Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SelectiveKD: A semi-supervised framework for cancer detection in DBT through Knowledge Distillation and Pseudo-labeling

Laurent Dillard, Hyeonsoo Lee, Weonsuk Lee, Tae Soo Kim, Ali Diba, Thijs Kooi

TL;DR

This paper presents SelectiveKD, a semi-supervised learning framework for building cancer detection models for DBT, which only requires a limited number of annotated slices to reach high performance and effectively utilizes unannotated slices from a DBT stack, leading to significantly improved cancer classification performance (AUC) and generalization performance.

Abstract

When developing Computer Aided Detection (CAD) systems for Digital Breast Tomosynthesis (DBT), the complexity arising from the volumetric nature of the modality poses significant technical challenges for obtaining large-scale accurate annotations. Without access to large-scale annotations, the resulting model may not generalize to different domains. Given the costly nature of obtaining DBT annotations, how to effectively increase the amount of data used for training DBT CAD systems remains an open challenge. In this paper, we present SelectiveKD, a semi-supervised learning framework for building cancer detection models for DBT, which only requires a limited number of annotated slices to reach high performance. We achieve this by utilizing unlabeled slices available in a DBT stack through a knowledge distillation framework in which the teacher model provides a supervisory signal to the student model for all slices in the DBT volume. Our framework mitigates the potential noise in the supervisory signal from a sub-optimal teacher by implementing a selective dataset expansion strategy using pseudo labels. We evaluate our approach with a large-scale real-world dataset of over 10,000 DBT exams collected from multiple device manufacturers and locations. The resulting SelectiveKD process effectively utilizes unannotated slices from a DBT stack, leading to significantly improved cancer classification performance (AUC) and generalization performance.

SelectiveKD: A semi-supervised framework for cancer detection in DBT through Knowledge Distillation and Pseudo-labeling

TL;DR

This paper presents SelectiveKD, a semi-supervised learning framework for building cancer detection models for DBT, which only requires a limited number of annotated slices to reach high performance and effectively utilizes unannotated slices from a DBT stack, leading to significantly improved cancer classification performance (AUC) and generalization performance.

Abstract

When developing Computer Aided Detection (CAD) systems for Digital Breast Tomosynthesis (DBT), the complexity arising from the volumetric nature of the modality poses significant technical challenges for obtaining large-scale accurate annotations. Without access to large-scale annotations, the resulting model may not generalize to different domains. Given the costly nature of obtaining DBT annotations, how to effectively increase the amount of data used for training DBT CAD systems remains an open challenge. In this paper, we present SelectiveKD, a semi-supervised learning framework for building cancer detection models for DBT, which only requires a limited number of annotated slices to reach high performance. We achieve this by utilizing unlabeled slices available in a DBT stack through a knowledge distillation framework in which the teacher model provides a supervisory signal to the student model for all slices in the DBT volume. Our framework mitigates the potential noise in the supervisory signal from a sub-optimal teacher by implementing a selective dataset expansion strategy using pseudo labels. We evaluate our approach with a large-scale real-world dataset of over 10,000 DBT exams collected from multiple device manufacturers and locations. The resulting SelectiveKD process effectively utilizes unannotated slices from a DBT stack, leading to significantly improved cancer classification performance (AUC) and generalization performance.
Paper Structure (11 sections, 1 equation, 2 figures, 1 table)

This paper contains 11 sections, 1 equation, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Method
  3. Practical Annotation of DBT data
  4. Incorporating Pseudo-Labeling and Knowledge Distillation
  5. Experiments
  6. Dataset
  7. Implementation Details
  8. Experimental setting
  9. Results
  10. Annotation cost efficiency
  11. Conclusion

Figures (2)

  • Figure 1: Overview of the SelectiveKD framework: Illustrating the application of knowledge distillation (KD) and SelectiveKD across varying levels of annotation of DBT slices. SelectiveKD leverages pseudo-labeling to select a few slices for training, minimizing the potential noise introduced by standard KD and reinforcing weak annotations. The approach allows use of all slices in the DBT stack during training.
  • Figure 2: Test set ROC AUC values versus the proportion of annotated exams: The scaling of model performance with respect to the proportion of annotated data is illustrated for the baseline and SelectiveKD. By using our approach, we show that the model achieves similar level of cancer detection performance while using less than one third of annotations (dotted line).