File size: 4,912 Bytes
d2f36ab
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e201451
 
 
 
 
 
fd7e84c
e201451
 
 
 
 
 
211dde0
e201451
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
---
configs:
- config_name: virtassist
  default: true
  data_files:
  - split: train
    path: virtassist/train.parquet
  - split: valid
    path: virtassist/valid.parquet
  - split: test1
    path: virtassist/test1.parquet
  - split: test2
    path: virtassist/test2.parquet
  - split: test3
    path: virtassist/test3.parquet
- config_name: aci
  data_files:
  - split: train
    path: aci/train.parquet
  - split: valid
    path: aci/valid.parquet
  - split: test1
    path: aci/test1.parquet
  - split: test2
    path: aci/test2.parquet
  - split: test3
    path: aci/test3.parquet
- config_name: virtscribe
  data_files:
  - split: train
    path: virtscribe/train.parquet
  - split: valid
    path: virtscribe/valid.parquet
  - split: test1
    path: virtscribe/test1.parquet
  - split: test2
    path: virtscribe/test2.parquet
  - split: test3
    path: virtscribe/test3.parquet
license: cc-by-4.0
task_categories:
- text-generation
tags:
- medical
- clinical
- dialogue
pretty_name: ACI-Bench
---

# ACI-Bench


HuggingFace upload of ACI-Bench, which evaluates a model's ability to convert clinical dialogue into structured clinical notes. This repo contains only the core benchmarking data. The full dataset is available on HuggingFace [here](https://huggingface.co/datasets/mkieffer/ACI-Bench-MedARC). If used, please cite the original authors using the citation below.



## Dataset Details

|                        | train | valid | test1 | test2 | test3 |
| ---------------------- | ----: | ----: | ----: | ----: | ----: |
| **aci**                |    35 |    11 |    22 |    22 |    22 |
| **virtassist**         |    20 |     5 |    10 |    10 |    10 |
| **virtscribe**         |    12 |     4 |     8 |     8 |     8 |
| **total**  |    67 |    20 |    40 |    40 |    40 |


### 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

### Dataset Sources 

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


### Direct Use

```python
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", name="virtassist")

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

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

    # ---------------------------------------------------------------------
    # 4) Load ALL subsets (virtassist, virtscribe, aci), all splits each
    # ---------------------------------------------------------------------
    all_subsets = {subset: load_dataset("mkieffer/ACI-Bench", 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", name="aci", split=["train", "test1+test2+test3"])
    
    # load each split separately
    aci_all_test_separate = load_dataset("mkieffer/ACI-Bench", 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}
}
```