the radiology report could also generate a modest time and cost savings.

Narrative(s)

A 40 year old women presents for her first screening mammogram.  An automated (AI-based) assessment algorithm generates a numeric breast density score and a BI-RADS breast composition category.  This results in a more standardized method of determining which patients should be counseled about the benefits of additional screening exams (such as ultrasound or breast MRI). The numeric breast density score could be further adjusted to age- and/or weight-matched controls.

Workflow Description

Digital data from standard mammographic and tomosynthesis breast exams are sent to a server, which can be accessed by the AI algorithm.  The algorithm then analyzes the data and dynamically generates breast density information for the interpreting radiologist, as well as automatically populating appropriate information into the radiologist’s report.

 

Considerations for Dataset Development


 

Procedure

Screening mammography and screening tomosynthesis

Views

Bilateral CC and MLO mammography views; bilateral CC and MLO tomosynthesis views from the entire 3D dataset

Age

over 40 years

Breast Anatomy

No exclusion criteria

Technical Specifications


Inputs

 

DICOM Study

Procedure

Screening Mammography

Views

Bilateral CC and MLO mammography images

Data Type

DICOM

Modality

MAM

Body Region

Chest

Anatomic Focus

Breast

 

Secondary Inputs

 

DICOM Study

Procedure

Screening Tomosynthesis

Views

Bilateral CC and MLO tomosynthesis views from the entire 3D dataset

Data Type

DICOM

Modality

MAM

Body Region

Chest

Anatomic Focus

Breast

Pharmaceutical

N/A

Scenario

N/A


Primary Outputs

 

BI-RADS Breast Composition Category

RadElement ID

RDES75

Definition

Assign a BI-RADS breast composition category

Data Type

Categorical

Value Set

0 -Unknown

1a-The breasts are almost entirely fatty.

2b-There are scattered areas of fibroglandular density.

3c-The breasts are heterogeneously dense, which

may obscure small masses.

4d-The breasts are extremely dense, which lowers the

sensitivity of mammography.

 

Units

N/A

 

Secondary Outputs

 

Aggregate Numeric Breast Density Score

RadElement ID

RDES75

Definition

Assign an aggregate numeric breast density score

Data Type

Numeric

Value Set

[0,1]

Units

N/A


Tertiary Outputs

 

Regional Breast Density Information

RadElement ID

RDES75

Definition

Regional breast density score

Data Type

Numeric

Value Set

List a score [0-1] for each region.

For example:

Right breast, upper outer quadrant, posterior third depth

Units

N/A

 

Future Development Ideas


The numeric scores outlined above could facilitate research into (1) the natural progression of breast density with age, (2) the identification of density change patterns likely to develop into cancer, and (3) the identification of density change patterns likely to be masking cancer.