Neoadjuvant chemotherapy can be given to breast cancer patients to treat micrometastatic disease and possibly downstage tumors in hopes of performing less radical surgery. Approximately 85% of patients are successfully treated with this method, while 15% have stable or progressive disease, said presenter Daniel Budreau, PhD, of the University of Chicago.
The researchers sought to employ dynamic contrast-enhanced MRI before neoadjuvant therapy to predict if response will occur.
"Ultimately, the physician can use this data when weighing the costs versus benefits of continuing neoadjuvant chemotherapy that might not appear to be successful," Budreau said.
Quantitative image analysis methods were employed to extract 30 MRI-based lesion features related to contrast kinetics, morphology, texture, and variance. These features, individually or merged, can be viewed as tumor signatures, he said. Merged features were obtained using two types of classifiers: linear discriminant analysis (LDA) and a Bayesian artificial neural network (BANN).
"In our initial study, we showed a significant difference for predicting response to therapy using pretreatment MR images and image-based tumor signatures consisting of a subset of kinetic features," Budreau said. "Our results are promising and warrant a larger clinical study."
Budreau noted that the ability to automatically extract lesion characteristics from MR images and combine them into a predictive classifier can be generalized.
"Our research aims to essentially create a virtual biopsy that potentially will provide useful information to clinicians for use in their clinical decision-making regarding cancer diagnosis, prognosis, or treatment," he said.
