ASTRA-Net model improves DISE segmentation with limited annotations

Researchers have developed ASTRA-Net, a novel system designed for segmenting drug-induced sleep endoscopy (DISE) images, particularly when real annotated data is scarce. The system employs a two-stage approach: first, it aligns intermediate representations from a large dataset of unlabeled virtual endoscopy frames with real DISE frames using ConvNeXt-Base. Second, it fine-tunes four independent UNet++ decoders on a limited set of 401 annotated real frames. This method achieved a mean Dice score of 0.8927 and a mean intersection over union of 0.8239 on a hold-out evaluation set, demonstrating its effectiveness in delineating airway boundaries with limited annotations. AI

IMPACT Enables more accurate medical image analysis in scenarios with limited annotated data.

RANK_REASON The cluster contains a research paper detailing a new model and methodology for image segmentation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• arXiv
• ASTRA-Net
• ConvNeXt-Base
• Hugging Face
• UNet++: A Nested U-Net Architecture for Medical Image Segmentation

• paper
• model release
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →