Classification Of Pulmonary Nodules Using Neural Network Ensemble

A neural network ensemble (NNE) scheme was designed for distinguishing probably benign, uncertain and probably malignant lung nodules on thin-section CT images. To construct the NNE scheme, a multilayer neural network with the back-propagation algorithm (BPNN), a radial basis probabilistic neural network (RBPNN) and a learning vector quantization neural network (LVQNN) were employed, and the Bayesian criterion was used as combination rule to integrate the outputs of individual neural networks. Experimental results illustrated that the proposed classification scheme had higher classification accuracy (78.7%) as compared to the best individual classifier (LVQNN: 68.1%), as well as to another NNE scheme with the same individual neural networks but with majority voting combination rule.