基于CT征象的卵巢肿块恶性风险预测模型研究

发布时间：2018-02-04 20:11

本文关键词： 卵巢肿块恶性风险指数体层摄影术 X线计算机 Logistic回归模型　出处：《遵义医学院》2017年硕士论文　论文类型：学位论文

【摘要】：目的:探讨单独基于CT征象的Logistic回归模型对卵巢肿块恶性风险的预测,以获得有效且无创的卵巢肿块术前定性诊断方法。方法:收集四川大学华西第二医院2011年1月1日至2016年4月30日经病理证实为卵巢恶性肿块的患者305例,并随机收集2015年1月1日至2016年4月30日经病理证实为卵巢良性病变的患者100例作为对照。按照时间顺序划分恶性病例及随机方式均分良性病例,将所有病例分为实验组与验证组,并对所有病例的CT及超声征象进行分析评估,通过对实验组卵巢肿块CT征象的提取,以病理诊断为金标准建立Logistic回归模型(LR-CT),并用验证组数据对该模型进行检验。同时,参照基于超声征象的恶性风险指数3(RMI3)及RMI4系统设计了一种基于CT征象的RMI模型(RMI3-CT及RMI4-CT)。利用受试者工作特征(ROC)曲线评价LR-CT、RMI3、RMI4、RMI3-CT及RMI4-CT五种模型的诊断效能。结果:1、LR-CT模型主要与实性成分、分隔、坏死、网膜和或肠系膜转移、强化程度有关。2、CT对卵巢肿块诊断的敏感性较超声高,其对病灶的分期与临床分期一致性高(K=0.827)。3、基于CT征象的RMI3及RMI4诊断效能并不绝对优于基于超声征象的RMI3及RMI4,且基于CT和超声征象的RMI4系统也并不绝对优于R MI3系统,但CT增强表现的卵巢血管蒂征(OVPS)对于鉴别盆腹腔巨大肿块的起源及判断卵巢病灶单侧或双侧来源有很大的帮助。4、通过绘制ROC曲线,LR-CT恶性风险预测模型阈值为0.8153(81.53%)。五种模型中LR-CT曲线下面积(AUC)最大(0.947),各模型间比较均没有统计学差异(p0.05),即基于CT征象的LRCT、RMI3-CT、RMI4-CT模型与基于超声征象的RMI3、RMI4模型一样可有效用于卵巢肿块的定性诊断,且LR-CT的诊断效能最佳。5、LR-CT相对于其他模型来说其诊断效能相对温和,但假阳性率和对早期病灶诊断的漏、误诊率仍相对较高。结论:基于CT征象的Logistic回归模型(LR-CT)是一种有效的卵巢肿块术前定性预测方法,或许可成为一种临床前瞻性诊断方法。
[Abstract]:Objective: to study the Logistic regression model based on CT signs alone in predicting the malignant risk of ovarian masses. Methods: to obtain an effective and noninvasive preoperative qualitative diagnosis method for ovarian masses. A total of 305 patients with malignant ovarian masses confirmed by pathology from January 1st 2011 to April 2016 in Huaxi second Hospital of Sichuan University were collected. From January 1st 2015 to April 2016, 100 patients with benign ovarian lesions were randomly selected as controls. The malignant cases were divided into malignant cases and randomly divided into benign cases according to time order. Case. All the cases were divided into experimental group and validation group, and the CT and ultrasound signs of all cases were analyzed and evaluated, and the CT signs of ovarian masses were extracted from the experimental group. The Logistic regression model was established according to the gold standard of pathological diagnosis, and the model was tested with the validation group data. According to the malignancy risk index (RMI3) based on ultrasound signs and the RMI4 system, a RMI model based on CT signs, RMI3-CT and RMI4-CT, is designed. Objective: to evaluate LR-CT by using the operating characteristics of subjects. The diagnostic efficacy of RMI3 / RMI4 / RMI3-CT and RMI4-CT models were analyzed. Results the RMI3 / RMI4 / RMI3-CT model was mainly separated from solid components and necrotic. The enhanced degree of omentum and mesenteric metastasis was more sensitive to the diagnosis of ovarian mass than that of ultrasound, and the correlation between the staging and clinical staging was higher than that of ultrasound. The diagnostic efficacy of RMI3 and RMI4 based on CT signs is not absolutely superior to that of RMI3 and RMI4 based on ultrasound signs. And the RMI4 system based on CT and ultrasound signs is not absolutely superior to R MI3 system. But CT enhanced ovarian vascular pedicle sign (OVPSS) is helpful in differentiating the origin of large pelvic and abdominal masses and judging the origin of ovarian lesions on one or both sides. 4. The ROC curve was plotted. The threshold value of LR-CT malignant risk prediction model was 0.8153 and 81.53.The maximum area under the LR-CT curve was 0.947 among the five models. There was no statistical difference between each model (p0.05), that is, LRCTT RMI3-CTN RMI4-CT model based on CT signs and RMI3 based on ultrasound signs. RMI4 model can be used in the qualitative diagnosis of ovarian masses as well, and the diagnostic efficacy of LR-CT is the best. 5% LR-CT compared with other models, the diagnostic efficacy of LR-CT is relatively mild. But false positive rate and leakage of diagnosis of early lesions. Conclusion: the Logistic regression model based on CT signs is an effective method to predict the preoperative quality of ovarian masses. It may be a prospective clinical diagnostic method.
【学位授予单位】：遵义医学院
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R737.31;R730.44

【相似文献】