mimic-ext-cxr-qba

KO | EN

gitlite — search

#hacktoberfest #react #typescript #machine-learning #rust #deep-learning #java #windows #awesome #cli #security #ios #openai #open-source

mimic-ext-cxr-qba

(★ 7)

MIMIC-Ext-CXR-QBA

mimic-ext-cxr-qba 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# CXR-QBA: A Structured, Tagged, and Localized Visual Question Answering Dataset with Full Sentence Answers and Scene Graphs for Chest X-ray Images
This is the official repository of the **MIMIC-Ext-CXR-QBA** dataset (ICLR 2026)

**[[Paper - OpenReview]](https://openreview.net/forum?id=LrmyW9JLYq) | [[ICLR 2026 Poster]](https://iclr.cc/virtual/2026/poster/10009998) | [[Dataset - PhysioNet]](https://physionet.org/content/mimic-ext-cxr-qba/1.0.0/)**

> **A Structured, Tagged, and Localized Visual Question Answering Dataset with Full Sentence Answers and Scene Graphs for Chest X-ray Images** <br>

> **Abstract:** Visual Question Answering (VQA) enables targeted and context-dependent analysis
of medical images, such as chest X-rays (CXRs). However, existing VQA datasets
for CXRs are typically constrained by simplistic and brief answer formats, lacking
localization annotations (e.g., bounding boxes) and structured tags (e.g., region
or radiological finding/disease tags). To address these limitations, we introduce
MIMIC-Ext-CXR-QBA (abbr. CXR-QBA), a large-scale CXR VQA dataset derived
from MIMIC-CXR, comprising 42 million QA-pairs with multi-granular, multi-part
answers, detailed bounding boxes, and structured tags. We automatically generated
our VQA dataset from scene graphs (also made available), which we constructed
using LLM-based information extraction from radiology reports. After automatic
quality assessment, we identified 31M pre-training and 7.5M fine-tuning grade
QA-pairs, providing the largest and most sophisticated VQA dataset for CXRs
to date.

**Figure 1:** Examples of question-answer (QA) pairs for each of our four different types of questions.
For each question for a given chest X-ray, a detailed answer with sentences in the style of free-text
radiology reports is given, supplemented by bounding boxes (for both positive and negative answers),
and a set of tags (e.g. regions, findings, certainty, etc.).

## How to Download the Dataset
The dataset can be downloaded via PhysioNet: [https://physionet.org/content/mimic-ext-cxr-qba/1.0.0/)](https://physionet.org/content/mimic-ext-cxr-qba/1.0.0/). You will need to have a PhysioNet credentialed account and accept the data license.

## How to Load, Filter, and Process the Dataset
We provide utilities for loading the MIMIC-Ext-CXR-QBA dataset (Python utilities) as well as filtering or exporting it (command-line tool).
Follow the instruction at [cxr_qba_utils](cxr_qba_utils/README.md). 
- **Recommended if you want to work with the dataset (scene graphs / QA-pairs) in your own code.**
- **Recommended if you want to export the dataset, e.g. into csv or hdf5 for seq2seq training or into classification/object detection datasets.**

## How to Train and Evaluate our VQA Model on the Dataset
We provide Python utilities and a command-line tool for training and evaluating VQA models on the MIMIC-Ext-CXR-QBA dataset. 
Follow the instruction at [cxr_qba_train](cxr_qba_train/README.md). 
- **Recommended if you want to train our example VQA model on the Structured VQA Task on the MIMIC-Ext-CXR-QBA dataset.**
- **Recommended if you want to use our training code to train your own model (alternatively, you can use our export and loading utilities in your own training code).**
- **Recommended if you want to use our RadStrucVQA Metric** (see [RadStrucVQA](cxr_qba_train/README.md#radstrucvqa-metric) for more details).

## How to Construct the Dataset
We provide a Python-based command-line tool to construct the MIMIC-Ext-CXR-QBA dataset.
**You do not need this tool if you just want to use the final dataset. Download it instead at [https://physionet.org/content/mimic-ext-cxr-qba/1.0.0/)](https://physionet.org/content/mimic-ext-cxr-qba/1.0.0/).
If you want to reconstruct the dataset or understand the construction process the code and instructions can be found at [cxr_qba_extract](cxr_qba_extract/README.md).
- **Recommended if you want to re-create the MIMIC-Ext-CXR-QBA dataset on your own.**
- **Recommended if you want to understand how the dataset was created.**
- **Recommended if you want to apply or modify the construction pipeline to other data sources.**

## More utilities
We also provide a utility library for distributed and robust batch processing using LLMs, e.g. to automatically extract information from large datasets. This was used in our extraction process.
See [llm_processor](llm_processor/README.md) for details.

mimic-ext-cxr-qba

mimic-ext-cxr-qba 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

mimic-ext-cxr-qba

mimic-ext-cxr-qba 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README