Learned Image Compression and Restoration for Digital Pathology

TL;DR

Digital pathology WSIs pose extreme storage and transmission challenges. CLERIC tackles this with a pathology-tailored LIC framework that combines a learnable lifting scheme to separate frequency bands, two-branch encoders, deformable and recurrent residual blocks, and MEM++ entropy modeling, all optimized via rate-distortion loss: L = R(ŷ) + R ˆz + λ D(x, x̂). The approach demonstrates superior rate-distortion performance over state-of-the-art LIC methods and traditional codecs on both in-house and open pathology datasets, while preserving fine tissue structures essential for diagnostics. This work offers practical implications for data management, long-term storage, and seamless integration into clinical workflows and AI-assisted pathology systems, with potential extensions to raw data evaluation and real-time visualization.

Abstract

Digital pathology images play a crucial role in medical diagnostics, but their ultra-high resolution and large file sizes pose significant challenges for storage, transmission, and real-time visualization. To address these issues, we propose CLERIC, a novel deep learning-based image compression framework designed specifically for whole slide images (WSIs). CLERIC integrates a learnable lifting scheme and advanced convolutional techniques to enhance compression efficiency while preserving critical pathological details. Our framework employs a lifting-scheme transform in the analysis stage to decompose images into low- and high-frequency components, enabling more structured latent representations. These components are processed through parallel encoders incorporating Deformable Residual Blocks (DRB) and Recurrent Residual Blocks (R2B) to improve feature extraction and spatial adaptability. The synthesis stage applies an inverse lifting transform for effective image reconstruction, ensuring high-fidelity restoration of fine-grained tissue structures. We evaluate CLERIC on a digital pathology image dataset and compare its performance against state-of-the-art learned image compression (LIC) models. Experimental results demonstrate that CLERIC achieves superior rate-distortion (RD) performance, significantly reducing storage requirements while maintaining high diagnostic image quality. Our study highlights the potential of deep learning-based compression in digital pathology, facilitating efficient data management and long-term storage while ensuring seamless integration into clinical workflows and AI-assisted diagnostic systems. Code and models are available at: https://github.com/pnu-amilab/CLERIC.

Figures (9)

• Figure 1: Overview of the proposed workflow for pathology image compression. Whole slide images (WSIs) are digitized, segmented into patches, and processed through compression and decompression models. The framework supports multi-resolution storage and retrieval, ensuring efficient image reconstruction and real-time visualization in pathology software.
• Figure 2: Overview of the CLERIC framework. The model utilizes a learnable lifting scheme to decompose input images into frequency components, followed by separate encoding branches for low- and high-frequency features. Both the encoder and decoder incorporate deformable and recurrent residual blocks to enhance feature extraction and reconstruction. The compressed latent representation undergoes entropy coding, and the decoder reconstructs the image while preserving fine-grained pathological details.
• Figure 3: Illustration of the learned lifting scheme module. The input image is progressively decomposed into low- and high-frequency components using a row-wise and column-wise lifting scheme. The resulting four sub-bands capture multiscale structural information, facilitating more efficient compression and reconstruction. The module employs the Cohen-Daubechies-Feauveau (CDF) 9/7 wavelet and learnable convolutional operators to refine the decomposition process.
• Figure 4: (a) Structure of Deformable Convolution v2 (DCNv2). DCNv2 enhances standard convolution by introducing learnable offsets and modulation scalars, allowing adaptive sampling and improved spatial feature extraction. (b) Visualization of learned offset sampling positions in the Deformable Residual Block with Stride (DRBS). The dynamically adjusted receptive fields help preserve structural integrity and improve feature representation, particularly in pathology image compression.
• Figure 5: (a) Rate-distortion curves for the proposed CLERIC model and comparison methods on the in-house dataset. (b) Rate-distortion curves for the proposed CLERIC model and comparison methods on the public dataset. Higher curves indicate better compression efficiency and reconstruction quality.