Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SIFT-DBT: Self-supervised Initialization and Fine-Tuning for Imbalanced Digital Breast Tomosynthesis Image Classification

Yuexi Du, Regina J. Hooley, John Lewin, Nicha C. Dvornek

TL;DR

A novel method using view-level contrastive Self-supervised Initialization and Fine-Tuning for identifying abnormal DBT images, namely SIFT-DBT is proposed, which achieves 92.69% volume-wise AUC on an evaluation of 970 unique studies.

Abstract

Digital Breast Tomosynthesis (DBT) is a widely used medical imaging modality for breast cancer screening and diagnosis, offering higher spatial resolution and greater detail through its 3D-like breast volume imaging capability. However, the increased data volume also introduces pronounced data imbalance challenges, where only a small fraction of the volume contains suspicious tissue. This further exacerbates the data imbalance due to the case-level distribution in real-world data and leads to learning a trivial classification model that only predicts the majority class. To address this, we propose a novel method using view-level contrastive Self-supervised Initialization and Fine-Tuning for identifying abnormal DBT images, namely SIFT-DBT. We further introduce a patch-level multi-instance learning method to preserve spatial resolution. The proposed method achieves 92.69% volume-wise AUC on an evaluation of 970 unique studies.

SIFT-DBT: Self-supervised Initialization and Fine-Tuning for Imbalanced Digital Breast Tomosynthesis Image Classification

TL;DR

A novel method using view-level contrastive Self-supervised Initialization and Fine-Tuning for identifying abnormal DBT images, namely SIFT-DBT is proposed, which achieves 92.69% volume-wise AUC on an evaluation of 970 unique studies.

Abstract

Digital Breast Tomosynthesis (DBT) is a widely used medical imaging modality for breast cancer screening and diagnosis, offering higher spatial resolution and greater detail through its 3D-like breast volume imaging capability. However, the increased data volume also introduces pronounced data imbalance challenges, where only a small fraction of the volume contains suspicious tissue. This further exacerbates the data imbalance due to the case-level distribution in real-world data and leads to learning a trivial classification model that only predicts the majority class. To address this, we propose a novel method using view-level contrastive Self-supervised Initialization and Fine-Tuning for identifying abnormal DBT images, namely SIFT-DBT. We further introduce a patch-level multi-instance learning method to preserve spatial resolution. The proposed method achieves 92.69% volume-wise AUC on an evaluation of 970 unique studies.
Paper Structure (14 sections, 2 equations, 4 figures, 3 tables)

This paper contains 14 sections, 2 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Contrastive Learning Paradigm Review
  4. DBT Metadata Based Positive Pair Selection
  5. Local Multi-Patch Fine-tuning
  6. Volume-Level Classification from Patch Prediction
  7. Experiments
  8. Experiment Setup
  9. Slice-level Evaluation
  10. Ablation Experiments
  11. Volume-level Evaluation
  12. Discussion and Conclusion
  13. Compliance with Ethical Standards
  14. Acknowledgements

Figures (4)

  • Figure 1: Proposed SIFT-DBT framework. (a) We use random augmentation $t\sim\mathcal{T}$ with inter-view and inter-slice sampling to get the positive pair inputs for pre-training. (b) During fine-tuning, we updated the model with only one sub-patch sampled from the full image. The blue gradient color within the model indicates we use the discriminative learning rate for each block.
  • Figure 2: Slice-level ROC curve. We plot the slice-level ROC curve for the baselines. Our method is plotted with a solid line.
  • Figure 3: Evaluation with different number of patches. We evaluate the effect of using different numbers of patches on slice classification performance (AUC) for each baseline with the ResNet50 backbone.
  • Figure 4: Volume-level ROC curve. We plot the ROC curve for the volume-level evaluation. Our method is plotted with a solid line.