Prediction And Clinical Prognosis Of Breast Cancer Tertiary Lymphoid Structures Expression Based On Radiomics And Clinicopathology

Objective: The predictive impact of radiomics and clinicopathological indicators on the expression of TLSs in breast cancer patients was explored,and a prediction model of breast cancer TLSs based on MRI radiomics and clinicopathology was established through retrospective analysis of the correlation between tertiary lymphoid structures(TLSs)and the survival and prognosis of breast cancer patients.Methods: From January 2016 to January 2019,basic information on patients at Sichuan Cancer Hospital who received a postoperative pathology diagnosis of breast cancer was gathered.A total of 242 breast cancer patients were enrolled.The enrolled patients were randomly divided into training set(n=169)and test set(n=73)according to the ratio of 7:3.The basic data of the training set and the test set were analyzed to clarify the representativeness of the test set.Some samples with high expression of TLSs were subjected to m IHC and the corresponding parts under HE staining were marked after the TLSs were positively marked,so as to identify atypical TLSs in tissue samples of other patients.After two senior pathologists judged the presence of TLSs,they were divided into TLSs-positive group(n=122)and TLSs-negative group(n=120).The baseline characteristics of patients in the TLSs-positive and TLSs-negative groups were compared using a univariate logistic regression analysis,accounting for any clinical factors that might have contributed to the differences.The blood parameters were split into two groups and the cut-off value was obtained using the ROC curve.These parameters were then associated to TLSs and incorporated in the single factor logistic regression analysis.After univariate regression analysis,variables with P<0.05 were screened,and multivariate logistic regression was included to select clinical independent risk factors.Before delineating the region of interest(ROI),the collected images need to be adjusted for intensity and corrected for bias field.The preprocessed DCE images and T2 WI images were imported into 3D Slicer4.11 software to delineate multi-sectional regions of interest,and extract the radiomics features of the original image,Gaussian Laplace transform and wavelet transform.In the later stage,the intraclass correlation coefficient(ICC)was used to evaluate the repeatability within observers.In the training set,use one-way analysis of variance and LASSO algorithm to screen out highly correlated features,and then calculate the Rad-score.The selected clinical independent risk factors were analyzed by multivariate logistic regression,combined with Rad-score to establish a nomogram model,and the area under the curve(AUC),sensitivity,specificity and accuracy of the model ROC curve were calculated.The performance and clinical utility of the model were assessed using calibration curve and decision curve analysis.We are able to investigate the connection between TLSs and prognosis using the Kaplan-Meier technique and log-rank analysis.Results:1.Clinical baseline of enrolled patients The 242 female breast cancer patients were 27-78 years old,with a median age of 49 years.Among them,142 patients were premenopausal and 100 patients were postmenopausal;the median follow-up time was 52 months,ranging from 15 to 76 months.Through follow-up,34 of all patients progressed,30 had distant metastasis,and 7 died(4 died of distant metastasis,2 died of recurrence,and 1 died of other causes).There was no significant difference in the basic data(age,tumor diameter,lymph node metastasis and shape)between the training set and the test set.2.Detection of TLSs expression and prognostic analysis in breast cancer According to the existence,they were divided into two groups for prognosis analysis.The 3-year and 5-year OS of TLSs-positive group and TLSs-negative group were 98.4 % vs.100.0 % and 98.4 % vs.90.7 %,respectively.The 3-year and 5-year DMFS of TLSs-positive group and TLSs-negative group were95.1 % vs.84.2 % and 91.4 % vs.82.5 %,respectively.The 3-year and 5-year DFS of TLSs-positive group and TLSs-negative group were 95.1 % vs.83.3 %and 90.2 % vs.78.5 %,respectively.There was no statistical difference in OS between the two groups,but there were significant differences in DFS(P=0.006)and DMFS(P=0.013)between the two groups,and the survival rate of the TLSs-positive group was significantly better than that of the negative group.3.Analysis of clinicopathological and hematological parameters related to TLSs expression in breast cancer There were significant differences in ER(P=0.001),PR(P<0.001),HER-2 (P=0.001),Ki-67 index(P<0.001)and molecular subtypes(P=0.001)between TLSs-positive group and TLSs-negative group.Among them,when HER-2 is overexpressed or Ki-67 index is highly expressed,the possibility of TLSs is high.TLSs correlation analysis was performed on blood parameters,and 11 hematological indexes were statistically different.The P value was set to less than 0.05,and the above 16 indicators were included in the logistic regression analysis.Finally,a total of six variables(HER-2,Ki-67 index,LYMPH%,MONO%,BASO%,PLT)were independent clinical risk factors for TLSs to construct a clinical prediction model.It showed that the expression of TLSs was inversely connected with BASO% or PLT,and that the greater the LYMPH% or MONO% score,the higher the positive rate of TLSs.4.Correlation analysis between TLSs expression level and MRI radiomics in breast cancer In the training set,586 radiomics features were extracted from the ROIs of DCE and T2 WI of each patient.After ANOVA and LASSO regression screening,10 best radiomics features were finally selected,7 from DCE phase and 3 from T2 WI phase to construct a simple radiomics model.After three months,the ICCs≥0.75 accounted for 94.2%,and the reproducibility of the model was good.In the training set,the sensitivity,specificity and accuracy of the radiomics model were 72.50%,71.91% and 72.18%,respectively.In the test set,the sensitivity of the model reached 77.78%,the specificity reached 75.68%,and the accuracy reached 76.12%.It suggested that MRI radiomics had a good discrimination ability for TLSs expression.5.Breast cancer TLSs expression prediction model based on radiomics,clinicopathological and blood parameters A nomogram model for predicting TLSs was constructed based on radiomics,clinical pathology and blood parameters.In the training set,the sensitivity of the nomogram model incorporating Rad-score,HER-2,Ki-67 index,LYMPH%,MONO%,BASO%,and PLT reached 73.17%,the specificity reached 73.56%,and the accuracy reached 73.37%.Compared with the clinical model,the nomogram model(training set: AUC 0.820 vs.0.735;test set: AUC0.749 vs.0.642)had better predictive ability.The predictive performance of the nomogram model was similar to that of the radiomics model(training set: AUC0.820 vs.0.766;test set: AUC 0.749 vs.0.749).However,after DCA analysis,it was found that the nomogram model had higher overall net income and was optimal in clinical practicality.At the same time,the calibration curve shows that the nomogram model has a small error in predicting TLSs.Conclusion: The prognosis of breast cancer patients is highly correlated with TLS expression.Patients with positive TLSs had a better prognosis and a higher long-term disease control rate.The clinicopathological indicator and radiomics feature-based nomogram model has a good predictive effect on TLSs expression and can be utilized to enhance clinical decision-making.