Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ProPL: Universal Semi-Supervised Ultrasound Image Segmentation via Prompt-Guided Pseudo-Labeling

Yaxiong Chen, Qicong Wang, Chunlei Li, Jingliang Hu, Yilei Shi, Shengwu Xiong, Xiao Xiang Zhu, Lichao Mou

TL;DR

ProPL tackles the challenge of universal semi-supervised ultrasound image segmentation by introducing a prompt-guided pseudo-labeling framework with a shared encoder, a prompt encoder, and dual decoders. It leverages prompting-upon-decoding and an uncertainty-driven pseudo-label calibration (UPLC) to learn from both labeled and unlabeled data across multiple organs and tasks, validated on a 6,400-image ultrasound dataset. Empirically, ProPL outperforms state-of-the-art single-task and universal segmentation methods by substantial margins in Dice and mIoU, and its ablations show the critical role of prompts and UPLC, along with favorable memory efficiency. The approach demonstrates strong cross-task generalization and practical potential for data-scarce clinical settings, advancing universal ultrasound segmentation toward real-world deployment.

Abstract

Existing approaches for the problem of ultrasound image segmentation, whether supervised or semi-supervised, are typically specialized for specific anatomical structures or tasks, limiting their practical utility in clinical settings. In this paper, we pioneer the task of universal semi-supervised ultrasound image segmentation and propose ProPL, a framework that can handle multiple organs and segmentation tasks while leveraging both labeled and unlabeled data. At its core, ProPL employs a shared vision encoder coupled with prompt-guided dual decoders, enabling flexible task adaptation through a prompting-upon-decoding mechanism and reliable self-training via an uncertainty-driven pseudo-label calibration (UPLC) module. To facilitate research in this direction, we introduce a comprehensive ultrasound dataset spanning 5 organs and 8 segmentation tasks. Extensive experiments demonstrate that ProPL outperforms state-of-the-art methods across various metrics, establishing a new benchmark for universal ultrasound image segmentation.

ProPL: Universal Semi-Supervised Ultrasound Image Segmentation via Prompt-Guided Pseudo-Labeling

TL;DR

ProPL tackles the challenge of universal semi-supervised ultrasound image segmentation by introducing a prompt-guided pseudo-labeling framework with a shared encoder, a prompt encoder, and dual decoders. It leverages prompting-upon-decoding and an uncertainty-driven pseudo-label calibration (UPLC) to learn from both labeled and unlabeled data across multiple organs and tasks, validated on a 6,400-image ultrasound dataset. Empirically, ProPL outperforms state-of-the-art single-task and universal segmentation methods by substantial margins in Dice and mIoU, and its ablations show the critical role of prompts and UPLC, along with favorable memory efficiency. The approach demonstrates strong cross-task generalization and practical potential for data-scarce clinical settings, advancing universal ultrasound segmentation toward real-world deployment.

Abstract

Existing approaches for the problem of ultrasound image segmentation, whether supervised or semi-supervised, are typically specialized for specific anatomical structures or tasks, limiting their practical utility in clinical settings. In this paper, we pioneer the task of universal semi-supervised ultrasound image segmentation and propose ProPL, a framework that can handle multiple organs and segmentation tasks while leveraging both labeled and unlabeled data. At its core, ProPL employs a shared vision encoder coupled with prompt-guided dual decoders, enabling flexible task adaptation through a prompting-upon-decoding mechanism and reliable self-training via an uncertainty-driven pseudo-label calibration (UPLC) module. To facilitate research in this direction, we introduce a comprehensive ultrasound dataset spanning 5 organs and 8 segmentation tasks. Extensive experiments demonstrate that ProPL outperforms state-of-the-art methods across various metrics, establishing a new benchmark for universal ultrasound image segmentation.

Paper Structure

This paper contains 17 sections, 5 equations, 5 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Single-Task Segmentation for Ultrasound Images
  4. Single-Task Semi-Supervised Image Segmentation
  5. Universal Medical Image Segmentation
  6. Universal Semi-Supervised Image Segmentation
  7. Methodology
  8. Network Architecture
  9. Prompting-upon-Decoding
  10. Training and Pseudo-Label Calibration
  11. Experiments
  12. Datasets and Evaluation Metrics
  13. Implementation Details
  14. Comparison with State-of-the-Art Methods
  15. Discussion
  16. ...and 2 more sections

Figures (5)

  • Figure 1: Overview of our proposed universal semi-supervised model ProPL.
  • Figure 2: Comparison of our prompting-upon-decoding method against established prompting approaches implemented in CLIP-UM ref_clip and DoDNet ref_dodnet.
  • Figure 3: Impact of varying the number of stochastic perturbations in UPLC.
  • Figure 4: Performance-efficiency trade-off analysis across various models.
  • Figure 5: Qualitative segmentation results comparing the proposed method against competitive approaches across four segmentation tasks: breast cancer, left atrium, ovarian tumor, and thyroid gland.