A Study Of First-line EGFR Mutant Advanced Lung Adenocarcinoma Efficacy Prediction Method Based On CT Images And Clinical Features

Objective: Tyrosine kinase inhibitor(TKI)monotherapy and combination therapy significantly prolong progression-free survival(PFS)in patients with advanced epidermal growth factor receptor(EGFR)-positive lung adenocarcinoma(LUAD).However,not all patients with mutations benefit from TKI and drug resistance is their inevitable outcome,therefore accurate prediction of treatment response and prognosis is crutial in clinical practice.Due to the specific response pattern of targeted drugs,traditional efficacy evaluation methods cannot always assess the treatment response accurately in above patients.This study explores the prognostic efficacy of baseline CT and clinical data including serum tumor biomarkers based on advanced first-line EGFR mutant LUAD patients based on radiomics and deep learning methods,aiming to provide a reference for the accurate treatment of EGFR mutant LUAD patients.Methods: We retrospectively collected Lung CTs,clinical datas and follow-up datas at baseline from 133 patients with advanced first-line EGFR mutant LUAD who attended the First Hospital of China Medical University and Shengjing Hospital from December 9,2011 to March 18,2022.The median follow-up time was 17.3 months.We constructed an imagingomics-based efficacy prediction model in advanced first-line EGFR-mutant LUAD.Univariate and multivariate analysis regression using COX proportional risk regression model was performed to explore the predictive role of each clinical information on PFS.A joint model of imaging histology and screening clinical features was constructed to predict efficacy.Finally,a deep learning efficacy prediction model based on CT images was developed using Res Net convolutional neural network.Results: A total of 133 patients with advanced first-line EGFR mutant LUAD receiving TKI monotherapy or combination therapy with a median follow-up of 17.3 months were enrolled in this study.Median(14.0 months)and mean(16.497 months)progression-free survival(PFS)were used as the grouping criteria,respectively.In the efficacy prediction model for imaging histological features,the prediction model using the median grouping of PFS was better than the mean grouping model(AUC: 0.741 vs 0.735).Multifactorial analysis of the COX risk regression model showed that the presence of brain metastases(HR=1.97,P=0.002)and bone metastases(HR=2.03,P=0.001)were both important factors for advanced first-line EGFR mutant LUAD patients as independent prediction factors for PFS.Combining the above clinical factors with imaging histology to construct a combined prediction model was better than the imaging histology model alone(AUC:0.798 vs 0.741),and the prediction model using median PFS grouping was better than the mean grouping model(AUC: 0.798 vs 0.749).Deep learning models were less effective in prediction than imaging histology models alone(0.725 vs 0.741).Conclusion: CT-based imaging models have good predictive efficacy for efficacy in patients with advanced first-line EGFR-mutant LUAD.PFS was shorter in patients with advanced first-line EGFR-mutated LUAD with brain and bone metastases.The combined imaging and clinical models were more effective than the imaging model alone in predicting efficacy in patients with advanced first-line EGFR mutated LUAD.Deep learning combined with CT images was able to predict efficacy in LUAD patients,which was not as effective as the imaging histology model.Adequate integration of imaging histology and clinical features can be used to predict the outcome of advanced LUAD treatment and help assist physicians in making more accurate medical decisions.