数字化肺癌细胞病理诊断系统的研制和临床应用  

Implementation and application of a digital cytopathological lung cancer diagnosing system

作　　者：汪栋[1] 叶玉坤[1] 张缨[1] 高阳[2] 杨育彬[2] 周源[1] 郭爱勇[1] 赵波[2] 朱亮[2] 陈世福[2] 舒志强 钱雪庆 

机构地区：[1]解放军第八一医院全军肿瘤研究中心,南京210002 [2]南京大学计算机科学技术系软件新技术国家重点实验室 [3]南京长城信息系统有限公司

出　　处：《中华外科杂志》2007年第20期1417-1419,共3页Chinese Journal of Surgery

基　　金：国家"九五"重点科技攻关计划资助项目(96-906-01-18);国家"十五"重点科技攻关资助项目(2001BA703811);江苏省"十一五"科技攻关资助项目(BE2006011)

摘　　要：目的探讨数字化肺癌细胞病理诊断系统在临床肺癌细胞病理诊断中的应用效果。方法自动提取涂片上的细胞图像,运用 B 样条和改进 deBoor-Cox 方法对重叠细胞区域进行分离和可视化重构;运用基于强化学习技术的图像分割法将细胞区域从背景中分离出来,实现对目标图像的正确提取分割预处理;将细胞病理专家知识数字化用以提取较精确的细胞特征信息;运用决策树、支持向量机、贝叶斯等先进的分类算法,使系统拥有高精度和强分类能力,能同时进行癌与非癌的判断、肺癌细胞的分类(鳞癌、腺癌、小细胞癌及未分类癌)及核异型细胞评估。结果初步研制的数字化肺癌细胞病理诊断系统运行顺利,判断较为快速准确,随机应用于临床120例肺部病灶穿刺所得224幅细胞学涂片,肺癌识别诊断准确率92.3%,肺癌细胞的分类诊断符合率82.5%,核异型细胞判断识别率71.6%。结论数字化肺癌细胞病理诊断系统操作可行、对肺癌细胞学涂片判断准确率高,克服了重叠细胞识别率低、涂片染色差异和不良以及背景杂质等干扰因素,提供了相对客观统一的肺癌细胞病理学诊断策略,可用于肺部病灶穿刺细胞学识别分类诊断,为肺癌的早诊早治提供了一个重要的科学手段。Objective To evaluate the efficacy of the digital cytopathological lung cancer diagnosing system （DCLCDS） utilizing the latest computer technologies （including reinforcement learning, image segmentation and classifier） and the cytopathological knowledge on lung cancer cells. Methods Separate the overlapped lung cancer cells in a slice image applying the improved deBoor-Cox B-Spline algorithm; Segment cell regions in a slice image using an image segmentation algorithm based on reinforcement learning; Ensemble different classifiers, including Decision Tree classifier, Support Vector Machine （SVM） classifier and Bayesian classifier, to achieve an accurate result of cytopathological lung cancer diagnosis. Results The accurate diagnosis rate for lung cancer identification of 224 images of small lung lesions aspiration biopsy from 120 cases randomly selected was 92.3%. The accurate diagnosis rate for type classification of lung cancer was 82. 5%. The identification rate for abnormal nuclear cells was 71.6%. Condusions The DCLCDS achieves a high accuracy on cytopathological lung cancer diagnosis by solving some major problems on the cytology smears, including cell overlapping, uneven coloration and impurity. It provides a relatively objective, standard tool on cytopathological lung cancer diagnosis. It has good efficacy on early diagnosis of lung cancer.

关 键 词：肺肿瘤 诊断 计算机辅助 细胞病理 

分 类 号：R734.2[医药卫生—肿瘤]