| Gastric cancer is one of the most common cancers worldwide and a leading cause of cancer-related death.Gastric adenocarcinoma is the most common type.Due to the insidious onset of gastric cancer,most gastric adenocarcinoma have entered the advanced stage when patients seeking medical treatment.Operation is the main treatment for advanced gastric adenocarcinoma.Accurate preoperative diagnosis,assessment and prognosis are the basis for accurate treatment decisions.Due to the temporal and spatial heterogeneity of tumors,postoperative pathological grading,preoperative biopsy,genetic testing and other prognostic evaluation methods have certain limitations.With the wide application of artificial intelligence technology in the field of medical images,radiomics and pathomics technologies emerge at the right moment and provide new ideas for solving the above problems.At present,excellent research achievements have been made in the diagnosis and differential diagnosis of gastric cancer,classification and staging,prognosis prediction and efficacy evaluation,and relatively mature technical framework has been formed in radiomics and pathomics.The construction of a model to predict the pathological grade and prognosis of advanced gastric adenocarcinoma by combining radiomics and pathomics techniques can not only reflect the heterogeneity of tumors from multi-dimensional,real-time and non-invasive perspectives,but also provide an important basis for the formulation of precise diagnosis and treatment plans for clinicians,which has not been reported yet.There are two parts in our research.The specific methods and achievements are as follows:In the first part of this study,based on preoperative enhanced CT venous images of patients with advanced gastric adenocarcinoma,a preoperative grading model for advanced gastric adenocarcinoma model was constructed.a total of 186 pathologically confirmed advanced gastric adenocarcinoma patients from Guizhou Provincial People’s Hospital were included in the study,including 36 patients in the high-grade gastric cancer group and 150 patients in the low-grade gastric cancer group.All patients were randomly divided into training set and independent validation set according to ratio 7:3.Firstly,preoperative enhanced CT venous phase images,postoperative pathological reports and clinical information of the patients were collected.The radiologist manually segmented the tumor area of the preoperative enhanced CT venous phase images and extracted the radiomics features of the ROI.Based on single-factor and multi-factor regression analysis and minimum absolute contraction and selection operator regression algorithm,feature screening was performed to obtain the radiomics features highly correlated with the pathological grade of gastric adenocarcinoma and obtain the radscore.Finally,combining clinical information,the fusion model of radscore and clinical information was established and verified.Finally,receiver operating characteristic curve was drawn to evaluate the preoperative prediction performance of the radiomics model and the fusion model for gastric adenocarcinoma grade classification.Calibration curve and decision curve were drawn to evaluate and compare the clinical application value of the two models.Based on the radiomics features and clinical information,this study constructed a radiomics model and a fusion model for predicting grade of gastric adenocarcinoma patients.A total of 17 radiomics features highly correlated with grading of gastric adenocarcinoma were screened out,including 5 first-order features and 12 texture features.Clinical information included age,sex,surgical procedure,AFP,CEA,and CA199.The AUC value of radiomics model in training set was 0.763(accuracy:0.664;Sensitivity: 0.638;Specificity: 0.769).The AUC in the validation set is 0.747(accuracy: 0.655;Sensitivity: 0.622;Specificity: 0.800).The AUC obtained by fusion model in training set was 0.753(accuracy: 0.779;Sensitivity: 0.819;Specificity: 0.615),and AUC value of validation set was 0.724(accuracy: 0.818;Sensitivity: 0.889;Specificity: 0.500).Finally,the De Long test is used to evaluate the prediction efficiency of radiomics model and fusion model.The results showed that the radiomics model and fusion model have good performance in preoperative prediction of advanced gastric adenocarcinoma grade,and there was no significant difference between the two models(P = 0.88).In the second part,based on preoperative enhanced CT venous images and postoperative digital pathological images of patients with advanced gastric adenocarcinoma,a postoperative 2-year survival risk predicting model for advanced gastric adenocarcinoma was constructed by using radiomics and pathomics techniques.126 pathologically confirmed patients with advanced gastric adenocarcinoma who underwent gastrectomy in Guizhou Provincial People’s Hospital were included.The patients were divided into high risk group(52 cases)and low risk group(74 cases)according to the criteria of whether death occurred within 2 years after surgery.All patients were randomly divided into the training set(87 cases)and the independent validation set(39 cases)in a 7:3 ratio.Firstly,preoperative enhanced CT images of abdominal venous phase were collected and the ROIs were delineated manually,and the radiomics features were extracted and screened to construct the radiomics model.At the same time,postoperative pathological sections of patients were collected,scanned into digital pathological images,and the ROIs were roughly segmented manually.After image normalization,shearing and amplification and other pre-processing,pathomics features were extracted from the pathological images and the pathomics model was constructed.The above-mentioned extracted radiomics features and pathomics features were fused for feature screening,and the features highly correlated with the patient’s 2-year postoperative survival risk were obtained,and the fusion model was constructed.Finally,the efficacy of these three models in predicting the 2-year survival risk of gastric adenocarcinoma patients after surgery was evaluated.First of all,44 radiomics features with high correlation with 2-year survival risk of patients with advanced gastric adenocarcinoma were extracted from the abdominal enhanced CT venous phase images of enrolled patients,including 14first-order features and 30 texture features.Then we construct an radiomics model to predict the 2-year survival risk of patients with advanced gastric adenocarcinoma.The AUC value of the training set was 0.770(accuracy: 0.579;Sensitivity: 0.647;Specificity: 0.857).The AUC in the validation set was 0.714(accuracy 0.577;Sensitivity: 0.667;Specificity: 0.786).Secondly,two pathomics features highly correlated with patients’ 2-year survival risk were finally extracted from the postoperative digital pathological images of the enrolled cases,including 1 first-order feature and 1 texture feature.A pathomics model was established to predict the 2-year postoperative survival risk of patients with advanced gastric adenocarcinoma,and the AUC value in the training set was 0.822(accuracy: 0.614;Sensitivity: 0.743;Specificity: 0.830);The AUC in the validation set was 0.864(accuracy: 0.632;Sensitivity: 0.867;Specificity: 0.739).Finally,the fusion model with one radiomics feature and two pathomics features obtained an AUC of 0.830(accuracy: 0.614;Sensitivity: 0.743;Specificity: 0.743)in training set,and AUC value of validation set was 0.904(accuracy: 0.632;Sensitivity: 0.867;Specificity: 0.867).The results showed that the fusion model was slightly more efficient than the radiomics or pathomics models,and the pathomics features played a slightly more important role in the fusion model than the radiomics features.In conclusion,pathological grade and 2-year survival risk of patients with advanced gastric adenocarcinoma are highly correlated with prognosis of patients with gastric adenocarcinoma.The preoperative prediction grading model of advanced gastric adenocarcinoma obtained in this study can objectively and noninvasively reflect the pathological grading status of tumors.The 2-year survival risk predicting model for advanced gastric adenocarcinoma combined with radiomics and pathomics can objectively and accurately predict the prognosis of patients.These two models confirmed the efficacy of radiomics and pathomics features in the classification and prognosis of gastric cancer patients,providing an important basis for assisting clinical decision-making and the formulation of diagnosis and treatment plans,thus improving the prognosis of patients and improving the quality of a patients life. |
PDF Full Text Request