深度学习重建算法在超高分辨力颅脑CT中的图像质量改善与剂量降低研究OA

Objective:This study aimed to investigate the effect of ultra-high-resolution(UHR)detector computed tomography(CT)combined with a deep learning reconstruction algorithm(ClearInfinity(CI))on cranial CT image quality and its potential for radiation dose reduction.Methods:A NeuViz Epoch Elite CT scanner was used to scan a Catphan 600 phantom(with volume CT dose index(CTDIvol)set to 50,37.5,and 25 mGy)and three rhesus monkeys(CTDIvol=50 mGy).The collimation width was 128×0.312 5 mm.Images were reconstructed using filtered back projection(FBP),adaptive iterative reconstruction(ClearView,CV30%and CV60%),and deep learning reconstruction(ClearInfinity,CI30%and CI60%).Image quality was evaluated using objective metrics,such as modulation transfer function(MTF),contrast-to-noise ratio(CNR),and artifact severity,as well as double-blind subjective scoring on a 5-point scale.Statistical analyses were then performed.Results:(1)Phantom experiments:At all dose levels,the CNR increased significantly with higher reconstruction levels,with the CI60%images showing a significantly higher CNR than the other algorithms.At 25 mGy,the CNR of CI60%was comparable to that of FBP at 50 mGy,and no significant decrease was observed for MTF10%or MTF50%.(2)Animal experiments:At the centrum semiovale level,the CNR of the CI60%images was significantly higher than that obtained with other algorithms,and artifacts tended to decrease with increasing iteration levels.Inter-observer agreement for image quality assessment was good(Kappa≥0.75).Overall,the subjective scores increased with higher CV/CI levels,with CI60%achieving the highest scores.Conclusion:In UHR detector CT,deep learning reconstruction can improve cranial CT image contrast and reduce noise and artifacts without compromising high-contrast spatial resolution,showing significant potential for radiation dose reduction and demonstrating good clinical application value.