基于深度学习的CT量化指标在间质性肺疾病进展预测中的临床价值OA

Objective:To construct a quantitative index DL-ILD for interstitial lung disease(ILD)based on deep learning(DL),and to explore its clinical value in predicting the progression of ILD.Methods:A retrospective study was conducted on 112 patients with ILD.A three-dimensional U-Net(U-shaped convolutional neural network)model was constructed based on high resolution computerized tomography(HRCT)to automatically identify lesions and calculate DL-ILD.The evaluation criterion was whether ILD had progressed after a 24-month follow-up.The correlation between DL-ILD,visual scores,and lung function was assessed,and the predictive performance of the two methods for predicting progression and prognosis was compared.Logistic regression,receiver operating characteristic(ROC)curve analysis,Cox analysis,and five-fold cross-validation were used to validate the model.Results:Among the 112 ILD patients,42 experienced progression events.DL-ILD showed a negative correlation with forced vital capacity(FVC％)and diffusion capacity of the lung for carbon monoxide(DLCO％)(r=-0.762,r=-0.685,P＜0.001),outperforming visual scores(Z=2.593,P=0.010;Z=2.598,P=0.009).Logistic regression analysis indicated that DL-ILD was an independent predictor of ILD progression(OR=1.242,P=0.001),with its predictive performance(AUC=0.864)superior to that of visual scores(AUC=0.710).The 1-year and 2-year progression-free survival rates of the high DL-ILD group were 82.1％and 53.7％,respectively,which were significantly lower than 98.4％and 84.6％of the low DL-ILD group(P＜0.01),and the Cox model confirmed it as an independent risk factor for disease progression(HR=2.872,P=0.002).The AUC of the five-fold cross-validation was 0.833,indicating stable model performance.Conclusion:DL-ILD can accurately reflect the prediction of ILD disease progression,outperforming traditional visual scores,and has good reproducibility and clinical application potential.