The Application Of Radiomics Analyzing In Predicting Thoracic Tumor Heterogeneity And Radiotherapy Response

PART?The Value of CBCT-based Tumor Density and Volume Variations in Prediction of Early Response to Chemoradiat-ion Therapy in NSCLCPurpose and BackgroundsDue to tumor heterogeneity,patients who received the same treatment modality had significant difference in therapeutic response and prognosis.Although the Response Evaluation Criteria in Solid Tumors(RECIST)has been widely applied in the clinical setting and introduced in tumor therapeutic response evaluation,some limitations are still existed.Firstly,changes in the longest axis of targeted lesion is the main evaluation criterion in RECIST,which ignores variations in other forms to a certain extent.Secondly,the morphologic changes based on uni-dimensional measurements neglects molecular alternations and cannot completely reflect biologic alternations in tumors,which may result in treatment response underestimation.Thirdly,the RECIST criteria are generally performed two or three months after treatment,which might delay the detection of disease progression and recurrence.With advances in medical imaging analysis and radiotherapy technology,quantitative ki lovoltage Cone-beam Computed Tomography(KV-CBCT)images not only contribute to radiotherapy plan optimization,but also dynamically observe tumor changes.The first aim is to investigate the correlations between primary tumor physical density changes and radiation dose or tumor volume(TV)variations.Secondly,this study proposed to investigate the correlations between chemoradiation treatment response and variations in primary CT number(CTN)and TV by KV-CBCT in NSCLC patients and to determine the appropriate time points for early response prediction.Materials and MethodsFifty-four patients with inoperable and locally advanced NSCLC who received Chemoradiotherapy(CRT)and CBCT scan were involved in this study.Primary tumors were mannually delineated and modified on CBCT images on 1st,5th,10th,15th,20th,25th,and 30th fractions through treatment planning system.CTN and TV were measured on each of the seven observation points by in-house software.Variation ratios of CTN mean values(HU)and TV(cm3)during the treatment course were analyzed and correlated with clinical outcomes evaluated by RECIST criteria.The correlation between the CTN and volume changes for the primary tumor,and the correlation between the CTN change and radiation dose were analyzed using linear analysis.Categorized characteristics were compared between two groups using a Chi-squared test,and continuous variables were compared using a t-test.ROC curve was applied to assess the predictive abilities of CTN and TV for clinical outcomes.ResultsAll patients had various degrees of changes in CTN and TV during radiation delivery.For all patients in the study,the mean CTN value reduction was significantly related to the radiation dose with R2 = 0.879 ±0.164 and p=0.002.While the relationship between mean CTN value change and volume reduction ratio was only R2=0.343 and p<0.001.Out of the 54 patients,33 patients were response group and 21 patients presented non-response according to RECIST criteria.After 60Gy of conventional fractionation radiotherapy,the CTN variations were more significant in the response group(28.44±13.12HU vs.19.63±8.67 HU,p=0.016).The tumor volume in patients who had a response diminished by 32.01%(range,8.46-61.67%),and patients who were categorized as non-response had tumor volumes that dro-pped by 23.20%(range,-15.57-38.00%)(p = 0.048).Further,logistic regression model and ROC analysis implied that the combination of changes of CT density value and tumor volume had a higher AUC(AUC = 0.751)than CTN(AUC = 0.666)or tumor size(AUC=0.693)alone for assessing treatment early response(p = 0.002).The differences between response and non-response were most significant between Fraction 10 and Fraction 15 for CTN changes and between Fraction 5 and Fraction 10 for the TV regression ratio.ConclusionsFor NSCLC tumor target,CTN variation was linearly correlated with the radiation dose received.The changes of TV and CTN obtained from CBCT images were considered as predictors of early response,which could be able to identify NSCLC patients who benefit from CRT.The prediction capability may be improved by the combination of the changes on TV and CTN.PART ?CT-based Radiomics Features Predict Platinum-sensitivity Status in Limited-stage Small Cell Lung Cancer Patients Treatedwith ChemoradiotherapyPurpose and BackgroundsSmall cell lung cancer(SCLC)is characterized by short doubling time,high incidence of metastasis and poor prognosis.Patients with limited stage small-cell lung cancer(LS-SCLC)respond well to a combination of chemo-and radio-therapy,but high recurrence rates result in low survival rate.Theoretically,LS-SCLC patients could be denoted as platinum refractory or platinum sensitive based on shorter or longer relapse times(90 days from last platinum administered),respectively.Furthermore,patients with longer treatment-free interval have a better progression-free survival(PFS)time and objective response rate(ORR)to the same chemotherapy regimens used in initial treatment.According to the National Comprehensive Cancer Network(NCCN 2016 edition)guidelines,the chosen second-line chemotherapy regimens was mainly dependent on the length of time from the end of chemotherapy to disease progression.Although platinum-sensitivity status is so important in clinical treatment,the risk factors for LS-SCLC are still unknown and controversial.With advances in medical imaging technology,tumor phenotype heterogeneity could be quantitatively analyzed by CT radiomics.CT radiomics analysis not only provided a non-invasive approach to assess inter-tumoral heterogeneity,but also covered the shortages of current method to predict treatment sensitivity.The purpose of this retrospective study is to investigate the CT-based radiomics features and clinicopathological parameters to predict platinum-based sensitivity status in LS-SCLC,and explore the significance of CT radiomics in predicting platinum-sensitivity status.Materials and MethodsWe retrospectively enrolled 200 LS-SCLC patients who were diagnosed by cytology or histology between January 2005 and December 2010 at Shandong Cancer Hospital.Furthermore,patients must receive diagnostic CT scan before anti-cancer treatment and at least one cycle of chemotherapy.The clinical data were obtained from medical records,including age,smoking status and treatment modality;biochemical laboratory test involved white blood count,hemoglobin and tumor-specific biomarkers.Primary tumors were manually delineated on treatment planning system and modified if necessary by two independent radiation oncologists.Radiomics features were high-throughput extracted from regions of primary tumor by in-house software.508 radiomics features were obtained from diagnostic CT of each tumor,and 127 of these were considered as independent features through inter-observers test and inner-correlation analysis.The correlations between platinum-sensitivity status and radiomics features with clinicopathological factors were performed using SPSS and R software.LASSO algorithm-based logistic regression was used for independent predictive factors identification.Area under the curve(AUC)was applied to evaluate radiomics features and clinicopathological.parameters predictive ability.ResultsAmong all of patients,124 patients were platinum sensitive and 76 patients were platinum refractory.T-test analysis shown that the differences between sensitive and refractory patients were detected in 19 radiomics features.Corresponding to clinicopathological parameters,such as PS score,treatment modality,response to first-line treatment and the level of NSE,CEA and NLR have potential abilities to distinguish the sensitive from the refractory.LASSO logistic regression model demonstrated that Maximum Probability(AUC = 0.648),Long Run Emphasis(AUC?0.665),Maximum 2D diameter Row(AUC = 0.670)and Correlation-HHH(AUC ?0.730)were correlated with the sensitivity status.Multivariate analysis illustrated that pretreatment NSE level(OR= 1.742;95%CI= 1.057-2.746.p = 0.03),NLR(OR?1.812;95%CI = 1.456-2.273,p<0.001)and response to first-line therapy(OR =0.388;95%CI=0.208-0.741,p = 0.003)were independent predictors for platinum-sensitivity status.Moreover,ROC analysis implied that the combination of them had a higher AUC(AUC = 0.791,p = 0.001)than radiomics features(AUC = 0.738,p<0.001)or clinicopathological factors(AUC = 0.683,p = 0.004)alone for assessing platinum-sensitivity status.ConclusionsRadiomics features could be considered as independent predictive factors for platinum-sensitivity status in LS-SCLC.Compared with traditional clinicopathological factors and laboratory test parameters,features extracted from CT images performed better for prediction,which might enable a step forward in individualized treatment.PART ?Quantitative Imaging Features of Computed Tomography Allow Prediction of PD-L1,FOXP3+TILs and CD8+TILs Expression and Prognosis in ESCCPurpose and BackgroundsNearly 90%of esophageal cancer are esophageal squamous cell carcinoma(ESCC)in China.Although prevention,diagnosis and treatment modality improved gradually in recent years,patients with ESCC have not yet increased dramatically in prognosis.Therefore,it is necessary to explore a novel treatment modality.Immune checkpoint inhibitors blocked PD-1(Programmed Cell Death-1)/PD-L1(Programmed Cell Death-Ligand 1)pathway and regulated its expression by binding to immunosuppressive proteins,then restoring the effects of T cells on tumor cells.While compared to widespread applications in melanoma and lung cancer treatment,the utilities of checkpoint inhibitors in esophageal cancer were still in clinical trials.On the other hand,tumor-infiltrating lymphocytes(TILs)had cytotoxic effect on tumor cells as well.Nevertheless,the association between TILs and patients'prognosis was still controversial.As a consequence,the participation of radiomics features might improve the predictive ability of prognosis.As we all known,tumor heterogeneity was a crucial influence factor on ESCC prognosis.Despite patients were given similar treatment modalities,clinical outcomes were various apparently.With the development of medical imaging analysis,it is now possible that high-throughput extraction of large numbers of image parameters could reflect tumor heterogeneity,capture phenotype information and predict prognosis.The aim of this study is to e-valuate the association between CT radiomics features and expressions of PD-L1,FOXP3+TILs and CD8+TILs,and establish a prognostic model consisting of clinicopathological factors,immunological parameters and radiomics features for esophageal cancer.Materials and MethodsWe retrospectively enrolled 160 ESCC patients who were diagnosed by cytology or histology between January 2005 and December 2012 at Shandong Cancer Hospital and received diagnostic CT scan.Computer-generated random numbers were used to assign 100 patients to the training set and 60 patients to the validation set.The clinical data was obtained from medical records.The expressions of PD-L1,FOXP3+TILs and CD8+TILs were identified as high-expression and low-expression group,and the cut-off values were defined as the average value of stained cells per field per section.Primary tumors were manually delineated on treatment planning system and modified if necessary by two independent radiation oncologists.In-house software extracted numerous radiomics features from diagnostic CT,involving histogram intensity,size and shape based-features,texture features and wavelet features.mRMR(Minimum Redundancy Maximum Relevance)and logistic regression with backward stepwise method were performed to identify independent predictive factors of PD-L1,FOXP3+TILs and CD8+TILs expressions.LASSO Cox regression model was used for evaluating the prognostic capability of clinicopathological factors,immunological factors expression and radiomics features.Variables with p<0.05 were considered statistically significant(two-sided).ResultsThere was no significant difference between training set and validation set excepted the following time(p = 0.022).Tumor staging(OR = 0.521,95%Cl ?0.295-0.922,p = 0.025)and differentiated grade(OR = 0.336,95%CI=0.126-0.894,p = 0.029)were independent predictors for PD-L1 high expression.FOXP3+TILs expression was correlated with differentiated grade(OR = 0.211 95%CI ?0.08-0.560,p = 0.002).For CT radiomics features,the predictive biomarkers were Run Length Non Uniformity for PD-L1,Interquartile Range-LHH and 90 Percentile-HHH for FOXP3+TILs,Run Percentage for CD8+TILs,respectively.The overall survival(OS)of ESCC patients was 40.73 months and 1-,3-,5-year OS rate were 92.31%?52.57%and 23.08%.Kplan-Meier analysis illustrated that ECOG PS scores,tumor staging,differentiated grade and the expressions of PD-L1 and CD8+TILs were related with clinical outcomes.Multivariate Cox proportional hazard models shown that PD-L(HR = 2.479,95%CI = 1.461-4.205,p = 0.001),CD8+TILs(HR = 0.524,95%CI ? 0.310-0.885,p = 0.016),tumor staging(HR ?1.559,95%CI=1.124-2.163,p = 0.008)and differentiated grade(HR = 1.618,95%CI? 1.151-2.275,p = 0.006)were correlated with survival.Furthermore,49 radiomics features were selected and 5 independent prognositc radiomcis features were identified through LASSO COX regression analysis.The radiomics signature shown predictive effects on prognosis,which could classify patients into low-risk group(?12.604)and high-risk group(>12.604).Kaplan-Meier survival analysis demonstrated that significant separations between high-and low-risk groups in both of training(35.0m vs.47.8m,high-risk vs low-risk,p<0.001)and validation cohorts(33.8m vs.44.6m,high-risk vs low-risk,p<0.001).ConclusionsThe utility of CT-based radiomics features could accurately predict immune expression and capture tumor phenotype information in esophageal cancer.This study suggested that radiomics features participation not only contributed to better performance,but also provided decision-making for tumor immunotherapy.It is now possible to evaluate patient survival benefit through the multi-omics model and promote development in precision therapy.