Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Domain-Guided YOLO26 with Composite BCE-Dice-Lovász Loss for Multi-Class Fetal Head Ultrasound Segmentation

M. Fazri Nizar

Abstract

Segmenting fetal head structures from prenatal ultrasound remains a practical bottleneck in obstetric imaging. The current state-of-the-art baseline, proposed alongside the published dataset, adapts the Segment Anything Model with per-class Dice and Lovász losses but still depends on bounding-box prompts at test time. We build a prompt-free pipeline on top of YOLO26-Seg that jointly detects and segments three structures, Brain, Cavum Septi Pellucidi (CSP), and Lateral Ventricles (LV), in a single forward pass. Three modifications are central to our approach: (i) a composite BCE-Dice-Lovász segmentation loss with inverse-frequency class weighting, injected into the YOLO26 training loop via runtime monkey-patching; (ii) domain-guided copy-paste augmentation that transplants minority-class structures while respecting their anatomical location relative to the brain boundary; and (iii) inter-patient stratified splitting to prevent data leakage. On 575 held-out test images, the composite loss variant reaches a mean Dice coefficient of 0.9253, exceeding the baseline (0.9012) by 2.68 percentage points, despite reporting over three foreground classes only, whereas the baseline's reported mean includes the easy background class. We further ablate each component and discuss annotation-quality and class-imbalance effects on CSP and LV performance.

Domain-Guided YOLO26 with Composite BCE-Dice-Lovász Loss for Multi-Class Fetal Head Ultrasound Segmentation

Abstract

Segmenting fetal head structures from prenatal ultrasound remains a practical bottleneck in obstetric imaging. The current state-of-the-art baseline, proposed alongside the published dataset, adapts the Segment Anything Model with per-class Dice and Lovász losses but still depends on bounding-box prompts at test time. We build a prompt-free pipeline on top of YOLO26-Seg that jointly detects and segments three structures, Brain, Cavum Septi Pellucidi (CSP), and Lateral Ventricles (LV), in a single forward pass. Three modifications are central to our approach: (i) a composite BCE-Dice-Lovász segmentation loss with inverse-frequency class weighting, injected into the YOLO26 training loop via runtime monkey-patching; (ii) domain-guided copy-paste augmentation that transplants minority-class structures while respecting their anatomical location relative to the brain boundary; and (iii) inter-patient stratified splitting to prevent data leakage. On 575 held-out test images, the composite loss variant reaches a mean Dice coefficient of 0.9253, exceeding the baseline (0.9012) by 2.68 percentage points, despite reporting over three foreground classes only, whereas the baseline's reported mean includes the easy background class. We further ablate each component and discuss annotation-quality and class-imbalance effects on CSP and LV performance.

Paper Structure

This paper contains 23 sections, 8 equations, 9 figures, 11 tables.

Table of Contents

  1. Introduction
  2. Materials and Methods
  3. Dataset
  4. Inter-Patient Splitting
  5. YOLO26-Seg Architecture
  6. Composite BCE-Dice-Lovász Loss
  7. Automatic Optimizer Selection (MuSGD)
  8. Domain-Guided Copy-Paste Augmentation
  9. Training Configuration
  10. Evaluation Protocol
  11. Results
  12. Overall Performance
  13. Per-Class Breakdown
  14. Training Dynamics
  15. Comparison with FetSAM and Other Models
  16. ...and 8 more sections

Figures (9)

  • Figure 1: Class distribution and imbalance ratio across the full dataset.
  • Figure 2: Per-split class counts, showing that the stratified split preserves proportional representation.
  • Figure 3: Grouped bar chart of detection, segmentation, and classification metrics.
  • Figure 4: Per-class DSC across the three models, with numeric labels.
  • Figure 5: Metric heatmap for the Domain-Guided model; warm colours highlight the Brain--CSP--LV performance disparity.
  • ...and 4 more figures