基于神经网络集成的肺癌早期诊断

将病理性诊断与计算机技术相结合以实现肺癌的早期诊断，首先利用数字图像技术对肺癌穿刺样本进行处理，提出取形态和色度特征，然后通过一种二级集成结构和特殊的投票方式，用神经网络集成对细胞图象进行分析，实验和原型系统试用表明，方法的总误诊率和肺癌患者漏诊率均低于单一神经网络方法和常用的神经网络集成方法。

EARLY STAGE LUNG CANCER DIAGNOSIS BASED ON NEURAL NETWORK ENSEMBLE

In this paper, pathological diagnosis is combined with computer techniques for early stage diagnosis of lung cancer. Firstly, punctured samples of lung cancer are processed by digital image technique, extracting morphologic and chromatic features. Then, the cell images are analyzed by neural network ensemble with a two-layered architecture and a specific voting scheme. Experiments and the probation of a prototype system show that both the overall misdiagnosis rate and the rate of missed diagnosis of lung cancer sufferers are lower than that of the single neural network and commonly-used neural network ensemble methods.