基于新型支持向量机的影像组学在肝脏结节分类中的应用

肝癌是最常见的恶性肿瘤之一,亚洲地区最为常见的肝癌演变过程为肝炎-肝硬化结节-异型增生结节-肝细胞性肝癌．判断肝脏结节在演变过程所处分期,并采取干预措施,对降低肝癌的发生率非常关键．本文针对影像组学提出了更精确的支持向量机（SVM）分类算法——LFOA-F-SVM,用于对120名患者的腹部动态增强磁共振图像的肝脏结节进行四分类．该算法利用了考虑半径与几何间距的F-SVM,并结合莱维飞行策略（LF）的果蝇优化算法（FOA）寻求超参．为了验证方法的有效性,本文另外添加了5个UCI分类数据集（心脏、帕金森疾病、虹膜、葡萄酒和动物园）,并与SVM、PSO-SVM、FOA-SVM、F-SVM进行比较．结果表明,在6个分类数据集（包括肝脏结节数据集和5个UCI分类数据集）中,相对于其他分类算法,LFOA-F-SVM的分类准确率最高,在肝脏结节数据集中的四分类精确率和查全率也较高．     

Region of interest and pixel-by-pixel analysis of dynamic contrast enhanced magnetic resonance imaging parameters and time-intensity curve shapes: a comparison in chondroid tumors

Dynamic contrast enhanced (DCE) MRI is widely acknowledged to be a helpful tool in the diagnosis and differentiation of tumors. In common clinical settings, the dynamic changes described by the time-intensity curves (TICs) are evaluated to find patterns of atypical tissue behavior, i.e., areas characterized by rapid contrast wash-in and wash-out. Despite the ease of this approach, there is no consensus about the specificity of the TIC shapes in discriminating tumor grades. We explore a new way of looking at TICs, where these are not averaged over a selected region of interest (ROI), but rendered pixel-by-pixel. In this way, the characteristic of the tissue is not given as a single TIC classification but as a distribution of the different TIC patterns. We applied this method in a group of patients with chondroid tumors and compared its outcome with the outcome of the standard ROI-based averaged TIC analysis. Furthermore, we focused on the problem of ROI selection in these tumors and how this affects the outcome of the TIC analysis. Finally, we investigated what relationship exists between the "standard" DCE-MRI parameter maximum enhancement (ME) and the TIC shape. CONCLUSIONS: We demonstrate that, where the ROI approach fails to show the presence of areas of rapid contrast wash-in and wash-out, the pixel-by-pixel approach reveals the coexistence of a heterogeneous pattern of TIC shapes. Secondly, we point out the differences in the DCE MRI parameters and tumor volume that can result when selecting the tumor based on DCE parameter maps or post-contrast T1-weighted images. Finally, we show that ME maps and TIC shape maps highlight different tissue areas and, therefore, the use of the ME maps is not appropriate for the correct identification of areas of atypical TICs.     

Decision Rules Construction: Algorithm Based on EAV Model

In the paper, an approach for decision rules construction is proposed. It is studied from the point of view of the supervised machine learning task, i.e., classification, and from the point of view of knowledge representation. Generated rules provide comparable classification results to the dynamic programming approach for optimization of decision rules relative to length or support. However, the proposed algorithm is based on transformation of decision table into entity–attribute–value (EAV) format. Additionally, standard deviation function for computation of averages’ values of attributes in particular decision classes was introduced. It allows to select from the whole set of attributes only these which provide the highest degree of information about the decision. Construction of decision rules is performed based on idea of partitioning of a decision table into corresponding subtables. In opposite to dynamic programming approach, not all attributes need to be taken into account but only these with the highest values of standard deviation per decision classes. Consequently, the proposed solution is more time efficient because of lower computational complexity. In the framework of experimental results, support and length of decision rules were computed and compared with the values of optimal rules. The classification error for data sets from UCI Machine Learning Repository was also obtained and compared with the ones for dynamic programming approach. Performed experiments show that constructed rules are not far from the optimal ones and classification results are comparable to these obtained in the framework of the dynamic programming extension.     

A novel video fusion framework using surfacelet transform

A novel video fusion framework based on the three-dimensional surfacelet transform (3D-ST) is proposed in this paper. Different from the traditional individual-frame based video fusion methods, the proposed framework fused multi-frame images of input videos as a whole rather than frame by frame independently with the 3D-ST. Furthermore, under the proposed framework, two ST-based video fusion algorithms are proposed. In the first algorithm, no special treatment is performed on the temporal motion information in input videos, and only a spatial-temporal region energy-based fusion rule is employed. While in the second algorithm, a modified z-score based motion detection is performed to distinguish the temporal motion information from the spatial geometry information, and then a motion-based fusion rule is present. Experimental results demonstrate that, with the motion selectivity of the 3D-ST, existing static image fusion rules can be extended to video fusion under the proposed framework. Both of the two proposed fusion algorithms significantly outperform some traditional individual-frame based and motion-based methods in spatial-temporal information extraction as well as in temporal stability and consistency. In addition, the second proposed algorithm is with high computation efficiency and can be applied to real-time video fusion.     

Spatio-temporal segmentation of moving objects using edge features in infrared videos

In this paper, a novel method is proposed for spatio-temporal segmentation of moving objects using edge features in infrared videos. We define motion saliency of edge (MSoE) to generate the MSoE-map. The seeds of moving objects are extracted from the MSoE-map by using Otsu's method and subsequently compensated by historical data. An improved layer-based region growing method is applied to the seeds to achieve spatial segmentation of moving objects. The region growing method has an adjustable growing threshold. So, one of the focuses of our work is how to determine the best growing threshold. A Markov Random Field (MRF) based criterion with maximum a posterior (MAP) estimation principle is proposed for performance evaluation of moving object segmentation without ground truth (GT) in infrared videos. This criterion can be considered as an object function of threshold determination during global searching. The global optimum is accomplished by using simulated annealing (SA) algorithm to obtain the best growing threshold. The final segmentation mask of moving objects is grown from the seeds with the best growing threshold. Experimental results are provided to illustrate that the proposed method has better performance for moving object segmentation with fewer effects of object-background misclassification in infrared videos.     

基于自适应无参经验小波变换和选择集成分类模型的运动想象

运动想象模式识别率的提高对脑机接口(BCI)技术的应用具有重要意义,本文采用自适应无参经验小波变换(APEWT)和选择集成分类模型相结合的方法提高脑电(EEG)信号的分类识别准确率.首先,通过APEWT将EEG信号分解成不同的模态;然后,使用最优模态重构后的信号计算其能量谱(ES)特征,使用最优模态分量计算其边际谱(MS)特征;最后,将不同时间段的ES特征和不同频段的MS特征输入到构建的选择集成分类模型中,从而得到其分类结果,并将该方法与其他4种组合方法进行比较.实验结果表明,本文方法具有较好分类准确率和实时性,其平均分类正确率高于其他4种方法,同时较近期使用相同数据的文献也有优势.本文为在线运动想象类BCI的应用提供了新的方法和思路.     

Predominant environmental noise classification over sound mixing based on source-specific dictionary

This paper presents a methodology to classify predominant urban acoustic sources in real mixed signals. This is based on a source-specific dictionary with atoms in the time–frequency domain using the Orthogonal Matching Pursuit (OMP) algorithm and identifying the class through a proposed selection criterion with a dynamic number of iterations involving a lower algorithm complexity. Several time–frequency atoms were evaluated considering retained energy and relative error to build a source-specific dictionary in the relevant classes. The source-specific dictionary has better results up to 7% in retained energy than to use an individual dictionary such as based on wavelet or Gabor functions, improving classification of predominant sources over sound mixing up to 9% compared to using standard dictionaries. Experimental results on classification are applied to mixture inter-class signals of two or more sources recorded by a real permanent monitoring system in an urban soundscape. The classification performance has successfully achieved identifying a predominant source in real inter-class mixtures of urban soundscapes.     

Detecting Unfavorable Driving States in Electroencephalography Based on a PCA Sample Entropy Feature and Multiple Classification Algorithms

Unfavorable driving states can cause a large number of vehicle crashes and are significant factors in leading to traffic accidents. Hence, the aim of this research is to design a robust system to detect unfavorable driving states based on sample entropy feature analysis and multiple classification algorithms. Multi-channel Electroencephalography (EEG) signals are recorded from 16 participants while performing two types of driving tasks. For the purpose of selecting optimal feature sets for classification, principal component analysis (PCA) is adopted for reducing dimensionality of feature sets. Multiple classification algorithms, namely, K nearest neighbor (KNN), decision tree (DT), support vector machine (SVM) and logistic regression (LR) are employed to improve the accuracy of unfavorable driving state detection. We use 10-fold cross-validation to assess the performance of the proposed systems. It is found that the proposed detection system, based on PCA features and the cubic SVM classification algorithm, shows robustness as it obtains the highest accuracy of 97.81%, sensitivity of 96.93%, specificity of 98.73% and precision of 98.75%. Experimental results show that the system we designed can effectively monitor unfavorable driving states.     

基于多模态MRI与深度学习的乳腺病变良恶性鉴别

为提高基于动态增强磁共振成像（DCE-MRI）的计算机辅助（CAD）方法对乳腺病变良恶性鉴别的精度,本文基于多模态特征融合,提出一种联合非对称卷积和超轻子空间注意模块的卷积神经网络AC_Ulsam_CNN.首先,采用迁移学习方法预训练模型,筛选出对乳腺病变良恶性鉴别最为有效的DCE-MRI扫描时序.而后,基于最优扫描时序图像,搭建基于AC_Ulsam_CNN网络的模型,以增强分类模型的特征表达能力和鲁棒性.最后,将影像特征与乳腺影像数据报告和数据系统（BI-RADS）分级、表观扩散系数（ADC）和时间-信号强度曲线（TIC）类型等多模态信息进行特征融合,以进一步提高模型对病灶的预测性能.采用五折交叉验证方法进行模型验证,本文方法获得了0.826的准确率（ACC）和0.877的受试者工作曲线下面积（AUC）.这表明该算法在小样本量数据下可较好区分乳腺病变的良恶性,而基于多模态数据的融合模型也进一步丰富了特征信息,从而提高病灶的检出精度,为乳腺病灶良恶性的自动鉴别诊断提供了新方法.     

Effects of k-space orders on the time-intensity curves in dynamic contrast-enhanced magnetic resonance imaging of the breast based on simulation study

PurposeThis study aims to investigate the influence of time-intensity curves (TICs) on the shapes using a dynamic contrast-enhanced magnetic resonance imaging (MRI) study depending on the Cartesian and radial orders for benign and cancerous breast tumors.MethodsBased on kinetic curve parameters, the signal intensities of six concentration gradients comprising two benign and four cancer models were used. The study aimed to construct a dynamic simulated image by creating a digital phantom image according to the following steps: (1) creating a simple numerical phantom, (2) setting the signal intensity in the contrast area, (3) creating the k-space in each time phase, (4) extracting data from k-space in each time phase, (5) filling in the k-space and adding data to the k-space assembly, and (6) creating a magnitude image. The TICs of Cartesian (centric and sequential) and radial (full-length [RFL] and half-length [RHL]) orders were created and sigmoid curve fitting was performed to compare these curves. Maximum slope (MS, s⁻¹), width of the response (WOR, s), and primary signal response (PSR) were then calculated. Phase encode steps were set for 512 and 256.ResultsMS was significantly decreased by radial order in the cancer model. No change was observed in WOR in Cartesian order, whereas RFL and RHL orders increased in the cancer models. PSR increased remarkably in the radial orders of cancer models.The difference in the fill slope in radial orders was remarkable when the TIC was steeper compared with when it was gentle, especially RHL. In WOR, both radial RFL and RHL were well matched except for the one benign model, and the shape of radial TIC was similar to sequential order as compared to centric order in 256 steps.ConclusionThe effects of Cartesian and radial orders on the patterns of TICs in a dynamic contrast-enhanced MRI study of benign and cancerous breast tumors were revealed. Interestingly, the TIC gradient of radial orders became gentler, particularly in the breast cancer MRI.     

Forest Fire Detection via Feature Entropy Guided Neural Network

Forest fire detection from videos or images is vital to forest firefighting. Most deep learning based approaches rely on converging image loss, which ignores the content from different fire scenes. In fact, complex content of images always has higher entropy. From this perspective, we propose a novel feature entropy guided neural network for forest fire detection, which is used to balance the content complexity of different training samples. Specifically, a larger weight is given to the feature of the sample with a high entropy source when calculating the classification loss. In addition, we also propose a color attention neural network, which mainly consists of several repeated multiple-blocks of color-attention modules (MCM). Each MCM module can extract the color feature information of fire adequately. The experimental results show that the performance of our proposed method outperforms the state-of-the-art methods.     

基于信息增益率的WNB水下机器人故障分类

提高故障诊断能力对于确保水下机器人系统的稳定运行具有重要意义,故障分类是目前水下机器人故障诊断所面临的一个重要问题。针对水下机器人推进器系统数据特征,提出一种基于信息增益率的加权朴素贝叶斯故障分类算法。首先,计算故障训练样本的先验概率,将各属性的信息增益率作为权值;其次,构建基于增益率加权的朴素贝叶斯分类模型;然后,对检测的故障数据利用分类模型获取具有最大后验概率的故障模式,实现故障分类。与朴素贝叶斯算法和决策树算法相比,仿真实验结果表明基于信息增益率加权的朴素贝叶斯算法的分类成功率更高,能够有效地实现水下机器人的故障分类。     

基于I-BGLAM纹理和光谱融合的高光谱显微成像木材树种分类

为了提高木材树种分类的正确率,提出了一种基于I-BGLAM纹理特征和光谱特征融合的高光谱图像的木材树种分类方法。实验数据是利用SOC710VP高光谱成像仪获取的可见光/近红外（372.53~1 038.57 nm）范围内的高光谱图像。首先,利用基于OIF的特征波段选择方法降低高光谱图像的维数,选择出含有信息量大的波段。其次,对选择出的波段图像使用NSCT及NSCT逆变换得到融合图像,对得到的融合图像使用I-BGLAM提取其纹理特征。与此同时,对高光谱图像的全波段求取平均光谱并进行S-G（Savitzky-Golay）平滑得到光谱特征。最后,将得到的纹理特征和光谱特征融合后送进极限学习机（ELM）中进行分类。此外,还和基于灰度共生矩阵（GLCM）的木材识别的传统方法以及近几年木材树种识别领域内被提出的主流方法进行了比较。该研究主要创新点有两个：一是将强纹理提取器I-BGLAM用于高光谱图像中提取其纹理特征;二是提出一种新的特征融合的模型用于高光谱图像的分类。针对8个树种的实验结果表明,单独使用I-BGLAM提取的纹理特征来进行分类的正确率最高可到达88.54%,而使用GLCM提取纹理特征的传统方法正确率最高只有76.04%,该结果可以得出本文使用I-BGLAM在纹理特征提取方面要优于GLCM,这为后面建立的融合模型打下很好的基础,单独使用平均光谱特征来分类的正确率最高可以达到92.71%,使用所提出的特征融合方法所得到的分类正确率最高可达到100%,这说明使用所提出的融合模型来分类要比以前单独使用某一种特征的分类模型要好。此外,使用所提出的方法得到的分类正确率要高于本领域内其他两种主流的识别方法。因此,所提出的基于I-BGLAM纹理特征和光谱特征融合的方法能够提高木材树种分类的正确率,该方法在木材树种分类方面有着一定的利用价值。     

Preference-Tree-Based Real-Time Recommendation System

In the current era of online information overload, recommendation systems are very useful for helping users locate content that may be of interest to them. A personalized recommendation system presents content based on information such as a user’s browsing history and the videos watched. However, information filtering-based recommendation systems are vulnerable to data sparsity and cold-start problems. Additionally, existing recommendation systems suffer from the large overhead incurred in learning regression models used for preference prediction or in selecting groups of similar users. In this study, we propose a preference-tree-based real-time recommendation system that uses various tree models to predict user preferences with a fast runtime. The proposed system predicts preferences based on two balance constants and one similarity threshold to recommend content with a high accuracy while balancing generalized and personalized preferences. The results of comparative experiments and ablation studies confirm that the proposed system can accurately recommend content to users. Specifically, we confirmed that the accuracy and novelty of the recommended content were, respectively, improved by 12.1% and 27.2% compared to existing systems. Furthermore, we verified that the proposed system satisfies real-time requirements and mitigates both cold-start and overfitting problems.     

一种新型广义RBF神经网络在混沌时间序列预测中的研究

提出了一种新颖的广义径向基函数神经网络模型,其径向基函数(RBF)的形式由生成函数确定.然后,给出了易实现的梯度学习算法,同时为了进一步提高网络的收敛速度和网络性能,又给出了基于卡尔曼滤波的动态学习算法.为了验证网络的学习性能，采用基于卡尔曼滤波算法的新型广义RBF网络预测模型对Mackey-Glass混沌时间序列和Henon映射进行了仿真.结果表明,所提出的新型广义RBF神经网络模型能快速、精确地预测混沌时间序列,是研究复杂非线性动力系统辨识和控制的一种有效方法. 关键词： 广义径向基函数神经网络 卡尔曼滤波 梯度下降学习算法 混沌时间序列 预测     

基于分子光谱图像识别的食用油快速分类研究

分子光谱分析技术结合化学计量学已成为一种非常活跃的食用油鉴别方法。然而,当不同类型的样本之间的光谱差异极其微小时,利用传统的分类技术也很难将其分开。为了完成相似品种食用油的快速识别和分类,收集了包括芝麻油、玉米油、油菜籽油、调和油、葵花油、花生油、橄榄油七种食用油的衰减全反射红外光谱,在此基础上,采用图像识别的方法对七种食用油进行快速分类。在所提出的图像识别方法中,首先,将通过多元散射校正预处理后的红外光谱吸光度矩阵进行自相关运算,利用等高线原理根据吸光度强度值的不同生成光谱图像,以扩大的光谱差异并提高光谱可视化。然后,根据图像膨胀的原理找到光谱图像的局部特征点,将其作为图像特征。最后,使用BP神经网络对特征点进行训练和分类预测。为了对比所提出的方法,PCA-BP和KL-BP的方法被用于与图像识别的方法进行比较,实验结果表明,图像识别方法的正确识别率为94.4%,高于PCA-BP的66.7%和KL-BP的83.3%。所提方法为实现食用油的快速识别和检测提供了一条新的有效途径。     

Type II quantum chaos

A classification of quantum systems into three categories, type I, II and III, is proposed. The classification is based on the degree of sensitivity upon initial conditions, and the appearance of chaos. The quantum dynamics of type I systems is quasi periodic displaying no exponential sensitivity. They arise, e.g., as the quantized versions of classical chaotic systems. Type II systems are obtained when classical and quantum degrees of freedom are coupled. Such systems arise naturally in a dynamic extension of the first step of the Born-Oppenheimer approximation, and are of particular importance to molecular and solid state physics. Type II systems can show exponential sensitivity in the quantum subsystem. Type III systems are fully quantized systems which show exponential sensitivity in the quantum dynamics. No example of a type III system is currently established. This paper presents a detailed discussion of a type II quantum chaotic system which models a coupled electronic-vibronic system. It is argued that type II systems are of importance for any field systems (not necessarily quantum) that couple to classical degrees of freedom.     

How can automotive friction-induced noises be related to physical mechanisms?

Three main physical mechanisms are found in the literature to explain the occurrence of friction-induced noises: the stick–slip, the sprag-slip and the mode-coupling instabilities. In order to improve the understanding of the automotive friction-induced noises and regarding the variety of these noises and the systems concerned, the consideration of these three physical mechanisms in a unique model, called phenomenological model, is proposed. The relationships between the mechanisms at the origin of friction-induced noises and the different kinds of friction-induced noises that can be perceived in a vehicle are particularly investigated. First, a simple classification of automotive-friction induced noises is proposed and highlights three noise categories: squeal, squeak and creak noises. Time simulations carried out on the phenomenological model show the qualitative reproduction of the vibrational behaviors at the origin of these three noise categories. Conditions are then proposed to define the three noise categories, based on the contact states ratios encountered in the time response. In order to understand the relationships between the three physical mechanisms and the three noise categories, a fullfact design of experiments is carried out with the phenomenological model. A system with realistic dynamic properties is used and submitted to a large number of conditions of use, allowing the appearance of a wide diversity of responses. The results show that the three mechanisms as well as the three noise categories can be obtained on a same dynamic system. They also show that creak is caused by a stick–slip phenomenon, squeal is mainly due to a mode-coupling phenomenon, while squeak can be caused either by mode-coupling or stick–slip phenomena. Finally, the occurrence of each mechanism and noise category is independently analyzed for the given dynamic system, giving quite significant trends towards model parameters. These trends highlight some interesting design levers to reduce the propensity of noise for an automotive structure.