Pub Date : 2022-09-27DOI: 10.1109/BHI56158.2022.9926884
L. Dui, Simone Toffoli, Christopher Speziale, C. Termine, Matteo Matteucci, Simona Ferrante
Handwriting difficulties need to be addressed early to avoid several problems to children, both at school and in everyday life, but dysgraphia diagnosis cannot be performed before handwriting maturation. To solve this issue, we hypothesize that the analysis of drawings produced in a pre-literacy stage can predict handwriting problems that will occur years later. We designed a three-year longitudinal study from the last year of kindergarten to the end of second grade with two aims: (1) to longitudinally assess the evolution of drawing features, and (2) to understand if the features collected at pre-literacy can predict future handwriting problems. Hence, features were tested for statistically significant variation among the five time points available to assess their longitudinal evolution in time. Moreover, we trained machine learning models to select the most important features collected at pre-literacy and to assess their predictive capabilities, with dysgraphia risk assessed at the end of second grade. 202 children completed the longitudinal study. We found that 81% of the feature was sensitive to longitudinal maturation and that it is possible to predict the difficulties with a weighted area under the precision-recall curve of 0.72. This is a step forward towards an early intervention for handwriting problems.
{"title":"Can Free Drawing Anticipate Handwriting Difficulties? A Longitudinal Study","authors":"L. Dui, Simone Toffoli, Christopher Speziale, C. Termine, Matteo Matteucci, Simona Ferrante","doi":"10.1109/BHI56158.2022.9926884","DOIUrl":"https://doi.org/10.1109/BHI56158.2022.9926884","url":null,"abstract":"Handwriting difficulties need to be addressed early to avoid several problems to children, both at school and in everyday life, but dysgraphia diagnosis cannot be performed before handwriting maturation. To solve this issue, we hypothesize that the analysis of drawings produced in a pre-literacy stage can predict handwriting problems that will occur years later. We designed a three-year longitudinal study from the last year of kindergarten to the end of second grade with two aims: (1) to longitudinally assess the evolution of drawing features, and (2) to understand if the features collected at pre-literacy can predict future handwriting problems. Hence, features were tested for statistically significant variation among the five time points available to assess their longitudinal evolution in time. Moreover, we trained machine learning models to select the most important features collected at pre-literacy and to assess their predictive capabilities, with dysgraphia risk assessed at the end of second grade. 202 children completed the longitudinal study. We found that 81% of the feature was sensitive to longitudinal maturation and that it is possible to predict the difficulties with a weighted area under the precision-recall curve of 0.72. This is a step forward towards an early intervention for handwriting problems.","PeriodicalId":347210,"journal":{"name":"2022 IEEE-EMBS International Conference on Biomedical and Health Informatics (BHI)","volume":"30 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":"131673823","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/BHI56158.2022.9926840
Yongshen Zeng, Xiaoyan Song, Hongwu Chen, Weimin Huang, Wenjin Wang
The comprehensive monitoring of cardio-respiratory and neurological events of premature infants is desired for the Neonatal Intensive Care Unit (NICU). Video-based infant monitoring is an emerging tool for NICU as it eliminates skin irritations and enables new measurements like pain assessment. A multi-modal clinical dataset across the measurement of EEG and videos will be helpful in developing novel monitoring solutions for infant care. In this paper, we created such a dataset by simultaneously collecting the EEG signals and videos data from critically ill and preterm infants in NICU. Along with the recordings, we used the video-based cardio-respiratory measurements (heart rate and respiratory rate) to examine the validity of video recordings. We employed a classical video-based physiological measurement framework called Spatial Redundancy in combination with living-skin detection to measure the vital signs of recorded infants. The pilot measurements show the feasibility as well as the challenges that need to be addressed in algorithmic design in the next step. The dataset will be made publicly available to facilitate the research in this area. It will be useful for studying the video-based infant monitoring and its fusion with EEG, which may lead to new measurements such as a neonatal PSG for infant sleep staging and disease analysis (e.g. neonatal encephalopathy, neonatal respiratory distress syndrome).
{"title":"A Multi-modal Clinical Dataset for Critically-Ill and Premature Infant Monitoring: EEG and Videos","authors":"Yongshen Zeng, Xiaoyan Song, Hongwu Chen, Weimin Huang, Wenjin Wang","doi":"10.1109/BHI56158.2022.9926840","DOIUrl":"https://doi.org/10.1109/BHI56158.2022.9926840","url":null,"abstract":"The comprehensive monitoring of cardio-respiratory and neurological events of premature infants is desired for the Neonatal Intensive Care Unit (NICU). Video-based infant monitoring is an emerging tool for NICU as it eliminates skin irritations and enables new measurements like pain assessment. A multi-modal clinical dataset across the measurement of EEG and videos will be helpful in developing novel monitoring solutions for infant care. In this paper, we created such a dataset by simultaneously collecting the EEG signals and videos data from critically ill and preterm infants in NICU. Along with the recordings, we used the video-based cardio-respiratory measurements (heart rate and respiratory rate) to examine the validity of video recordings. We employed a classical video-based physiological measurement framework called Spatial Redundancy in combination with living-skin detection to measure the vital signs of recorded infants. The pilot measurements show the feasibility as well as the challenges that need to be addressed in algorithmic design in the next step. The dataset will be made publicly available to facilitate the research in this area. It will be useful for studying the video-based infant monitoring and its fusion with EEG, which may lead to new measurements such as a neonatal PSG for infant sleep staging and disease analysis (e.g. neonatal encephalopathy, neonatal respiratory distress syndrome).","PeriodicalId":347210,"journal":{"name":"2022 IEEE-EMBS International Conference on Biomedical and Health Informatics (BHI)","volume":"146 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":"116467238","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/BHI56158.2022.9926896
R. Sharan, Hao Xiong, S. Berkovsky
The sound of cough is an important indicator of the condition of the respiratory system. Automatic cough sound evaluation can aid the diagnosis of respiratory diseases. Large crowdsourced cough sound datasets have recently been used by several groups around the world to develop cough classification models. However, not all recordings in these datasets contain cough sounds. As such, it is important to screen the recordings for the presence of cough sounds before developing cough classification models. This work proposes a method to screen crowdsourced audio recordings for cough sounds using deep learning methods. The proposed approach divides the audio recording into overlapping frames and converts each frame into a mel-spectrogram representation. A pretrained convolutional neural network for audio classification is trained to learn the spectral characteristics of cough and non-cough frames from its mel-spectrogram representation. It is combined with a recurrent neural network to learn the dependencies between the sequence of frames. The proposed method is evaluated on 400 crowdsourced audio recordings, manually annotated as cough or non-cough. An accuracy of 0.9800 (AUC of 0.9973) is achieved in classifying cough and non-cough recordings using the proposed method. The trained network is used to analyze the remaining audio recordings in the dataset, identifying only about 67% of recordings as containing usable cough sounds. This shows the need to exercise caution when using crowdsourced cough data.
{"title":"Detecting Cough Recordings in Crowdsourced Data Using CNN-RNN","authors":"R. Sharan, Hao Xiong, S. Berkovsky","doi":"10.1109/BHI56158.2022.9926896","DOIUrl":"https://doi.org/10.1109/BHI56158.2022.9926896","url":null,"abstract":"The sound of cough is an important indicator of the condition of the respiratory system. Automatic cough sound evaluation can aid the diagnosis of respiratory diseases. Large crowdsourced cough sound datasets have recently been used by several groups around the world to develop cough classification models. However, not all recordings in these datasets contain cough sounds. As such, it is important to screen the recordings for the presence of cough sounds before developing cough classification models. This work proposes a method to screen crowdsourced audio recordings for cough sounds using deep learning methods. The proposed approach divides the audio recording into overlapping frames and converts each frame into a mel-spectrogram representation. A pretrained convolutional neural network for audio classification is trained to learn the spectral characteristics of cough and non-cough frames from its mel-spectrogram representation. It is combined with a recurrent neural network to learn the dependencies between the sequence of frames. The proposed method is evaluated on 400 crowdsourced audio recordings, manually annotated as cough or non-cough. An accuracy of 0.9800 (AUC of 0.9973) is achieved in classifying cough and non-cough recordings using the proposed method. The trained network is used to analyze the remaining audio recordings in the dataset, identifying only about 67% of recordings as containing usable cough sounds. This shows the need to exercise caution when using crowdsourced cough data.","PeriodicalId":347210,"journal":{"name":"2022 IEEE-EMBS International Conference on Biomedical and Health Informatics (BHI)","volume":"10 6","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"120843469","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/BHI56158.2022.9926740
Athanasios Kallipolitis, Melina Tziomaka, Dimitris Papadopoulos, I. Maglogiannis
Infertility significantly affects the quality of life on social and psychological levels and is estimated to expand in the coming years. In vitro fertilization is the applied answer of modern medicine to the ever-rising problem of low fertility in economically developed countries. Designated experts base their decision on selecting the most suitable embryo for transfer in the uterus by reviewing blastocysts images. Therefore, subjectivity and erroneous judgement can influence the progress of the whole fertilization process since no repeatable criteria exist to characterize the quality of each embryo. Towards the quantization of the visual content of ‘wannabe babies’ embryos, a comparative study between traditional machine and deep learning techniques is conducted in this paper. The utilization of a novel unsupervised segmentation scheme for the separation of trophectoderm and inner cell mass area provides a significant boost to the performance of traditional machine learning techniques. Moreover, an explainability technique that is based on the information retrieved by the Fisher Vector's generative model provides the necessary connection between the visual stimuli and the predicted results. The classification results of the proposed methodology are comparable with state-of the-art deep learning techniques and are accompanied by corresponding visual explanations that reveal the inner workings of each model and provide useful insight concerning the predictions' validity.
{"title":"Explainable computer vision analysis for embryo selection on blastocyst images","authors":"Athanasios Kallipolitis, Melina Tziomaka, Dimitris Papadopoulos, I. Maglogiannis","doi":"10.1109/BHI56158.2022.9926740","DOIUrl":"https://doi.org/10.1109/BHI56158.2022.9926740","url":null,"abstract":"Infertility significantly affects the quality of life on social and psychological levels and is estimated to expand in the coming years. In vitro fertilization is the applied answer of modern medicine to the ever-rising problem of low fertility in economically developed countries. Designated experts base their decision on selecting the most suitable embryo for transfer in the uterus by reviewing blastocysts images. Therefore, subjectivity and erroneous judgement can influence the progress of the whole fertilization process since no repeatable criteria exist to characterize the quality of each embryo. Towards the quantization of the visual content of ‘wannabe babies’ embryos, a comparative study between traditional machine and deep learning techniques is conducted in this paper. The utilization of a novel unsupervised segmentation scheme for the separation of trophectoderm and inner cell mass area provides a significant boost to the performance of traditional machine learning techniques. Moreover, an explainability technique that is based on the information retrieved by the Fisher Vector's generative model provides the necessary connection between the visual stimuli and the predicted results. The classification results of the proposed methodology are comparable with state-of the-art deep learning techniques and are accompanied by corresponding visual explanations that reveal the inner workings of each model and provide useful insight concerning the predictions' validity.","PeriodicalId":347210,"journal":{"name":"2022 IEEE-EMBS International Conference on Biomedical and Health Informatics (BHI)","volume":"202 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":"132317479","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/BHI56158.2022.9926806
S. VivekB., A. Adarsh, Jay Gubbi, Kartik Muralidharan, R. K. Ramakrishnan, Arpan Pal
Brain-computer interface (BCI) systems play an important role in medical applications such as stroke rehabilitation and neural prosthesis. These systems aim to decode the neural activity of the human brain measured using an Electroencephalogram (EEG). In this work, we consider the task of EEG-based motor imagery (intent) classification. Motor imagery (MI) refers to the imagination of the limb movement in the brain without actual action. Classification of motor imagery forms the basis for BCI-based prosthetic control. Existing approaches either use handcrafted features or features extracted from a deep neural network to interpret EEG-based MI. However, majority of the existing works fail to harness the functional connectivity within the brain that is captured using multiple EEG channels. In our work, we represent the input EEG signal as a graph where the nodes represent the EEG channels. The proposed approach uses a graph representation with a trainable weighted adjacency matrix to learn the optimal connectivity between nodes. Spatio-temporal features of the EEG signal are extracted via the proposed model that consists of a temporal convolution module and a graph convolution network. Experimental results and ablation study highlight the effectiveness of the proposed approach on the PhysioNet EEG motor movement and imagery dataset (EEG-MMIDB).
{"title":"ST-GNN for EEG Motor Imagery Classification","authors":"S. VivekB., A. Adarsh, Jay Gubbi, Kartik Muralidharan, R. K. Ramakrishnan, Arpan Pal","doi":"10.1109/BHI56158.2022.9926806","DOIUrl":"https://doi.org/10.1109/BHI56158.2022.9926806","url":null,"abstract":"Brain-computer interface (BCI) systems play an important role in medical applications such as stroke rehabilitation and neural prosthesis. These systems aim to decode the neural activity of the human brain measured using an Electroencephalogram (EEG). In this work, we consider the task of EEG-based motor imagery (intent) classification. Motor imagery (MI) refers to the imagination of the limb movement in the brain without actual action. Classification of motor imagery forms the basis for BCI-based prosthetic control. Existing approaches either use handcrafted features or features extracted from a deep neural network to interpret EEG-based MI. However, majority of the existing works fail to harness the functional connectivity within the brain that is captured using multiple EEG channels. In our work, we represent the input EEG signal as a graph where the nodes represent the EEG channels. The proposed approach uses a graph representation with a trainable weighted adjacency matrix to learn the optimal connectivity between nodes. Spatio-temporal features of the EEG signal are extracted via the proposed model that consists of a temporal convolution module and a graph convolution network. Experimental results and ablation study highlight the effectiveness of the proposed approach on the PhysioNet EEG motor movement and imagery dataset (EEG-MMIDB).","PeriodicalId":347210,"journal":{"name":"2022 IEEE-EMBS International Conference on Biomedical and Health Informatics (BHI)","volume":"118 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":"133293604","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/BHI56158.2022.9926880
Florian Allender, Rémi Allègre, Cédric Wemmert, J. Dischler
In a context of limited data availability, we consider the supervised segmentation of glomerular structures in patches of renal histopathological whole slide images. These structures are complex, include multiple substructures, and exhibit great variability in their shape, making their robust segmentation challenging. In this context, using appropriate data augmentation techniques is crucial to obtain more robust results. We investigate data augmentation based on random spatial deformations and conditional image synthesis for the training of a U-Net model. We rely on a SPADE model to perform the synthesis, using label maps built from the real patches available for training as input. Synthesis from ground truth masks only results in noisy patches, where substructures are absent, whereas additional structure information yield more realistic patches. We show that the best improvements of the segmentation performances are obtained by mixing real patches with synthetic patches generated from ground truth masks only, which yields an increase of up to 0.76 of average dice score w.r.t. augmentation based on spatial deformations only. We conclude that, using conditional image synthesis, patches synthesized with no additional structure information better contribute to the robustness of glomeruli segmentation than patches synthesized with structure information extracted from available real patches.
{"title":"Conditional image synthesis for improved segmentation of glomeruli in renal histopathological images","authors":"Florian Allender, Rémi Allègre, Cédric Wemmert, J. Dischler","doi":"10.1109/BHI56158.2022.9926880","DOIUrl":"https://doi.org/10.1109/BHI56158.2022.9926880","url":null,"abstract":"In a context of limited data availability, we consider the supervised segmentation of glomerular structures in patches of renal histopathological whole slide images. These structures are complex, include multiple substructures, and exhibit great variability in their shape, making their robust segmentation challenging. In this context, using appropriate data augmentation techniques is crucial to obtain more robust results. We investigate data augmentation based on random spatial deformations and conditional image synthesis for the training of a U-Net model. We rely on a SPADE model to perform the synthesis, using label maps built from the real patches available for training as input. Synthesis from ground truth masks only results in noisy patches, where substructures are absent, whereas additional structure information yield more realistic patches. We show that the best improvements of the segmentation performances are obtained by mixing real patches with synthetic patches generated from ground truth masks only, which yields an increase of up to 0.76 of average dice score w.r.t. augmentation based on spatial deformations only. We conclude that, using conditional image synthesis, patches synthesized with no additional structure information better contribute to the robustness of glomeruli segmentation than patches synthesized with structure information extracted from available real patches.","PeriodicalId":347210,"journal":{"name":"2022 IEEE-EMBS International Conference on Biomedical and Health Informatics (BHI)","volume":"131 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":"133279758","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/BHI56158.2022.9926766
Ilaria Siviero, L. Brusini, G. Menegaz, S. Storti
Motor-imagery (MI) electroencephalography (EEG) signal decomposition is an emerging technique for improving the performance of brain computer interfaces (BCIs), We proposed a multichannel-empirical wavelet transform (EWT) representation combined with a scattering convolution network (SCN) to efficiently decode the brain activity and extract relevant wave patterns for MI-based BCI. Two different preprocessing steps were tested: the first (PM1) included a bandpass Butterworth filter (1–40 Hz) and the independent component analysis (ICA), the second one (PM2) consisted only of a bandpass Butterworth filter (8–30 Hz). A binary support vector machine (SVM) classifier was used and the performance was evaluated in terms of classification accuracy. The proposed framework was assessed using the BCI competition IV dataset IIa, which contains EEG from 9 healthy subjects. PMI presented a maximum mean accuracy over all subjects of 82.05% in the classification of the tongue and the left-hand MI tasks. PM2 achieved an average accuracy over all subjects of 88.40% and a standard deviation of 3.01 outperforming other state of the art methods in classifying right-hand and left-hand MI tasks. Finally, we observed that the best channels, intended as the channels holding the highest discrimination power between two MI tasks, were highly subject-specific and thus enabling task-based channel selection is crucial.
{"title":"Motor-imagery EEG signal decoding using multichannel-empirical wavelet transform for brain computer interfaces","authors":"Ilaria Siviero, L. Brusini, G. Menegaz, S. Storti","doi":"10.1109/BHI56158.2022.9926766","DOIUrl":"https://doi.org/10.1109/BHI56158.2022.9926766","url":null,"abstract":"Motor-imagery (MI) electroencephalography (EEG) signal decomposition is an emerging technique for improving the performance of brain computer interfaces (BCIs), We proposed a multichannel-empirical wavelet transform (EWT) representation combined with a scattering convolution network (SCN) to efficiently decode the brain activity and extract relevant wave patterns for MI-based BCI. Two different preprocessing steps were tested: the first (PM1) included a bandpass Butterworth filter (1–40 Hz) and the independent component analysis (ICA), the second one (PM2) consisted only of a bandpass Butterworth filter (8–30 Hz). A binary support vector machine (SVM) classifier was used and the performance was evaluated in terms of classification accuracy. The proposed framework was assessed using the BCI competition IV dataset IIa, which contains EEG from 9 healthy subjects. PMI presented a maximum mean accuracy over all subjects of 82.05% in the classification of the tongue and the left-hand MI tasks. PM2 achieved an average accuracy over all subjects of 88.40% and a standard deviation of 3.01 outperforming other state of the art methods in classifying right-hand and left-hand MI tasks. Finally, we observed that the best channels, intended as the channels holding the highest discrimination power between two MI tasks, were highly subject-specific and thus enabling task-based channel selection is crucial.","PeriodicalId":347210,"journal":{"name":"2022 IEEE-EMBS International Conference on Biomedical and Health Informatics (BHI)","volume":"31 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":"133666651","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/BHI56158.2022.9926925
Thi-Thu-Hong Phan, Duc Le, P. Brijesh, D. Adjeroh, Jingxian Wu, M. Jensen, Ngan T. H. Le
Electrocardiogram (ECG) signal is one of the most effective sources of information mainly employed for the diagnosis and prediction of cardiovascular diseases (CVDs) connected with the abnormalities in heart rhythm. Clearly, single modality ECG (i.e. time series) cannot convey its complete characteristics, thus, exploiting both time and time-frequency modalities in the form of time-series data and spectrogram is needed. Leveraging the cutting-edge self-supervised learning (SSL) technique on unlabeled data, we propose SSL-based multimodality ECG classification. Our proposed network follows SSL learning paradigm and consists of two modules corresponding to pre-stream task, and down-stream task, respectively. In the SSL-pre-stream task, we utilize self-knowledge distillation (KD) techniques with no labeled data, on various transformations and in both time and frequency domains. In the down-stream task, which is trained on labeled data, we propose a gate fusion mechanism to fuse information from multimodality. To evaluate the effectiveness of our approach, ten-fold cross validation on the 12-lead PhysioNet 2020 dataset has been conducted. https://github.com/UARK-AICV/ECG-SSL.
{"title":"Multimodality Multi-Lead ECG Arrhythmia Classification using Self-Supervised Learning","authors":"Thi-Thu-Hong Phan, Duc Le, P. Brijesh, D. Adjeroh, Jingxian Wu, M. Jensen, Ngan T. H. Le","doi":"10.1109/BHI56158.2022.9926925","DOIUrl":"https://doi.org/10.1109/BHI56158.2022.9926925","url":null,"abstract":"Electrocardiogram (ECG) signal is one of the most effective sources of information mainly employed for the diagnosis and prediction of cardiovascular diseases (CVDs) connected with the abnormalities in heart rhythm. Clearly, single modality ECG (i.e. time series) cannot convey its complete characteristics, thus, exploiting both time and time-frequency modalities in the form of time-series data and spectrogram is needed. Leveraging the cutting-edge self-supervised learning (SSL) technique on unlabeled data, we propose SSL-based multimodality ECG classification. Our proposed network follows SSL learning paradigm and consists of two modules corresponding to pre-stream task, and down-stream task, respectively. In the SSL-pre-stream task, we utilize self-knowledge distillation (KD) techniques with no labeled data, on various transformations and in both time and frequency domains. In the down-stream task, which is trained on labeled data, we propose a gate fusion mechanism to fuse information from multimodality. To evaluate the effectiveness of our approach, ten-fold cross validation on the 12-lead PhysioNet 2020 dataset has been conducted. https://github.com/UARK-AICV/ECG-SSL.","PeriodicalId":347210,"journal":{"name":"2022 IEEE-EMBS International Conference on Biomedical and Health Informatics (BHI)","volume":"60 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":"123845049","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/BHI56158.2022.9926797
Reihaneh Hassanzadeh, A. Abrol, V. Calhoun
Neuroimaging studies in Alzheimer's disease (AD) and schizophrenia (SZ) have compared AD or SZ subjects against control (CN) subjects. However, it is also of interest and more critical to identify potential biomarkers by comparing these disorders, which can share some overlap, to each other directly. In this study, we investigated similarities and differences in resting-state functional network connectivity (rs-FNC) between 162 AD + late mild cognitive impairment (LMCI) and 181 SZ subjects from two well-known datasets - Alzheimer's Disease Neuroimaging Initiative (ADNI) and Bipolar and Schizophrenia Network on Intermediate Phenotypes (B-SNIP). We applied standard machine learning algorithms on confounder-controlled FNC features to distinguish groups of subjects, achieving an accuracy of 89% in classifying AD+LMCI vs. SZ subjects and an accuracy of 68% in a three-way classification of AD+LMCI, SZ, and CN subjects. Our results indicate that support vector machine (SVM) with an RBF kernel outperforms linear SVM and other machine learning methods, including random forest (RF), logistic regression (LR), and k-nearest neighbor (KNN). Furthermore, we conducted experiments for monitoring the potential impact of biases and showed that our trained models perform reasonably in a dataset-agnostic way. Finally, our findings highlight cerebellum and cognitive control networks as notable domains in common and unique FNC changes in AD and SZ disorders.
{"title":"Classification of Schizophrenia and Alzheimer's Disease using Resting-State Functional Network Connectivity","authors":"Reihaneh Hassanzadeh, A. Abrol, V. Calhoun","doi":"10.1109/BHI56158.2022.9926797","DOIUrl":"https://doi.org/10.1109/BHI56158.2022.9926797","url":null,"abstract":"Neuroimaging studies in Alzheimer's disease (AD) and schizophrenia (SZ) have compared AD or SZ subjects against control (CN) subjects. However, it is also of interest and more critical to identify potential biomarkers by comparing these disorders, which can share some overlap, to each other directly. In this study, we investigated similarities and differences in resting-state functional network connectivity (rs-FNC) between 162 AD + late mild cognitive impairment (LMCI) and 181 SZ subjects from two well-known datasets - Alzheimer's Disease Neuroimaging Initiative (ADNI) and Bipolar and Schizophrenia Network on Intermediate Phenotypes (B-SNIP). We applied standard machine learning algorithms on confounder-controlled FNC features to distinguish groups of subjects, achieving an accuracy of 89% in classifying AD+LMCI vs. SZ subjects and an accuracy of 68% in a three-way classification of AD+LMCI, SZ, and CN subjects. Our results indicate that support vector machine (SVM) with an RBF kernel outperforms linear SVM and other machine learning methods, including random forest (RF), logistic regression (LR), and k-nearest neighbor (KNN). Furthermore, we conducted experiments for monitoring the potential impact of biases and showed that our trained models perform reasonably in a dataset-agnostic way. Finally, our findings highlight cerebellum and cognitive control networks as notable domains in common and unique FNC changes in AD and SZ disorders.","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":"130233531","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/BHI56158.2022.9926873
H. Moradi, N. Roth, Ann-Kristin Seifer, Bjoern M. Eskofier
Levodopa (L-dopa) is the gold-standard medication and the most commonly used substance in the treatment of motor complications in Parkinson's disease (PD) patients. The “Wearing-off” phenomenon is the most frequent complication developed by long-term L-dopa therapy, which results in the reemergence of PD symptoms and lower quality of life in patients. Detecting and monitoring the onset and the duration of wearing-off alongside the persistence of the symptoms, known as “delayed-on”, would enable the patients to receive the medication changes in the required time while preventing them from extravagant use of L-dopa. Home monitoring systems using inertial measurement units have enabled us to measure gait parameters in unsupervised environments. By using patients' medication diaries and their gait parameters obtained from continuous real-world data in the course of two weeks, we developed a system to identify the distorted gait spans during L-dopa therapy utilizing personalized machine learning. Our algorithm differentiates between the two states of medication in effect and the distorted gait states with the mean accuracy of 77% ± 3.37. Furthermore, through each model's feature importance, we found that maximum sensor lift was the most prominent feature affected in the distorted gait sequences. We contribute to a better understanding of the repercussions of wearing-off episodes on gait parameters during L-dopa therapy. Moreover, our proposed system facilitates clinicians in monitoring the severity of these episodes more efficiently.
{"title":"Detection of distorted gait and wearing-off phenomenon in Parkinson's disease patients during Levodopa therapy","authors":"H. Moradi, N. Roth, Ann-Kristin Seifer, Bjoern M. Eskofier","doi":"10.1109/BHI56158.2022.9926873","DOIUrl":"https://doi.org/10.1109/BHI56158.2022.9926873","url":null,"abstract":"Levodopa (L-dopa) is the gold-standard medication and the most commonly used substance in the treatment of motor complications in Parkinson's disease (PD) patients. The “Wearing-off” phenomenon is the most frequent complication developed by long-term L-dopa therapy, which results in the reemergence of PD symptoms and lower quality of life in patients. Detecting and monitoring the onset and the duration of wearing-off alongside the persistence of the symptoms, known as “delayed-on”, would enable the patients to receive the medication changes in the required time while preventing them from extravagant use of L-dopa. Home monitoring systems using inertial measurement units have enabled us to measure gait parameters in unsupervised environments. By using patients' medication diaries and their gait parameters obtained from continuous real-world data in the course of two weeks, we developed a system to identify the distorted gait spans during L-dopa therapy utilizing personalized machine learning. Our algorithm differentiates between the two states of medication in effect and the distorted gait states with the mean accuracy of 77% ± 3.37. Furthermore, through each model's feature importance, we found that maximum sensor lift was the most prominent feature affected in the distorted gait sequences. We contribute to a better understanding of the repercussions of wearing-off episodes on gait parameters during L-dopa therapy. Moreover, our proposed system facilitates clinicians in monitoring the severity of these episodes more efficiently.","PeriodicalId":347210,"journal":{"name":"2022 IEEE-EMBS International Conference on Biomedical and Health Informatics (BHI)","volume":"47 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":"127463539","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：