Deep Learning-Based Image Reconstruction for CT Angiography of the Aorta

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• CT angiography is crucial for aortic assessment.
• Current iterative reconstruction methods have limitations in image quality.
• Deep learning offers potential for advanced image reconstruction.

Purpose of the Study:

• To evaluate a novel deep-learning-based image reconstruction (DLIR) algorithm.
• To assess DLIR's impact on image quality in CT angiography of the aorta.
• To compare DLIR with adaptive statistical iterative reconstruction (ASIR-V).

Main Methods:

• Retrospective analysis of 51 patients undergoing CT angiography.
• Image reconstruction using ASIR-V and DLIR on a 256-detector-row CT.
• Quantification of image noise, SNR, and CNR in various aortic segments.
• Subjective image quality scoring by two independent readers.

Main Results:

• DLIR reduced image noise by 51-54% in the ascending and descending thoracic aorta.
• DLIR approximately doubled CNR in the ascending and descending thoracic aorta.
• DLIR improved CNR by 38% in the abdominal aorta and iliac arteries, with improved subjective quality.

Conclusions:

• DLIR substantially improves objective and subjective image quality in aortic CT angiography.
• DLIR surpasses state-of-the-art iterative reconstruction in image enhancement.
• DLIR may enable future reductions in radiation and contrast agent doses.