loading page

Automatic detection of microcalcifications in whole slide image - comparison of deep learning and standard computer vision approaches
  • +4
  • Marceau Clavel ,
  • Stéphane Sockeel ,
  • Marie Sockeel ,
  • Catherine Miquel ,
  • Julien Adam ,
  • Elisabeth Lanteri ,
  • Nicolas Pozin
Marceau Clavel
Primaa, Primaa

Corresponding Author:[email protected]

Author Profile
Stéphane Sockeel
Author Profile
Marie Sockeel
Author Profile
Catherine Miquel
Author Profile
Julien Adam
Author Profile
Elisabeth Lanteri
Author Profile
Nicolas Pozin
Author Profile

Abstract

Early detection of breast cancer through mammography is crucial for successful treatment. Microcalcifications are small deposits of calcium in breast ducts, they can be an indication of breast cancer and are first detected by mammography. However, their presence must be confirmed by an histopathologist through slides examination. As a help to practitioners, we present an automatic microcalcification detection pipeline in Whole Slide Images (WSI). This pipeline is a patch-based approach, in which patches from epithelial regions are analyzed by a classifier to determine if they contain microcalcifications. We have developed two classifiers for this pipeline, one based on image processing techniques and the other on deep learning. In addition to standard classification metrics, we compare the performance of these two classifiers using custom metrics specifically designed to capture the value brought by the pipeline to pathologists. Both approaches provide promising results. The deep learning classifier has better metrics, but requires more labeled data to be implemented.