期刊文献+

神经网络集成在肺癌细胞识别中的应用 被引量:19

Applications of Neural Network Ensemble in Lung Cancer Cell Identification
下载PDF
导出
摘要 通过一种特殊的二级集成结构将神经网络集成应用于肺癌细胞识别 .集成中的个体网络由Bagging方法产生 ,第一级集成的归纳结论由本文提出的完全投票法合成 ,第二级集成的归纳结论由相对多数投票法合成 .实验和原型系统试用表明 ,该方法的总体误识率较低 ,更重要的是 ,其将癌细胞判别为非癌细胞的误识率非常低 。 Lung cancer is one of the most deathful diseases in the world. In general, 80% of the patients were in the terminal stages when their illnesses were diagnosed as lung cancer. However, research reveals that if their illnesses can be diagnosed in early stages, the 5 year survival rate of the patients can be improved to about 40%. Therefore trying to find out the canceration at the earliest opportunity is a very pressing task. Until now, most work on computer aided lung cancer diagnosis exploit X ray chest images or CT images. Some exploit bronchoscopy. But the most reliable process of diagnosing lung cancer, i.e. pathologic diagnosis, has not been utilized. In this paper, an early stage diagnosis system LCDS for lung cancer is presented. LCDS employs digital image processing and neural network techniques to analyze the cell images of puncturing specimen from the patients, and uses the proof of the existence of cancer cells to perform diagnosis. Therefore the pathologic techniques are tightly combined with computer techniques. LCDS employs CCD and image acquisitor to get cell images of high quality that are gradually processed by cleaning, smoothing, sharpening, segmenting and filtering procedures. Then the morphological features and colorimetric features are extracted, which are fed to a neural network ensemble to identify cancer cells. In this paper, a specific two layer ensemble architecture is devised to accomplish this task. The neural model used here is FTART2, which is a fast neural classifier that combines the advantages of the adaptive resonance theory and the field theory. The individual networks in the ensemble are generated by Bagging that is capable of improving the predictive accuracy of unsteady learning algorithms such as neural networks and decision trees. The individual networks in the first layer ensemble are combined by a strategy named full voting, which is proposed in this paper. Only when all the individual networks admit that there is no cancer cell, the output of full voting is 'normal cell'; otherwise the output is 'may be cancer cell'. The individual networks in the second layer ensemble are combined by plurality voting, which is a popular way of ensembling multiple classifiers. Experiments show that the overall error identification rate of the specific neural network ensemble of LCDS system is quite low. What is more important is that its error identification rate of identifying normal cancer cells is so low that the probability of misdiagnosing lung cancer patients to be healthy persons is greatly decreased.
出处 《南京大学学报(自然科学版)》 CAS CSCD 北大核心 2001年第5期529-534,共6页 Journal of Nanjing University(Natural Science)
基金 国家自然科学基金 (6 9875 0 0 6 ) 江苏省自然科学基金 (BK990 36 )
关键词 神经网络集成 计算机辅助诊断 模式识别 图象处理 肺癌 细胞识别 完全投票法 neural networks, ensemble, computer aided diagnosis, pattern recognition, image processing
  • 相关文献

参考文献4

二级参考文献10

共引文献23

同被引文献267

引证文献19

二级引证文献237

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部