Predicting 5-Year Survival in Gastric Cancer Patients Using Iliopsoas Muscle CT Radiomics and Machine Learning Techniques

Abstract

Objectives Sarcopenia, linked to postoperative survival in cancer patients, was investigated in this study. The research explored the relationship between CT imaging features of muscles in gastric cancer patients and their survival. Additionally, the study aimed to create a quantifiable survival prediction model using artificial intelligence. Methods In a retrospective study, 100 patients who underwent radical gastrectomy for gastric cancer were analyzed. After identifying sarcopenia using the psoas muscle index, clinical factors related to patient survival were investigated. Imaging features were extracted from manually delineated iliopsoas muscles and used in 11 machine learning algorithms. After completing the model training, we used a dataset comprising 34 patients from a secondary center as an external validation set to evaluate the model’s classification performance. After identifying the optimal model, we further explored the fusion methods of clinical omics and radiomics. Based on this, we constructed a predictive model for estimating the five-year survival rate of patients. Results Clinical survival analysis highlighted age and tumor M stage as relevant factors. For the task of predicting five-year survival, we found that the Logistic Regression (LR) model without clinical feature fusion exhibited the most balanced and superior performance. Specifically, the AUC (Area Under Curve) values of this model on the training set, internal validation set, and external validation set were 0.82, 0.72, and 0.69, respectively. Additionally, the model’s accuracy remained relatively stable, approximately around 70%. Conclusions In this study, we developed a machine learning model based on preoperative CT imaging data of gastric cancer patients to predict their five-year survival rate. The model can achieve about 70% accuracy. Additionally, we explored the necessity and rationale of incorporating clinical independent factors into this predictive model. The results indicated a significant correlation between muscle imaging features and overall patient survival, highlighting the importance of sarcopenia in the clinical management of gastric cancer patients.