Application value of the deep learning-based image reconstruction algorithm in combined head and neck CT angiography with low radiation dose / 中华放射医学与防护杂志

ABSTRACT

Objective:To explore the differences between the deep learning-based image reconstruction (DLIR) and the adaptive statistical iterative reconstruction V (ASiR-V) algorithms in the radiation dose and image quality of head and neck CT angiography (CTA).Methods:The data of 80 patients undergoing head and neck CTA due to vascular diseases in the head and neck were prospectively collected. These patients were randomly divided into groups A and B based on their examination sequence. The CTA images of group A were reconstructed based on ASiR-V 50%, with a tube voltage of 120 kV and a noise index of 11.0. In contrast, those of group B were reconstructed based on ASiR-V 50% (for group B1) and DLIR-H (for group B2), with a tube voltage of 80 kV and a noise index of 9.0. Then, the radiation doses and image quality of both groups were compared using the independent-sample t-test. The radiation doses, and both subjective and objective image quality of the two imaging method were compared through the Kruskal-Wallis test and the Wilcoxon rank-sum test. The independent- or paired-sample t-test was employed to measure inter-group vascular enhanced CT values, as well as signals and noise from regions of interest (ROIs), with signal-to-noise ratios (SNRs) and contrast-to-noise ratios (CNRs) calculated. Results:The effective doses of groups A and B were (0.77±0.08) and (0.45±0.05) mSv, respectively, with a statistically significant difference ( t = 21.96, P < 0.001). The vascular enhanced CT values, SDs, SNRs, and CNRs in the arch of the aorta, the initial and bifurcation parts of the common carotid artery, and the M1 segment of the middle cerebral artery showed statistically significant differences among groups A, B1, and B2 ( F = 67.69, 68.50, 50.52, 74.10, 63.10, 91.22, 69.16, P < 0.001). Additionally, statistically significant differences were observed in the subjective scores of image quality among groups A, B1, and B2 ( Z = 71.06, P < 0.05). Conclusions:The DLIR algorithm can further reduce the radiation dose in head and neck CTA examination while significantly reducing image noise and ensuring image quality, thus demonstrating high clinical application value.