Imaging as a Surrogate for the Early Prediction and Assessment of Treatment Response through the Analysis of 4-D Texture Ensembles (ISEPARATE)

In order to facilitate the use of imaging as a surrogate endpoint for the early prediction and assessment of treatment response, we present a quantitative image analysis system to process the anatomical and functional images acquired over the course of treatment. The key features of our system are deformable registration, texture analysis via texton histograms, feature selection using the minimal-redundancy-maximal-relevance method, and classification using support vector machines. The objective of the proposed image analysis and machine learning methods in our system is to permit the identification of multi-parametric imaging phenotypic properties that have superior diagnostic and prognostic value as compared to currently used morphometric measurements. We evaluate our system for predicting treatment response of breast cancer patients undergoing neoadjuvant chemotherapy using a series of MRI acquisitions.