Related Experiment Video
Updated: Jan 11, 2026
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
Automatic segmentation and reconstruction of lower-extremity arteries from computed tomography angiography images via
Xiang Liu1, Huijun Hu1, Guoxiong Lu1
1Department of Radiology, Sun Yat-sen Memorial Hospital, Sun Yat-sen University, Guangzhou, China.
Background:
Preoperative lower extremity computed tomography angiography (CTA) is indispensable for planning free skin flap transplantation in patients with oral and maxillofacial tumors. However, there is a lack of dedicated automatic three-dimensional (3D) reconstruction tools of lower-extremity arteries from CTA images. Our study aimed to develop and validate an artificial intelligence (AI) model based on a 3D convolutional neural network (CNN) for automatic reconstruction of lower-extremity arteries from preoperative lower extremity CTA images.
Methods:
This retrospective study included a dataset of lower extremity CTA images from 1,201 patients with oral or maxillofacial tumors between January 2015 and December 2023. A deep learning-based lower-extremity artery segmentation network (LEAS-Net) was proposed for 3D reconstruction of the lower-extremity arteries from CTA images, which had a three-stage network architecture consisting of a coarse-resolution network for initial vessel localization, a refinement network for skeleton extraction, and a fine-resolution network for precise segmentation. The segmentation performance of LEAS-Net was assessed via the Dice similarity coefficient and center-line Dice coefficient (clDice). The quality of reconstructed images and the time needed for reconstruction were compared between LEAS-Net and three human radiologists.
Results:
The LEAS-Net exhibited high accuracy in segmenting large vessels and small perforator vessels at the voxel level, with average Dice and clDice coefficients exceeding 0.65. The LEAS-Net achieved a higher or comparable image quality score for the reconstruction of large vessels and a higher quality score for the perforator vessel reconstruction compared with human radiologists (P<0.01). The processing time of LEAS-Net was reduced by 9.7 to 22 times compared with the three human radiologists (P<0.05).
Conclusions:
The LEAS-Net can be used as an AI tool to reconstruct lower-extremity arteries from CTA images for planning perforator flap transplantation.
