基于深度学习重建算法的低剂量CT脑灌注扫描可行性研究

Feasibility of low-dose CT brain perfusion scanning based on deep learning reconstruction algorithm:a preliminary study

目的比较不同扫描条件下全脑CT灌注检查的图像质量和诊断参数,评估深度学习图像重建算法(DLIR)在降低管电流低剂量扫描中的应用价值。方法前瞻性收集2022年3月至2023年3月郑州大学第一附属医院疑似急性缺血性脑卒中(AIS)患者105例并记录其基线信息。所有患者均行头颅CT平扫及CT灌注(CTP)检查,CTP扫描条件分别为80 kV,分为150(常规剂量)和100 mA(低剂量)两组。将150 mA组CTP图像采用滤波反投影算法(FBP)、全模型迭代重建算法(ASIR-V)40%和80%强度进行重建,分别命名为A、B、C组;将100 mA组CTP图像分别采用ASIR-V80%、DLIR-M和DLIR-H进行重建,分别命名为D、E、F组。对不同扫描条件下两组患者进行临床基线特征及辐射剂量的比较,并比较6组重建CTP图像的主客观图像质量、常规灌注参数及AIS患者的异常灌注参数。结果150和100 mA扫描条件下两组确诊AIS患者分别为47例、48例,两组患者基线特征差异均无统计学意义(P>0.05);平均有效剂量分别为5.71、3.80 mSv,(t=2768.30,P<0.001)。6组重建图像中灰质及白质区域的噪声(SD)、信号噪声比(SNR)及对比噪声比(CNR)差异有统计学意义(F=40.58~212.13,P<0.001)。脑灰质区域,C、D和F组的SD值低于其他各组(P<0.05),C组和F组的SNR值高于其他各组(P<0.05);脑白质区域,C和F组的SD、SNR值与其他各组间差异均有统计学意义(P<0.05)。C和F组的CNR值也高于其他各组(P<0.05)。B、C和F组的主观评分差异无统计学意义(P>0.05)。灌注参数比较中,脑灰质中D和E组的脑血容量(CBV)值均低于A~C组(P<0.05),F组的CBV值低于B组(P<0.05);脑白质中D组的平均通过时间(MTT)值均低于其他各组(P<0.05)。6组图像的AIS病灶检出率及相关灌注参数差异无统计学意义(P<0.05)。结论低管电流CTP扫描结合DLIR-H算法可在不影响灌注参数(包括CBF和MTT)的前提下提高图像质量,减少30%的辐射剂量,可将此算法常规应用于脑CTP检查中。

Objective To compare image quality and diagnostic parameters of whole-brain CT perfusion scans under different scanning conditions and assess the utility of deep learning image reconstruction algorithm(DLIR)in reducing tube current during low-dose scans.Methods Method A total of 105 patients with suspected acute ischemic stroke(AIS)were prospectively enrolled in the First Affiliated Hospital of Zhengzhou University from March,2022 to March,203 and their baseline information was recorded.All patients underwent head non-contrast CT and CT perfusion(CTP)examinations.CTP scanning was performed at 80 kV in two groups with the tube current of 150 mA(regular dose)and 100 mA(low dose),respectively.The CTP images of 150 mA group were reconstructed using filtered back-projection algorithm as well as adaptive statistical iterative reconstruction-V(ASIR-V)at 40%and 80%strength levels,which were denoted as groups A-C.The CTP images of 100 mA group were reconstructed using ASIR-V80%,DLIR-M,and DLIR-H,which were denoted as groups D-F.Clinical baseline characteristics and radiation doses were compared between the two groups under different scanning conditions.Furthermore,we assessed the subjective and objective image quality,conventional perfusion parameters,and abnormal perfusion parameters of AIS patients across the six groups of reconstructed CTP images.Results Under the scanning conditions of 150 mA and 100 mA,47 and 48 patients were diagnosed with AIS,respectively.There were no significant differences in the baseline characteristics between the two groups.However,there was a significant difference in the mean effective radiation dose(5.71 mSv vs.3.80 mSv,t=2768.30,P<0.001).The standard deviation(SD)of noise,signal-to-noise ratio(SNR),and contrast-to-noise ratio(CNR)of gray matter(GM)and white matter(WM)were significantly different among the six groups of reconstructed images(F=40.58-212.13,P<0.001).In GM,the SD values in groups C,D,and F were lower than those in other groups(P<0.05),and the SNR values in groups C and F were higher than those in other groups(P<0.05).In WM,the SD and SNR values in groups C and F were significantly different from those in other groups(P<0.05).Additionally,CNR values in groups C and F were higher than those in other groups(P<0.05).There was no significant difference in subjective scores among groups B,C,and F(P>0.05).Regarding perfusion parameters in the brain GM,groups D and E had lower cerebral blood volume(CBV)values compared to groups A to C(P<0.05),and group F had lower CBV values than group B(P<0.05).In the brain WM,group D had consistently lower mean transit time(MTT)values compared to the other groups(P<0.05).Notably,there were no significant differences in AIS lesion detection rates and relevant diagnostic parameters across the six image groups.Conclusions Low-tube current CTP scan combined with the DLIR-H algorithm can enhance image quality without affecting perfusion parameters such as CBV and MTT,while reducing radiation dose by 30%.This algorithm can be routinely applied in brain CTP examinations.