Research map / Vision-language and multimodal / Cross-modal learning
GECKO: Gigapixel Vision-Concept Contrastive Pretraining in Histopathology
Saarthak Kapse, Pushpak Pati, Srikar Yellapragada, Srijan Das, Rajarsi R. Gupta, Joel Saltz, Dimitris Samaras, Prateek Prasanna
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 |
|---|---|---|---|
| use self-interpretable linear concept aggregator instead of ABMIL in concept branch (ablation) | GECKO ensemble · LUAD vs LUSC (k=10) | AUC: 92.3 -> 95.3 | Improved |
| use self-interpretable linear concept aggregator instead of ABMIL in concept branch (ablation) | GECKO ensemble · EBV+MSI vs Others (k=10) | AUC: 75.1 -> 79.8 | Improved |
| use self-interpretable linear concept aggregator instead of ABMIL in concept branch (ablation) | GECKO concept · LUAD vs LUSC (k=10) | AUC: 88.2 -> 93.5 | Improved |
| add false negative elimination with keep ratio 0.7 (ablation) | GECKO-Zero · five TCGA tasks | AUC | Improved |
| add concept prior alignment with WSI only vs Intra pretraining | GECKO ensemble · TCGA-STAD EBV+MSI vs Others (full supervision) | AUC: 83.5 -> 86.4 | Improved |
| add gene modality to GECKO pretraining (ablation) | GECKO ensemble · TCGA-STAD EBV+MSI vs Others (full supervision) | AUC: 86.4 -> 87.1 | Improved |
| add gene modality to GECKO pretraining (ablation) | GECKO ensemble · TCGA-BRCA HER2 (full supervision) | AUC: 77.2 -> 78.4 | Improved |
| add gene modality to GECKO pretraining (ablation) | GECKO ensemble · TCGA LUAD vs LUSC (full supervision) | AUC: 97.6 -> 97.9 | No clear effect |
| ensemble deep and concept embeddings instead of deep only (ablation) | GECKO (WSI only) · TCGA-STAD MSI vs Others (full supervision) | AUC: 83.8 -> 86.5 | Improved |
| concept-based zero-shot with pretrained concept embeddings instead of MI-Zero | GECKO-Zero · TCGA-STAD MSI vs Others | AUC: 42.3 -> 77.1 | Improved |
| GECKO-Zero vs MI-Zero on lung subtyping | GECKO-Zero · TCGA LUAD vs LUSC | AUC: 96.6 -> 95.0 | No clear effect |
| GECKO ensemble vs TITAN in multimodal setting | GECKO ensemble · TCGA-STAD EBV+MSI vs Others (k=10) | AUC: 78.7 -> 84.4 | Improved |
Closest papers by meaning
- SI-MIL: Taming Deep MIL for Self-Interpretability in Gigapixel Histopathology (CVPR 2024 · similarity 0.87)
- Quilt-LLaVA: Visual Instruction Tuning by Extracting Localized Narratives from Open-Source Histopathology Videos (CVPR 2024 · similarity 0.87)
- Unsupervised Foundation Model-Agnostic Slide-Level Representation Learning (CVPR 2025 · similarity 0.87)
- Few-Shot Learning from Gigapixel Images via Hierarchical Vision-Language Alignment and Modeling (NeurIPS 2025 · similarity 0.87)
- Multi-Resolution Pathology-Language Pre-training Model with Text-Guided Visual Representation (CVPR 2025 · similarity 0.86)
- SlideChat: A Large Vision-Language Assistant for Whole-Slide Pathology Image Understanding (CVPR 2025 · 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.
