| Summary: | 博士 === 元智大學 === 機械工程學系 === 106 === Human physiological signals are critical in human body system. Bio-potential signals are produced by the electrochemical function. This signal can be electrocardiography (ECG) and electroencephalography (EEG). However, the raw ECG and EEG are mostly noisy and have hidden materials. Initially, the filtering algorithm need to be applied to the original signal. It is very important order to extract the information from the signal before the further evaluations. The empirical mode decomposition (EMD) filtering is utilized as the filtering algorithm. The fast Fourier transform (FFT) is one of the most evaluator for the signal processing. In further, the time frequency evaluation based-on the FFT can be one utilized for the short-time investigation. Other algorithms are entropy and detrended fluctuation analysis (DFA). Meanwhile, the DFA evaluates the integrated of the original signal to its local trends. In order to evaluate the given signal from our human body to the corresponding conditions, the classification algorithms should be selected.
This thesis implemented these algorithms to the several applications related the human physiological system. The first application is the anesthesia. This evaluation is conducted using the 63 patient surgical data. In this study, the EMD filter is applied for filtering the EEG signal and the sample entropy is used as the feature extraction algorithm. This sample entropy is combined to the intermittent 5-second data to train the artificial neural network (ANN) with the output from averaged consciousness level given by 5 doctors. The main achievement of this application is the important of the electromyography (EMG) signal from the sensitivity analysis. This also help the classification result.
The second application is the evaluation of the cardiopulmonary resuscitation (CPR) in the emergency medical system (EMS). This investigation is conducted from about a thousand asystole patient data. The EMD is used for the initial stage for the filtering of the CPR signal. The time frequency evaluation is utilized to display the corresponding signal to its frequency in a time axis evaluation. This application also uses the sample entropy, multiscale entropy and complexity index (CI) in order to evaluate the quality of the CPR given to the asystole patients. From this application, we found that the CI provides a significant difference result from the younger patients. This result may be applied for designing the automated CPR machine with the dynamic force.
The third application from this study is the arrhythmia evaluation in wearable device. This research is conducted using four PhysioNet databases and is evaluated using the wearable device and the smartphone. The evaluated classes are the atrial premature complex (APC), ventricular premature complex (VPC), atrial fibrillation (AF) and ventricular fibrillation (VF) based on ANSI/AAMI EC57:2012. The APC and VPC are the beat-based arrhythmia. Meanwhile, AF and VF are the rhythm-based arrhythmia. From this study, a less complexity algorithms are selected in order to fit the device and the smartphone specifications. For this application, we conclude that our integrated algorithm detection can achieve a good accuracy in comparison to other previous studies.
The last application is ensemble genetic fuzzy neuro model applied for the emergency medical service via unbalanced data evaluation. The raw data after the first filter is used consists of 4,408,187 patient datapoints. A linguistic algorithm is applied to evaluate the input and output relationship, namely Fuzzy c-Means (FCM), which is applied as a clustering algorithm for the majority class to balance the minority class data. Each cluster is used to train several ANN models. Different techniques are applied to generate an ensemble genetic fuzzy neuro model (EGFNM) in order to select the models. The first ensemble technique, the intra-cluster EGFNM, works by evaluating the best combination from all the models generated by each cluster. Another ensemble technique is the inter-cluster model EGFNM, which is based on selecting the best model from each cluster. The accuracy of these techniques is evaluated using the receiver operating characteristic (ROC) via its area under the curve (AUC). For the result, this study achieved improved results by performing the EGFNM method compared with the unbalanced training. This study concludes that selecting several best models will produce a better result compared with all models combined.
In summary, for bio-signal processing, the EMD-filter is important to raw physiological signal purification. ANN, one of the intelligent systems, produces an acceptable result for some applications of the bio-signal processing. In further, the feature extraction methods, entropy-based algorithms, DFA and FFT, display some characteristics from the signal that may helpful for the classification or further evaluation. Meanwhile, for the unbalanced data modelling, EGFNM is developed for the model evaluation. This method is useful for the ensemble technique in selecting a set of the models for the future detection.
|
|---|
