AI segmentation study highlights PE detection challenges, offers open-weight model

Researchers have identified significant limitations in current pulmonary embolism (PE) segmentation algorithms, citing issues with small datasets, lack of reproducibility, and insufficient comparative evaluations. Their study, which involved curating a new dataset of 490 CTPA scans and evaluating nine segmentation architectures, found that 3D U-Net models with ResNet encoding blocks performed best. The paper highlights that distal emboli remain particularly challenging due to task complexity and data scarcity, and makes the best-performing model's architecture and weights publicly available to foster reproducibility. AI

IMPACT Provides a reproducible baseline and open-weight model for PE segmentation, addressing data scarcity and model evaluation challenges.

RANK_REASON Academic paper with an open-weight model release and new dataset.

Read on arXiv cs.CV →

• CNN
• Pulmonary Embolism
• ResNet
• U-Net
• ViT
• Yixin Zhang

• paper
• other

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