Predicting Initial Efficacy Of Neoadjuvant Chemotherapy For Advanced Gastric Cancer Based On Radiomics

Objective: Gastric cancer is one of the most common malignant tumors in the world,with high incidence and mortality.Gastric cancer patients generally have the characteristics of "three high,three low",that is,high incidence,metastasis rate and mortality rate,and low early diagnosis rate,radical resection rate and 5-year survival rate.Peritoneal metastasis is the main cause of death in patients with advanced gastric cancer(AGC).Currently,the main therapeutic methods for these patients are surgery with chemoradiotherapy.Different from traditional adjuvant chemotherapy after surgery,neoadjuvant chemotherapy is preoperative chemotherapy,which can effectively reduce tumor stage,improve resection rate and reduce recurrence rate.However,not all patients can benefit from the corresponding chemotherapy regimen,so we need to have a effective prediction of the efficacy in order to make better treatment decisions.Currently,computed tomography(CT)is mainly used to diagnose and evaluate the efficacy of gastric cancer.Therefore,this study intends to explore the value of radiomics in predicting the initial efficacy of neoadjuvant chemotherapy in patients with advanced gastric cancer with peritoneal metastasis(PM)based on multi-phase enhanced CT.Material and Method: The clinical baseline and imaging data of 93 AGC patients with pathologically confirmed peritoneal metastasis were retrospectively analyzed,including58 males and 35 females,aged 27-77 years old,with an average of(56.54±9.90)years old.Among them,48 cases(PR)had obvious chemotherapy effect,and 45 cases(PD+SD)had no obvious chemotherapy effect.The omentum was taken as the target of the study.ITK-SNAP software was used to manually delineate the region of interest(ROI)on the portal phase CT images,and imported it to AK software for feature extraction.R software was used for feature screening.Firstly,correlation matrix was constructed for preliminary screening,and then the importance of radiomics features was sorted based on random forest algorithm to get the features with predictive value.Four radiomics models were constructed by using the selected features,and each model was cross-validated by ten folds,and the ROC curves of each model were drawn to evaluate the prediction efficiency of the models.Result: 18 features were selected,and the main ones were from Gray Level Run Length Matrix(GLRLM).The AUC values of the radiomics models were 0.85(linear discriminant analysis,LDA),0.73(random forest,RF),0.72(support vector machine,SVM)and 0.66(K nearest neighbor,KNN),respectively.The comparative analysis showed that the LDA model had the best predictive performance.The sensitivity and specificity of the LDA model were 76.9% and 64.2%,respectively,indicating that the model was sensitive to the initial efficacy of neoadjuvant chemotherapy.Conclusion: Multi-phase enhanced CT-based radiomics models have great potential in predicting the initial efficacy of adjuvant chemotherapy for AGC patients with PM,and has important reference value for helping clinical decision-making and realizing individualized treatment.