Accurately detecting and examining lung nodules early is key in diagnosing lung cancers and thus one of the best ways to prevent lung cancer deaths. Radiologists spend countless hours detecting small spherical-shaped nodules in computed tomography (CT) images. In addition, even after detecting nodule candidates, a considerable amount of effort and time is required for them to determine whether they are real nodules. The aim of this paper is to introduce a high performance nodule classification method that uses three dimensional deep convolutional neural networks (DCNNs) and an ensemble method to distinguish nodules between non-nodules. In this paper, we use a three dimensional deep convolutional neural network (3D DCNN) with shortcut connections and a 3D DCNN with dense connections for lung nodule classification. The shortcut connections and dense connections successfully alleviate the gradient vanishing problem by allowing the gradient to pass quickly and directly. Connections help deep structured networks to obtain general as well as distinctive features of lung nodules. Moreover, we increased the dimension of DCNNs from two to three to capture 3D features. Compared with shallow 3D CNNs used in previous studies, deep 3D CNNs more effectively capture the features of spherical-shaped nodules. In addition, we use an alternative ensemble method called the checkpoint ensemble method to boost performance. The performance of our nodule classification method is compared with that of the state-of-the-art methods which were used in the LUng Nodule Analysis 2016 Challenge. Our method achieves higher competition performance metric (CPM) scores than the state-of-the-art methods using deep learning. In the experimental setup ESB-ALL, the 3D DCNN with shortcut connections and the 3D DCNN with dense connections using the checkpoint ensemble method achieved the highest CPM score of 0.910. The result demonstrates that our method of using a 3D DCNN with shortcut connections, a 3D DCNN with dense connections, and the checkpoint ensemble method is effective for capturing 3D features of nodules and distinguishing nodules between non-nodules.

/pdf/classification-of-lung-nodules-in-ct-scans-using-three-2v7x8vqwvn.pdf

Classification of lung nodules in CT scans using three-dimensional deep convolutional neural networks with a checkpoint ensemble method

The double-peaked profiles in spectral data are very rare and valuable for the astronomers. Their recognitions are largely depended on visually inspect. The main reasons for auto-search of such spectra are the complex and distinct characteristics of astronomical data. In this paper, we address the problems by a double-peaked profiles search algorithm (called DoPS) based on relevant subspace (RS) and support vector machine (SVM). First, characteristics subspace is extracted by using the relevant subspace mining algorithm, in which the local density factor λ is particularly defined to measure the data sparsity. The characteristics of double-peaked profiles are represented by using the locations, the interval spaces, and strength ratio of double peaks. Second, the characteristics set is analyzed and grouped into three subsets according to the correlations among the characteristics based on the frequent patterns and rough set theory. Third, the double-peaked profiles search algorithm is proposed by using the support vectors trained from the labeled samples as thresholds. Finally, several spectral data sets from the LAMOST survey are employed to test the DoPS. The experimental results indicate that DoPS presents high performance than other similar algorithms in terms of time efficiency, noise immunity and recall, and reduced rates.

/pdf/dops-a-double-peaked-profiles-search-method-based-on-the-rs-58n70a7or7.pdf

DoPS: A Double-Peaked Profiles Search Method Based on the RS and SVM

An improved capuchin search algorithm optimized hybrid CNN-LSTM architecture for malignant lung nodule detection

Cancer is a threatening disease for the human race affecting most people around the world. The topmost reason for cancer demise across the globe is lung cancer and therefore there were many algorithms applied in predicting the survival rate of lung cancer patients. As a result of which the survival rate of lung cancer patients is increasing gradually. Support Vector Machine (SVM) technique have high accuracy than other technique in prediction. The performance of the SVM algorithm depends on the kernel function. In this paper, a comparison of three different kernel functions predicting the survival rate of a lung cancer patient with an efficient normalization technique is studied. Experiments are conducted in the dataset obtained from Cancer Imaging Archive (TCIA). Along with SVM kernel functions, five machine learning techniques were also used in predicting the survival rate of lung cancer. RBF_SVM with normalized data produced high accuracy of 97.72% compared to another algorithm. Various performance metrics such as accuracy, precision, recall, F1 score are used to evaluate the performance of the SVM kernel function.

SVM kernel Methods with Data Normalization for Lung Cancer Survivability Prediction Application

The kernel function in SVM enables linear segmentation in a feature space for a large number of linear inseparable data. The kernel function that is selected directly affects the classification performance of SVM. To improve the applicability and classification prediction effect of SVM in different areas, in this paper, we propose a weighted p-norm distance t kernel SVM classification algorithm based on improved polarization. A t-class kernel function is constructed according to the t distribution probability density function, and its theoretical proof is presented. To find a suitable mapping space, the t-class kernel function is extended to the p-norm distance kernel. The training samples are obtained by stratified sampling, and the affinity matrix is redefined. The improved local kernel polarization is established to obtain the optimal kernel weights and kernel parameters so that different kernel functions are weighted combinations. The cumulative optimal performance rate is constructed to evaluate the overall classification performance of different kernel SVM algorithms, and the significant effects of different p-norms on the classification performance of SVM are verified by 10 times fivefold cross-validation statistical comparison tests. In most cases, the results using 6 real datasets show that compared with the traditional kernel function, the proposed weighted p-norm distance t kernel can improve the classification prediction performance of SVM. 

/pdf/weighted-p-norm-distance-t-kernel-svm-classification-34h1run6.pdf

Weighted p-norm distance t kernel SVM classification algorithm based on improved polarization

Pulmonary nodule recognition is the core module of lung CAD. The Support Vector Machine (SVM) algorithm has been widely used in pulmonary nodule recognition, and the algorithm of Multiple Kernel Learning Support Vector Machine (MKL-SVM) has achieved good results therein. Based on grid search, however, the MKL-SVM algorithm needs long optimization time in course of parameter optimization; also its identification accuracy depends on the fineness of grid. In the paper, swarm intelligence is introduced and the Particle Swarm Optimization (PSO) is combined with MKL-SVM algorithm to be MKL-SVM-PSO algorithm so as to realize global optimization of parameters rapidly. In order to obtain the global optimal solution, different inertia weights such as constant inertia weight, linear inertia weight, and nonlinear inertia weight are applied to pulmonary nodules recognition. The experimental results show that the model training time of the proposed MKL-SVM-PSO algorithm is only 1/7 of the training time of the MKL-SVM grid search algorithm, achieving better recognition effect. Moreover, Euclidean norm of normalized error vector is proposed to measure the proximity between the average fitness curve and the optimal fitness curve after convergence. Through statistical analysis of the average of 20 times operation results with different inertial weights, it can be seen that the dynamic inertial weight is superior to the constant inertia weight in the MKL-SVM-PSO algorithm. In the dynamic inertial weight algorithm, the parameter optimization time of nonlinear inertia weight is shorter; the average fitness value after convergence is much closer to the optimal fitness value, which is better than the linear inertial weight. Besides, a better nonlinear inertial weight is verified.

/pdf/pulmonary-nodule-recognition-based-on-multiple-kernel-4xvsscnne1.pdf

Pulmonary Nodule Recognition Based on Multiple Kernel Learning Support Vector Machine-PSO.

The invention discloses an MKL-SVM-PSO algorithm-based lung nodule image processing method. The method comprises the steps of extracting a region of interest out of a lung nodule image, performing feature selection for the region of interest, and obtaining a data sample, wherein the data sample comprises a training set for optimizing parameters and a test set for testing models; subjecting the training set of the data sample to optimization treatment based on the MKL-SVM-PSO algorithm so as to obtain an optimal parameter set, and establishing an MKL-SVM mathematical model; applying the optimal parameter set to the MKL-SVM mathematical model for recognition and calculation so as to obtain a recognition result of a lung nodule. Based on the method, the optimal parameter set of the MKL-SVM algorithm can be quickly and accurately searched, and the optimal parameter set is applied to the recognition of the lung nodule. Meanwhile, the PSO algorithm is introduced to the MKL-SVM algorithm and is applied to the benign and malignant identification of the lung nodule.

MKL-SVM-PSO algorithm-based lung nodule image processing method

The invention relates to a brain tumor segmentation algorithm based on combination of deep learning and a traditional segmentation algorithm, in particular to a full-connection CRF cascaded FCN and Kmean value brain tumor segmentation algorithm. The DenseCRF matches all the pixels in the original image with each pixel in the segmentation result of the FCN algorithm, finds the pixels with the sameattribute, supplementarily smoothes the input, improves the detail information of the segmentation result, and improves the segmentation precision. Meanwhile, in different segmentation algorithms, the segmentation standards are different, and the segmentation results obtained based on the algorithms of the different segmentation standards are complementary to each other by fusing the deep learning algorithm FCN with the different segmentation standards and the traditional segmentation algorithm K-means clustering, so that the segmentation results are closer to real segmentation images. Therefore, the brain tumor nuclear magnetic resonance image is segmented more accurately, and a more accurate tumor image is provided for high-precision repetitive measurement and evaluation of the brain tumor nuclear magnetic resonance image.

Li Yang

Papers

Pulmonary Nodule Recognition Based on Multiple Kernel Learning Support Vector Machine-PSO.

MKL-SVM-PSO algorithm-based lung nodule image processing method

Full-connection CRF cascaded FCN and K mean value brain tumor segmentation algorithm