Pub Date : 2022-09-27DOI: 10.1109/BHI56158.2022.9926767
Umair Mohammad, Fahad Saeed
Epilepsy is a chronic condition that causes repeat unprovoked seizures and many epileptics either develop resistance to medications and/or are not suitable candidates for surgical solutions. Hence, these recurring unpredictable seizures can have a severely negative impact on quality of life including an elevated risk of injury, social stigmatization, inability to take part in essential activities such as driving and possibly reduced access to healthcare. A predictive system that informs patients and caregivers about a potential upcoming seizure ahead of time is not only desirable but an urgent necessity. In this paper, we contribute by designing and developing patient-specific epileptic seizure (ES) prediction models using only electroencephalography (EEG) data with residual neural networks (ResNets) and transfer learning (TL) - (SPERTL). We train our proposed model on EEG data from 20 patients with a seizure prediction horizon (SPH) of 5 minutes and use the validation data to plot precision-recall curves for selecting the best thresholds. Testing on unseen data shows our model outperforms the state-of-the-art methods by achieving the highest average sensitivity of 88.1%, specificity of 92.3%, and accuracy of 92.3%. Our results also demonstrate the proposed model is less susceptible to false positives while maintaining a high positive prediction rate.
{"title":"SPERTL: Epileptic Seizure Prediction using EEG with ResNets and Transfer Learning","authors":"Umair Mohammad, Fahad Saeed","doi":"10.1109/BHI56158.2022.9926767","DOIUrl":"https://doi.org/10.1109/BHI56158.2022.9926767","url":null,"abstract":"Epilepsy is a chronic condition that causes repeat unprovoked seizures and many epileptics either develop resistance to medications and/or are not suitable candidates for surgical solutions. Hence, these recurring unpredictable seizures can have a severely negative impact on quality of life including an elevated risk of injury, social stigmatization, inability to take part in essential activities such as driving and possibly reduced access to healthcare. A predictive system that informs patients and caregivers about a potential upcoming seizure ahead of time is not only desirable but an urgent necessity. In this paper, we contribute by designing and developing patient-specific epileptic seizure (ES) prediction models using only electroencephalography (EEG) data with residual neural networks (ResNets) and transfer learning (TL) - (SPERTL). We train our proposed model on EEG data from 20 patients with a seizure prediction horizon (SPH) of 5 minutes and use the validation data to plot precision-recall curves for selecting the best thresholds. Testing on unseen data shows our model outperforms the state-of-the-art methods by achieving the highest average sensitivity of 88.1%, specificity of 92.3%, and accuracy of 92.3%. Our results also demonstrate the proposed model is less susceptible to false positives while maintaining a high positive prediction rate.","PeriodicalId":347210,"journal":{"name":"2022 IEEE-EMBS International Conference on Biomedical and Health Informatics (BHI)","volume":"71 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122508766","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 : 2022-09-27DOI: 10.1109/BHI56158.2022.9926741
J. Arenas, Hugo F. Posada-Quintero
Measuring pain objectively, namely, based on physiological signals instead of self-reported measures, would be highly valuable for better treating people with chronic pain. The subjectivity of the gold standard to quantify pain, which is based upon subjects' self-reported assessment using numerical or visual scales, makes pain management extremely complicated and, in many cases, has led to abuse of pain medication. Electrodermal activity (EDA) is a highly sensitive measure of sympathetic activity and has been increasingly used to objectively assess pain. In this study, we evaluated convolutional neural networks (CNN) and long short-term memory (LSTM) architectures for the task of detecting pain continuously. Additionally, we tested the use of the time-frequency spectrum of the phasic component of the electrodermal activity, as feature for this task. We used a merged database composed of thirty-six healthy subjects that underwent heat pain stimuli by means of a thermal grill. The LSTM models obtained better performance than the CNN ones by more of 3% in the F1-Score. Moreover, the best performance was achieved by a stacked bi- and uni-directional LSTM architecture, with 75.3% F1-Score, being able to accurately detect the onset and end of the pain response on EDA. Continuous objective pain detection using deep learning can contribute to continuous monitoring pain sensation and to reduce the consequences of subjectiveness of current pain assessment methods.
{"title":"Towards Continuous Acute Pain Detection using Deep Learning and Electrodermal Activity","authors":"J. Arenas, Hugo F. Posada-Quintero","doi":"10.1109/BHI56158.2022.9926741","DOIUrl":"https://doi.org/10.1109/BHI56158.2022.9926741","url":null,"abstract":"Measuring pain objectively, namely, based on physiological signals instead of self-reported measures, would be highly valuable for better treating people with chronic pain. The subjectivity of the gold standard to quantify pain, which is based upon subjects' self-reported assessment using numerical or visual scales, makes pain management extremely complicated and, in many cases, has led to abuse of pain medication. Electrodermal activity (EDA) is a highly sensitive measure of sympathetic activity and has been increasingly used to objectively assess pain. In this study, we evaluated convolutional neural networks (CNN) and long short-term memory (LSTM) architectures for the task of detecting pain continuously. Additionally, we tested the use of the time-frequency spectrum of the phasic component of the electrodermal activity, as feature for this task. We used a merged database composed of thirty-six healthy subjects that underwent heat pain stimuli by means of a thermal grill. The LSTM models obtained better performance than the CNN ones by more of 3% in the F1-Score. Moreover, the best performance was achieved by a stacked bi- and uni-directional LSTM architecture, with 75.3% F1-Score, being able to accurately detect the onset and end of the pain response on EDA. Continuous objective pain detection using deep learning can contribute to continuous monitoring pain sensation and to reduce the consequences of subjectiveness of current pain assessment methods.","PeriodicalId":347210,"journal":{"name":"2022 IEEE-EMBS International Conference on Biomedical and Health Informatics (BHI)","volume":"34 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122150425","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 : 2022-09-27DOI: 10.1109/BHI56158.2022.9926938
C. Siebra, Igor Matias, K. Wac
Transformers are recent deep learning (DL) models used to capture the dependence between parts of sequential data. While their potential was already demonstrated in the natural language processing (NLP) domain, emerging research shows transformers can also be an adequate modeling approach to relate longitudinal multi-featured continuous behavioral data to future health outcomes. As transformers-based predictions are based on a domain lexicon, the use of categories, commonly used in specialized areas to cluster values, is the likely way to compose lexica. However, the number of categories may influence the transformer prediction accuracy, mainly when the categorization process creates imbalanced datasets, or the search space is very restricted to generate optimal feasible solutions. This paper analyzes the relationship between models' accuracy and the sparsity of behavioral data categories that compose the lexicon. This analysis relies on a case example that uses mQoL-Transformer to model the influence of physical activity behavior on sleep health. Results show that the number of categories shall be treated as a further transformer's hyperparameter, which can balance the literature-based categorization and optimization aspects. Thus, DL processes could also obtain similar accuracies compared to traditional approaches, such as long short-term memory, when used to process short behavioral data sequences.
{"title":"Behavioral Data Categorization for Transformers-based Models in Digital Health","authors":"C. Siebra, Igor Matias, K. Wac","doi":"10.1109/BHI56158.2022.9926938","DOIUrl":"https://doi.org/10.1109/BHI56158.2022.9926938","url":null,"abstract":"Transformers are recent deep learning (DL) models used to capture the dependence between parts of sequential data. While their potential was already demonstrated in the natural language processing (NLP) domain, emerging research shows transformers can also be an adequate modeling approach to relate longitudinal multi-featured continuous behavioral data to future health outcomes. As transformers-based predictions are based on a domain lexicon, the use of categories, commonly used in specialized areas to cluster values, is the likely way to compose lexica. However, the number of categories may influence the transformer prediction accuracy, mainly when the categorization process creates imbalanced datasets, or the search space is very restricted to generate optimal feasible solutions. This paper analyzes the relationship between models' accuracy and the sparsity of behavioral data categories that compose the lexicon. This analysis relies on a case example that uses mQoL-Transformer to model the influence of physical activity behavior on sleep health. Results show that the number of categories shall be treated as a further transformer's hyperparameter, which can balance the literature-based categorization and optimization aspects. Thus, DL processes could also obtain similar accuracies compared to traditional approaches, such as long short-term memory, when used to process short behavioral data sequences.","PeriodicalId":347210,"journal":{"name":"2022 IEEE-EMBS International Conference on Biomedical and Health Informatics (BHI)","volume":"58 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114870003","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 : 2022-09-27DOI: 10.1109/BHI56158.2022.9926897
Lei Lu, T. Zhu, Yuan-ting Zhang, D. Clifton
Heart rate variability (HRV) is an important non-invasive parameter to assess the cardiac autonomic nervous system. In particular, spectrum matrices of HRV data have been widely used for physical and mental health monitoring. However, measurement uncertainties from data acquisition and physiological factors can easily affect the HRV spectrum and degrade outcomes of health monitoring. In this paper, we propose a new model for incomplete spectrum estimation of the HRV data based on matrix completion (MC). We show that our model performs efficiently when estimating missing entries for HRV spectra. Moreover, a refined model of matrix completion (RMC) is proposed that can be derived from correlation analysis of the HRV spectra. Two benchmark electrocardiography (ECG) datasets are retrieved and used to derive the HRV data, which are employed to evaluate the performance of our RMC method on the estimation of missing entries in the spectra. Furthermore, four different types of deep recurrent neural networks and the traditional MC method are used for a comparison study, and our RMC method obtains the least estimation error with different masking ratios. The experimental studies and comparison results demonstrate the advantages and robustness of our developed method for the estimation of incomplete HRV spectra.
{"title":"Spectrum Estimation of Heart Rate Variability Using Low-rank Matrix Completion","authors":"Lei Lu, T. Zhu, Yuan-ting Zhang, D. Clifton","doi":"10.1109/BHI56158.2022.9926897","DOIUrl":"https://doi.org/10.1109/BHI56158.2022.9926897","url":null,"abstract":"Heart rate variability (HRV) is an important non-invasive parameter to assess the cardiac autonomic nervous system. In particular, spectrum matrices of HRV data have been widely used for physical and mental health monitoring. However, measurement uncertainties from data acquisition and physiological factors can easily affect the HRV spectrum and degrade outcomes of health monitoring. In this paper, we propose a new model for incomplete spectrum estimation of the HRV data based on matrix completion (MC). We show that our model performs efficiently when estimating missing entries for HRV spectra. Moreover, a refined model of matrix completion (RMC) is proposed that can be derived from correlation analysis of the HRV spectra. Two benchmark electrocardiography (ECG) datasets are retrieved and used to derive the HRV data, which are employed to evaluate the performance of our RMC method on the estimation of missing entries in the spectra. Furthermore, four different types of deep recurrent neural networks and the traditional MC method are used for a comparison study, and our RMC method obtains the least estimation error with different masking ratios. The experimental studies and comparison results demonstrate the advantages and robustness of our developed method for the estimation of incomplete HRV spectra.","PeriodicalId":347210,"journal":{"name":"2022 IEEE-EMBS International Conference on Biomedical and Health Informatics (BHI)","volume":"45 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127890975","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 : 2022-09-27DOI: 10.1109/BHI56158.2022.9926869
Dimitrios Boucharas, Christos Androutsos, N. Tachos, E. Tripoliti, Dimitrios Manousos, Vasileios Skaramagkas, Emmanouil Ktistakis, K. Marias, M. Tsiknakis, D. Fotiadis
Smart wearables are becoming an irreplaceable part of daily living by supporting their users to maintain or adopt healthier lifestyles and monitor their current status. While the trend is increasing, little has been accomplished in the field of personalized solutions. In the present study, two models derived from distinct conceptual themes were developed, and the performance was evaluated utilizing a wearable prototype in the form of smart glasses. A statistical and a reinforcement learning approach were adopted to construct a personalization layer in terms of a predefined system reaction upon specific user behavior. The settings of the present study involve the user behavior derived from Artificial Intelligence (AI) based human activity recognition, among others, and the system reaction being a supportive Augmented Reality (AR) based functionality. Each approach yielding different benefits and drawbacks, imminently leads to a comparative analysis based on the efficiency offered by assessing the inference, update, and trend handling time. Both models are built upon the user's previous data, resulting in a data driven approach that is entirely different for each user and tailored to the user preferences. The results derived from the comparative analysis indicate that both approaches offer the personalization seeked, with the reinforcement learning approach to adapt faster.
{"title":"AI Methods for Personalized Suggestions on Smart Glasses Based on Human Activity Recognition*","authors":"Dimitrios Boucharas, Christos Androutsos, N. Tachos, E. Tripoliti, Dimitrios Manousos, Vasileios Skaramagkas, Emmanouil Ktistakis, K. Marias, M. Tsiknakis, D. Fotiadis","doi":"10.1109/BHI56158.2022.9926869","DOIUrl":"https://doi.org/10.1109/BHI56158.2022.9926869","url":null,"abstract":"Smart wearables are becoming an irreplaceable part of daily living by supporting their users to maintain or adopt healthier lifestyles and monitor their current status. While the trend is increasing, little has been accomplished in the field of personalized solutions. In the present study, two models derived from distinct conceptual themes were developed, and the performance was evaluated utilizing a wearable prototype in the form of smart glasses. A statistical and a reinforcement learning approach were adopted to construct a personalization layer in terms of a predefined system reaction upon specific user behavior. The settings of the present study involve the user behavior derived from Artificial Intelligence (AI) based human activity recognition, among others, and the system reaction being a supportive Augmented Reality (AR) based functionality. Each approach yielding different benefits and drawbacks, imminently leads to a comparative analysis based on the efficiency offered by assessing the inference, update, and trend handling time. Both models are built upon the user's previous data, resulting in a data driven approach that is entirely different for each user and tailored to the user preferences. The results derived from the comparative analysis indicate that both approaches offer the personalization seeked, with the reinforcement learning approach to adapt faster.","PeriodicalId":347210,"journal":{"name":"2022 IEEE-EMBS International Conference on Biomedical and Health Informatics (BHI)","volume":"2011 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125619497","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 : 2022-09-27DOI: 10.1109/BHI56158.2022.9926953
Nikolaos Siopis, Anastasios Alexiadis, Georgios Gerovasilis, Andreas K Triantafyllidis, K. Votis, D. Tzovaras
Hypertension is a serious disorder which contributes to an increased risk of cardiovascular disease and death. However, digital health systems dealing with the complexity of long-term hypertension self-management and remote medical management have been scarce. HSmartBPM provides a modular web platform for tailored management of hypertension. The HSmartBPM components include a virtual agent for patient guidance, a Decision Support System (DSS) for individualized monitoring of health parameters, risk prediction for cardiovascular disease, and shared care plan activities for patient treatment, contributing to a personalized approach for the therapeutic management of hypertension. Overall, the HSmartBPM solution aims to assist both patients and healthcare professionals with the everyday management of hypertension through the provision of an intelligent and tailored system.
{"title":"HSmartBPM: A modular web platform for tailored management of hypertension","authors":"Nikolaos Siopis, Anastasios Alexiadis, Georgios Gerovasilis, Andreas K Triantafyllidis, K. Votis, D. Tzovaras","doi":"10.1109/BHI56158.2022.9926953","DOIUrl":"https://doi.org/10.1109/BHI56158.2022.9926953","url":null,"abstract":"Hypertension is a serious disorder which contributes to an increased risk of cardiovascular disease and death. However, digital health systems dealing with the complexity of long-term hypertension self-management and remote medical management have been scarce. HSmartBPM provides a modular web platform for tailored management of hypertension. The HSmartBPM components include a virtual agent for patient guidance, a Decision Support System (DSS) for individualized monitoring of health parameters, risk prediction for cardiovascular disease, and shared care plan activities for patient treatment, contributing to a personalized approach for the therapeutic management of hypertension. Overall, the HSmartBPM solution aims to assist both patients and healthcare professionals with the everyday management of hypertension through the provision of an intelligent and tailored system.","PeriodicalId":347210,"journal":{"name":"2022 IEEE-EMBS International Conference on Biomedical and Health Informatics (BHI)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125726487","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}
Coughing is a common symptom across different clinical conditions and has gained further relevance in the past years due to the COVID-19 pandemic. An automated cough detection for continuous health monitoring could be developed using Earbud, a wearable sensor platform with audio and inertial measurement unit (IMU) sensors. Though several previous works have investigated audio-based automated cough detection, audio-based methods can be highly power-consuming for wearable sensor applications and raise privacy concerns. In this work, we develop IMU-based cough detection using a template matching-based algorithm. IMU provides a low-power privacy-preserving solution to complement audio-based algorithms. Similarly, template matching has low computational and memory needs, suitable for on-device implementations. The proposed method uses feature transformation of IMU signal and unsupervised representative template selection to improve upon our previous work. We obtained an AUC (AUC-ROC) of 0.85 and 0.83 for cough detection in a lab-based dataset with 45 participants and a controlled free-living dataset with 15 participants, respectively. These represent an AUC improvement of 0.08 and 0.10 compared to the previous work. Additionally, we conducted an uncontrolled free-living study with 7 participants where continuous measurements over a week were obtained from each participant. Our cough detection method achieved an AUC of 0.85 in the study, indicating that the proposed IMU-based cough detection translates well to the varied challenging scenarios present in free-living conditions.
{"title":"A Template Matching Based Cough Detection Algorithm Using IMU Data From Earbuds","authors":"Bishal Lamichhane, Ebrahim Nemati, Tousif Ahmed, Md. Mahbubur Rahman, Jilong Kuang, A. Gao","doi":"10.1109/BHI56158.2022.9926839","DOIUrl":"https://doi.org/10.1109/BHI56158.2022.9926839","url":null,"abstract":"Coughing is a common symptom across different clinical conditions and has gained further relevance in the past years due to the COVID-19 pandemic. An automated cough detection for continuous health monitoring could be developed using Earbud, a wearable sensor platform with audio and inertial measurement unit (IMU) sensors. Though several previous works have investigated audio-based automated cough detection, audio-based methods can be highly power-consuming for wearable sensor applications and raise privacy concerns. In this work, we develop IMU-based cough detection using a template matching-based algorithm. IMU provides a low-power privacy-preserving solution to complement audio-based algorithms. Similarly, template matching has low computational and memory needs, suitable for on-device implementations. The proposed method uses feature transformation of IMU signal and unsupervised representative template selection to improve upon our previous work. We obtained an AUC (AUC-ROC) of 0.85 and 0.83 for cough detection in a lab-based dataset with 45 participants and a controlled free-living dataset with 15 participants, respectively. These represent an AUC improvement of 0.08 and 0.10 compared to the previous work. Additionally, we conducted an uncontrolled free-living study with 7 participants where continuous measurements over a week were obtained from each participant. Our cough detection method achieved an AUC of 0.85 in the study, indicating that the proposed IMU-based cough detection translates well to the varied challenging scenarios present in free-living conditions.","PeriodicalId":347210,"journal":{"name":"2022 IEEE-EMBS International Conference on Biomedical and Health Informatics (BHI)","volume":"121 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134335672","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 : 2022-09-27DOI: 10.1109/BHI56158.2022.9926814
T. Koike-Akino, Ye Wang
In this paper, we introduce an emerging quantum machine learning (QML) framework to assist classical deep learning methods for biosignal processing applications. Specifically, we propose a hybrid quantum-classical neural network model that integrates a variational quantum circuit (VQC) into a deep neural network (DNN) for electroencephalogram (EEG), electromyogram (EMG), and electrocorticogram (ECoG) analysis. We demonstrate that the proposed quantum neural network (QNN) achieves state-of-the-art performance while the number of trainable parameters is kept small for VQC.
{"title":"quEEGNet: Quantum AI for Biosignal Processing","authors":"T. Koike-Akino, Ye Wang","doi":"10.1109/BHI56158.2022.9926814","DOIUrl":"https://doi.org/10.1109/BHI56158.2022.9926814","url":null,"abstract":"In this paper, we introduce an emerging quantum machine learning (QML) framework to assist classical deep learning methods for biosignal processing applications. Specifically, we propose a hybrid quantum-classical neural network model that integrates a variational quantum circuit (VQC) into a deep neural network (DNN) for electroencephalogram (EEG), electromyogram (EMG), and electrocorticogram (ECoG) analysis. We demonstrate that the proposed quantum neural network (QNN) achieves state-of-the-art performance while the number of trainable parameters is kept small for VQC.","PeriodicalId":347210,"journal":{"name":"2022 IEEE-EMBS International Conference on Biomedical and Health Informatics (BHI)","volume":"28 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133000740","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 : 2022-09-27DOI: 10.1109/BHI56158.2022.9926877
Orestis D. Papagiannopoulos, C. Papaloukas, V. Pezoulas, Harmen van de Werken, C. Poulet, Y. Mueller, P. Katsikis, D. Seny, D. Fotiadis
A cross-analysis study was conducted to compare proteomic platforms in classifying patients with Systemic Autoinflammatory diseases, using proteins extracted from different profiling experiments. The datasets used were obtained from SomaScan assays and Mass Spectrometry (MS). A separate analysis was performed to each dataset based on the false discovery rate (FDR) in order to extract statistically important proteins. Conventional machine learning algorithms were subsequently employed to evaluate the denoted proteins as candidate biomarkers and compare the predictive capabilities of the two proteomic platforms. Using the SomaScan assay, we managed to achieve higher classification metrics compared to the MS dataset. An improvement was also attained on the classification results when the features used were extracted from the MS data and applied on the SomaScan dataset, compared to the opposite combination. Finally, the proteins derived from the FDR analysis in both datasets proved to be highly correlated regarding their importance score.
{"title":"Comparison of Proteomic Approaches in Autoinflammatory Disease Classification","authors":"Orestis D. Papagiannopoulos, C. Papaloukas, V. Pezoulas, Harmen van de Werken, C. Poulet, Y. Mueller, P. Katsikis, D. Seny, D. Fotiadis","doi":"10.1109/BHI56158.2022.9926877","DOIUrl":"https://doi.org/10.1109/BHI56158.2022.9926877","url":null,"abstract":"A cross-analysis study was conducted to compare proteomic platforms in classifying patients with Systemic Autoinflammatory diseases, using proteins extracted from different profiling experiments. The datasets used were obtained from SomaScan assays and Mass Spectrometry (MS). A separate analysis was performed to each dataset based on the false discovery rate (FDR) in order to extract statistically important proteins. Conventional machine learning algorithms were subsequently employed to evaluate the denoted proteins as candidate biomarkers and compare the predictive capabilities of the two proteomic platforms. Using the SomaScan assay, we managed to achieve higher classification metrics compared to the MS dataset. An improvement was also attained on the classification results when the features used were extracted from the MS data and applied on the SomaScan dataset, compared to the opposite combination. Finally, the proteins derived from the FDR analysis in both datasets proved to be highly correlated regarding their importance score.","PeriodicalId":347210,"journal":{"name":"2022 IEEE-EMBS International Conference on Biomedical and Health Informatics (BHI)","volume":"17 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133587436","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 : 2022-09-27DOI: 10.1109/BHI56158.2022.9926860
Adriana Anido-Alonso, D. Álvarez-Estévez
Deep learning has demonstrated its usefulness in reaching top-level performance on a number of application domains. However, the achievement of robust prediction capabilities on multi-database scenarios referring to a common task is still a broad of concern. The problem arises associated with different sources of variability modulating the respective database generative processes. Hence, even though great performance can be obtained during validation on a local (source) dataset, maintenance of prediction capabilities on external databases, or target domains, is usually problematic. Such scenario has been studied in the past by the authors in the context of inter-database generalization in the domain of sleep medicine. In this work we build up over past work and explore the use of different local deep-learning model's combination strategies to analyze their effects on the resulting inter-database generalization performance. More specifically, we investigate the use of three different ensemble combination strategies, namely max-voting, output averaging, and weighted Nelder-Mead output combination, and compare them to the more classical database-aggregation approach. We compare the performance resulting from each of these strategies using six independent, heterogeneous and open sleep staging databases. Based on the results of our experimentation we analyze and discuss the advantages and disadvantages of each of the examined approaches.
{"title":"Analysis of ensemble-combination strategies for improving inter-database generalization of deep-learning-based automatic sleep staging","authors":"Adriana Anido-Alonso, D. Álvarez-Estévez","doi":"10.1109/BHI56158.2022.9926860","DOIUrl":"https://doi.org/10.1109/BHI56158.2022.9926860","url":null,"abstract":"Deep learning has demonstrated its usefulness in reaching top-level performance on a number of application domains. However, the achievement of robust prediction capabilities on multi-database scenarios referring to a common task is still a broad of concern. The problem arises associated with different sources of variability modulating the respective database generative processes. Hence, even though great performance can be obtained during validation on a local (source) dataset, maintenance of prediction capabilities on external databases, or target domains, is usually problematic. Such scenario has been studied in the past by the authors in the context of inter-database generalization in the domain of sleep medicine. In this work we build up over past work and explore the use of different local deep-learning model's combination strategies to analyze their effects on the resulting inter-database generalization performance. More specifically, we investigate the use of three different ensemble combination strategies, namely max-voting, output averaging, and weighted Nelder-Mead output combination, and compare them to the more classical database-aggregation approach. We compare the performance resulting from each of these strategies using six independent, heterogeneous and open sleep staging databases. Based on the results of our experimentation we analyze and discuss the advantages and disadvantages of each of the examined approaches.","PeriodicalId":347210,"journal":{"name":"2022 IEEE-EMBS International Conference on Biomedical and Health Informatics (BHI)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132760364","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：