| Objective:To explore the value of CT quantitative parameters in predicting future growth trends of pure ground glass nodules(pGGNs)in the lung using artificial intelligence software,aiming to help the clinic to differentiate indolent and potentially growing pGGNs early in follow-up,and then to provide a certain theoretical basis for individualized follow-up protocol development and surgical timing selection.Methods:A total of 105 single pGGNs that underwent at least 2 chest high-resolution CT(HRCT)examinations at our hospital between 2015.01 to 2020.12 were retrospectively included,and all pGGNs presented as single pGGNs with a diameter between 5-20 mm without malignant morphological signs at the initial CT examination,and their follow-up was required to be≥2 years,or the follow-up period was<2 years but pGGNs showed growth.Patients’ general data including gender,age,number of CT scans,follow-up time and pathological findings of pGGNs(only for those who underwent puncture or surgery)were recorded.Automated accurate segmentation and quantification of the initial and final CT images of the enrolled patients were performed using artificial intelligence(AI)software,and subsequently 14 quantitative parameter values of the initial CT of the pGGNs were obtained,including:initial volumetric features(three-dimensional diameter,volume,surface area,mass),histogram features(maximum CT value,minimum CT value,mean CT value,variance in CT value,CT value kurtosis,CT value skewness,entropy,energy)and morphological features(sphericity,compactness).For pGGNs grown at follow-up,volume doubling time(VDT)and mass doubling time(MDT)were calculated.All CT images were reviewed by two radiologists to observe the AI autosegmentation effect and the appearance of solid components within the pGGNs.Based on the growth of pGGNs,pGGNs were divided into growth and non growth groups,with growth defined as an increase in the volume of pGGNs by≥20%or the appearance of a solid component within it,and the remainder were classified into the non growth group,with more than 2 years of follow-up required for the non growth group.Logistic regression and Cox regression were used to determine independent predictors of pGGNs growth,predictive efficacy was assessed with receiver operating characteristic curve(ROC)and cut off values were determined.Kaplan Meier survival curves(cumulative risk curves)of pGGNs were plotted by the Kaplan Meier method,and differences between the curves were compared using the log rank test.Results:1.Comparison of general datas:a total of 105 patients,40 males(38.1%),65 females(61.9%),with an initial mean age(53.2±10.5)years,with a mean follow-up time(47.2±11.1)months,with a median of 5 CT scans.During follow-up,37 pGGNs showed growth(growth group),68 pGGNs showed no evidence of growth(non growth group),and the growth ratio of pGGNs was 35.2%(37/105).The median VDT in the growth group was 1901.8 days,and the median MDT was 1263.5 days.Surgical resection was performed in 18 cases(48.6%,18/37)in the growth group,and the pathology was all primary lung adenocarcinoma,and IAC was the most common(66.7%,12/18).In the non growth group,19 patients(27.9%,19/68)underwent surgical resection,and the pathology was mostly AAH and AIS(73.6%,14/19).Except for the growth group,which was more likely to undergo surgical resection(P=0.034),there were no significant differences in patient sex,age,number of CT scans,and duration of follow-up between the 2 groups(P<0.05).2.Univariate analysis of quantitative parameters in the initial CT of pGGNs:the differences in the initial volumetric characteristics(three-dimensional diameter,volume,surface area,and mass)between the two groups were statistically significant,and the initial three-dimensional diameter,volume,surface area,and mass of the pGGNs in the growth group were significantly greater than those in the non growth group(P<0.001).However,there were no statistically significant differences in other histogram features(maximum CT value,minimum CT value,kurtosis,skewness,entropy,energy)and shape features(sphericity,compactness)except for mean CT value(P<0.001)and variance in CT value(P<0.001)between the two groups.3.Multivariate analysis of quantitative parameters in initial CT of pGGNs both logistic regression and Cox regression analysis revealed that larger initial three-dimensional diameter,volume,mass and higher mean CT value were independent predictors of pGGNs growth(P<0.05).4.ROC analysis:the AUC values of initial three-dimensional diameter,volume,mass,and mean CT values for predicting pGGNs growth were 0.863,0.814,0.874,and 0.797,respectively,with cutoff values of 10.5mm,299.5mm3,105.3mg,and-681.6HU.The combination of the above four indicators improved the predictive efficacy with an AUC value of 0.927.5.Kaplan Meier survival curves(cumulative risk curves):the cumulative growth rate of pGGNs at 24 months,36 months,and 60 months was 4.8%,11.4%,and 33.3%,respectively.In the subgroup categorized by initial three-dimensional diameter(10.5 mm),volume(299.5 mm3),mass(105.3 mg),and mean CT value(-681.6 HU),the difference in the cumulative growth rate of pGGNs in each subgroup was statistically significant(P<0.001).Conclusion:1.Compared with traditional manual or semi-automatic measurements,AI can quickly and automatically provide quantitative data such as 3D volumes,histograms and shape characteristics of a large number of pGGNs,which can be a useful tool for assessing pGGN growth.2.PGGNs were slow growing with a median VDT of 1901.8 days and a median MDT of 1263.5 days,and approximately 35.2%of the pGGNs were within a relatively long follow-up period(47.2±11.1 months)appeared growth.The cumulative growth ratio of pGGNs at 24,36,and 60 months was 4.8%,11.4%,and 33.3%,respectively.Therefore,5-year follow-up of pGGNs is warranted.3.Quantitative parameters of initial CT of pGGNs can help to predict their future growth trends,and the initial three-dimensional diameter,volume,mass,and average CT values are independent predictors of pGGNs growth,and pGGNs with larger initial three-dimensional diameter(≥ 10.5 mm),volume(≥ 299.5 mm3),mass(≥ 105.3 mg),or higher average CT value(≥-681.6 HU)are easier to grow and should be monitored more intensively. |
PDF Full Text Request