Research map / Label-efficient and robust learning / Self-supervised and contrastive
MTNeuro: A Benchmark for Evaluating Representations of Brain Structure Across Multiple Levels of Abstraction
Jorge Quesada, Lakshmi Sathidevi, Ran Liu, Nauman Ahad, Joy M Jackson, Mehdi Azabou, Jingyun Xiao, Chris Liding, Matthew Jin, Carolina Urzay, William Gray-Roncal, Erik Christopher Johnson and 1 more
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 |
|---|---|---|---|
| Train with an additional cube of data | BYOL ROI classifier · ROI-C1 to ROI-C3 | accuracy: 0.88 ± 0.02 -> 0.97 ± 0.01 | Improved |
| Train with an additional cube of data | MYOW ROI classifier · ROI-C1 to ROI-C3 | accuracy: 0.90 ± 0.02 -> 0.98 ± 0.01 | Improved |
| Train with an additional cube of data | Supervised ROI classifier · ROI-C1 to ROI-C3 | accuracy: 0.88 ± 0.03 -> 0.88 ± 0.02 | No clear effect |
| Train supervised ResNet18 with Mixup | Supervised ROI classifier · ROI-C1 | accuracy: 0.88 ± 0.03 -> 0.90 ± 0.04 | Improved |
| Increase image resolution from 2x downsampling to full resolution | Self-supervised ROI classifiers · ROI-C1 | accuracy: BYOL: 0.88 ± 0.02 -> 0.98 ± 0.01; MYOW: 0.90 ± 0.02 -> 0.96 ± 0.03; MYOW-m: 0.94 ± 0.02 -> 0.99 ± 0.01 | Improved |
| Increase image resolution from 2x downsampling to full resolution | Supervised ROI classifiers · ROI-C1 | accuracy: Resnet18: 0.88 ± 0.03 -> 0.87 ± 0.1; Resnet18-Mixup: 0.90 ± 0.04 -> 0.91 ± 0.03 | No clear effect |
| Use the 3-class setting instead of 4-class segmentation | 2D segmentation models · Pixel-level segmentation | mean IoU: 0.72 to 0.79 | Improved |
| Use 3D inputs instead of 2D inputs | U-Net segmentation · 3-class segmentation | mean IoU: 0.79 ± 0.014 -> 0.80 ± 0.007 | No clear effect |
| Train on two subvolumes rather than one | BYOL feature representation · Semantic feature readout | R², vessels: 0.85 ± 0.01 -> 0.88 ± 0.00 | Improved |
| Train on two subvolumes rather than one | Supervised feature representation · Semantic feature readout | R², cell count: 0.67 ± 0.06 -> 0.73 ± 0.02 | Improved |
| Use MYOW's merged single-projector and predictor design (ablation) | MYOW · ROI-C1 | accuracy: 0.90 ± 0.02 -> 0.94 ± 0.02 | Improved |
Closest papers by meaning
- Neural Regression, Representational Similarity, Model Zoology & Neural Taskonomy at Scale in Rodent Visual Cortex (NeurIPS 2021 · similarity 0.89)
- CLIP-MSM: A Multi-Semantic Mapping Brain Representation for Human High-Level Visual Cortex (AAAI 2025 · similarity 0.86)
- Brain Mapping with Dense Features: Grounding Cortical Semantic Selectivity in Natural Images With Vision Transformers (ICLR 2025 · similarity 0.86)
- In Silico Mapping of Visual Categorical Selectivity Across the Whole Brain (NeurIPS 2025 · similarity 0.86)
- Learning to cluster neuronal function (NeurIPS 2025 · similarity 0.86)
- TopoNets: High performing vision and language models with brain-like topography (ICLR 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.
