Daniel Budreau PhD, Presenter: Nothing to Disclose

Maryellen L. Giger PhD, Abstract Co-Author: Stockholder, Hologic, Inc Shareholder, Quantitative Insights, Inc Royalties, Hologic, Inc Royalties, General Electric Company Royalties, MEDIAN Technologies Royalties, Riverain Medical Royalties, Mitsubishi Corporation Royalties, Toshiba Corporation Researcher, Koninklijke Philips Electronics NV Researcher, U-Systems, Inc

Li Lan, Abstract Co-Author: Nothing to Disclose

To predict response to therapy by applying linear discriminant analysis (LDA) and Bayesian artifical neural networks (BANN) to MRI computer-extracted characteristics of pre- and post-treatment primary breast lesions.

A database of pre-treatment DCE-MR images were obtained from thirty-six patients who received neoadjuvant chemotherapy between January 2003 and December 2010 using an institutional review-board approved protocol. Thirty of these patients additionally recieved post-treatment MR images. A radiologist determined response to therapy and identified the breast lesion; the computer then performed automatic lesion segmentation and extracted lesion-based features. Of the patients with post-treatment DCE-MRIs, the change in feature values as a percentage of baseline was evaluated. LDA and BANN with automatic relevance determination for joint feature selection and classification were used to merge features. Area under the receiver operating characteristic curve (AUC) served as the figure of merit in predicting response to therapy for each individual feature as well as output from the LDA and BANN.

The LDA classifier based on pre-treatment lesions yielded an AUC=0.81 (95% confidence interval; 0.74 – 0.88 [CI] ) and the BANN classifier had an AUC=0.90 (95% CI: 0.85 – 0.95), both of which were significantly different from guessing. Among individual pre-treatment features, time-to-peak, uptake rate, and normalized total rate variation demonstrated a significant difference under Bonferroni correction. The LDA classifer based on change between pre- and post-treatment lesions had an AUC=0.82 (95% CI: 0.73 – 0.91); the BANN classifier gave an AUC=0.71 (95% CI: 0.81 – 0.61). Among individual features, time-to-peak and uptake rate demonstrated a significant difference under a Bonferroni correction.

The results demonstrate that LDA, BANN, and some kinetic features can distinguish response to therapy in patients based on pre-treatment features extracted from pre-treatment lesions as well as change in features between pre- and post-treatment lesions.

Classifers using features extracted from breast DCE-MRI could be used to predict response to neoadjuvant therapy, providing clinicians a tool when weighing risk against benefit in patient management.

Budreau, D, Giger, M, Lan, L, Breast MRI-based Feature Analysis in Predicting Neoadjuvant Therapy Response.  Radiological Society of North America 2012 Scientific Assembly and Annual Meeting, November 25 - November 30, 2012 ,Chicago IL. http://archive.rsna.org/2012/12034375.html