Application Of Energy Spectrum CT Combined With Artificial Intelligence To The Invasion Of Ground Glass Nodular Lung Adenocarcinoma

Objective: To investigate the predictive value of quantitative parameters of energy spectrum CT combined with artificial intelligence(AI)on the invasion of ground-glass nodules(GGN)lung adenocarcinoma.Methods: A total of 113 patients with lung ground-glass nodule were collected from October 2021 to October 2022 in the First Affiliated Hospital of Shihezi University,who underwent energy spectrum CT for plain and enhanced scans,and had pathologically confirmed lung adenocarcinoma.According to the pathological results,the patients were divided into two groups,a non-invasive group,including atypical adenomatous hyperplasia(AAH),adenocarcinoma in situ(AIS)and minimally invasive adenocarcinoma(MIA),and the invasive adenocarcinoma(IAC)was divided into an invasive group.The quantitative parameters of energy spectrum CT included water concentration(WC)and slope of energy spectrum curve(K value)in three phases;iodine concentration(IC),normalized iodine concentration(NIC)both in arterial phase and venous phase;The quantitative parameters of AI included 3D length diameter,maximum surface area,surface area,quality,volume,mean CT value.The correlation between parameter and the invasion of GGN lung adenocarcinoma was analyzed by single-factors and multiple-factors analysis,and the parameters with statistical differences between the two groups were selected.The energy spectrum CT prediction model,AI prediction model and spectral CT + AI combined prediction model were constructed and the prediction efficiency of each model was analyzed.A Nomogram based on the combined prediction model were assessed by C-index,calibration curve and decision curve analysis for accuracy,consistency and clinical utility.Results:(1)There were significant differences between the non-invasive group and invasive group in the slope of energy spectrum curve of non-contrast scan(K-NC),water concentration of non-contrast phase(WC-NC),water concentration in arterial phase(WC-AP),and water concentration in venous phase(WCVP)among the quantitative parameters of energy spectrum CT(all P<0.05),multi-factor Logistic regression showed that K-NC and WC-AP were independent risk factors for predicting the invasion of GGN lung adenocarcinoma(all OR>1,P<0.05),the AUC of the energy spectrum CT prediction model was 0.847,the sensitivity and specificity were 0.816 and 0.750;(2)There were significant differences between the noninvasive group and invasive group in the 3D length diameter,maximum surface area,surface area,quality,volume and mean CT value among the quantitative parameters of AI(all OR>1,P<0.05),multi-factor Logistic regression showed that 3D length diameter and mean CT value were independent risk factors for predicting the invasion of GGN lung adenocarcinoma,the AUC of the AI prediction model was 0.897,the sensitivity was 0.776,and the specificity was 0.969.(3)Multi-factor Logistic regression of all quantitative parameters of energy-spectrum CT and AI showed that K-NC,WC-AP,3D length diameter and mean CT value were independent risk factors associated with the invasion of GGN lung adenocarcinoma(all OR>1,P<0.05),the AUC of the combined prediction model based on energy spectrum CT+AI was 0.965 with a sensitivity of 0.918 and a specificity of 0.953;The AUC of the combined prediction model was higher than the AUC of independent factors,the energy spectrum CT prediction model and the AI prediction model(all P<0.05).(4)A nomogram was plotted based on energy spectrum CT+AI combined prediction model,and the C-index is 0.965,the calibration curve showed that the curve predicting the probability of infiltration of GGN lung adenocarcinoma fitted well with the curve of the probability of actual occurrence of invasion.Conclusion: Quantitative parameters of energy spectrum CT and AI can predict the invasion of GGN lung adenocarcinoma.K-NC,WC-AP,3D length diameter and mean CT value were independent risk factors associated with the invasion of GGN lung adenocarcinoma.Based on the above factors,the predictive efficiency of the combined prediction model is higher than that of each single parameter and prediction model.The Nomogram based on the combined prediction model is useful for individual-level risk assessment of invasive lung adenocarcinoma.