Wang Peizhe, Yan Jingrui, Qiu Hui, Huang Jingying, Yang Zhe, Shi Qiang, Yan Chengxin
Department of Medical Imaging, The Second Affiliated Hospital of Shandong First Medical University, Taian, 271000, Shandong, China.
Department of Gastroenterology, Shandong Provincial Hospital Affiliated to Shandong First Medical University, Jinan, 250021, Shandong, China.
J Cancer Res Clin Oncol. 2023 Nov;149(14):12993-13003. doi: 10.1007/s00432-023-05170-7. Epub 2023 Jul 19.
To discriminate the risk stratification in gastrointestinal stromal tumors (GISTs) by preoperatively constructing a model of nonenhanced computed tomography (NECT).
A total of 111 GISTs patients (77 in the training group and 34 in the validation Group) from two hospitals between 2015 and 2022 were collected retrospectively. One thousand and thirty-seven radiomics features were extracted from non-contract CT images, and the optimal radiomics signature was determined by univariate analysis and LASSO regression. The radiomics model was developed and validated from the ten optimal radiomics features by three methods. Covariates (clinical features, CT findings, and immunohistochemical characteristics) were collected to establish the clinical model, and both the radiomics features and the covariates were used to build the combined model. The effectiveness of the three models was evaluated by the Delong test.
The experimental results showed that the clinical models (75.3%, 70.6%), the radiomics models (79.2%, 79.4%) and the combined models (81.8%, 82.4%) all had high accuracy in predicting the pathological risk of GIST in both training and validation groups. The AUC values of the combined models were significantly higher in both the training groups (0.921 vs 0.822, p= 0.032) and the validation groups (0.913 vs 0.792, p= 0.019) than that of the clinical models. According to the calibration curve, the combined model nomogram is clinically useful.
The clinical-radiomics combined model and based on NECT performed well in discriminating the risk stratification in GISTs. As a quantitative technique, radiomics is capable of predicting the malignant potential and guiding treatment preoperatively.
通过术前构建非增强计算机断层扫描（NECT）模型来鉴别胃肠道间质瘤（GIST）的风险分层。
回顾性收集2015年至2022年两家医院的111例GIST患者（训练组77例，验证组34例）。从非增强CT图像中提取1037个放射组学特征，并通过单因素分析和LASSO回归确定最佳放射组学特征。通过三种方法从十个最佳放射组学特征中开发并验证放射组学模型。收集协变量（临床特征、CT表现和免疫组化特征）以建立临床模型，并将放射组学特征和协变量用于构建联合模型。通过德龙检验评估三种模型的有效性。
实验结果表明，临床模型（75.3%，70.6%）、放射组学模型（79.2%，79.4%）和联合模型（81.8%，82.4%）在训练组和验证组中预测GIST病理风险方面均具有较高的准确性。联合模型在训练组（0.921对0.822，p = 0.032）和验证组（0.913对0.792，p = 0.019）中的AUC值均显著高于临床模型。根据校准曲线，联合模型列线图具有临床实用性。
基于NECT的临床-放射组学联合模型在鉴别GIST风险分层方面表现良好。作为一种定量技术，放射组学能够预测恶性潜能并在术前指导治疗。
