Computerized textural analysis of DCE-MRI to enable identification of HER2-enriched breast cancers.

Nathaniel Braman, Prateek Prasanna, Donna Plecha, Hannah Gilmore, Lyndsay N. Harris, Vinay Varadan, Anant MadabhushiJournal of Clinical Oncology 2016 · DOI

Abstract

598 Background: HER2+ breast cancer is biologically and clinically heterogeneous. PAM50 profiling of HER2+ breast cancer reliably identifies the HER2-Enriched (HER2-E) subtype as most responsive to HER2-targeted antibody therapy (trastuzumab). As this subtype is currently only identifiable using molecular profiling of breast tumor tissue, a non-invasive HER2-E identification method could reduce future patient morbidity. We present initial findings involving a new computational imaging feature (CIF) that captures the disorder among pixel level gradient directions on dynamic contrast-enhanced (DCE) MRI. We show that this CIF is able to discriminate between HER2-E and other HER2+ breast cancer subtypes on DCE-MRI. Methods: 25 baseline DCE-MRI HER2+ breast cancer cases were split into two cohorts by PAM50-confirmed subtype (10 HER2-E, 15 non-HER2-E). 13 CIFs were computed on peak contrast image within expert-annotated lesions. This feature set was narrowed down to the 8 most differentially expressed features. Pharmacokinetic (PK) parameters (Kep, Ktrans, Ve), which quantify tumor-associated permeability changes, were also computed using Toft’s model. Consensus clustering was performed using 1) CIFs and 2) PK parameters and peak intensities. Results: Results of 2-group consensus clustering are shown in Table 1. CIFs yielded two distinct clusters, corresponding to HER2-E and non-HER2-E groups, with 70% sensitivity and 60% specificity. Studies sorted to the HER2-E cluster were characterized by under-expression of entropy and over-expression of texture homogeneity features relative to those of the non-HER2-E cluster. In contrast, PK parameters and peak intensities produced poorly defined groups that were uncorrelated with HER2 subtypes. Conclusions: CIFs more accurately identified HER2-E status compared with standard PK parameters on DCE-MRI. CIF expression patterns in HER2-E tumors suggest a less disordered imaging phenotype than other HER2+ tumors. Future validation studies of the HER2-E CIF signature identified here could potentially enable non-invasive prediction of response to trastuzumab. Feature Set Cluster % HER2-E Present % Non-HER2-E Present CIF A 70 40 B 30 60 PK + Intensity A 40 20 B 60 80
BibTeX
@article{braman2016computerized,
  author = {Nathaniel Braman and Prateek Prasanna and Donna Plecha and Hannah Gilmore and Lyndsay N. Harris and Vinay Varadan and Anant Madabhushi},
  title = {{Computerized textural analysis of DCE-MRI to enable identification of HER2-enriched breast cancers.}},
  journal = {Journal of Clinical Oncology},
  year = {2016},
  doi = {10.1200/jco.2016.34.15\_suppl.598},
}

Loading related content...

Figure