Abstract:Breast cancer remains a global challenge, causing over 600,000 deaths in 2018 (ref. <a href="/articles/s41591-020-01174-9#ref-CR1">1</a>). To achieve earlier cancer detection, health organizations worldwide recommend screening mammography, which is estimated to decrease breast cancer mortality by 20–40% (refs. <a href="/articles/s41591-020-01174-9#ref-CR2">2</a>,<a href="/articles/s41591-020-01174-9#ref-CR3">3</a>). Despite the clear value of screening mammography, significant false positive and false negative rates along with non-uniformities in expert reader availability leave opportunities for improving quality and access<a href="/articles/s41591-020-01174-9#ref-CR4">4</a>,<a href="/articles/s41591-020-01174-9#ref-CR5">5</a>. To address these limitations, there has been much recent interest in applying deep learning to mammography<a href="#ref-CR6">6</a>,<a href="#ref-CR7">7</a>,<a href="#ref-CR8">8</a>,<a href="#ref-CR9">9</a>,<a href="#ref-CR10">10</a>,<a href="#ref-CR11">11</a>,<a href="#ref-CR12">12</a>,<a href="#ref-CR13">13</a>,<a href="#ref-CR14">14</a>,<a href="#ref-CR15">15</a>,<a href="#ref-CR16">16</a>,<a href="#ref-CR17">17</a>,<a href="/articles/s41591-020-01174-9#ref-CR18">18</a>, and these efforts have highlighted two key difficulties: obtaining large amounts of annotated training data and ensuring generalization across populations, acquisition equipment and modalities. Here we present an annotation-efficient deep learning approach that (1) achieves state-of-the-art performance in mammogram classification, (2) successfully extends to digital breast tomosynthesis (DBT; '3D mammography'), (3) detects cancers in clinically negative prior mammograms of patients with cancer, (4) generalizes well to a population with low screening rates and (5) outperforms five out of five full-time breast-imaging specialists with an average increase in sensitivity of 14%. By creating new 'maximum suspicion projection' (MSP) images from DBT data, our progressively trained, multiple-instance learning approach effectively trains on DBT exams using only breast-level labels while maintaining localization-based interpretability. Altogether, our results demonstrate promise towards software that can improve the accuracy of and access to screening mammography worldwide.

Mammographic Image Classification System via Active Learning

Minimization of Annotation Work: Diagnosis of Mammographic Masses Via Active Learning.

Deep Co-Training Active Learning for Mammographic Images Classification

Breast Cancer Histopathological Image Classification Via Deep Active Learning And Confidence Boosting

A comprehensive survey on deep active learning in medical image analysis

Label-Efficient Breast Cancer Histopathological Image Classification

Active Learning on Medical Image

DSAL: Deeply Supervised Active Learning from Strong and Weak Labelers for Biomedical Image Segmentation.

Computerized breast cancer analysis system using three stage semi-supervised learning method

Classification of Mammography based on Semi-supervised Learning

AL-Annotator: an Active Learning-based Cervical Cell Annotation System

An Efficient High-Quality Medical Lesion Image Data Labeling Method Based on Active Learning.

A novel breast cancer image classification model based on multiscale texture feature analysis and dynamic learning

Clustered Microcalcification Detection in Digital Mammograms Based on an Active Learning with Support Vector Machine

Using Machine Learning to Automate Mammogram Images Analysis

Robust breast cancer detection in mammography and digital breast tomosynthesis using an annotation-efficient deep learning approach

Breast mass classification based on supervised contrastive learning and multi-view consistency penalty on mammography

Mammograms Classification: A Review

Su-Sampling Based Active Learning for Large-Scale Histopathology Image.

PCDAL: A Perturbation Consistency-Driven Active Learning Approach for Medical Image Segmentation and Classification

Predictive Accuracy-Based Active Learning for Medical Image Segmentation