Mobile health framework based on adaptive feature selection of deep convolutional neural network and QoS optimisation for benign-malignant lung nodule classification

Abstract

In order to explore the potential of Deep Learning (DL) methods in mobile health, we propose a novel framework combined enhanced DL methods and Quality of Service (QoS) optimisation for lung nodule classification. First, for classification-based DL methods, the methods of feature extraction and feature selection are widely used as the key steps in the classification of lung nodules. This paper proposes an adaptive feature selection method based on Deep Convolution Neural Network (DCNN). Based on the idea of transfer learning, we firstly use DCNN model pre-trained on ImageNet database to extract the features of multi-channel lung nodules images and then we use adaptive feature selection method extract sparse activation features. The experimental results show that the proposed method does improve the performance of benign and malignant lung nodule classification, which can achieve the classification accuracy of 89.30% and the AUC of 0.94.