FLARE-MedFM/FLARE-Task5-MLLM-2D

FLARE25 Medical Multimodal Dataset

This repository contains a multimodal medical imaging dataset for FLARE 2025 with question-answer pairs across various medical imaging modalities.

Dataset Structure

The dataset is organized into the following main directories:

• training/: Training data
• validation-public/: Public validation data
• validation-hidden/: Hidden validation data (answer not released)
• testing/: Hidden testing data (not released)

Dataset Statistics

• Total datasets: 19
• Medical imaging modalities: 8
• Task types: 10
• Total images: 50996
• Total questions: 58112
• Data sources: 9

Modalities

The dataset includes the following medical imaging modalities:

Clinical, Dermatology, Endoscopy, Mammography, Microscopy, Retinography, Ultrasound, Xray

Tasks

The dataset supports the following tasks:

Classification, Counting, Detection, Multi-label Classification, Regression, Report_Generation

Dataset Overview

Dataset	Modality	Images	Tasks	Questions	Sources
Dermatology_bcn20000	Dermatology	12413	Classification	3576	https://doi.org/10.6084/m9.figshare.24140028.v1
Xray_IUXRay	Xray	5908	Report_Generation	9742	https://doi.org/10.1093/jamia/ocv080
Ultrasound_iugc	Ultrasound	5125	Classification, Detection, Regression	13302	https://codalab.lisn.upsaclay.fr/competitions/18413
Xray_chestdr	Xray	4848	Classification, Multi-label Classification	4848	https://doi.org/10.6084/m9.figshare.c.6476047.v1
Endoscopy_endo	Endoscopy	3865	Classification	80	https://doi.org/10.6084/m9.figshare.c.6476047.v1
Mammography_CMMD	Mammography	3582	Classification	4493	https://doi.org/10.7937/tcia.eqde-4b16
Xray_periapical	Xray	2317	Classification, Multi-label Classification	4656	Private
Clinical_neojaundice	Clinical	2235	Classification	745	https://doi.org/10.6084/m9.figshare.c.6476047.v1
Microscopy_chromosome	Microscopy	1785	instance_detection	1785	Private
Retinography_retino	Retinography	1392	Classification	1392	https://doi.org/10.6084/m9.figshare.c.6476047.v1
Microscopy_neurips22cell	Microscopy	1100	Counting	1100	N/A
Microscopy_bone_marrow	Microscopy	1045	classification	1045	PRIVATE
Xray_boneresorption	Xray	1004	regression	1004	PRIVATE
Xray_dental	Xray	1001	Classification	5998	Private
Retinography_fundus	Retinography	987	Classification	1974	Private
Ultrasound_BUSI	Ultrasound	780	classification	780	https://doi.org/10.1016/j.dib.2019.104863
Ultrasound_BUS-UCLM	Ultrasound	682	classification	682	https://doi.org/10.1038/s41597-025-04562-3
Ultrasound_BUSI-det	Ultrasound	647	detection	647	https://doi.org/10.1016/j.dib.2019.104863
Ultrasound_BUS-UCLM-det	Ultrasound	263	detection	263	https://doi.org/10.1038/s41597-025-04562-3

Note: The numbers shown in the above table include data from all subsets: training, validation-public, validation-hidden, and testing.

Directory Structure

Each dataset typically follows this structure:

modality/
└── dataset_name/
    ├── images[Tr|Val|Ts]/
    │   └── image_files.png
    └── dataset_questions_[train|val].json

Question Format

Questions are formatted as JSON arrays with the following structure:

[
    {
        "TaskType": "Classification",
        "Modality": "X-ray",
        "ImageName": "imagesTr/image001.png",
        "Question": "What abnormality is visible in this image?",
        "Answer": "Fracture",
        "Split": "train"
    }
]