Toward general AI harmonization with real mammogram imaging breast cancer dataset

Abstract Objectives This dataset was specifically collected to support the development and evaluation of computer-aided diagnosis (CAD) systems for breast cancer detection. The dataset, Ma’amon Diagnostic Center Mammogram Imaging for Breast Cancer (MDCMI-BC), was created to provide high-quality mammogram images with expert annotations to improve AI-based diagnostic tools. This data is intended to aid in the validation of CAD systems, ensuring they can more effectively detect breast cancer by integrating with other available multimodal datasets. Data description The MDCMI-BC dataset contains 133 mammogram cases with 266 images, categorized by the BI-RADS (Breast Imaging-Reporting and Data System) classification. These include 100 normal, 66 benign, and 100 malignant cases, with images captured from both left and right sides using craniocaudal (CC) and mediolateral oblique (MLO) views. The dataset includes detailed annotations made by expert radiologists, which provide ground truth (GT) for training and validating CAD systems. The images are converted from DICOM format to PNG, de-identified to ensure patient privacy, and made available for public access at the Mendeley repository. The data will assist in refining diagnostic algorithms and fostering AI technologies in breast cancer detection.