0%

(2018) Deep multiscale convolutional feature learning for weakly supervised localization of chest pathologies in X-ray images

Posted on In , ,

Keyword [ChestX-ray14]

Sedai S, Mahapatra D, Ge Z, et al. Deep multiscale convolutional feature learning for weakly supervised localization of chest pathologies in X-ray images[C]//International Workshop on Machine Learning in Medical Imaging. Springer, Cham, 2018: 26


1. Overview

In this paper, it proposes weakly supervised method to localize chest pathologies

  • intermediate feature maps from different stages
  • leared layer relevant weight
  • weighted CAM
  • improves the location performance of small size pathologies (nodule, mass)

1.1. Model



  • Train C-CNN



  • learned weight. (b, c), each (1, c) initialize to 1/B

  • B. number of scales

1.2. Loss Function



  • β. percentage of 0


1.3. Details

  • Adam. LR 1e-3, decay by 0.1 when valid loss plateau
  • randomly split
  • 256x256
  • others. Xavier init