Pulmonary Nodule Classification Based on Heterogeneous Features Learning

Pulmonary cancer is one of the most dangerous cancers with a high incidence and mortality. An early accurate diagnosis treatment pulmonary can observably increase survival rates, where computer-aided systems largely improve efficiency radiologists. In this article, we propose deep automated lung nodule system based on three-dimensional convolutional neural network (3D-CNN) support vector machine (SVM) multiple kernel learning (MKL) algorithms. The not only explores computed tomography (CT) scans, but also clinical information patients like age, smoking history history. To extract deeper image features, 34-layers 3D Residual Network (3D-ResNet) employed. Heterogeneous features including extracted data are learned MKL. experimental results prove effectiveness proposed feature extractor combination heterogeneous in task diagnosis.