LA-Breast: A Latin American multiparametric breast DCE-MRI dataset with benign and malignant annotations

This dataset was collected from single-institution retrospective examinations from Latin American patients. Patients were anonymized and contains 15 DCE-MR imaging sequences: pre-contrast T1 fat saturated Dynamic (d0), five postcontrast T1 fat saturated dynamics (d1 to d5), T1 with no fat saturation (t1), T2 with no fat saturation (t2), the apparent diffusion coefficient image (ADC) and the diffusion image (Diff), and five post-contrast phase sequences (f1 to f5). All image sequences were obtained and stored using the standard DICOM 3.0 and converted to TIFF format to reduce required storage space. Besides, all images were obtained using multiple 1.5T scanners and all contrast agents were gadolinium-based with dosages between 0.014 and 0.016 l/mol under different acquisition conditions as they were obtained retrospectively. Each patient data was filtered according to their available clinical finding, were benign and malignant lesions were prioritized to ensure at least one relevant clinical finding across all images. Thereupon, the data contains balanced train, test and validation sets in terms of benign/malignant lesions, as well as non-dense/dense tissue. Additionally, annotations per image are provided with lesionlocation (<i>x</i> and <i>y</i> coordinates) BIRADS and tissue density. This data can be used for multiple purposes, including image synthesis, image characterization, lesion classification, tissue segmentation, among others.