Pulmonary nodule is the early symptoms of lung cancer. Diagnosis classification of pulmonary nodule from CT is a challenging task. Recently, Convolution Neural Networks (CNNs) have been proposed to address this task. However, the inexplicability of deep networks make it is difficult for Radiolo- gists to accept. In this paper we present a Multi-scale and Multi-input DenseNet (MsMi-DenseNet). First, MsMi-DensNet mixes multiple window widths and levels together to gain more information from the CT images. Second, MsMi- DenseNet combines the features of two scales of nodule images. Third, in order to improve the interpretability of deep networks, the manual features of nodule are added to the network to be trained. We tested MsMi-DenseNet on the LIDC- IDRI dataset, the sensitivity, specificity, accuracy of the network and the area of ROC curve are 96.65%, 92.93%, 94.17% and 0.9820 respectively. Comparing with the existing methods, MsMi-DenseNet has made significant improvements in the diagnosis of pulmonary nodules.
CITATION STYLE
Huo, D., Wang, R., & Ding, J. (2019). Image and Graphics Technologies and Applications. Communications in Computer and Information Science (Vol. 1043, pp. 360–369). Retrieved from http://link.springer.com/10.1007/978-981-13-9917-6
Mendeley helps you to discover research relevant for your work.