一种用于前列腺区域分割的改进水平集算法

前列腺区域的精确分割是提高计算机辅助前列腺癌诊断准确率的重要前提.本文提出了一种新的精确的前列腺区域分割模型,分为4个步骤：首先,读取T₂加权磁共振（MR）图像;其次,利用半径为5个像素的8邻域模板（8x5）的局部二值模式（LBP）特征模板计算前列腺磁共振图像的LBP特征图;然后,利用改进的距离正则化水平集（DRLSE）模型对特征图进行分割,提取前列腺粗轮廓;最后将原始水平集能量函数进行优化,构造一个新的能量函数,提取局部灰度信息和梯度信息,并在此新的能量函数的基础上,将粗轮廓迭代演化为最终的细轮廓.本文将该模型在203组来自于国际光学与光子学学会-美国医学物理学家协会-国家癌症研究所（SPIE-AAPM-NCI）前列腺MR分类挑战数据库的T2W磁共振图像上进行了测试,并与医生手工分割结果进行了比较,结果表明本文提出模型得到的分割结果的Dice系数为0.94±0.01,相对体积差（RVD）为-1.21%±2.44%,95% Hausdorff距离（HD）为6.15±0.66 mm;与文献中现有的分割模型相比,使用本文提出的模型得到的前列腺区域分割结果更接近于手工分割的结果.     

基于AI+MRI的影像诊断的样本增广与批量标注方法

训练样本是所有领域人工智能（AI）研发的关键因素.目前，基于人工智能+磁共振成像（AI+MRI）的影像诊断存在着训练样本的有效标注数量和类型无法满足研发需求的瓶颈问题.本文利用临床MRI设备对志愿者或阳性病例进行正常或重点病灶区的定量扫描，获取高分辨率各向同性的纵向弛豫时间（T₁）、横向弛豫时间（T₂）、质子密度（Pd）和表观扩散系数（ADC）等物理信息的多维数据矩阵，作为原始数据.开发虚拟MRI技术平台，对原始数据（相当于数字人体样本）进行虚拟扫描，实现不同序列不同参数下的多种类磁共振图像输出.选择感兴趣组织具有最好边界区分度的图像种类，经有经验的影像医生对其进行手动勾画并轨迹跟踪形成三维MASK标注矩阵，作为其他种类图像的图像勾画标注模板，从而实现低成本、高效率的MRI样本增广和批量标注.该平台以临床少量阳性病例作为输入，进行样本增广和标注，极大地减少AI对实际扫描样本的要求，降低了影像医生的精力和时间投入，极大地节省了成本，并输出了数量足够的磁共振图像，为基于AI+MRI的影像诊断研发提供低成本的训练数据解决方案.     

基于多输出的 3D 卷积神经网络诊断阿尔兹海默病

随着人口老龄化的加深，阿尔兹海默疾病更加大众化地出现在我们生活中，而早期精准诊断阿尔兹海默疾病并进行正向干预可有效延缓阿尔兹海默疾病的进程．基于磁共振图像的阿尔兹海默疾病的精准诊断需要综合利用多个感兴趣区域（ROIs）的信息，而单个ROI无法体现不同ROIs之间存在的联系与影响．本文首先提出三输入3D卷积神经网络（CNN），综合利用大脑3D磁共振图像中海马体、灰质（无海马体）和白质3个ROIs的信息．此外，随着神经网络的加深，原始图像的重要特征信息会部分丢失，因此我们又提出一种多输出3D CNN，通过增加中间层的连接和输出，缩短输入和输出之间的距离，增强特征传播，减少特征信息的丢失．结果显示采用多输出3DCNN模型实现整个测试集三分类的准确率为90.5%、精确率为91.0%、灵敏度为90.4%、特异性为95.2%、F1-score为90.5%，诊断性能优于单输出3D CNN模型．     

基于改进DRLSE模型的前列腺磁共振图像分割

在图像引导下的前列腺磁共振图像分割的介入诊断与治疗具有重要意义.本文对距离正则化水平集演化（DRLSE）方法进行了改进并用于前列腺磁共振图像分割.前列腺磁共振图像中靠近膀胱一侧边界较为模糊，靠近尿道一侧及左右两侧边界较为清晰，仅用传统的梯度信息指示函数无法达到理想分割结果.本研究分别采用两个指示函数控制边界清晰段及模糊段的演化，以达到准确分割的目的.此外，还在外部能量函数中增加了能量牵制项，避免演化在虚假边界停止，驱使水平集向灰度波动较大的区域移动，并能在模糊边界停止演化.实验表明利用本方法进行前列腺磁共振图像分割的效果较好；Dice相似性系数（DSC）均值达到96%，接近专家手动分割结果.     

基于 Mask RCNN 的桥小脑角区脑膜瘤与听神经瘤分类定位研究

由于人体桥小脑角区的脑膜瘤与听神经瘤在影像学的表现以及发病位置极其相似,所以临床诊断极易发生误诊．针对此问题,本文应用掩膜区域卷积神经网络（Mask RCNN）对两类肿瘤进行分类定位研究．首先采集89名脑膜瘤与218名听神经瘤患者的T₁WI-SE序列的磁共振图像,对其进行预处理,再结合改进的特征金字塔网络（FPN）算法进行网络训练．本文对比了三种不同的Mask RCNN主干网络对两者分类定位的效果．结果表明,结合改进的FPN算法和ResNet101作为主干网络的Mask RCNN分类定位模型能够有效实现对两类肿瘤的分类定位,精确率为0.918 2、召回率为0.856 9、特异性为0.876 2、均值平均精度（mAP）为0.90．     

结合小波融合和深度学习的脑胶质瘤自动分割

针对水肿区域边界模糊和瘤内结构复杂多变导致的脑胶质瘤分割不精确问题，本文提出了一种基于小波融合和3D-UNet网络的脑胶质瘤磁共振图像自动分割算法.首先，对脑胶质瘤磁共振图像的T1、T1ce、T2、Flair四种模态进行小波融合以及偏置场校正；然后，提取待分类的图像块；再利用提取的图像块训练3D-UNet网络以对图像块中的像素进行分类；最后加载损失率较小的网络模型进行分割，并采用基于连通区域的轮廓提取方法，以降低假阳性率.对57组Brats2018（Brain Tumor Segmentation 2018）磁共振图像测试集进行分割的结果显示，肿瘤的整体、核心和水肿部分的平均分割准确率（DSC）分别达到90.64%、80.74%和86.37%，这表明该算法分割脑胶质瘤准确率较高，与金标准相近.相比多模态图像融合前，该算法在减少输入网络数据量和图像冗余信息的同时，还一定程度上解决了胶质瘤边界模糊、分割不精确的问题，提高了分割的准确度和鲁棒性.     

基于Faster-RCNN和Level-Set的桥小脑角区肿瘤自动精准分割

桥小脑角区（CPA）肿瘤的精准分割在手术治疗、放疗中有重要影响,本文结合更快速区域卷积神经网络（Faster-RCNN）和水平集（Level-Set）方法对CPA肿瘤的自动分割进行了研究.首先,采集317名CPA肿瘤患者的T₁WI-SE序列磁共振图像,使用基于Faster-RCNN主干网络VGG16提取特征,结合区域建议网络（RPN）进行学习训练,建立带有CPA肿瘤位置信息的定位模型,再应用Level-Set对肿瘤进行精准分割.本文对比了不同CPA肿瘤区域勾画范围对分割结果产生的影响,并以精确率、召回率、均值平均精度值（mAP）和戴斯系数（Dice系数）等指标评估了模型定位和分割的性能.实验结果表明,结合Faster-RCNN和Level-Set建立的模型能更有效对CPA肿瘤进行精准分割,减轻临床医生的负担,并提升治疗效果.     

一种新的磁共振图像处理流水线的设计与实现

磁共振图像的重建、后处理及可视化是磁共振成像（MRI）系统的重要组成部分.本文开发了一个新的用于磁共振图像重建、后处理及可视化的开源框架YAP（Yet Another Pipeline），利用此框架可以方便地构建图像处理流水线.与现有的一些其他开源框架相比，本文开发的框架具有如下特点：（1）采用基于接口的设计，可使用基于接口的插件对流水线的功能进行扩展；（2）允许用户使用编写脚本的方式构建图像处理流水线，编辑与修改流水线都很方便；（3）支持带有分支结构的流水线，便于流水线的构建与调试.目前，该框架已经在商用系统中获得了应用.     

用于SAR估计的基于U-Net网络的快速膝关节模型重建

膝关节高场磁共振成像（MRI）时，射频功率沉积（SAR）是一个关键的安全指标.目前对于局部SAR的准确估计只能通过电磁仿真实现，这就要求得到每一个个体的膝关节模型.本文提出一种针对低场磁共振图像的基于卷积神经网络的分割方法，以实现膝关节磁共振图像的快速重建.数据集来自于矢位T₁加权自旋回波图像，将膝关节组织按照"肌肉-脂肪-骨骼"模型进行简化，除脂肪与骨骼之外的其他组织归类为肌肉.采用一种全卷积的神经网络，即U-Net进行逐层的图像分割，卷积层数为4，训练采用交叉熵函数.本文对图像的自动分割结果与手动标注结果进行了定量的比较.此外，采用3 T正交鸟笼线圈进行了SAR仿真，结果验证了组织简化对于SAR估计的可行性，并且所提方法构建的模型可以得到较为精准的局部SAR分布.     

基于虚拟线圈和卷积神经网络的多层同时激发图像重建

本文提出一种基于虚拟共轭线圈（Virtual Coil Concept，VCC）技术和k空间插值鲁棒人工神经网络（Robust Artificial-neural-networks for k-space Interpolation，RAKI）的图像重建方法，用于磁共振多层同时激发成像（Simultaneous Multi-Slice imaging，SMS），该方法能够有效提升重建图像的质量，被命名为VIRGINIA（VIRtual conjuGate coIls Neural-networks InterpolAtion）.为了得到更高质量的SMS图像，本文提出的VIRGINIA方法利用磁共振线圈数据的复数共轭对称性质扩展了SMS所获取的多通道数据，并将扩展后的数据用于RAKI网络的训练，利用训练后的网络实现高质量的SMS图像重建.本文将VIRGINIA方法和其他SMS图像重建方法（RAKI和Slice-GRAPPA方法）进行了对比，并采用结构相似指数（Structural Similarity Index，SSIM）、峰值信噪比（Peak Signal-to-Noise Ratio，PSNR）和均方根误差（Root Mean Square Error，RMSE）对不同方法的重建图像进行了量化对比分析.结果显示，在相同的SMS加速倍数下，使用VIRGINIA方法进行重建的图像质量均好于RAKI方法，且远好于传统Slice-GRAPPA方法.     

Optimization of the idler wavelength tunable cascaded optical parametric oscillator based on chirp-assisted aperiodically poled lithium niobate crystal

We present the numerical results for the optimization of the pump-to-idler conversion efficiencies of nanosecond idler wavelength tunable cascaded optical parametric oscillators(OPO) in different wavelength tuning ranges, where the primary signals from the OPO process are recycled to enhance the pump-to-idler conversion efficiencies via the simultaneous difference frequency generation(DFG) process by monolithic aperiodically poled, magnesium oxide doped lithium niobate(APMg LN) crystals. The APMg LN crystals are designed with different chirp parameters for the DFG process to broaden their thermal acceptance bandwidths to different extents. The idler wavelength tuning of the cascaded OPO is realized by changing the temperature of the designed APMg LN crystal and the cascaded oscillation is achieved in a single pump pass singly resonant linear cavity. The pump-to-idler conversion efficiencies with respect to the pump pulse duration and ratio of OPO coefficient to DFG coefficient are calculated by numerically solving the coupled wave equations. The optimal working conditions of the tunable cascaded OPOs pumped by pulses with energies of 350 μJ and 700 μJ are compared to obtain the general rules of optimization. It is concluded that the optimization becomes the interplay between the ratio of OPO coefficient to DFG coefficient and the pump pulse duration when the idler wavelength tuning range and the pump pulse energy are fixed. Besides, higher pump pulse energy is beneficial for reaching higher optimal pump-to-idler conversion efficiency as long as the APMg LN crystal is optimized according to this pump condition. To the best of our knowledge,this is the first numerical analysis of idler wavelength tunable cascaded OPOs based on chirp-assisted APMg LN crystals.     

基于级联网络的膝关节图像分割与模型构建

采用膝关节模型进行电磁仿真是计算膝关节局部射频功率沉积（SAR）的主要方法,为了构建膝关节模型,本文提出了一种包含两个卷积神经网络——U-Net的级联网络结构,用于膝关节磁共振图像的分割.第一个网络在整幅图上分割肌肉、脂肪等占比较大的组织,并从分割结果中预测软骨与半月板的大致位置信息,第二个网络基于该信息在一个更小的子图上分割小组织以提高分割精度.两个网络均采用焦点损失函数,它们的分割结果合并在一起构成膝关节模型.我们将该方法与其它4种方法的分割结果进行了定量指标的对比研究,并分别构建膝关节模型,计算局部SAR值.结果表明本文提出的级联网络结构可以更精确的构建用于SAR仿真的膝关节模型.     

Proposal and simulation for all-optical format conversion between differential phase-shift keying signals based on cascaded second-order nonlinearities

We propose and simulate simple realizations of all-optical format conversion between differential phase-shift keying (DPSK) signals based on cascaded second-order nonlinearities in a periodically poled lithium niobate (PPLN) waveguide. Four kinds of 40 Gb/s all-optical format conversion from non-return-to-zero differential phase-shift keying (NRZ-DPSK) to return-to-zero differential phase-shift keying (RZ-DPSK) are investigated based on cascaded second-harmonic generation and difference-frequency generation (cSHG/DFG) or cascaded sum- and difference-frequency generation (cSFG/DFG). The optical spectra, temporal waveforms, eye diagrams, constellation diagrams, and time-related phase distribution are analyzed, which indicate successful implementation of NRZ-DPSK-to-RZ-DPSK format conversion. The obtained results also confirm the phase preservation characteristic of PPLN.     

Deep learning and expert knowledge based underwater acoustic OFDM receiver

In this paper, a deep learning and expert knowledge based receiver is proposed for underwater acoustic (UWA) orthogonal frequency division multiplexing (OFDM). Different from the existing deep learning based UWA OFDM receivers, the proposed receiver combines deep learning with the classical expert knowledge of block-based signal processing in UWA OFDM to improve system performance and interpretability. It performs joint channel estimation and signal detection by designing skip connection (SC) convolutional neural network (CNN) cascaded attention mechanism (AM) enhanced bi-directional long short-term memory (BiLSTM) network, abbreviated as SC-CNN-AM-BiLSTM network (SCABNet). Specifically, the channel estimation subnet is designed with SC-CNN to utilize the thought of image super-resolution to reconstruct the entire channel frequency response of all subcarriers. The signal detection subnet is designed with AM-BiLSTM to extract the correlations of received sequential data for signal detection. Especially with the AM, the signal detection subnet can focus more on effective information of the received distorted signal to train the optimal network weights to improve the accuracy of data recovery. The proposed SCABNet is evaluated by experimental data, and the results have demonstrated that the SCABNet has the lowest BER and robust performance compared to the traditional linear algorithm, deep learning based black-box receiver, and ComNet receiver. And the proposed SCABNet is effective and robust when multiple nonideal factors co-exist.     

Photon Generation Scheme of 32-Fold Millimeter-Wave Signal Based on Mach-Zehnder Modulator

This paper proposes a 32-tupling frequency millimeter-wave (MMW) filter-free system based on four Mach-Zehnder Modulators (MZM) connected in parallel and cascaded with a simple radio-fiber (RoF) link structure. The four MZMs are all at the maximum transmission point (MATP), and the radio frequency (RF) driving voltage phase difference between MZMs is π /2. The center carrier is suppressed by using an optical attenuator (OATT) and an optical phase shifter (OPS). Two parallel MZMs can generate ±8th order and ±12th order optical sidebands, and the ±4th order optical sidebands can be suppressed by adjusting the modulation index m of the MZM, using cascaded two dual-parallel MZMS(DPMZM) and the phase difference of the RF signal source is π/4 to generate ±16th order optical sidebands. The theoretical analysis and simulation experiments are performed for the scheme proposed in this paper. The results show that the simulated and theoretical values of the optical sideband suppression ratio (OSSR) for ±16th order optical sideband signals are 60.02 and 59.96 dB, respectively, and the simulated and theoretical values of the RF sideband suppression ratio (RFSSR) for the 32-tupling MMW signal are 56.34 and 53.94 dB, respectively.     

基于卷积神经网络与光谱特征的夏威夷果品质鉴定研究

夏威夷果含油量高,在开缝之后容易发生变质,现有关于夏威夷果品质鉴定的方法多为传统的破坏性检验,很难满足无损检测的需求。卷积神经网络(CNN)作为应用最广泛的深度学习网络模型之一,具有比浅层学习方法更强的特征提取与模型表达能力,在光谱数据方面的应用拥有很大潜力。基于夏威夷果在可见-近红外的光谱特征分析,研究用于提取夏威夷果光谱特征的卷积神经网络模型,并提出一种高效无损鉴定夏威夷果品质的方法。首先以三种不同品质的夏威夷果(好籽、哈籽及霉籽)为研究对象,分析样本在500~2 100 nm的光谱信息;在光谱数据预处理中引入白化处理方法,用以增强数据的相关性差异;然后在模型训练过程中,将样本随机分为训练集和预测集,探讨不同CNN结构、卷积层数、卷积核大小及个数、池化层类型、全连接层神经元个数以及激活函数对分类结果的影响,并采用激活函数ReLU和Dropout方法,预防样本数据过少引起的过拟合现象;最后通过分析模型分类准确率和计算效率,确定了一个6层结构的CNN模型： 输入层-卷积层-池化层-全连接层(200神经元)-全连接层(100神经元)-输出层。实验结果表明： 上述网络模型对校正集和预测集的分类准确率均达到100％。因此,改进后的卷积神经网络模型可充分学习夏威夷果的光谱特征并有效分类,将深度学习理论与光谱分析相结合的方法能够实现对夏威夷果品质的准确鉴定,同时为夏威夷果等坚果类食品的高效、无损、实时在线检测提供了新思路。     

基于PLC的PMD补偿器中PDL效应的分析

应用穆勒（Müller）矩阵对硅基平面光波导线路（Planar lightwave circuit PLC）补偿器中PDL对差分群时延(Differential group delay DGD)分布的影响进行了分析，在不同的PDL值下对各种不同级次的PLC单元级联情况的DGD值分布进行了讨论，并与Maxwell分布进行了比较，得到了在特定PDL值情况下最为优化的PLC级次.     

SET-MOS混合结构的细胞神经网络及其应用

基于细胞神经网络（CNN）细胞单元的等效电路及其电学特性模型,利用SET-MOS混合结构反相器实现了模型中的激活函数电路,用耦合电容单元实现CNN细胞的系统模板,构建了SET-MOS CNN细胞硬件电路,并将其应用在图像处理中.仿真结果表明,所设计的CNN硬件电路具有结构简单、功耗低、响应速度快等特点,可用于构成各种规模的CNN电路,进一步满足大规模信号处理的需求及提高集成电路的集成度. 关键词： 单电子晶体管 MOS管 细胞神经网络 图像处理     

GIS Partial Discharge Pattern Recognition Based on a Novel Convolutional Neural Networks and Long Short-Term Memory

Distinguishing the types of partial discharge (PD) caused by different insulation defects in gas-insulated switchgear (GIS) is a great challenge in the power industry, and improving the recognition accuracy of the relevant models is one of the key problems. In this paper, a convolutional neural network and long short-term memory (CNN-LSTM) model is proposed, which can effectively extract and utilize the spatiotemporal characteristics of PD input signals. First, the spatial characteristics of higher-level PD signals can be obtained through the CNN network, but because CNN is a deep feedforward neural network, it does not have the ability to process time-series data. The PD voltage signal is related to the time dimension, so LSTM saves and analyzes the previous voltage signal information, realizes the modeling of the time dependence of the data, and improves the accuracy of the PD signal pattern recognition. Finally, the pattern recognition results based on CNN-LSTM are given and compared with those based on other traditional analysis methods. The results show that the pattern recognition rate of this method is the highest, with an average of 97.9%, and its overall accuracy is better than that of other traditional analysis methods. The CNN-LSTM model provides a reliable reference for GIS PD diagnosis.     

All-optical analog-to-digital converters based on cascaded 3-dB power splitters in 2D photonic crystals

In this paper, a new design of a photonic crystal (PC) all-optical analog-to-digital converter (ADC), which is based on cascaded 3-dB power splitters, is presented. It has been shown that by choosing appropriate threshold values in the output ports of the 3-dB power splitters, a set of unique states corresponding to different power levels of input analog signal can be obtained. It has been demonstrated that based on the proposed approach, the design of 2 and 3-bit PC ADC can be achievable.