Meeting Banner
Abstract #1050

Self-Organizing Map Kinetic Features as Prognostic Markers for Classifying Gene Expression Risk for Breast Cancer Recurrence

Majid Mahrooghy 1 , Ahmed B. Ashraf 1 , Dania Daye 1 , Carolyn Mies 2 , Mark Rosen 1 , Michael Feldman 2 , and Despina Kontos 1

1 Department of Radiology, University of Pennsylvania, Philadelphia, PA, United States, 2 Department of Pathology and Laboratory Medicine, University of Pennsylvania, Philadelphia, PA, United States

We developed DCE-MRI kinetic heterogeneity features using self-organizing map (SOM) neural networks. We use SOM to cluster tumor pixels based on kinetics and extract features including variance and entropy of cluster size, variance of cluster kinetic features, mean and variance of weighted cluster kinetics, and the kinetic features of the cluster having maximum peak enhancement. We evaluated these features for classifying tumor recurrence risk as determined by a validated gene expression assay, and compared their performance to current standard kinetics. Our features have ROC AUC=0.80 for classifying tumors at low- versus high- risk of recurrence, outperforming standard kinetics with AUC=0.65.

This abstract and the presentation materials are available to members only; a login is required.

Join Here