论文部分内容阅读
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 investigate the application of digital lung cancer cell pathological diagnosis system in the diagnosis of clinical lung cancer cells. Methods The cell images on the smears were extracted automatically. The B-spline and modified deBoor-Cox method were used to separate and visualize the overlapping cell regions. The image segmentation method based on reinforcement learning was used to separate the cell regions from the background and to achieve The correct extraction and segmentation of the target image preprocessing; digital knowledge of cytopathology experts to extract more accurate cell characteristics of information; the use of decision trees, support vector machines, Bayesian and other advanced classification algorithms, the system has high precision and strong The ability to classify can simultaneously make both cancerous and non-cancerous determinations, lung cancer cell classifications (squamous cell carcinoma, adenocarcinoma, small cell carcinoma and undifferentiated carcinoma) and nuclear atypical cell assessment. Results The preliminary diagnosis of lung cancer cell pathological diagnosis system was successful and the diagnosis was rapid and accurate. The results of this method were applied to 224 cytology smears obtained from 120 pulmonary lesions punctured randomly. The diagnostic accuracy of lung cancer was 92.3%. The classification of lung cancer was consistent Rate of 82.5%, nuclear abnormalities to determine the recognition rate of 71.6%. Conclusions Digital lung cancer cell pathological diagnosis system is feasible and feasible. It has high accuracy in judging cytology smears of lung cancer, and overcomes the interference factors such as low recognition rate of overlapping cells, differences and smears of smears and background impurities, and provides relatively objective and uniform lung cancer cells Pathological diagnosis strategy can be used for lung lesions puncture cytology classification diagnosis, early diagnosis and treatment of lung cancer provides an important scientific means.