基于痰涂片巴氏染色色度学特征的肺癌脱落细胞的模式识别研究

目的探讨基于巴氏染色色度学参数的痰涂片肺癌脱落细胞的模式识别方法,为肺癌脱落细胞的计算机诊断奠定基础。方法研究分为学习组和评价组,两组均进一步分为正常组和肺癌组;两个正常组又均分为正常的柱状上皮细胞、鳞状上皮细胞和组织细胞组;两个肺癌组又均分为鳞癌、腺癌、大细胞癌及小细胞癌细胞组。选取的参数包括上述细胞的红、绿、蓝三基色及三基色系数,核的红、绿、蓝三基色及其三基色系数。根据学习组测试结果建立判别函数,计算回代判别符合率;用评价组细胞判别诊断的准确率。结果所建立的判别函数对学习组正常与肺癌细胞判别的符合率分别为62.5%、95.9%,对评价组两组细胞判别诊断的准确率分别为63.4%、95.5%;对学习组鳞癌、腺癌、大细胞癌和小细胞癌癌细胞判别的符合率分别为72.5%、46.2%、81.2%、56.5%,对评价组四类癌细胞判别的准确率分别为71.8%、44.7%、82.7%、57.9%;对正常组柱状上皮细胞、组织细胞及鳞状上皮细胞判别的符合率分别为100.0%、95.8%、94.2%,对评价组三类正常细胞判别诊断的准确率分别为100%、95.8%、94.2%。学习组的判别符合率与评价组的诊断准确率两者间差异无显著性。结论用整个细胞及核的色度学参数建立的逐步判别函数对于判别痰涂片中的正常细胞与肺癌细胞,正常的柱状上皮细胞、组织细胞和鳞状上皮细胞,以及肺的鳞癌、腺癌、小细胞癌和大细胞癌四类亚型肺癌细胞有重要的价值,对于肺的小细胞癌及腺癌细胞的判别效果不佳;基于所建立的判别函数,提出了痰涂片肺癌脱落细胞色度学定量判别分析路径。

Study on pattern recognition for exfoliative cells of lung cancer based on chromaticity

Objective The aim of this study was to explore the methods of pattern recognition on lung cancer exfoliative cells in sputum smears based on the chromatic parameters,to establish a foundation of discrimination and classification of lung cancer exfoliative cells applying computer.Method The samples were divided into leaning group and evaluation group.The exfoliative cells included columnar epithelial cells,squamous epithelial cells,histiocytes,and cells of squamous cell carcinoma,adenocarcinoma,small cell carcinoma and large cell carcinoma.The chromatic parameters included three primary colors of cells and nuclei and their coefficient.The parameters submitted to stepwise discriminate analysis.The effects of discrimination between learning and evaluation group were compared.Results The coincidence of the function discriminating normal and cancer cells in learning group was 62.5% and 95.9%,respectively,that in evaluated group was 63.4% and 95.5%,respectively.The coincidence of the function discriminating cells of SqCa,AD,LCC and SCC in learning group was 72.5%,46.2%,81.2%,56.5%,respectively,that in evaluated group was 71.8%,44.7%,82.7% and 57.9%,respectively.The coincidence of the function discriminating cells of CE,HC,SqE in learning group is 100.0%,95.8%,94.2%,respectively,that in evaluated group was 100.0%,95.8%,94.2%,respectively.There was no significant difference on coincidence of the stepwise discrimination function between learning and evaluated group.Conclusions The discriminate function based on three primary colors and their coefficient has an important significance to discriminate between normal cells and lung cancer cells,and help to classify the lung cancer cells.However,there isn't noticeable value in discriminating the cells of AD and SCC.Therefore,a strategy has been proposed for discrimination of exfoliative cells in sputum smears based on chromatic features.