MITCriticalData/mimic-iv-echo-jepa-embeddings

MIMIC-IV-Echo V-JEPA2 Embeddings

Pre-computed video embeddings for MIMIC-IV-Echo echocardiography videos, extracted using V-JEPA2 (Meta's self-supervised video encoder).

Access requirement: This dataset includes metadata from MIMIC-IV-Echo (subject IDs, study IDs, timestamps, clinical note references). You must have PhysioNet credentialed access and a signed DUA for MIMIC-IV-Echo before requesting access.

Dataset

Videos	525,328 echocardiography clips
Subjects	~4,800 patients
Studies	~7,200 echo studies
Embedding model	V-JEPA2 ViT-L (300M params)
Embedding dim	1024 (float32)
Format	Sharded Parquet (10 shards, ~2.1 GB total)

Structure

mimic-iv-echo-jepa-embeddings/
└── jepa-l-embeddings/
    ├── train-00000-of-00010.parquet   (p10, 51K rows, 202 MB)
    ├── train-00001-of-00010.parquet   (p11, 52K rows, 205 MB)
    ├── ...
    └── train-00009-of-00010.parquet   (p19, 53K rows, 208 MB)

Each shard corresponds to one MIMIC-IV patient folder (p10-p19).

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]	V-JEPA2 ViT-L	1024-dim video embedding

Metadata is joined from two MIMIC-IV-Echo CSV files 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

Usage

from datasets import load_dataset

ds = load_dataset("MITCriticalData/mimic-iv-echo-jepa-embeddings")
print(ds["train"][0])  # {'subject_id': 10002221, 'embedding': [...], ...}

import pyarrow.parquet as pq

table = pq.read_table("jepa-l-embeddings/")
print(table.num_rows)   # 525328
print(table.schema)

Extraction

Embeddings were extracted using the pipeline described in readme-embeddings.md.

Step	Command
Extract (SLURM)	sbatch extract_slurm.sh vitl
Merge .pt files	python merge_embeddings.py --model vitl
Convert to Parquet	python to_parquet.py --model vitl

Config: L40S GPU, batch=256, 8 DataLoader workers, ~30 min per folder.

Citation

If you use this dataset, please cite both the original MIMIC-IV-Echo dataset and V-JEPA2:

@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}
}