目的 探讨基于压缩感知(compressed sensing-Sensitivity Encoding,CS-SENSE)技术,人工智能重建(artificial intelligence reconstruction)对头颅3D FLAIR图像的影响.方法 收集2021年10月-12月在南京大学医学院附属鼓楼医院检查头颅3D FLAIR显示异常高信号的患者共43例.采用不同压缩因子(4和8)获取3D FLAIR图像,分别以常规重建和人工智能重建方法重建出CS4、CS8、CS_AI4、CS_AI8四种图像.采用单因素重复测量方差分析法定量比较不同序列图像信噪比(signal to noise ratio,SNR)和对比噪声比(contrast to noise ratio,CNR).两位具有5年以上神经影像诊断经验的医师对图像质量进行定性评分,采用Kappa检验判断两名医生对图像质量定性评估的一致性.结果 CS_AI4(SNRLesions:801.32±318.79,SNRWM:463.50±209.23,CNR:337.83±158.30)及CS_AI8(SNRLesions:887.94±445.27,SNRWM:500.99±261.71,CNR:386.95±224.98)的CNR 及SN R 显著优于 CS4(SN RLesions:553.53±135.79,SNRWM:320.86±93.46,CNR:232.67±91.67)及CS8(SN RLesions:482.50±132.29,SNRWM:279.41±85.92,CNR:203.09±86.52)(P值均＜0.001).CS4的CNR及SNR显著优于CS8(P＜0.001).定性分析结果显示,CS_AI4(8.87±0.40)、CS_AI8(8.55±0.63)及CS4(8.34±0.82)之间无统计学差异(P均＞0.008),但是显著优于CS8(5.07±0.78,P均＜0.001).结论 Al重建之后的3D FLAIR图像信噪比及对比噪声比显著优于常规重建的FLAIR图像.
Optimization of Intracranial 3D FLAIR Image Based on Artificial Intelligence Reconstruction Algorithm
Objective To explore the effect of different reconstruction methods(conventional reconstruction and artificial intelligence(Al)reconstruction)on 3D FLAIR brain image based on the study of Compressed Sensing-Sensitivity Encoding(CS-SENSE)technology.Methods A total of 43 patients with high signal on 3D FLAIR were enrolled in the Affiliated Drum Tower Hospital of Nanjing University Medical School from October to December 2021.Different compression factors(4 and 8)were used to obtain 3D FLAIR images,and CS4,CS8,CS_AI4 and CS_AI8 images were reconstructed using conventional and Al reconstruction,respectively.Single factor repeated measure ANOVA was adopted to analyze the differences of the signal to noise ratio(SNR)and contrast to noise ratio(CNR)in different sequences.Two neuroradiologists scored theimages in each sequence,and Kappa test was used to calculate the consistency.Results CNR and SNR of CS_AI4(SNRLesions:801.32±318.79,SNRWM:463.50±209.23,CNR:337.83±158.30)and CS_AI8(SNRLesions:887.94±445.27,SNRWM:500.99±261.71,CNR:386.95±224.98)were significantly higher than those of CS4(SNRLesions,553.53±135.79,SNRWM:320.86±93.46,CNR:232.67±91.67)and CS8(SNRLesions:482.50±132.29,SNRWM:279.41±85.92,CNR:203.09±86.52)(P＜0.001).CNR and SNR of CS4 were significantly higher than those of CS8(P＜0.001).The qualitative analysis results showed that CS_AI4(8.8710.40),CS_AI8(8.55±0.63)and CS4(8.34±0.82)had no statistical difference(P＞0.008),but were significantly higher than CS8(5.07±0.78,P＜0.001).Conclusion The SNR and CNR of Al reconstructed 3D FLAIR images are significantly higher than those of conventional reconstructed FLAIR images.
NETL
NSTL
万方数据
