Radiomics And Multimodal Functional Imaging In Predicting The Response Of Esophageal Cancer To Definitive Chemoradiotherapy

Objective:1.To establish and validate a radiomics nomogram for predicting 2-3 years local recurrence/uncontrolled rate and survival rate of patients with esophageal cancer undergoing definitive chemoradiotherapy.2.To explore the value of pre-treatment SUV value of PET-CT in predicting the response rate of definitive chemoradiotherapy for esophageal cancer.3.To explore the value of multimodal functional imaging combined with chemoradiotherapy resistance molecular molecules for prediction of response to chemoradiotherapy in esophageal cancer.Method:1.A retrospective analysis was made of pre-radiotherapy enhanced CT images of 311 patients with esophageal cancer receiving definitive chemo-radiotherapy(232 in training and 79 in validation cohort).A total of 2338 image features were extracted.Lasso-based Cox survival regression model was used to select radiomics features for patients in training cohort.Finally,4 features were selected.Each feature had a weight value,which was used to calculate radiomics score(feature multiplied by weight sum)and construct radiomics signature.Cox regression model was used to establish the combined prediction model of radiohistology and clinical factors,and calibration curve and C-index were used to evaluate the effectiveness of the model.2.Pre-treatment PET-CT images of 67 patients with esophageal cancer undergoing definitive chemoradiotherapy were retrospectively collected.The average SUV value and the maximum SUV value(SUVmean and SUVmax)for the most obvious level of the lesion were measured.The response of chemoradiotherapy was evaluated by imaging data within three months after radiotherapy(CR,PR,SD and PD were evaluated by RECIST standard,respectively).CR+PR was defined as effective group,while PD+SD was defined as noneffective group.The ROC curve analysis was used to evaluate the effectiveness of pretreatment initial SUV value in predicting the response of definitive chemoradiotherapy for esophageal cancer.3.Ten patients with esophageal cancer who received definitive chemoradiotherapy in our department were prospectively enrolled.Blood samples,ADC values of DWI-MRI,ADC values before,during and after radiotherapy,ADC value changes before and during radiotherapy(?ADCduring),ADC value changes before and after radiotherapy(?ADCpost),PET-CT SUV values before and during radiotherapy(SUVmean?SUVmax)were collected The expression of vascular endothelial growth factor and the level of CTC in blood samples before,during and after radiotherapy were detected.Gastroscopy biopsy,PET-CT and DWI-MRI were performed 6 weeks after radiotherapy.According to RECIST criteria,the patients were divided into effective treatment group(CR+PR)and noneffective treatment group(PD+SD).The therapeutic effect(cCR and non-cCR)was evaluated according to the results of gastroscopic biopsy.The distribution difference of each measured value in different therapeutic groups was analyzed.Spearman correlation analysis was used to evaluate the relationship between the above measurements and the response of esophageal cancer to definitive chemoradiotherapy.Result:1.The radiomics signature composed of 4 image features were selected from pre-radiotherapy enhanced CT images,which showed good performance in predicting local recurrence-free rate(LRFS)for 2-3 years.The patients could be divided into high-risk and low-risk groups according to radiomics score(Rad-score threshold:-0.1152).The prognosis of the two groups was significantly different.The LRFS of low-risk group determined by radiomics signature was significantly better than that of high-risk group in training cohort(2-y LRFS 73.1%vs 54.4%,P<0.05;3-y LRFS 82.5%vs 45.74%,P<0.05)and verification cohort(2-y LRFS 87.5%vs 46.6%,P<0.05;3-y LRFS 82.5%vs 45.74%,P<0.05).Survival analysis of OS also showed the same results in training cohort(2-y OS 69.3%vs 42%;3-y OS 57.8%vs 35.6%,P<0.001)and verification cohort(2-y OS 51.9%vs 48.1%;3-y OS 51.9%vs 33.4%,P<0.05).The clinical-radiomics model established by combining radiomics signature with clinical data is obviously superior to the single clinical model(verification of the centralized C-index,0.729 vs 0.69;P<0.01),and the calibration curve also shows good consistency.2.Pre-radiotherapy SUV values was not capable of predicting the response of definitive chemoradiotherapy for esophageal cancer.Three months after treatment,8 patients(11.9%)reached CR,46 patients(68.7%)reached PR,2 patients(3.0%)reached PD and 11 patients(16.4%)reached SD,respectively.Between CR(N=8)and non CR(N=59)groups,effective(CR+PR,N=54)and noneffective(PD+SD,N=13)groups,there was no significant difference in the SUVmean before radiotherapy(5.8±2.5 vs 7.3±2.9,P=0.379;6.9±2.8vs7.8±3.2,P=0.547)as well as SUVmean before radiotherapy(9.6±4.6 vs 12.7±5.8,P=0.292;12.1 ±5.7 vs 13.4±5.7,P=0.423).The correlation analysis showed that the average and maximum of SUV before radiotherapy were not significantly correlated with the response after radiotherapy(both correlation coefficient absolute value<0.2;both P value>0.05).ROC curve analysis indicated that SUV value before radiotherapy had a low predictive effect on the response of esophageal cancer after chemoradiotherapy,but there was a trend indicating the worse the SUV value,the worse response rate.3.Effective treatment group(CR+PR,N=7)and noneffective group(PD+SD,N=3)had significant difference in ?ADCpost,?CTCduring and ?VEGFduring(60.55%121.41%vs 27.10%,P=0.033;2±6 vs-16,P=0.017;39.78±41.94 vs-40.00,P=0.017).The initial ADC values(1.09±0.22 vs 1.12;P=0.517),?ADCduring(27.72%±71.38%vs 17.76%;P=0.667),?SUVmax(48%±134%vs 41%;P=0.833),?SUVmean(42%±139%vs 35%;P=0.833)and the initial CTC level(6±5 vs 6;P=0.833)were not significantly different between the two groups.The correlation analysis also showed that ?ADCpost,?CTCduring and ?VEGFduring had strong correlations with the short-term efficacy(correlation coefficient were 0.722,0.803,0.80;P value were 0.018,0.005,0.005 respectively),and there was a significant correlation between the post-treatment ADC,?CTCpost and the short-term efficacy(correlation coefficient were 0.646,0.650;P value were 0.044,0.042,respectively).Conclusion1.Adding radiomics features to clinical data can significantly improve the prediction power in terms of 2-3 year local recurrence/uncontrolled rate and survival rate.The radiomics analysis of contrast enhanced CT before radiotherapy is helpful for personalized risk stratification and treatment of esophageal cancer patients undergoing definitive chemoadiotherapy.2.SUVmean(SUVmax)alone before radiotherapy could not predict the treatment response of esophageal cancer patients undergoing definitive chemoradiotherapy.3.?ADCpost,?CTCduring,?CTCpost could predict treatment response of esophageal cancer patients undergoing definitive chemoradiotherapy.Multimodal functional imaging combined with expression of vascular endothelial growth factor and CTC levels may improve the prediction accuracy and provide clues for stratified management and treatment of patients.The next step of this study is to further expand the sample size and conduct in-depth research.