Image Quality Assessment of the External Carotid Artery and Its Branches on Ultra-High-Resolution Head and Neck Computed Tomography Angiography Using a High-Resolution 0.25-mm Detector and Deep Learning Reconstruction

BACKGROUND AND PURPOSE: Deep learning reconstruction can improve image quality of CTA, but its benefit for visualizing small-caliber external carotid artery branches on ultra-high-resolution CTA remains unclear. We evaluated the image quality of ultra-high-resolution CTA of the external carotid artery using filtered back projection, hybrid iterative reconstruction, and deep learning reconstruction, and assessed its added value for visualizing small-caliber external carotid artery branches and tumor-feeding vessels. MATERIALS AND METHODS: In this single-center study, we retrospectively analyzed 24 patients who underwent ultra-high-resolution CTA for evaluation of head/neck tumors or an elongated styloid process. Axial images (0.25-mm slice thickness) were reconstructed with each method. The quantitative analyses included attenuation, image noise, SNR, contrast-to-noise ratio, edge rise distance, and edge rise slope for the external carotid artery trunk and 11 branches. Two radiologists (each with 8 years of experience) independently assessed overall image quality, peripheral vessel sharpness, artifact severity, and clarity of the tumor-margin and intratumoral vessels, using a four-point Likert scale. Objective and subjective metrics were compared using nonparametric repeated-measures tests. Interobserver agreement was assessed using quadratic-weighted κ-value. RESULTS: Deep learning reconstruction provided the lowest image noise, highest SNR, and highest contrast-to-noise ratio for the external carotid artery trunk (all P <.001). For the 11 branches, deep learning reconstruction demonstrated the shortest edge rise distance (0.795 [0.751-0.824] mm vs. 0.982 [0.942-1.017] for hybrid iterative reconstruction and 1.044 [1.002-1.157] for filtered back projection), the highest edge rise slope(613.4 [586.1-686.5] vs. 393.1 [366.7-442.6] and 477.4 [446.9-524.7]), and the highest contrast-to-noise ratio (40.6 [36.0-43.2] vs. 16.7 [14.9-20.2] and 9.8 [8.3-11.0]) (all P <.001). Subjective scores were highest for deep learning reconstruction across all categories (median overall image quality: 4 for deep learning reconstruction, 3 for hybrid iterative reconstruction, and 2 for filtered back projection). Deep learning reconstruction also provided significantly superior visualization of tumor-margin and intratumoral vessels (P <.001). Interobserver agreement was substantial for all qualitative metrics (κ = 0.67-0.76). CONCLUSIONS: Deep learning reconstruction markedly improves the quality of ultra-high-resolution CTA images of the external carotid artery system by enhancing vessel sharpness, contrast resolution, and tumor-feeding vessel conspicuity, although further validation in larger cohorts is needed to confirm these findings.