Clustering Approach Combining Contrast-enhanced CT And FDG PET To Predict Outcomes In Gastric Cancer

Purpose: Gastric cancer is highly heterogeneous.We aimed to identify these subregions using clustering,determining prognosis based on pretreatment contrastenhanced CT and FDG PET in stage Ⅲ gastric cancer.Methods and Materials:Thirty-seven patients with stage Ⅲ who underwent both baseline contrast-enhanced CT and PET were recruited.They all underwent surgical treatment and postoperative pathology was confirmed as stage Ⅲ.Patients are also followed up for more than one year.New hematogenous metastases,local recurrences,peritoneal metastases and lymphatic metastases during follow-up are defined as gastric cancer progression.Patients are divided into progressive（n=14）and non-progressive groups（n=13）.CT images are retrieved from within the PACS system and all patient PET information is stored on a disk from within the Nuclear Medicine Department system.As far as possible,the maximum level of tumors with similar CT axis positions in CT arterial and venous phase images and PET is taken;PET CT is used as a standard to resample CT arterial phase,venous phase images,pet nuclide tomography images so that both have the same number of pixels;Matlab is used for registration,and the selected method is control point registration,so that the tumor spatial position in CT arterial and venous phase images is aligned with PET CT images;the respective image information is converted into numerical information,and three sets of values are obtained,namely the CT value of the arterial and venous phase images and the corresponding SUV value.The k-means clustering method is used to cluster 5 classes so that the data within the same class have a high degree of similarity and the different classes have a high degree of difference.Images obtained from cluster analysis were processed using FROGSTATS landscape pattern index analysis software.Two independent nonparametric rank tests were performed on the progression group and the non-progression group,and the variables with statistical differences were included in the binary logistic regression model for analysis.Construct a ROC curve to evaluate diagnostic efficacy.Results: The tumor was divided into five different subregions using the k-means clustering method,namely metabolically active region with the highest SUV value,relatively high arterial phase and portal phase CT values;Blood-rich supply region with the highest arterial CT value and the relatively high SUV value portal phase CT value;with the highest portal CT value,the arterial phase CT value was relatively high,and the SUV value was at the intermediate level;the low metabolism,blood supply region where the portal phase was relatively high and the arterial CT value and SUV value were relatively low;the necrotic region where the SUV value,arterial phase and portal phase CT value were low.According to the progress group and nonprogress group,the numerical results of the FRAGSTATS landscape pattern index analysis software processing images to obtain five types of regions were examined by two independent samples and non-parametric ranks,and many statistically different variables were obtained.Construct a binary logistic regression analysis model for two independent samples with nonparametric ranks and meaningful variables obtained by checking.The model fits well,with significance greater than 0.05.For gastric cancer progression,blood-rich supply region CONTIG_MN had a significant negative correlation with the progression of gastric cancer patients,that is,the larger the CONTIG_MN value,the less likely it was to progress.CONTIG,called the proximity exponential distribution,is a shape indicator that evaluates spatial connectivity or continuity,with larger values indicating larger connectivity plaques.The diagnostic efficacy of the constructed model was evaluated by using the receiver operating characteristic curve（ROC）and the area under curve（AUC）,and AUC value was0.824 [95% confidence interval（CI）: 0.666~0.982].Conclusion: Clustering approach combining contrast-enhanced CT and PET is associated with the survival status of patients in gastric cancer.