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

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Biomedical Data Science

Background:

• Retrospective collection of 15 DCE-MR imaging sequences from Latin American patients.
• Anonymized patient data includes pre-contrast, post-contrast, T1, T2, ADC, diffusion, and post-contrast phase sequences.
• Utilized multiple 1.5T scanners and gadolinium-based contrast agents under varying acquisition conditions.

Purpose of the Study:

• To present a comprehensive dataset for advancing AI in medical imaging.
• To facilitate research in image synthesis, characterization, and lesion classification.
• To provide a balanced dataset for training and validating machine learning models.

Main Methods:

• Data collected retrospectively from a single institution, anonymized.
• Standard DICOM 3.0 format images converted to TIFF for storage efficiency.
• Patient data filtered for relevant clinical findings, prioritizing benign and malignant lesions.
• Ensured balanced train, test, and validation sets for lesion types and tissue density.
• Provided annotations including lesion location (x, y coordinates), BIRADS, and tissue density.

Main Results:

• A dataset comprising 15 distinct MRI sequences per patient.
• Balanced representation of benign/malignant lesions and non-dense/dense tissues.
• Detailed annotations available for each image, including lesion characteristics.
• Data suitable for diverse AI applications such as image synthesis and segmentation.

Conclusions:

• The presented dataset is a valuable resource for AI-driven medical image analysis.
• Facilitates the development and validation of algorithms for lesion detection and classification.
• Enables further research into advanced imaging techniques and diagnostic tools.