概率神经网络在胃镜样品红外光谱检测中的应用

Application of Probabilistic Neural Networks Method to Gastric Endoscope Samples Diagnosis Based on FTIR Spectroscopy

将人工神经网络方法应用于人体胃镜样品红外光谱检测,以克服常规线性判别分析方法的局限性,从而提高了胃镜样品判别的准确率。概率神经网络是一种适用模式分类的径向基神经网络,采用样本的先验概率和最优判定原则对新的样本进行分类,具有识别率高、训练速度快、不会陷入局部极值等优点。文章采用概率神经网络进行胃镜样品红外光谱模式识别,将预处理后的胃镜样品光谱进行主成分分析,将得分值作为输入,建立概率神经网络判别模型。文中选取118例胃镜离体样品进行红外光谱判别分析,其中正常胃组织19例,胃炎组织64例,胃癌35例,选取其中59例样品建立概率神经网络校正模型,其余样品作为预测集来检验模型。实验结果表明,正常、炎症及癌症胃镜样品检测的总体准确率达到81.4%,对胃镜样品的判别取得了较好的结果。

In the present paper, probabilistic neural network method was applied to the classification of gastric endoscope samples based on FTIR spectroscopy for higher discrimination correctness than the conventional linear discriminant analysis algorithm. The probabilistic neural network （PNN） is a kind of radial basis network suitable for discriminant analysis. There are several advantages of PNN method： less time is needed to train the model, higher correctness could be achieved, global optimal solution could be obtained and so on. In this paper, PNN method was utilized to classify gastric endoscopic biopsies into healthy, gastritis, and malignancy. Firstly, principal component analysis was carried out for the pretreated sample spectra. Principal components analysis is a quantitatively rigorous method for achieving the simplification. The method generates a new set of vari- ables, called principal components. Each principal component is a linear combination of the original variables. All the principal components are orthogonal to each other, so there is no redundant information. The principal components as a whole form an or- thogonal basis for the space of the data. And then, the scores of principal components were selected as input to train the PNN model. Finally, PNN model was established. In this experiment, a total of 118 gastric endoscopic biopsies, including 35 cases of cancer, 64 cases of gastritis, and 19 healthy tissue samples, were obtained at the First Hospital of Xi＇an Jiaotong University, China. Fifty nine samples were selected to establish the PNN classification model. The rest of the samples were used as the test set to valid the discriminant analysis model. The total discrimination correctness of normal, inflammation and gastric cancer achieved 81.4%.