Temporal consistency-aware network for renal artery segmentation in X-ray angiography

Area of Science:

• Medical imaging analysis
• Deep learning for medical applications
• Cardiovascular imaging

Background:

• Accurate segmentation of renal arteries in X-ray angiography is vital for assessing renal sympathetic denervation (RDN) procedures.
• Challenges include dynamic contrast changes and varying vessel morphology across video frames.
• Existing methods struggle with segmentation consistency.

Purpose of the Study:

• To introduce TCA-Net, a novel deep learning model designed to enhance segmentation consistency in renal artery angiography videos.
• To leverage local and global contextual information for improved vessel segmentation.
• To provide a more reliable tool for evaluating RDN procedures.

Main Methods:

• A deep learning framework incorporating a local temporal window vessel enhancement module and a global vessel refinement module (GVR).
• The local module fuses multi-scale temporal-spatial features for enhanced vessel representation.
• The GVR module uses decoupled attention and gating mechanisms for global refinement and redundancy reduction, trained with a temporal perception consistency loss.

Main Results:

• TCA-Net was evaluated on 195 renal artery angiography sequences for development and an external dataset from 44 patients.
• Achieved a high F1-score of 0.8678 for renal artery segmentation.
• Demonstrated superior performance compared to existing state-of-the-art segmentation methods.

Conclusions:

• TCA-Net significantly improves segmentation consistency in renal artery angiography videos.
• Effectively utilizes local and global temporal contextual information.
• Offers a reliable deep learning-based solution for assessing RDN procedures.