Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Learned Image Compression for HE-stained Histopathological Images via Stain Deconvolution

Maximilian Fischer, Peter Neher, Tassilo Wald, Silvia Dias Almeida, Shuhan Xiao, Peter Schüffler, Rickmer Braren, Michael Götz, Alexander Muckenhuber, Jens Kleesiek, Marco Nolden, Klaus Maier-Hein

TL;DR

The paper addresses the large storage burden of digital histopathology WSIs and the inadequacy of standard JPEG for DL-based compression. It introduces Stain Quantized Latent Compression (SQLC), a two-stage DL framework that fuses RGB and staining-domain information via a stain encoder and compresses the latent space with a neural image compression model, yielding quantized latents that can be decoded to six-channel RGB+HED data. Empirical results show SQLC achieves superior compression with preserved perceptual quality (MS-SSIM) and higher downstream classification accuracy (e.g., on Camelyon16) compared with JPEG and baseline NICM methods, especially at high compression ratios. The approach offers a modular, hardware-aware solution for clinical deployment and opens avenues for applying stain-aware compression to other histopathology tasks and stainings.

Abstract

Processing histopathological Whole Slide Images (WSI) leads to massive storage requirements for clinics worldwide. Even after lossy image compression during image acquisition, additional lossy compression is frequently possible without substantially affecting the performance of deep learning-based (DL) downstream tasks. In this paper, we show that the commonly used JPEG algorithm is not best suited for further compression and we propose Stain Quantized Latent Compression (SQLC ), a novel DL based histopathology data compression approach. SQLC compresses staining and RGB channels before passing it through a compression autoencoder (CAE ) in order to obtain quantized latent representations for maximizing the compression. We show that our approach yields superior performance in a classification downstream task, compared to traditional approaches like JPEG, while image quality metrics like the Multi-Scale Structural Similarity Index (MS-SSIM) is largely preserved. Our method is online available.

Learned Image Compression for HE-stained Histopathological Images via Stain Deconvolution

TL;DR

The paper addresses the large storage burden of digital histopathology WSIs and the inadequacy of standard JPEG for DL-based compression. It introduces Stain Quantized Latent Compression (SQLC), a two-stage DL framework that fuses RGB and staining-domain information via a stain encoder and compresses the latent space with a neural image compression model, yielding quantized latents that can be decoded to six-channel RGB+HED data. Empirical results show SQLC achieves superior compression with preserved perceptual quality (MS-SSIM) and higher downstream classification accuracy (e.g., on Camelyon16) compared with JPEG and baseline NICM methods, especially at high compression ratios. The approach offers a modular, hardware-aware solution for clinical deployment and opens avenues for applying stain-aware compression to other histopathology tasks and stainings.

Abstract

Processing histopathological Whole Slide Images (WSI) leads to massive storage requirements for clinics worldwide. Even after lossy image compression during image acquisition, additional lossy compression is frequently possible without substantially affecting the performance of deep learning-based (DL) downstream tasks. In this paper, we show that the commonly used JPEG algorithm is not best suited for further compression and we propose Stain Quantized Latent Compression (SQLC ), a novel DL based histopathology data compression approach. SQLC compresses staining and RGB channels before passing it through a compression autoencoder (CAE ) in order to obtain quantized latent representations for maximizing the compression. We show that our approach yields superior performance in a classification downstream task, compared to traditional approaches like JPEG, while image quality metrics like the Multi-Scale Structural Similarity Index (MS-SSIM) is largely preserved. Our method is online available.
Paper Structure (14 sections, 1 equation, 5 figures, 3 tables)

This paper contains 14 sections, 1 equation, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. SQLC Framework
  4. Neural Image Compression (NIC)
  5. Stain Quantized Latent Compression (SQLC)
  6. Experiments
  7. Dataset
  8. Neural Image compression models
  9. Experimental setup
  10. Results
  11. Training of compression models
  12. Classification on compressed images
  13. Discussion
  14. Appendix

Figures (5)

  • Figure 1: Proposed SQLC method. A: Preprocessing for the fusion of RGB image features and staining features HED. B: Lossy compression of the latent representations, derived from part A. C: Finetuning of the decoder part of A to reconstruct the color channels from the quantized latents.
  • Figure 2: Architecture of the SE model.
  • Figure 3: Perceptual Image quality score, evaluated with the MS-SSIM metric. The plot shows the achieved score for each compression scheme for the quality factors 8 - 1, which refer to increasing compression ratio.
  • Figure 4: H-channel for the compression schemes. On the right side an example reconstruction from the SQLC method is shown. The figure shows that the proposed method merely alters the information in the staining channels in contrast to other not finetuned compression schemes.
  • Figure 5: Visualizations of reconstructed images with some of the compared compression schemes.