Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Breaking the Barrier: Selective Uncertainty-based Active Learning for Medical Image Segmentation

Siteng Ma, Haochang Wu, Aonghus Lawlor, Ruihai Dong

TL;DR

This work tackles the shortcomings of traditional uncertainty-based active learning in medical image segmentation, where aggregating pixel-level uncertainties can overlook small target regions and introduce redundancy. It introduces a Selective Uncertainty-based AL framework with two complementary streams: Target-Aware Uncertainty Sampling using $T_x = \{ p|p\in pixels_x, P(c|p)>T \}$ and Boundary-Driven Uncertainty Sampling using $U_x = \{ p|p\in pixels_x, |P(c|p)-B|<U \}$, then fuses them via TopK to select 2k samples per round. The approach is demonstrated to be compatible with multiple uncertainty baselines across BraTS and MSD datasets, achieving higher Dice scores with fewer labeled samples and surpassing random sampling in most rounds. The findings suggest substantial practical gains in labeling efficiency and segmentation performance, with code available for replication.

Abstract

Active learning (AL) has found wide applications in medical image segmentation, aiming to alleviate the annotation workload and enhance performance. Conventional uncertainty-based AL methods, such as entropy and Bayesian, often rely on an aggregate of all pixel-level metrics. However, in imbalanced settings, these methods tend to neglect the significance of target regions, eg., lesions, and tumors. Moreover, uncertainty-based selection introduces redundancy. These factors lead to unsatisfactory performance, and in many cases, even underperform random sampling. To solve this problem, we introduce a novel approach called the Selective Uncertainty-based AL, avoiding the conventional practice of summing up the metrics of all pixels. Through a filtering process, our strategy prioritizes pixels within target areas and those near decision boundaries. This resolves the aforementioned disregard for target areas and redundancy. Our method showed substantial improvements across five different uncertainty-based methods and two distinct datasets, utilizing fewer labeled data to reach the supervised baseline and consistently achieving the highest overall performance. Our code is available at https://github.com/HelenMa9998/Selective\_Uncertainty\_AL.

Breaking the Barrier: Selective Uncertainty-based Active Learning for Medical Image Segmentation

TL;DR

This work tackles the shortcomings of traditional uncertainty-based active learning in medical image segmentation, where aggregating pixel-level uncertainties can overlook small target regions and introduce redundancy. It introduces a Selective Uncertainty-based AL framework with two complementary streams: Target-Aware Uncertainty Sampling using and Boundary-Driven Uncertainty Sampling using , then fuses them via TopK to select 2k samples per round. The approach is demonstrated to be compatible with multiple uncertainty baselines across BraTS and MSD datasets, achieving higher Dice scores with fewer labeled samples and surpassing random sampling in most rounds. The findings suggest substantial practical gains in labeling efficiency and segmentation performance, with code available for replication.

Abstract

Active learning (AL) has found wide applications in medical image segmentation, aiming to alleviate the annotation workload and enhance performance. Conventional uncertainty-based AL methods, such as entropy and Bayesian, often rely on an aggregate of all pixel-level metrics. However, in imbalanced settings, these methods tend to neglect the significance of target regions, eg., lesions, and tumors. Moreover, uncertainty-based selection introduces redundancy. These factors lead to unsatisfactory performance, and in many cases, even underperform random sampling. To solve this problem, we introduce a novel approach called the Selective Uncertainty-based AL, avoiding the conventional practice of summing up the metrics of all pixels. Through a filtering process, our strategy prioritizes pixels within target areas and those near decision boundaries. This resolves the aforementioned disregard for target areas and redundancy. Our method showed substantial improvements across five different uncertainty-based methods and two distinct datasets, utilizing fewer labeled data to reach the supervised baseline and consistently achieving the highest overall performance. Our code is available at https://github.com/HelenMa9998/Selective\_Uncertainty\_AL.
Paper Structure (12 sections, 5 equations, 2 figures, 3 tables)

This paper contains 12 sections, 5 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Proposed Method
  3. Proposed Method
  4. Target-Aware Uncertainty Sampling
  5. Boundary-Driven Uncertainty Sampling
  6. Selective Uncertainty Sampling
  7. EXPERIMENT
  8. Implementation Details
  9. Ablation Studies
  10. Experimental Results
  11. Conclusions
  12. Acknowledgement

Figures (2)

  • Figure 1: Comparison between the conventional and our proposed methods. (a) The general framework of the conventional uncertainty-based query strategy; (b) Our proposed Selective Uncertainty-based method.
  • Figure 2: Distribution of selected samples in the unlabeled pool: Random sampling selects diverse samples, but with many blank slices. The existing entropy-based method selects samples with targets but introduces redundancy. Our modified entropy-based method considers both the model's inherent prediction uncertainty and sample distribution.