Preprocessed Mammogram Images and Metadata for Multi-View Classification

Ahmed elmotaz published this dataset on figshare in 2026 to accompany a paper on mammogram classification. It contains 1,592 preprocessed CBIS-DDSM images and 410 preprocessed INbreast images, along with metadata and model predictions. The images underwent a YOLOv5-based preprocessing pipeline for breast region extraction and contrast enhancement.

Use Cases

• Training multi-view ensemble classifiers based on the provided preprocessed mammogram images.
• Benchmarking model performance using the included misclassified cases and ground truth labels.
• Analyzing clinical metadata such as BI-RADS scores and lesion annotations for correlation studies.
• Developing preprocessing pipelines for medical imaging, informed by the described YOLOv5 and CLAHE methods.

Strengths

• Includes 1,592 preprocessed images from the CBIS-DDSM dataset and 410 from the INbreast dataset.
• Provides a codebook mapping BI-RADS codes to clinical labels and binary classification targets.
• Contains a file detailing class distribution, showing 310 benign and 100 malignant cases in the INbreast subset.
• Documents a specific preprocessing pipeline using YOLOv5 and CLAHE for reproducibility.

Limitations

• Row count for the primary image datasets is unknown, which may limit suitability assessment.
• Column-level documentation for the metadata files is absent; field semantics must be inferred after download.
• Data may reflect bias inherent to the original source datasets from specific medical institutions.

Provenance

Source

Derived from The Cancer Imaging Archive (CBIS-DDSM) and the INbreast dataset by Moreira et al. (2012).

Collection Method

Images were preprocessed using a YOLOv5 model trained on the RSNA Screening Mammography dataset, followed by CLAHE contrast enhancement and resizing.

Freshness

Last updated 2026-05-03 21:15:21.