基于影像组学鉴别BorrmannⅣ型胃癌和原发性胃淋巴瘤的研究

发布时间：2018-04-26 08:55

本文选题：BorrmannⅣ型胃癌 + 原发性胃淋巴瘤　；参考：《南方医科大学》2017年硕士论文

【摘要】：背景:胃癌和原发性胃淋巴瘤(PGL)是胃部最常见的两种恶性肿瘤,内镜和CT成像难以将其鉴别,尤其是Borrmann Ⅳ型胃癌与胃淋巴瘤的表现较为相似,诊断时容易混淆。然而,胃癌和原发性胃淋巴瘤的治疗方法及预后存在很大差异,因此,临床需探讨其他有效途径对其进行鉴别。近年来兴起的影像组学(radiomics)成为影像研究领域的热点,在疾病的分型诊断、疗效评估和预后预测方面体现出重要价值,为鉴别胃癌和原发性胃淋巴瘤提供了新的思路和方法。目前,影像组学在鉴别Borrmann Ⅳ型胃癌与胃淋巴瘤方面的价值如何,尚未见报道。目的:探讨基于CT的影像组学方法在鉴别Borrmann Ⅳ型胃癌和原发性胃淋巴瘤方面的价值。方法:本课题来源于国家自然科学基金(N0.81271569),该回顾性研究通过了广东省人民医院(广东省医学科学院)伦理委员会的批准(批准号:GDREC2012020H)。本研究最终纳入确诊为Borrmann Ⅳ型胃癌患者40例及原发性胃淋巴瘤患者30例,均具备完整的临床病理资料及治疗前的上腹部CT增强扫描图像。由两名在腹部影像诊断具有丰富经验的副主任医师独立评估CT图像的主观征象,包括胃壁蠕动性、胃周脂肪浸润情况、肾门平面以下肿大淋巴结及强化模式,并通Kappa检验评估观察者间的可重复性。通过多元回归分析联合以上CT征象,构建基于CT主观征象的鉴别诊断模型。另外,两名医师对肿瘤进行手动分割,进而提取影像组学特征(共485个),并通过Lasso-logistic回归模型对影像组学特征进行筛选,构建影像组学标签。通过组间一致性检验(interclass correlation coefficient,ICC)比较两名测量者在影像组学评估上的可重复性。以独立样本t检验或Mann-WhitneyU检验比较两组间在年龄、影像组学标签的差异。采用卡方检验或Fisher's确切概率检验法比较两组患者在性别、CT主观征象的差异。将单因素比较有统计学差异的参数纳入多元回归分析,构建联合诊断预测模型。使用受试者操作曲线(ROC)检验CT主观征象模型、影像组学标签及联合诊断模型在诊断Borrmann Ⅳ型胃癌和原发性胃淋巴瘤中的诊断效能,参数包括:曲线下面积(area under the curve,AUC)、敏感性、特异性和准确性。模型的诊断效能采用七倍交叉验证法进行验证。各诊断模型间的诊断效能比较采用Delong检验。结果:Borrmann Ⅳ型胃癌和胃淋巴瘤两组患者在性别、胃壁蠕动性及强化模式、影像组学标签上,差异存在统计学意义(所有P0.01)。CT主观征象模型、影像组学标签、联合诊断模型的AUC值及95%CI(confidence interval)分别为 0.806(0.696-0.917)、0.886(0.809-0.963)和 0.903(0.831-0.975);敏感性分别为63.33%、86.67%和70%;特异性分别为95%、82.5%和100%;准确性分别为81.43%、84.29%和87.14%。CT主观征象模型的AUC值、敏感性及准确性均低于影像组学标签及联合诊断模型。影像组学标签的敏感性及特异性均较高。联合诊断模型的AUC值、准确性及特异性在三个模型中均最高。然而,三个诊断模型之间的AUC值比较无统计学意义(CT主观征象模型vs影像组学标签,P=0.188;CT主观征象模型vs联合诊断模型,P=0.051;影像组学标签vs联合诊断模型,P=0.422)。交叉验证后的AUC值进行两两比较,差异亦无统计学意义(P=0.065-0.279)。结论:基于治疗前CT图像的影像组学标签可用于鉴别Borrmann Ⅳ型胃癌和原发性胃淋巴瘤。作为定量方法,影像组学分析对评估者的经验依赖性更小,且易于并入现有的影像工作流程,有望成为影像医师的重要辅助工具。
[Abstract]:Background: gastric cancer and primary gastric lymphoma (PGL) are the two most common malignant tumors in the stomach. Endoscopy and CT imaging are difficult to identify them. Especially, the manifestations of Borrmann IV type gastric cancer and gastric lymphoma are similar, and the diagnosis is easily confused. However, there are great differences in the treatment and prognosis of gastric cancer and primary gastric lymphadenoma. In recent years, the rise of radiomics has become a hot spot in the field of imaging research, which is of great value in the diagnosis of disease, evaluation of curative effect and prediction of prognosis, and provides new ideas and methods for the identification of gastric cancer and primary gastric lymphoma. The value of non Borrmann type IV gastric cancer and gastric lymphoma has not been reported. Objective: To explore the value of CT based imaging group method in the identification of Borrmann IV gastric cancer and primary gastric lymphoma. Methods: this topic was derived from the National Natural Science Foundation (N0.81271569), and the retrospective study was carried out through the people's medicine in Guangdong province. The Institute (Guangdong Academy of Medical Sciences) Ethics Committee approved (approval number: GDREC2012020H). This study finally included 40 patients with Borrmann IV gastric cancer and 30 cases of primary gastric lymphoma. All of them have complete clinicopathological data and CT enhanced scan images of the upper abdomen before treatment. Two in the abdominal imaging diagnosis are abundant. The experienced deputy chief physician assessed the subjective signs of CT images independently, including the peristalsis of the stomach wall, the infiltration of the stomach peririphal, the enlarged lymph nodes and intensification patterns below the renal portal plane, and evaluated the repeatability between the observers through the Kappa test. Through multiple regression analysis, the above CT signs were combined to construct a differential diagnosis model based on the subjective signs of CT. In addition, two physicians split the tumor manually, and then extracted the image group characteristics (485), and screened the image histology characteristics by Lasso-logistic regression model, and constructed the image group label. Through the interclass correlation coefficient (ICC), two surveyors were compared to the image histology evaluation. The difference between two groups of age and image group labels was compared by independent sample t test or Mann-WhitneyU test. The differences in gender and CT subjective signs were compared with the chi square test or the exact Fisher's test method of the two groups. Diagnostic prediction model. The diagnostic effectiveness of the CT subjective image model, image group label and joint diagnostic model in the diagnosis of Borrmann type IV gastric cancer and primary gastric lymphoma using the subject operator curve (ROC), parameters including the area under the curve (area under the curve, AUC), sensitivity, specificity and accuracy. The diagnostic effectiveness of the model. The seven times cross validation method was used to verify the results. The diagnostic effectiveness of the diagnostic models was compared with the Delong test. Results: there were statistically significant differences in gender, gastric wall peristalsis and intensification patterns in two groups of gastric cancer and gastric lymphoma. The difference was statistically significant (all P0.01).CT subjective signs, image group label, The AUC value and 95%CI (confidence interval) of the combined diagnostic model were 0.806 (0.696-0.917), 0.886 (0.809-0.963) and 0.903 (0.831-0.975), and the sensitivity was 63.33%, 86.67% and 70%, respectively, and the specificity was 95%, 82.5% and 100%, respectively. The accuracy of the model was 81.43%, 84.29% and 87.14%.CT, respectively, and the sensitivity and accuracy were low. The sensitivity and specificity of imaging group labels were high. The AUC value, accuracy and specificity of the combined diagnostic model were the highest among the three models. However, the AUC values between the three diagnostic models were not statistically significant (CT subjective image Model vs image group label, P=0.188; CT subjective sign) Image Model vs joint diagnosis model, P=0.051, image group label vs joint diagnostic model, P=0.422). The AUC value after cross validation was 22 compared, and the difference was not statistically significant (P=0.065-0.279). Conclusion: the image group label based on CT image before treatment can be used to identify Borrmann type IV gastric cancer and primary gastric lymphoma. The method of image analysis has less experience dependence on the assessor, and is easy to incorporate into the existing image workflow. It is expected to become an important auxiliary tool for the image doctor.

【学位授予单位】：南方医科大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R735.2;R730.44

【参考文献】