声空化对巨噬细胞损伤效应的人工神经网络自适应模型辨识

目的针对超声空化过程中声参数与空化效应之间的影响关系问题,将多模型思想与人工神经网络相结合,构建了声空化对巨噬细胞损伤效应的辨识模型。方法(1)将巨噬细胞视为黑箱,把超声强度和辐照时间作为神经网络的输入变量,巨噬细胞存活率作为输出变量。采用人工神经网络算法训练神经网络,构建输入与输出变量之间的数值映射关系。(2)基于多模型思想选取训练样本,从而克服因训练样本选取不当而造成的模型失配。结果相比传统的神经网络模型,改进网络模型的辨识值与实验值更为接近,具有较高的辨识精度(EI=0.0137,PA=100%)。结论建立的改进人工神经网络模型实现了对超声空化效应的量化分析,有利于缩减生物学实验的成本。     

基于Hilbert-Huang变换的1 000kV输电线路雷电绕击与反击识别方法

在考虑雷电先导发展的基础上，提出了一种基于希尔伯特-黄变换(HHT)的1 000kV特高压输电线路雷电绕击与反击的识别方法。首先基于ATP-EMTP软件建立了考虑雷电先导发展过程的1 000kV输电线路雷电绕击与反击模型，研究了先导发展对输电线路雷击过电压的影响；然后对雷击输电线路后绕击和反击过电压的暂态过程进行了分析，基于信号的局部特征时间尺度，实现了线路雷电绕、反击信号的经验模式分解(EMD)，对包含大部分高频信号的前四阶固有模式函数(IMF)进行Hilbert变换，将变换后的IMF瞬时幅值作为特征量，并计算其相应的方差贡献率的大小，进而完成输电线路雷电绕击和反击的判别。仿真结果和现场实测数据验证表明，该方法能够正确有效地识别输电线路雷电绕击与反击，不受线路接地电阻和雷电流幅值的影响，且克服了目前雷电流参数实测技术中获取数据工作量大和易误判的缺点，能为输电线路的防雷分析提供理论参考。     

基于动态差值法的直流系统绝缘监测技术

针对常用直流系统绝缘监测方法的不足，提出了一种新型直流系统绝缘在线监测方法—动态差值法。即利用两次投切电阻后的电压变化量与漏电流变化量的比值来检测支路接地故障。该方法克服了绝缘监测装置的检测死区，克服了由于传感器零点漂移影响支路绝缘电阻计算的准确性造成的漏报和误报，提高了检测灵敏度和准确性。现场运行表明了此方法的准确与可靠。     

用脑电双谱分析和人工神经网络预测癫痫发作的研究

目的：根据癫痫患者脑电信号具有非高斯非线性随机特性，应用高阶累积量技术对癫痫患者的脑电信号进行双谱估计，进而结合神经网络研究发作前脑电对癫痫发作预报的价值，以寻求更加敏感和准确的发作预报参量和临床监护方法。方法：对7例癫痫患者在不同发作阶段特别是发作前夕的八导脑电信号进行双谱估计，提取各导脑电的双相关指数和加权双谱权重中心，研究了在不同发病阶段的脑电信号的高斯偏离程度，使用一个四层(24-10-2-1)的神经网络实现分类。神经网络的训练和测试采用去一循环对比法。结果：不同发作阶段时癫痫脑电信号的高斯偏离程度明显不同，其中双相关系数能够敏感区分癫痫的不同发作阶段；双相关系数和加权双谱中心作为人工神经网络输入时可提前12-24s预报癫痫的发作。结论：双谱分析、双相关系数为癫痫脑电信号的研究提供了一些新的思路，有望成为临床监护预报癫痫发作的一个指标。     

模糊神经网络预测电针镇痛过程中生化指标变化的应用研究

目的针对电针镇痛过程中生化指标检验的时滞性、费用高等不足，本文提出基于模糊神经网络预测电针镇痛过程中生化指标变化的方法。方法模糊神经网络的性能通过调整隶属函数的类型和隶属函数的个数进行优化。模糊神经网络的训练数据来源于电针镇痛患者的ECG、EEG等生理电信号以及检测的生化指标。我们从中选取45个样本为训练样本，剩余的15个样本作为测试样本。结果从模型训练结果的变化来看，训练数据预测的结果比较接近实际检测的结果。结论模糊神经网络预测的结果与实际检测的结果非常吻合，这为电针镇痛过程中生化指标变化预测提供了一种新的方法和思路。     

血清蛋白质谱与人工神经网络模型诊断卵巢癌的应用性研究

目的 建立筛选卵巢癌血清蛋白质谱与人工神经网络诊断模型的研究。方法 用H4(疏水表面)蛋白芯片结合表面增强激光解吸电离飞行时间质谱(SELDI-TOF-MS)技术检测卵巢癌患者和健康人血清样本的蛋白质谱，同时采用人工神经网络筛选差异蛋白以建立诊断模型。结果 用SELDI-TOF-MS技术和H4蛋白芯片从47例卵巢癌和29名健康人血清中，筛选出4个有明显表达差异的蛋白，其质荷比(m／z)分别为5881、7553、6652和9391。用其中的18名健康人和29例卵巢癌患者样本作训练集和交叉验证后，再用筛选出的4个差异蛋白质建立人工神经网络预测模型。然后，对11名健康人和18例卵巢癌患者样本进行盲法测试，以验证该模型。结果显示，我们建立的诊断模型对卵巢癌检测的敏感性为100％，特异性为90．9％，阳性率为94．7％。结论 血清蛋白质谱与人工神经网络模型对小样本的卵巢癌诊断具有较高的敏感性和特异性，可扩大样本进行深入的应用性研究。     

基于改进的广义谐波小波包分解和混沌振子的小电流接地系统故障选线

小电流系统发生故障时,各线路零序电流具有非平稳、非线性等复杂特性。本文提出一种基于改进的广义谐波小波包分解算法和混沌振子检测的小电流接地系统故障选线方法。广义谐波小波包分解算法经过扩频,能进行任意频段及任意频宽的信号特征提取,且计算量小,实现信号的快速无混叠分离。Duffing振子系统不受噪声影响且对于与内驱动力同频的外界信号具有高度敏感性,对Duffing振子系统进行时标变换,可以适应不同频率信号的检测要求,通过观察相图的变化可以准确选出故障线路。数值仿真验证了该算法的准确性及可靠性。     

人工神经网络在脑卒中早期快速分类诊断中的应用

目的:通过人工神经网络对出血与缺血两类不同性质的脑卒中(acute cerebral vascular disease,ACVD)做出快速的早期分类诊断,以达到提高早期诊断率,从而减少脑卒中的病残率。方法:编制调查表,通过调查员的快速调查获得患者基本信息及血常规信息。结合评分法的指标及瞳孔改变和神经系统体征(巴彬斯基征),把这些信息经过数据处理作为神经网络的输入;出血性、缺血性脑卒中患者两个结局变量作为神经网络的输出。把样本随机分成训练样本、检验样本和测试样本,利用软件statistica neural network(STNN)建立误差反向传播网络预测模型,然后训练网络,进行结果预测。同时对样本进行Siriraj评分,其结果与神经网络模型的结果进行比较。结果:Siriraj评分法对脑出血诊断的准确率为91.6%;神经网络为97.6%,χ2=10.7,P=0.001。Siriraj评分法对脑出血诊断的误诊率为4.7%;神经网络为2.4%,χ2=2.4,P=0.12。Siriraj评分法对脑出血诊断的漏诊率为3.7%。Siriraj评分法对脑梗死诊断的准确率为92.3%;神经网络为97.7%,χ2=14.8,P...     

超声影像下基于ResNet34-Unet的静脉分割方法研究

静脉超声图像存在噪点多、阈值分割效果不佳的问题，对此本文提出一种基于ResNet34主干网络的ResNet34-UNet分割网络模型，利用ResNet34网络残差学习的结构特点，在保证网络能够提取充足图像特征的前提下，有效避免梯度消失和网络退化问题，且34层的网络深度维持了较小的网络规模；利用U-Net结构特有的长连接（Skip Connection）模块，将静脉超声图像的深层特征与浅层特征有效融合，使静脉的识别精度得以较大幅度的提升，实现了静脉边缘的平滑分割。将300张静脉超声图像作为训练集，200张作为测试集，通过随机旋转、翻转、投影等操作进行数据集的增强，经过十轮迭代训练后得到模型的准确度（ACC）达96.3%，较全卷积神经网络（FCN）高5.9%，较DeepLab v3+高5.2%。结果表明基于ResNet34-UNet的静脉分割方法能够准确地分割静脉超声图像，为后续超声影像下静脉的自动识别与跟踪提供了技术参考。     

基于V-Net卷积神经网络深度学习模型自动分割腰椎CT图像中的椎旁肌

目的 观察基于V-Net卷积神经网络(CNN)的深度学习(DL)模型自动分割腰椎CT图像中的椎旁肌的价值。方法 收集471例接受腰椎CT检查患者，按7∶3比例将其分为训练集(n=330)和测试集(n=141);采用2D V-Net进行训练，建立DL模型；观察其分割腰大肌、腰方肌、椎后肌群及椎旁肌的价值。结果 基于V-Net CNN的DL模型分割椎旁肌精度良好，戴斯相似系数(DSC)均较高、肌肉横截面积误差率(CSA error)均较低；其分割训练集图像中的腰大肌、腰方肌及椎旁肌的DSC均高于测试集(P均<0.05),而分割训练集中4组肌肉的CSA error均低于测试集(P均<0.05)。测试集内两两比较结果显示，该模型分割椎后肌群的DSC最高、腰方肌的DSC最低；分割腰方肌的CSA error最高、椎旁肌的CSA error最低(P均<0.05)。结论 以基于V-Net的DL模型自动分割椎旁肌的效能较佳。     

医用超声系统增益补偿电路的设计

背景:为使深度不同但性质相同的界面反射的回波信号能显示出相同的幅值,必须根据深度(即时间)逐步增大放大器的增益,而使超声波在传播衰减过程中所引起远距离反射波弱的情况得到相应补偿.目的:为了补偿医学超声系统中回波信号的传输衰减问题,提出了一种时间增益补偿电路设计方案,阐述电路设计依据和原理.方法:针对医用超声系统的特点选择高信噪比、高带宽的可变增益放大器件VCA610,实现超声的增益补偿电路.结果与结论:该设计方案有效地解决了医用超声软组织测量过程中由声程导致的回波信号的非线性补偿问题.与传统的分立元件电路相比,该方案具有电路简单,TGC控制信号稳定可靠以及调节灵活等优点,能准确地补偿超声波在人体内的衰减,从而为医学测量系统的设计提供了一个新的可靠方法.     

Propagation velocity and attenuation of a shear wave pulse measured by ultrasound detection in agarose and polyacrylamide gels

The aim of our research was to measure and analyze phase velocity and pulse attenuation of a shear wave in two media: well-known agarose-gelatin gel and seldom-used polyacrylamide gel. These quantities were determined at three temperatures by the method of transmission sonoelastography described by Catheline et al. (1999). The shear wave was generated with a shaker stimulated by an electric pulse, with a length of one sinusoidal period with a preset frequency. The calculation method is based on a cross-correlation algorithm used for consecutive A-scans of signals of backwards scattered ultrasonic pulses. It allows determination of the local displacement of scattering elements in the medium, caused by a propagating shear wave, and determination of viscoelastic properties of gels. The results of the measurements of shear wave phase velocity and attenuation, obtained for agarose-gelatin and polyacrylamide gels that simulate biological systems depending on frequency and amplitude of vibrations, are presented. The comparison of the measured characteristic properties of gels has revealed that polyacrylamide gel is more useful in viscoelastic investigations of tissue-like phantoms.     

Spectral Shift Originating from Non-linear Ultrasonic Wave Propagation and Its Effect on Imaging Resolution

An amplitude-dependent downshift in the fundamental wave spectrum of a propagating ultrasonic pulse caused by non-linear wave propagation is described. The effects of non-linearity and the associated downshift on spatial resolution are also studied. The amounts of downshift and spatial resolution are extracted from the numerically simulated beam profile based on the KZK equation. Results for a 25-MHz transducer reveal that non-linear effects can lead to 58% additional downshift in the centre frequency of a pulse compared with a linear case with downshift caused only by attenuation. This additional downshift causes about 50% degradation in axial resolution. However, as the beam becomes narrower from the non-linear effects, the overall effect of non-linearity still leads to improved lateral resolution (≤26%). Therefore, as non-linearity increases with wave pressure, it is concluded that the increase in source pressure improves lateral resolution and degrades axial resolution.     

Fault diagnosis in a current sensor and its application to fault-tolerant control for an air supply subsystem of a 50 kW-Grade fuel cell engine

The safety, reliability and stability of air supply subsystems are still problems for the commercial applications of fuel cells; therefore, engine fault diagnosis and fault-tolerant control are essential to protect the fuel cell stack. In this study, a fault diagnosis and fault-tolerant control method based on artificial neural networks (ANNs) has been proposed. The offline ANN modification model was trained with a Levenberg–Marquardt (LM) algorithm based on other sensors'' signals relevant to the current sensor of a 50 kW-grade fuel cell engine test bench. The output current was predicted via the ANN identification model according to other relevant sensors and compared with the sampled current sensor signal. The faults in the current sensor were detected immediately once the difference exceeded the given threshold value, and the invalid signals of the current sensor were substituted with the predictive output value of the ANN identification model. Finally, the reconstructed current sensor signals were sent back to a fuel cell controller unit (FCU) to adjust the air flow and rotate speeds of the air compressor. Experimental results show that the typical faults in the current sensor can be diagnosed and distinguished within 0.5 s when the threshold value is 15 A. The invalid signal of current sensor can be reconstructed within 0.1 s. Which ensures that the air compressor operate normally and avoids oxygen starvation. The proposed method can protect the fuel cell stack and enhance the fault-tolerant performance of air supply subsystem used in the fuel cell engine, and it is promising to be utilized in the fault diagnosis and fault-tolerant control of various fuel cell engines and multiple sensor systems.

A fault diagnosis and fault-tolerant control method based on artificial neural networks is proposed. Faults were detected immediately once the difference exceeded the set threshold, and the invalid signals were substituted with the predictive output value.     

Application of artificial neural network to fMRI regression analysis

We used an artificial neural network (ANN) to detect correlations between event sequences and fMRI (functional magnetic resonance imaging) signals. The layered feed-forward neural network, given a series of events as inputs and the fMRI signal as a supervised signal, performed a non-linear regression analysis. This type of ANN is capable of approximating any continuous function, and thus this analysis method can detect any fMRI signals that correlated with corresponding events. Because of the flexible nature of ANNs, fitting to autocorrelation noise is a problem in fMRI analyses. We avoided this problem by using cross-validation and an early stopping procedure. The results showed that the ANN could detect various responses with different time courses. The simulation analysis also indicated an additional advantage of ANN over non-parametric methods in detecting parametrically modulated responses, i.e., it can detect various types of parametric modulations without a priori assumptions. The ANN regression analysis is therefore beneficial for exploratory fMRI analyses in detecting continuous changes in responses modulated by changes in input values.     

Frequency dependence of ultrasonic attenuation in bovine cortical bone: an in vitro study

Recent progress in quantitative ultrasonic (QUS) techniques enables the in vivo evaluation of cortical bone, which is determinant in bone fragility. However, the interaction between ultrasound and cortical bone remains poorly understood. Most ultrasonic studies have been confined to longitudinal wave speed analysis and the frequency dependence of ultrasonic wave attenuation in this complex multiscale structure has not been extensively investigated. Our objective was to evaluate in vitro the frequency dependence of attenuation in bovine femoral cortical bone samples obtained from three specimens at different anatomical locations along the diaphysis axis and around the circumference. The frequency-dependent attenuation coefficient was evaluated after correction of transmission effects using a transmission device operating at 10 MHz. Attenuation exhibits a non linear variation versus frequency. However, the quasi-linearity of attenuation on a 1 MHz restricted bandwidth around 4 MHz enables broadband ultrasonic attenuation (BUA) evaluation. Our study demonstrates the feasibility of BUA measurements in the three directions (axial, radial and tangential) with reasonable precision (standardized coefficient of variation: 10% to 12%). Significant differences in BUA are obtained according to the anatomical location. BUA values are higher in the distal and proximal parts of the bone than in the midshaft and in the posterior and lateral parts than in the medial and anterior parts. Findings are consistent with results previously obtained and may be explained primarily by scattering phenomena but also by bone viscoelasticity. (E-mail: haiat@univ-paris12.fr)     

Assessment of lipid peroxidation and artificial neural network models in early Alzheimer Disease diagnosis

Objective: Lipid peroxidation constitutes a molecular mechanism involved in early Alzheimer Disease (AD) stages, and artificial neural network (ANN) analysis is a promising non-linear regression model, characterized by its high flexibility and utility in clinical diagnosis. ANN simulates neuron learning procedures and it could provide good diagnostic performances in this complex and heterogeneous disease compared with linear regression analysis. Design and Methods: In our study, a new set of lipid peroxidation compounds were determined in urine and plasma samples from patients diagnosed with early Alzheimer Disease (n = 70) and healthy controls (n = 26) by means of ultra-performance liquid chromatography coupled with tandem mass-spectrometry. Then, a model based on ANN was developed to classify groups of participants. Results: The diagnostic performances obtained using an ANN model for each biological matrix were compared with the corresponding linear regression model based on partial least squares (PLS), and with the non-linear (radial and polynomial) support vector machine (SVM) models. Better accuracy, in terms of receiver operating characteristic-area under curve (ROC-AUC), was obtained for the ANN models (ROC-AUC 0.882 in plasma and 0.839 in urine) than for PLS and SVM models. Conclusion: Lipid peroxidation and ANN constitute a useful approach to establish a reliable diagnosis when the prognosis is complex, multidimensional and non-linear.     

In vivo potentiostatic studies at the electrode tissue interface: filter properties of the monophasic action potential (Ag/AgCl) electrode in living rat heart

The monophasic action potential (Franz) catheter is regarded as the criterion standard for high fidelity recording of a class of physiological signals. However, its signal modulation characteristics have never been reported. Broadband impedance spectroscopy was performed in perfused living rat heart in a three-electrode potentiostatic configuration to determine the filtering characteristics of the MAP and model Ag/AgCl electrode-tissue interfaces. The filter transfer function H(f) (attenuation [dB] vs log(f) [log(Hz)]) was derived for the frequency range 10 Hz-10(6) Hz. As a filter, the MAP interface is characterized by two ranges of filtering behavior. At high frequency the MAP interface is a high-pass filter with passband frequency 54 kHz-549 kHz (median 321 kHz) and with -3 dB cutoff points ranging from 10 kHz to 302 kHz. In this high frequency range the transfer function is characterized by decreasing attenuation per decade. However, in the lower frequency range relevant to physiological signals (the monophasic action potential, 0.1-40 Hz), it is a severely attenuating nondiodic high-pass filter element with an average attenuation of 16.87 dB relative to passband. In this physiological range, rolloff is nonlinear with increasing attenuation per decade. While the MAP electrode and model Ag/AgCl electrodes are high-pass filters with robust transfer functions for high frequency signals in the living heart, the attenuation of signals in a frequency range relevant to in vivo physiological recording imparts extreme attenuation that may distort physiological signals unpredictably. This disadvantage may be mitigated by amplitude scaling to a calibrated pure tone signal within the physiological frequency band to recover a reproducible signal.     

Contrast-Enhanced Transcranial Two-Dimensional Ultrasound Imaging Using Shear-Mode Conversion at Low Frequency

The distortion and attenuation of transcranial ultrasound (US) signals are significant problems in US imaging of the brain. Of the variety of proposed solutions, shear-mode transmission through the skull is one of the more recent options and has been shown to reduce distortion of the US beam. This study examined the effects of transcranial shear-mode transmission on the images of a contrast-agent–filled polytetrafluoroethylene tube produced by a 32-element 750 kHz linear phased array transducer through an ex vivo human skull section. Although the tube was successfully imaged using shear-mode transmission with subharmonic imaging in 6 of 9 cases, the tube was visible in only 1 of 9 cases for both the fundamental and the second harmonic frequencies. Some improvement in the location of the axial image was seen at the fundamental frequency using shear mode. No improvement was seen at the other two frequencies, but this may be due to low transducer sensitivity. As well, neither the presence of the skull nor the incident angle changed the distance at which signals from the two tubes could be resolved. With this transducer, these distances were found to be 5 mm laterally and 3 mm axially for the fundamental and second harmonic images, and 10 mm and 5 mm for the subharmonic images. The results show that the subharmonic signal was the most successful of the three examined in penetrating a thick skull but that the success comes at the cost of image resolution.     

Attenuation estimations using envelope echo data: analysis and simulations

Previously we described a video signal analysis (VSA) method for measuring backscatter and attenuation from B-Mode image data. VSA computes depth-dependent ratios of the mean echo intensity from a sample to the mean echo intensity from a reference phantom imaged using identical scanner settings. The slope of a line-fit of this ratio (expressed in dB) versus depth is related to the attenuation of the sample. This paper investigates conditions for which the echo intensity ratio versus depth is independent of transducer pulsing characteristics and instrument settings, and depends only on the properties of the sample and the reference. A theoretical model is described for the echo signal power versus depth from a uniform medium containing scatterers. The model incorporates bandwidth, frequency and media attenuation. Results show that the sample-to-reference echo intensity ratio versus depth is a curve, the departure of which from a straight line is a function of the relative attenuation of the two media, the imaging system bandwidth and the initial frequency. The model also leads to a depth-dependent "effective frequency" determination in the VSA method. Model predictions are verified using RF signals computed by an acoustic pulse-echo simulation program.