A medical diagnostic tool based on radial basis function classifiers and evolutionary simulated annealing

ObjectiveThe profusion of data accumulating in the form of medical records could be of great help for developing medical decision support systems. The objective of this paper is to present a methodology for designing data-driven medical diagnostic tools, based on neural network classifiers.MethodsThe proposed approach adopts the radial basis function (RBF) neural network architecture and the non-symmetric fuzzy means (NSFM) training algorithm, which presents certain advantages including better approximation capabilities and shorter computational times. The novelty in this work consists of adapting the NSFM algorithm to train RBF classifiers, and suitably tailoring the evolutionary simulated annealing (ESA) technique to optimize the produced RBF models. The integration of ESA is critical as it helps the optimization procedure to escape from local minima, which could arise from the application of the traditional simulated annealing algorithm, and thus discover improved solutions. The resulting method is evaluated in nine different medical benchmark datasets, where the common objective is to train a suitable classifier. The evaluation includes a comparison with two different schemes for training classifiers, including a standard RBF training technique and support vector machines (SVMs). Accuracy% and the Matthews Correlation Coefficient (MCC) are used for comparing the performance of the three classifiers.ResultsResults show that the use of ESA helps to greatly improve the performance of the NSFM algorithm and provide satisfactory classification accuracy. In almost all benchmark datasets, the best solution found by the ESA-NSFM algorithm outperforms the results produced by the SFM algorithm and SVMs, considering either the accuracy% or the MCC criterion. Furthermore, in the majority of datasets, the average solution of the ESA-NSFM population is statistically significantly higher in terms of accuracy% and MCC at the 95% confidence level, compared to the global optimum solution that its rivals could achieve. As far as computational times are concerned, the proposed approach was found to be faster compared to SVMs.ConclusionsThe results of this study suggest that the ESA-NSFM algorithm can form the basis of a generic method for knowledge extraction from data originating from different kinds of medical records. Testing the proposed approach on a number of benchmark datasets, indicates that it provides increased diagnostic accuracy in comparison with two different classifier training methods.     

基于GA-RBF神经网络和sEMG的下肢动作识别方法研究

为了提高人体肌电信号对于下肢动作识别的准确率,提出一种基于遗传算法(GA)优化的径向基(RBF)神经网络分类模型.通过采集人体日常8种下肢动作的表面肌电信号并选择"sym6"系小波函数对肌电信号进行滤波预处理,使用主成分分析法(PCA)对时频域特征降维,把特征向量输入GA算法优化的RBF神经网络进行训练和识别.实验结果...     

计算机处理睡眠数据中2导脑电和1导眼电的非周期波形分析

背景：睡眠中记录的数据量很大,不用计算机自动处理不能满足实用需要,而现有的对睡眠数据进行分期的方法准确率都不高。 目的：考察仅依据脑电与眼电,基于非周期波形分析和径向基函数遗传神经网络的睡眠数据分期新方法。 方法：实验数据来自MIT的PhysioBank中的S1eep-EDF数据库,共8名被试,各记录2导脑电和1导眼电。原始数据经零相位数字滤波后,进行非周期波形分析,得出每个Epoch的特征向量,经预处理后送遗传径向基函数。神经网络配合专家手工分类结果进行训练,训练好的神经网络再对测试数据进行分析。 结果与结论：总的分期符合率为95.6%,超出已知文献研究结果(70%~90%),具有很高的实用价值,能满足睡眠研究与临床使用。     

Classifying coronary dysfunction using neural networks through cardiovascular auscultation

The paper applies artificial neural networks (ANNs) to the analysis of heart sound abnormalities through auscultation. Audio auscultation samples of 16 different coronary abnormalities were collected. Data pre-processing included down-sampling of the auscultated data and use of the fast Fourier transform (FFT) and the Levinson-Durbin autoregression algorithms for feature extraction and efficient data encoding. These data were used in the training of a multi-layer perceptron (MLP) and radial basis function (RBF) neural network to develop a classification mechanism capable of distinguishing between different heart sound abnormalities. The MLP and RBF networks attained classification accuracies of 84% and 88%, respectively. The application of ANNs to the analysis of respiratory auscultation and consequently the development of a combined cardio-respiratory analysis system using auscultated data could lead to faster and more efficient treatment.     

基于灰度共生矩阵和人工神经网络的肺癌CT图像的分类研究

目的研究小细胞肺癌（SCLC）和非小细胞肺癌（NSCLC）的分类问题。方法217例肺癌患者．其中男性165例．殳性52例;年龄35～80岁,平均年龄61．5岁。其中SCLC108例,NSCLC109例。提取患者764幅肺癌CT图像的灰度共生矩阵,选取对比度、熵、能量和逆差矩4个特征值,借助临床确诊结果,利用多层前向（BP）、径向基函数（RBF）人工神经网络对特征进行训练测试。结果BP人工神经网络对10％的78例样本进行测试,SCLC42例预测正确．NSCLC33例预测正确．3例预测失败。RBF神经网络对10％的78例测试样本进行测试,SCLC42例预测正确．NSCLC36例预测正确、类似方法对样本总数的70％进行训练,用30％的230例进行测试;BP人工神经网络有209例预测正确。正确率为90．9％：其中SCLC111例预测正确,正确检出率为88．8％;NSCLC98例预测正确,正确检出率为93.3％。RBF人工神经网络有216例预测正确．正确率为93．9％,其中SCLC117例预测正确,正确率为93．6％;NSCLC99例预测正确,止确检出率为94.3％。可见BP、RBF人1二神经网络对SCLC和NSCLC均具有90％以上的正确率,高于人工诊断结果。结论基于灰度共生矩阵的对比度、熵、能量和逆差矩4个特征值能反映SCLC和NSCLC的有效特征参量．通过人工神经网络能达到分类目的,辅助临床治疗。     

Neural network modeling for surgical decisions on traumatic brain injury patients

Computerized medical decision support systems have been a major research topic in recent years. Intelligent computer programs were implemented to aid physicians and other medical professionals in making difficult medical decisions. This report compares three different mathematical models for building a traumatic brain injury (TBI) medical decision support system (MDSS). These models were developed based on a large TBI patient database. This MDSS accepts a set of patient data such as the types of skull fracture, Glasgow Coma Scale (GCS), episode of convulsion and return the chance that a neurosurgeon would recommend an open-skull surgery for this patient. The three mathematical models described in this report including a logistic regression model, a multi-layer perceptron (MLP) neural network and a radial-basis-function (RBF) neural network. From the 12,640 patients selected from the database. A randomly drawn 9480 cases were used as the training group to develop/train our models. The other 3160 cases were in the validation group which we used to evaluate the performance of these models. We used sensitivity, specificity, areas under receiver-operating characteristics (ROC) curve and calibration curves as the indicator of how accurate these models are in predicting a neurosurgeon's decision on open-skull surgery. The results showed that, assuming equal importance of sensitivity and specificity, the logistic regression model had a (sensitivity, specificity) of (73%, 68%), compared to (80%, 80%) from the RBF model and (88%, 80%) from the MLP model. The resultant areas under ROC curve for logistic regression, RBF and MLP neural networks are 0.761, 0.880 and 0.897, respectively (P < 0.05). Among these models, the logistic regression has noticeably poorer calibration. This study demonstrated the feasibility of applying neural networks as the mechanism for TBI decision support systems based on clinical databases. The results also suggest that neural networks may be a better solution for complex, non-linear medical decision support systems than conventional statistical techniques such as logistic regression.     

3D asymmetric expectation-maximization attention network for brain tumor segmentation

Automatic brain tumor segmentation on MRI is a prerequisite to provide a quantitative and intuitive assistance for clinical diagnosis and treatment. Meanwhile, 3D deep neural network related brain tumor segmentation models have demonstrated considerable accuracy improvement over corresponding 2D methodologies. However, 3D brain tumor segmentation models generally suffer from high computation cost. Motivated by a recently proposed 3D dilated multi-fiber network (DMF-Net) architecture that pays more attention to reduction of computation cost, we present in this work a novel encoder-decoder neural network, ie a 3D asymmetric expectation-maximization attention network (AEMA-Net), to automatically segment brain tumors. We modify DMF-Net by introducing an asymmetric convolution block into a multi-fiber unit and a dilated multi-fiber unit to capture more powerful deep features for the brain tumor segmentation. In addition, AEMA-Net further incorporates an expectation-maximization attention (EMA) module into the DMF-Net by embedding the EMA block in the third stage of skip connection, which focuses on capturing the long-range dependence of context. We extensively evaluate AEMA-Net on three MRI brain tumor segmentation benchmarks of BraTS 2018, 2019 and 2020 datasets. Experimental results demonstrate that AEMA-Net outperforms both 3D U-Net and DMF-Net, and it achieves competitive performance compared with the state-of-the-art brain tumor segmentation methods.     

Detection of Hard Exudates in Retinal Images Using a Radial Basis Function Classifier

Diabetic retinopathy (DR) is one of the most important causes of visual impairment. Automatic recognition of DR lesions, like hard exudates (EXs), in retinal images can contribute to the diagnosis and screening of the disease. The aim of this study was to automatically detect these lesions in fundus images. To achieve this goal, each image was normalized and the candidate EX regions were segmented by a combination of global and adaptive thresholding. Then, a group of features was extracted from image regions and the subset which best discriminated between EXs and retinal background was selected by means of logistic regression (LR). This optimal subset was subsequently used as input to a radial basis function (RBF) neural network. To improve the performance of the proposed algorithm, some noisy regions were eliminated by an innovative postprocessing of the image. The main novelty of the paper is the use of LR in conjunction with RBF and the proposed postprocessing technique. Our database was composed of 117 images with variable color, brightness and quality. The database was divided into a training set of 50 images (from DR patients) and a test set of 67 images (40 from DR patients and 27 from healthy retinas). Using a lesion-based criterion (pixel resolution), a mean sensitivity of 92.1% and a mean positive predictive value of 86.4% were obtained. With an image-based criterion, a mean sensitivity of 100%, mean specificity of 70.4% and mean accuracy of 88.1% were achieved. These results suggest that the proposed method could be a diagnostic aid for ophthalmologists in the screening for DR.     

FRAN and RBF-PSO as two components of a hyper framework to recognize protein folds

In this paper, an intelligent hyper framework is proposed to recognize protein folds from its amino acid sequence which is a fundamental problem in bioinformatics. This framework includes some statistical and intelligent algorithms for proteins classification. The main components of the proposed framework are the Fuzzy Resource-Allocating Network (FRAN) and the Radial Bases Function based on Particle Swarm Optimization (RBF-PSO). FRAN applies a dynamic method to tune up the RBF network parameters. Due to the patterns complexity captured in protein dataset, FRAN classifies the proteins under fuzzy conditions. Also, RBF-PSO applies PSO to tune up the RBF classifier. Experimental results demonstrate that FRAN improves prediction accuracy up to 51% and achieves acceptable multi-class results for protein fold prediction. Although RBF-PSO provides reasonable results for protein fold recognition up to 48%, it is weaker than FRAN in some cases. However the proposed hyper framework provides an opportunity to use a great range of intelligent methods and can learn from previous experiences. Thus it can avoid the weakness of some intelligent methods in terms of memory, computational time and static structure. Furthermore, the performance of this system can be enhanced throughout the system life-cycle.     

A hierarchical neural network algorithm for robust and automatic windowing of MR images

A novel hierarchical neural network based algorithm for automatic adjustment of display window width and center for a wide range of magnetic resonance (MR) images is presented in this paper. The algorithm consists of a feature generator utilizing both wavelet histogram and compact spatial statistical information computed from a MR image, a competitive layer based neural network for clustering MR images into different subclasses, two pairs of a radial basis function (RBF) network and a bi-modal linear estimator for each subclass, as well as a data fusion process using estimates from both estimators to compute the final display parameters. Both estimators can adapt to new kinds of MR images simply by training them with those images, which make the algorithm adaptive and extendable. The RBF based estimator performs very well for images that are similar to those in the training data set. The bi-modal linear estimator provides reasonable estimations for a wide range of images that may not be included in the training data set. The data fusion step makes the final estimation of the display parameters accurate for trained images and robust for the unknown images. The algorithm has been tested on a wide range of MR images and has shown satisfactory results.     

基于双谱分析—BP神经网络的ICU急性低血压发生预测模型研究

目的:ICU中,急性低血压的发生严重威胁着患者的生命安全,能够及时准确地对其预测具有重要临床意义。为了提高医生对患者发病提前判断的准确性,本文研究了一种基于双谱分析和BP神经网络的急性低血压发生的预测模型。方法:应用双谱分析提取动脉血压数据特征,构建特征向量,利用BP神经网络训练出分类预测模型,实现对急性低血压发生的提前预测。结果:经过不断的优化调整,本文最终构造了一个三层的BP神经网络预测模型,具有良好的自动预测能力。结论:实验表明,本方法能达到比较好的分类预测效果,可为ICU中急性低血压发生的提前预测和干预提供辅助参考。     

Radial basis function classifiers to help in the diagnosis of the obstructive sleep apnoea syndrome from nocturnal oximetry

The aim of this study is to assess the ability of radial basis function (RBF) classifiers as an assistant tool for the diagnosis of the obstructive sleep apnoea syndrome (OSAS). A total of 187 subjects suspected of suffering from OSAS were available for our research. The initial population was divided into training, validation and test sets for deriving and testing our neural classifiers. We used nonlinear features from nocturnal oxygen saturation (SaO₂) to perform patients’ classification. We evaluated three different RBF construction techniques based on the following algorithms: k-means (KM), fuzzy c-means (FCM) and orthogonal least squares (OLS). A diagnostic accuracy of 86.1, 84.7 and 85.5% was provided by the networks developed with KM, FCM and OLS, respectively. The three proposed networks achieved an area under the receiver operating characteristic (ROC) curve over 0.90. Our results showed that a useful non-invasive method could be applied to diagnose OSAS from nonlinear features of SaO₂ with RBF classifiers.     

Comparative study between DD-HMM and RBF in ventricular tachycardia and ventricular fibrillation recognition

This paper deals with automatic recognition of cardiac arrhythmias that require immediate electrical defibrillation therapy (ventricular fibrillation and ventricular tachycardia), through ECG (electrocardiogram) samples. The DD-HMM (discrete density hidden Markov model) and RBF (radial basis function) neural network algorithms were compared in the following aspects: precision, defined as correct recognition percentage and process time, defined as the delay since the ECG input until the result, indicating shock or non-shock events. The results show that RBF is more precise than DD-HMM but not so fast to evaluate. PhysioNet database files were used to train and to validate the algorithms.     

Radial basis function neural networks for the characterization of heart rate variability dynamics

This study introduces new neural network based methods for the assessment of the dynamics of the heart rate variability (HRV) signal. The heart rate regulation is assessed as a dynamical system operating in chaotic regimes. Radial-basis function (RBF) networks are applied as a tool for learning and predicting the HRV dynamics. HRV signals are analyzed from normal subjects before and after pharmacological autonomic nervous system (ANS) blockade and from diabetic patients with dysfunctional ANS. The heart rate of normal subjects presents notable predictability. The prediction error is minimized, in fewer degrees of freedom, in the case of diabetic patients. However, for the case of pharmacological ANS blockade, although correlation dimension approaches indicate significant reduction in complexity, the RBF networks fail to reconstruct adequately the underlying dynamics. The transient attributes of the HRV dynamics under the pharmacological disturbance is elucidated as the explanation for the prediction inability.     

Efficient prediction methods for selecting effective siRNA sequences

Although short interfering RNA (siRNA) has been widely used for studying gene functions in mammalian cells, its gene silencing efficacy varies markedly and there are only a few consistencies among the recently reported design rules/guidelines for selecting siRNA sequences effective for mammalian genes. Another shortcoming of the previously reported methods is that they cannot estimate the probability that a candidate sequence will silence the target gene. This paper first reviewed the recently reported siRNA design guidelines and clarified the problems concerning the guidelines. It then proposed two prediction methods—Radial Basis Function (RBF) network and decision tree learning—and their combined method for selecting effective siRNA target sequences from many possible candidate sequences. They are quite different from the previous score-based siRNA design techniques and can predict the probability that a candidate siRNA sequence will be effective. The methods imply high estimation accuracy for selecting candidate siRNA sequences.     

Prediction on lengths of stay in the postanesthesia care unit following general anesthesia: preliminary study of the neural network and logistic regression modelling

The length of stay in the postanesthesia care unit (PACU) following general anesthesia in adults is an important issue. A model, which can predict the results of PACU stays, could improve the utilization of PACU and operating room resources through a more efficient arrangement. The purpose of study was to compare the performance of neural network to logistic regression analysis using clinical sets of data from adult patients undergoing general anesthesia. An artificial neural network was trained with 409 clinical sets using backward error propagation and validated through independent testing of 183 records. Twenty-two inputs were used to find determinants and to predict categorical values. Logistic regression analysis was performed to provide a comparison. The neural network correctly predicted in 81.4% of situations and identified discriminating variables (intubated state, sex, neuromuscular blocker and intraoperative use of opioid), whereas the figure was 65.0% in logistic regression analysis. We concluded that the neural network could provide a useful predictive model for the optimization of limited resources. The neural network is a new alternative classifying method for developing a predictive paradigm, and it has a higher classifying performance compared to the logistic regression model.     

Holter系统的技术要求及神经网络方法

为了长时间、准确的记录分分析体表ＥＣＧ，从各个方面满足基础研究、临床、康复以及特殊生理需要，对Ｈｏｌｔｅｒ系统提出了愈来愈高的要求。作者从技术实现的角度讨论了Ｈｏｌｔｅｒ的主要技术要求，包括记录部分和分析部分；还介绍了几种用神经网络模型满足Ｈｏｌｔｅｒ系统技术要求的原理，包括用多层感知器实现数据压缩及分类，用高阶神经网络实现ＥＣＧ的分类，以及用ＡＲＴ模型识别ＥＣＧ中的Ｐ波与Ｔ波。     

Prediction of Seizure Onset in an <Emphasis Type="Italic">In-Vitro</Emphasis> Hippocampal Slice Model of Epilepsy Using Gaussian-Based and Wavelet-Based Artificial Neural Networks

We propose that artificial neural networks (ANNs) can be used to predict seizure onsets in an in-vitro hippocampal slice model capable of generating spontaneous seizure-like events (SLEs) in their extracellular field recordings. This paper assesses the effectiveness of two ANN prediction schemes: Gaussian-based artificial neural network (GANN) and wavelet-based artificial neural network (WANN). The GANN prediction system consists of a recurrent network having Gaussian radial basis function (RBF) nonlinearities capable of extracting the estimated manifold of the system. It is able to classify the underlying dynamics of spontaneous in-vitro activities into interictal, preictal and ictal modes. It is also able to successfully predict the onsets of SLEs as early as 60 s before. Improvements can be made to the overall seizure predictor design by incorporating time-varying frequency information. Consequently, the idea of WANN is considered. The WANN design entails the assumption that frequency variations in the extracellular field recordings can be used to compute the times at which onsets of SLEs are most likely to occur in the future. Progressions of different frequency components can be captured by the ANN using appropriate frequency band adjustments via pruning, after the initial wavelet transforms. In the off-line processing comprised of 102 spontaneous SLEs generated from 14 in-vitro rat hippocampal slices, with half of them used for training and the other half for testing, the WANN is able to predict the forecoming ictal onsets as early as 2 min prior to SLEs with over 75% accuracy within a 30 s precision window.     

Determining Top Fully Connected Layer’s Hidden Neuron Count for Transfer Learning,Using Knowledge Distillation: a Case Study on Chest X-Ray Classification of Pneumonia and COVID-19

Deep convolutional neural network (CNN)-assisted classification of images is one of the most discussed topics in recent years. Continuously innovation of neural network architectures is making it more correct and efficient every day. But training a neural network from scratch is very time-consuming and requires a lot of sophisticated computational equipment and power. So, using some pre-trained neural network as feature extractor for any image classification task or “transfer learning” is a very popular approach that saves time and computational power for practical use of CNNs. In this paper, an efficient way of building full model from any pre-trained model with high accuracy and low memory is proposed using knowledge distillation. Using the distilled knowledge of the last layer of pre-trained networks passes through fully connected layers with different hidden layers, followed by Softmax layer. The accuracies of student networks are mildly lesser than the whole models, but accuracy of student models clearly indicates the accuracy of the real network. In this way, the best number of hidden layers for dense layer for that pre-trained network with best accuracy and no-overfitting can be found with less time. Here, VGG16 and VGG19 (pre-trained upon “ImageNet” dataset) is tested upon chest X-rays (pneumonia and COVID-19). For finding the best total number of hidden layers, it saves nearly 44 min for VGG19 and 36 min and 37 s for VGG16 feature extractor.     

Formulation of effects of atropine, pralidoxime and magnesium sulfate on cardiac tissue levels of nitric oxide, malondialdehyde and glutathione in organophosphate poisoning using artificial neural network

Anticholinesterase poisoning is an important health problem in our country, and a complete understanding of its underlying mechanisms is essential for the emergency physician. So, this study focused on two purposes. First one was aimed to investigate the biochemical analysis to determine the tissue levels of malondialdehyde (MDA), glutathione and nitric oxide (NO). Secondly, it was planned to model and formulate the effects of some drugs on cardiac tissues levels of NO, MDA and glutathione in acute organophosphate poisoning in rats by the application of neural network based on experimental results. It has been planned to determine whether artificial neural network (ANN) is appropriate tool to analyze and formulate it. As a result, it has been considered that ANN can be effectively used to model NO, MDA and glutathione level. The performances of ANN formulation versus target experimental values are found to be quite high. It is concluded that, proposed NN models are also presented as simple explicit mathematical functions for further use by researchers.