Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Semi-Supervised Contrastive VAE for Disentanglement of Digital Pathology Images

Mahmudul Hasan, Xiaoling Hu, Shahira Abousamra, Prateek Prasanna, Joel Saltz, Chao Chen

TL;DR

This paper proposes the first disentanglement method for pathology images for detecting tumor-infiltrating lymphocytes (TIL) and achieves superior performance on complex pathology images, thus improving the interpretability and even generalization power of TIL detection deep learning models.

Abstract

Despite the strong prediction power of deep learning models, their interpretability remains an important concern. Disentanglement models increase interpretability by decomposing the latent space into interpretable subspaces. In this paper, we propose the first disentanglement method for pathology images. We focus on the task of detecting tumor-infiltrating lymphocytes (TIL). We propose different ideas including cascading disentanglement, novel architecture, and reconstruction branches. We achieve superior performance on complex pathology images, thus improving the interpretability and even generalization power of TIL detection deep learning models. Our codes are available at https://github.com/Shauqi/SS-cVAE.

Semi-Supervised Contrastive VAE for Disentanglement of Digital Pathology Images

TL;DR

This paper proposes the first disentanglement method for pathology images for detecting tumor-infiltrating lymphocytes (TIL) and achieves superior performance on complex pathology images, thus improving the interpretability and even generalization power of TIL detection deep learning models.

Abstract

Despite the strong prediction power of deep learning models, their interpretability remains an important concern. Disentanglement models increase interpretability by decomposing the latent space into interpretable subspaces. In this paper, we propose the first disentanglement method for pathology images. We focus on the task of detecting tumor-infiltrating lymphocytes (TIL). We propose different ideas including cascading disentanglement, novel architecture, and reconstruction branches. We achieve superior performance on complex pathology images, thus improving the interpretability and even generalization power of TIL detection deep learning models. Our codes are available at https://github.com/Shauqi/SS-cVAE.
Paper Structure (9 sections, 2 equations, 5 figures, 2 tables)

This paper contains 9 sections, 2 equations, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Unsupervised disentangling methods.
  4. Contrastive analysis based disentangling methods.
  5. Method
  6. Experiments and Results
  7. Conclusions
  8. Acknowledgement.
  9. Disclosure of Interests.

Figures (5)

  • Figure 1: Whole SS-cVAE Framework. I) Module to disentangle discriminative factors from common factors. II) Module for disentangling factors related to density. Note that, the same salient encoder from (I) is fine-tuned here. III) Module for image reconstruction from the disentangled latent generated from (I) and (II).
  • Figure 2: The first module of SS-cVAE. It performs disentanglement of discriminative factor from common factor given any patch of $C$ or $B$.
  • Figure 3: 2nd and 3rd modules of SS-cVAE. II) Module to disentangle the factor corresponding to density. During InfoNCE loss calculation in (II) we consider one of the $s_{high}$ samples as $s_{anchor}$ III) GAN-based reconstruction using disentangled latent.
  • Figure 4: Data Preprocessing. A) Patch extraction and patch labeling using ground truth cell annotation. B) Synthetic Data Creation using a copy-paste mechanism.
  • Figure 5: Example of Latent Swap. Row 1 and 2 represent salient latent swapping between the samples of $C$ and $B$. Row 3 and 4 represent the salient latent swapping between the samples of $H.T$ and $L.T$.