Objective. Sleep apnea significantly decreases the quality of life. The apnea hypopnea index (AHI... more Objective. Sleep apnea significantly decreases the quality of life. The apnea hypopnea index (AHI) is the main indicator for sleep apnea diagnosis. This study explored a novel automatic algorithm to diagnose sleep apnea from nasal airflow (AF) and pulse oximetry (SpO2) signals. Approach. Of the 988 polysomnography (PSG) records from the sleep heart health study (SHHS), 45 were randomly selected for the development of an algorithm and the remainder for validation (n = 943). The algorithm detects apnea events by a digitization process, following the determination of the peak excursion (peak-to-trough amplitude) from AF envelope. Hypopnea events were determined from the AF envelope and oxygen desaturation with correction to time lag in SpO2. Total sleep time (TST) was estimated from an optimized percentage of artefact-free total recording time. AHI was estimated from the number of detected events divided by the estimated TST. The estimated AHI was compared to the scored SHHS data for p...
IEEE Transactions on Evolutionary Computation, 2021
This paper investigates deep neural networks (DNNs) based lung nodule classification with hyperpa... more This paper investigates deep neural networks (DNNs) based lung nodule classification with hyperparameter optimization. Hyperparameter optimization in DNNs is a computationally expensive problem, and a surrogate-assisted evolutionary algorithm has been recently introduced to automatically search for optimal hyperparameter configurations of DNNs, by applying computationally efficient surrogate models to approximate the validation error function of hyperparameter configurations. Different from existing surrogate models adopting stationary covariance functions (kernels) to measure the difference between hyperparameter points, this paper proposes a non-stationary kernel that allows the surrogate model to adapt to functions whose smoothness varies with the spatial location of inputs. A multi-level convolutional neural network (ML-CNN) is built for lung nodule classification, and the hyperparameter configuration is optimized by the proposed non-stationary kernel-based Gaussian surrogate model. Our algorithm searches with a surrogate for optimal setting via a hyperparameter importance based evolutionary strategy, and the experiments demonstrate our algorithm outperforms manual tuning and several well-established hyperparameter optimization methods, including random search, grid Search, the Tree-structured Parzen Estimator Approach (TPE), Gaussian processes (GP) with stationary kernels, and the recently proposed Hyperparameter Optimization via RBF and Dynamic coordinate search (HORD).
IEEE journal of biomedical and health informatics, May 19, 2016
This paper presents a two-class electroencephalography (EEG)-based classification for classifying... more This paper presents a two-class electroencephalography (EEG)-based classification for classifying of driver fatigue (fatigue state vs. alert state) from 43 healthy participants. The system uses independent component by entropy rate bound minimization analysis (ERBM-ICA) for the source separation, autoregressive (AR) modeling for the features extraction and Bayesian neural network for the classification algorithm. The classification results demonstrate a sensitivity of 89.7%, a specificity of 86.8% and an accuracy of 88.2%. The combination of ERBM-ICA (source separator), AR (feature extractor) and Bayesian neural network (classifier) provides the best outcome with a p-value < 0.05 with the highest value of area under the receiver operating curve (AUC-ROC=0.93) against other methods such as power spectral density (PSD) as feature extractor (AUC-ROC=0.81). The results of this study suggest the method could be utilized effectively for a countermeasure device for driver fatigue identi...
2007 Australasian Universities Power Engineering Conference, 2007
This paper studies the effects of the presence of a current-sharing loop in a system of parallel-... more This paper studies the effects of the presence of a current-sharing loop in a system of parallel-connected dc/dc buck converters under democratic current-sharing control. Under this control scheme, the reference current is generated by a typical proportional- integral (PI) controller for voltage regulation and the currents in individual converters are programmed by a built-in interleaving switching control. Comparisons are made
IECON 2007 - 33rd Annual Conference of the IEEE Industrial Electronics Society, 2007
Abstract Border collision bifurcations have been given much attention in recent years. This paper... more Abstract Border collision bifurcations have been given much attention in recent years. This paper demonstrates an alternative method of analyzing border collision bifurcation, using a method called symbolic analysis. It shows how symbolic analysis can be used to ...
International Journal of Machine Learning and Computing, 2011
Preventive maintenance plays an important role in Heating, Ventilation and Air Conditioning (HVAC... more Preventive maintenance plays an important role in Heating, Ventilation and Air Conditioning (HVAC) system. One cost effective strategy is the development of analytic fault detection and isolation (FDI) module by online monitoring the key variables of HAVC systems. This paper investigates realtime FDI for HAVC system by using online Support Vector Machine (SVM), by which we are able to train a FDI system with manageable complexity under real time working conditions. It is also proposed a new approach which allows us to detect unknown faults and updating the classifier by using these previously unknown faults. Based on the proposed approach, a semi unsupervised fault detection methodology has been developed for HVAC systems
Proceedings of the International Joint Conference on Neural Networks, 2008
ecently, new kinds of neural networks known as the wavelet neural networks (WNNs), which combine ... more ecently, new kinds of neural networks known as the wavelet neural networks (WNNs), which combine feed-forward neural networks with the wavelet theory [2], have been proposed [3-5]. The wavelet theory provides a multi-resolution approximation for discriminate functions. The WNN can thus exhibit better performance in function learning than the conventional feed forward neural networks. Researchers have successfully applied WNNs in function approximation [1], robotics [4], and power systems [5]. Using neural networks to achieve learning [6] usually involves two steps: designing a network structure and deriving an algorithm for the learning process. The structure of the neural network governs the non-linearity of the modelled function. The learning algorithm determines the rules for optimizing the weight values of the network within the training period. A typical wavelet neural network structure offers a fixed set of weights after the learning process. This single set of weights is used to capture the characteristics of all input data. However, a fixed set of weights may not be enough to learn the data set if the data are distributed in a vast domain separately and/or the number of network parameters is too small. This also applies to a typical wavelet neural network structure.
Proceedings of the International Joint Conference on Neural Networks, 2009
To improve cancer diagnosis and drug development, the classification of tumor types based on geno... more To improve cancer diagnosis and drug development, the classification of tumor types based on genomic information is important. As DNA microarray studies produce a large amount of data, expression data are highly redundant and noisy, and most genes are believed to be uninformative with respect to the studied classes. Only a fraction of genes may present distinct profiles for different classes of samples. Classification tools to deal with these issues are thus important. These tools should learn to robustly identify a subset of informative genes embedded in a large dataset that is contaminated with high dimensional noises. In this paper, an integrated approach of support vector machine (SVM) and particle swarm optimization (PSO) is proposed for this purpose. The proposed approach can simultaneously optimize the selection of feature subset and the classifier through a common solution coding mechanism. As an illustration, the proposed approach is applied to search the combinational gene signatures for predicting histologic response to chemotherapy of osteosarcoma patients. Crossvalidation results show that the proposed approach outperforms other existing methods in terms of classification accuracy. Further validation using an independent dataset shows misclassification of only one out of fourteen patient samples, suggesting that the selected gene signatures can reflect the chemoresistance in osteosarcoma.
Conference proceedings : ... Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Annual Conference, 2011
For patients with Type 1 Diabetes Mellitus (T1DM), hypoglycemia is a very common but dangerous co... more For patients with Type 1 Diabetes Mellitus (T1DM), hypoglycemia is a very common but dangerous complication which can lead to unconsciousness, coma and even death. The variety of hypoglycemia symptoms is originated from the inadequate supply of glucose to the brain. In this study, we explore the connection between hypoglycemic episodes and the electrical activity of neurons within the brain or electroencephalogram (EEG) signals. By analyzing EEG signals from a clinical study of five children with T1DM, associated with hypoglycemia at night, we find that some EEG parameters change significantly under hypoglycemia condition. Based on these parameters, a method of detecting hypoglycemic episodes using EEG signals with a feed-forward multi-layer neural network is proposed. In our application, the classification results are 72% sensitivity and 55% specificity when the EEG signals are acquired from 2 electrodes C3 and O2. Furthermore, signals from different channels are also analyzed to o...
Abstract This book focuses on computational intelligence techniques and its applications-fast-gro... more Abstract This book focuses on computational intelligence techniques and its applications-fast-growing and promising research topics that have drawn a great deal of attention from researchers over the years. It brings together many different aspects of the current research on intelligence technologies such as neural networks, support vector machines, fuzzy logic and evolutionary computation, and covers a wide range of applications from pattern recognition and system modeling, to intelligent control problems and biomedical ...
The 2010 International Joint Conference on Neural Networks (IJCNN), 2010
Nocturnal hypoglycaemia in type 1 diabetic patients can be dangerous in which symptoms may not be... more Nocturnal hypoglycaemia in type 1 diabetic patients can be dangerous in which symptoms may not be apparent while blood glucose level decreases to very low level, and for this reason, an effective detection system for hypoglycaemia is crucial. This research work proposes a detection system for the hypoglycaemia based on the classification of electrocardiographic (ECG) parameters. The classification uses a Fuzzy Support Vector Machine (FSVM) with inputs of heart rate, corrected QT (QT c) interval and corrected TpTe (TpTe c) interval. Three types of kernel functions (radial basis function (RBF), exponential radial basis function (ERBF) and polynomial function) are investigated in the classification. Moreover, parameters of the kernel functions are tuned to find the optimum of the classification. The results show that the FSVM classification using RBF kernel function demonstrates better performance than using SVM. However, both classifiers result approximately same performance if ERBF and polynomial kernel functions are used.
2011 Annual International Conference of the IEEE Engineering in Medicine and Biology Society, 2011
Hypoglycemia is the most acute and common complication of Type 1 diabetes and is a limiting facto... more Hypoglycemia is the most acute and common complication of Type 1 diabetes and is a limiting factor in a glycemic management of diabetes. In this paper, two main contributions are presented; firstly, ventricular repolarization variabilities are introduced for hypoglycemia detection, and secondly, a swarm-based support vector machine (SVM) algorithm with the inputs of the repolarization variabilities is developed to detect hypoglycemia. By using the algorithm and including several repolarization variabilities as inputs, the best hypoglycemia detection performance is found with sensitivity and specificity of 82.14% and 60.19%, respectively.
2010 Annual International Conference of the IEEE Engineering in Medicine and Biology, 2010
Hypoglycaemia is a serious side effect of insulin therapy in patients with diabetes. We measure p... more Hypoglycaemia is a serious side effect of insulin therapy in patients with diabetes. We measure physiological parameters (heart rate, corrected QT interval of the electrocardiogram (ECG) signal) continuously to provide early detection of hypoglycemic episodes in Type 1 diabetes mellitus (T1DM) patients. Based on the physiological parameters, an evolved fuzzy reasoning model (FRM) to recognize the presence of hypoglycaemic episodes is developed. To optimize the fuzzy rules and the fuzzy membership functions of FRM, an evolutionary algorithm called hybrid particle swarm optimization with wavelet mutation operation is investigated. All data sets are collected from Department of Health, Government of Western Australia for a clinical study. The results show that the proposed algorithm performs well in terms of the clinical sensitivity and specificity.
This paper proposes a genetic algorithm (GA) based beamformer to optimize speech recognition accu... more This paper proposes a genetic algorithm (GA) based beamformer to optimize speech recognition accuracy for a pretrained speech recognizer. The proposed beamformer is designed to tackle the non-differentiable and non-linear natures of speech recognition by employing the GA algorithm to search for the optimal beamformer weights. Specifically, a population of beamformer weights is reproduced by crossover and mutation until the optimal beamformer weights are obtained. Results show that the speech recognition accuracies can be greatly improved even in noisy environments.
International Journal of Computational Intelligence and Applications, 2008
This paper presents a method on how to estimate main effects of gene representation. This estimat... more This paper presents a method on how to estimate main effects of gene representation. This estimate can be used not only to understand the domination of genes in the representation but also to design the mutation rate in genetic algorithms (GAs). A new approach of dynamic mutation rate is proposed by integrating the information of the main effects into the genes. By introducing the proposed method in GAs, both solution quality and solution stability can be improved in solving a set of parametrical test functions. The algorithm was applied to two illustrative applications to evaluate the performance of the proposed method, where the first application is on solving uncapacitated facility location problems and the next is on optimal power flow problems, which are employed. Results indicate that the proposed method yields significantly better results than the existing methods.
Objective. Sleep apnea significantly decreases the quality of life. The apnea hypopnea index (AHI... more Objective. Sleep apnea significantly decreases the quality of life. The apnea hypopnea index (AHI) is the main indicator for sleep apnea diagnosis. This study explored a novel automatic algorithm to diagnose sleep apnea from nasal airflow (AF) and pulse oximetry (SpO2) signals. Approach. Of the 988 polysomnography (PSG) records from the sleep heart health study (SHHS), 45 were randomly selected for the development of an algorithm and the remainder for validation (n = 943). The algorithm detects apnea events by a digitization process, following the determination of the peak excursion (peak-to-trough amplitude) from AF envelope. Hypopnea events were determined from the AF envelope and oxygen desaturation with correction to time lag in SpO2. Total sleep time (TST) was estimated from an optimized percentage of artefact-free total recording time. AHI was estimated from the number of detected events divided by the estimated TST. The estimated AHI was compared to the scored SHHS data for p...
IEEE Transactions on Evolutionary Computation, 2021
This paper investigates deep neural networks (DNNs) based lung nodule classification with hyperpa... more This paper investigates deep neural networks (DNNs) based lung nodule classification with hyperparameter optimization. Hyperparameter optimization in DNNs is a computationally expensive problem, and a surrogate-assisted evolutionary algorithm has been recently introduced to automatically search for optimal hyperparameter configurations of DNNs, by applying computationally efficient surrogate models to approximate the validation error function of hyperparameter configurations. Different from existing surrogate models adopting stationary covariance functions (kernels) to measure the difference between hyperparameter points, this paper proposes a non-stationary kernel that allows the surrogate model to adapt to functions whose smoothness varies with the spatial location of inputs. A multi-level convolutional neural network (ML-CNN) is built for lung nodule classification, and the hyperparameter configuration is optimized by the proposed non-stationary kernel-based Gaussian surrogate model. Our algorithm searches with a surrogate for optimal setting via a hyperparameter importance based evolutionary strategy, and the experiments demonstrate our algorithm outperforms manual tuning and several well-established hyperparameter optimization methods, including random search, grid Search, the Tree-structured Parzen Estimator Approach (TPE), Gaussian processes (GP) with stationary kernels, and the recently proposed Hyperparameter Optimization via RBF and Dynamic coordinate search (HORD).
IEEE journal of biomedical and health informatics, May 19, 2016
This paper presents a two-class electroencephalography (EEG)-based classification for classifying... more This paper presents a two-class electroencephalography (EEG)-based classification for classifying of driver fatigue (fatigue state vs. alert state) from 43 healthy participants. The system uses independent component by entropy rate bound minimization analysis (ERBM-ICA) for the source separation, autoregressive (AR) modeling for the features extraction and Bayesian neural network for the classification algorithm. The classification results demonstrate a sensitivity of 89.7%, a specificity of 86.8% and an accuracy of 88.2%. The combination of ERBM-ICA (source separator), AR (feature extractor) and Bayesian neural network (classifier) provides the best outcome with a p-value < 0.05 with the highest value of area under the receiver operating curve (AUC-ROC=0.93) against other methods such as power spectral density (PSD) as feature extractor (AUC-ROC=0.81). The results of this study suggest the method could be utilized effectively for a countermeasure device for driver fatigue identi...
2007 Australasian Universities Power Engineering Conference, 2007
This paper studies the effects of the presence of a current-sharing loop in a system of parallel-... more This paper studies the effects of the presence of a current-sharing loop in a system of parallel-connected dc/dc buck converters under democratic current-sharing control. Under this control scheme, the reference current is generated by a typical proportional- integral (PI) controller for voltage regulation and the currents in individual converters are programmed by a built-in interleaving switching control. Comparisons are made
IECON 2007 - 33rd Annual Conference of the IEEE Industrial Electronics Society, 2007
Abstract Border collision bifurcations have been given much attention in recent years. This paper... more Abstract Border collision bifurcations have been given much attention in recent years. This paper demonstrates an alternative method of analyzing border collision bifurcation, using a method called symbolic analysis. It shows how symbolic analysis can be used to ...
International Journal of Machine Learning and Computing, 2011
Preventive maintenance plays an important role in Heating, Ventilation and Air Conditioning (HVAC... more Preventive maintenance plays an important role in Heating, Ventilation and Air Conditioning (HVAC) system. One cost effective strategy is the development of analytic fault detection and isolation (FDI) module by online monitoring the key variables of HAVC systems. This paper investigates realtime FDI for HAVC system by using online Support Vector Machine (SVM), by which we are able to train a FDI system with manageable complexity under real time working conditions. It is also proposed a new approach which allows us to detect unknown faults and updating the classifier by using these previously unknown faults. Based on the proposed approach, a semi unsupervised fault detection methodology has been developed for HVAC systems
Proceedings of the International Joint Conference on Neural Networks, 2008
ecently, new kinds of neural networks known as the wavelet neural networks (WNNs), which combine ... more ecently, new kinds of neural networks known as the wavelet neural networks (WNNs), which combine feed-forward neural networks with the wavelet theory [2], have been proposed [3-5]. The wavelet theory provides a multi-resolution approximation for discriminate functions. The WNN can thus exhibit better performance in function learning than the conventional feed forward neural networks. Researchers have successfully applied WNNs in function approximation [1], robotics [4], and power systems [5]. Using neural networks to achieve learning [6] usually involves two steps: designing a network structure and deriving an algorithm for the learning process. The structure of the neural network governs the non-linearity of the modelled function. The learning algorithm determines the rules for optimizing the weight values of the network within the training period. A typical wavelet neural network structure offers a fixed set of weights after the learning process. This single set of weights is used to capture the characteristics of all input data. However, a fixed set of weights may not be enough to learn the data set if the data are distributed in a vast domain separately and/or the number of network parameters is too small. This also applies to a typical wavelet neural network structure.
Proceedings of the International Joint Conference on Neural Networks, 2009
To improve cancer diagnosis and drug development, the classification of tumor types based on geno... more To improve cancer diagnosis and drug development, the classification of tumor types based on genomic information is important. As DNA microarray studies produce a large amount of data, expression data are highly redundant and noisy, and most genes are believed to be uninformative with respect to the studied classes. Only a fraction of genes may present distinct profiles for different classes of samples. Classification tools to deal with these issues are thus important. These tools should learn to robustly identify a subset of informative genes embedded in a large dataset that is contaminated with high dimensional noises. In this paper, an integrated approach of support vector machine (SVM) and particle swarm optimization (PSO) is proposed for this purpose. The proposed approach can simultaneously optimize the selection of feature subset and the classifier through a common solution coding mechanism. As an illustration, the proposed approach is applied to search the combinational gene signatures for predicting histologic response to chemotherapy of osteosarcoma patients. Crossvalidation results show that the proposed approach outperforms other existing methods in terms of classification accuracy. Further validation using an independent dataset shows misclassification of only one out of fourteen patient samples, suggesting that the selected gene signatures can reflect the chemoresistance in osteosarcoma.
Conference proceedings : ... Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Annual Conference, 2011
For patients with Type 1 Diabetes Mellitus (T1DM), hypoglycemia is a very common but dangerous co... more For patients with Type 1 Diabetes Mellitus (T1DM), hypoglycemia is a very common but dangerous complication which can lead to unconsciousness, coma and even death. The variety of hypoglycemia symptoms is originated from the inadequate supply of glucose to the brain. In this study, we explore the connection between hypoglycemic episodes and the electrical activity of neurons within the brain or electroencephalogram (EEG) signals. By analyzing EEG signals from a clinical study of five children with T1DM, associated with hypoglycemia at night, we find that some EEG parameters change significantly under hypoglycemia condition. Based on these parameters, a method of detecting hypoglycemic episodes using EEG signals with a feed-forward multi-layer neural network is proposed. In our application, the classification results are 72% sensitivity and 55% specificity when the EEG signals are acquired from 2 electrodes C3 and O2. Furthermore, signals from different channels are also analyzed to o...
Abstract This book focuses on computational intelligence techniques and its applications-fast-gro... more Abstract This book focuses on computational intelligence techniques and its applications-fast-growing and promising research topics that have drawn a great deal of attention from researchers over the years. It brings together many different aspects of the current research on intelligence technologies such as neural networks, support vector machines, fuzzy logic and evolutionary computation, and covers a wide range of applications from pattern recognition and system modeling, to intelligent control problems and biomedical ...
The 2010 International Joint Conference on Neural Networks (IJCNN), 2010
Nocturnal hypoglycaemia in type 1 diabetic patients can be dangerous in which symptoms may not be... more Nocturnal hypoglycaemia in type 1 diabetic patients can be dangerous in which symptoms may not be apparent while blood glucose level decreases to very low level, and for this reason, an effective detection system for hypoglycaemia is crucial. This research work proposes a detection system for the hypoglycaemia based on the classification of electrocardiographic (ECG) parameters. The classification uses a Fuzzy Support Vector Machine (FSVM) with inputs of heart rate, corrected QT (QT c) interval and corrected TpTe (TpTe c) interval. Three types of kernel functions (radial basis function (RBF), exponential radial basis function (ERBF) and polynomial function) are investigated in the classification. Moreover, parameters of the kernel functions are tuned to find the optimum of the classification. The results show that the FSVM classification using RBF kernel function demonstrates better performance than using SVM. However, both classifiers result approximately same performance if ERBF and polynomial kernel functions are used.
2011 Annual International Conference of the IEEE Engineering in Medicine and Biology Society, 2011
Hypoglycemia is the most acute and common complication of Type 1 diabetes and is a limiting facto... more Hypoglycemia is the most acute and common complication of Type 1 diabetes and is a limiting factor in a glycemic management of diabetes. In this paper, two main contributions are presented; firstly, ventricular repolarization variabilities are introduced for hypoglycemia detection, and secondly, a swarm-based support vector machine (SVM) algorithm with the inputs of the repolarization variabilities is developed to detect hypoglycemia. By using the algorithm and including several repolarization variabilities as inputs, the best hypoglycemia detection performance is found with sensitivity and specificity of 82.14% and 60.19%, respectively.
2010 Annual International Conference of the IEEE Engineering in Medicine and Biology, 2010
Hypoglycaemia is a serious side effect of insulin therapy in patients with diabetes. We measure p... more Hypoglycaemia is a serious side effect of insulin therapy in patients with diabetes. We measure physiological parameters (heart rate, corrected QT interval of the electrocardiogram (ECG) signal) continuously to provide early detection of hypoglycemic episodes in Type 1 diabetes mellitus (T1DM) patients. Based on the physiological parameters, an evolved fuzzy reasoning model (FRM) to recognize the presence of hypoglycaemic episodes is developed. To optimize the fuzzy rules and the fuzzy membership functions of FRM, an evolutionary algorithm called hybrid particle swarm optimization with wavelet mutation operation is investigated. All data sets are collected from Department of Health, Government of Western Australia for a clinical study. The results show that the proposed algorithm performs well in terms of the clinical sensitivity and specificity.
This paper proposes a genetic algorithm (GA) based beamformer to optimize speech recognition accu... more This paper proposes a genetic algorithm (GA) based beamformer to optimize speech recognition accuracy for a pretrained speech recognizer. The proposed beamformer is designed to tackle the non-differentiable and non-linear natures of speech recognition by employing the GA algorithm to search for the optimal beamformer weights. Specifically, a population of beamformer weights is reproduced by crossover and mutation until the optimal beamformer weights are obtained. Results show that the speech recognition accuracies can be greatly improved even in noisy environments.
International Journal of Computational Intelligence and Applications, 2008
This paper presents a method on how to estimate main effects of gene representation. This estimat... more This paper presents a method on how to estimate main effects of gene representation. This estimate can be used not only to understand the domination of genes in the representation but also to design the mutation rate in genetic algorithms (GAs). A new approach of dynamic mutation rate is proposed by integrating the information of the main effects into the genes. By introducing the proposed method in GAs, both solution quality and solution stability can be improved in solving a set of parametrical test functions. The algorithm was applied to two illustrative applications to evaluate the performance of the proposed method, where the first application is on solving uncapacitated facility location problems and the next is on optimal power flow problems, which are employed. Results indicate that the proposed method yields significantly better results than the existing methods.
Uploads
Papers by Steve Ling