TN-Mammo: A Multi-view Mammography Dataset for Breast Density Classification

Breast cancer is one of the most common types of cancer among women, leading
to a growing and essential need for early and precise detection. A variety of
machine learning techniques have been demonstrating great promise in improving
diagnostic accuracy integrated with digital mammography, which remains the
gold standard screening technique in the detection of early breast cancer.
Since the performance of machine learning models generally depends on the
quality of training data, it is essential to build a high-quality mammogram
dataset. Although there exists a number of public datasets, there are still
significant limitations due to both their quality and quantity. We present an
extensive Vietnamese mammogram dataset with breast density annotations in an
effort to bridge the gap between availability and practical usability. Our
mammogram dataset named TNMammo, consists of bilateral craniocaudal (CC) and
mediolateral oblique (MLO) views for 676 subjects, each with paired left and
right breast views. The breast density and orientation for each case have been
independently assessed by two radiologists in a double-blind manner, ensuring
consistency and reliability in the annotations. Each case is evaluated based
on breast density, categorized into four levels: A, B, C, and D.