README.md

---
license: mit
task_categories:
- visual-question-answering
language:
- en
tags:
- medical
pretty_name: PathVQA
size_categories:
- 10K<n<100K
---

# Dataset Card for PathVQA
![image](image.png)

## Dataset Description
PathVQA is a dataset of question-answer pairs on pathology images. The dataset is intended to be used for training and testing 
Medical Visual Question Answering (VQA) systems. The questions contained in the dataset are similar to those in the American 
Board of Pathology (ABP) test. The dataset includes both open-ended questions and binary "yes/no" questions. The dataset is
built from two publicly-available pathology textbooks: "Textbook of Pathology" and "Basic Pathology", and a publicly-available 
digital library: "Pathology Education Informational Resource" (PEIR). The copyrights of images and captions belong to the 
publishers and authors of these two books, and the owners of the PEIR digital library.<br> 

**Repository:** [PathVQA Official GitHub Repository](https://github.com/UCSD-AI4H/PathVQA)<br>
**Paper:** [PathVQA: 30000+ Questions for Medical Visual Question Answering](https://arxiv.org/abs/2003.10286)<br>
**Leaderboard:** [Papers with Code Leaderboard](https://paperswithcode.com/sota/medical-visual-question-answering-on-pathvqa)

### Dataset Summary
The data was obtained from the updated Google Drive link shared by the authors on Feb 15, 2023, 
see the [commit](https://github.com/UCSD-AI4H/PathVQA/commit/117e7f4ef88a0e65b0e7f37b98a73d6237a3ceab)
in the GitHub repository. This version of the dataset contains a total of 5,004 images and 32,795 question-answer pairs. 
Out of the 5,004 images, 4,289 images are referenced by a question-answer pair, while 715 images are not used.
There are a few image-question-answer triplets which occur more than once in the same split (training, validation, test). 
After dropping the duplicate image-question-answer triplets, the dataset contains 32,632 question-answer pairs on 4,289 images.

#### Supported Tasks and Leaderboards
This dataset has an active leaderboard which can be found on [Papers with Code](https://paperswithcode.com/sota/medical-visual-question-answering-on-pathvqa) 
and ranks models based on three metrics: "Yes/No Accuracy", "Free-form accuracy" and "Overall accuracy". "Yes/No Accuracy" is
the accuracy of a model's generated answers for the subset of binary "yes/no" questions. "Free-form accuracy" is the accuracy 
of a model's generated answers for the subset of open-ended questions. "Overall accuracy" is the accuracy of a model's generated 
answers across all questions.

#### Languages
The question-answer pairs are in English.

## Dataset Structure

### Data Instances
Each instance consists of an image-question-answer triplet.
```
{
  'image': {'bytes': b'\xff\xd8\xff\xee\x00\x0eAdobe\x00d..., 'path': None},
  'question': 'What does immunoperoxidase staining reveal that marks positively with anti-CD4 antibodies?',
  'answer': 'a predominantly perivascular cellular infiltrate'
}
```
### Data Fields
- `'image'`: the image referenced by the question-answer pair, as a byte array. 
- `'question'`: the text of the question about the image.
- `'answer'`: the text of the expected answer.

### Data Splits
The dataset is split into training, validation and test. The split is provided directly by the authors.

## Additional Information

### Licensing Information
The authors have released the dataset under the [MIT License](https://github.com/UCSD-AI4H/PathVQA/blob/master/LICENSE).

### Citation Information
```
@article{he2020pathvqa,
    title={PathVQA: 30000+ Questions for Medical Visual Question Answering},
    author={He, Xuehai and Zhang, Yichen and Mou, Luntian and Xing, Eric and Xie, Pengtao},
    journal={arXiv preprint arXiv:2003.10286},
    year={2020}
}
```