Multi-Level Cross Residual Network for Lung Nodule Classification

Abstract

Computer-aided algorithm plays an important role in disease diagnosis through medical images. As one of the major cancers, lung cancer is commonly detected by computer tomography. To increase the survival rate of lung cancer patients, an early-stage diagnosis is necessary. In this paper, we propose a new structure, multi-level cross residual convolutional neural network (ML-xResNet), to classify the different types of lung nodule malignancies. ML-xResNet is constructed by three-level parallel ResNets with different convolution kernel sizes to extract multi-scale features of the inputs. Moreover, the residuals are connected not only with the current level but also with other levels in a crossover manner. To illustrate the performance of ML-xResNet, we apply the model to process ternary classification (benign, indeterminate, and malignant lung nodules) and binary classification (benign and malignant lung nodules) of lung nodules, respectively. Based on the experiment results, the proposed ML-xResNet achieves the best results of 85.88% accuracy for ternary classification and 92.19% accuracy for binary classification, without any additional handcrafted preprocessing algorithm.

Keywords: binary; computed tomography; lung nodule classification; residual convolutional neural network; ternary.

Figure 1

Samples for three classes: (a) benign nodules; (b) indeterminate nodules; and (c) malignant nodules.

Figure 2

The illustration of the multi-level cross residual block. The residuals x are not only connected to the output of their existing layers, but also added to the cross layers. The residual in the last layer is connected to the output of the first layer. The conv means convolution + BN + ReLu.

Figure 3

The diagram of the multi-level cross residual neural network. The input x is inputted into three parallel levels which have the same structure but different convolution kernel sizes. ML-xResNet contains two xRes blocks and three normal convolutional layers as well as max-pooling layers. Then, fusing the outputs of three levels by the concatenate layer, and through a global average pooling layer, the final result can be obtained by the softmax classifier.