Datasetpublic

ctmedtech/RFMID

🩺 RFMiD — Retinal Fundus Multi-Disease Image Dataset Image: Dataset Samples. The Retinal Fundus Multi-Disease Image Dataset (RFMiD) is designed for multi-disease detection and classification in retinal fundus photographs.It includes 3,200 high-quality color images with 46 labeled retinal disease conditions, curated by expert ophthalmologists from India.This dataset enables development of generalized deep… See the full description on the dataset page: https://huggingface.co/datasets/ctmedtech/RFMID.

sourceHugging Facecc-by-4.0updated 11mo agoView on Hugging Face

1likes1kdownloads

Dataset Card

🩺 RFMiD — Retinal Fundus Multi-Disease Image Dataset

<table align="center"> <tr> <td width="100%" align="center"> <img src="rmimages/MergedFundusImageswith_Caption.jpg" alt="Merged Dataset Samples" style="max-width: 100%; height: auto;"> <br> <p><strong>Image:</strong> Dataset Samples.</p> </td> </tr> </table>

The Retinal Fundus Multi-Disease Image Dataset (RFMiD) is designed for multi-disease detection and classification in retinal fundus photographs. It includes 3,200 high-quality color images with 46 labeled retinal disease conditions, curated by expert ophthalmologists from India. This dataset enables development of generalized deep learning models for comprehensive retinal disease screening.

📘 Overview

Field	Details
Full Name	Retinal Fundus Multi-Disease Image Dataset (RFMiD)
Focus	Multi-label classification of retinal diseases
Condition Types	46 disease classes including diabetic retinopathy, glaucoma, AMD, hypertensive retinopathy, myopia, and others
Collection Site	Ophthalmology centers in Maharashtra, India
Devices Used	TOPCON 3D OCT-2000 (~2144×1424), Kowa VX-10α (~4288×2848), TOPCON TRC-NW300 (~2048×1536)
Field of View (FOV)	~45°–50°
Image Type	Color fundus photographs (JPG, RGB)
Total Images	3,200
Annotations	Expert ophthalmologist-verified, multi-label (each image may contain multiple conditions)
License	CC BY 4.0
Source	MDPI Paper · IEEE Dataport

🗂️ Dataset Structure

The RFMiD dataset includes images and corresponding metadata files organized as follows:

text

RFMiD/
│
├── Images/
│ ├── Training_Set/
│ │ ├── IDRiD_001.jpg
│ │ ├── IDRiD_002.jpg
│ │ └── ...
│ │
│ ├── Validation_Set/
│ │ ├── IDRiD_801.jpg
│ │ ├── IDRiD_802.jpg
│ │ └── ...
│ │
│ └── Test_Set/
│ ├── IDRiD_901.jpg
│ ├── IDRiD_902.jpg
│ └── ...
│
├── Groundtruths/
│ ├── RFMiD_Training_Labels.csv
│ ├── RFMiD_Validation_Labels.csv
│ └── RFMiD_Test_Labels.csv
│
└── Metadata/
└── RFMiD_Clinical_Information.csv

📄 File Description

File / Folder	Description
Images/	Contains all RGB fundus images grouped into train, validation, and test sets
Groundtruths/	CSV files with disease labels for each image ID
Metadata/	Contains additional information like patient age, gender, and diagnostic notes (if available)

🧾 Label Format (CSV Example)

Each row in RFMiD_Training_Labels.csv includes binary indicators (0 or 1) for each of the 46 disease categories:

ImageID	DR	ARMD	MH	DN	MYA	...	HR	Others
0001	1	0	0	1	0	...	0	0
0002	0	0	0	0	0	...	1	0

Total columns: 46 disease labels + 1 ImageID column.

📊 Dataset Composition

Split	Number of Images	Description
Training Set	1,920	Used to train AI models
Validation Set	640	Used to tune hyperparameters
Test Set	640	Held-out evaluation set
Total	3,200	All high-quality fundus images

🧠 Research Applications

Primary Use Cases

—Multi-label retinal disease classification
—Generalized ophthalmic AI screening
—Rare disease detection (long-tail recognition)
—Domain adaptation across imaging devices
—Quality-aware retinal analysis

Recommended Tasks

—Classification: Healthy vs Abnormal
—Multi-label Detection: 46 retinal diseases
—Transfer Learning: Adaptation to real-world clinical data
—Explainability: Visualizing disease localization with Grad-CAM or attention maps

⚙️ Technical Notes

—Input format: RGB fundus images, JPG
—Recommended preprocessing: Center-cropping, illumination correction, resizing to 512×512 or 1024×1024
—Label imbalance: Some diseases have <50 samples; use focal loss or weighted sampling
—Multi-device domain variation: Apply histogram equalization or color normalization

🧩 Quick Summary Table

Dataset	Description (conditions, source, etc.)	Size
RFMiD	Multi-disease retinal fundus dataset with 46 labeled conditions from Indian ophthalmic clinics	3,200 images

📚 Citation

If you use this dataset, please cite:

Pachade, S.; Porwal, P.; Thulkar, D.; Kokare, M.; Deshmukh, G.; Sahasrabuddhe, V.; Giancardo, L.; Quellec, G.; Mériaudeau, F. Retinal Fundus Multi-Disease Image Dataset (RFMiD): A Dataset for Multi-Disease Detection Research. Data 2021, 6(2), 14. DOI: 10.3390/data6020014

🪪 License

This dataset is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license. You may share and adapt the dataset, provided appropriate credit is given.