Comparative Study of Deep Learning Networks for Lung Nodule Detection

Comparative Study of Deep Learning Networks for Lung Nodule Detection
Vaishnawi singh, Indu saini and Mayank kumar singh, B R Ambedkar National Institute of Technology, India
Abstract
With this rising pollution and population lung cancer may become a forthcoming pandemic. Thanks to Convolutional Neural Network (CNN) it is possible to create a reliable automated system for lung nodule detection. In past few years lots of research has been done to implement automated detection using CNN, but very few have tested deeper CNN architectures. Since there are large public datasets available, it is now possible to train and test deep-er CNN architectures. In this paper we have adopted architectures like googlenet, Xception, and resnet-101, for lung nodule detection without any pre-processing step. The depth
and trainable parameters of these networks ranges from 22 to 101 and 7 million to 44.6 million, respectively. For training and TESTING, we have used the benchmark datasets like LIDC-IDRI and LUNA-16. The maximum accuracy of 98.76% was achieved by googlenet. A detailed comparison of these networks and previously proposed CNN is also presented in this paper. We have also presented confusion matrices for comparison of the fore-mentioned deep networks. After comparing the true class and predicted class the sensitivity of googlenet was found to be 100% and specificity was 82.08%.
Keywords : Lung cancer, Convolutional Neural Network, Deep Learning, Automation, Classification.
Full Text : wireilla.com/e...
Volume URL : wireilla.com/e...
#lungcancer #convolutionalneuralnetwork #deeplearning #automation #classification