Bi-attention DoubleUNet: A deep learning approach for carotid artery segmentation in transverse view images for non-invasive stenosis diagnosis
The carotid artery is a vital blood vessel that supplies oxygenated blood to the brain. Blockages in this artery can lead to life-threatening illnesses, making accurate diagnosis essential. While ultrasound (US) imaging is the primary diagnostic tool, evaluations by trained operators can be subjecti...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | , , , , |
| Format: | article |
| Published: |
2024
|
| Subjects: | |
| Online Access: | http://dx.doi.org/10.1016/j.bspc.2024.106350 https://www.sciencedirect.com/science/article/pii/S1746809424004087 http://hdl.handle.net/10576/55190 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The carotid artery is a vital blood vessel that supplies oxygenated blood to the brain. Blockages in this artery can lead to life-threatening illnesses, making accurate diagnosis essential. While ultrasound (US) imaging is the primary diagnostic tool, evaluations by trained operators can be subjective and imprecise. An automated approach can provide a more reliable and accurate evaluation of the carotid artery’s condition. To achieve this, the artery must first be identified and isolated from US images. This paper proposes an automated segmentation method for the carotid artery in transverse B-mode ultrasound images, using a Bi-attention DoubleUnet architecture which incorporates spatial attention and channel wise attention using Bottleneck attention module. The method is evaluated on a dataset of transverse images acquired from two devices. The obtained results exhibit higher performance than existing methods with a dice index of 95.96%, IoU of 97.92%, Precision of 98.35% and recall of 97.57% on combined dataset. Another variant of doubleUnet with SE layer and without ASPP (Artous Spatial Pyramidal Pooling) is presented where the modified DoubleUnet with SE realized an average IoU of 92.805%, Dice of 96.215%, precision of 98.82%, and recall of 93.84% representing average results across datasets. These approaches can significantly improve the efficiency and accuracy of carotid artery disease diagnosis and treatment, ultimately improving patient outcomes. |
|---|