Classifying Suspicious Microcalcifications

Purpose	To automate classification of breast microcalcifications into categories based on level of suspicion at time of diagnostic mammography
Tag(s)
Panel	Breast Imaging
Define-AI ID	18060001
Originator	Christoph Lee
Panel Chair	Elizabeth Burnside
Panel Reviewers	Breast Imaging Panel
License	Creative Commons 4.0
Status	Published

                

Clinical Implementation

Value Proposition

Microcalcifications identified on screening mammography may signify ductal carcinoma in situ (DCIS). However, the majority of microcalcifications are eventually found to be benign, and thus constitute a leading cause of false-positive screens and benign biopsies. There is variability in radiologist interpretation and BI-RADS assessment of microcalcifications at the time of diagnostic imaging (spot magnification evaluation of microcalcifications). AI can help improve accuracy and use quantitative imaging features to more accurately categorize microcalcifications by level of suspicion for DCIS, potentially decreasing the rate of unnecessary benign biopsies.

Narrative(s)

A 40-year-old female presents for her baseline screening mammogram and is found to have a group of microcalcifications in the left breast. She is recalled from screening and undergoes standard diagnostic imaging with spot magnification views in the cranial-caudal (CC) and mediolateral (ML) projections. The AI algorithm provides the radiologist with an automated interpretation of BI-RADS categorization as well as a continuous numerical risk score at the time of the interpretation to help the radiologist provide the most suitable clinical recommendation.

Workflow Description

Spot magnification mammography images obtained at diagnostic workup are sent from PACS to the AI engine. Images are analyzed by the AI engine, and the areas of microcalcification are given a BI-RADS classification and a numerical malignancy risk score. A message is sent to PACS from the engine with this information that can then be used by the interpreting radiologist to make a final assessment and recommendation.

Considerations for Dataset Development

Procedures(s): Diagnostic mammography

View(s): Spot magnification CC and ML mammography images

 

Age: over 40 years

 

Breast Anatomy: no prior surgery or implants

Technical Specifications

Inputs

DICOM Study

Procedure	Diagnostic mammography
Views	Spot magnification CC and ML mammography images
Data Type	DICOM
Modality	Mammo
Body Region	Chest
Anatomic Focus	Breast

 

Primary Outputs

Suspicious Microcalcification Detection

RadElement ID	RDE204
Definition	Identify suspicious microcalcifications
Data Type	Categorical
Value Set	0-Unknown 1-Suspicious microcalcification present 2-Suspicious microcalcification absent
Units	N/A

 

Probability of malignant microcalcification

RadElement ID	RDE205
Definition	For a selected microcalcification, identify the probability of malignancy
Data Type	Numeric
Value Set	[0,1] 0-Benign 1-Malignant
Units	N/A

Future Development Ideas

Develop a robust algorithm to handle microcalcification evaluations for screening images. Consider aggregating malignancy scoring based on evaluation of magnification and screening.

Related Datasets

The Cancer Imaging Archive (TCIA)

• CBIS-DDSM

Know of more related datasets? Please let us know