| Part ?: The value of spectral CT imaging for the preoperative prediction of lymph node metastasis in gastric cancerObjective 1.To explore the diagnostic efficacy of spectral CT for the preoperative prediction of lymph node metastasis(LNM)of gastric cancer(GC),and to develop and validate an iodine concentration based nomogram for the preoperative prediction of lymph node metastasis in gastric cancer.2.To explore the value of the developed nomogram for the prediction of prognosis in gastric cancer patients.Material and methods This study retrospectively included the clinical and imaging data of 210 patients with surgically confirmed gastric adenocarcinoma,who received spectral CT scan before surgery.Among them,159 cases were males and 51 cases were females,with an average age of 59.8±7.7 years and a range of 28-79 years.Patients were randomly divided into a training set(140 cases)and test set(70 cases).Clinical data and patient survival information were collected,including overall survival(OS)and progressive-free survival(PFS).The iodine concentration(IC)of gastric cancer tumor at arterial and venous phase was measured on the iodine-water material deposition images(MD)automatically generated on GSI Viewer software,and was normalized by dividing IC value of tumor to that of the aorta on the same slice.The tumor thickness was measured and Borrmann classification was evaluated according to tumor size,ulceration and infiltration.Univariate and multivariate logistic regression analysis was used to screen independent predictors of lymph node metastasis,and a nomogram was developed to predict lymph node metastasis preoperatively in the training set.ROC curve,accuracy and Harrell's C-index were used in the test set to evaluate the predictive effectiveness of the model.Cox survival analysis was used to evaluate the relationship between the model and the prognosis of gastric cancer patients.Results 1.The mean values of tumor thickness,the ratio of Borrmann classification ?-?,ICAP,ICVP,n ICAP,n ICVP were 19.68±8.83 mm,42.50%,14.52±5.75(100 ?g/ml),23.60±9.29(100 ?g/ml),0.1806±0.0876,0.4982±0.1895,respectively.Except ICAP(P > 0.05);the other parameters in lymph node metastasis group were higher than those in non-metastatic group(P < 0.05).2.Tumor thickness,Borrmann classification and ICVP were independent predictors for lymph node metastasis in gastric cancer.3.The nomogram based on the above predictors was developped in the training set and was assiociated with lymph node metastasis status(P <0.001).4.In the test set,the area under the curve(AUC)of this model was 0.793 [(95% confidence interval,CI);0.678-0.908] and the accuracy was 0.757(95% CI,0.640-0.852),showing a high ability for preoperative prediction of lymph node metastasis.5.The model has a good prognostic ability,and the C-index for predicting PFS and OS was 0.675(95% CI,0.571-0.779;P <0.001)and 0.643(95% CI,0.518-0.768;P = 0.025).Conclusion 1.This study firstly developed and validated a nomogram which included tumor thickness,Borrmann chassification and ICVP.2.The developed model was simple and easy to use,and successfully stratified gastric cancer patients according to the risk of lymph node metastasis,which in turn result to improve preoperative individualized N staging diagnosis of gastric cancer.3.The model was also assiociated with patients' survival with a promising prognostic ability.Part ?: The value of a CT radiomic nomogram for the preoperative prediction of lymph node metastasis in gastric cancerObjective 1.To develop and validate a CT based radiomic nomogram for the preoperative prediction of lymph node metastasis in gastric cancer,and to explore its additional diagnostic efficacy by comparing with the clinical prediction model.2.To explore the value of this model in the prediction for patients prognosis.Material and methods A total of 193 patients with surgical pathologically confirmed gastric adenocarcinoma were retrospectively enrolled.There were 150 males and 43 females,aged from 28 to 79 years,with an average of 58.62±10.22 years.Patients were randomly divided into a training set(97 cases)and a test set(96 cases)by computer random numbers.Clinical indicators,including age,gender,tumor location,were collected.Tumor thickness was measured and Borrmann classification was evaluated according to tumor size,ulceration and infiltration.CT reported LN status positive was defined as lymph nodes with short diameter > 10 mm or with heterogeneous enhancement and necrosis.ITK-SNAP 3.6.0.(www.itksnap.org)software was used to complete two dimensional tumor segmentations on the largest section of the tumor on the dual phasic enhanced images.Machine learning method was used for radiocmic feature extraction and signature building.Univariate analysis was used to evaluate the relationship between clinical characteristics and lymph node status.Chi-square test or Fisher's exact test were used to compare the differences between categorical variables,and Mann-Whitney U test or independent t test were used for continuous variables.Multivariate logistic regression analysis was performed based on meaningful clinical features and radiomic signatures to identify independent predictors of lymph node metastasis and then to construct a personalized prediction model.Firstly,multivariate logistic regression analysis was conducted based on clinical characteristics and a clinical prediction model was developed to simulate the clinical benchmark decision making and compared with the radiomic nomogram.The receiver operating characteristic curve(ROC)and the area under curve(AUC)were used to evaluate the efficacy of the radiomic model.The optimal diagnostic threshold is the point corresponding to Youden's index on the ROC curve of the training set.Hazard ratio(HR)and P value were obtained by univariate Cox survival regression analysis.Harrell's consistency index(C-index)was used to quantify the predictive effectiveness of the model.Results 1.Tumor thickness and CT reported LN status in lymph node metastasis group were significantly higher than those in non-lymph node metastasis group(P<0.05).2.Construction and validation of radiomic signatures: arterial phase signature contained 2 deep learning features and 1 artificial definition feature,while venous phase signature contained 1 deep learning feature and 2 artificial definition features.The two radiomic signatures showed a high predictive value for lymph node metastasis in the test set,and the AUC of AP and VP signature was 0.688(95% CI: 0.554-0.822)and 0.643(95% CI: 0.505-0.781),respectively.3.Development and validation the radiomic nomogram: CT reported LN status and VP radiomic signature were included in model construction.4.The nomogram showed high diagnostic efficacy in both training and test set,with AUC was 0.810(95%CI: 0.734-0.886)and 0.776(95%CI: 0.664-0.888),higher than that of clinical model.5.When taking 0.63 as the optimal threshold,the nomogram showed a high diagnostic efficacy in the training set and test set,with accuracy,sensitivity and specificity was 0.765(95%CI: 0.684-0.833),0.786,0.731,and 0.677(95% CI: 0.552-0.785),0.711,0.6336.6.The C-index of the correlation between nomogram and PFS and OS was 0.593(95% CI,0.494-0.692;P=0.0662),0.611(95% CI,0.495-0.727;P=0.0603).Patients with high nomograms had poor prognosis.Conclusion 1.This study firstly developed and validated a CT based radiomic normogram for the preoperative prediction of lymph node metastasis of gastric cancer.2.The normogram integrated CT reported LN status and VP radiomic signatures,which helped optimizing preoperative prediction of lymph node metastasis in gastric cancer patients and can provide more information for radiomic studies.3.The nomogram showed a higher predictive ability than clinical model.4.The nomogram was associated with prognosis and showed a high predictive ability for both overall survival and progression free survival.Part ?: A spectral CT based radiomic nomogram to predict lymph node metastasis preoperatively in gastric cancerObjective 1.To develop and validate a spectral CT multi-energy image based radiomic nomogram for the preoperative prediction of lymph node metastasis in gastric cancer.2.To explore the additional value of multi-energy image based radiomic nomogram for the preoperative prediction of lymph node metastasis by comparing with single-energy image-based radiomic model and clinical model.3.To explore the ability of the developed radiomic model for patients prognosis prediction.Material and methods A total of 204 patients with surgical pathologically confirmed gastric adenocarcinoma were retrospectively enrolled.There were 147 males and 57 females,age range 28 ~ 81 years,mean age of 58.83±11.34 years.Patients were randomly divided into a training set(136 cases)and a test set(68 cases)by computer random numbers.Clinical parameters,including age,gender,tumor location,were collected.CT parameters included: iodine concentration(IC)of gastric cancer at arterial and venous phase was measured on the iodine water images automatically generated by GSI Viewer software,and normalized iodine concentration(n ICs)was calculated.Tumor thickness was measured and Borrmann classification was evaluated according to tumor size,ulceration and infiltration.CT reported LN status positive was defined as lymph nodes with short diameter > 10 mm or with heterogeneous enhancement and necrosis.ITK-SNAP 3.6.0.(www.itksnap.org)software was used to complete two dimensional tumor segmentations on the largest section from 21 sets of energy images from 40 to 140 ke V with 5 ke V intervals.Machine learning method was used for radiocmic feature extraction and signature building.Univariate analysis was used to evaluate the relationship between clinical characteristics and lymph node status.Chi-square test or Fisher's exact test were used to compare the differences between categorical variables,and Mann-Whitney U test or independent t test were used for continuous variables.Multivariate logistic regression analysis was performed based on meaningful clinical features and radiomic signatures to identify independent predictors of lymph node metastasis and then to construct a personalized prediction model.Firstly,multivariate logistic regression analysis was conducted based on clinical characteristics and a clinical prediction model was developed.Then,a multi-energy image based radiomic nomogram were developed and compared with clinical model and the 65 ke V single-energy image based radiomic nomogram.The receiver operating characteristic curve(ROC)and the area under curve(AUC)were used to evaluate the efficacy of the radiomic model.The optimal diagnostic threshold is the point corresponding to Youden's index on the ROC curve of the training set.Hazard ratio(HR)and P value were obtained by univariate Cox survival regression analysis.Harrell's consistency index(C-index)was used to quantify the predictive effectiveness of the model.Results 1.Tumor thickness,n ICVP and CT reported LN status in lymph node metastasis group were significantly higher than those in non-lymph node metastasis group(P<0.05).There were no statistically significant differences in age,gender,tumor site and lymph node metastasis prevalence between the training set and test set.2.Arterial phase radiomic signature contained 1 deep learning feature and 2 artificial definition features,while venous phase radiomic signature contained 1 deep learning feature and 1 artificial definition feature.The two radiomic signature showed promising predictive ability of lymph node metastasis in the test set,and the AUC was 0.711(95% CI: 0.585-0.838)and 0.755(95% CI: 0.638-0.873).3.CT reported LN status and the two radiomic signatures were included in the model construction.4.The developed multi-energy image based nomogram showed encouraging diagnostic efficacy in both training and test set,with the AUC of 0.839(95% CI: 0.773-0.904)and 0.821(95% CI: 0.722-0.920),higher than the single-energy image based radiomic nomogram and clinical model.5.The accuracy,sensitivity and specificity of the nomogram in training and test set was 0.772(95% CI: 0.692-0.840),0.774,0.769 and 0.765(95% CI: 0.646-0.859),0.737,0.800,respectively,when using 0.62 as the threshold value.6.The diagnostic accuracy of the multi-energy image based model was significantly higher than that of the single-energy image based model and the clinical model.7.The nomogram was associated with prognosis,with the C-index was 0.637(95% CI,0.544-0.730,P=0.0038)for predicting progression free survival and 0.669(95% CI,0.560-0.778,P = 0.0023)for overall survival.Conclusion 1.This study firstly developed and validated a spectral CT multi-energy image based radiomic normogram for the preoperative prediction of lymph node metastasis in gastric cancer.2.The nomogram integrated the biphasic CT radiomic signatures based on multi-energy image and CT reported LN status,which wass helpful to optimize the preoperative prediction of lymph node metastasis in gastric cancer and improve the potential of radomic analyses.3.The nomogram was associated with prognosis and showed a good predictive ability for both overall survival and progression free survival. |
PDF Full Text Request