We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

q-bio.QM

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Quantitative Biology > Quantitative Methods

Title: TopoTxR: A Topological Biomarker for Predicting Treatment Response in Breast Cancer

Abstract: Characterization of breast parenchyma on dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) is a challenging task owing to the complexity of underlying tissue structures. Current quantitative approaches, including radiomics and deep learning models, do not explicitly capture the complex and subtle parenchymal structures, such as fibroglandular tissue. In this paper, we propose a novel method to direct a neural network's attention to a dedicated set of voxels surrounding biologically relevant tissue structures. By extracting multi-dimensional topological structures with high saliency, we build a topology-derived biomarker, TopoTxR. We demonstrate the efficacy of TopoTxR in predicting response to neoadjuvant chemotherapy in breast cancer. Our qualitative and quantitative results suggest differential topological behavior of breast tissue on treatment-na\"ive imaging, in patients who respond favorably to therapy versus those who do not.
Comments: 12 pages, 5 figures, 2 tables, accepted to International Conference on Information Processing in Medical Imaging (IPMI) 2021
Subjects: Quantitative Methods (q-bio.QM); Computer Vision and Pattern Recognition (cs.CV); Image and Video Processing (eess.IV)
Cite as: arXiv:2105.06049 [q-bio.QM]
  (or arXiv:2105.06049v1 [q-bio.QM] for this version)

Submission history

From: Fan Wang [view email]
[v1] Thu, 13 May 2021 02:38:48 GMT (2335kb,D)

Link back to: arXiv, form interface, contact.