Classification and Uncertainty Visualization of Dendritic Spines from Optical Microscopy Imaging

Neuronal dendrites and their spines affect the connectivity of neural networks, and play a significant role in many neurological conditions. Neuronal function is observed to be closely correlated with the appearance, disappearance and morphology of the spines. Automatic 3‐D reconstruction of neurons from light microscopy images, followed by the identification, classification and visualization of dendritic spines is therefore essential for studying neuronal physiology and biophysical properties. In this paper, we present a method to reconstruct dendrites using a surface representation of the dendrite. The 1‐D skeleton of the dendritic surface is then extracted by a medial geodesic function that is robust and topologically correct. This is followed by a Bayesian identification and classification of the spines. The dendrite and spines are visualized in a manner that displays the spines' types and the inherent uncertainty in identification and classification. We also describe a user study conducted to validate the accuracy of the classification and the efficacy of the visualization.     

Automated algorithms for multiscale morphometry of neuronal dendrites

We describe the synthesis of automated neuron branching morphology and spine detection algorithms to provide multiscale three-dimensional morphological analysis of neurons. The resulting software is applied to the analysis of a high-resolution (0.098 microm x 0.098 microm x 0.081 microm) image of an entire pyramidal neuron from layer III of the superior temporal cortex in rhesus macaque monkey. The approach provides a highly automated, complete morphological analysis of the entire neuron; each dendritic branch segment is characterized by several parameters, including branch order, length, and radius as a function of distance along the branch, as well as by the locations, lengths, shape classification (e.g., mushroom, stubby, thin), and density distribution of spines on the branch. Results for this automated analysis are compared to published results obtained by other computer-assisted manual means.     

基于骨架树的荧光共焦图像神经树突棘自动分割与检测方法*

提出一个荧光共焦图像中神经树突棘自动分割与检测方法。该方法采用新的自适应区域生长法对神经树突棘目标进行预分割,基于种子点的路径规划算法,以计算给定点到目标点的最短路径来获取初始主骨架;通过建立最小生成树描述模型对骨架进行修剪,利用种子点间的矢量角度变化及顶点距离值对突棘进行检测提取。实验结果表明,该方法能很好地提取树突骨架,并取得了较好的突棘检测效果。     

Investigation of different neural models for blood cell type identification

The analysis of blood cells in microscope images can provide useful information concerning the health of patients; however, manual classification of blood cells is time-consuming and susceptible to error due to the different morphological features of the cells. Therefore, a fast and automated method for identifying the different blood cells is required. In this paper, we investigate the use of different neural network models for the purpose of cell identification. The neural models are based on the back propagation learning algorithm and differ in design according to the way data features are extracted from the cell microscopic images. Three different topologies of neural networks are investigated, and a comparison between these models is drawn. Experimental results suggest that the proposed method performs well in identifying blood cell types regardless of their irregular shapes, sizes, and orientation.     

卷积神经网络金相组织自动识别

为了降低人工分辨金相组织图像类别的误差率,提高分辨效率,采用卷积神经网络模型对金相组织图像进行自动辨识。对制备金相样块所得铁素体与马氏体两种金相组织图像进行分析,提出符合金相组织图像分布特征的预处理方案。通过采用图像尺寸归一化、灰度值归一化以及高斯平滑处理等方法,对原始金相组织图像进行预处理,建立金相组织图像数据集。针对建立的铁素体和马氏体金相组织图像数据集,提出了适合金相组织图像辨识的改进模型,分别记为LeNet-MetStr模型、AlexNet-MetStr模型和VGGNet-MetStr模型。对3种改进卷积神经网络进行模型训练及分析,结果表明VGGNet-MetStr模型对2种金相组织图像自动辨识具有更高的准确度。     

A neurocomputing model for SODAR structure classification

The manual identification of different types of atmospheric microstructures recorded by SODAR (SOund Detection And Ranging) is a tedious task and can be performed only by an expert with broad experience. To avoid this manual task, a neural network based method of SODAR structure classification system is proposed. This method is developed based on past observations of various meteorological parameters such as temperature, relative humidity and vapour pressure, along with different features computed from the SODAR structure data, which are images representing the dynamics of the planetary boundary layer (PBL). The patterns of these images indicate the structure of different thermal patterns of the atmosphere. We propose a neural network model whose architecture combines multilayer perceptron networks (MLPs) to realize better performance after capturing the seasonality and other related effects in the atmospheric data. We also demonstrate that the use of appropriate features can further improve performance of the prediction system. These observations inspired us to use a feature selection neural network which can select good features online while learning the prediction task. The feature selection neural network is used as a preprocessor to select good features. The combined use of feature selection neural network and MLP, i.e. FSMLP (feature selection multilayer perception) results in a neural network system that uses only very few inputs but can produce a good classifier. Here we develop a real-time system that classifies the SODAR patterns automatically.     

Extending cable theory to heterogeneous dendrites

Dendrites may exhibit many types of electrical and morphological heterogeneities at the scale of a few micrometers. Models of neurons, even so-called detailed models, rarely consider such heterogeneities. Small-scale fluctuations in the membrane conductances and the diameter of dendrites are generally disregarded and spines merely incorporated into the dendritic shaft. Using the two-scales method known as homogenization, we establish explicit expressions for the small-scale fluctuations of the membrane voltage, and we derive the cable equation satisfied by the voltage when these fluctuations are averaged out. This allows us to rigorously establish under what conditions a heterogeneous dendrite can be approximated by a homogeneous cable. We consider different distributions of synapses, orderly or random, on a passive dendrite, and we investigate when replacing excitatory and inhibitory synaptic conductances by their local averages leads to a small error in the voltage. This indicates in which regimes the approximations made in compartmental models are justified. We extend these results to active membranes endowed with voltage-dependent conductances or NMDA receptors. Then we examine under which conditions a spiny dendrite behaves as a smooth dendrite. We discover a new regime where this holds true, namely, when the conductance of the spine neck is small compared to the conductance of the synapses impinging on the spine head. Spines can then be taken into account by an effective excitatory current, the capacitance of the dendrite remaining unchanged. In this regime, the synaptic current transmitted from a spine to the dendritic shaft is strongly attenuated by the weak coupling conductance, but the total current they deliver can be quite substantial. These results suggest that pedunculated spines and stubby spines might play complementary roles in synaptic integration. Finally, we analyze how varicosities affect voltage diffusion in dendrites and discuss their impact on the spatiotemporal integration of synaptic input.     

Analogue pattern matching in a dendritic spine model based on phosphorylation of potassium channels

Modification of potassium channels by protein phosphorylation has been shown to play a role in learning and memory. If such memory storage machinery were part of dendritic spines, then a set of spines could act as an 'analogue pattern matching' device by learning a repeatedly presented pattern of synaptic activation. In this study, the plausibility of such analogue pattern matching is investigated in a detailed circuit model of a set of spines attached to a dendritic branch. Each spine head contains an AMPA synaptic channel in parallel with a calcium-dependent potassium channel whose sensitivity depends on its phosphorylation state. Repeated presentation of synaptic activity results in calcium activation of protein kinases and subsequent channel phosphorylation. Simulations demonstrate that signal strength is greatest when the synaptic input pattern is equal to the previously learned pattern, and smaller when components of the synaptic input pattern are either smaller or larger than corresponding components of the previously learned pattern. Therefore, our results indicate that dendritic spines may act as an analogue pattern matching device, and suggest that modulation of potassium channels by protein kinases may mediate neuronal pattern recognition.     

Medical image classification based on artificial intelligence approaches: A practical study on normal and abnormal confocal corneal images

Corneal images can be acquired using confocal microscopes which provide detailed views of the different layers inside a human cornea. Some corneal problems and diseases can occur in one or more of the main corneal layers: the epithelium, stroma and endothelium. Consequently, for automatically extracting clinical information associated with corneal diseases, identifying abnormality or evaluating the normal cornea, it is important to be able to automatically recognise these layers reliably. Artificial intelligence (AI) approaches can provide improved accuracy over the conventional processing techniques and save a useful amount of time over the manual analysis time required by clinical experts. Artificial neural networks (ANNs), adaptive neuro fuzzy inference systems (ANFIS) and a committee machine (CM) have been investigated and tested to improve the recognition accuracy of the main corneal layers and identify abnormality in these layers. The performance of the CM, formed from ANN and ANFIS, achieves an accuracy of 100% for some classes in the processed data sets. Three normal corneal data sets and seven abnormal corneal images associated with diseases in the main corneal layers have been investigated with the proposed system. Statistical analysis for these data sets is performed to track any change in the processed images. This system is able to pre-process (quality enhancement, noise removal), classify corneal images, identify abnormalities in the analysed data sets and visualise corneal stroma images as well as each individual keratocyte cell in a 3D volume for further clinical analysis.     

基于多尺度特征融合与注意力机制的脊柱CT图像分割

本文针对医学脊柱CT图像因骨密度不均匀、骨骼结构复杂或图像成像分辨率低等因素造成的分割精度较低的问题,提出一种基于卷积-反卷积神经网络的CT图像脊柱分割方法.通过引入多尺度残差模块及注意力机制改进U-Net网络,训练特征模型并进行测试.在真实数据集上的实验结果表明,该方法能有效提高CT图像中脊柱的分割精度及分割效率,Dice系数评估值为0.97,IOU系数评估值为0.94.     

Recursive Bayesian Recurrent Neural Networks for Time-Series Modeling

This paper develops a probabilistic approach to recursive second-order training of recurrent neural networks (RNNs) for improved time-series modeling. A general recursive Bayesian Levenberg–Marquardt algorithm is derived to sequentially update the weights and the covariance (Hessian) matrix. The main strengths of the approach are a principled handling of the regularization hyperparameters that leads to better generalization, and stable numerical performance. The framework involves the adaptation of a noise hyperparameter and local weight prior hyperparameters, which represent the noise in the data and the uncertainties in the model parameters. Experimental investigations using artificial and real-world data sets show that RNNs equipped with the proposed approach outperform standard real-time recurrent learning and extended Kalman training algorithms for recurrent networks, as well as other contemporary nonlinear neural models, on time-series modeling.      

Learning regularity in an economic time-series for structure prediction

Although an economic time-series has an apparently random fluctuation over time, there exists certain regularity in the functional behavior of the series. This paper attempts to identify the regularly occurring structures in an economic time-series with an aim to represent the series as a specific sequence of such structures for forecasting applications. The applications include prediction of the most probable structure with its expected duration, along with predicted values lying thereon. Representation of a time-series by a set of regularly recurring structures is undertaken by invoking three main steps: (i) non-uniform length segmentation of the series, (ii) identification of the recurrent patterns by clustering of the generated segments, and (iii) representing the sequence of regular structures using a specially designed automaton. The automaton is used here to both encode the sequence of structures representing the time-series and also to act as an inference engine for stochastic forecasting about the time-series. Experiments undertaken on large (28 years’) daily economic time-series data sets confirm the success in automated structure prediction with an average prediction accuracy of 88.05%, average precision of 91.24% and average recall of 93.42%.     

Retinal vessel segmentation employing ANN technique by Gabor and moment invariants-based features

Diabetic retinopathy (DR) is the major ophthalmic pathological cause for loss of eye sight due to changes in blood vessel structure. The retinal blood vessel morphology helps to identify the successive stages of a number of sight threatening diseases and thereby paves a way to classify its severity. This paper presents an automated retinal vessel segmentation technique using neural network, which can be used in computer analysis of retinal images, e.g., in automated screening for diabetic retinopathy. Furthermore, the algorithm proposed in this paper can be used for the analysis of vascular structures of the human retina. Changes in retinal vasculature are one of the main symptoms of diseases like hypertension and diabetes mellitus. Since the size of typical retinal vessel is only a few pixels wide, it is critical to obtain precise measurements of vascular width using automated retinal image analysis. This method segments each image pixel as vessel or nonvessel, which in turn, used for automatic recognition of the vasculature in retinal images. Retinal blood vessels are identified by means of a multilayer perceptron neural network, for which the inputs are derived from the Gabor and moment invariants-based features. Back propagation algorithm, which provides an efficient technique to change the weights in a feed forward network is utilized in our method. The performance of our technique is evaluated and tested on publicly available DRIVE database and we have obtained illustrative vessel segmentation results for those images.     

PG-RNN:一种基于递归神经网络的密码猜测模型

用户名—密码（口令）是目前最流行的用户身份认证方式，鉴于获取真实的大规模密码明文非常困难，利用密码猜测技术来生成大规模密码集，可以评估密码猜测算法效率、检测现有用户密码保护机制的缺陷等，是研究密码安全性的主要方法。本文提出了一种基于递归神经网络的密码猜测概率模型（password guessing RNN， PG-RNN），区别于传统的基于人为设计规则的密码生成方法，递归神经网络能够自动地学习到密码集本身的分布特征和字符规律。因此，在泄露的真实用户密码集上训练后的递归神经网络，能够生成非常接近训练集真实数据的密码，避免了人为设定规则来破译密码的局限性。实验结果表明，PG-RNN生成的密码在结构字符类型、密码长度分布上比Markov模型更好地接近原始训练数据的分布特征，同时在真实密码匹配度上，本文提出的PG-RNN模型比目前较好的基于生成对抗网络的PassGAN模型提高了1.2%。     

Neural recognition in a pyramidal structure

In recent years, there have been several proposals for the realization of models inspired to biological solutions for pattern recognition. In this work we propose a new approach, based on a hierarchical modular structure, to realize a system capable to learn by examples and recognize objects in digital images. The adopted techniques are based on multiresolution image analysis and neural networks. Performance on two different data sets and experimental timings on a single instruction multiple data (SIMD) machine are also reported.     

A mathematical analysis of the effects of Hebbian learning rules on the dynamics and structure of discrete-time random recurrent neural networks

We present a mathematical analysis of the effects of Hebbian learning in random recurrent neural networks, with a generic Hebbian learning rule, including passive forgetting and different timescales, for neuronal activity and learning dynamics. Previous numerical work has reported that Hebbian learning drives the system from chaos to a steady state through a sequence of bifurcations. Here, we interpret these results mathematically and show that these effects, involving a complex coupling between neuronal dynamics and synaptic graph structure, can be analyzed using Jacobian matrices, which introduce both a structural and a dynamical point of view on neural network evolution. Furthermore, we show that sensitivity to a learned pattern is maximal when the largest Lyapunov exponent is close to 0. We discuss how neural networks may take advantage of this regime of high functional interest.     

大数据分析的无限深度神经网络方法

深度神经网络(deep neural networks, DNNs)及其学习算法,作为成功的大数据分析方法,已为学术界和工业界所熟知.与传统方法相比,深度学习方法以数据驱动、能自动地从数据中提取特征(知识),对于分析非结构化、模式不明多变、跨领域的大数据具有显著优势.目前,在大数据分析中使用的深度神经网络主要是前馈神经网络(feedforward neural networks, FNNs),这种网络擅长提取静态数据的相关关系,适用于基于分类的数据应用场景.但是受到自身结构本质的限制,它提取数据时序特征的能力有限.无限深度神经网络(infinite deep neural networks)是一种具有反馈连接的回复式神经网络(recurrent neural networks, RNNs),本质上是一个动力学系统,网络状态随时间演化是这种网络的本质属性,它耦合了“时间参数”,更加适用于提取数据的时序特征,从而进行大数据的预测.将这种网络的反馈结构在时间维度展开,随着时间的运行,这种网络可以“无限深”,故称之为无限深度神经网络.重点介绍这种网络的拓扑结构和若干学习算法及其在语音识别和图像理解领域的成功实例.     

An investigation of neural networks for linear time-series forecasting

This study examines the capability of neural networks for linear time-series forecasting. Using both simulated and real data, the effects of neural network factors such as the number of input nodes and the number of hidden nodes as well as the training sample size are investigated. Results show that neural networks are quite competent in modeling and forecasting linear time series in a variety of situations and simple neural network structures are often effective in modeling and forecasting linear time series.Scope and purposeNeural network capability for nonlinear modeling and forecasting has been established in the literature both theoretically and empirically. The purpose of this paper is to investigate the effectiveness of neural networks for linear time-series analysis and forecasting. Several research studies on neural network capability for linear problems in regression and classification have yielded mixed findings. This study aims to provide further evidence on the effectiveness of neural network with regard to linear time-series forecasting. The significance of the study is that it is often difficult in reality to determine whether the underlying data generating process is linear or nonlinear. If neural networks can compete with traditional forecasting models for linear data with noise, they can be used in even broader situations for forecasting researchers and practitioners.     

Rough sets for adapting wavelet neural networks as a new classifier system

Classification is an important theme in data mining. Rough sets and neural networks are two techniques applied to data mining problems. Wavelet neural networks have recently attracted great interest because of their advantages over conventional neural networks as they are universal approximations and achieve faster convergence. This paper presents a hybrid system to extract efficiently classification rules from decision table. The neurons of such hybrid network instantiate approximate reasoning knowledge gleaned from input data. The new model uses rough set theory to help in decreasing the computational effort needed for building the network structure by using what is called reduct algorithm and a rules set (knowledge) is generated from the decision table. By applying the wavelets, frequencies analysis, rough sets and dynamic scaling in connection with neural network, novel and reliable classifier architecture is obtained and its effectiveness is verified by the experiments comparing with traditional rough set and neural networks approaches.     

Rough Neural Network of Variable Precision

In this paper, a new method is described to construct rough neural networks. On the base of rough set model, we present a method to develop rough neural network of variable precision and train it using Levenberg–Marquart algorithm. The method is particularly attractive because it combines the advantages of both rough logic networks and neural networks. In our system, weak generalization in rough sets theory and complexity in neural network are avoided while anti-jamming performance is highly improved and the network structure is also simplified. In experiments, the network is applied to classification of remote sensing images. The results show that our method is more effective and successful than application of rough sets and neural network separately.