Automated detection of cyclic alternating pattern and classification of sleep stages using deep neural network

The visual sleep stages scoring by human experts is the current gold standard for sleep analysis. However, this method is tedious, time-consuming, prone to human errors, and unable to detect microstructure of sleep such as cyclic alternating pattern (CAP) which is an important diagnostic factor for the detection of sleep disorders such as insomnia and obstructive sleep apnea (OSA). The CAP is only observed as subtle changes in the electroencephalogram (EEG) signals during non-rapid eye movement (NREM) sleep, making it very difficult for human experts to discern. Hence, it is important to have an automated system developed using artificial intelligence for accurate and robust detection of CAP and sleep stages classification. In this study, a deep learning model based on 1-dimensional convolutional neural network (1D-CNN) is proposed for CAP detection and homogenous 3-class sleep stages classification, namely wakefulness (W), rapid eye movement (REM) and NREM sleep. The proposed model is developed using standardized EEG recordings. Our developed CNN network achieved good model performance for 3-class sleep stages classification with a classification accuracy of 90.46%. Our proposed model also yielded a classification accuracy of 73.64% using balanced CAP dataset, and sensitivity of 92.06% with unbalanced CAP dataset. Our proposed model correctly identified majority of A-phases which comprised of only 12.6% in the unbalanced dataset. The performance of the developed prototype is ready to be tested with more data before clinical application.

     

基于CNN-BiLSTM的自动睡眠分期算法

睡眠分期是睡眠数据分析的基础,针对目前睡眠分期存在的依赖人工提取、人工判别效率低、自动睡眠分期准确率不高等问题,本文研究模型是基于卷积神经网络和双向长短时记忆神经网络2个深度学习神经网络相结合的,利用脑电信号来进行自动睡眠分期的模型方法.算法能提取得到原始脑电信号的梅尔频谱,利用卷积神经网络和双向长短时记忆神经网络进行...     

Artificial neural network model for effective cancer classification using microarray gene expression data

Microarray gene expression profile shall be exploited for the efficient and effective classification of cancers. This is a computationally challenging task because of large quantity of genes and relatively small amount of experiments in gene expression data. The repercussion of this work is to devise a framework of techniques based on supervised machine learning for discrimination of acute lymphoblastic leukemia and acute myeloid leukemia using microarray gene expression profiles. Artificial neural network (ANN) technique was employed for this classification. Moreover, ANN was compared with other five machine learning techniques. These methods were assessed on eight different classification performance measures. This article reports a significant classification accuracy of 98% using ANN with no error in identification of acute lymphoblastic leukemia and only one error in identification of acute myeloid leukemia on tenfold cross-validation and leave-one-out approach. Furthermore, models were validated on independent test data, and all samples were correctly classified.

     

Does spatial resolution matter? A multi-scale comparison of object-based and pixel-based methods for detecting change associated with gas well drilling operations

An implicit assumption of the geographic object-based image analysis (GEOBIA) literature is that GEOBIA is more accurate than pixel-based methods for high spatial resolution image classification, but that the benefits of using GEOBIA are likely to be lower when moderate resolution data are employed. This study investigates this assumption within the context of a case study of mapping forest clearings associated with drilling for natural gas. The forest clearings varied from 0.2 to 9.2 ha, with an average size of 0.9 ha. National Aerial Imagery Program data from 2004 to 2010, with 1 m pixel size, were resampled through pixel aggregation to generate imagery with 2, 5, 15, and 30 m pixel sizes. The imagery for each date and at each of the five spatial resolutions was classified into Forest and Non-forest classes, using both maximum likelihood and GEOBIA. Change maps were generated through overlay of the classified images. Accuracy evaluation was carried out using a random sampling approach. The 1 m GEOBIA classification was found to be significantly more accurate than the GEOBIA and per-pixel classifications with either 15 or 30 m resolution. However, at any one particular pixel size (e.g. 1 m), the pixel-based classification was not statistically different from the GEOBIA classification. In addition, for the specific class of forest clearings, accuracy varied with the spatial resolution of the imagery. As the pixel size coarsened from 1 to 30 m, accuracy for the per-pixel method increased from 59% to 80%, but decreased from 71% to 58% for the GEOBIA classification. In summary, for studying the impact of forest clearing associated with gas extraction, GEOBIA is more accurate than pixel-based methods, but only at the very finest resolution of 1 m. For coarser spatial resolutions, per-pixel methods are not statistically different from GEOBIA.     

基于双向LSTM卷积网络与注意力机制的自动睡眠分期模型

针对现阶段深度睡眠分期模型存在的梯度消失、对时序信息学习能力较弱等问题,提出一种基于双向长短时记忆卷积网络与注意力机制的自动睡眠分期模型。将少样本类别的睡眠脑电数据通过过采样方式进行数据增强后,利用带残差块的卷积神经网络学习数据特征表示,再通过带注意力层的双向长短时记忆网络挖掘深层时序信息,使用Softmax层实现睡眠分期的自动判别。实验使用Sleep-EDF数据集中19晚单通道脑电信号对模型进行交叉验证,取得了较高的分类准确率和宏平均F₁值,优于对比方法。该方法能够有效缓解睡眠分期判别中少数类分类性能较低的问题,并提高了深度睡眠分期模型的整体分类性能。     

Complex networks approach for EEG signal sleep stages classification

Sleep stage scoring is a challenging task. Most of existing sleep stage classification approaches rely on analysing electroencephalography (EEG) signals in time or frequency domain. A novel technique for EEG sleep stages classification is proposed in this paper. The statistical features and the similarities of complex networks are used to classify single channel EEG signals into six sleep stages. Firstly, each EEG segment of 30 s is divided into 75 sub-segments, and then different statistical features are extracted from each sub-segment. In this paper, feature extraction is important to reduce dimensionality of EEG data and the processing time in classification stage. Secondly, each vector of the extracted features, which represents one EEG segment, is transferred into a complex network. Thirdly, the similarity properties of the complex networks are extracted and classified into one of the six sleep stages using a k-means classifier. For further investigation, in the statistical features extraction phase two statistical features sets are tested and ranked based on the performance of the complex networks. To investigate the classification ability of complex networks combined with k-means, the extracted statistical features were also forwarded to a k-means and a support vector machine (SVM) for comparison. We also compare the proposed method with other existing methods in the literature. The experimental results show that the proposed method attains better classification results and a reasonable execution time compared with the SVM, k-means and the other existing methods. The research results in this paper indicate that the proposed method can assist neurologists and sleep specialists in diagnosing and monitoring sleep disorders.     

Sleep spindle detection using artificial neural networks trained with filtered time-domain EEG: a feasibility study

An artificial neural network (ANN) based on the Multi-Layer Perceptron (MLP) architecture is used for detecting sleep spindles in band-pass filtered electroencephalograms (EEG), without feature extraction. Following optimum classification schemes, the sensitivity of the network ranges from 79.2% to 87.5%, while the false positive rate ranges from 3.8% to 15.5%. Furthermore, due to the operation of the ANN on time-domain EEG data, there is agreement with visual assessment concerning temporal resolution. Specifically, the total inter-spindle interval duration and the total duration of spindles are calculated with 99% and 92% accuracy, respectively. Therefore, the present method may be suitable for investigations of the dynamics among successive inter-spindle intervals, which could provide information on the role of spindles in the sleep process, and for studies of pharmacological effects on sleep structure, as revealed by the modification of total spindle duration.     

Automated Dataset Generation for Training Peer-to-Peer Machine Learning Classifiers

Peer-to-peer (P2P) classifications based on flow statistics have been proven accurate in detecting P2P traffic. A machine learning classification is affected by the quality and recency of the training dataset used. Hence, to classify P2P traffic on-line requires the removal of these limitations. In this paper, an automated training dataset generation for an on-line P2P traffic classification is proposed to allow frequent classifier retraining. A two-stage training dataset generator (TSTDG) is proposed by combining a 3-class heuristic and a 3-class statistical classification to automatically generate a training dataset. In the heuristic stage, traffic is classified as P2P, non-P2P, or unknown. In the statistical stage, a dual Decision Tree is built based on a dataset generated in the heuristic stage to reduce the amount of classified unknown traffic. The final training dataset is generated based on all flows that are classified in these two stages. The proposed system has been evaluated on traces captured from a campus network. The overall results show that the TSTDG can generate an accurate training dataset by classifying around 94 % of total flows with high accuracy (98.59 %) and a low false positive rate (1.27 %).     

Comparison of support vector machine,artificial neural network,and spectral angle mapper algorithms for crop classification using LISS IV data

The Resourcesat-2 is a highly suitable satellite for crop classification studies with its improved features and capabilities. Data from one of its sensors, the linear imaging and self-scanning (LISS IV), which has a spatial resolution of 5.8 m, was used to compare the relative accuracies achieved by support vector machine (SVM), artificial neural network (ANN), and spectral angle mapper (SAM) algorithms for the classification of various crops and non-crop covering a part of Varanasi district, Uttar Pradesh, India. The separability analysis was performed using a transformed divergence (TD) method between categories to assess the quality of training samples. The outcome of the present study indicates better performance of SVM and ANN algorithms in comparison to SAM for the classification using LISS IV sensor data. The overall accuracies obtained by SVM and ANN were 93.45% and 92.32%, respectively, whereas the lower accuracy of 74.99% was achieved using the SAM algorithm through error matrix analysis. Results derived from SVM, ANN, and SAM classification algorithms were validated with the ground truth information acquired by the field visit on the same day of satellite data acquisition.     

A genetic algorithm-based multi-objective optimization of an artificial neural network classifier for breast cancer diagnosis

The conventional technique for diagnosing the breast cancer disease relies on human experiences to identify the presence of certain pattern from the database. It is time-consuming and incurs unnecessary burden to radiologists. This work proposes a genetic algorithm-based multi-objective optimization of an Artificial Neural Network classifier, namely GA-MOO-NN, for the automatic breast cancer diagnosis. It performs a simultaneous search for the significant feature subsets and the optimum architecture of the network. The combination of ANN’s parameters with feature selection to be optimized by Genetic Algorithm is novel. The Pareto-optimality with new ranking approach is applied for simultaneous minimizations of two competing objectives: the number of network‘s connections and squared error percentage of the validation data. Result shows that the algorithm with the proposed combination of objectives has achieved the best and average, 98.85 and 98.10 % accuracy of classification, respectively, on breast cancer dataset which outperform most systems of other works found in the literature.     

Pre-determination of OSA degree using morphological features of the ECG signal

Obstructive sleep apnea (OSA) is a very common, but a difficult sleep disorder to diagnose. Recurrent obstructions form in the airway during sleep, such that OSA can threaten a breathing capacity of patients. Clinically, continuous positive airway pressure (CPAP) is the most specific and effective treatment for this. In addition, these patients must be separated according to its degree, with CPAP treatment applied as a result. In this study, 30 OSA patients from two different databases were automatically classified using electrocardiogram (ECG) data, identified as mild, moderate, and severe. One of the databases was original recordings which had 9 OSA patients with 8303 epochs and the other one was Physionet benchmark database which had 21 patients with 20,824 epochs. Fifteen morphological features could be identified when apnea was seen, both before and after it presented. Five data groups in total for first dataset and second dataset were prepared with these features and 10-fold cross validation was used to effectively determine the test data. Then, sequential backward feature selection (SBFS) algorithm was applied to understand the more effective features. The prepared data groups were evaluated with artificial neural networks (ANN) to obtain optimum classification performance. All processes were repeated for ten times and error deviation was calculated for the accuracy. Furthermore, different classifiers which are frequently used in the literature were tested with selected features. The degree of OSA was estimated from three epochs in pre-apnea data, yielding the success rates of 97.20 ± 2.15% and 90.18 ± 8.11% with the SBFS algorithm for the first and second datasets, respectively. Also, SVM classifier followed ANN system in the success rates of 96.23 ± 3.48% and 88.75 ± 8.52% for used datasets.     

A framework for diagnosing cervical cancer disease based on feedforward MLP neural network and ThinPrep histopathological cell image features

In this paper, Levenberg–Marquardt feedforward MLP neural network (LMFFNN) was proposed to classify cervical cell images obtained from 100 patients including healthy, low-grade intraepithelial squamous lesion and high-grade intraepithelial squamous lesion cases. This neural network along with extracted cell image features is a new model for cervical cell image classification. The semiautomated cervical cancer diagnosis system is composed of two phases: image preprocessing/processing and feedforward MLP neural network. In the first stage, image preprocessing is done to reduce the existing noises without lowering the resolution. After that, image processing algorithms were applied to manually cropped cell images to achieve a linear plot which includes real components, were used as LMFFNN inputs for classification of cervical cell images. Based on the results, cervical cell images were classified successfully with 100 % correct classification rate using the proposed method. Moreover, the rates of sensitivity and specificity were calculated as 100 % using LMFFNN method. It was shown there was a good agreement between the expert decision and values gained from the ANN model.     

回归分析最小化睡眠信号中的ECG干扰

简要介绍了回归分析的方法,并将其应用于睡眠EEG自动分期研究中,用于抑制所采集的各通道信号中的ECG干扰。实验结果表明：回归分析方法可以极大地消除各通道信号中的ECG干扰,对研究睡眠EEG的计算机自动分期有很大的意义。     

Face recognition Face2vec based on deep learning: Small database case

Object classification is a common problem in artificial intelligence and now it is usually approached by deep learning. In the paper the artificial neural network (ANN) architecture is considered. According to described ANN architecture, the ANN models are trained and tested on a relatively small Color-FERET facial image database under different conditions. The best fine-tuned ANN model provides 94% face recognition accuracy on Color-FERET frontal images and 98% face recognition accuracy within 3 attempts. However, for improving recognition system accuracy large data sets are still necessary preferably consisting of millions of images.     

Modeling of Trivalent Chromium Sorption onto Commercial Resins by Artificial Neural Network

In this research, artificial neural network (ANN) model having three layers was developed for precise estimation of Cr(III) sorption rate varying from 17% to 99% by commercial resins as a result of obtaining 38 experimental data. ANN was trained by using the data of sorption process obtained at different pH (2–7) values with Amberjet 1200H and Diaion CR11 amount (0.01–0.1 g) dosage, initial metal concentration (4.6–31.7 ppm), contact time (5–240 min), and a temperature of 25°C. A feed-forward back propagation network type with one hidden layer, different algorithm (transcg, trainlm, traingdm, traincgp, and trainrp), different transfer function (logsig, tansig, and purelin) for hidden layer and purelin transfer function for output layer were used, respectively. Each model trained for cross-validation was compared with the data that were not used. The trainlm algorithm and purelin transfer functions with five neurons were well fitted to training data and cross-validation. After the best suitable coefficient of determination and mean squared error values were found in the current network, optimal result was searched by changing the number of neurons range from 1 to 20 in the current network hidden layer.     

An approach for automatic sleep stage scoring and apnea-hypopnea detection

In this article we present an application of data mining to the medical domain sleep research, an approach for automatic sleep stage scoring and apnea-hypopnea detection. By several combined techniques (Fourier and wavelet transform, derivative dynamic time warping, and waveform recognition), our approach extracts meaningful features (frequencies and special patterns like k-complexes and sleep spindles) from physiological recordings containing EEG, ECG, EOG and EMG data. Based on these pieces of information, an ensemble of decision trees is constructed using the principle of bagging, which classifies sleep epochs in their sleep stages according to the rules by Rechtschaffen and Kales and annotates occurrences of apnea-hypopnea (total or partial cessation of respiration). After that, casebased reasoning is applied in order to improve quality. We tested and evaluated our approach on several large public databases from PhysioBank, which showed an overall accuracy of 95.2% for sleep stage scoring and 94.5% for classifying minutes as apneic or non-apneic.     

ADAPTIVE NEURO-FUZZY INFERENCE SYSTEM FOR AUTOMATIC SLEEP MULTISTAGE LEVEL SCORING EMPLOYING EEG,EOG, AND EMG EXTRACTED FEATURES

A new system for sleep multistage level scoring by employing extracted features from twenty five polysomnographic recording is presented. For the new system, an adaptive neuro-fuzzy inference system (ANFIS) is developed for each sleep stage. Initially, three types of electrophysiological signals including electroencephalogram (EEG), electrooculogram (EOG), and electromyogram (EMG) were collected from twenty five healthy subjects. The input pattern used for training the ANFIS subsystem is a set of extracted features based on the entropy measure which characterize the recorded signals. Finally an output selection subsystem is utilized to provide the appropriate sleep stage according to the ANFIS stage subsystems outputs. The developed system was able to provide an acceptable estimation for six sleep stages with an average accuracy of about 76.43% which confirmed its ability for multistage sleep level scoring based on the extracted features from the EEG, EOG and EMG signals compared to other approaches.     

Automatic recognition of alertness level from EEG by using neural network and wavelet coefficients

Electrophysiological recordings are considered a reliable method of assessing a person's alertness. Sleep medicine is asked to offer objective methods to measure daytime alertness, tiredness and sleepiness. As EEG signals are non-stationary, the conventional method of frequency analysis is not highly successful in recognition of alertness level. This paper deals with a novel method of analysis of EEG signals using wavelet transform, and classification using ANN. EEG signals were decomposed into the frequency sub-bands using wavelet transform and a set of statistical features was extracted from the sub-bands to represent the distribution of wavelet coefficients. Then these statistical features were used as an input to an ANN with three discrete outputs: alert, drowsy and sleep. The error back-propagation neural network is selected as a classifier to discriminate the alertness level of a subject. EEG signals were obtained from 30 healthy subjects. The group consisted of 14 females and 16 males with ages ranging from 18 to 65 years and a mean age of 33.5 years, and a Body Mass Index (BMI) of 32.4±7.3 kg/m². Alertness level and classification properties of ANN were tested using the data recorded in 12 healthy subjects, whereby the EEG recordings were not used to train the ANN. The statistics were used as a measure of potential applicability of the ANN. The accuracy of the ANN was 95±3% alert, 93±4% drowsy and 92±5% sleep.     

A hybrid neural network approach for credit scoring

The development of an effective credit scoring model has become a very important issue as the credit industry is confronted with ever‐intensifying competition and aggravating bad debt problems. During the past few years, a substantial number of studies in the field of statistics have been conducted to improve the accuracy of credit scoring models. In order to refine the classification and decrease misclassification, this paper presents a two‐stage model. Focusing on classification, the first stage aims at constructing an artificial neural network (ANN)‐based credit scoring model to categorize applicants into the group of accepted (good) credit and the group of rejected (bad) credit. Switching from classification to reassignment, the second stage proceeds to reduce the Type I error by retrieving the originally rejected good credit applicants to conditional acceptance using the Case‐Based Reasoning (CBR) classification technique. The proposed model (RST–ANN–CBR) is applied to a credit card dataset to verify its effectiveness. As the results indicate, the proposed model is able to achieve more accurate credit scoring than four other methods; more importantly, it is validated to recover potentially lost customers and to increase business revenues.     

Neural network nonlinear modeling for hydrogen production using anaerobic fermentation

The potential of utilizing artificial neural network (ANN) model approach for simulate and predict the hydrogen yield in batch model using Clostridium saccharoperbutylacetonicum N1-4 (ATCC 13564) was investigated. A unique architecture has been introduced in this research to mimic the inter-relationship between three input parameters initial substrate, initial medium pH and reaction temperature (37 °C, 6.0 ± 0.2, 10), respectively, to predict hydrogen yield. Sixty data records from the experiment have been utilized to develop the ANN model. The results showed that the proposed ANN model provided significant level of accuracy for prediction with maximum error (10 %). Furthermore, a comparative analysis with a traditional approach Box–Wilson design (BWD) has proved that the ANN model output significantly outperformed the BWD. ANN model overcomes the limitation of the BWD approach with respect to the number of records, which is merely considering limited length of stochastic pattern for hydrogen yield (15 records).