基于CT影像特征构建鉴别胃神经鞘瘤与胃间质瘤的列线图模型

A nomogram model for differentiating gastric schwannoma from gastric stromal tumor based on CT imaging features

目的探讨基于CT影像特征构建术前鉴别诊断胃神经鞘瘤(GS)与胃间质瘤(GST)(长径2~5 cm)的列线图模型。方法回顾性分析2009年7月至2023年4月济宁医学院附属医院和2017年6月至2022年9月广东省人民医院49例GS患者和240例GST患者的临床和影像资料。采用多因素Logistic回归分析鉴别GS与GST的独立因素。采用R4.3.1软件构建鉴别GS和GST的列线图模型,绘制受试者工作特征(ROC)曲线,评估列线图模型鉴别GS与GST的效能;通过校准曲线和决策曲线评估列线图模型的预测效能和临床应用价值。结果GS患者和GST患者有临床症状率、钙化率、溃疡率、肿瘤血管率、长短径比值和动脉期与平扫CT值差值(CTV A-N)比较差异无统计学意义(P>0.05)。GS患者女性比例、发生于胃中下部率、混合或腔外型生长率、肿瘤相关淋巴结率、明显强化率、静脉期与平扫CT值差值(CTV P-N)、延迟期与平扫CT值差值(CTV D-N)、静脉期与动脉期CT值差值(CTV P-A)和延迟期与静脉期CT值差值(CTV D-P)明显高于GST患者[75.51%(37/49)比58.33%(140/240)、85.71%(42/49)比54.17%(130/240)、75.51%(37/49)比45.00%(108/240)、44.90%(22/49)比5.42%(13/240)、51.02%(25/49)比27.08%(65/240)、32.0(26.0,43.5)HU比29.0(22.0,37.7)HU、(44.59±13.46)HU比(32.94±12.47)HU、20.0(11.5,25.0)HU比10.0(5.0,17.0)HU和9.0(6.0,12.0)HU比4.0(-2.7,7.0)HU],年龄、形态不规则率、囊变率和不均匀强化率明显低于GST患者[(58.12±12.59)岁比(62.05±11.22)岁、16.33%(8/49)比38.33%(92/240)、18.37%(9/49)比51.25%(123/240)和34.69%(17/49)比56.25%(135/240)],差异有统计学意义(P<0.05或<0.01)。多因素Logistic回归分析结果显示,部位、囊变、肿瘤相关淋巴结、CTV P-A和CTV D-P是鉴别GS与GST的独立影响因素(OR=3.599、0.201、19.031、1.124和1.160,95%CI 1.184~10.938、0.070~0.578、6.159~58.809、1.066~1.185和1.094~1.231,P<0.05或<0.01)。基于部位、囊变、肿瘤相关淋巴结、CTV P-A和CTV D-P构建鉴别GS与GST的列线图模型。列线图鉴别GS与GST的曲线下面积为0.924(95%CI 0.887~0.951)。校准曲线分析结果显示,列线图模型预测GS曲线与实际GS曲线之间一致性良好(平均绝对误差为0.033),Hosmer-Lemeshow检验提示该列线图模型不存在过度拟合(χ^(2)=2.52,P=0.961)。临床决策曲线分析结果显示,当列线图模型鉴别两种肿瘤的阈值>0.03时,列线图模型较"所有患者视为GS"方案或"所有患者视为GST"方案获得更多的净收益。结论基于CT影像特征的列线图模型可用于术前鉴别GS与GST。

ObjectiveTo construct a nomogram model for differentiating gastric schwannoma(GS)from gastric stromal tumor(GST)(diameters 2 to 5 cm)based on CT imaging features before surgery.MethodsThe clinical and imaging data of 49 patients with GS and 240 patients with GST in the Affiliated Hospital of Jining Medical University from July 2009 to April 2023 and Guangdong Provincial People′s Hospital from June 2017 to September 2022 were analyzed retrospectively.The independent factors for differentiating GS from GST were obtained by multivariate Logistic regression analysis.The nomogram model was constructed by R4.3.1 software.The efficacy of the nomogram model for differentiating GS from GST was evaluated by the receiver operating characteristics(ROC)curve,and calibration curve and decision curve analysis were used to evaluate the predictive efficacy and clinical application value of the nomogram model.ResultsThere were no statistical differences in the clinical symptom rate,calcification rate,ulcer rate,tumor vessel rate,ratio of long diameter to short diameter and CT value difference during the arterial and nonenhanced phases(CTV A-N)between GS patients and GST patients(P>0.05).The proportion of female,incidence of lesions located in central or lower part of stomach,extraluminal or mixed growth rate,tumor-associated lymph node rate,strong enhancement rate,CT value difference during the portal and nonenhanced phases(CTV P-N),CT value difference during the delayed and nonenhanced phases(CTV D-N),CT value difference during the portal and arterial phases(CTV P-A)and CT value difference during the delayed and portal phases(CTV D-P)in GS patients were significantly higher than those in GST patients:75.51%(37/49)vs.58.33%(140/240),85.71%(42/49)vs.54.17%(130/240),75.51%(37/49)vs.45.00%(108/240),44.90%(22/49)vs.5.42%(13/240),51.02%(25/49)vs.27.08%(65/240),32.0(26.0,43.5)HU vs.29.0(22.0,37.7)HU,(44.59±13.46)HU vs.(32.94±12.47)HU,20.0(11.5,25.0)HU vs.10.0(5.0,17.0)HU and 9.0(6.0,12.0)HU vs.4.0(-2.7,7.0)HU,the age,irregular shape rate,cystic degeneration rate and heterogeneous enhancement rate were significantly lower than those in GST patients:(58.12±12.59)years old vs.(62.05±11.22)years old,16.33%(8/49)vs.38.33%(92/240),18.37%(9/49)vs.51.25%(123/240)and 34.69%(17/49)vs.56.25%(135/240),and there were statistical differences(P<0.05 or<0.01).Multivariate Logistic regression analysis result showed that location,cystic degeneration,tumor-associated lymph node,CTV P-A and CTV D-P were the independent factors for differentiating GS from GST(OR=3.599,0.201,19.031,1.124 and 1.160;95%CI 1.184 to 10.938,0.070 to 0.578,6.159 to 58.809,1.066 to 1.185 and 1.094 to 1.231;P<0.05 or<0.01).The nomogram model for differentiating GS from GST was constructed based on location,cystic degeneration,tumor-associated lymph node,CTV P-A and CTV D-P.The area under curve of the nomogram model for differentiating GS from GST was 0.924(95%CI 0.887 to 0.951).The calibration curve analysis result showed that there was a good agreement between the predicted GS curve and the actual GS curve(the mean absolute error was 0.033).The result of the Hosmer-Lemeshow goodness-of-fit test indicated that the calibration of the nomogram model was appropriate(χ^(2)=2.52,P=0.961).The clinical decision curve analysis result showed that when the threshold for the nomogram model for differentiating the two tumors was>0.03,the nomogram yielded more net benefits than the"all patients treated as GS"or"all patients treated as GST"scenarios.ConclusionsThe nomogram model based on CT imaging features can be used to differentiate GS from GST before surgery.