Datasets:
language:
- en
license: cc-by-4.0
multilinguality:
- monolingual
pretty_name: Nirschl et al 2018
tags:
- H&E
- cardiac pathology
- heart
- heart disease
- heart failure
- histology
- pathology
task_categories:
- image-classification
Dataset Card for nirschl_et_al_2018
This is a dataset card for nirschl_et_al_2018 dataset, which has been used under the CC BY 4.0 license. The original data have been updated, extended, and incorporated into the Biomedical Reasoning And image Understanding for Robust AI agents (BRAVURA) benchmark.
Table of Contents
- Table of Contents
- Dataset Description
- Dataset Structure
- Dataset Creation
- Considerations for Using the Data
- Additional Information
Dataset Description
The original nirschl_et_al_2018 has been cleaned, updated, and extended to include additional metadata and then converted to a Hugging Face dataset by jnirschl and lozanoe. This is part of the Biomedical Reasoning And image Understanding for Robust AI agents (BRAVURA) benchmark. If you use this updated and extended dataset in your research, please cite both the original paper and the BRAVURA benchmark paper.
Dataset Summary
| Dataset Name | Nirschl et al 2018 |
|---|---|
| Dataset description | Classification of clinical chronic heart failure from cardiac histopathology images. |
| Tasks | multi_class |
| Languages | en |
| Homepage | |
| Paper | Paper |
| Leaderboard | Leaderboard |
| Dataset curator | jnirschl |
| License | CC-BY-4.0 |
| Last updated | 2024-04-17 20:38:57 |
| Version | 0.1.0 |
| Comment |
Dataset Statistics
| Missing | Overall | Test set | Train | Validation | ||
|---|---|---|---|---|---|---|
| n | 2299 | 1155 | 770 | 374 | ||
| Age (yrs), median [Q1,Q3] | 22 | 58.0 [48.0,63.0] | 57.0 [48.0,62.0] | 58.0 [51.0,63.0] | 58.0 [52.0,62.0] | |
| Sex, n (%) | Female | 11 | 825 (36.1) | 462 (40.0) | 231 (30.0) | 132 (36.4) |
| Male | 1463 (63.9) | 693 (60.0) | 539 (70.0) | 231 (63.6) | ||
| Institution, n (%) | UPenn | 0 | 2299 (100.0) | 1155 (100.0) | 770 (100.0) | 374 (100.0) |
| Label, n (%) | Chronic heart failure | 0 | 1034 (45.0) | 517 (44.8) | 352 (45.7) | 165 (44.1) |
| Heart tissue pathology | 22 (1.0) | 22 (1.9) | ||||
| Not chronic heart failure | 1243 (54.1) | 616 (53.3) | 418 (54.3) | 209 (55.9) | ||
| Domain, n (%) | Pathology | 0 | 2299 (100.0) | 1155 (100.0) | 770 (100.0) | 374 (100.0) |
| Subdomain, n (%) | Cardiovascular pathology | 0 | 2299 (100.0) | 1155 (100.0) | 770 (100.0) | 374 (100.0) |
| Stain, n (%) | H&E | 0 | 2299 (100.0) | 1155 (100.0) | 770 (100.0) | 374 (100.0) |
| Modality, n (%) | Light microscopy | 0 | 2299 (100.0) | 1155 (100.0) | 770 (100.0) | 374 (100.0) |
| Submodality, n (%) | Brightfield | 0 | 2299 (100.0) | 1155 (100.0) | 770 (100.0) | 374 (100.0) |
| Tasks, n (%) | multi_class | 0 | 2299 (100.0) | 1155 (100.0) | 770 (100.0) | 374 (100.0) |
[More Information Needed]
Supported Tasks and Leaderboards
multi_class [More Information Needed]
Languages
en
Dataset Structure
Data Instances
An example of 'train' looks as follows.
{
"image_id": d54bb7ec-284f-4218-a47d-af87bb371de5,
"image": np.array([250, 250, 3], dtype="uint8"),
"label": datasets.ClassLabel(names={'chronic heart failure': 0, 'heart tissue pathology': 1, 'not chronic heart failure': 2},
num_classes=3),
"label_name": datasets.Value("string"),
"domain": pathology,
"subdomain": cardiovascular pathology,
"modality": light microscopy,
"submodality": brightfield microscopy,
"stain": H&E,
"microns_per_pixel": 2.0,
...
}
[More Information Needed]
Data Fields
[More Information Needed]
{
"image_id": datasets.Value("string"),
"image": datasets.Array3D(shape=[250, 250, 3], dtype="uint8"),
"label": datasets.ClassLabel(names={'chronic heart failure': 0, 'heart tissue pathology': 1, 'not chronic heart failure': 2},
num_classes=3),
"label_name": datasets.Value("string"),
"domain": datasets.Value("string"),
"subdomain": datasets.Value("string"),
"modality": datasets.Value("string"),
"submodality": datasets.Value("string"),
"stain": datasets.Value("string"),
"microns_per_pixel": datasets.Value("float32"),
...
}
Data Splits
[More Information Needed]
| Split | # Instances |
|---|---|
| Train | 770 |
| Dev | 374 |
| Test | 1155 |
| Total | 1155 |
LICENSE
Dataset Creation
Curation Rationale
There is a need for well curated and annotated biomedical datasets to train and evaluate biomedical machine learning models. The original dataset was cleaned, updated, and extended to include additional metadata. The dataset was then converted to a Hugging Face dataset by jnirschl and is incorporated into the Biomedical Reasoning And image Understanding for Robust AI agents (BRAVURA) benchmark. If you use this extended dataset in your research, please cite both the original paper and the BRAVURA benchmark paper.
Each instance has been assigned a UUID and the original identifiers and filenames have been retained in the metadata to ensure that the data can be traced back to the original source if needed.
[More Information Needed]
Source Data
Initial Data Collection and Normalization
[More Information Needed]
Who are the source language producers?
[More Information Needed]
Annotations
Annotation process
[More Information Needed]
Who are the annotators?
[More Information Needed]
Personal and Sensitive Information
The original dataset authors have anonymized the data and removed any personal or sensitive information prior to making the dataset public. The original dataset authors obtained appropriate institutional review board (IRB) approval for the data collection process, and users are referred to the original data sources for more information on the ethical considerations and data collection process.
[More Information Needed]
Considerations for Using the Data
Social Impact of Dataset
[More Information Needed]
Discussion of Biases
[More Information Needed]
Other Known Limitations
[More Information Needed]
Additional Information
Dataset Curators
Jeff Nirschl Alejandro Lozano
[More Information Needed]
Licensing Information
The datasets were curated and extended while adhering to the original copyright and licensing rules. We specifically avoided materials that restricted or prohibited copying, adapting, remixing, redistributing, or otherwise using the data for research purposes. We recommend users refer to the original data sources' terms of use and licensing information. In case of any concerns or issues, please contact us at jnirschl Should you encounter any oversight or data samples potentially breaching the copyright or licensing regulations of any site, we encourage you to notify us.
[More Information Needed]
Citation Information
The original authors of nirschl_et_al_2018 have requested that the dataset be cited as follows:
@ARTICLE{Nirschl2018-pc,
title = "A deep-learning classifier identifies patients with clinical
heart failure using whole-slide images of {H&E} tissue",
author = "Nirschl, Jeffrey J and Janowczyk, Andrew and Peyster, Eliot G and
Frank, Renee and Margulies, Kenneth B and Feldman, Michael D and
Madabhushi, Anant",
journal = "PLoS One",
volume = 13,
number = 4,
pages = "e0192726",
month = apr,
year = 2018,
language = "en"
}
The extended nirschl_et_al_2018 dataset is part of the Biomedical Reasoning And image Understanding for Robust AI agents (BRAVURA) benchmark. If you use this extended dataset in your research, please cite both the original paper and the BRAVURA benchmark paper.
@article{TODO,
}
[More Information Needed]
Contributions
Thanks to @jnirschl for adding this dataset.