Cerebral arteries segmentation based on projection domain in single exposure computed tomographic angiography

Area of Science:

• Medical Imaging
• Radiology
• Artificial Intelligence in Medicine

Background:

• Cerebrovascular diseases pose significant global health challenges, necessitating accurate and timely diagnosis.
• Conventional CT angiography (CTA) methods face limitations such as invasiveness, radiation exposure, and artifacts from metal and bone.

Purpose of the Study:

• To evaluate the feasibility of segmenting cerebral arteries using a novel projection-domain framework in single-exposure CT angiography (CTA).
• To assess the potential of this AI-driven approach for improved vascular imaging.

Main Methods:

• Utilized the DeepLab V3+ model for direct brain vessel segmentation in the projection domain.
• Trained and tested the model on a dataset of 103 patient CTA scans.
• The framework minimizes motion and beam-hardening artifacts while preserving clinical data.

Main Results:

• Achieved high performance metrics, including Intersection over Union (0.89) and Dice similarity coefficient (0.90), indicating strong agreement with reference labels.
• Demonstrated the ability to reconstruct clinically relevant vascular information, even in the presence of metallic implants.
• Quantitatively validated the accuracy of the reconstructed cerebral artery images.

Conclusions:

• The proposed projection-domain framework is feasible for generating 3D cerebral arterial images in CTA systems.
• This method offers potential for enhanced vascular visualization and improved image quality.
• Further clinical validation is recommended to confirm its utility in patient care.