CT Quantitative Analysis Of Early-stage Lung Adenocarcinoma And The Application Of Computer Aided Detection For Early Lung Cancer

Part ? Quantitative CT Analysis of Early-stage Lung Adenocarcinoma with Pure Ground-Glass opacity?Objective?To discuss if quantitative analysis of early-stage lung adenocarcinoma manifesting as a pure ground-glass nodule(pGGN)on computed tomography(CT)can predict its pathological grading.?Materials and Methods?One hundred and five patients who had undergone curative resection for lung adenocarcinoma,manifesting as a pure ground-glass nodule,were enrolled in this retrospective study.Among 104 lesions,there are 19 typical adenomatous hyperplasia(AAH),25 adenocarcinoma in situ(AIS),31 minimally invasive adenocarcinoma(MIA)and 29 invasive adenocarcinoma(IAC).We evaluated all CT images on using computer aided diagnosis software,and all pGGNs were analyzed as follows: long diameter and area of maximum section,volume,mean CT value,mass,minimum CT value,maximum CT value,and 2%,5%,25%,50%,75%,95%,98%percentile CT value.variables between different pathological grades satisfying the law of normal distribution and homoscedasticity were compared using analysis of variance(AVOVA),other variables were compared using Kruskal—Wallis H test.Logistic regression analysis was performed by taking if pGGN is invasive lesion as the dependent variable,and variables with statistical differences were taken as independent variables.?Results?The lesion size(including long diameter and area of maximum section,volume),mean CT value,mass,50%,75%,95%,98% percentile CT value and maximum CT value were significantly statistical different between four pathological types of AAH?AIS?MIA and IAC(P<0.01).Between preinvasive lesion and invasive lesion,The lesion size,mean CT value,mass,50%,75%,95%,98% percentile CT value and maximum CT value were also significantly statistical different(P<0.01).Logistic regression analysis showed that the maximum CT value(OR=1.005,95% CI 1.004~1.010,P=0.001)and the long diameter of maximum section(OR=1.143,95% CI 1.027~1.273,P=0.015)predicted if pGGN is invasive lesion,the ROC curve analysis was performed based on the predicted probability of Logistic regression model,and the area under the curve was0.793(P=0.000).For the maximum CT value,diameter of maximum section and mass,under the curve were 0.759?0.754 and 0.822(P<0.01);The cutoff value were-369 HU,11.5mm,120 mg,respectively.?Conclusion?The research show that quantitative analysis of early-stage lung adenocarcinoma manifesting as a pure ground-glass nodule on CT to get its size,the maximum CT value and mass,can be useful for predicting pathological grading.Part II Acomparative study of three different methods for measuring the pure ground glass nodule?Objective?To compare the differences between three different methods: conventional Single slice measurement,Computer automatic segmentation measurement and Manual layer-by-layer outline measurement,for evaluating pure ground glass nodules,and to evaluate the clinical applicability of computer automatic nodule segmentation.?Materials and Methods?50 cases of pure ground glass nodules in part I were measured using 3 methods:(1)Manual measurement :the doctor manually measure the long diameter and average CT value on the maximum surface.(2)Automatic measurement: the doctor signs the center of the nodule,and the computer automatic segmentation of nodule,gets the maximum surface length diameter,volume,and average CT value of three-dimensional volume.(3)Precise measurement: the doctor by layer outline of the nodule edge,then the computer calculates to get the maximum surface length diameter,volume,and average CT value of three-dimensional volume.Sets precise measurement result as the reference standard,use intraclass correlation coefficient(ICC)to evaluate the consistency of the measurement of long diameter,mean CT value between three methods,and use paired samples T test to compare the differences between parameters with reliability below good.?Results?Manual measurement of long diameter and precise measurement of consistency is good,ICC=0.979.Computer automatic measurement of long diameter,volume and precise measurement of consistency is Acceptable.ICC=0.821 and 0.963 respectively.The average CT value of manual measurement(ICC=0.324)is not consistent with the precise measurement.Computer automatic measurement of the average CT value has general consistency with precise measurement.The average CT value of the manual measurement and the average CT value of the computer automatic measurement are larger than those of the precise measurement(P<0.001;P=0.004).?Conclusion?Nodules' long diameter measured manually can be instead of precise 3D measurement.nodule computer automatic measurement can evaluate approximately pGGN's long diameter and volume.the average CT value of the manual measurement and automatic measuring is higher than precise measurement,accurate measurement of the CT value requires measurement by layers to outline.Part III The Application Value of Computer Aided Detection System Based on LIDC/IDRI Database?Objective?Using a world public medical image database,explore the clinical application value of computer aided detection software to help doctors to read CT images,to improve the detection rate of pulmonary nodules.?Materials and Methods?Downloaded and used 40 cases of chest CT scan image from LIDC/IDRI database,selecting the unenhanced with under 3mm thickness of 29 cases.Two junior radiology doctors with 2 years working experience read 29 cases,nodules were recorded separately in the mode of independent reading and with reference by computer aided detection software.A careful study was taken on nodules identified in the annotation document,all nodules more than 3 radiologists identified set as a reference standard,recorded the density,size,and location of the reference nodules at the same time.Counted the detection rate of nodules in independent reading,and with the use of computer-aided detection software,and compare the sensitivity of different types of nodules in different mode using CADe software and false its positive rate.?Results?A total of 127 nodules reference standard.The software automatic detection rate was 65.4% of all nodule.Two doctors' independent reading detection rate was found in 58.3%,56.7%,and after the reference of CADe automatic reading were increased to 79.5% and 85%(P<0.001).The detection rate of the nonsolid nodule detected by the software is lower than that of the solid(P=0.035).For nodules less than 4mm automatic detection and two physicians independent read the detection rate is relatively low,respectively,46.4%,39.3%,46.4%,but combined the rate has increased to 60.7%,73.2%.For more than 5mm nodule doctors and software sensitivity were respectively high.Two doctors had a low detection rate of hilar nodule,and the sensitivity with CADe was significantly increased.The number of false positive of automatic detection per scan is 1.0,with IQR of 1.5.?Conclusion?This study based on the LIDC/IDRI database and its annotation data,shows that the new CADe software help the junior doctors to improve the detection rate of lung nodules,maintaining a low false positive rate.