超声图像中胆嚢真假性息肉计算机辅助鉴别  

Computer Aided Identification of True and Pseudo Gallbladder Polyps on Ultrasound Images

作　　者：余其徽 袁海霞 张燕群 俞清 季正标 张麒 王文平[3,4] YU Qihui;YUAN Haixia;ZHANG Yanqun;YU Qing;JI Zhengbiao;ZHANG Qi;WANG Wenping(Shanghai Institute for Advanced Communication and Data Science,Shanghai University,Shanghai 200444,China;School of Communication and Information Engineering,Shanghai University,Shanghai 200444,China;Department of Ultrasound,Zhongshan Hospital of Fudan University,Shanghai 200032,China;Department of Ultrasound,Xiamen Branch,Zhongshan Hospital of Fudan University,Xiamen 361015,China;Hangzhou YITU Healthcare Research Institute,Hangzhou 310000,China)

机构地区：[1]上海先进通信与数据科学研究院(上海大学),上海200444 [2]上海大学通信与信息工程学院,上海200444 [3]复旦大学附属中山医院超声科,上海200032 [4]复旦大学附属中山医院厦门医院超声科,福建厦门361015 [5]杭州依图医疗研究院,浙江杭州310000

出　　处：《自动化仪表》2021年第4期86-90,共5页Process Automation Instrumentation

基　　金：国家自然科学基金资助项目(61671281、61911530249);上海市临床重点专科基金资助项目(shsk:zdzk03501);福建省卫生健康科研人才基金资助项目(2019-ZQNB-39);厦门市科技计划(医疗卫生项目)基金资助项目(3502Z20184002)。

摘　　要：超声检查是胆囊息肉样病变首选的影像学检查方法,但仅根据常规超声对病变回声、形态、血流的检测结果来区分真假性息肉是不可靠的,难以满足快速准确的术前鉴别需求。基于计算机辅助分析,对经病理证实的胆囊腺瘤真性息肉31例31个病灶和胆囊胆固醇假性息肉37例38个病灶进行研究。从术前超声图像中提取病灶的空域和形态特征,接着对特征进行统计分析并用于支持向量机分类,以鉴别胆囊腺瘤及胆囊胆固醇性息肉。结果表明,胆囊腺瘤组的均一度显著高于胆固醇息肉组(P<0.001),胆囊腺瘤组的像素分布较胆固醇息肉更均匀。相较于空域或形态特征对应的分类模型,所有特征对应的集成模型性能提升。分类准确率,敏感性、特异性分别达到90.5%、91.0%、90.0%,曲线下面积为0.927。因此,计算机辅助分析超声图像有助于提高胆囊真性和假性息肉的诊断准确性。Ultrasonography is the preferred imaging method for gallbladder polyp lesion examination.However,it is unreliable to distinguish true-and pseudo-polyps only based on the results of lesion echo,morphology,and blood flow obtained by conventional ultrasound and it is difficult to meet the needs of rapid and accurate preoperative identification.A study of 31 pathologically proven gallbladder adenoma true-polyps(31 patients)and 38 pathologically proven gallbladder cholesterol pseudo-polyps(37 patients)was carried out based on computer-aided diagnosis.Spatial and morphological features were acquired from the preoperative ultrasound images for the lesion.Following this,these features were analyzed statistically and thus apply to the differential diagnosis with the support vector machine.Results show that the Homogeneity value in the gallbladder adenoma group is significantly higher than that in the cholesterol polyp group(P<0.001),indicating that the pixel distribution of gallbladder adenoma is more even.Relative to the classification model using spatial or morphological features,the integration model using all features perform better.The accuracy,sensitivity and specificity values are 90.5%,91.0%and 90.0%,respectively,and the value of area under the curve is 0.921.Therefore,it is helpful to improve the diagnostic accuracy of gallbladder true-and pseudo-polyps based on computer-aided analysis of ultrasound images.

关 键 词：胆囊真性息肉 胆囊假性息肉 胆囊腺瘤 胆囊胆固醇性息肉 术前鉴別 模式识别 

分 类 号：TH-77[机械工程]