Contents 1. INTRODUCTION
The lung cancer growth rate and mortality have expanded forcefully regardless of the incredible development of the imaging multimodality and careful treatment. It is the main cause of cancer-related deaths in the US and all over the world, because it is usually diagnosed at late stages [1]. If lung cancer is diagnosed early, it will help deciding the proper way for its treatment and as a result will increase the survival rate. The utilization of the Computed Tomography (CT) scans in lung cancer screening decreases the mortality by 20% [2]. However, the detection of lung cancer in CT is a very tough mission for the radiologists and time consuming because of the huge data stored in CT screening that needs to be analyzed by radiologists. In order to help radiologists, many researchers start developing their computer aided diagnostic (CADx) systems for lung nodules by utilizing machine learning and image processing techniques which will assist in cancer detection in a fast and accurate manner [3]. Shen et al. [4] developed a new interpretable deep hierarchical semantic convolutional neural network (HS-CNN), to classify nodules into benign versus malignant, that overcome the lack of interpretability related to traditional deep learning algorithms. They used the Lung Image Database Consortium (LIDC) dataset to validate their framework and compared their results to the 3D-CNN algorithm. Lakshmanaprabu et al. [5] presented an Optimal Deep Neural Network (ODNN) to classify lung nodules as benign or malignant from CT images. They used Modified Gravitational Search Algorithm (MGSA) as an optimization framework and Linear Discriminant analysis (LDA) to reduce the dimensionality of deep features extracted from their algorithm and proved that their method is competitive with other state of the art algorithms. Yuan et al. [6] used geometrical features from Fisher vector (FV) encodings found on scale-invariant feature transform (SIFT) and hybrid descriptors composed of statistical features from multi-view, multi-scale convolutional neural networks (CNNs) to classify lung nodules into benign or malignant. They experiment their algorithm on the LIDC and early lung cancer action program (ELCAP) datasets and prove that their method has promising results compared to other algorithms. Liu et al. [7] proposed multi-view multi-scale CNNs for the classification of four solid nodule types: pleural-tail, vascularized, well-circumscribed and juxta-pleural. They proved their method could also achieve high classification rates on non-nodules and ground glass optical (GGO) nodules in CT scans. Zhao et al. [8] proposed a hybrid CNN of AlexNet and LeNet by linking the parameter settings of AlexNet and the layer settings of LeNet. They used 743 nodules from the LIDC to test their model and achieved an accuracy of 82.23% when the kernel size is set to 7 × 7, the batch size is set to 32 and the learning rate is 0.005. Nishio et al. [9] used a variant of the local binary pattern as algorithm for feature extraction and support vector machine (SVM) and XGBoost with a Leave-one-out cross-validation approach for the classification of 99 lung nodules (62 cancerous and 37 benign lung nodules) from CT images. They evaluated their algorithm using AUC and got an average over 10 runs of 0.850 and 0.896 for the SVM and XGBoost respectively. Arulmurugan et al. [10] implemented a CAD system for the classification of lung nodules into benign versus malignant. They used Entropy(Wavelet attributes), Energy, Contrast, and Auto-correlation as an input to the artificial neural network classifier to get the diagnosis. Lyu et al. [11] developed a multi-level convolutional neural network (ML-CNN) for lung nodules classification. They applied their algorithm to ternary classification of lung nodules (benign, indeterminate and malignant lung nodules).
