Dataset Viewer
Auto-converted to Parquet Duplicate
files
list
schema
int64
[ { "bytes": 527492, "name": "shard-00000.npy", "sha256": "1cfc8897ccc92136d5f2920be576eed47dc431465bb2100c071b6fcfc950cc8d" } ]
1

UQ Vital Signs Dataset Preprocessed

A tokenized version of the University of Queensland Vital Signs Dataset (UQVSD), prepared to test a training workflow for temporal, multimodal patient-state models (digital twins). It is small: one shard of about 0.5 MB.

Non-commercial use only. The original dataset is licensed under CC BY-NC 3.0, and so is this derived version. For commercial use, contact the original authors.

Source

This derived dataset is not endorsed by the original authors or the University of Queensland.

Changes from the original

  • Channels kept (8, in this order): HR, SpO2, etCO2, awRR, NBP (Mean), ART (Mean), BIS, MAC. All other columns, including the wall-clock Clock column and alarm text, were dropped.

  • Time steps: 10-second windows on RelativeTimeMilliseconds.

  • Values: the median of a channel's readings in the window, clipped to a fixed clinical range and binned into 15 equal-width bins.

    Channel Range
    HR 30–150
    SpO2 85–100
    etCO2 0–60
    awRR 0–30
    NBP (Mean) 30–130
    ART (Mean) 30–150
    BIS 0–100
    MAC 0–2
  • Missing values: each channel has its own absent token, used when nothing was recorded in the window. Missing values are not imputed.

  • Intermittent NBP: a value appears only in the window where a new cuff reading occurs. The monitor's held value is not carried forward.

  • Tokens: token = channel_index * 16 + slot. Slot 0 is absent and slots 1–15 are value bins, which gives a vocabulary of 128.

  • Rows: windows of 257 tokens (32 steps × 8 channels + 1), stride 16 steps. Rows never cross a case boundary.

  • Split by case (seed 25501039), done before tokenization:

    • train: 24 cases, 1,026 rows (this repository)
    • val: cases 02, 13, 17, 25 (not published here)
    • test: cases 06, 07, 22, 28 (not published here)

The preprocessing code is research/uqvsd.py in the author's research repository.

Files

  • shard-00000.npy: uint16 array of shape [1026, 257]
  • manifest.json: the size and SHA-256 of every shard

Pin experiments to a full commit SHA of this repository, not to main.

Citation

@article{liu2012uqvsd,
  title   = {University of Queensland Vital Signs Dataset: development of an accessible repository of anesthesia patient monitoring data for research},
  author  = {Liu, David and G{\"o}rges, Matthias and Jenkins, Simon A.},
  journal = {Anesthesia \& Analgesia},
  volume  = {114},
  number  = {3},
  pages   = {584--589},
  year    = {2012},
  doi     = {10.1213/ANE.0b013e318241f7c0}
}
Downloads last month
55