mkieffer/ACI-Bench-MedARC

ACI-Bench

HuggingFace upload of ACI-Bench, which evaluates a model's ability to convert clinical dialogue into structured clinical notes. This dataset includes the benchmark itself, as well as data from ablation studies testing different transcription methods. If used, please cite the original authors using the citation below.

Dataset Details

subset	transcript_version	train	valid	test1	test2	test3	Total
aci	asr	35	11	22	22	22	112
aci	asrcorr	35	11	22	22	22	112
aci	humantrans	0	0	0	0	0	0
virtassist	asr	0	0	0	0	0	0
virtassist	asrcorr	0	0	0	0	0	0
virtassist	humantrans	20	5	10	10	10	55
virtscribe	asr	12	4	8	8	8	40
virtscribe	asrcorr	0	0	0	0	0	0
virtscribe	humantrans	12	4	8	8	8	40
ALL	ALL	114	35	70	70	70	359

Dataset Description

The dataset consists of different subsets capturing different clinical workflows

1. ambient clinical intelligence (aci): doctor-patient dialogue
2. virtual assistant (virtassist): doctor-patient dialogue with queues to trigger Dragon Copilot, e.g., "hey, dragon. show me the chest x-ray"
3. virtual scribe (virtscribe): doctor-patient dialogue with a short dictation from the doctor about the patient at the very beginning

There are three different transcription versions:

1. asr: machine-transcribed
2. asrcorr: human corrections to asr, for example: "nonsmile" in D2N081 --> "non-small" in ACI006
3. humantrans: transcribed by a human

The subsets have the following transcription versions

1. aci: asr and asrcorr
2. virtassist: humantrans only
3. virtscribe: asr and humantrans

Dataset Sources

• GitHub: https://github.com/wyim/aci-bench
• Paper: https://www.nature.com/articles/s41597-023-02487-3

Direct Use

from datasets import load_dataset

SUBSETS = ["virtassist", "virtscribe", "aci"]
SPLITS = ["train", "valid", "test1", "test2", "test3"]

if __name__ == "__main__":
    # ---------------------------------------------------------------------
    # 1) Load ONE subset (config) with ALL splits
    # ---------------------------------------------------------------------
    virtassist_all = load_dataset("mkieffer/ACI-Bench-MedARC", name="virtassist")

    # ---------------------------------------------------------------------
    # 2) Load ONE subset (config) with ONE split
    # ---------------------------------------------------------------------
    virtassist_train = load_dataset("mkieffer/ACI-Bench-MedARC", name="virtassist", split="train")

    # ---------------------------------------------------------------------
    # 3) Load TWO subsets (configs), all splits for each
    # ---------------------------------------------------------------------
    two_subsets = {
        "virtassist": load_dataset("mkieffer/ACI-Bench-MedARC", name="virtassist"),
        "aci": load_dataset("mkieffer/ACI-Bench-MedARC", name="aci"),
    }

    # ---------------------------------------------------------------------
    # 4) Load ALL subsets (virtassist, virtscribe, aci), all splits each
    # ---------------------------------------------------------------------
    all_subsets = {subset: load_dataset("mkieffer/ACI-Bench-MedARC", name=subset) for subset in SUBSETS}
    aci_all = all_subsets["aci"]  # DatasetDict
    aci_train = aci_all["train"]  # Dataset
    aci_valid = aci_all["valid"]

    # ---------------------------------------------------------------------
    # 5) Load multiple splits at once
    # ---------------------------------------------------------------------
    # load each split, concatenated
    aci_all_test_concat = load_dataset("mkieffer/ACI-Bench-MedARC", name="aci", split=["train", "test1+test2+test3"])
    
    # load each split separately
    aci_all_test_separate = load_dataset("mkieffer/ACI-Bench-MedARC", name="aci", split=["train", "test1", "test2", "test3"])

Citation

@article{aci-bench,
  author = {Wen{-}wai Yim and
                Yujuan Fu and
                Asma {Ben Abacha} and
                Neal Snider and Thomas Lin and Meliha Yetisgen},
  title = {ACI-BENCH: a Novel Ambient Clinical Intelligence Dataset for Benchmarking Automatic Visit Note Generation},
  journal = {Nature Scientific Data},
  year = {2023}
}