Domain-Guided YOLO26 with Composite BCE-Dice-Lov\'{a}sz Loss for Multi-Class Fetal Head Ultrasound Segmentation

TL;DR

This paper introduces a prompt-free, domain-guided YOLO26-based pipeline with a composite loss function for accurate multi-class fetal head ultrasound segmentation, outperforming existing methods.

Contribution

The authors propose a novel YOLO26-based segmentation pipeline with a composite BCE-Dice-Lovász loss, domain-guided augmentation, and stratified splitting, advancing fetal ultrasound segmentation accuracy.

Findings

Achieved a mean Dice coefficient of 0.9253 on test images.

Outperformed the baseline Dice score by 2.68 percentage points.

Validated the effectiveness of each proposed component through ablation studies.

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 Lov\'{a}sz 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-Lov\'{a}sz 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.