Classification of Lung Nodules Based on Convolutional Deep Belief Network

A Convolutional Deep Belief Network model is proposed. Three CRBM layers were stacked to build a CDBN. We trained the three CRBMs layer by layer using unsupervised methods. And the feature vectors generated by the CDBN were used to train a softmax classifier. It is used to do the classification of lung nodule ROIs. The accuracy, precision, recall and F1 value of the best classification results reached 92.83%, 90%, 97.30% and 93.51% respectively. The features extracted from lung nodule ROIs by the CDBN deep network model proposed in this paper have been as good as or even better than ones extracted by experienced researchers, which shows the potential of deep learning in this field.