Summary
Batch effects—systematic technical variations unrelated to the biological signal—are the central obstacle in biomedical imaging, undermining reproducibility and causing deep learning models to fail on new experimental batches. This paper introduces CS-ARM-BN (Control-Stabilized Adaptive Risk Minimization via Batch Normalization), a meta-learning adaptation method that leverages negative control samples: unperturbed reference images present in every experimental batch by design, used as stable context for adaptation. On Mechanism-of-Action (MoA) classification from the large-scale JUMP-CP dataset, a standard ResNet drops from 0.939±0.005 accuracy on the training domain to 0.862±0.060 on new experimental batches; foundation models with typical variation normalization fail to close this gap. The proposed meta-learning approach reaches 0.935±0.018, closing the domain gap for the first time. When strong domain shifts occur (e.g., data generated in a different lab), adaptation is stabilized by control samples, which are always available in biomedical experiments. The work shows batch effects can be neutralized with principled, practical in-context adaptation. Source: arXiv 2604.20824, posted on zhichai.net.
Paper Overview
Field: Machine Learning
Authors: Ana Sanchez-Fernandez, Thomas Pinetz, Werner Zellinger
Published: 2026-04-22
arXiv: 2604.20824
Summary
The central problem in biomedical imaging is batch effects: systematic technical variations unrelated to the biological signal of interest. These batch effects critically undermine experimental reproducibility and are the primary cause of deep learning system failures on new experimental batches, preventing practical real-world use. Despite years of research, no method has succeeded in closing this performance gap for deep learning models.
The authors propose CS-ARM-BN (Control-Stabilized Adaptive Risk Minimization via Batch Normalization), a meta-learning adaptation method that exploits negative control samples. Such unperturbed reference images are present in every experimental batch by design and serve as stable context for adaptation.
Results
Validated on Mechanism-of-Action (MoA) classification using the large-scale JUMP-CP dataset:
- Standard ResNet: accuracy drops from 0.939±0.005 on the training domain to 0.862±0.060 on new experimental batches.
- Foundation models, even with typical variation normalization, fail to close this gap.
- The proposed meta-learning approach achieves 0.935±0.018, closing the domain gap.
When new experimental batches exhibit strong domain shift (e.g., data generated in a different laboratory), the meta-learning method can be stabilized using control samples, which are always available in biomedical experiments.
Conclusion
Batch effects in biomedical imaging data can be effectively neutralized through principled in-context adaptation, in a way that is both practical and efficient.
---
*Auto-collected on 2026-04-24*
This page is an English static mirror generated for search and AI citation.
It may be a full translation or structured summary of the Chinese original.
Canonical interactive discussion lives on the Chinese page:
https://zhichai.net/topic/177618686