Related Experiment Video
Updated: Aug 10, 2026
12:50
Lesion Explorer: A Video-guided, Standardized Protocol for Accurate and Reliable MRI-derived Volumetrics in Alzheimer's Disease and Normal Elderly
Published on: April 14, 2014
Deep Learning-based Automated Vessel Extraction for VR Image Creation From Cerebral TOF-MRA
Kota Kawahara1, Shinpei Sato1, Daisuke Oura2
1Department of Radiology, Otaru General Hospital, 1-1-1 Wakamatsu, Otaru, Hokkaido, Japan (K.K., S.S., D.O.).
Academic Radiology
|August 8, 2026
Summary
A new deep learning model accurately segments cerebral vasculature from MRA scans, automating vessel extraction for efficient 3D visualization. This method improves reproducibility and significantly reduces time for creating volume renderings in clinical practice.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Medicine
- Neuroimaging
Background:
- Accurate 3D visualization of cerebral vasculature is crucial for clinical evaluation.
- Manual vessel extraction from time-of-flight (TOF) magnetic resonance angiography (MRA) is time-consuming and operator-dependent.
Purpose of the Study:
- To develop a deep learning-based cerebrovascular segmentation model.
- To create an automated vessel extraction method.
- To evaluate the accuracy, volumetric reliability, and workflow efficiency of the developed method for volume rendering (VR).
Main Methods:
- A 3D U-Net model was trained for vessel segmentation on TOF-MRA images.
- Automated vessel extraction was achieved by dilating predicted vessel regions.
- Segmentation performance was assessed using Dice Similarity Coefficient (DSC), Normalized Surface Dice (NSD), and Centerline Distance (CLD).
- Aneurysm volumes were compared between original and extracted images.
- VR image creation time was measured by radiological technologists.
Main Results:
- The model achieved high segmentation accuracy with an inter-rater DSC of 0.916.
- Dilated vessel masks showed significantly improved DSC, recall, and precision.
- Excellent performance was noted for NSD (0.982 ± 0.015) and CLD (0.196 ± 0.182 mm).
- Aneurysm volumes demonstrated strong correlation (r=0.999) and equivalence.
- VR image creation time was significantly reduced.
Conclusions:
- The proposed deep learning method enables accurate, reproducible, and time-efficient generation of cerebral vascular VR images.
- The method shows potential for enhancing clinical practice in neurovascular evaluation.
