VinDr-Mammo: A large-scale benchmark dataset for computer- aided diagnosis in full-feld digital mammography

Abstract

Mammography, or breast X-ray imaging, is the most widely used imaging modality to detect cancer and other breast diseases. Recent studies have shown that deep learning-based computer-assisted detection and diagnosis (CADe/x) tools have been developed to support physicians and improve the accuracy of interpreting mammography. A number of large-scale mammography datasets from different populations with various associated annotations and clinical data have been introduced to study the potential of learning-based methods in the field of breast radiology. With the aim to develop more robust and more interpretable support systems in breast imaging, we introduce VinDr-Mammo, a Vietnamese dataset of digital mammography with breast-level assessment and extensive lesion-level annotations, enhancing the diversity of the publicly available mammography data. The dataset consists of 5,000 mammography exams, each of which has four standard views and is double read with disagreement (if any) being resolved by arbitration. The purpose of this dataset is to assess Breast Imaging Reporting and Data System (BI-RADS) and breast density at the individual breast level. In addition, the dataset also provides the category, location, and BI-RADS assessment of non-benign findings. We make VinDr-Mammo publicly available as a new imaging resource to promote advances in developing CADe/x tools for mammography interpretation.

Background & Summary Breast Cancer is among the most prevalent cancers and accounts for the largest portion of cancer deaths, with an estimated 2.2 million new cases in 2020. Treatment is most successful when breast cancer is at its early stage. Biennial screening can reduce breast cancer mortality rate by 30%. Among standard imaging examinations for breast cancer diagnosis, namely mammography, ultrasound, digital breast tomosynthesis, and magnetic resonance, mammography is the recommended modality for cancer screening. Interpreting mammography for breast cancer screening is a challenging task. The recall rate of mammogram screening is around 11% with a sensitivity of 86.9%, while the cancer detection rate is 5.1 per 1,000 screens. It means that a large portion of cases called back for further examinations eventually result in non-cancer. Improving cancer screening results may help reduce the cost of follow-up examinations and unnecessary mental burdens on patients. With recent advancements of learning-based algorithms for image analysis, several works have adapted deep learning networks for mammography interpretation and showed potential to use in clinical practices. In retrospective settings, the CAD tool as an independent reader can achieve a performance comparable to an average mammographer. It can be leveraged as a decision support tool that helps enhance radiologists’ cancer detection with the reading time being unchanged. In another human-machine hybrid setting, where radiologists and machine-learning algorithm independently estimate the malignancy of the lesions, the linear combination of human and machine prediction shows higher performance than a single human or machine reader.