Nilanjana Banerjee PhD, Presenter: Employee, Koninklijke Philips Electronics NV

Vinay Varadan, Abstract Co-Author: Employee, Koninklijke Philips Electronics NV

Sitharthan Kamalakaran PhD, Abstract Co-Author: Employee, Koninklijke Philips Electronics NV

Angel Janevski PhD, Abstract Co-Author: Employee, Koninklijke Philips Electronics NV

William Sikov MD, Abstract Co-Author: Nothing to Disclose

Maysa M. Abu-Khalaf MD, Abstract Co-Author: Nothing to Disclose

Veerle Bossuyt MD, Abstract Co-Author: Nothing to Disclose

Donald R. Lannin MD, Abstract Co-Author: Nothing to Disclose

Lyndsay Harris MD, Abstract Co-Author: Nothing to Disclose

Daniel Cornfeld MD, Abstract Co-Author: Nothing to Disclose

Nevenka Dimitrova, Abstract Co-Author: Employee, Koninklijke Philips Electronics NV

Texture features are important in lesion characterization but their relationship to molecular phenotypes is unknown. Molecular stratification of breast cancer into luminal, basal, ERBB2, and normal-like can be made based on gene expression profiles. We investigate how imaging-based texture features relate to tumor biology and genetic subtypes using MRI, histopathological and RNA-sequencing data.

Data from 74 Stage IIA to IIIB breast cancer patients enrolled in neo-adjuvant clinical trials NCT00617942 and NCT00723125 were retrospectively reviewed. We evaluated the following gray-level co-occurrence matrix features (GLCM) on post contrast T1 fat suppressed images of 42 tumors in 42 patients: angular second moment, contrast correlation, first diagonal moment, entropy, as well as fractal dimension. Tumors were profiled based on estrogen receptor (ER), progesterone receptor (PR), and HER2 status for validation of tumor subtypes within the PAM50 gene set. Wilcoxon signed rank test was used to determine whether the distribution of values for each texture feature differed between luminal and basal subtypes. We then performed hierarchical clustering on our patient data set based on multiple texture features and evaluated their association with hormone receptor status (ER and PR) using the statistical package R. We also performed RNA-sequencing on 23 tumors and compared RNAseq-based PAM50 clustering with texture-based clustering. Statistical significance of clusters was determined by hypergeometric test.

First diagonal moment was most significantly associated with tumor subtypes (p=0.04). There was an association between texture features and tumor hormone status (14/20 ER negatives clustered, p=0.0008 and 15/23 PR negatives clustered, p=0.0005). Furthermore, the texture-based clusters differentiated between basal and luminal PAM50 subtypes (4 out of 5 basals clustered, p=0.0007).

Our results indicate that certain texture features from MRI images can identify biological heterogeneity in breast tumors. They suggest that local imaging features are correlated with tissue-level molecular phenotypes. This could potentially impact clinical management decisions and therapy selection.

This study suggests that texture features can provide insights into biological heterogeneity of breast tumors.

Banerjee, N, Varadan, V, Kamalakaran, S, Janevski, A, Sikov, W, Abu-Khalaf, M, Bossuyt, V, Lannin, D, Harris, L, Cornfeld, D, Dimitrova, N, Evaluation of GLCM Texture Features in MRI and Association with Molecular Phenotypes in Breast Cancer.  Radiological Society of North America 2012 Scientific Assembly and Annual Meeting, November 25 - November 30, 2012 ,Chicago IL. http://archive.rsna.org/2012/12034218.html