Annotated Mammography Dataset for Breast Cancer Detection

This dataset is derived from The mini-MIAS database of mammograms [J. Suckling et al., 1994], a well-known collection of digitized mammographic images. The original dataset, available at MIAS Database (http://peipa.essex.ac.uk/info/mias.html), consists of 322 images.

For this dataset, the images were manually annotated using Roboflow, with two annotation classes:
1) BC (Breast Cancer)
2) Inconclusive

i) Preprocessing & Augmentation:
To standardize the dataset, images were preprocessed using auto-orientation and resized to 640×640 pixels. Augmentation was applied to generate three outputs per training example, with transformations including:
1) Horizontal flip
2) Exposure adjustment (-15% to +15%)

ii) Dataset Split
The final dataset consists of:
917 images for training
10 images for validation
5 images for testing

iii) Formats & Compatibility
The dataset is available in four formats to support 23 different models:
1) JSON
2) XML
3) TXT
4) CSV

Disclaimer
The annotations in this dataset were made solely based on visual cues and were not conducted by medical professionals. As a result, they should not be considered clinically accurate or used for medical diagnosis. This dataset is intended for research and educational purposes only.

Reference: J Suckling et al (1994): The Mammographic Image Analysis Society Digital Mammogram Database Exerpta Medica. International Congress Series 1069 pp375-378.