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

Hui Li PhD, Abstract Co-Author: Nothing to Disclose

Elizabeth A. Morris MD, Abstract Co-Author: Nothing to Disclose

Ermelinda Bonaccio MD, Abstract Co-Author: Nothing to Disclose

Kathleen Rae Brandt MD, Abstract Co-Author: Nothing to Disclose

Elizabeth S. Burnside MD, MPH, Abstract Co-Author: Stockholder, Cellectar Biosciences, Inc Stockholder, NeuWave Medical Inc

Basak Erguvan Dogan MD, Abstract Co-Author: Nothing to Disclose

Brenda Fevrier-Sullivan BA, Abstract Co-Author: Nothing to Disclose

John B. Freymann BS, Abstract Co-Author: Nothing to Disclose

Marie Adele Ganott MD, Abstract Co-Author: Nothing to Disclose

Erich Huang PhD, Abstract Co-Author: Nothing to Disclose

C. Carl Jaffe MD, Abstract Co-Author: Nothing to Disclose

Yuan Ji, Abstract Co-Author: Nothing to Disclose

Justin Kirby, Abstract Co-Author: Stockholder, Myriad Genetics, Inc

Jose Miguel Net MD, Abstract Co-Author: Nothing to Disclose

Elizabeth J. Sutton MD, Abstract Co-Author: Nothing to Disclose

Gary J. Whitman MD, Abstract Co-Author: Nothing to Disclose

Margarita Louise Zuley MD, Abstract Co-Author: Research Grant, Hologic, Inc

To investigate the performance of MRI-based phenotypes in predicting the molecular classification of breast cancers in The Cancer Genome Atlas dataset of NCI.

Quantitative image analysis was performed on 98 de-identified, MRI studies depicting biopsy-proven breast cancers MRI studies from the NCI’s multi-institutional The Cancer Imaging Archive and The Cancer Genome Atlas project. Immunohistochemistry molecular classification determined estrogen (ER+82/ER-16), progesterone (PR+75/PR-23) and HER2 (HER2+16/HER2-16) receptor status for each case. Computerized image-based phenotyping included: 1) 3D lesion segmentation based on a fuzzy c-means clustering algorithm; 2) computerized feature extraction; 3) leave-one-out linear stepwise feature selection; and 4) Linear Discriminant Analysis (LDA) as the prognostic predictive classifier. The performance of the classifier model for molecular subtyping was evaluated using jackknifing ROC analysis with area under the ROC curve (AUC) as the figure of merit.

Use of computer-extracted tumor phenotypes in for the task of distinguishing between molecular prognostic indicators, yielded AUC values of 0.79 (p-value < 0.0001), 0.68 (p-value = 0.0066), and 0.61 (p-value =0.126) in the tasks of distinguishing ER- vs ER+, PR- vs PR+, and HER2- vs HER2+, respectively. Features selected for the predictive tasks included volumetrics, texture (entropy), and kinetics for the predictive tasks. 

The results from this study indicate that quantitative MRI analysis shows promise as a means for high-throughput image-based phenotyping in the discrimination of breast cancer subtypes, and potential. Merging imaging phenotypes with genomic data may lead to improved prognostic predictors.

Computerized image-based phenotyping may yield quantitative predictive models of breast cancer for precision medicine.

Giger, M, Li, H, Morris, E, Bonaccio, E, Brandt, K, Burnside, E, Dogan, B, Fevrier-Sullivan, B, Freymann, J, Ganott, M, Huang, E, Jaffe, C, Ji, Y, Kirby, J, Net, J, Sutton, E, Whitman, G, Zuley, M, Relationship of Quantitative MRI-based Phenotypes and the Molecular Classifications of Breast Cancers in the TCGA/TCIA Dataset.  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14015477.html