Wei-Hsin Chen

High-Performance Seizure Detection System Using a Wavelet-Approximate Entropy-fSVM Cascade With Clinical Validation

The classification of electroencephalography (EEG) signals is one of the most important methods for seizure detection. However, verification of an atypical epileptic seizure often can only be done through long-term EEG monitoring for 24 hours or longer. Hence, automatic EEG signal analysis for clinical screening is necessary for the diagnosis of epilepsy. We propose an EEG analysis system of seizure detection, based on a cascade of wavelet-approximate entropy for feature selection, Fisher scores for adaptive feature selection, and support vector machine for feature classification. Performance of the system was tested on open source data, and the overall accuracy reached 99.97%. We further tested the performance of the system on clinical EEG obtained from a clinical EEG laboratory and bedside EEG recordings. The results showed an overall accuracy of 98.73% for routine EEG, and 94.32% for bedside EEG, which verified the high performance and usefulness of such a cascade system for seizure detection. Also, the prediction model, trained by routine EEG, can be successfully generalized to bedside EEG of independent patients.

A Data-Mining Framework for Transnational Healthcare System

Medical resources are important and necessary in health care. Recently, the development of methods for improving the efficiency of medical resource utilization is an emerging problem. Despite evidence supporting the use of order sets in hospitals, only a small number of health information systems have successfully equipped physicians with analysis of complex order sequences from clinical pathway and clinical guideline. This paper presents a data-mining framework for transnational healthcare system to find alternative practices, including transfusion, pre-admission tests, and evaluation of liver diseases. However, individual countries vary with respect to geographical location, living habits, and culture, so disease risks and treatment methods also vary across countries. To realize the difference, a service-oriented architecture and cloud-computing technology are applied to analyze these medical data. The validity of the proposed system is demonstrated in including Taiwan and Mongolia, to ensure the feasibility of our approach.

Bio-signal analysis system design with support vector machines based on cloud computing service architecture

Today, many bio-signals such as Electroencephalography (EEG) are recorded in digital format. It is an emerging research area of analyzing these digital bio-signals to extract useful health information in biomedical engineering. In this paper, a bio-signal analyzing cloud computing architecture, called BACCA, is proposed. The system has been designed with the purpose of seamless integration into the National Taiwan University Health Information System. Based on the concept of. NET Service Oriented Architecture, the system integrates heterogeneous platforms, protocols, as well as applications. In this system, we add modern analytic functions such as approximated entropy and adaptive support vector machine (SVM). It is shown that the overall accuracy of EEG bio-signal analysis has increased to nearly 98% for different data sets, including open-source and clinical data sets.

Persistent photoconductivity in InxAlyGa1−x−yN quaternary alloys

The optical properties of InxAlyGa1−x−yN quaternary alloys were investigated by photoconductivity (PC), persistent photoconductivity (PPC), photoluminescence (PL), and photoluminescence excitation (PLE) measurements. Quite interestingly, persistent photoconductivity was observed. Through the combination of our optical studies, we show that the PPC effect arises from composition fluctuations in InxAlyGa1−x−yN quaternary alloys. From the analysis of the decay kinetics, the localization depth caused by composition fluctuations was determined. A comparison between the PL, PLE, and PC measurements gives a direct access to the Stokes’ shift. The Stokes’ shift can be explained in terms of localization due to the existence of nanoscale clusters, and it is consistent with the PPC result. The results shown here provide concrete evidence to support our previously proposed model that the existence of InGaN-like clusters is responsible for the strong luminescence in InxAlyGa1−x−yN quaternary alloys.

Web-based newborn screening system for metabolic diseases: machine learning versus clinicians.

Background A hospital information system (HIS) that integrates screening data and interpretation of the data is routinely requested by hospitals and parents. However, the accuracy of disease classification may be low because of the disease characteristics and the analytes used for classification. Objective The objective of this study is to describe a system that enhanced the neonatal screening system of the Newborn Screening Center at the National Taiwan University Hospital. The system was designed and deployed according to a service-oriented architecture (SOA) framework under the Web services .NET environment. The system consists of sample collection, testing, diagnosis, evaluation, treatment, and follow-up services among collaborating hospitals. To improve the accuracy of newborn screening, machine learning and optimal feature selection mechanisms were investigated for screening newborns for inborn errors of metabolism. Methods The framework of the Newborn Screening Hospital Information System (NSHIS) used the embedded Health Level Seven (HL7) standards for data exchanges among heterogeneous platforms integrated by Web services in the C# language. In this study, machine learning classification was used to predict phenylketonuria (PKU), hypermethioninemia, and 3-methylcrotonyl-CoA-carboxylase (3-MCC) deficiency. The classification methods used 347,312 newborn dried blood samples collected at the Center between 2006 and 2011. Of these, 220 newborns had values over the diagnostic cutoffs (positive cases) and 1557 had values that were over the screening cutoffs but did not meet the diagnostic cutoffs (suspected cases). The original 35 analytes and the manifested features were ranked based on F score, then combinations of the top 20 ranked features were selected as input features to support vector machine (SVM) classifiers to obtain optimal feature sets. These feature sets were tested using 5-fold cross-validation and optimal models were generated. The datasets collected in year 2011 were used as predicting cases. Results The feature selection strategies were implemented and the optimal markers for PKU, hypermethioninemia, and 3-MCC deficiency were obtained. The results of the machine learning approach were compared with the cutoff scheme. The number of the false positive cases were reduced from 21 to 2 for PKU, from 30 to 10 for hypermethioninemia, and 209 to 46 for 3-MCC deficiency. Conclusions This SOA Web service–based newborn screening system can accelerate screening procedures effectively and efficiently. An SVM learning methodology for PKU, hypermethioninemia, and 3-MCC deficiency metabolic diseases classification, including optimal feature selection strategies, is presented. By adopting the results of this study, the number of suspected cases could be reduced dramatically.

Integrating Human Genome Database into Electronic Health Record with Sequence Alignment and Compression Mechanism

With the initial completion of Human Genome Project, the post-genomic era is coming. Although the genome map of human has been decoded, the roles that each segment of sequences acts are not totally discovered. On the other hand, with the rapid expansion of sequence information, the issues of data compilation and data storage are increasingly important. In this paper, a “Human genome database system” is designed and implemented in National Taiwan University Hospital (NTUH). By accessing this system, the doctors can store and manage the experimental sequence data. The achievement of this system is that it integrates the modules of sequence alignment and data compression. By embedding with the NCBI alignment program—blastall [1], it automatically aligns the uploaded sequences and searches for the corresponding genomic positions. Besides, the system encodes the differences between sequences, effectively compresses them and decreases the demand of storage spaces by the compression ratio at 12.28. At the same time, it offers a variety of query methods. Users can quickly access the interesting data by inputting the keywords of specimen number, GI and sequence position, etc. The electronic health record (EHR) in Health Information System (HIS) of NTUH is also combined in this system and the doctors can utilize the valuable information to figure out the relation between the diseases and genes. With this system, a genetic personal healthcare environment will be established in the future.

A web-based telehealthcare system with mobile application and data analysis for diet people

A web-based telehealthcare system was integrated with the mobile application for diet people. In order to comprehensively deal with the over-weight problem, the user interface with four important aspects on weight, diet, exercise and sleep records is implemented. In addition, online courses and knowledge of diet are provided in the system. The electronic medical record of National Taiwan University Hospital is also embedded in the system for the case managers and doctors to evaluate the status of the over-weight patients. We utilized Pearsons Correlation Coefficient which represented the correlative degree of two variables to discover the relationship between the reduced weight of people and their lifestyle records. We found that three aspects of records (compliancy, violation, exercise) were important for people who wanted to lose weight.

Newborn Screening System Based on Adaptive Feature Selection and Support Vector Machines

The clinical symptoms of metabolic disorders during neonatal period are often not apparent, if not treated early irreversible damages such as mental retardation may occur, even death. Therefore, practicing newborn screening is very important to prevent neonatal from these damages. In this paper, the newborn screening system used support vector machines (SVM) classification technique is proposed in place of cut-off value decision to evaluate the metabolic substances concentration raw data obtained from tandem mass spectrometry (MS/MS) and determine whether the newborn has some kinds of metabolic disorder diseases. On the basis of the proposed features, new analytic combinations are identified with superior discriminatory performance compared with the best published combinations. Classifiers built with the feature selection to find C3/C2, C3 and C16 of three key point features achieved diagnostic sensitivities, specificities and accuracy approaching 100%.

Web Services Based Bio-signal System Leveraging Support Vector Machines

Bio-signal analysis is one of the most important approaches to biomedical engineering. The health information such as ECG, PCG, EMG and EEG are often recorded in digital format to be analyzed. In this paper, a bio-signal analyzing web service system using Support Vector Machines (SVM) classifier technique is proposed. The bio-signals are recorded in digital format as the input of the system. In addition, based on the concept of Service-Oriented Architecture (SOA), the system under web services .NET is designed. The system provides features including heterogeneous platform, protocols, as well as applications. The system has been designed with the purpose of seamless integration into the Health Information System.

Newborn Screening for Phenylketonuria: Machine Learning vs Clinicians

The metabolic disorders may hinder an infants normal physical or mental development during the neonatal period. The metabolic diseases can be treated by effective therapies if the diseases are discovered in the early stages. Therefore, newborn screening program is essential to prevent neonatal from these damages. In the paper, a support vector machine (SVM) based algorithm is introduced in place of cut-off value decision to evaluate the analyte elevation raw data associated with Phenylketonuria. The data were obtained from tandem mass spectrometry (MS/MS) for newborns. In addition, a combined feature selection mechanism is proposed to compare with the cut-off scheme. By adapting the mechanism, the number of suspected cases is reduced substantially, it also handles the medical resources effectively and efficiently.