Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

HDC: Hierarchical Distillation for Multi-level Noisy Consistency in Semi-Supervised Fetal Ultrasound Segmentation

Tran Quoc Khanh Le, Nguyen Lan Vi Vu, Ha-Hieu Pham, Xuan-Loc Huynh, Tien-Huy Nguyen, Minh Huu Nhat Le, Quan Nguyen, Hien D. Nguyen

TL;DR

A novel semi-supervised segmentation framework, called HDC, is proposed incorporating adaptive consistency learning with a single-teacher architecture, which reduces model complexity while enhancing generalization.

Abstract

Transvaginal ultrasound is a critical imaging modality for evaluating cervical anatomy and detecting physiological changes. However, accurate segmentation of cervical structures remains challenging due to low contrast, shadow artifacts, and indistinct boundaries. While convolutional neural networks (CNNs) have demonstrated efficacy in medical image segmentation, their reliance on large-scale annotated datasets presents a significant limitation in clinical ultrasound imaging. Semi-supervised learning (SSL) offers a potential solution by utilizing unlabeled data, yet existing teacher-student frameworks often encounter confirmation bias and high computational costs. In this paper, a novel semi-supervised segmentation framework, called HDC, is proposed incorporating adaptive consistency learning with a single-teacher architecture. The framework introduces a hierarchical distillation mechanism with two objectives: Correlation Guidance Loss for aligning feature representations and Mutual Information Loss for stabilizing noisy student learning. The proposed approach reduces model complexity while enhancing generalization. Experiments on fetal ultrasound datasets, FUGC and PSFH, demonstrate competitive performance with reduced computational overhead compared to multi-teacher models.

HDC: Hierarchical Distillation for Multi-level Noisy Consistency in Semi-Supervised Fetal Ultrasound Segmentation

TL;DR

A novel semi-supervised segmentation framework, called HDC, is proposed incorporating adaptive consistency learning with a single-teacher architecture, which reduces model complexity while enhancing generalization.

Abstract

Transvaginal ultrasound is a critical imaging modality for evaluating cervical anatomy and detecting physiological changes. However, accurate segmentation of cervical structures remains challenging due to low contrast, shadow artifacts, and indistinct boundaries. While convolutional neural networks (CNNs) have demonstrated efficacy in medical image segmentation, their reliance on large-scale annotated datasets presents a significant limitation in clinical ultrasound imaging. Semi-supervised learning (SSL) offers a potential solution by utilizing unlabeled data, yet existing teacher-student frameworks often encounter confirmation bias and high computational costs. In this paper, a novel semi-supervised segmentation framework, called HDC, is proposed incorporating adaptive consistency learning with a single-teacher architecture. The framework introduces a hierarchical distillation mechanism with two objectives: Correlation Guidance Loss for aligning feature representations and Mutual Information Loss for stabilizing noisy student learning. The proposed approach reduces model complexity while enhancing generalization. Experiments on fetal ultrasound datasets, FUGC and PSFH, demonstrate competitive performance with reduced computational overhead compared to multi-teacher models.

Paper Structure

This paper contains 19 sections, 15 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Method
  4. Theoretical Insights
  5. Overall Training Scheme
  6. Hierarchical Distillation for Multi-Level Noisy Consistency
  7. Correlation Guidance Loss
  8. Mutual Information Loss
  9. Pixel-Level Consistency Loss
  10. Final Objective Function
  11. Progressive Parameter Updates
  12. Experiments and Results
  13. Dataset
  14. Implementation Details
  15. Evaluation Metrics
  16. ...and 4 more sections

Figures (3)

  • Figure 1: An intuitive overview of our Hierarchical Distillation and Consistency scheme.
  • Figure 2: Our proposed semi-supervised pipeline for transvaginal ultrasound segmentation. The teacher generates pseudo-labels to guide the student, which processes both raw and noisy inputs. The dashed arrow (- - -) represents strong augmentation, the double slash arrow (//) denotes feature comparisons for $L_{CG}$ and $L_{MI}$, black arrows indicate forward propagation, and red arrows represent pixel-level consistency loss.
  • Figure 3: Qualitative comparison of different semi-supervised segmentation methods results.