Application of Multidimensional Features and GRNN in Electronic Nose for Detection of Hepatocellular Carcinoma

Abstract

A model based on multidimensional features and GRNN was designed for electronic nose (eNose) in the paper. It can be applied to distinguish hepatocellular carcinoma from normal controls. Hepatocellular carcinoma patients have altered composition of exhaled gas due to abnormal metabolism. Thus, we can detect them by the exhaled gas. In the paper, the exhaled gas signals of hepatocellular carcinoma patients and health controls were first collected with eNose. And then the features were extracted and the multidimensional combined features were achieved. Furthermore, the PCA method was applied to optimize the features. Next, the classification model based on GRNN was constructed for training and generalization ability testing. Finally, the constructed model was adopted to predict the test and the performance was calculated. The result shows that, with the limited training set, the performance of the GRNN model is better than the BPNN model. The prediction accuracy could reach to 91.3%. Therefore, the proposed model is well suited for the classification detection with small training set and this will contribute to the study of the practical application of the eNose in the clinic.