Structure-aware vessel enhancement network for low-dose contrast agent CT angiography imaging.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Radiology

Background:

• CT angiography (CTA) is vital for vascular condition assessment.
• High contrast agent doses in CTA pose risks like nephropathy.
• Low-dose CTA (LDCTA) compromises image quality and small vessel detection.

Purpose of the Study:

• To develop an advanced LDCTA image enhancement network.
• To improve reconstruction of vascular details under low contrast conditions.
• To overcome limitations of current low-dose contrast techniques.

Main Methods:

• Proposed a novel LDCTA network integrating structure-aware perceptual loss and adaptive deformable convolution.
• Perceptual loss uses a pre-trained vascular segmentation model for anatomical focus.
• Deformable convolution dynamically adapts kernel shapes for irregular vessel feature extraction.

Main Results:

• The network demonstrated superior image enhancement quality in head-neck and thoracic datasets.
• Significantly improved preservation of intricate vascular structures compared to existing methods.
• Enhanced detection of small and irregular vessels under low-dose conditions.

Conclusions:

• The proposed LDCTA network effectively enhances image quality and vascular detail reconstruction.
• This approach offers a promising solution for safer, high-quality low-dose contrast CTA.
• Improved visualization aids in clinical diagnosis, preoperative planning, and postoperative monitoring.