Continual Domain Incremental Learning for Privacy-aware Digital Pathology

TL;DR

This paper introduces a privacy-preserving continual learning method for digital pathology that uses generative models to simulate past data, improving robustness against data shifts without storing sensitive information.

Contribution

The paper proposes a novel Generative Latent Replay-based CL approach that avoids storing past data, addressing privacy concerns in medical image analysis.

Findings

Outperforms buffer-free CL methods in histopathology tasks

Performs comparably to rehearsal-based CL with large buffers

Effective against stain and organ data shifts

Abstract

In recent years, there has been remarkable progress in the field of digital pathology, driven by the ability to model complex tissue patterns using advanced deep-learning algorithms. However, the robustness of these models is often severely compromised in the presence of data shifts (e.g., different stains, organs, centers, etc.). Alternatively, continual learning (CL) techniques aim to reduce the forgetting of past data when learning new data with distributional shift conditions. Specifically, rehearsal-based CL techniques, which store some past data in a buffer and then replay it with new data, have proven effective in medical image analysis tasks. However, privacy concerns arise as these approaches store past data, prompting the development of our novel Generative Latent Replay-based CL (GLRCL) approach. GLRCL captures the previous distribution through Gaussian Mixture Models instead of storing past samples, which are then utilized to generate features and perform latent replay with new data. We systematically evaluate our proposed framework under different shift conditions in histopathology data, including stain and organ shift. Our approach significantly outperforms popular buffer-free CL approaches and performs similarly to rehearsal-based CL approaches that require large buffers causing serious privacy violations.