Study On The Effect Of Reconstruction Algorithm On The Quality Of Chest CT Image And Pulmonary Nodules Under Ultra-low Dose:A Phantom Study

Part One The chest CT image by ultra-low dose combined with iterative reconstruction algorithm:A phantom studyObjective:Using the latest GE revolution 256 slice CT scanner,and adopt high resolution chest scanning mode,to explore the low dose of Adaptive statistical iterative reconstruction-V in improving the value of high resolution chest CT image quality.Materials and methods:With GE Revolution CT,tube voltage 70 kV,respectively with 10 mA、20 mA、30 mA、40 mA、50 mA tube current scanning chest phantom,thick 0.625 mm combining adaptive statistical iterative reconstruction-V 50%(ASIR-V 50%)reconstruction high definition standard(HD-stnd)and high definition lung(HDlung),and subjective image(1-5,unable to diagnose-good)and objective evaluation(signal to noise ratio(SNR),contrast to noise ratio(CNR)mental image noise(SD1),the anterior chest wall noise(SD2),dose length product(DLP)and effective radiation dose(ED)were counted to find the combination of the lowest threshold value,in order to find the minimum threshold to meet clinical combination;Reconstruction of HDlung images with different weights ASIR-V0、30%、50%、80%and 100%under the threshold;objective and subjective scoring was performed to find the optimal iteration reconstruction range.Results:(1)Under constant tube voltage(70 kV),The difference of SD1 at the aver age level(10 mA、20 mA、30 mA、40 mA、50 mA)of different tube curren ts was statistically significant(F=118.893,P=0.000,P<0.05),SNR、CNR and S D2 had no statistical significance(respectively F=2.546、F=2.264、F=0.671,P>0.05);Pairwise comparison,according to the level of α ’=0.005,SD1 had statis tical differences in(10 mA and 20 mA(t=5.756),10 mA and 30 mA(t=15.563),10 mA and 40 mA(t=15.031),10 mA and 50 mA(t=18.228),20 mA and 30 mA(t=9.087),20 mA and 40 mA(t=9.274);20 mA and 50 mA(t=12.472))(after correction P<0.005),the subjective score consistency was equal and ca ntinuously improved(Kappa=0.701,P<0.05).In summary,SD1 values showed statistical difference,while CNR、SNR and SD2 showed no difference.In the pairwise comparison of SD1 values,there was no statistical difference between pairwise comparison of SD1 values at 30 mA、40 mA and 50 mA,and there was a difference between 10 mA、20 mA and 30 mA,with 30 mA being the best.Therefore,30 mA was considered as the lowest diagnostic threshold;(2)Under the 70 kV、30 mA,The image quality(SNR、CNR、SD1、SD2)of AS IR-V algorithm with different weight ratios had statistical significance(respecti vely F=3376.619、F=11.453、F=1182.877、F=5028.646,P<0.05),SNR,SD1 an d SD2 showed statistical significance in pairwise comparison between comparis on groups with different weights of ASIR-V(after correction P<0.005),CNR v alues were only different between 0 and 100%(t=6.298,after correction P<0.005).The subjective score of the image was improved gradually at 0、30%>50%,at 50%、80%、100%,the subjective score was decreased gradually,and th e image consistency was equal(Kappa=0.609,P=0.047).Conclusion:The chest high-resolution CT adopts the reconstruction algorithm of 70 kV、30 mA combined with ASIR-V 50%-80%,which can reduce the radiation dose and meet the diagnostic requirements.Part Two Effect of deep learning reconstruction on image quality and pulmonary nodules in the chest CT:a phantom studyObjective:Using the latest GE revolution 256 slice CT scanner,and adopt high resolution chest scanning mode,to explore the effect of deep learning reconstruction on the quality of high-resolution chest CT images and the performance of nodules detection rates under ultra-low dose and its clinical application value.Materials and methods:To choose GE Revolution 256 CT,tube voltage:70 kV,with different tube current(10 mA、20 mA、30 mA、40 mA and 50 mA)(embedded with 9 standard nodules of different densities(+100 HU、-630 HU、-800 HU)and different sizes(8 mm、10 mm、12 mm).After the scanning was completed,a 0.625mm layer thickness was adopted.Four reconstruction algorithms(DLIR-L、DLIR-M、DLIR-H and ASIR-V 50%)were adopted to generate different sequences,the image import GE aw4.7 workstation,in the case of two kinds of high definition scanning(HD-stnd)Lung window(HD-lung),image quality,and the image quality and pulmonary nodules detection rate,The ASIRV 50%was compared with DLIR-L、DLIR-M、DLIR-H to explore the value of deep learning algorithm in image quality and pulmonary nodules detection rates.Results:(1)Objective scores:When the tube voltage was consistent with the reconstruction algorithm,the objective scores of image quality(SNR、CNR、SD)were continuously optimized with the increase of the tube current.When the reconstruction was ASIR-V 50%,the objective scores of image quality(SNR,CNR、SD)were not statistically significant betweenthegroups(10mA、20mA、30mA、40mA、50mA)(P>0.05).The objective scores of image quality(SNR>CNR,SD)in the same reconstruction conditions(DLIR-L>DLIR-M and DLIR-H)were statistically significant between groups(10 mA、20 mA、30 mA、40 mA、50 mA)(P<0.05);The tube voltage and tube current were constant,In the same tube current(10 mA、20 mA、30 mA、40 mA、50 mA),the objective scores of image quality(SNR、CNR、SD)of different reconstruction algorithms(ASIR-V 50%、DLIR-L、DLIR-M and DLIR-H)were statistically different,and all of them were statistically significant(P<0.05).(2)Subjective scores:subjective scores for overall image quality,other condition is constant,the subjective rating of image with the constant improvement of the reconstruction algorithm(ASIR-V 50%、DLIR-L、DLIR-M、DLIR-H)and improve,under the condition of the DLIR-H reconstruction subjective score best,two doctors to equal the subjective rating of reconstruction image consistency(10 mA、20 mA、30 mA、40 mA and 50 mA of Kappa valuerespectively were 0.471、0.623、0.539、0.401 and 0.450,all P<0.05);Subjective score of pulmonary nodules:under constant tube voltage and reconstruction algorithm,with the increase of tube current,the detection rate of nodules increased gradually;When the tube voltage and tube current are constant,the detection rate of nodules changes with the improvement of reconstruction algorithm.Tube voltage and tube current under a constant,for subjective ratings of pulmonary nodules,such as image noise and texture of the diagnosis of pulmonary nodules confidence,nodules of sharpness and significance of nodules,with the improvement of reconstruction algorithm,subjective scores also gradually improve,two doctors for lung nodules consistency of subjective rating equal(10 mA、20 mA、30mA、40 mA and 50 mA,the Kappa value respectively were 0.490、0.571、0.500、0.483 and 0.498,all P<0.05).(3)The detection rate of pulmonary nodules under the four reconstruction algorithms reached 9/9(100.00%)at 70 kV、30 mA,and the subjective score of overall image quality at 70 kV、30 mA reached 3 or above within the four reconstruction algorithms,which met the clinical diagnostic requirements of images.The subjective score of pulmonary nodules reached the diagnostic requirements at 70 kV、50 mA.Under the condition of 70 kV and 50 mA,SNR、CNR and SD values were obj ectively scored.The DLIR-H was better than ASIR-V 50%,and there was a difference between ASIR-V 50%and DLIR-H(All after correction P<0.008),compared with ASIR-V 50%,SNR increased by 50.00%,CNR increased by 50.29%and SD decreased by 30.83%under DLIR-H.(4)Under the condition of 70 kV、50 mA,the objective evaluation indexes(SNRN,CNRN,SDN)of the three density pulmonary nodules(+100 HU、-630 HU、-800 HU)of 12 mm were different under different reconstruction algorithms(P<0.05),and the comprehensive evaluation was the best obtained by DLIR-H.(5)Under the condition of 70 kV、50 mA,the objective measurement values(VN、DN及CTN)of different densities(+100 HU、-630 HU、800 HU)of 8 mm diameter were different under different reconstruction algorithms(P<0.05).Under ’the DLIR-H,the APE values were the best under different reconstruction algorithms[+100 HU:VN(6.21%)、DN(7.50%)and CTN(32.20%),-630 HU:VN(3.00%)、DN(7.50%)and CTN(0.75%),-800HU:Vn(6.17%)、DN(8.75%)and CTN(0.20%)].Conclusion:For the diagnosis of pulmonary nodules,compared with ASIR-V 50%,chest highresolution CT is recommended to combine the deep learning reconstruction DLIR-H under the condition of 70 kV、50 mA,so as to meet the clinical diagnostic needs of pulmonary nodules at ultra-low dose.