206 - Using Ultrasound Features as Predictive & Prognostic Biomarkers

Friday, April 3, 2020

12:30 PM – 1:00 PM

Location: LinQ - Social C

BI-RADS features are now used as qualitative diagnostic biomarkers to determine whether or not to recommend biopsy (LOM >2%). When LOM > 2%, we currently simply "guesstimate" LOM and BI-RADS category. We present an ordinal scoring system of 5 US features, where LOM increases with score for each individual feature. We present combinations of feature scores, regression equation and machine learning approaches to as more objective and precise ways of predicting LOM with 90% CIs.

Learning Objectives:

Score 5 individual US features Know how each US feature score correlates with PPV3. Know limitations of individual US feature scores as negative predictors of malignancy.
Know how combining US feature scores improves both positive and negative predictive values. Understand how machine learning approaches can better assess LOM by combining feature scores.
Know how some US features correlate with receptors such as ER, PR, HER2 and Ki67. Know how some features correlate with molecular subtypes, particularly in distinguishing LUM A from TNBC.

Speaker(s)

A. Thomas Stavros, MD, FACR, FSBI, FSRU

Professor Specialist
University of Texas Health Sciences Center, San Antonio & Chief Medical Officer Seno Medical, San Antonio
U of Texas Health Sciences Center, San Antonio

Disclosure: Disclosure information not submitted.