Medical image recognition of lungs regions for multi-slice CT images by using the revised GMDH-type neural networks.

在这项研究中，我们提出了一种改进的GMDH型神经网络算法，该算法能够自动组织符合肺部X射线CT图像特征的最佳神经网络结构，并将该算法应用于肺部医学图像识别。在这些算法中，从Sigmoid函数型神经网络、RBF(径向基函数)型神经网络和多项式神经网络三种神经网络结构中自动选择最优的神经网络结构。此外，为了最小化定义为AIC(Akaike信息准则)或PSS(预测平方和)的预测误差准则，自动选择诸如层数、隐层中的神经元数目和最优神经元结构等结构参数。这些算法还具有从众多图像特征中自选择最佳输入变量的能力，从而使预测误差准则AIC或PSS最小化。因此，我们可以很容易地将这些修正的GMDH型神经网络算法应用到医学图像识别中，因为符合医学图像特征的最优神经网络结构是自动组织的。利用计算机中有组织的神经网络，自动提取X射线CT图像中的肺部区域，具有较高的精度。

In this study, we developed the revised GMDH (Group Method of Data Handling)-type neural network algorithms which can automatically organize the optimum neural network architectures fitting the characteristics of X-ray CT images of the lungs and applied this developed algorism to medical image recognition of the lungs. In these algorithms, the optimum neural network architecture is automatically selected from three types of neural network architectures such as sigmoid function type neural networks, RBF (Radial Basis Function) type neural networks and polynomial type neural networks. Furthermore, the structural parameters such as the number of layers, the number of neurons in the hidden layers and optimum neuron architectures are automatically selected so as to minimize the prediction error criterion defined as AIC ( Akaike's Information Criterion) or PSS (Prediction Sum of Squares). These algorithms have another ability of self-selecting the optimum input variables from many image characteristics so as to minimize the prediction error criterion AIC or PSS. Therefore, we can easily these revised GMDH-type neural network algorithms to the medical image recognition because the optimum neural network architecture fitting the medical image characteristics is automatically organized.In this study, we are applying these revised GMDH-type neural network algorithms to the medical image recognition of the lungs and we organize the optimum neural network architectures fitting the image characteristics of the lungs. By using the organized neural networks in the computer, the regions of the lungs of the X-ray CT images are automatically extracted with the good accuracy.