Deep Multi-instance Networks with Sparse Label Assignment for Whole Mammogram Classification

By Wentao Zhu, Qi Lou, Yeeleng Scott Vang, Xiaohui S. Xie. 2016

301 citations recorded by the source corpus.

Mammogram classification is directly related to computer-aided diagnosis of breast cancer. Traditional methods requires great effort to annotate the training data by costly manual labeling and specialized computational models to detect these annotations during test. Inspired by the success of using deep convolutional features for natural image analysis and multi-instance learning for labeling a set of instances/patches, we propose end-to-end trained deep multiinstance networks for mass classification based on whole mammogram without the aforementioned costly need to annotate the training data. We explore three different schemes to construct deep multi-instance networks for whole mammogram classification. Experimental results on the INbreast dataset demonstrate the robustness of proposed deep networks compared to previous work using segmentation and detection annotations in the training.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Oncology

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Jimeng Sun
2. P. Heng
3. Qi-Yan Dou
4. Jie Tang
5. Xiaohui S. Xie
6. Hao Chen
7. Jason Phang
8. Thibault Févry

In Computer Vision and Pattern Recognition

1. P. Heng
2. Qi-Yan Dou
3. Jie Tang
4. Luc Van Gool
5. Xiaohui S. Xie
6. Hao Chen
7. Trevor Darrell
8. Kyunghyun Cho

In Oncology

1. P. Heng
2. Hao Chen
3. Balázs Harangi

Citations in the corpus

• Visual Feature Attribution Using Wasserstein GANs (2017, background): “Pinpointing all evidence of a class is important for a variety of tasks such as weakly supervised localisation or segmentation of structures [41, 43, 65], and better understanding disease effects, and physiological or pathological processes in medical images [67, 18, 12, 19, 13, 27, 54, 30, 31, 63].”
• Opportunities and obstacles for deep learning in biology and medicine (2017, background): “For example, an image can easily be rotated, flipped, or translated and retain its label [59].”
• Attention-based Deep Multiple Instance Learning (2018, neutral): “Other applications of MIL methods in medical imaging are mammography (nodule) classification (Zhu et al., 2017) and microscopy cell detection (Kraus et al., 2016).”
• Adversarial Deep Structural Networks for Mammographic Mass Segmentation (2016, background): “Ngo and Carneiro used deep belief networks as the initialization of distance regularized level set method to segment ventricle in magnetic resonance (MR) images [1].”
• DeepLung: Deep 3D Dual Path Nets for Automated Pulmonary Nodule Detection and Classification (2018, methodology): “Deep transfer learning and multiinstance learning is used for patient-level lung CT diagnosis [25, 36].”
• Weakly Supervised Deep Learning for Whole Slide Lung Cancer Image Analysis (2020, methodology): “Recently, an MIL framework [48] utilized sparse label assignment in supervised training to classify the entire mammogram of breast cancers with size of about 3000 × 2000.”
• High-Resolution Breast Cancer Screening with Multi-View Deep Convolutional Neural Networks (2017, background): “Two groups [32], [33] went even further by proposing a single deep convolutional network that classifies a whole image, or a set of multiple views.”
• LEAP: Learning to Prescribe Effective and Safe Treatment Combinations for Multimorbidity (2017, methodology): “Multi-instance learning has been used for biomedical applications such as prediction of binding sites in mRNA molecules [4], and classification of mammogram images [43].”
• Deep Neural Networks Improve Radiologists’ Performance in Breast Cancer Screening (2019, background): “Such formulations can be paired with multiple-instance learning where each spatial location is treated as a single instance and associated with a score that is correlated with the existence of a malignant finding [28].”
• Thoracic Disease Identification and Localization with Limited Supervision (2017, background): “proposed an endto-end deep multi-instance network to achieve mass classification for whole mammogram [37].”
• DTFD-MIL: Double-Tier Feature Distillation Multiple Instance Learning for Histopathology Whole Slide Image Classification (2022, background): “MIL models generally can be divided into two groups, based on whether the final bag predictions are directly from instance predictions [3, 9, 12, 15, 17, 47], or from the aggregations of features of instance [14, 18, 23, 34, 35, 42, 53].”