ramankamran/retina-age-analysis

Retina Age Analysis Dataset

Dataset Description

This dataset contains 9,857 retinal fundus images from 5,393 patients for age prediction tasks.

Dataset Summary

• Task: Age prediction from retinal fundus images
• Images: 9,857 high-quality retinal images
• Patients: 5,393 unique patients
• Age Range: 5-97 years
• Image Format: JPEG
• Average Image Size: ~1 MB

Supported Tasks

1. Regression: Predict continuous age (5-97 years)
2. Classification: Predict age group (5 classes: pediatric, young adult, middle age, senior, elderly)

Data Splits

Split	Images	Patients	Percentage
Train	6,902	3,775	70%
Validation	1,493	809	15%
Test	1,462	809	15%

Note: Split at patient level to prevent data leakage.

Age Distribution

Age Group	Age Range	Count	Percentage
Pediatric	5-17	291	3.0%
Young Adult	18-39	1,447	14.7%
Middle Age	40-59	2,946	29.9%
Senior	60-74	3,484	35.3%
Elderly	75+	1,689	17.1%

Dataset Structure

retina-age-analysis/
├── images/           # 9,857 retinal fundus images
│   ├── img00001.jpg
│   ├── img00002.jpg
│   └── ...
│
└── splits/           # Train/val/test split CSV files
    ├── train.csv     # 6,902 samples
    ├── val.csv       # 1,493 samples
    └── test.csv      # 1,462 samples

Data Fields

Each CSV file contains:

• image_id: Image filename (without extension)
• patient_id: Unique patient identifier
• patient_age: Age in years (target variable for regression)
• age_group_broad: Age category name
• age_group_broad_numeric: Age category index (0-4, target for classification)
• patient_sex: Gender (1=Male, 2=Female)
• exam_eye: Eye examined (1=Right, 2=Left)
• diabetic_retinopathy: DR status (0=No, 1=Yes)
• camera: Camera type used
• Additional clinical features

Usage Example

from datasets import load_dataset
from PIL import Image
import pandas as pd

# Load dataset
dataset = load_dataset("ramankamran/retina-age-analysis")

# Load splits
train_df = pd.read_csv("hf://datasets/ramankamran/retina-age-analysis/splits/train.csv")
val_df = pd.read_csv("hf://datasets/ramankamran/retina-age-analysis/splits/val.csv")
test_df = pd.read_csv("hf://datasets/ramankamran/retina-age-analysis/splits/test.csv")

# Load an image
from huggingface_hub import hf_hub_download
img_path = hf_hub_download(
    repo_id="ramankamran/retina-age-analysis",
    filename="images/img00001.jpg",
    repo_type="dataset"
)
image = Image.open(img_path)

# Get corresponding label
label = train_df[train_df['image_id'] == 'img00001']['patient_age'].values[0]

PyTorch DataLoader

See the training code in the repository for PyTorch DataLoader implementation with:

• Data augmentation (rotation, flip, brightness, contrast)
• ImageNet normalization
• Batch loading

Baseline Results

Regression (Age Prediction):

• MAE: 7-10 years (baseline)
• Target: < 5 years (optimized)

Classification (Age Groups):

• Accuracy: 70-75% (baseline)
• Target: 85-90% (with semi-supervised learning)

License

MIT License

Citation

If you use this dataset, please cite:

@dataset{retina_age_analysis,
  author = {Raman Kamran},
  title = {Retina Age Analysis Dataset},
  year = {2025},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/ramankamran/retina-age-analysis}
}

Dataset Curators

Dataset cleaned and prepared by ramankamran.

Preprocessing

• Removed images with missing age labels (33.5% of original data)
• Removed inadequate quality images (8.9%)
• Verified all image files exist
• Created stratified train/val/test splits
• Patient-level splitting to prevent data leakage

Intended Use

• Medical image analysis research
• Age prediction from retinal images
• Transfer learning for ophthalmology tasks
• Semi-supervised learning experiments

Limitations

• Class imbalance (elderly patients over-represented, pediatric under-represented)
• Single imaging center data
• Requires domain knowledge for clinical interpretation

Additional Information

For training code and examples, see: GitHub Repository