What type of study is this?

This is a Quantitative Study study.

September 17, 2025

Automatic Labelling of Intracranial arteries: A Comparison of UNet-based Networks

Key Points

Residual UNet achieved the highest Dice score of 0.863, particularly excelling in larger artery segmentation.
Using 18 patients with intracranial atherosclerotic disease, three UNet-based models were compared.
The study aims to streamline hemodynamic quantification by automating the segmentation process.
Integrating Residual UNet could significantly enhance automated flow data processing in future applications.

Abstract

Motivation: 4D Flow MRI of intracranial arteries requires labor-intensive manual analysis for accurate hemodynamic quantification. Goal(s): This study seeks to streamline the segmentation of intracranial arteries in 4D Flow MRI to improve the speed and accuracy of cerebral blood flow analysis. Approach: Using data from 18 patients with intracranial atherosclerotic disease, we compared three UNet-based models: Residual UNet, UNETR, and nnUNet. Results: Residual UNet achieved the highest average Dice score (0.863), particularly excelling in segmenting larger arteries, in comparison with the other methods. Future work includes integrating into automated flow data analysis. Impact: This study identifies Residual UNet as an effective tool for automated intracranial artery labeling, enabling fully automatic processing of intracranial 4D Flow MRI data.

Mark Helpful

Bookmark

Relay