1024×1024重建矩阵结合Karl迭代重建算法对肾上腺CT图像质量和自动分割的影响  

The effect of 1024×1024 reconstruction matrix combined with Karl iterative reconstruction algorithm on adrenal gland image quality and auto-segmentation in CT images

作　　者：王诗耕 刘义军[1] 范勇 童小雨 魏巍 姜艳[1] WANG Shigeng;LIU Yijun;FAN Yong;TONG Xiaoyu;WEI Wei;JIANG Yan(Department of Radiology,the First Affiliated Hospital of Dalian Medical University,Dalian,Liaoning Province 116011,China)

机构地区：[1]大连医科大学附属第一医院放射科,辽宁大连116011

出　　处：《实用放射学杂志》2024年第8期1358-1362,共5页Journal of Practical Radiology

摘　　要：目的探讨大重建矩阵(1024×1024)结合Karl迭代重建算法对肾上腺CT图像质量和自动分割的影响。方法回顾性选取40例行肾上腺CT增强的患者。扫描完成后,对静脉期的原始数据进行重建分组,A组采用常规重建矩阵(512×512)结合Karl 5级进行重建;B组采用1024×1024重建矩阵结合不同等级的Karl(5、7、9级)进行重建,记为B1~B3组。2名观察者采用5分法评估各组肾上腺显示情况、图像整体质量和自动分割情况。在各组轴位图像上测量肾上腺、同层下腔静脉的CT值和标准差(SD)值,并计算信噪比(SNR)、对比噪声比(CNR)。采用基于深度学习(DL)的器官分割模型对各组重建图像的肾上腺进行分割,并计算Dice系数和体积差别率。结果在肾上腺显示、图像整体质量和自动分割情况方面,B2组得分最高且均优于A组(P<0.05)。随着Karl等级的提高,B组肾上腺、下腔静脉的SD值逐渐下降(P<0.05),SNR、CNR逐渐升高(P<0.05),其中B2组和A组SD值、SNR和CNR差异均无统计学意义(P>0.05)。所有组别Dice系数均>0.90,B组内的肾上腺(双侧)体积差别率均<5%,且低于A组(P<0.05)。结论采用1024×1024重建矩阵结合Karl 7级能够在不影响肾上腺图像分割的同时优化肾上腺图像质量。Objective To explore the effect of large reconstruction matrix(1024×1024)combined with Karl iterative reconstruction algorithm on adrenal gland image quality and auto-segmentation in CT images.Methods Retrospective analysis was performed on 40 patients with adrenal gland CT enhancement.After scanning,the original data of venous phase images were reconstructed and grouped.Group A was reconstructed using conventional 512×512 matrix combined with Karl 5;group B was reconstructed using 1024×1024 reconstruction matrix combined with Karl iterative reconstruction algorithm(level 5,7,9)of different levels,and was denoted as group B1-B3.Two radiologists assessed the display of adrenal glands,overall image quality,and auto-segmentation stability on a 5-point scale.The CT and standard deviation(SD)values of the adrenal gland and inferior vena cava in each group were measured.The signal-to-noise ratio(SNR)and contrast-to-noise ratio(CNR)were calculated.The deep learning(DL)-based organ segmentation model was used to segment the adrenal glands of each group of reconstructed images.The Dice coefficients and the volume difference rates were calculated.Results In terms of the display of adrenal glands,overall image quality,and auto-segmentation stability,group B2 had the highest score and was better than group A(P<0.05).With the increase of Karl levels,the SD values of adrenal gland and inferior vena cava in group B gradually decreased(P<0.05),SNR and CNR gradually increased(P<0.05),and there was no significant difference between group B2 and group A in SD value,SNR and CNR(P>0.05).Dice coefficients of all groups>0.90.The volume difference rate of adrenal gland(both sides)in group B was less than 5%,which was lower than that in group A(P<0.05).Conclusion The use of a 1024×1024 reconstruction matrix combined with Karl 7 is able to optimize the quality of the adrenal gland images without compromising the segmentation accuracy of the adrenal gland images.

关 键 词：重建矩阵 迭代重建算法 深度学习 肾上腺 

分 类 号：R814.42[医药卫生—影像医学与核医学] R335[医药卫生—放射医学]