深度学习图像重建算法在肝脏增强CT成像中改善图像质量及辐射剂量的应用价值  

The clinical value of deep learning reconstruction algorithm in image and radiation dose of contrast-enhanced liver CT examination

作　　者：杨硕 别依凡 庞国栋[1] 李行超 刘平[2] YANG Shuo;BIE Yifan;PANG Guodong;LI Xingchao;LIU Ping(Department of Radiology,The Second Hospital,Shandong University,Jinan 250033,China;Department of Radiology,The Second Hospital,Shandong University of Traditional Chinese Medicine,Jinan 250001,China)

机构地区：[1]山东大学第二医院医学影像中心 [2]山东中医药大学第二附属医院放射科

出　　处：《医学影像学杂志》2023年第5期785-789,共5页Journal of Medical Imaging

摘　　要：目的探讨深度学习图像重建算法(DLIR)对肝脏增强CT延迟期图像质量、辐射剂量的影响。方法选取因可疑肝肿块行腹部增强CT扫描患者70例,随机分为常规剂量组(A组)35例和低剂量组(B组)35例,对A、B两组患者延迟期数据分别进行30.0%迭代重建算法(ASIR-V 30.0%)、中等级DLIR(DLIR-M)、高等级DLIR(DLIR-H)重建,亚组分别命名为A_(AS-30)、A_(DL-M)、A_(DL-H),B_(AS-30)、B_(DL-M)、B_(DL-H)。比较A_(AS-30)、A_(DL-M)、A_(DL-H)算法间,B_(AS-30)、B_(DL-M)、B_(DL-H)算法间,以及A_(AS-30)与B_(DL-M)、B_(DL-H)算法间图像噪声、信噪比(SNR)、对比噪声比(CNR)及主观图像质量评分的统计学差异。结果在A组间和B组间,DLIR算法图像噪声、SNR、CNR和主观图像质量评分均优于ASIR-V 30.0%图像(均P<0.01),以DLIR-H图像噪声最低,SNR、主观评分最高。在有效辐射剂量降低81.0%时,BDL-M算法图像噪声、SNR、CNR与AAS-30算法差异无统计学意义(均P>0.05),但主观图像质量评分仍略高于A_(AS-30)算法(3.00±0.41 vs 2.32±0.47,P<0.01),B_(DL-H)算法图像噪声、SNR、CNR和主观图像质量评分均优于A_(AS-30)算法(均P<0.01),且B_(DL-M)、B_(DL-H)算法图像主观评分均能基本满足临床诊断需求(主观评分≥3分)。结论DLIR算法可显著提高肝脏增强CT图像质量,并可在保证临床诊断质量的同时,显著降低扫描辐射剂量。Objective To explore the influence on image quality and radiation dose in delay-enhanced phase of liver CT by using Deep learning reconstruction algorithm(DLIR). Methods Seventy patients with suspected hepatic masses underwent contrast-enhanced liver CT were enrolled in this prospective study. According to radiation dose, patients were randomly divided into group A(n=35, conventional-dose group) and group B(n=35, low-dose group). All delayed phase data of patients in groups A and B were reconstructed with 30.0% iterative reconstruction algorithm(ASIR-V 30.0%), middle-level DLIR(DLIR-M), and high-level DLIR(DLIR-H), respectively;and subgroups were named A_(AS-30), A_(DL-M), A_(DL-H) and B_(AS-30), B_(DL-M) B_(DL-H). Objective image quality including image noise, signal-to-noise(SNR), and the contrast-to-noise ratio(CNR), as well as subjective image scores was compared pairwise between A_(AS-30), A_(DL-M), A_(DL-H) algorithms, B_(AS-30), B_(DL-M), B_(DL-H) algorithms and A_(AS-30), B_(DL-M), B_(DL-H) algorithms. Results Either in group A or in group B, DLIR algorithms had better image noise, SNR, CNR, and subjective image quality score than ASIR-V 30.0%(all P<0.01), with DLIR-H having the lowest noise, the highest SNR and subjective image scores. With the effective dose reduced by 81.0%, low-dose DLIR-M algorithm had similar image noise, SNR and CNR compared to conventional-dose ASIR-V 30.0%(all P>0.05), but subjective image scores were still higher(3.00±0.41 vs 2.32±0.47, P<0.01). Low-dose DLIR-H algorithm had better image noise, SNR and CNR than conventional-dose ASIR-V 30.0%(all P<0.01). Regarding the subjective image quality, B_(DL-M) and B_(DL-H) algorithm images both can basically meet the needs of clinical diagnosis(subjective score ≥3 point). Conclusion DLIR algorithm can not only improve effective image quality, but also decrease radiation dose in liver CT examination, while ensuring the quality of clinical diagnosis.

关 键 词：肝脏 体层摄影术 X线计算机 深度学习重建 图像质量 辐射剂量 

分 类 号：R575[医药卫生—消化系统] R814.42[医药卫生—内科学]