ARTFEED — Contemporary Art Intelligence

SAM-Med2D and DINOv3 Enhance Fetal Cardiac Ultrasound Analysis

ai-technology · 2026-05-20

A semi-supervised framework integrating SAM-Med2D and DINOv3 improves fetal cardiac ultrasound segmentation and classification. Built on the EchoCare backbone, the method uses view-specific hard masking and a two-stage optimization strategy. On the FETUS 2026 leaderboard, it achieves a Dice Similarity Coefficient of 79.99%, Normalized Surface Distance of 61.62%, and F1-score of 41.20%, aiding prenatal congenital heart disease screening. Source code is available on GitHub.

Key facts

  • Semi-supervised framework for joint segmentation and classification of fetal cardiac ultrasound images.
  • Built upon the EchoCare multi-task backbone.
  • Integrates SAM-Med2D for boundary refinement and DINOv3 for pseudo-label enhancement.
  • Introduces view-specific hard masking and two-stage optimization: EMA phase then Classification Fine-Tuning phase.
  • Evaluated on the FETUS 2026 leaderboard.
  • Achieves Dice Similarity Coefficient of 79.99%, Normalized Surface Distance of 61.62%, F1-score of 41.20%.
  • Aims to improve prenatal congenital heart disease screening.
  • Source code publicly available at https://github.com/2826056177/zcst_fetus20.

Entities

Institutions

  • EchoCare
  • FETUS 2026

Sources