Classifying High-Risk Lesions - LN

 

Purpose

To automate classification of mammographic FEA into categories based on level of suspicion of upgrade to malignancy at the time of radiology-pathology correlation incorporating pathology results

Tag(s)


Panel

Breast Imaging

Define-AI ID

19060006

Originator

Yiming Gao

Lead

Yiming Gao

Panel Chair

Elizabeth Burnside

Panel Reviewers

Breast Imaging Panel

License

Creative Commons 4.0 

Status

Public Comment

RadElement Set RDES121 

Clinical Implementation


Value Proposition

Microcalcifications identified on mammography are a common manifestation of high-risk lesions at percutaneous biopsy, which routinely undergo surgical excision to exclude associated malignancy. Although high-risk lesions (such as LN) are considered nonobligate precursor lesions with variable upgrade rates to malignancy, majority of surgical excisions are ultimately benign, therefore potentially unnecessary. There is variability in radiology-pathology correlation by the radiologist, and in histologic classification by the pathologist. AI can help improve accuracy in predicting likelihood of malignancy in high risk lesions (specifically, LN in this case), by incorporating quantitative imaging features of originally biopsied microcalcifications at diagnostic imaging, and text features of pathology report and/or histopathologic slide imaging features, as well as patient risk factors and characteristics. This could serve as a more consistent and reproducible form of multi-disciplinary evaluation, to potentially decrease the rate of unnecessary surgical excisions. 

Narrative(s)

A 45-year old female found to have a new group of microcalcifications in the right breast at screening is recalled and undergoes diagnostic imaging with magnification views in CC and ML projections. The calcifications are recommended for stereotactic biopsy, which yields classic type LCIS as the highest grade lesion. At the time of radiology-pathology correlation, the AI algorithm provides the radiologist with an automated continuous numerical risk score for upgrade to malignancy based on radiologic lesion morphology (microcalcifications on CC and ML magnification views) and pathology result (text features in pathology report, and/or actual histopathologic slide features), as well as patient characteristics (age, family history, breast density, prior cancer, mutations) to help guide the best clinical recommendation. 

Workflow Description

Magnification mammography images obtained at diagnostic work-up are sent from PACS to the AI engine. Pathology reports and/or digitized histopathologic slide images are fed to the AI engine. Patient characteristics from EMR are also incorporated. Radiologic images are analyzed in the context of pathology report key words (text features); or, radiologic images and histopathologic slide images are analyzed in conjunction; also factoring in patient characteristics, to render a numerical malignancy risk score. A message is sent to PACS from the engine with this information which will be used by the interpreting radiologist to make a final assessment and appropriate recommendation of further surgical excision versus imaging follow up.


Considerations for Dataset Development



Procedure(s)

Stereotactic biopsy, Diagnostic mammography

View(s)

Magnification CC, ML mammography images

Age

40 years and older

Indication

High-risk lesion diagnosis (FEA)

Breast Anatomy

No prior surgery or implants

Digitized histopathology slide images

varied

Percutaneous biopsy pathology report

varied

Clinical Note: cases not always pure Flat Epithelial Atypia or pure Lobular Neoplasia, but often a mix

Technical Specifications


Inputs

DICOM Study

Procedure

Stereotactic biopsy, Diagnostic mammography

Views

Magnification CC and ML mammography images

Data Type

DICOM

Modality

MAMMO

Body Region

Chest

Anatomic Focus

Breast

 

Secondary Inputs

 

Pathology Report

Definition

Contents of the pathology report

Data Type

DICOM

Value Set

N/A

Units

N/A

 

Digitized Slide Image Data

Definition

Imaging data from biopsy slide

Data Type

DICOM

Value Set

N/A

Units

N/A

 

Primary Outputs

 

Detection of Suspicious Microcalcifications

RadElement ID

RDE790

Definition

Detect lesions with high risk of malignancy

Data Type

Categorical

Value Set

  • Unknown

  • Suspicious microcalcification present

  • Suspicious microcalcification absent

Units

N/A

  

 

Future Development Ideas


Develop robust algorithm to not only be able to provide a numerical malignancy risk score of immediate surgical upgrade rate, but provide an individualized long-term risk of breast cancer in a given patient.  


Related Datasets


No known related public datasets at this time,  please alert us if you know of any.