基于BM-TransUNet的咽后壁识别分割

图像分割经历了从基于传统的阈值分割等方法逐步发展到基于卷积神经网络的方法. 传统的卷积神经网络在分割领域中表现突出, 但训练速度慢、分割精度不够高等局限性也逐渐显现. 为了克服这些局限性, 本文在TransUNet网络的基础上进行改进, 提出了基于BM-TransUNet网络的图像分割识别方法, 在TransUNet网络的在第1层之后加上深度可分离卷积模块, 并在编码器下采样的卷积层后引入注意力机制模块, 让算法更好地探索分割对象特征, 同时在编码器与解码器之间引入多尺度特征融合模块FPN. 本文基于自制的咽后壁数据集, 用于图像分割训练, 并将训练后的BM-TransUNet网络与多种传统分割网络的效果进行对比. 实验结果表明, 相比于其他传统的深度学习模型, BM-TransUNet网络的识别方法具有较高的分类准确性和泛化能力, 精确度Precision和Dice系数分别达到了93.61%和90.76%, 显示出较好的计算效率, 能有效地应用于分割任务.     

融合空洞卷积与注意力的胃癌组织切片分割

目的 病理组织切片检查是诊断胃癌的金标准,准确发现切片中的病变区域有助于及时确诊并开展后续治疗。然而,由于病理切片图像的复杂性、病变细胞与正常细胞形态差异过小等问题,传统的语义分割模型并不能达到理想的分割效果。基于此,本文提出了一种针对病理切片的语义分割方法ADEU-Net （attention-dilated-efficient U-Net++）,提高胃癌区域分割的精度,实现端到端分割。方法 ADEU-Net使用经过迁移学习的EfficientNet作为编码器部分,增强图像特征提取能力。解码器采用了简化的U-Net++短连接方式,促进深浅层特征融合的同时减少网络参数量,并重新设计了其中的卷积模块提高梯度传递能力。中心模块使用空洞卷积对编码器输出结果进行多尺度的特征提取,增强模型对不同尺寸切片的鲁棒性。编码器与解码器的跳跃连接使用了注意力模块,以抑制背景信息的特征响应。结果 在2020年“华录杯”江苏大数据开发与应用大赛（简称“SEED”大赛）数据集中与其他经典方法比较,验证了一些经典模型在该分割任务中难以拟合的问题,同时实验得出修改特征提取方式对结果有较大提升,本文方法在分割准确度上比原始U-Net提高了18.96%。在SEED数据集与2017年中国大数据人工智能创新创业大赛（brain of things,BOT）数据集中进行了消融实验,验证了本文方法中各个模块均有助于提高病理切片的分割效果。在SEED数据集中,本文方法ADEU-Net比基准模型在Dice系数、准确度、敏感度和精确度上分别提升了5.17%、2.7%、3.69%、4.08%;在BOT数据集中,本文方法的4项指标分别提升了0.47%、0.06%、4.30%、6.08%。结论 提出的ADEU-Net提升了胃癌病理切片病灶点分割的精度,同时具有良好的泛化性能。     

Seg-CapNet:A Capsule-Based Neural Network for the Segmentation of Left Ventricle from Cardiac Magnetic Resonance Imaging

Deep neural networks (DNNs) have been extensively studied in medical image segmentation.However,existing DNNs often need to train shape models for each object to be segmented,which may yield results that violate cardiac anatomical structure when segmenting cardiac magnetic resonance imaging (MRI).In this paper,we propose a capsule-based neural network,named Seg-CapNet,to model multiple regions simultaneously within a single training process.The Seg-CapNet model consists of the encoder and the decoder.The encoder transforms the input image into feature vectors that represent objects to be segmented by convolutional layers,capsule layers,and fully-connected layers.And the decoder transforms the feature vectors into segmentation masks by up-sampling.Feature maps of each down-sampling layer in the encoder are connected to the corresponding up-sampling layers,which are conducive to the backpropagation of the model.The output vectors of Seg-CapNet contain low-level image features such as grayscale and texture,as well as semantic features including the position and size of the objects,which is beneficial for improving the segmentation accuracy.The proposed model is validated on the open dataset of the Automated Cardiac Diagnosis Challenge 2017 (ACDC 2017) and the Sunnybrook Cardiac Magnetic Resonance Imaging (MRI) segmentation challenge.Experimental results show that the mean Dice coefficient of Seg-CapNet is increased by 4.7％ and the average Hausdorff distance is reduced by 22％.The proposed model also reduces the model parameters and improves the training speed while obtaining the accurate segmentation of multiple regions.     

Seg-CapNet:A Capsule-Based Neural Network for the Segmentation of Left Ventricle from Cardiac Magnetic Resonance Imaging

Deep neural networks (DNNs) have been extensively studied in medical image segmentation.However,existing DNNs often need to train shape models for each object to be segmented,which may yield results that violate cardiac anatomical structure when segmenting cardiac magnetic resonance imaging (MRI).In this paper,we propose a capsule-based neural network,named Seg-CapNet,to model multiple regions simultaneously within a single training process.The Seg-CapNet model consists of the encoder and the decoder.The encoder transforms the input image into feature vectors that represent objects to be segmented by convolutional layers,capsule layers,and fully-connected layers.And the decoder transforms the feature vectors into segmentation masks by up-sampling.Feature maps of each down-sampling layer in the encoder are connected to the corresponding up-sampling layers,which are conducive to the backpropagation of the model.The output vectors of Seg-CapNet contain low-level image features such as grayscale and texture,as well as semantic features including the position and size of the objects,which is beneficial for improving the segmentation accuracy.The proposed model is validated on the open dataset of the Automated Cardiac Diagnosis Challenge 2017 (ACDC 2017) and the Sunnybrook Cardiac Magnetic Resonance Imaging (MRI) segmentation challenge.Experimental results show that the mean Dice coefficient of Seg-CapNet is increased by 4.7％ and the average Hausdorff distance is reduced by 22％.The proposed model also reduces the model parameters and improves the training speed while obtaining the accurate segmentation of multiple regions.     

融合残差注意力机制的UNet视盘分割

目的 青光眼和病理性近视等会对人的视力造成不可逆的损害,早期的眼科疾病诊断能够大大降低发病率。由于眼底图像的复杂性,视盘分割很容易受到血管和病变等区域的影响,导致传统方法不能精确地分割出视盘。针对这一问题,提出了一种基于深度学习的视盘分割方法RA-UNet（residual attention UNet）,提高了视盘分割精度,实现了自动、端到端的分割。方法 在原始UNet基础上进行了改进。使用融合注意力机制的ResNet34作为下采样层来增强图像特征提取能力,加载预训练权重,有助于解决训练样本少导致的过拟合问题。注意力机制可以引入全局上下文信息,增强有用特征并抑制无用特征响应。修改UNet的上采样层,降低模型参数量,帮助模型训练。对网络输出的分割图进行后处理,消除错误样本。同时,使用DiceLoss损失函数替代普通的交叉熵损失函数来优化网络参数。结果 在4个数据集上分别与其他方法进行比较,在RIM-ONE（retinal image database for optic nerve evaluation）-R1数据集中,F分数和重叠率分别为0.957 4和0.918 2,比UNet分别提高了2.89%和5.17%;在RIM-ONE-R3数据集中,F分数和重叠率分别为0.969和0.939 8,比UNet分别提高了1.5%和2.78%;在Drishti-GS1数据集中,F分数和重叠率分别为0.966 2和0.934 5,比UNet分别提高了1.65%和3.04%;在iChallenge-PM病理性近视挑战赛数据集中,F分数和重叠率分别为0.942 4和0.891 1,分别比UNet提高了3.59%和6.22%。同时还在RIM-ONE-R1和Drishti-GS1中进行了消融实验,验证了改进算法中各个模块均有助于提升视盘分割效果。结论 提出的RA-UNet,提升了视盘分割精度,对有病变区域的图像也有良好的视盘分割性能,同时具有良好的泛化性能。     

融合低阶特征与全局特征的图像语义分割方法

目前,深度全卷积网络在图像语义分割领域已经取得了瞩目的成就,但特征图的细节信息在多次下采样过程中会大量损失,对分割精度造成影响。针对该问题设计了一个用于图像语义分割的深度全卷积网络。该网络采用“编码器-解码器”结构,在编码器后端引入空洞卷积以降低细节信息的损失,在解码过程中融合对应尺寸的低阶语义特征,并在解码器末端融入全局特征以提升模型的分割精度。使用数据增强后的CamVid数据集对网络进行训练和测试,测试结果达到了90.14%的平均像素精度与71.94%的平均交并比。实验结果表明,该网络能充分利用低阶特征与全局特征,有效提升分割性能,并在区域平滑方面有很好的表现。     

基于深度学习高分辨率遥感影像语义分割

高分辨率遥感影像含有丰富的地理信息. 目前基于传统神经网络的语义分割模型不能够对遥感影像中小物体进行更高维度的特征提取, 导致分割错误率较高. 本文提出一种基于编码与解码结构特征连接的方法, 对DeconvNet网络模型进行改进. 模型在编码时, 通过记录池化索引的位置并应用于上池化中, 能够保留空间结构信息; 在解码时, 利用编码与解码对应特征层连接的方式使模型有效地进行特征提取. 在模型训练时, 使用设计的预训练模型, 可以有效地扩充数据, 来解决模型的过拟合问题. 实验结果表明, 在对优化器、学习率和损失函数适当调整的基础上, 使用扩充后的数据集进行训练, 对遥感影像验证集的分割精确度达到95%左右, 相对于DeconvNet和UNet网络模型分割精确度有显著提升.     

Fuzzy Neural Network Models for Classification

In this paper, we combine neural networks with fuzzy logic techniques. We propose a fuzzy-neural network model for pattern recognition. The model consists of three layers. The first layer is an input layer. The second layer maps input features to the corresponding fuzzy membership values, and the third layer implements the inference engine. The learning process consists of two phases. During the first phase weights between the last two layers are updated using the gradient descent procedure, and during the second phase membership functions are updated or tuned. As an illustration the model is used to classify samples from a multispectral satellite image, a data set representing fruits, and Iris data set.     

Maintaining filter structure: A Gabor-based convolutional neural network for image analysis

In image segmentation and classification tasks, utilizing filters based on the target object improves performance and requires less training data. We use the Gabor filter as initialization to gain more discriminative power. Considering the mechanism of the error backpropagation procedure to learn the data, after a few updates, filters will lose their initial structure. In this paper, we modify the updating rule in Gradient Descent to maintain the properties of Gabor filters. We use the Left Ventricle (LV) segmentation task and handwritten digit classification task to evaluate our proposed method. We compare Gabor initialization with random initialization and transfer learning initialization using convolutional autoencoders and convolutional networks. We experimented with noisy data and we reduced the amount of training data to compare how different methods of initialization can deal with these matters. The results show that the pixel predictions for the segmentation task are highly correlated with the ground truth. In the classification task, in addition to Gabor and random initialization, we initialized the network using pre-trained weights obtained from a convolutional Autoencoder using two different data sets and pre-trained weights obtained from a convolutional neural network. The experiments confirm the out-performance of Gabor filters comparing to the other initialization method even when using noisy inputs and a lesser amount of training data.     

A review on weight initialization strategies for neural networks

Over the past few years, neural networks have exhibited remarkable results for various applications in machine learning and computer vision. Weight initialization is a significant step employed before training any neural network. The weights of a network are initialized and then adjusted repeatedly while training the network. This is done till the loss converges to a minimum value and an ideal weight matrix is obtained. Thus weight initialization directly drives the convergence of a network. Therefore, the selection of an appropriate weight initialization scheme becomes necessary for end-to-end training. An appropriate technique initializes the weights such that the training of the network is accelerated and the performance is improved. This paper discusses various advances in weight initialization for neural networks. The weight initialization techniques in the literature adopted for feed-forward neural network, convolutional neural network, recurrent neural network and long short term memory network have been discussed in this paper. These techniques are classified as (1) initialization techniques without pre-training, which are further classified into random initialization and data-driven initialization, (2) initialization techniques with pre-training. The different weight initialization and weight optimization techniques which select optimal weights for non-iterative training mechanism have also been discussed. We provide a close overview of different initialization schemes in these categories. This paper concludes with discussions on existing schemes and the future scope for research.

     

Error Minimized Extreme Learning Machine With Growth of Hidden Nodes and Incremental Learning

One of the open problems in neural network research is how to automatically determine network architectures for given applications. In this brief, we propose a simple and efficient approach to automatically determine the number of hidden nodes in generalized single-hidden-layer feedforward networks (SLFNs) which need not be neural alike. This approach referred to as error minimized extreme learning machine (EM-ELM) can add random hidden nodes to SLFNs one by one or group by group (with varying group size). During the growth of the networks, the output weights are updated incrementally. The convergence of this approach is proved in this brief as well. Simulation results demonstrate and verify that our new approach is much faster than other sequential/incremental/growing algorithms with good generalization performance.      

量子门Elman神经网络及其梯度扩展的量子反向传播学习算法

针对Elman神经网络的学习速度和泛化性能, 提出一种具有量子门结构的新型Elman神经网络模型及其梯度扩展反向传播(Back-propagation)学习算法, 新模型由量子比特神经元和经典神经元构成. 新网络结构采用量子映射层以确保来自上下文单元的局部反馈与隐藏层输入之间的模式一致; 通过量子比特神经元输出与相关量子门参数的修正互补关系以提高网络更新动力. 新学习算法采用搜索然后收敛的策略自适应地调整学习率参数以提高网络学习速度; 通过将上下文单元的权值扩展到隐藏层的权值矩阵, 使其在与隐藏层权值同步更新过程中获取时间序列的额外信息, 从而提高网络上下文单元输出与隐藏层输入之间的匹配程度. 以峰值检波为例的数值实验结果显示, 在量子反向传播学习过程中, 量子门Elman神经网络具有较快的学习速度和良好的泛化性能.     

复杂噪声环境下语音识别研究

语音识别是人机交互的重要方式,针对传统语音识别系统对含噪语音识别性能较差、特征选择不恰当的问题,提出一种基于迁移学习的深度自编码器循环神经网络模型。该模型由编码器、解码器以及声学模型组成,其中,声学模型由堆栈双向循环神经网络构成,用于提升识别性能;编码器和解码器均由全连接层构成,用于特征提取。将编码器结构及参数迁移至声学模型进行联合训练,在含噪Google Commands数据集上的实验表明本文模型有效增强了含噪语音的识别性能,并且具有较好的鲁棒性和泛化性。     

CoT-TransUNet:轻量化的上下文Transformer医学图像分割网络

针对以往医学图像分割网络中卷积的感受野太小以及Transformer的特征丢失问题,提出了一种端到端的轻量化上下文Transformer医学图像分割网络（lightweight context Transformer medical image segmentation network,CoT-TransUNet)。该网络由编码器、解码器以及跳跃连接三部分组成。对于输入图像,编码器使用CoTNet-Transformer的混合模块,采用CoTNet作为特征提取器来生成特征图。Transformer块则把特征图编码为输入序列。解码器通过一个级联上采样器,将编码后的特征进行上采样。该上采样器级联了多个上采样块,每个上采样块都采用CARAFE上采样算子。通过跳跃连接实现编码器与解码器在不同分辨率上的特征聚合。CoT-TransUNet通过在特征提取阶段采用全局与局部上下文信息相结合的CoTNet;在上采样阶段采用具有更大感受野的CARAFE算子。实现了生成更好的输入特征图,以及基于内容的上采样,并保持轻量化。在多器官分割任务的实验中,CoT-TransUNet取得了优于其他网络的性能。     

改进预训练编码器U-Net模型的PET肿瘤自动分割

目的 为制定放疗计划并评估放疗效果,精确的PET（positron emission tomography）肿瘤分割在临床中至关重要。由于PET图像存在低信噪比和有限的空间分辨率等特点,为此提出一种应用预训练编码器的深度卷积U-Net自动肿瘤分割方法。方法 模型的编码器部分用ImageNet上预训练的VGG19编码器代替;引入基于Jaccard距离的损失函数满足对样本重新加权的需要;引入了DropBlock取代传统的正则化方法,有效避免过拟合。结果 PET数据库共包含1 309幅图像,专业的放射科医师提供了肿瘤的掩模、肿瘤的轮廓和高斯平滑后的轮廓作为模型的金标准。实验结果表明,本文方法对PET图像中的肿瘤分割具有较高的性能。Dice系数、Hausdorff距离、Jaccard指数、灵敏度和正预测值分别为0.862、1.735、0.769、0.894和0.899。最后,给出基于分割结果的3维可视化,与金标准的3维可视化相对比,本文方法分割结果可以达到金标准的88.5%,这使得在PET图像中准确地自动识别和连续测量肿瘤体积成为可能。结论 本文提出的肿瘤分割方法有助于实现更准确、稳定、快速的肿瘤分割。     

基于倒金字塔深度学习网络的三维医学图像分割

基于深度学习的医学图像分割对医学研究和临床疾病诊断具有重要意义。然而,现有三维脑图像分割网络仅依赖单一模态信息,且最后一层网络的特征表达不准确,导致分割精度降低。引入注意力机制,提出一种基于深度学习的多模态交叉重构的倒金字塔网络MCRAIP-Net。以多模态磁共振图像作为输入,通过三个独立的编码器结构提取各模态的特征信息,并将提取的特征信息在同一分辨率级进行初步融合。利用双通道交叉重构注意力模块实现多模态特征的细化与融合。在此基础上,采用倒金字塔解码器对解码器各阶段不同分辨率的特征进行整合,完成脑组织的分割任务。在MRBrainS13和IBSR18数据集上的实验结果表明,相比3D U-Net、MMAN、SW-3D-Unet等网络,MCRAIP-Net能够充分利用多模态图像的互补信息,获取更准确丰富的细节特征且具有较优的分割精度,白质、灰质、脑脊液的Dice系数分别达到91.67%、88.95%、84.79%。     

基于改进U-Net的宫颈细胞核图像分割

原始的U-Net采用跳跃结构结合高低层的图像信息,使得U-Net模型有良好的分割效果,但是分割结果在宫颈细胞核边缘依然存在分割欠佳、过分割和欠分割等不足.由此提出了改进型U-Net网络图像分割方法.首先将稠密连接的DenseNet引入U-Net的编码器部分,以解决编码器部分相对简单,不能提取相对抽象的高层语义特征.然后对二元交叉熵损失函数中的宫颈细胞核和背景给予不同的权重,使网络更加注重细胞核特征的学习.最后在池化操作过程中,对池化域内的像素值分配合理的权值,解决池化层丢失信息的问题.实验证明,改进型U-Net网络使宫颈细胞核分割效果更好,模型也越鲁棒,过分割和欠分割比率也越少.显然,改进型U-Net是更有效的图像分割方法.     

采用资格迹的神经网络学习控制算法

强化学习是解决自适应问题的重要方法,被广泛地应用于连续状态下的学习控制,然而存在效率不高和收敛速度较慢的问题.在运用反向传播(back propagation,BP)神经网络基础上,结合资格迹方法提出一种算法,实现了强化学习过程的多步更新.解决了输出层的局部梯度向隐层节点的反向传播问题,从而实现了神经网络隐层权值的快速更新,并提供一个算法描述.提出了一种改进的残差法,在神经网络的训练过程中将各层权值进行线性优化加权,既获得了梯度下降法的学习速度又获得了残差梯度法的收敛性能,将其应用于神经网络隐层的权值更新,改善了值函数的收敛性能.通过一个倒立摆平衡系统仿真实验,对算法进行了验证和分析.结果显示,经过较短时间的学习,本方法能成功地控制倒立摆,显著提高了学习效率.     

基于数据关联感知的无监督深度融合指针网络模型

为了提高组合优化问题可行解集合的收敛性和泛化性,根据不同无监督学习策略的特点,提出一种基于数据关联感知的深度融合指针网络模型(DMAG-PN),模型通过指针网络框架将Mogrifier LSTM、多头注意力机制与图卷积神经网络三者融合.首先,编码器模块中的嵌入层对输入序列进行编码,引入多头注意力机制获取编码矩阵中的特征信息;然后构建数据关联模型探索序列节点间的关联性,采用图卷积神经网络获取其多维度关联特征信息并融合互补,旨在生成多个嵌入有效捕捉序列深层的节点特征和边缘特征;最后,基于多头注意力机制的解码器模块以节点嵌入数据和融合图嵌入数据作为输入,生成选择下一个未访问节点的全局概率分布.采用对称旅行商问题作为测试问题,与当前先进算法进行对比,实验结果表明,所提出DMAG-PN模型在泛化性和求解精确性方面获得较大的改进与提高,预训练好的DMAG-PN模型能够直接对大规模实例进行端到端的求解,避免传统算法迭代搜索的过程,具有较高的求解效率.