Research map / Label-efficient and robust learning / Recognition and long-tail
Deep Learning for BioImaging: What Are We Really Learning?
Ivan Svatko, Maxime Sanchez, Ihab Bendidi, Gilles Cottrell, Auguste Genovesio
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| pretrain the image encoder instead of leaving it untrained | full-image encoders · HEST-1k-1NN CCRCC | gene-wise PCC: 0.12 -> 0.37 | Improved |
| pretrain the image encoder instead of leaving it untrained | full-image encoders · HEST-1k-1NN COAD | gene-wise PCC: 0.27 -> 0.38 | Improved |
| use in-domain foundation model instead of best structure-only model | image vs structure · HEST-1k-1NN CCRCC | gene-wise PCC: 0.16 -> 0.37 | Improved |
| use structure-only model instead of out-of-domain image encoder | structure-only encoder · HEST-1k-1NN COAD | gene-wise PCC: 0.31 vs 0.30 | No clear effect |
| pretrain GCN structure encoder (vs cell count baseline) (ablation) | GCN · HEST-1k-1NN SKCM | gene-wise PCC: 0.40 -> 0.45 | Improved |
| pretrain GCN structure encoder (vs cell count baseline) (ablation) | GCN · HEST-1k-1NN CCRCC | gene-wise PCC: 0.11 -> 0.04 | Hurt |
| use EGNN instead of GCN (ablation) | structure-only encoder · HEST-1k-1NN SKCM | gene-wise PCC: 0.45 -> 0.22 | Hurt |
| pretrain image encoders on RxRx3-core gene retrieval (ablation) | ViT · RxRx3-core recall@5% | recall@5% | No clear effect |
| use in-domain MAE-L/8 and MAE-G/8 pretraining | MAE backbones · RxRx3-core recall@5% | recall@5%: 1.5 times higher averaged top 5% recall | Improved |
| use ResNet instead of ViT | image encoders · RxRx3-core recall@5% | recall@5% | Hurt |
| use deeper layers for representation (ablation) | image encoders · RxRx3-core recall@5% | recall@5% | Hurt |
| pretrain on cell-painting data (in-domain) rather than ImageNet | OpenPhenom vs OOD · JUMP-CP mAP | mAP | Improved |
Closest papers by meaning
- ViTally Consistent: Scaling Biological Representation Learning for Cell Microscopy (ICML 2025 · similarity 0.90)
- MorphoHELM: A Comprehensive Benchmark for Evaluating Representations for Microscopy-Based Morphology Assays (NeurIPS 2026 · similarity 0.89)
- Micro-Bench: A Microscopy Benchmark for Vision-Language Understanding (NeurIPS 2024 · similarity 0.89)
- CellCLIP - Learning Perturbation Effects in Cell Painting via Text-Guided Contrastive Learning (NeurIPS 2025 · similarity 0.87)
- Masked Autoencoders for Microscopy are Scalable Learners of Cellular Biology (CVPR 2024 · similarity 0.86)
- Beyond ViT Tokens: Masked-Diffusion Pretrained Convolutional Pathology Foundation Model for Cell-Level Dense Prediction (NeurIPS 2026 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
