基于CT的影像组学模型用于预测胃癌生存预后

CT-Based Radiomics Model for Predicting Gastric Cancer Survival Prognosis

摘要: 目的：胃癌(gastric cancer, GC)仍是我国的主要死亡原因之一。本研究的目的是建立基于计算机断层扫描图像(computed tomography, CT)的影像组学模型结合临床特征来预测胃癌术后患者总生存期(overall survival, OS)。方法：本研究回顾性纳入2015年1月至2020年12月青岛大学附属医院确诊GC并接受手术治疗的117例患者，按7:3随机分为训练集与验证集。收集患者临床资料及术前腹部CT图像，经GAN重建预处理后勾画感兴趣区域，提取放射组学特征并LASSO回归筛选，分别构建影像组学模型、临床风险模型及综合模型。采用C-index、Kaplan-Meier生存分析等评估模型效能。结果：影像组学模型预测效能最优，在验证集中C-index为0.758，优于临床预测模型(0.703)及综合模型(0.739)。Kaplan-Meier分析显示，各模型高低风险组OS差异均有统计学意义(P < 0.05)。同时，Cox多因素分析证实，淋巴血管浸润是OS的独立危险因素。结论：基于术前CT的影像组学模型可有效、无创预测GC患者术后OS，为临床预后评估及个体化治疗提供重要参考。

Abstract: Objective: Gastric cancer (GC) remains one of the major causes of death in China. The purpose of this study was to establish a radiomics model based on computed tomography (CT) images combined with clinical features to predict the overall survival (OS) of patients after gastric cancer surgery. Methods: A total of 117 patients diagnosed with GC and treated with surgery at the Affiliated Hospital of Qingdao University from January 2015 to December 2020 were retrospectively enrolled. They were randomly divided into a training set and a validation set at a ratio of 7:3. Clinical data of patients and preoperative abdominal CT images were collected. After preprocessing with GAN reconstruction, the regions of interest (ROI) were delineated, radiomic features were extracted and screened by LASSO regression, and a radiomics model, a clinical risk model, and an integrated model were constructed respectively. The model performance was evaluated using C-index, Kaplan-Meier survival analysis, and other methods. Results: The radiomics model showed the best predictive performance, with a C-index of 0.758 in the validation set, which was superior to the clinical prediction model (0.703) and the integrated model (0.739). Kaplan-Meier analysis showed that there were statistically significant differences in OS between the high-risk and low-risk groups of each model (P < 0.05). Meanwhile, multivariate Cox analysis confirmed that lymph vascular invasion was an independent risk factor for OS. Conclusion: The radiomics model based on preoperative CT can effectively and non-invasively predict the postoperative OS of GC patients, providing important reference for clinical prognosis evaluation and individualized treatment.