Research on Automatic Screening System for Lung Cancer Detection.

我们建议开发一种利用CT扫描图像检测肺癌的自动筛查系统。实现这一目的的最重要课题是计算机对肺结节的检测精度要比目前的技术水平提高10倍以上。在本次研究活动中，我们开发了以下三种不同的特征提取和识别算法，并将这三种算法结合起来实现了高效的检测精度。(1)基于三维人工结构模型的异常结节与正常血管分类算法的开发。异常结节模型生成为具有各种半径的三维球体。正常的容器模型是由两个或三个具有不同连接角度和半径的圆柱体组合而成的。将输入图像与所有这些模型进行比较，选择与最终决策（异常或正常）相对应的最佳拟合模型。(2)基于感兴趣区域（ROI） CT强度信息主成分分析法（PCA）的分类算法。(3)基于启发式特征提取和统计判别函数的分类算法。

We proposed to develop an automatic screening system for lung cancer detection utilizing CT scanner images. The most important subject to perform this purpose is that the detection accuracy of lung nodules by computer should be realized more than 10 times higher level of present technical status.In this research activity, we developed the following three different algorithms for feature ectraction and recognition, and realized efficient detection accuracy by combining these three algorithms.(1)Development of a classification algorithm between abnormal nodules and normal vessels based on 3-D artificial structure models. Abnormal nodule models are produced as 3D sphere with various kinds of radii. Normal vessel models are produced as the combination of two or three cylinders with various kinds of connecting angles and radii. Input image is compared with all of these models and best fitting model is selected which correspond to final decision (abnormal or normal).(2)Development of a classification algorithm based on principal component analysis method (PCA) of CT intensity information in the region of interest(ROI).(3)Development of a classification algorithm based on heuristic feature extraction and statistical discriminant function.