eduzhai > Applied Sciences > Engineering >

StyPath Style-Transfer Data Augmentation For Robust Histology Image Classification

  • king
  • (0) Download
  • 20210505
  • Save

... pages left unread,continue reading

Document pages: 12 pages

Abstract: The classification of Antibody Mediated Rejection (AMR) in kidney transplantremains challenging even for experienced nephropathologists; this is partlybecause histological tissue stain analysis is often characterized by lowinter-observer agreement and poor reproducibility. One of the implicated causesfor inter-observer disagreement is the variability of tissue stain qualitybetween (and within) pathology labs, coupled with the gradual fading ofarchival sections. Variations in stain colors and intensities can make tissueevaluation difficult for pathologists, ultimately affecting their ability todescribe relevant morphological features. Being able to accurately predict theAMR status based on kidney histology images is crucial for improving patienttreatment and care. We propose a novel pipeline to build robust deep neuralnetworks for AMR classification based on StyPath, a histological dataaugmentation technique that leverages a light weight style-transfer algorithmas a means to reduce sample-specific bias. Each image was generated in 1.84 +-0.03 seconds using a single GTX TITAN V gpu and pytorch, making it faster thanother popular histological data augmentation techniques. We evaluated our modelusing a Monte Carlo (MC) estimate of Bayesian performance and generate anepistemic measure of uncertainty to compare both the baseline and StyPathaugmented models. We also generated Grad-CAM representations of the resultswhich were assessed by an experienced nephropathologist; we used thisqualitative analysis to elucidate on the assumptions being made by each model.Our results imply that our style-transfer augmentation technique improveshistological classification performance (reducing error from 14.8 to 11.5 )and generalization ability.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...