Biosignal › VitalDB Arrhythmia Database › vitaldb-arrhythmia-abnormal@0

Abnormal rhythm in 5-second intraoperative ECG (lead II) preview

Data: khdp.net/database/data-search-detail/738/vitaldb-arrhythmia/1.0.0 · CC BY 4.0 · binary · primary metric: auroc · spec bd4a0b735b2c

Follows the technical validation of the dataset paper (Eun et al., Sci Data 2026): a 5-second ECG lead II segment is abnormal if it carries any rhythm label other than normal sinus rhythm, and normal only inside 45 s of uninterrupted normal sinus rhythm without premature beats. Noise is not scored, and test segments with bad signal quality are left out. Patients are split as in every VitalDB task; scores are on the test split, with 95% intervals from a bootstrap over patients. Deviations from the paper are listed in the specification below.

Data

VitalDB Arrhythmia Database (biosignal). License: CC BY 4.0. Source: https://khdp.net/database/data-search-detail/738/vitaldb-arrhythmia/1.0.0

Eun D, Shim K, Lee H, Lim Y, Lim H, Lee H, Lee HC. VitalDB Arrhythmia Database: An Anesthesiologist-Validated Large-Scale Intraoperative Arrhythmia Dataset with Beat and Rhythm Labels (version 1.0.0). PhysioNet. 2026. doi:10.13026/axd6-wm13

Task specification

{
  "bad_signal_quality": {
    "exclude_from": [
      "test"
    ]
  },
  "dataset": "vitaldb-arrhythmia",
  "deviations": [
    "Split: the dataset-level VitalDB subject split (70/10/20 over all VitalDB subjects, shared with other VitalDB tasks) instead of the paper's 8:2 patient-level split of this database.",
    "Segmentation: the paper does not give the segment grid; segments are non-overlapping and start at multiples of 5 s.",
    "Normal context: 'part of a continuous 45-second window entirely annotated as normal sinus rhythm' is read as: the segment lies inside an uninterrupted run of N annotation rows spanning at least 45 s.",
    "Noise is not scored (it is a signal-quality label, not one of the ten rhythm labels); Unclassifiable counts as abnormal (it is one of the ten).",
    "Premature beats are beats labelled S or V.",
    "Bad signal quality: the paper does not say; flagged segments are excluded from the test split, as the dataset group's openecg validation does, and kept with a flag in train and validation.",
    "Segments overlapping annotation gaps over 10 s, segments without an annotated beat, and segments with more than 10% missing ECG are not scored."
  ],
  "input": {
    "duration_s": 5,
    "hz": 500,
    "loader": "vitaldb.load_case(caseid, ['SNUADC/ECG_II'], 1/500) with vitaldb 1.7.2 (open-dataset vital files 1.0.1)",
    "reference": [
      "caseid",
      "start_s",
      "duration_s"
    ],
    "track": "SNUADC/ECG_II"
  },
  "kind": "binary",
  "label": {
    "abnormal": [
      "AFIB/AFL",
      "AVB",
      "Patterned Atrial Ectopy",
      "Patterned Ventricular Ectopy",
      "SND",
      "SVTA",
      "Unclassifiable",
      "VT",
      "WAP/MAT"
    ],
    "normal": [
      "N"
    ],
    "normal_context_s": 45,
    "not_scored": [
      "Noise"
    ],
    "premature_beats": [
      "S",
      "V"
    ]
  },
  "metrics": {
    "bootstrap": {
      "by": "subjectid",
      "n": 1000
    },
    "by_category": "share flagged abnormal per rhythm (sensitivity for abnormal rhythms, false-positive rate for N)",
    "primary": "auroc",
    "secondary": [
      "auprc",
      "brier",
      "accuracy",
      "sensitivity",
      "specificity",
      "f1"
    ],
    "threshold": 0.5
  },
  "preview": true,
  "primary_metric": "auroc",
  "protocol": {
    "annotations": {
      "commit": "59b7722674e76592900cb8131d25337f2dd084b8",
      "release": "vitaldb-arrhythmia 1.0.0 (KHDP; PhysioNet doi:10.13026/axd6-wm13)",
      "repository": "https://github.com/vitaldb/arrdb"
    },
    "bad_signal_quality_precedent": {
      "commit": "0d8610fdac1ec2bae25e2f9674d177c124041638",
      "file": "scripts/validate_vitaldb_arrhythmia.py",
      "repository": "https://github.com/vitaldb/openecg"
    },
    "source": "Eun DI, Shim K, Lee H, Lim Y, Lim H, Lee H, Lee J, Lee HC. VitalDB Arrhythmia Database: an anesthesiologist-validated large-scale intraoperative arrhythmia dataset with beat and rhythm labels. Sci Data. 2026;13. doi:10.1038/s41597-026-07076-8 (section Technical Validation)",
    "task": "Binary classification of 5-second ECG segments as normal or abnormal rhythm. A segment is normal only if it is part of a continuous 45-second window annotated entirely as normal sinus rhythm; normal sinus rhythm segments that contain premature beats are omitted; a segment with any rhythm label other than normal sinus rhythm is abnormal."
  },
  "segments": {
    "grid_s": 5,
    "max_annotation_gap_s": 10,
    "max_ecg_missing_fraction": 0.1,
    "min_beats": 1
  },
  "slug": "vitaldb-arrhythmia-abnormal",
  "split": {
    "by": "subjectid",
    "fractions": [
      0.7,
      0.1,
      0.2
    ],
    "scope": "all VitalDB subjects (dataset-level split)",
    "seed": 20261029
  },
  "title": "Abnormal rhythm in 5-second intraoperative ECG (lead II)",
  "version": 0
}

JSON: /v1/tasks/vitaldb-arrhythmia-abnormal · /v1/leaderboard?task=vitaldb-arrhythmia-abnormal@0