---
license: other
license_name: physionet-credentialed-health-data-license-150
license_link: https://physionet.org/content/mimic-iv-echo/view-license/0.1/
extra_gated_heading: "Access to MIMIC-IV-Echo V-JEPA & EchoJEPA Embeddings"
extra_gated_description: >
  This dataset contains metadata derived from MIMIC-IV-Echo, which is a credentialed dataset on PhysioNet.
  To access this dataset, you must have an active PhysioNet credentialed account with signed
  Data Use Agreement (DUA) for MIMIC-IV-Echo.
extra_gated_fields:
  PhysioNet username: text
  I have an active credentialed account on PhysioNet: checkbox
  I have signed the MIMIC-IV-Echo Data Use Agreement: checkbox
  I agree to not redistribute this data: checkbox
  Affiliation: text
tags:
- medical-imaging
- echocardiography
- embeddings
- mimic-iv-echo
- v-jepa2
- video-embeddings
- self-supervised-learning
---

# MIMIC-IV-Echo V-JEPA & EchoJEPA Embeddings

Pre-computed video embeddings for all ~525K [MIMIC-IV-Echo](https://physionet.org/content/mimic-iv-echo/) echocardiography clips using [V-JEPA2](https://github.com/facebookresearch/jepa) / V-JEPA2.1 and EchoJEPA fine-tuned variants.

> **Access requirement:** You must have [PhysioNet credentialed access](https://physionet.org/settings/credentialing/) and a signed DUA for [MIMIC-IV-Echo](https://physionet.org/content/mimic-iv-echo/) before requesting access.

## Models

Folder names map directly to source checkpoints. V-JEPA **2** and V-JEPA **2.1** are distinct — see the Description column. EchoJEPA models are all V-JEPA 2.1 fine-tuned on echocardiography.

<table>
<thead>
<tr>
<th>Status</th>
<th style="white-space:nowrap">Folder</th>
<th>Arch</th>
<th>Dim</th>
<th>Description</th>
<th style="white-space:nowrap">Checkpoint</th>
</tr>
</thead>
<tbody>
<tr>
<td>✅</td>
<td style="white-space:nowrap"><b>vjepa2-vitl</b></td>
<td>ViT-L 300M</td><td>1024</td>
<td>V-JEPA <b>2</b> base model, pretrained on natural video — no echo fine-tuning</td>
<td><code>vitl.pt</code></td>
</tr>
<tr>
<td>✅</td>
<td style="white-space:nowrap"><b>vjepa2-vitl-scratch-pt-210-c25</b></td>
<td>ViT-L 300M</td><td>1024</td>
<td>V-JEPA <b>2</b> architecture, trained from random init on MIMIC-IV-Echo. pt-210 = 210 pretraining epochs; c25 = checkpoint 25</td>
<td><code>vitl-scratch-pt-210-c25.pt</code></td>
</tr>
<tr>
<td>✅</td>
<td style="white-space:nowrap"><b>vjepa2.1-vitl-mimic-pt-117</b></td>
<td>ViT-L 300M</td><td>1024</td>
<td><b>EchoJEPA</b> — ViT-L fine-tuned from V-JEPA <b>2.1</b> on MIMIC-IV-Echo. pt-117 = 117 fine-tuning epochs</td>
<td><code>vjepa21_vitl_mimic_pt117.pt</code></td>
</tr>
<tr>
<td>✅</td>
<td style="white-space:nowrap"><b>vjepa2.1-vitb-mimic-pt-169-c60</b></td>
<td>ViT-B 86M</td><td>768</td>
<td><b>EchoJEPA</b> — ViT-B fine-tuned from V-JEPA <b>2.1</b> on MIMIC-IV-Echo. pt-169 = 169 epochs; c60 = checkpoint 60</td>
<td><code>vjepa2_1_vitb_mimic_pt169_c60.pt</code></td>
</tr>
<tr>
<td>✅</td>
<td style="white-space:nowrap"><b>vjepa2.1-vitl-mimic-pt-100</b></td>
<td>ViT-L 300M</td><td>1024</td>
<td><b>EchoJEPA</b> — ViT-L fine-tuned from V-JEPA <b>2.1</b> on MIMIC-IV-Echo. pt-100 = 100 fine-tuning epochs</td>
<td><code>vjepa21_vitl_mimic_pt100.pt</code></td>
</tr>
<tr>
<td>✅</td>
<td style="white-space:nowrap"><b>vjepa2-vitl-vmix22m-pt-220-c55</b></td>
<td>ViT-L 300M</td><td>1024</td>
<td>V-JEPA <b>2</b> ViT-L fine-tuned on a 22M-clip video mix. pt-220 = 220 epochs; c55 = checkpoint 55</td>
<td><code>vitl-vmix22m-pt220-c55.pt</code></td>
</tr>
</tbody>
</table>

## Dataset

| | |
|---|---|
| **Videos** | 525,328 echocardiography clips |
| **Subjects** | ~4,800 patients |
| **Studies** | ~7,200 echo studies |
| **Embedding dim** | 1024 (ViT-L) / 768 (ViT-B) float32 |
| **Format** | Sharded Parquet (10 shards per model, ~2 GB per model) |

## Structure

```
mimic-iv-echo-jepa-embeddings/
├── vjepa2-vitl/                          ✅ V-JEPA2 base, pretrained on natural video
│   ├── train-00000-of-00010.parquet
│   └── ... (10 shards)
├── vjepa2-vitl-scratch-pt-210-c25/       ✅ V-JEPA2 arch, trained from scratch on echo
│   ├── train-00000-of-00010.parquet
│   └── ... (10 shards)
├── vjepa2.1-vitl-mimic-pt-117/           ✅ fine-tuned 117 epochs
│   ├── train-00000-of-00010.parquet
│   └── ... (10 shards)
├── vjepa2.1-vitb-mimic-pt-169-c60/       ✅ fine-tuned 169 epochs, ViT-B (768-dim)
│   ├── train-00000-of-00010.parquet
│   └── ... (10 shards)
├── vjepa2.1-vitl-mimic-pt-100/           ✅ EchoJEPA, fine-tuned 100 epochs
│   ├── train-00000-of-00010.parquet
│   └── ... (10 shards)
└── vjepa2-vitl-vmix22m-pt-220-c55/       ✅ V-JEPA2, video mix fine-tune (22M-clip video mix)
    ├── train-00000-of-00010.parquet
    └── ... (10 shards)
```

Each shard corresponds to one MIMIC-IV patient folder (p10–p19). Shard index is consistent across all models.

## Columns

| Column | Type | Source | Description |
|--------|------|--------|-------------|
| `subject_id` | int64 | echo-record-list.csv | MIMIC patient ID |
| `study_id` | int64 | echo-record-list.csv | Echo study ID |
| `dicom_id` | str | filename | Original DICOM identifier (e.g. `94106955_0001`) |
| `file_path` | str | embedding key | Relative path to source MP4 |
| `acquisition_datetime` | str | echo-record-list.csv | Per-video acquisition timestamp |
| `study_datetime` | str | echo-study-list.csv | Per-study timestamp |
| `note_id` | str | echo-study-list.csv | Clinical note reference (nullable) |
| `note_seq` | str | echo-study-list.csv | Note sequence number (nullable) |
| `note_charttime` | str | echo-study-list.csv | Note chart time (nullable) |
| `embedding` | list[float32] | encoder | 1024-dim (ViT-L) or 768-dim (ViT-B) video embedding |

Metadata joined from MIMIC-IV-Echo CSVs so each row is self-contained:
- **echo-record-list.csv** (525K rows) — per-video: subject, study, acquisition time
- **echo-study-list.csv** (7K rows) — per-study: study time, clinical notes

## Embedding representation

Each `embedding` column stores a **mean-pooled** video-level descriptor: the encoder outputs a sequence of spatiotemporal patch tokens (`[num_patches, embed_dim]`) and we average across all tokens (`features.mean(dim=1)`) to produce a single fixed-size vector per video.

This keeps storage tractable (~2 GB per model for 525K videos) and works well for classification, retrieval, and most clinical outcome prediction tasks.

> **Note on full-token embeddings:** Retaining the full token sequence (e.g. ~1500 tokens × 1024 dim) instead of mean-pooling would give better performance on tasks that require fine-grained spatiotemporal structure — in particular, **right ventricular (RV) function tasks**, where temporal dynamics across frames carry discriminative information that mean-pooling discards. The trade-off is storage: full-token embeddings for 525K videos would require on the order of several terabytes per model, and dimensionality reduction (PCA or token selection) would be needed before training. Extracting and publishing full-token variants is an open contribution opportunity.

## Usage

```python
from datasets import load_dataset

# V-JEPA2 base model (pretrained on natural video)
ds = load_dataset("MITCriticalData/mimic-iv-echo-jepa-embeddings",
                  data_dir="vjepa2-vitl")

# EchoJEPA — V-JEPA2.1 fine-tuned on MIMIC-IV-Echo
ds = load_dataset("MITCriticalData/mimic-iv-echo-jepa-embeddings",
                  data_dir="vjepa2.1-vitl-mimic-pt-100")

print(ds["train"][0])  # {'subject_id': 10002221, 'embedding': [...], ...}
```

## Extraction Pipeline

From [EchoJEPA-VE](https://github.com/sebasmos/EchoJEPA-VE):

| Step | Command |
|------|---------|
| Extract (SLURM) | `sbatch scripts/extract-embeddings/extract_echo_slurm.sh <model>` |
| Merge .pt files | `python scripts/extract-embeddings/merge_embeddings.py --model <model>` |
| Convert to Parquet | `python scripts/extract-embeddings/to_parquet.py --model <model> --subfolder <vjepa2-or-vjepa2.1-variant>` |

Config: L40S GPU, batch=256, 8 DataLoader workers, ~465 vids/min per GPU.

## Citation

```bibtex
@article{mimic-iv-echo,
  title={MIMIC-IV-Echo: A Large-Scale Echocardiography Dataset},
  note={PhysioNet, https://physionet.org/content/mimic-iv-echo/}
}

@article{bardes2025vjepa2,
  title={Revisiting Feature Prediction for Learning Visual Representations from Video},
  author={Bardes, Adrien and Garrido, Quentin and Ponce, Jean and Chen, Xinlei and Rabbat, Michael and LeCun, Yann and Assran, Mahmoud and Ballas, Nicolas},
  year={2025}
}
```