MEDEC-MS

HuggingFace upload of the MEDEC-MS dataset with fixes. If used, please cite the original authors using the citation below.

Dataset Details

Dataset Description

The dataset contains three splits: train, validation, test.

Dataset Sources

• Repository: https://github.com/abachaa/MEDEC
• Paper: https://arxiv.org/pdf/2412.19260

Corrections to the original dataset:

• Added pipes to delimit sentence IDs and sentences in the sentences column, as specified in Prompt #1.

• Removed non-standard characters.

  For example, the original ms-train-19 text column had

  ...“something crawling on my hand.” The...
  

  which is corrected in this dataset to be

  ..."something crawling on my hand." The...
  

• Fixed many broken sentences in the original dataset.

  For example, the original ms-test-15 row had a malformed sentences column:

  5 Her temperature is 37.5 C 
  6 (99.5 F ); pulse is 75/min; respiratory rate is 13/min, and blood pressure is
  7 115/70 mm
  8 Hg.
  9 A maculopapular rash is observed over the trunk and limbs.
  

  which is corrected in this dataset to be

  5 | Her temperature is 37.5C (99.5 F); pulse is 75/min; respiratory rate is 13/min, and blood pressure is 115/70 mm Hg.
  6 | A maculopapular rash is observed over the trunk and limbs.
  

• Ensured that the error_sentence and corrected_sentence column contain the full sentences.

  For example, ms-test-551 originally had

  error_sentence: "Prazosin is prescribed as his temperature is 99.3 F (37.4 C), blood pressure is 165/118"
  corrected_sentence: "Phenoxybenzamine is prescribed as his temperature is 99.3 F (37.4 C), blood pressure is 165/118"
  

  with sentences

  0 A 45-year-old-man presents to the physician with complaints of intermittent episodes of severe headaches and palpitations.
  1 During these episodes, he notices that he sweats profusely and becomes pale in complexion.
  2 He describes the episodes as coming and going within the past 2 months.
  3 Prazosin is prescribed as his temperature is 99.3 F (37.4 C), blood pressure is 165/118
  4 mmHg, pulse is
  5 126/min, respirations are 18/min, and oxygen saturation is 90% on room air.
  

  which is corrected in this dataset to include the full sentences

  error_sentence: "Prazosin is prescribed as his temperature is 99.3 F (37.4 C), blood pressure is 165/118 mmHg, pulse is 126/min, respirations are 18/min, and oxygen saturation is 90% on room air."
  corrected_sentence: "Phenoxybenzamine is prescribed as his temperature is 99.3 F (37.4 C), blood pressure is 165/118 mmHg, pulse is 126/min, respirations are 18/min, and oxygen saturation is 90% on room air."
  

  and

  0 | A 45-year-old-man presents to the physician with complaints of intermittent episodes of severe headaches and palpitations.
  1 | During these episodes, he notices that he sweats profusely and becomes pale in complexion.
  2 | He describes the episodes as coming and going within the past 2 months.
  3 | Prazosin is prescribed as his temperature is 99.3 F (37.4 C), blood pressure is 165/118 mmHg, pulse is 126/min, respirations are 18/min, and oxygen saturation is 90% on room air.
  

  As you can see, the original sentences column had incorrect sentence splits, which the error_sentence and corrected_sentence at least conformed to. But if you want to specifically test for sentence extraction, especially in the case of raw, unmarked text, we feel it's important to respect sentence boundaries.

Direct Use

import json
from datasets import load_dataset

if __name__ == "__main__":
    # load all data
    dataset = load_dataset("mkieffer/MEDEC")

    # load only train split
    dataset_train = load_dataset("mkieffer/MEDEC", split="train")

    # load only test split
    dataset_test = load_dataset("mkieffer/MEDEC", split="test")

    # load only validation split
    dataset_val = load_dataset("mkieffer/MEDEC", split="validation")

    print("\nfull dataset:\n", dataset)
    print("\ntrain split:\n", dataset_train)
    print("\test split:\n", dataset_test)
    print("\validation split:\n", dataset_val)

    print("\ntrain sample:\n", json.dumps(dataset_train[0], indent=2))
    print("\test sample:\n", json.dumps(dataset_test[0], indent=2))
    print("\validation sample:\n", json.dumps(dataset_val[0], indent=2))

Citation

  @inproceedings{medec,
    author       = {Asma {Ben Abacha} and Wen{-}wai Yim and Yujuan Fu and Zhaoyi Sun and Meliha Yetisgen and Fei Xia and Thomas Lin}, 
    title        = {MEDEC: A Benchmark for Medical Error Detection and Correction in Clinical Notes}, 
    booktitle    = {Findings of the Association for Computational Linguistics, ACL 2025, Vienna, Austria, July 27 - August 1, 2025}, 
    pages        = {22539--22550}, 
    publisher    = {Association for Computational Linguistics}, 
    year         = {2025}, 
    url          = {https://aclanthology.org/2025.findings-acl.1159/} 
    }