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


Current browse context:


Change to browse by:

References & Citations


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

Electrical Engineering and Systems Science > Image and Video Processing

Title: Improving Mammography Malignancy Segmentation by Designing the Training Process

Abstract: We work on the breast imaging malignancy segmentation task while focusing on the training process instead of network complexity. We designed a training process based on a modified U-Net, increasing the overall segmentation performances by using both, benign and malignant data for training. Our approach makes use of only a small amount of annotated data and relies on transfer learning from a self-supervised reconstruction task, and favors explainability.
Subjects: Image and Video Processing (eess.IV)
Report number: MIDL/2020/ExtendedAbstract/vVsWe9-s0G
Cite as: arXiv:2006.00060 [eess.IV]
  (or arXiv:2006.00060v1 [eess.IV] for this version)

Submission history

From: Mickael Tardy [view email]
[v1] Fri, 29 May 2020 20:14:45 GMT (638kb,D)

Link back to: arXiv, form interface, contact.