Nj-1111/Glaucoma_Dataset

Glaucoma Dataset

Dataset Summary

The Glaucoma Dataset is a comprehensive collection of retinal fundus images designed for the automated detection and classification of glaucoma. Containing between 10,000 and 100,000 high-quality images, this dataset aims to support the development and evaluation of machine learning and deep learning models in the field of ophthalmic medical imaging.

The dataset is organized using the standard imagefolder format, making it highly accessible for immediate integration with standard computer vision pipelines.

Supported Tasks and Leaderboards

• image-classification: The primary task is to classify images into different categories (e.g., Glaucomatous vs. Healthy/Normal).

Languages

The metadata and descriptive annotations are provided in English.

---

Dataset Structure

Data Instances

Each data point represents a single retinal fundus image of a patient. A typical instance includes:

• image: A PIL Image object containing the retinal scan.
• label: An integer classification label (e.g., 0 for Normal, 1 for Glaucoma). (Note: Update labels based on your exact classes)

Data Fields

• image: A highly-detailed color image of the retina.
• label: The diagnostic class associated with the image.

Data Splits

The dataset is divided into standard splits for machine learning workflows. (Note: Update these numbers with your exact split counts)

Split	Number of Images
Train	12,000
Validation	770
Test	770

---

Dataset Creation

Curation Rationale

Early detection of Glaucoma is critical in preventing irreversible vision loss. This dataset was curated to provide researchers with a robust, large-scale resource to train reliable diagnostic models, overcoming the limitations of smaller, fragmented medical datasets.

Source Data

Initial Data Collection and Normalization

• Source: [Describe where the images came from: e.g., specific hospitals, public clinics, or combined from other public sources.]
• Equipment: [Describe the fundus cameras or imaging equipment used, if known.]
• Preprocessing: [Mention any cropping, resizing, or lighting normalization applied to the raw images.]

Annotations

• Annotation process: [Describe how the images were labeled. e.g., "Labels were assigned by a panel of board-certified ophthalmologists..."]
• Who are the annotators? [e.g., Medical professionals, automated extraction from patient records, etc.]

Personal and Sensitive Information

All images have been rigorously anonymized to remove personally identifiable information (PII). No patient names, birthdates, or direct identifiers are included in this dataset, complying with standard medical data sharing privacy protocols.

---

Usage

You can easily load this dataset using the Hugging Face datasets library. Since it is structured as an imagefolder, the labels are automatically inferred from the directory names.

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("Nj-1111/Glaucoma_Dataset")

# Access the first image in the training set
example = dataset["train"][0]
image = example["image"]
label = example["label"]

# Display the image and label
image.show()
print(f"Label: {label}")