Uncertainty-Guided Selective Adaptation Enables Cross-Platform Predictive Fluorescence Microscopy

TL;DR

This paper introduces a novel selective adaptation method for deep learning in microscopy, focusing on adapting only early layers to improve cross-platform transfer without disrupting semantic features.

Contribution

It proposes SIT-ADDA-Auto, a self-configuring framework that automatically selects adaptation depth using uncertainty, enhancing robustness in microscopy image translation.

Findings

Improved cross-instrument transfer and segmentation accuracy.

Reduced semantic drift compared to full-network adaptation.

Robust performance across various imaging conditions.

Abstract

Deep learning is transforming microscopy, yet models often fail when applied to images from new instruments or acquisition settings. Conventional adversarial domain adaptation (ADDA) retrains entire networks, often disrupting learned semantic representations. Here, we overturn this paradigm by showing that adapting only the earliest convolutional layers, while freezing deeper layers, yields reliable transfer. Building on this principle, we introduce Subnetwork Image Translation ADDA with automatic depth selection (SIT-ADDA-Auto), a self-configuring framework that integrates shallow-layer adversarial alignment with predictive uncertainty to automatically select adaptation depth without target labels. We demonstrate robustness via multi-metric evaluation, blinded expert assessment, and uncertainty-depth ablations. Across exposure and illumination shifts, cross-instrument transfer, and multiple stains, SIT-ADDA improves reconstruction and downstream segmentation over full-encoder adaptation and non-adversarial baselines, with reduced drift of semantic features. Our results provide a design rule for label-free adaptation in microscopy and a recipe for field settings; the code is publicly available.