← Back to Projects

Project

Metal artifact reduction for tomosynthesis

Oct 1, 2024

Built an AI-assisted annotation and refinement pipeline for fluoroscopy data, then trained and deployed an nnU-Net system for metal artifact reduction across multiple C-arm platforms.

Built an AI-assisted annotation and refinement pipeline for fluoroscopy data, then trained and deployed an nnU-Net system for metal artifact reduction across multiple C-arm platforms.

  • Built an AI-assisted fluoroscopy annotation and human-in-the-loop refinement pipeline that reduced per-video labeling time by 83%, from 12 hours to 2 hours.
  • Trained and deployed an nnU-Net for metal artifact reduction using the resulting scalable dataset.
  • Covered 11 C-arm systems and removed about 95% of metal-related artifacts, including artifacts from scopes and WFGs, while preserving underlying tissue structures.
  • Metal Artifact Reduction
  • Tomosynthesis
  • nnU-Net
  • Human-in-the-loop