Yu-te Wu - Academia.edu (original) (raw)

Papers by Yu-te Wu

European radiology, May 22, 2024

Purpose To compare the diagnostic performance of standalone deep learning (DL) algorithms and hum... more Purpose To compare the diagnostic performance of standalone deep learning (DL) algorithms and human experts in lung cancer detection on chest computed tomography (CT) scans. Materials and methods This study searched for studies on PubMed, Embase, and Web of Science from their inception until November 2023. We focused on adult lung cancer patients and compared the efficacy of DL algorithms and expert radiologists in disease diagnosis on CT scans. Quality assessment was performed using QUADAS-2, QUADAS-C, and CLAIM. Bivariate random-effects and subgroup analyses were performed for tasks (malignancy classification vs invasiveness classification), imaging modalities (CT vs low-dose CT [LDCT] vs high-resolution CT), study region, software used, and publication year. Results We included 20 studies on various aspects of lung cancer diagnosis on CT scans. Quantitatively, DL algorithms exhibited superior sensitivity (82%) and specificity (75%) compared to human experts (sensitivity 81%, specificity 69%). However, the difference in specificity was statistically significant, whereas the difference in sensitivity was not statistically significant. The DL algorithms' performance varied across different imaging modalities and tasks, demonstrating the need for tailored optimization of DL algorithms. Notably, DL algorithms matched experts in sensitivity on standard CT, surpassing them in specificity, but showed higher sensitivity with lower specificity on LDCT scans. Conclusion DL algorithms demonstrated improved accuracy over human readers in malignancy and invasiveness classification on CT scans. However, their performance varies by imaging modality, underlining the importance of continued research to fully assess DL algorithms' diagnostic effectiveness in lung cancer. Clinical relevance statement DL algorithms have the potential to refine lung cancer diagnosis on CT, matching human sensitivity and surpassing in specificity. These findings call for further DL optimization across imaging modalities, aiming to advance clinical diagnostics and patient outcomes. Key Points • Lung cancer diagnosis by CT is challenging and can be improved with AI integration. • DL shows higher accuracy in lung cancer detection on CT than human experts. • Enhanced DL accuracy could lead to improved lung cancer diagnosis and outcomes.

Frontiers in Neurology, 2022

ObjectivesTo identify the neuroimaging predictors for the responsiveness of patients to sumatript... more ObjectivesTo identify the neuroimaging predictors for the responsiveness of patients to sumatriptan and use an independent cohort for external validation.MethodsStructuralized headache questionnaire and 3-Tesla brain magnetic resonance imaging were performed in migraine patients. Regional brain volumes were automatically calculated using FreeSurfer version 6.0, including bilateral amygdala, anterior cingulated cortex, caudate, putamen, precuneus, orbitofrontal cortex, superior frontal gyri, middle frontal gyri, hippocampus, and parahippocampus. A sumatriptan-responder was defined as headache relief within 2 h after the intake of sumatriptan in at least two out of three treated attacks. We constructed a prediction model for sumatriptan response using the regional brain volume and validated it with an independent cohort of migraine patients.ResultsA total of 105 migraine patients were recruited, including 73 sumatriptan responders (69.5%) and 32 (30.5%) non-responders. We divided the ...

Entropy, Mar 29, 2022

This article is an open access article distributed under the terms and conditions of the Creative... more This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY

In this study, we aimed to use the motor-imagery electroencephalography (EEG) signal to construct... more In this study, we aimed to use the motor-imagery electroencephalography (EEG) signal to construct a Brain Computer Interface (BCI) system. We developed an EEG-based real-time cursor control system on LabVIEW platform. EEG signals of left or right motor imagery on C3, and C4 channels were collected using OpenBCI amplifier system. The experimental protocol consisted of a training stage and a self-controlled stage with several runs in each stage. In the training stage, EEG signals were collected while subjects were asked to look at the moving cursor with a constant speed toward left/right, and pretended the cursor was controlled by their imagination. In the self-controlled stage, the movement of the cursor was controlled based on the concordance between the classification results and preset direction. Ten subjects were enrolled in this study. We found that the classification rate was associated with the consistency of the individual ERD/ERS frequency bands across different runs. If a subject presented a stable frequency band for the motor imaginary, the classification rate of the developed BCI system can reach a satisfactory performance with the highest classification rate of 93%. In conclusion, our results showed that the efficacy of our BCI system highly relied on the stability of the individual frequency pattern.

Long-term aerobic exercise can effectively improve the heart and lung function, stabilize mood an... more Long-term aerobic exercise can effectively improve the heart and lung function, stabilize mood and reduce the incidence of cardiopulmonary diseases. Brain activity can properly reflect physical and mental status of subjects during prolonged exercise, and long-term exercise may affect the power spectrum of EEG. Many studies showed that ECG, EMG and EEG, can effectively and accurately assess the status change during exercise. Safety and efficiency are the main concerns for promoting the aerobics for the elder. In this study, we aim to investigate the EEG and ECG features that can reveal status change during cycling exercise. Twenty-nine healthy subjects participated in this study. After four-minute resting stage, participants were asked to take cycling exercise continuously for twenty minutes, and the EEG, ECG signals were recorded and analyzed. The EEG data were divided into one-minute epoch and the wavelet transform was used to analyze five frequency bands, namely, theta (T), low alpha (LA), high alpha (HA), low beta (LB) and high beta (HB). The ECG signal was used to establish the average maximum heart rate ratio (AMHRR) and cardiac stress index (CSI). We found variations of RR intervals decreases during sustained cycling exercise. The CSI plot of a less frequent exerciser showed steeper than a frequent exerciser. If a participant has a steeper slope of CSI curve may imply an increase in cardiac stress. The AMHRR score at 65% could be a threshold for the occurrence of feeling hard during exercise. The CSI, HA and LB are the most proper features for assessing status change during exercise.

Expert Systems With Applications, Sep 1, 2023

In this study, we used two machine learning algorithms, namely, linear support vector machine (SV... more In this study, we used two machine learning algorithms, namely, linear support vector machine (SVM) and convolutional neural network (CNN), to classify the BCI (Brain Computer interface) competition IV-2a 2-class MI (motor imagery) data set which consists of EEG data from 9 subjects. For each subject, 5 sessions of signals from three electrodes (C3, Cz, and C4) were recorded with sampling rate 250Hz. The training data, which consisted of the first 3 sessions, included 400 trials. The evaluation data, which consisted of the last 2 sessions, included 320 trials. Each trial started with gazing at fix cross on screen for 3 seconds followed by a one-second visual cue pointing either to the left or right to instruct the subject for left or right motor imagery over a period of 4 seconds, and then followed by a short break of at least 1.5 seconds. Features were extracted from the 0.5 to 2.5 second signals after the cue for each trial from C3 and C4. Each EEG trial was band pass filtered into different frequency bands, namely, delta (0.5-3Hz), theta (4-8Hz), alpha (8-12Hz), beta bands (13-30Hz), gamma bands (31-60Hz). Those filtered signals were then used as the input data for training the linear SVM. In addition, we generated a 2 by 500 matrix by down sampling the training data from each trial. There are 5760 such matrices in total generated from all subjects and serve as the input data for training CNN and the trained model was evaluated by another 340 matrices from each subject. Our CNN architecture consisted of 2 convolution layer and 2 fully connect layers, and there was a batch normalization layer before the activated layer and a dropout layer with a probability of 50% after the activated layer. The classification accuracies evaluated by averaged kappa values obtained from linear SVM and CNN are 0.5 and 0.621, respectively, suggesting the deep learning CNN method is superior to the classical linear SVM on the EEG classification.

IEEE Transactions on Industrial Electronics, Jul 1, 2013

This paper proposes an asynchronous steady-state visual evoked potential (SSVEP)-based hospital b... more This paper proposes an asynchronous steady-state visual evoked potential (SSVEP)-based hospital bed nursing system without training from the users. The proposed system can be easily applied to any kinds of electrical hospital beds and can help the user to control the attitude of the hospital bed by only staring at the flashing panel. Different from most brain-computer interface (BCI) systems consisting of commercial instruments, this study provides a total design solution. The system firstly presents a light-emitting-diode stimulation panel to induce the user's SSVEP signal used as the input signal of the proposed system. Then, an SSVEP-amplifier/filter circuit and a field-programmable-gate-array-based SSVEP signal processor are respectively designed to acquire and process the subject's SSVEP. Moreover, H-bridge dc motor drive circuit is implemented to adjust the attitude of the hospital bed. Eventually, 15 subjects are invited to demonstrate the effectiveness of the proposed BCI-based hospital bed control nursing system. The total design BCI system shows good performance with an average accuracy of 92.5% and an average command transfer interval of 5.22 s per command.

Computer Methods and Programs in Biomedicine, Feb 1, 2023

Springer optimization and its applications, 2010

An electroencephalogram (EEG) is commonly used to study changes in brain activity during exercise... more An electroencephalogram (EEG) is commonly used to study changes in brain activity during exercise. In this study, we used recorded electrocardiography to define the average-to-maximal heart rate ratio (AMHRR) as a gauge of relative exercise workload, and explored how an increase in the AMHRR affected brain activity. EEG signals were recorded from 44 healthy subjects at four scalp sites (i.e., C1, C2, P1, and P2) during 9-min nonstop cycling exercise. The relationships among the AMHRR, EEG spectral power, and EEG fuzzy entropy (FuzzyEn) were established. The EEG spectral power and FuzzyEn results displayed similarly increasing patterns with the AMHRR at all electrodes. However, FuzzyEn exhibited superior specificity in selecting effective frequency bands (i.e., theta, alpha, and beta). The FuzzyEn method can be applied in a wearable device for a human machine interface, which can monitor the EEG during exercise.

IFMBE proceedings, 2009

Multiple system atrophy-cerebellar (MSA-C) is a degenerative neurological disease of brain. In th... more Multiple system atrophy-cerebellar (MSA-C) is a degenerative neurological disease of brain. In this study, we used three-dimensional (3D) fractal dimension (FD) method to investigate the structural complexity change of human cerebellum white matter (CBWM) and gray matter (CBGM) for MSA-C disease diagnosis. The original T-1 weighted magnetic resonance (MR) images of twenty-three MSA-C patients and twenty-one normal subjects were processed and the 3D CBGM and 3D CBGM were quantitatively analysed by the FD method. Results showed that FD values of the CBWM and CBGM of the MSA-C patients were significantly smaller comparing with the control group. Furthermore, the results of this study also demonstrated that the FD method is superior to the conventionally volumetric method in terms of better accuracy and sensitivity. The use of FD analysis in conjunction with the conventionally volumetric methods further allows us to establish a separating line in the FD-volume scatter plot of CBWM to identify the occurrence of cerebellum atrophy, which can serve as an early-stage indicator for detecting the MSA-C disease.

In this study, we employed Morlet wavelet, sample entropy, and fractal dimension on EEG and EMG s... more In this study, we employed Morlet wavelet, sample entropy, and fractal dimension on EEG and EMG signal to extract the feature of physical fatigue in the exercise. The result may be helpful for rehabilitation in effectiveness evaluation. Twenty healthy subjects participated in cycling exercise, and their physiological signals, including EEG, EMG, and ECG were recorded. In addition, we recorded subjects' feeling of fatigue since each subject has different physical strength and tolerance of non-stopping exercise. Signals in different stages, namely, resting, early, middle and late stages of exercising, were analyzed. ECG signal was used to categorize subjects into two groups, namely, moderate fatigue and severe fatigue. In EEG results, the averaged power, sample entropy, and fractal dimension of signals indicated that resting stages before and after the exercise were distinct from exercising stage. In severe fatigue, the averaged power within each frequency band of EEG increased with the duration of exercise whereas the power ratio, denoted by (theta+ alpha)/ beta, decreased gradually from the beginning of exercise until the resting after exercise. In addition, the EEG (C3) results of SE complexity ratio and FD complexity ratio decreased gradually from resting to last session of exercise in the moderate fatigue whereas in severe fatigue these ratios increased at the late exercising stage. Our results demonstrate that different patterns between moderate fatigue and severe fatigue can be effectively extracted by using the proposed methods.

Cancer Imaging

Background The epidermal growth factor receptor (EGFR) tyrosine kinase inhibitors (TKIs) are a fi... more Background The epidermal growth factor receptor (EGFR) tyrosine kinase inhibitors (TKIs) are a first-line therapy for non-small cell lung cancer (NSCLC) with EGFR mutations. Approximately half of the patients with EGFR-mutated NSCLC are treated with EGFR-TKIs and develop disease progression within 1 year. Therefore, the early prediction of tumor progression in patients who receive EGFR-TKIs can facilitate patient management and development of treatment strategies. We proposed a deep learning approach based on both quantitative computed tomography (CT) characteristics and clinical data to predict progression-free survival (PFS) in patients with advanced NSCLC after EGFR-TKI treatment. Methods A total of 593 radiomic features were extracted from pretreatment chest CT images. The DeepSurv models for the progression risk stratification of EGFR-TKI treatment were proposed based on CT radiomic and clinical features from 270 stage IIIB-IV EGFR-mutant NSCLC patients. Time-dependent PFS pred...

2017 International Automatic Control Conference (CACS), 2017

Bipolar disorder and schizophrenia are two prevailing psychiatric disorders with significant over... more Bipolar disorder and schizophrenia are two prevailing psychiatric disorders with significant overlaps in symptoms, abnormalities, and disease progression. Therefore, it is difficult to differentiate these two diseases without repeated clinical visits. Previous studies demonstrated high accuracy of classification for bipolar disorder and schizophrenia at the individual level by functional connectivity, but few studies focused on classifying between these two diseases directly. In order to assist diagnosis, we investigated further the feasibility of classifying bipolar disorder and schizophrenia by the structure of functional networks. The results revealed 90.0% accuracy of the classification with the sensitivity 1.0 and the specificity 0.80 for the patients with bipolar disorder. The present study indicated that the differences between the characteristics of brain network structures in bipolar disorder and schizophrenia could be the reliable features for the classification and may be the diagnostic indicators in the future.

IFMBE Proceedings

This study aims to recover the somatosensory evoked potentials (SSEPs) from the smearing electroe... more This study aims to recover the somatosensory evoked potentials (SSEPs) from the smearing electroencephalography (EEG) recordings using independent component analysis (ICA) in conjunction with the proposed timefrequency SSEP template (TF-SSEP). The SSEPs induced from patients with the impaired motor functions exhibit longer latency and lower amplitude than the normal SSEPs and are inevitably contaminated by artifacts and environmental noise. Although ICA has been demonstrated as a novel technique to segregate the EEG into independent sources, the selection of task-related components needs to be further elaborated. The TF-SSEP template, generated by the Morelet wavelet transformation of the averaged SSEPs from three normal subjects, was used to automatically extract the SSEP-related features. The performance of the TF-SSEP template was further validated using EEGs through the left and right peroneal nerve stimulation of four stroke patients. After ICA decomposition, the sources were selected for reconstruction if their correlation coefficients with the TF-SSEP template were higher than the predetermined threshold. On the other hand, the unselected sources were considered as the event-unrelated components or artifacts. Among all patients, the topography maps at four peak times, namely P40, N45, P60 and N75, showed higher contrast in the vicinity of the foot-associated motor area, and the resolved SSEPs demonstrated uncontaminated waveforms in comparison with the conventionally averaging method. This indicated that the proposed method can remarkably suppress artifacts and effectively extracted the SSEP-related features.

Medicine, 2016

In this study, we aimed to investigate the reactive changes in diffusion tensor imaging (DTI)-der... more In this study, we aimed to investigate the reactive changes in diffusion tensor imaging (DTI)-derived diffusion metrics of the anterior thalamic nucleus (AN), a relaying center for the Papez circuit, in early idiopathic normal pressure hydrocephalus (iNPH) patients with memory impairment, as well as its correlation with the patients' neuropsychological performances. In total, 28 probable iNPH patients with symptom onset within 1 year and 17 control subjects were prospectively recruited between 2010 and 2013 for this institutional review board-approved study. Imaging studies including DTI and a neuropsychological assessment battery were performed in all subjects. Diffusion metrics were measured from the region of the AN using tract-deterministic seeding method by reconstructing the mammillo-thalamo-cingulate connections within the Papez circuit. Differences in diffusion metrics and memory assessment scores between the patient and control group were examined via the Mann-Whitney U...

IFMBE proceedings, 2009

Multiple system atrophy-cerebellar (MSA-C) is a degenerative neurological disease of brain. In th... more Multiple system atrophy-cerebellar (MSA-C) is a degenerative neurological disease of brain. In this study, we used three-dimensional (3D) fractal dimension (FD) method to investigate the structural complexity change of human cerebellum white matter (CBWM) and gray matter (CBGM) for MSA-C disease diagnosis. Twenty patients and twenty-three normal subjects as the control group participated in this study. The T-1 weighted magnetic resonance (MR) images were processed and 3D CBGM and 3D CBGM were analyzed by the FD method. Results demonstrated that the FD values of patients’ CBWM and CBGM decreased significantly, and that their CBWM FD values decreased more significant than that of CBGM. Results also showed that the 3D FD method was superior to the conventional volumetric method in terms of better reliability, accuracy, and sensitivity.

Medicine, 2016

The studies regarding to the comparisons between major depressive disorder (MDD) and panic disord... more The studies regarding to the comparisons between major depressive disorder (MDD) and panic disorder (PD) in the microintegrity of white matter (WM) are uncommon. Therefore, we tried to a way to classify the MDD and PD. Fifty-three patients with 1st-episode medication-naive PD, 54 healthy controls, and 53 patients with 1st-episode medication-naive MDD were enrolled in this study. The controls and patients were matched for age, gender, education, and handedness. The diffusion tensor imaging scanning was also performed. The WM microintegrity was analyzed and compared between 3 groups of participants (ANOVA analysis) with age and gender as covariates. The MDD group had lower WM microintegrity than the PD group in the left anterior thalamic radiation, left uncinate fasciculus, left inferior fronto-occipital fasciculus, and bilateral corpus callosum. The MDD group had reductions in the microintegrity when compared to controls in the bilateral superior longitudinal fasciculi, inferior longitudinal fasciculi, inferior fronto-occipital fasciculi, and corpus callosum. The PD group had lower microintegrity in bilateral superior longitudinal fasciculi and left inferior fronto-occipital fasciculus when compared to controls. The widespread pattern of microintegrity alterations in frontolimbic WM circuit for MDD was different from restrictive pattern of alterations for PD.

European radiology, May 22, 2024

Purpose To compare the diagnostic performance of standalone deep learning (DL) algorithms and hum... more Purpose To compare the diagnostic performance of standalone deep learning (DL) algorithms and human experts in lung cancer detection on chest computed tomography (CT) scans. Materials and methods This study searched for studies on PubMed, Embase, and Web of Science from their inception until November 2023. We focused on adult lung cancer patients and compared the efficacy of DL algorithms and expert radiologists in disease diagnosis on CT scans. Quality assessment was performed using QUADAS-2, QUADAS-C, and CLAIM. Bivariate random-effects and subgroup analyses were performed for tasks (malignancy classification vs invasiveness classification), imaging modalities (CT vs low-dose CT [LDCT] vs high-resolution CT), study region, software used, and publication year. Results We included 20 studies on various aspects of lung cancer diagnosis on CT scans. Quantitatively, DL algorithms exhibited superior sensitivity (82%) and specificity (75%) compared to human experts (sensitivity 81%, specificity 69%). However, the difference in specificity was statistically significant, whereas the difference in sensitivity was not statistically significant. The DL algorithms' performance varied across different imaging modalities and tasks, demonstrating the need for tailored optimization of DL algorithms. Notably, DL algorithms matched experts in sensitivity on standard CT, surpassing them in specificity, but showed higher sensitivity with lower specificity on LDCT scans. Conclusion DL algorithms demonstrated improved accuracy over human readers in malignancy and invasiveness classification on CT scans. However, their performance varies by imaging modality, underlining the importance of continued research to fully assess DL algorithms' diagnostic effectiveness in lung cancer. Clinical relevance statement DL algorithms have the potential to refine lung cancer diagnosis on CT, matching human sensitivity and surpassing in specificity. These findings call for further DL optimization across imaging modalities, aiming to advance clinical diagnostics and patient outcomes. Key Points • Lung cancer diagnosis by CT is challenging and can be improved with AI integration. • DL shows higher accuracy in lung cancer detection on CT than human experts. • Enhanced DL accuracy could lead to improved lung cancer diagnosis and outcomes.

Frontiers in Neurology, 2022

ObjectivesTo identify the neuroimaging predictors for the responsiveness of patients to sumatript... more ObjectivesTo identify the neuroimaging predictors for the responsiveness of patients to sumatriptan and use an independent cohort for external validation.MethodsStructuralized headache questionnaire and 3-Tesla brain magnetic resonance imaging were performed in migraine patients. Regional brain volumes were automatically calculated using FreeSurfer version 6.0, including bilateral amygdala, anterior cingulated cortex, caudate, putamen, precuneus, orbitofrontal cortex, superior frontal gyri, middle frontal gyri, hippocampus, and parahippocampus. A sumatriptan-responder was defined as headache relief within 2 h after the intake of sumatriptan in at least two out of three treated attacks. We constructed a prediction model for sumatriptan response using the regional brain volume and validated it with an independent cohort of migraine patients.ResultsA total of 105 migraine patients were recruited, including 73 sumatriptan responders (69.5%) and 32 (30.5%) non-responders. We divided the ...

Entropy, Mar 29, 2022

This article is an open access article distributed under the terms and conditions of the Creative... more This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY

In this study, we aimed to use the motor-imagery electroencephalography (EEG) signal to construct... more In this study, we aimed to use the motor-imagery electroencephalography (EEG) signal to construct a Brain Computer Interface (BCI) system. We developed an EEG-based real-time cursor control system on LabVIEW platform. EEG signals of left or right motor imagery on C3, and C4 channels were collected using OpenBCI amplifier system. The experimental protocol consisted of a training stage and a self-controlled stage with several runs in each stage. In the training stage, EEG signals were collected while subjects were asked to look at the moving cursor with a constant speed toward left/right, and pretended the cursor was controlled by their imagination. In the self-controlled stage, the movement of the cursor was controlled based on the concordance between the classification results and preset direction. Ten subjects were enrolled in this study. We found that the classification rate was associated with the consistency of the individual ERD/ERS frequency bands across different runs. If a subject presented a stable frequency band for the motor imaginary, the classification rate of the developed BCI system can reach a satisfactory performance with the highest classification rate of 93%. In conclusion, our results showed that the efficacy of our BCI system highly relied on the stability of the individual frequency pattern.

Long-term aerobic exercise can effectively improve the heart and lung function, stabilize mood an... more Long-term aerobic exercise can effectively improve the heart and lung function, stabilize mood and reduce the incidence of cardiopulmonary diseases. Brain activity can properly reflect physical and mental status of subjects during prolonged exercise, and long-term exercise may affect the power spectrum of EEG. Many studies showed that ECG, EMG and EEG, can effectively and accurately assess the status change during exercise. Safety and efficiency are the main concerns for promoting the aerobics for the elder. In this study, we aim to investigate the EEG and ECG features that can reveal status change during cycling exercise. Twenty-nine healthy subjects participated in this study. After four-minute resting stage, participants were asked to take cycling exercise continuously for twenty minutes, and the EEG, ECG signals were recorded and analyzed. The EEG data were divided into one-minute epoch and the wavelet transform was used to analyze five frequency bands, namely, theta (T), low alpha (LA), high alpha (HA), low beta (LB) and high beta (HB). The ECG signal was used to establish the average maximum heart rate ratio (AMHRR) and cardiac stress index (CSI). We found variations of RR intervals decreases during sustained cycling exercise. The CSI plot of a less frequent exerciser showed steeper than a frequent exerciser. If a participant has a steeper slope of CSI curve may imply an increase in cardiac stress. The AMHRR score at 65% could be a threshold for the occurrence of feeling hard during exercise. The CSI, HA and LB are the most proper features for assessing status change during exercise.

Expert Systems With Applications, Sep 1, 2023

In this study, we used two machine learning algorithms, namely, linear support vector machine (SV... more In this study, we used two machine learning algorithms, namely, linear support vector machine (SVM) and convolutional neural network (CNN), to classify the BCI (Brain Computer interface) competition IV-2a 2-class MI (motor imagery) data set which consists of EEG data from 9 subjects. For each subject, 5 sessions of signals from three electrodes (C3, Cz, and C4) were recorded with sampling rate 250Hz. The training data, which consisted of the first 3 sessions, included 400 trials. The evaluation data, which consisted of the last 2 sessions, included 320 trials. Each trial started with gazing at fix cross on screen for 3 seconds followed by a one-second visual cue pointing either to the left or right to instruct the subject for left or right motor imagery over a period of 4 seconds, and then followed by a short break of at least 1.5 seconds. Features were extracted from the 0.5 to 2.5 second signals after the cue for each trial from C3 and C4. Each EEG trial was band pass filtered into different frequency bands, namely, delta (0.5-3Hz), theta (4-8Hz), alpha (8-12Hz), beta bands (13-30Hz), gamma bands (31-60Hz). Those filtered signals were then used as the input data for training the linear SVM. In addition, we generated a 2 by 500 matrix by down sampling the training data from each trial. There are 5760 such matrices in total generated from all subjects and serve as the input data for training CNN and the trained model was evaluated by another 340 matrices from each subject. Our CNN architecture consisted of 2 convolution layer and 2 fully connect layers, and there was a batch normalization layer before the activated layer and a dropout layer with a probability of 50% after the activated layer. The classification accuracies evaluated by averaged kappa values obtained from linear SVM and CNN are 0.5 and 0.621, respectively, suggesting the deep learning CNN method is superior to the classical linear SVM on the EEG classification.

IEEE Transactions on Industrial Electronics, Jul 1, 2013

This paper proposes an asynchronous steady-state visual evoked potential (SSVEP)-based hospital b... more This paper proposes an asynchronous steady-state visual evoked potential (SSVEP)-based hospital bed nursing system without training from the users. The proposed system can be easily applied to any kinds of electrical hospital beds and can help the user to control the attitude of the hospital bed by only staring at the flashing panel. Different from most brain-computer interface (BCI) systems consisting of commercial instruments, this study provides a total design solution. The system firstly presents a light-emitting-diode stimulation panel to induce the user's SSVEP signal used as the input signal of the proposed system. Then, an SSVEP-amplifier/filter circuit and a field-programmable-gate-array-based SSVEP signal processor are respectively designed to acquire and process the subject's SSVEP. Moreover, H-bridge dc motor drive circuit is implemented to adjust the attitude of the hospital bed. Eventually, 15 subjects are invited to demonstrate the effectiveness of the proposed BCI-based hospital bed control nursing system. The total design BCI system shows good performance with an average accuracy of 92.5% and an average command transfer interval of 5.22 s per command.

Computer Methods and Programs in Biomedicine, Feb 1, 2023

Springer optimization and its applications, 2010

An electroencephalogram (EEG) is commonly used to study changes in brain activity during exercise... more An electroencephalogram (EEG) is commonly used to study changes in brain activity during exercise. In this study, we used recorded electrocardiography to define the average-to-maximal heart rate ratio (AMHRR) as a gauge of relative exercise workload, and explored how an increase in the AMHRR affected brain activity. EEG signals were recorded from 44 healthy subjects at four scalp sites (i.e., C1, C2, P1, and P2) during 9-min nonstop cycling exercise. The relationships among the AMHRR, EEG spectral power, and EEG fuzzy entropy (FuzzyEn) were established. The EEG spectral power and FuzzyEn results displayed similarly increasing patterns with the AMHRR at all electrodes. However, FuzzyEn exhibited superior specificity in selecting effective frequency bands (i.e., theta, alpha, and beta). The FuzzyEn method can be applied in a wearable device for a human machine interface, which can monitor the EEG during exercise.

IFMBE proceedings, 2009

Multiple system atrophy-cerebellar (MSA-C) is a degenerative neurological disease of brain. In th... more Multiple system atrophy-cerebellar (MSA-C) is a degenerative neurological disease of brain. In this study, we used three-dimensional (3D) fractal dimension (FD) method to investigate the structural complexity change of human cerebellum white matter (CBWM) and gray matter (CBGM) for MSA-C disease diagnosis. The original T-1 weighted magnetic resonance (MR) images of twenty-three MSA-C patients and twenty-one normal subjects were processed and the 3D CBGM and 3D CBGM were quantitatively analysed by the FD method. Results showed that FD values of the CBWM and CBGM of the MSA-C patients were significantly smaller comparing with the control group. Furthermore, the results of this study also demonstrated that the FD method is superior to the conventionally volumetric method in terms of better accuracy and sensitivity. The use of FD analysis in conjunction with the conventionally volumetric methods further allows us to establish a separating line in the FD-volume scatter plot of CBWM to identify the occurrence of cerebellum atrophy, which can serve as an early-stage indicator for detecting the MSA-C disease.

In this study, we employed Morlet wavelet, sample entropy, and fractal dimension on EEG and EMG s... more In this study, we employed Morlet wavelet, sample entropy, and fractal dimension on EEG and EMG signal to extract the feature of physical fatigue in the exercise. The result may be helpful for rehabilitation in effectiveness evaluation. Twenty healthy subjects participated in cycling exercise, and their physiological signals, including EEG, EMG, and ECG were recorded. In addition, we recorded subjects' feeling of fatigue since each subject has different physical strength and tolerance of non-stopping exercise. Signals in different stages, namely, resting, early, middle and late stages of exercising, were analyzed. ECG signal was used to categorize subjects into two groups, namely, moderate fatigue and severe fatigue. In EEG results, the averaged power, sample entropy, and fractal dimension of signals indicated that resting stages before and after the exercise were distinct from exercising stage. In severe fatigue, the averaged power within each frequency band of EEG increased with the duration of exercise whereas the power ratio, denoted by (theta+ alpha)/ beta, decreased gradually from the beginning of exercise until the resting after exercise. In addition, the EEG (C3) results of SE complexity ratio and FD complexity ratio decreased gradually from resting to last session of exercise in the moderate fatigue whereas in severe fatigue these ratios increased at the late exercising stage. Our results demonstrate that different patterns between moderate fatigue and severe fatigue can be effectively extracted by using the proposed methods.

Cancer Imaging

Background The epidermal growth factor receptor (EGFR) tyrosine kinase inhibitors (TKIs) are a fi... more Background The epidermal growth factor receptor (EGFR) tyrosine kinase inhibitors (TKIs) are a first-line therapy for non-small cell lung cancer (NSCLC) with EGFR mutations. Approximately half of the patients with EGFR-mutated NSCLC are treated with EGFR-TKIs and develop disease progression within 1 year. Therefore, the early prediction of tumor progression in patients who receive EGFR-TKIs can facilitate patient management and development of treatment strategies. We proposed a deep learning approach based on both quantitative computed tomography (CT) characteristics and clinical data to predict progression-free survival (PFS) in patients with advanced NSCLC after EGFR-TKI treatment. Methods A total of 593 radiomic features were extracted from pretreatment chest CT images. The DeepSurv models for the progression risk stratification of EGFR-TKI treatment were proposed based on CT radiomic and clinical features from 270 stage IIIB-IV EGFR-mutant NSCLC patients. Time-dependent PFS pred...

2017 International Automatic Control Conference (CACS), 2017

Bipolar disorder and schizophrenia are two prevailing psychiatric disorders with significant over... more Bipolar disorder and schizophrenia are two prevailing psychiatric disorders with significant overlaps in symptoms, abnormalities, and disease progression. Therefore, it is difficult to differentiate these two diseases without repeated clinical visits. Previous studies demonstrated high accuracy of classification for bipolar disorder and schizophrenia at the individual level by functional connectivity, but few studies focused on classifying between these two diseases directly. In order to assist diagnosis, we investigated further the feasibility of classifying bipolar disorder and schizophrenia by the structure of functional networks. The results revealed 90.0% accuracy of the classification with the sensitivity 1.0 and the specificity 0.80 for the patients with bipolar disorder. The present study indicated that the differences between the characteristics of brain network structures in bipolar disorder and schizophrenia could be the reliable features for the classification and may be the diagnostic indicators in the future.

IFMBE Proceedings

This study aims to recover the somatosensory evoked potentials (SSEPs) from the smearing electroe... more This study aims to recover the somatosensory evoked potentials (SSEPs) from the smearing electroencephalography (EEG) recordings using independent component analysis (ICA) in conjunction with the proposed timefrequency SSEP template (TF-SSEP). The SSEPs induced from patients with the impaired motor functions exhibit longer latency and lower amplitude than the normal SSEPs and are inevitably contaminated by artifacts and environmental noise. Although ICA has been demonstrated as a novel technique to segregate the EEG into independent sources, the selection of task-related components needs to be further elaborated. The TF-SSEP template, generated by the Morelet wavelet transformation of the averaged SSEPs from three normal subjects, was used to automatically extract the SSEP-related features. The performance of the TF-SSEP template was further validated using EEGs through the left and right peroneal nerve stimulation of four stroke patients. After ICA decomposition, the sources were selected for reconstruction if their correlation coefficients with the TF-SSEP template were higher than the predetermined threshold. On the other hand, the unselected sources were considered as the event-unrelated components or artifacts. Among all patients, the topography maps at four peak times, namely P40, N45, P60 and N75, showed higher contrast in the vicinity of the foot-associated motor area, and the resolved SSEPs demonstrated uncontaminated waveforms in comparison with the conventionally averaging method. This indicated that the proposed method can remarkably suppress artifacts and effectively extracted the SSEP-related features.

Medicine, 2016

In this study, we aimed to investigate the reactive changes in diffusion tensor imaging (DTI)-der... more In this study, we aimed to investigate the reactive changes in diffusion tensor imaging (DTI)-derived diffusion metrics of the anterior thalamic nucleus (AN), a relaying center for the Papez circuit, in early idiopathic normal pressure hydrocephalus (iNPH) patients with memory impairment, as well as its correlation with the patients' neuropsychological performances. In total, 28 probable iNPH patients with symptom onset within 1 year and 17 control subjects were prospectively recruited between 2010 and 2013 for this institutional review board-approved study. Imaging studies including DTI and a neuropsychological assessment battery were performed in all subjects. Diffusion metrics were measured from the region of the AN using tract-deterministic seeding method by reconstructing the mammillo-thalamo-cingulate connections within the Papez circuit. Differences in diffusion metrics and memory assessment scores between the patient and control group were examined via the Mann-Whitney U...

IFMBE proceedings, 2009

Multiple system atrophy-cerebellar (MSA-C) is a degenerative neurological disease of brain. In th... more Multiple system atrophy-cerebellar (MSA-C) is a degenerative neurological disease of brain. In this study, we used three-dimensional (3D) fractal dimension (FD) method to investigate the structural complexity change of human cerebellum white matter (CBWM) and gray matter (CBGM) for MSA-C disease diagnosis. Twenty patients and twenty-three normal subjects as the control group participated in this study. The T-1 weighted magnetic resonance (MR) images were processed and 3D CBGM and 3D CBGM were analyzed by the FD method. Results demonstrated that the FD values of patients’ CBWM and CBGM decreased significantly, and that their CBWM FD values decreased more significant than that of CBGM. Results also showed that the 3D FD method was superior to the conventional volumetric method in terms of better reliability, accuracy, and sensitivity.

Medicine, 2016

The studies regarding to the comparisons between major depressive disorder (MDD) and panic disord... more The studies regarding to the comparisons between major depressive disorder (MDD) and panic disorder (PD) in the microintegrity of white matter (WM) are uncommon. Therefore, we tried to a way to classify the MDD and PD. Fifty-three patients with 1st-episode medication-naive PD, 54 healthy controls, and 53 patients with 1st-episode medication-naive MDD were enrolled in this study. The controls and patients were matched for age, gender, education, and handedness. The diffusion tensor imaging scanning was also performed. The WM microintegrity was analyzed and compared between 3 groups of participants (ANOVA analysis) with age and gender as covariates. The MDD group had lower WM microintegrity than the PD group in the left anterior thalamic radiation, left uncinate fasciculus, left inferior fronto-occipital fasciculus, and bilateral corpus callosum. The MDD group had reductions in the microintegrity when compared to controls in the bilateral superior longitudinal fasciculi, inferior longitudinal fasciculi, inferior fronto-occipital fasciculi, and corpus callosum. The PD group had lower microintegrity in bilateral superior longitudinal fasciculi and left inferior fronto-occipital fasciculus when compared to controls. The widespread pattern of microintegrity alterations in frontolimbic WM circuit for MDD was different from restrictive pattern of alterations for PD.