Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DSTCS: Dual-Student Teacher Framework with Segment Anything Model for Semi-Supervised Pubic Symphysis Fetal Head Segmentation

Yalin Luo, Shun Long, Huijin Wang, Jieyun Bai

TL;DR

This work tackles pubic symphysis–fetal head segmentation in intrapartum ultrasound, addressing data scarcity, boundary ambiguity, and noise. It introduces DSTCS, a Dual-Student-Teacher framework that fuses a CNN-based branch with a Segment Anything Model adapter, guided by edge-preserving augmentation and a boundary-aware loss suite. Key contributions include Edge-Patch In-situ Superposition (EPIS), Neighborhood Weighted Dice (NW-Dice) loss, cross-supervision and teacher–student consistency, and the AoP-SAM integration within a dual-student setup, all optimized under the loss L_total = L_sup + α L_h + β L_s + γ L_cdd + μ L_cr with $(α,β,γ,μ) = (0.5,1.0,3.0,0.1)$. Experiments on MICCAI 2023 and 2024 PSFH benchmarks demonstrate state-of-the-art segmentation accuracy and boundary precision, with robustness under high noise and low contrast, indicating strong potential for clinical adoption.

Abstract

Segmentation of the pubic symphysis and fetal head (PSFH) is a critical procedure in intrapartum monitoring and is essential for evaluating labor progression and identifying potential delivery complications. However, achieving accurate segmentation remains a significant challenge due to class imbalance, ambiguous boundaries, and noise interference in ultrasound images, compounded by the scarcity of high-quality annotated data. Current research on PSFH segmentation predominantly relies on CNN and Transformer architectures, leaving the potential of more powerful models underexplored. In this work, we propose a Dual-Student and Teacher framework combining CNN and SAM (DSTCS), which integrates the Segment Anything Model (SAM) into a dual student-teacher architecture. A cooperative learning mechanism between the CNN and SAM branches significantly improves segmentation accuracy. The proposed scheme also incorporates a specialized data augmentation strategy optimized for boundary processing and a novel loss function. Extensive experiments on the MICCAI 2023 and 2024 PSFH segmentation benchmarks demonstrate that our method exhibits superior robustness and significantly outperforms existing techniques, providing a reliable segmentation tool for clinical practice.

DSTCS: Dual-Student Teacher Framework with Segment Anything Model for Semi-Supervised Pubic Symphysis Fetal Head Segmentation

TL;DR

This work tackles pubic symphysis–fetal head segmentation in intrapartum ultrasound, addressing data scarcity, boundary ambiguity, and noise. It introduces DSTCS, a Dual-Student-Teacher framework that fuses a CNN-based branch with a Segment Anything Model adapter, guided by edge-preserving augmentation and a boundary-aware loss suite. Key contributions include Edge-Patch In-situ Superposition (EPIS), Neighborhood Weighted Dice (NW-Dice) loss, cross-supervision and teacher–student consistency, and the AoP-SAM integration within a dual-student setup, all optimized under the loss L_total = L_sup + α L_h + β L_s + γ L_cdd + μ L_cr with . Experiments on MICCAI 2023 and 2024 PSFH benchmarks demonstrate state-of-the-art segmentation accuracy and boundary precision, with robustness under high noise and low contrast, indicating strong potential for clinical adoption.

Abstract

Segmentation of the pubic symphysis and fetal head (PSFH) is a critical procedure in intrapartum monitoring and is essential for evaluating labor progression and identifying potential delivery complications. However, achieving accurate segmentation remains a significant challenge due to class imbalance, ambiguous boundaries, and noise interference in ultrasound images, compounded by the scarcity of high-quality annotated data. Current research on PSFH segmentation predominantly relies on CNN and Transformer architectures, leaving the potential of more powerful models underexplored. In this work, we propose a Dual-Student and Teacher framework combining CNN and SAM (DSTCS), which integrates the Segment Anything Model (SAM) into a dual student-teacher architecture. A cooperative learning mechanism between the CNN and SAM branches significantly improves segmentation accuracy. The proposed scheme also incorporates a specialized data augmentation strategy optimized for boundary processing and a novel loss function. Extensive experiments on the MICCAI 2023 and 2024 PSFH segmentation benchmarks demonstrate that our method exhibits superior robustness and significantly outperforms existing techniques, providing a reliable segmentation tool for clinical practice.
Paper Structure (9 sections, 10 equations, 4 figures, 4 tables)

This paper contains 9 sections, 10 equations, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Experiments and Results
  4. Dataset and implementations
  5. Comparison with other methods
  6. Generalization experiment
  7. Ablation study
  8. Impact of annotation ratio on segmentation performance
  9. Conclusion

Figures (4)

  • Figure 1: An overview of the DSTCS architecture, where the black and grey parts are the loss functions for student1 and student2
  • Figure 2: Box plots with overlaid scatter points comparing the performance distribution of different methods tested with 20% labeled data.
  • Figure 3: Visual comparison of different methods using 20% labeled data for testing. Red indicates ground truth annotations, green represents PS prediction results, and yellow denotes FH prediction results.
  • Figure 4: Segmentation performance versus annotation ratios. Red, green and yellow curves denote the FH, PS, and PSFH, respectively.