基于超声指标建立卵巢肿瘤性质的Fisher判别模型

Fisher Discriminant Analysis for Diagnosis Characters of Ovarian Tumors Based on Ultrasonography Index

目的:利用超声指标构建卵巢肿瘤性质Fisher判别模型,以实现良性、交界性与恶性卵巢肿瘤的客观、准确鉴别。方法:共纳入经病理证实的良性卵巢肿瘤116个、交界性肿瘤23个、恶性肿瘤75个,观察彩色多普勒超声影像表现,并建立Fisher判别模型,回代法验证。结果:(1)单因素分析显示,良性、交界性和恶性卵巢肿瘤在超声显示的形态、最大径、物理性质、内壁分隔光滑度、乳头状突起数、腹水和血流信号方面的差异有统计学意义(P<0.05)。(2)Fisher逐步判别得到判别函数:Function 1=0.956形态+0.269最大径+0.305物理性质+1.279内壁分隔光滑度+1.172乳头状突起数+0.886腹水+0.541血流信号-3.589;Function 2=0.103形态+0.241最大径-0.136物理性质-0.410内壁分隔光滑度-0.051乳头状突起数+1.207腹水-0.770血流信号-0.105。(3)良性、交界性和恶性卵巢肿瘤在二维坐标图中的散点投影位置具有聚类性,但也存在部分交叉重叠。(4)模型诊断良性、交界性和恶性卵巢肿瘤的敏感度分别是86.21%(100/116)、60.87%(14/23)和81.33%(61/75),总体符合率81.78%(175/214)。结论:超声显示的肿瘤形态、最大径、物理性质、内壁分隔、乳头状突起数、腹水和血流信号是鉴别卵巢肿瘤性质的重要指标,以此建立的Fisher判别模型有助于良性、交界性和恶性卵巢肿瘤的鉴别诊断。

Objective:To construct Fisher discrminant model for diagnosis characters of ovarian tumors with index of ultrasonography,so as to objectively and accurately distinguish the benign,borderline and malignant ovarian tumors.Methods:A total of 116 benign,23 borderline and 75 malignant ovarian tumors by pathology were enrolled in this study.The ultrasonic manifestations of patients were analyzed,a diagnosis model was developed using stepwise discriminant analysis,and the diagnostic ability of the model was verified with jackknife method.Results:(1)Univariate analysis indicated that ovarian tumors shape,maximum diameter,physical property,inner wall and septation smoothness,the number of papillary projections,ascites and blood flow signal have statistical difference(P<0.05)among the three kinds of ovarian tumors.(2)The Fisher discriminant functions were as following:Function 1=0.956 shape+0.269 maximum diameter+0.305 physical property+1.279 inner wall and septation smoothness+1.172 number of papillary projections+0.886 ascites+0.541 blood flow signal-3.589;Function 2=0.103 shape+0.241 maximum diameter-0.136 physical property-0.410 inner wall and septation smoothness-0.051 number of papillary projections+1.207 ascites-0.770 blood flow signal-0.105.(3)The projective positions of benign,borderline and malignant ovarian tumors at 2D coordinates were clustering,but there were some projective positions crossed and overlapped.(4)The sensibility of model for diagnosis benign,borderline and malignant ovarian tumors was 86.21%(100/116),60.87%(14/23)and 81.33%(61/75),the total diagnostic accuracy for the patients was 81.78%(175/214).Conclusion:Shape,maximum diameter,physical property,inner wall and septation smoothness,number of papillary projections,ascites and blood flow signal are the significant differential prognostic variables.Fisher discrimination model based on theses index is helpful to differential diagnosis of benign,borderline and malignant ovarian tumors.