Pub Date : 2022-12-07DOI: 10.1109/IECBES54088.2022.10079420
M.Z. Syatirah, M. Fatanah, M.Z. N. Jihan, M.M. Zulfakar, E. Seniz, M. Farhah
Several studies have been venturing into developing a model for controlling blood glucose among diabetes patients. It is because diabetes mellitus is a severe and common chronic disease affecting almost all populations in many countries. This study collected retrospective clinical data from five patients receiving insulin therapy in the ICU of HUSM. The auto-regressive with exogenous (ARX) and auto-regressive moving average with exogenous (ARMAX) model structure techniques were used to generate a model converter that best describes the glucose and insulin relationship of the subject. The simulation of ARX were started from model order (1,1,1) to model order (5,5,10) while, for ARMAX the simulation was started from model order (1,1,1,1) until model order (5,5,5,10). The three best model orders from ARX and ARMAX models were chosen. The best model fits for ARX and ARMAX were compared to identify the best model order in predicting the glucose-insulin system. The finding indicated that the ARX model recorded the best model fit for all patients in the 5th model order. Meanwhile, the ARMAX model recorded patients with different medical backgrounds and produced a different model order. Besides, the ARMAX model was considered the best option for most of the patients in this study due to the highest model fit, time-delay and lowest percentage of peak error. A more extensive data set may be required to ensure the structure of the model precisely describe the glucose-insulin interaction of the patient.Clinical Relevance– This study establishes a prediction model of the glucose-insulin system that can assist clinicians in providing appropriate insulin value and consequently reduce the incidence of hypoglycemia and hyperglycemia.
{"title":"Predictive Modeling using ARX and ARMAX Models for Glycemic Control in Intensive Care Unit Patients","authors":"M.Z. Syatirah, M. Fatanah, M.Z. N. Jihan, M.M. Zulfakar, E. Seniz, M. Farhah","doi":"10.1109/IECBES54088.2022.10079420","DOIUrl":"https://doi.org/10.1109/IECBES54088.2022.10079420","url":null,"abstract":"Several studies have been venturing into developing a model for controlling blood glucose among diabetes patients. It is because diabetes mellitus is a severe and common chronic disease affecting almost all populations in many countries. This study collected retrospective clinical data from five patients receiving insulin therapy in the ICU of HUSM. The auto-regressive with exogenous (ARX) and auto-regressive moving average with exogenous (ARMAX) model structure techniques were used to generate a model converter that best describes the glucose and insulin relationship of the subject. The simulation of ARX were started from model order (1,1,1) to model order (5,5,10) while, for ARMAX the simulation was started from model order (1,1,1,1) until model order (5,5,5,10). The three best model orders from ARX and ARMAX models were chosen. The best model fits for ARX and ARMAX were compared to identify the best model order in predicting the glucose-insulin system. The finding indicated that the ARX model recorded the best model fit for all patients in the 5th model order. Meanwhile, the ARMAX model recorded patients with different medical backgrounds and produced a different model order. Besides, the ARMAX model was considered the best option for most of the patients in this study due to the highest model fit, time-delay and lowest percentage of peak error. A more extensive data set may be required to ensure the structure of the model precisely describe the glucose-insulin interaction of the patient.Clinical Relevance– This study establishes a prediction model of the glucose-insulin system that can assist clinicians in providing appropriate insulin value and consequently reduce the incidence of hypoglycemia and hyperglycemia.","PeriodicalId":146681,"journal":{"name":"2022 IEEE-EMBS Conference on Biomedical Engineering and Sciences (IECBES)","volume":"21 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133761835","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-12-07DOI: 10.1109/IECBES54088.2022.10079466
Reemt Hinrichs, Felix Ortmann, Jörn Ostermann
Cochlear implants (CIs) are battery-powered, surgically implanted hearing-aids capable of restoring a sense of hearing in people suffering from moderate to profound hearing loss. Wireless transmission of audio from or to signal processors of cochlear implants can be used to improve speech understanding and localization of CI users. Data compression algorithms can be used to conserve battery power in this wireless transmission. However, very low latency is a strict requirement, limiting severly the available source coding algorithms. Previously, instead of coding the audio, coding of the electrical stimulation patterns of CIs was proposed to optimize the trade-off between bit-rate, latency and quality. In this work, a zero-delay deep autoencoder (DAE) for the coding of the electrical stimulation patters of CIs is proposed. Combining for the first time bayesian optimization with numerical approximated gradients of a nondifferential speech intelligibility measure for CIs, the short-time intelligibility measure (STOI), an optimized DAE architecture was found and trained that achieved equal or superior speech understanding at zero delay, outperforming well-known audio codecs. The DAE achieved reference vocoder STOI scores at 13.5 kbit/s compared to 33.6 kbit/s for Opus and 24.5 kbit/s for AMR-WB.
{"title":"Vector-Quantized Zero-Delay Deep Autoencoders for the Compression of Electrical Stimulation Patterns of Cochlear Implants using STOI","authors":"Reemt Hinrichs, Felix Ortmann, Jörn Ostermann","doi":"10.1109/IECBES54088.2022.10079466","DOIUrl":"https://doi.org/10.1109/IECBES54088.2022.10079466","url":null,"abstract":"Cochlear implants (CIs) are battery-powered, surgically implanted hearing-aids capable of restoring a sense of hearing in people suffering from moderate to profound hearing loss. Wireless transmission of audio from or to signal processors of cochlear implants can be used to improve speech understanding and localization of CI users. Data compression algorithms can be used to conserve battery power in this wireless transmission. However, very low latency is a strict requirement, limiting severly the available source coding algorithms. Previously, instead of coding the audio, coding of the electrical stimulation patterns of CIs was proposed to optimize the trade-off between bit-rate, latency and quality. In this work, a zero-delay deep autoencoder (DAE) for the coding of the electrical stimulation patters of CIs is proposed. Combining for the first time bayesian optimization with numerical approximated gradients of a nondifferential speech intelligibility measure for CIs, the short-time intelligibility measure (STOI), an optimized DAE architecture was found and trained that achieved equal or superior speech understanding at zero delay, outperforming well-known audio codecs. The DAE achieved reference vocoder STOI scores at 13.5 kbit/s compared to 33.6 kbit/s for Opus and 24.5 kbit/s for AMR-WB.","PeriodicalId":146681,"journal":{"name":"2022 IEEE-EMBS Conference on Biomedical Engineering and Sciences (IECBES)","volume":"17 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114553959","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-12-07DOI: 10.1109/IECBES54088.2022.10079373
Guorui Xu, S. Mahmoud, Akshay Kumar, Qiang Fang
Stroke is a severe cerebrovascular disease caused by the disruption of blood supply to the brain. To provide Inhome rehabilitation, it is essential to solve the problem of high rehabilitation cost and improve the rehabilitation efficacy. The training quality assessment is the core part of an in-home based training system as it provides important feedback which can be used by both the doctors and the patients. Furthermore, patients can conduct training in a comfortable, familiar environment. In this paper, a fast-training motion quality evaluation algorithm is proposed for the motion data captured by an accelerometer-based sensor network. The experiment results show that the evaluation result from the proposed system is within a satisfactory error range compared with that obtained by using the commercial X-sens system. The presented fast and efficient quantitative assessment method could be used for implementing a networked regional in-home rehabilitation system covering multiple users. Clinical Relevance– A fast and efficient upper limb assessment framework is proposed to support the development of a massive telerehabilitation system. Clinical Relevance – A fast and efficient upper limb assessment framework is proposed to support the development of a massive telerehabilitation system.
{"title":"A FastDTW Based Lightweight Limb Motion Function Assessment Method","authors":"Guorui Xu, S. Mahmoud, Akshay Kumar, Qiang Fang","doi":"10.1109/IECBES54088.2022.10079373","DOIUrl":"https://doi.org/10.1109/IECBES54088.2022.10079373","url":null,"abstract":"Stroke is a severe cerebrovascular disease caused by the disruption of blood supply to the brain. To provide Inhome rehabilitation, it is essential to solve the problem of high rehabilitation cost and improve the rehabilitation efficacy. The training quality assessment is the core part of an in-home based training system as it provides important feedback which can be used by both the doctors and the patients. Furthermore, patients can conduct training in a comfortable, familiar environment. In this paper, a fast-training motion quality evaluation algorithm is proposed for the motion data captured by an accelerometer-based sensor network. The experiment results show that the evaluation result from the proposed system is within a satisfactory error range compared with that obtained by using the commercial X-sens system. The presented fast and efficient quantitative assessment method could be used for implementing a networked regional in-home rehabilitation system covering multiple users. Clinical Relevance– A fast and efficient upper limb assessment framework is proposed to support the development of a massive telerehabilitation system. Clinical Relevance – A fast and efficient upper limb assessment framework is proposed to support the development of a massive telerehabilitation system.","PeriodicalId":146681,"journal":{"name":"2022 IEEE-EMBS Conference on Biomedical Engineering and Sciences (IECBES)","volume":"96 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"117220812","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-12-07DOI: 10.1109/IECBES54088.2022.10079531
W. Mansor, A. Z. Ahmad Zainuddin, M. F. Mohd Hanafi
Dyslexia diagnosis in children could not be performed in the absence of a specialist. This issue can be overcome with the use of advanced technology. Using convolution neural networks (CNN), the automatic classification of dyslexia from electroencephalogram (EEG) can be achieved. The role of the CNN and EEG in distinguishing dyslexia in children has not been explored. This study reveals the performance of CNN in recognising EEG signals of dyslexic and normal children using raw signals. The recorded EEG signals were passed through Short-Time Fourier Transform Analysis to transform the signals into images, which were then served as the input of the CNN. It was found that the CNN could recognise the EEG signals of dyslexic children with an accuracy of 80.9% and 72.1% using training and testing data, respectively.
{"title":"Convolution Neural Network Performance in Recognising EEG Signals of Dyslexic Children","authors":"W. Mansor, A. Z. Ahmad Zainuddin, M. F. Mohd Hanafi","doi":"10.1109/IECBES54088.2022.10079531","DOIUrl":"https://doi.org/10.1109/IECBES54088.2022.10079531","url":null,"abstract":"Dyslexia diagnosis in children could not be performed in the absence of a specialist. This issue can be overcome with the use of advanced technology. Using convolution neural networks (CNN), the automatic classification of dyslexia from electroencephalogram (EEG) can be achieved. The role of the CNN and EEG in distinguishing dyslexia in children has not been explored. This study reveals the performance of CNN in recognising EEG signals of dyslexic and normal children using raw signals. The recorded EEG signals were passed through Short-Time Fourier Transform Analysis to transform the signals into images, which were then served as the input of the CNN. It was found that the CNN could recognise the EEG signals of dyslexic children with an accuracy of 80.9% and 72.1% using training and testing data, respectively.","PeriodicalId":146681,"journal":{"name":"2022 IEEE-EMBS Conference on Biomedical Engineering and Sciences (IECBES)","volume":"68 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115776337","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-12-07DOI: 10.1109/IECBES54088.2022.10079329
N. F. Ismail, K. Y. Lee, L. N. Ismail, A. F. Abdul Rahim, N. S. Mohamad Hadis, A. Radzol
Surface Enhanced Raman Spectroscopy (SERS) is a specific and sensitive analytic technique suitable for detection of low concentration analyte. However, the performance of SERS is highly dependent on the type of SERS substrate used. In this study, solid base SERS substrates are fabricated for detection of low concentration dengue non-structural protein 1 (NS1) in saliva. Using an n-type phosphorous dopant, microstructural porous silicon (PSi) was fabricated using direct current electrochemical method. The PSi was deposited with different sizes of silver nanoparticles (AgNP) to increase the strength of electromagnetic field on the PSi surface. Here, the structural, electrical and Raman characterization of the fabricated AgNP coated PSi are presented. FESEM images show the cross-shaped surface structure of the substrate. The I-V curve reveals that the 75nm-AgNP samples produce better electrical conductivity property than the others. It is also observed that etching longer than a threshold reduces the conductivity performance of the substrate substantially, due to increase in the surface porosity. From Raman spectrum, the silicon peak at 520cm-1 shows a decreasing trend in intensity for samples with 30 min of etching. Interestingly, this observation complements that reported in our previous paper, where etching time of more than 28 min is found not suitable for producing uniform structure of PSi. The consistency between the structural, conductivity and Raman intensity can be used as indicators in developing good SERS substrate for non-invasive detection of low concentration dengue NS1 protein in saliva.
{"title":"Structural, Electrical and Raman Characterization of AgNP-coated Porous Silicon SERS Substrate for Detection of Dengue NS1 Protein","authors":"N. F. Ismail, K. Y. Lee, L. N. Ismail, A. F. Abdul Rahim, N. S. Mohamad Hadis, A. Radzol","doi":"10.1109/IECBES54088.2022.10079329","DOIUrl":"https://doi.org/10.1109/IECBES54088.2022.10079329","url":null,"abstract":"Surface Enhanced Raman Spectroscopy (SERS) is a specific and sensitive analytic technique suitable for detection of low concentration analyte. However, the performance of SERS is highly dependent on the type of SERS substrate used. In this study, solid base SERS substrates are fabricated for detection of low concentration dengue non-structural protein 1 (NS1) in saliva. Using an n-type phosphorous dopant, microstructural porous silicon (PSi) was fabricated using direct current electrochemical method. The PSi was deposited with different sizes of silver nanoparticles (AgNP) to increase the strength of electromagnetic field on the PSi surface. Here, the structural, electrical and Raman characterization of the fabricated AgNP coated PSi are presented. FESEM images show the cross-shaped surface structure of the substrate. The I-V curve reveals that the 75nm-AgNP samples produce better electrical conductivity property than the others. It is also observed that etching longer than a threshold reduces the conductivity performance of the substrate substantially, due to increase in the surface porosity. From Raman spectrum, the silicon peak at 520cm-1 shows a decreasing trend in intensity for samples with 30 min of etching. Interestingly, this observation complements that reported in our previous paper, where etching time of more than 28 min is found not suitable for producing uniform structure of PSi. The consistency between the structural, conductivity and Raman intensity can be used as indicators in developing good SERS substrate for non-invasive detection of low concentration dengue NS1 protein in saliva.","PeriodicalId":146681,"journal":{"name":"2022 IEEE-EMBS Conference on Biomedical Engineering and Sciences (IECBES)","volume":"23 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122611746","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-12-07DOI: 10.1109/IECBES54088.2022.10079643
Yang Feng, L. Chow, S. S. Tiang, N. Ramli, Nadia Muhammad Gowdh, L. Tan, Suhailah Abdullah
The post-processing of optic nerve Magnetic Resonance (MR) images is very challenging due to their small size and the surrounding cerebrospinal fluid (CSF). This study proposed a new segmentation method called gradient-based edge detection with skeletonization (GES), which is specifically designed to segment the optic nerve acquired with T1-weighted magnetization-prepared 180° radio-frequency pulses and rapid gradient-echo (MPRAGE) without fat saturation (FATSAT). GES identifies the edges of the optic nerve based on the largest gradient changes of signal intensity from one region (optic nerve) to another region (CSF). The proposed GES method performed better than the well-known level set method (LSM) with higher Dice similarity coefficient (DSC) of 0.80 - 0.85 compared to 0.61 – 0.77 using LSM. Bicubic interpolation with a factor of 8 was applied before the segmentation process to increase the spatial resolution of the optic nerve. Five datasets of NMOSD patients, clinically diagnosed optic neuritis, were used in this study. The bicubic-GES processed optic nerve images were used for the area and volume measurements on the intraorbital portion of both left and right optic nerves. The measurement results were used to study the effect of Neuromyelitis Optica Spectrum Disorder (NMOSD) on the optic nerve The NMOSD causes optic neuritis and demyelination in the optic nerve. This study found that the affected optic nerve has a smaller volume than the normal contralateral optic nerve. Clinical Relevance — This study provides an additional tool to confirm the diagnosis of optic neuritis in NMOSD patients through the volume measurement on bicubic-GES processed optic nerve MR images.
{"title":"Bicubic Interpolation and Gradient-based Edge Detection with Skeletonization Segmentation (Bicubic-GES) on MR Optic Nerve Images for Examining NMOSD","authors":"Yang Feng, L. Chow, S. S. Tiang, N. Ramli, Nadia Muhammad Gowdh, L. Tan, Suhailah Abdullah","doi":"10.1109/IECBES54088.2022.10079643","DOIUrl":"https://doi.org/10.1109/IECBES54088.2022.10079643","url":null,"abstract":"The post-processing of optic nerve Magnetic Resonance (MR) images is very challenging due to their small size and the surrounding cerebrospinal fluid (CSF). This study proposed a new segmentation method called gradient-based edge detection with skeletonization (GES), which is specifically designed to segment the optic nerve acquired with T1-weighted magnetization-prepared 180° radio-frequency pulses and rapid gradient-echo (MPRAGE) without fat saturation (FATSAT). GES identifies the edges of the optic nerve based on the largest gradient changes of signal intensity from one region (optic nerve) to another region (CSF). The proposed GES method performed better than the well-known level set method (LSM) with higher Dice similarity coefficient (DSC) of 0.80 - 0.85 compared to 0.61 – 0.77 using LSM. Bicubic interpolation with a factor of 8 was applied before the segmentation process to increase the spatial resolution of the optic nerve. Five datasets of NMOSD patients, clinically diagnosed optic neuritis, were used in this study. The bicubic-GES processed optic nerve images were used for the area and volume measurements on the intraorbital portion of both left and right optic nerves. The measurement results were used to study the effect of Neuromyelitis Optica Spectrum Disorder (NMOSD) on the optic nerve The NMOSD causes optic neuritis and demyelination in the optic nerve. This study found that the affected optic nerve has a smaller volume than the normal contralateral optic nerve. Clinical Relevance — This study provides an additional tool to confirm the diagnosis of optic neuritis in NMOSD patients through the volume measurement on bicubic-GES processed optic nerve MR images.","PeriodicalId":146681,"journal":{"name":"2022 IEEE-EMBS Conference on Biomedical Engineering and Sciences (IECBES)","volume":"76 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124768903","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-12-07DOI: 10.1109/IECBES54088.2022.10079591
Hannan N. Riaz, H. Nisar, K. Yeap
Neurofeedback Training (NFT) is an effective way for the participants to self-regulate the Electroencephalography (EEG) activity based on real-time feedback. This procedure has been proven to improve the neurological disorders in mentally ill patients and the psychological behavior of healthy individuals. Despite the considerable success of neurofeedback techniques, it is observed that some subjects fail to learn how to control their brain activities during neurofeedback training. This study is aimed to investigate the EEG learning process in alpha neurofeedback as an early-stage predictor of learners and non-learners in terms of the enhancement of alpha-band activities. 25 healthy participants have been trained using alpha upregulations. 8 of them were unable to regulate their alpha band within each session. Hence in this work resting state eyes-open EEG is used to predict the learning performance of the NFT participants. Using machine learning. A comparison of three machine learning algorithms; LDA, SVM, and GBM is performed to predict the non-learners based on the absolute alpha power and its Daubechies (level-4) wavelet decompositions eyes-open resting state EEG signals.
{"title":"Inefficacy Prediction of Alpha Up-Regulation Neurofeedback Training Using Eyes-Open Resting State Wavelet Features and Machine Learning","authors":"Hannan N. Riaz, H. Nisar, K. Yeap","doi":"10.1109/IECBES54088.2022.10079591","DOIUrl":"https://doi.org/10.1109/IECBES54088.2022.10079591","url":null,"abstract":"Neurofeedback Training (NFT) is an effective way for the participants to self-regulate the Electroencephalography (EEG) activity based on real-time feedback. This procedure has been proven to improve the neurological disorders in mentally ill patients and the psychological behavior of healthy individuals. Despite the considerable success of neurofeedback techniques, it is observed that some subjects fail to learn how to control their brain activities during neurofeedback training. This study is aimed to investigate the EEG learning process in alpha neurofeedback as an early-stage predictor of learners and non-learners in terms of the enhancement of alpha-band activities. 25 healthy participants have been trained using alpha upregulations. 8 of them were unable to regulate their alpha band within each session. Hence in this work resting state eyes-open EEG is used to predict the learning performance of the NFT participants. Using machine learning. A comparison of three machine learning algorithms; LDA, SVM, and GBM is performed to predict the non-learners based on the absolute alpha power and its Daubechies (level-4) wavelet decompositions eyes-open resting state EEG signals.","PeriodicalId":146681,"journal":{"name":"2022 IEEE-EMBS Conference on Biomedical Engineering and Sciences (IECBES)","volume":"37 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131969793","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-08-15DOI: 10.1109/IECBES54088.2022.10079267
Khiem H. Le, Hieu Pham, Thao Nguyen, Tu Nguyen, Cuong D. Do
An increasing number of people are being diagnosed with cardiovascular diseases (CVDs), the leading cause of death globally. The gold standard for identifying these heart problems is via electrocardiogram (ECG). The standard 12-lead ECG is widely used in clinical practice and most of the current research. However, using fewer leads can make ECG more pervasive as it can be integrated with portable or wearable devices. This article introduces two novel techniques to improve the performance of the current deep learning system for 3-lead ECG classification, making it comparable with models that are trained using standard 12-lead ECG. Specifically, we propose a multi-task learning scheme in the form of the number of heartbeats regression and an effective mechanism to integrate patient demographic data into the system. With these two advancements, we got classification performance in terms of F1 scores of 0.9796 and 0.8140 on two large-scale ECG datasets, i.e., Chapman and CPSC2018, respectively, which surpassed current state-of-the-art ECG classification methods, even those trained on 12-lead data. Our source code is available at github.com/lhkhiem28/LightX3ECG.
{"title":"Enhancing Deep Learning-based 3-lead ECG Classification with Heartbeat Counting and Demographic Data Integration","authors":"Khiem H. Le, Hieu Pham, Thao Nguyen, Tu Nguyen, Cuong D. Do","doi":"10.1109/IECBES54088.2022.10079267","DOIUrl":"https://doi.org/10.1109/IECBES54088.2022.10079267","url":null,"abstract":"An increasing number of people are being diagnosed with cardiovascular diseases (CVDs), the leading cause of death globally. The gold standard for identifying these heart problems is via electrocardiogram (ECG). The standard 12-lead ECG is widely used in clinical practice and most of the current research. However, using fewer leads can make ECG more pervasive as it can be integrated with portable or wearable devices. This article introduces two novel techniques to improve the performance of the current deep learning system for 3-lead ECG classification, making it comparable with models that are trained using standard 12-lead ECG. Specifically, we propose a multi-task learning scheme in the form of the number of heartbeats regression and an effective mechanism to integrate patient demographic data into the system. With these two advancements, we got classification performance in terms of F1 scores of 0.9796 and 0.8140 on two large-scale ECG datasets, i.e., Chapman and CPSC2018, respectively, which surpassed current state-of-the-art ECG classification methods, even those trained on 12-lead data. Our source code is available at github.com/lhkhiem28/LightX3ECG.","PeriodicalId":146681,"journal":{"name":"2022 IEEE-EMBS Conference on Biomedical Engineering and Sciences (IECBES)","volume":"45 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-08-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128425867","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：