Research map / Vision-language and multimodal / Human and embodied modeling
ETAB: A Benchmark Suite for Visual Representation Learning in Echocardiography
Ahmed Alaa, Anthony Philippakis, David Sontag
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 |
|---|---|---|---|
| Pre-train on source task instead of training from scratch (ablation) | SegFormer head with ResNet-50 · a0-A4-E/a0-A4-C and a0-A4-C/a0-A2-C | DICE score | Improved |
| Pre-train on source task instead of training from scratch (ablation) | SegFormer head with ResNet-50 · a0-A4-C/a2-A4-C | DICE score | No clear effect |
| Label shift between source and target (a0-A4-C/a1-A4-C, a0-A4-C/a2-A4-C) vs data set or view shift | segmentation heads with ResNet-50 · transfer benchmarks | DICE score: 0.540 vs 0.885 | Hurt |
| Switch head from U-Net to DeepLabV3 | ResNet-50 encoder · a0-A4-C/a2-A4-C | DICE score: 0.540 -> 0.666 | Improved |
| Switch head from U-Net to Linknet | ResNet-50 encoder · a0-A4-C/a1-A4-C | DICE score: 0.729 -> 0.481 | Hurt |
| Switch head from U-Net to U-Net++ | ResNet-50 encoder · a0-A4-E/a0-A4-C | DICE score: 0.909 -> 0.804 | Hurt |
| Switch head from U-Net to PSPNet | ResNet-50 encoder · a0-A4-E/a2-A4-C | DICE score: 0.634 -> 0.707 | Improved |
| Different segmentation head across benchmarks | ResNet-50 encoder · six transfer benchmarks | DICE score | Mixed |
| Use CNN backbones instead of ViT backbones (MiT, PoolFormer) | ETAB backbone evaluation · ETAB benchmark | ETAB score: 0.689 vs 0.653 | Improved |
| Fully finetune ResNet-50 instead of freezing ImageNet-1K weights (ablation) | ResNet-50 · ETAB benchmark | ETAB score: 0.689 -> 0.769 | Improved |
| Fully finetune MiT-B2 instead of freezing ImageNet-1K weights (ablation) | MiT-B2 · ETAB benchmark | ETAB score: 0.653 -> 0.691 | Improved |
| Fully finetune the entire backbone instead of tuning the head only (ablation) | backbone models · ETAB benchmark | ETAB score | No clear effect |
Closest papers by meaning
- EchoXFlow: A Beamspace Echocardiography Dataset for Cardiac Motion, Flow, and Function (NeurIPS 2026 · similarity 0.86)
- Self-supervised Learning of Echocardiographic Video Representations via Online Cluster Distillation (NeurIPS 2025 · similarity 0.86)
- CardioLens: Revealing the Clinical Reality Gap of MLLMs via Multi-Sequence Cardiac MRI Evaluations (NeurIPS 2026 · similarity 0.85)
- EchoWorld: Learning Motion-Aware World Models for Echocardiography Probe Guidance (CVPR 2025 · similarity 0.85)
- ECHOPulse: ECG Controlled Echocardio-gram Video Generation (ICLR 2025 · similarity 0.84)
- BenchX: A Unified Benchmark Framework for Medical Vision-Language Pretraining on Chest X-Rays (NeurIPS 2024 · similarity 0.84)
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.
