Related Experiment Video
Updated: Aug 5, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
Automated Craniofacial Artery Segmentation with Vessel Enhancement-Guided Deep Learning
Hyeonju Park1, Young Chul Kim2, Kyoyeong Koo1
1SKIA Inc., 1502, 288 Digital-ro, Guro-gu, Seoul 08390, Republic of Korea.
Bioengineering (Basel, Switzerland)
|July 28, 2026
Summary
This study introduces a deep learning method to accurately segment superficial temporal and facial arteries using CT angiography. The new approach improves vessel boundary delineation and reduces errors, aiding surgical planning.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Medicine
- Neurosurgery
Background:
- Accurate segmentation of superficial temporal arteries (STAs) and facial vessels (FVs) from computed tomography angiography (CTA) is crucial for neurosurgical and reconstructive procedures.
- Challenges in STA and FV segmentation include their small size, complex paths, and proximity to bone, leading to inaccuracies.
Purpose of the Study:
- To develop an automated deep learning framework for precise segmentation of craniofacial vessels (STAs and FVs) using CTA data.
- To enhance segmentation accuracy, particularly for vessel boundaries and distal branches, compared to existing methods.
Main Methods:
- A 3D nnU-Net v2 model was trained on raw CTA volumes.
- A Fusion-based Vesselness Map (FVM) was created using multiscale filters to highlight vessels and suppress non-vascular structures like bone and skin.
- The FVM was integrated into the loss function as a spatial prior to guide the deep learning model, rather than as an additional input.
Main Results:
- The FVM-guided deep learning model significantly improved segmentation accuracy in 72 clinical cases compared to a baseline model.
- Average Symmetric Surface Distance for STAs decreased from 6.543 mm to 2.941 mm, indicating better boundary delineation.
- Qualitative analysis revealed reduced segmentation noise and fewer false positives, especially near bony structures and fine vascular branches.
Conclusions:
- Integrating classical vessel enhancement techniques (FVM) into deep learning supervision enhances craniofacial vessel segmentation.
- The proposed method yields morphologically consistent segmentations, offering improved support for preoperative neurosurgical and reconstructive planning.