Brain computer interface (BCI) based on visual stimulus is widely used, however, subjects have to focus on the stimulus rather than the object they want to control. Therefore, a video feedback car control system based on steady state visual evoked potential (SSVEP) was designed in this paper. We added a video feedback screen surround by the visual stimulators. As a result, subject could know the location as well as the status of the car. Meanwhile, we studied an improved multiple signal classification (MUSIC) method to classify SSVEP signal to improve the performance of frequency-domain analysis, and compared it with canonical correlation analysis and cyclic convolution method, it showed the highest accuracy. Moreover, we added an online training session to ensure that subject could master the using of the system, and according to the result of training session, the average online accuracy for four directions is 87.5%. Experiment results show that in our video feedback car control system, subjects could control the smart car by adjusting their distribution of the attention and drive the car through an obstacle fluently.
{"title":"Design of a video feedback SSVEP-BCI system for car control based on improved MUSIC method","authors":"Chang Liu, Songyun Xie, Xinzhou Xie, Xu Duan, Wei Wang, K. Obermayer","doi":"10.1109/IWW-BCI.2018.8311499","DOIUrl":"https://doi.org/10.1109/IWW-BCI.2018.8311499","url":null,"abstract":"Brain computer interface (BCI) based on visual stimulus is widely used, however, subjects have to focus on the stimulus rather than the object they want to control. Therefore, a video feedback car control system based on steady state visual evoked potential (SSVEP) was designed in this paper. We added a video feedback screen surround by the visual stimulators. As a result, subject could know the location as well as the status of the car. Meanwhile, we studied an improved multiple signal classification (MUSIC) method to classify SSVEP signal to improve the performance of frequency-domain analysis, and compared it with canonical correlation analysis and cyclic convolution method, it showed the highest accuracy. Moreover, we added an online training session to ensure that subject could master the using of the system, and according to the result of training session, the average online accuracy for four directions is 87.5%. Experiment results show that in our video feedback car control system, subjects could control the smart car by adjusting their distribution of the attention and drive the car through an obstacle fluently.","PeriodicalId":6537,"journal":{"name":"2018 6th International Conference on Brain-Computer Interface (BCI)","volume":"30 1","pages":"1-4"},"PeriodicalIF":0.0,"publicationDate":"2018-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"74576036","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 : 2018-01-01DOI: 10.1109/IWW-BCI.2018.8311530
Minju Kim, Sung-Phil Kim
Preprocessing of scalp electroencephalogram (EEG) signals to remove artifacts is essential to the reliable operation of non-invasive brain-computer interfaces (BCIs). One of the EEG-based BCIs leverages event-related potentials (ERPs) elicited by changes in specific external stimuli, which are sensitive to artifacts. To date, numerous methods have been proposed to remove artifacts from EEG. In this paper, we compare different artifact rejection methods for the operation of a BCI utilizing the ERP components such as P300 and N200, including independent component analysis (ICA), adaptive filtering, and artifact subspace reconstruction. We investigate the effect artifact removal by each method on the ERP waveform as well as BCI classification accuracy. The result demonstrates that the ERP waveforms through ICA showed a less across-trial variability in P300 amplitudes compared to other methods, as well as higher BCI classification accuracy. Our results may help the design of signal processing pipeline for EEG-based BCI systems.
{"title":"A comparsion of artifact rejection methods for a BCI using event related potentials","authors":"Minju Kim, Sung-Phil Kim","doi":"10.1109/IWW-BCI.2018.8311530","DOIUrl":"https://doi.org/10.1109/IWW-BCI.2018.8311530","url":null,"abstract":"Preprocessing of scalp electroencephalogram (EEG) signals to remove artifacts is essential to the reliable operation of non-invasive brain-computer interfaces (BCIs). One of the EEG-based BCIs leverages event-related potentials (ERPs) elicited by changes in specific external stimuli, which are sensitive to artifacts. To date, numerous methods have been proposed to remove artifacts from EEG. In this paper, we compare different artifact rejection methods for the operation of a BCI utilizing the ERP components such as P300 and N200, including independent component analysis (ICA), adaptive filtering, and artifact subspace reconstruction. We investigate the effect artifact removal by each method on the ERP waveform as well as BCI classification accuracy. The result demonstrates that the ERP waveforms through ICA showed a less across-trial variability in P300 amplitudes compared to other methods, as well as higher BCI classification accuracy. Our results may help the design of signal processing pipeline for EEG-based BCI systems.","PeriodicalId":6537,"journal":{"name":"2018 6th International Conference on Brain-Computer Interface (BCI)","volume":"2 1","pages":"1-4"},"PeriodicalIF":0.0,"publicationDate":"2018-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"81376954","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 : 2018-01-01DOI: 10.1109/IWW-BCI.2018.8311515
G. Choi, Soo-In Choi, Han-Jeong Hwang
Traditional electroencephalography (EEG)-based authentication systems generally use external stimuli that require user attention and relatively long time for authentication. The aim of this study is to investigate whether EEGs measured in resting state without using external stimuli can be used to develop a biometric authentication system. Seventeen subjects participated in the experiment in which EEG data were measured while the subjects repetitively closed and opened their eyes. Changes in alpha activity (8–13 Hz) during eyes open and closed were extracted for each channel as features, and inter- and intra-subject cross-correlation was calculated for identifying each subject. Increase in alpha activity was observed for all subjects at most channels. Most importantly, spatio-spectral patterns of changed alpha activity were different between the subjects, which led to a high mean identification accuracy of 88.4 %. Our experimental results demonstrate the feasibility of the proposed authentication method based on resting state EEGs.
{"title":"Individual identification based on resting-state EEG","authors":"G. Choi, Soo-In Choi, Han-Jeong Hwang","doi":"10.1109/IWW-BCI.2018.8311515","DOIUrl":"https://doi.org/10.1109/IWW-BCI.2018.8311515","url":null,"abstract":"Traditional electroencephalography (EEG)-based authentication systems generally use external stimuli that require user attention and relatively long time for authentication. The aim of this study is to investigate whether EEGs measured in resting state without using external stimuli can be used to develop a biometric authentication system. Seventeen subjects participated in the experiment in which EEG data were measured while the subjects repetitively closed and opened their eyes. Changes in alpha activity (8–13 Hz) during eyes open and closed were extracted for each channel as features, and inter- and intra-subject cross-correlation was calculated for identifying each subject. Increase in alpha activity was observed for all subjects at most channels. Most importantly, spatio-spectral patterns of changed alpha activity were different between the subjects, which led to a high mean identification accuracy of 88.4 %. Our experimental results demonstrate the feasibility of the proposed authentication method based on resting state EEGs.","PeriodicalId":6537,"journal":{"name":"2018 6th International Conference on Brain-Computer Interface (BCI)","volume":"5 1","pages":"1-4"},"PeriodicalIF":0.0,"publicationDate":"2018-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"75218087","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 : 2018-01-01DOI: 10.1109/IWW-BCI.2018.8311487
Taeho Kang, Yiyu Chen, S. Fazli, C. Wallraven
This study examines whether it is possible to predict successful memorization of previously-learned words in a language learning context from brain activity alone. Participants are tasked with memorizing German-Korean word association pairs, and their retention performance is tested on the day of and the day after learning. To investigate whether brain activity recorded via multi-channel EEG is predictive of memory formation, we perform statistical analysis followed by single-trial classification: first by using linear discriminant analysis, and then with convolutional neural networks. Our preliminary results confirm previous neurophysiological findings, that above-chance prediction of vocabulary memory formation is possible in both LDA and deep neural networks.
{"title":"Decoding of human memory formation with EEG signals using convolutional networks","authors":"Taeho Kang, Yiyu Chen, S. Fazli, C. Wallraven","doi":"10.1109/IWW-BCI.2018.8311487","DOIUrl":"https://doi.org/10.1109/IWW-BCI.2018.8311487","url":null,"abstract":"This study examines whether it is possible to predict successful memorization of previously-learned words in a language learning context from brain activity alone. Participants are tasked with memorizing German-Korean word association pairs, and their retention performance is tested on the day of and the day after learning. To investigate whether brain activity recorded via multi-channel EEG is predictive of memory formation, we perform statistical analysis followed by single-trial classification: first by using linear discriminant analysis, and then with convolutional neural networks. Our preliminary results confirm previous neurophysiological findings, that above-chance prediction of vocabulary memory formation is possible in both LDA and deep neural networks.","PeriodicalId":6537,"journal":{"name":"2018 6th International Conference on Brain-Computer Interface (BCI)","volume":"36 1","pages":"1-5"},"PeriodicalIF":0.0,"publicationDate":"2018-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"90295948","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 : 2018-01-01DOI: 10.1109/IWW-BCI.2018.8311521
Nayoung Kim, C. Nam
The main goal of the present study was to investigate how individual differences in working-memory capacity influence the participants' performance and brain networks in a dual-task paradigm. An important function of working memory is to integrate incoming information into an appropriate cognitive model by using two executive functions — updating and inhibition. We hypothesized that individual variability in working-memory function (estimated using operation-span measure) may affect to differential reactivity to both performance and brain connectivity. EEG signals and reaction times were recorded during a dual task that combined n-back and flanker tasks. In these tasks, participants with high working-memory span scores showed a better performance than those with low span scores. This finding suggests that a group with high working memory capacity is more affected by the cognitive control network than a low capacity group, possibly because people with high span utilize more efficient brain network during dual or multitasking situations. These findings contribute to perceiving cognitive control network as an individual trait, which can reflect neural efficiency to allow augmented human cognition, as well as a significant predictor of brain-computer interface performance.
{"title":"Working memory capacity influences performance and brain networks: Evidence from effective connectivity analysis","authors":"Nayoung Kim, C. Nam","doi":"10.1109/IWW-BCI.2018.8311521","DOIUrl":"https://doi.org/10.1109/IWW-BCI.2018.8311521","url":null,"abstract":"The main goal of the present study was to investigate how individual differences in working-memory capacity influence the participants' performance and brain networks in a dual-task paradigm. An important function of working memory is to integrate incoming information into an appropriate cognitive model by using two executive functions — updating and inhibition. We hypothesized that individual variability in working-memory function (estimated using operation-span measure) may affect to differential reactivity to both performance and brain connectivity. EEG signals and reaction times were recorded during a dual task that combined n-back and flanker tasks. In these tasks, participants with high working-memory span scores showed a better performance than those with low span scores. This finding suggests that a group with high working memory capacity is more affected by the cognitive control network than a low capacity group, possibly because people with high span utilize more efficient brain network during dual or multitasking situations. These findings contribute to perceiving cognitive control network as an individual trait, which can reflect neural efficiency to allow augmented human cognition, as well as a significant predictor of brain-computer interface performance.","PeriodicalId":6537,"journal":{"name":"2018 6th International Conference on Brain-Computer Interface (BCI)","volume":"99 1","pages":"1-4"},"PeriodicalIF":0.0,"publicationDate":"2018-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"80574136","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 : 2018-01-01DOI: 10.1109/IWW-BCI.2018.8311496
KIRAN KUMAR G R, M. Reddy
Traditional multichannel detection algorithms use reference signals that are a generalisation of the steady-state visual evoked potential (SSVEP) components. This leads to the suboptimal performance of the algorithms. For the first time, periodic component analysis (nCA) has been applied for the extraction of SSVEP components from background electroencephalogram (EEG). Data from six test subjects were used to evaluate the proposed method and compare it to standard canonical correlation analysis (CCA). The results demonstrate that the periodic component analysis acts as a reliable spatial filter for SSVEP extraction, and significantly outperforms traditional CCA even in low SNR conditions. The mean detection accuracy of nCA was higher than CCA across subjects, various window lengths and harmonics. The detection scores obtained from nCA provide reliable discrimination between control and idle states compared to CCA.
{"title":"Exploiting the temporal structure of EEG data for SSVEP detection","authors":"KIRAN KUMAR G R, M. Reddy","doi":"10.1109/IWW-BCI.2018.8311496","DOIUrl":"https://doi.org/10.1109/IWW-BCI.2018.8311496","url":null,"abstract":"Traditional multichannel detection algorithms use reference signals that are a generalisation of the steady-state visual evoked potential (SSVEP) components. This leads to the suboptimal performance of the algorithms. For the first time, periodic component analysis (nCA) has been applied for the extraction of SSVEP components from background electroencephalogram (EEG). Data from six test subjects were used to evaluate the proposed method and compare it to standard canonical correlation analysis (CCA). The results demonstrate that the periodic component analysis acts as a reliable spatial filter for SSVEP extraction, and significantly outperforms traditional CCA even in low SNR conditions. The mean detection accuracy of nCA was higher than CCA across subjects, various window lengths and harmonics. The detection scores obtained from nCA provide reliable discrimination between control and idle states compared to CCA.","PeriodicalId":6537,"journal":{"name":"2018 6th International Conference on Brain-Computer Interface (BCI)","volume":"7 1","pages":"1-4"},"PeriodicalIF":0.0,"publicationDate":"2018-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"90586628","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 : 2018-01-01DOI: 10.1109/IWW-BCI.2018.8311507
T. Milekovic
Paralysis has a severe impact on a patient's quality of life and entails a high emotional burden and life-long social and financial costs (‘One Degree of Separation, Paralysis and Spinal Cord Injury in the United States’ 2009; “Towards concerted efforts for treating and curing spinal cord injury” 2002; Arno, Levine, and Memmott 1999). Restoring movement and independence for people with paralysis remains a challenging clinical problem, currently with no viable solution. Recent demonstrations of intracortical brain-computer interfaces, neuroprosthetic devices that create a link between a person and a computer based on invasive recordings of a person's brain activity, have brought hope for their potential to restore movement and communication (Ajiboye et al. 2017; Pandarinath et al. 2017; Gilja et al. 2015; Jarosiewicz et al. 2015; Hochberg et al. 2012; Wodlinger et al. 2015; Collinger et al. 2013; Bouton et al. 2016; Aflalo et al. 2015). While the intracortical brain-computer interfaces have steadily improved over the last decade, the recent success in linking brain activity with the newly developed techniques for spinal cord stimulation look to revolutionize locomotor rehabilitation (Moraud et al. 2016; Wenger et al. 2016; Wenger et al. 2014; van den Brand et al. 2012; Rejc et al. 2016; Angeli et al. 2014; Harkema et al. 2011). Specifically, in a recent study a brain-spine interface — a neuroprostheses using gait states decoded from intracortically recorded neuronal activity to control spinal cord stimulation — restored weight-bearing locomotion of the paralyzed leg as early as six days post-injury in rhesus macaques (Capogrosso et al. 2016). The talk will discuss our progress towards enhancing the capabilities of brain-spine interfaces and demonstrating their use to alleviate motor deficits in other neurological disorders. In parallel, there is an ongoing search for identifying neural features and designing decoding algorithms with the aim to deliver both stable and accurate brain-computer interface control over clinically relevant periods of several months (Jarosiewicz et al. 2015; Vansteensel et al. 2016). The talk will also present our progress in developing techniques to identify stable neural features from intracortical neural recordings of people with tetraplegia and locked-in syndrome. The talk will show the use of these techniques to deliver stable long-term control of neural interfaces. This abstract is based on join work with Flavio Raschella, Giuseppe Schiavone, Matthew Perich, Marco Capogrosso, David Borton, Anish A. Sarma, Fabien Wagner, Eduardo Martin Moraud, Christopher Hitz, Jean-Baptiste Mignardot, Daniel Bacher, John D. Simeral, Jad Saab, Chethan Pandarinath, Brittany L. Sorice, Christine Blabe, Erin M. Oakley, Kathryn R. Tringale, Nicolas Buse, Jerome Gandar, Quentin Barraud, David Xing, Elodie Rey, Simone Duis, Yang Jianzhong, Wai Kin D. Ko, Qin Li, Chuan Qin, Emad Eskandar, Sydney S. Cash, Jaimie M. Henderson, Peter Detemple,
瘫痪严重影响患者的生活质量，并带来沉重的情感负担和终生的社会和经济成本(《美国的一度分离、瘫痪和脊髓损伤》，2009;“共同努力治疗和治愈脊髓损伤”，2002年;Arno, Levine, and Memmott 1999)。恢复瘫痪患者的活动和独立仍然是一个具有挑战性的临床问题，目前没有可行的解决方案。最近皮层内脑机接口的展示，神经假肢装置基于对人脑活动的侵入性记录在人与计算机之间建立联系，为他们恢复运动和沟通的潜力带来了希望(Ajiboye等人，2017;Pandarinath et al. 2017;Gilja et al. 2015;Jarosiewicz et al. 2015;Hochberg et al. 2012;Wodlinger et al. 2015;Collinger et al. 2013;Bouton et al. 2016;Aflalo et al. 2015)。虽然皮质内脑机接口在过去十年中稳步改善，但最近成功地将大脑活动与新开发的脊髓刺激技术联系起来，有望彻底改变运动康复(Moraud et al. 2016;Wenger et al. 2016;Wenger et al. 2014;van den Brand et al. 2012;Rejc et al. 2016;Angeli et al. 2014;Harkema et al. 2011)。具体来说，在最近的一项研究中，脑-脊柱界面——一种神经假体，利用从皮质内记录的神经元活动中解码的步态状态来控制脊髓刺激——在猕猴受伤后6天就恢复了瘫痪腿的负重运动(Capogrosso et al. 2016)。这次演讲将讨论我们在增强脑-脊柱接口能力方面的进展，并展示它们在缓解其他神经系统疾病的运动缺陷方面的应用。与此同时，目前正在研究识别神经特征和设计解码算法，目的是在临床相关的几个月内提供稳定和准确的脑机接口控制(Jarosiewicz等人，2015;Vansteensel et al. 2016)。本次演讲还将介绍我们在开发技术方面的进展，以从四肢瘫痪和闭锁综合征患者的皮质内神经记录中识别稳定的神经特征。这次演讲将展示如何使用这些技术来实现对神经接口的长期稳定控制。这篇摘要是基于与弗拉维奥·拉斯切拉、朱塞佩·斯齐亚沃尼、马修·佩里奇、马可·卡波格罗索、大卫·博尔顿、阿尼什·萨尔马、法比安·瓦格纳、爱德华多·马丁·莫罗德、克里斯托弗·希茨、让-巴蒂斯特·米格纳多、丹尼尔·巴切尔、约翰·d·西莫勒尔、杰德·萨伯、切坦·潘达里纳特、布列塔尼·l·索里斯、克里斯汀·布拉贝、埃琳·m·奥克利、凯瑟琳·r·特林格尔、尼古拉斯·布斯、杰罗姆·甘达尔、昆汀·巴罗、大卫·星、Elodie Rey、西蒙·杜伊斯、杨建中、Wai Kin D. Ko、李琴、琴川、Emad Eskandar, Sydney S. Cash, Jaimie M. Henderson, Peter Detemple, Tim Denison, Silvestro Micera, Erwan Bezard, Jocelyne Bloch, Krishna V. Shenoy, John P. Donoghue, Leigh R. Hochberg和gracimgoire Courtine。
{"title":"Brain-computer interfaces based on intracortical recordings of neural activity for restoration of movement and communication of people with paralysis","authors":"T. Milekovic","doi":"10.1109/IWW-BCI.2018.8311507","DOIUrl":"https://doi.org/10.1109/IWW-BCI.2018.8311507","url":null,"abstract":"Paralysis has a severe impact on a patient's quality of life and entails a high emotional burden and life-long social and financial costs (‘One Degree of Separation, Paralysis and Spinal Cord Injury in the United States’ 2009; “Towards concerted efforts for treating and curing spinal cord injury” 2002; Arno, Levine, and Memmott 1999). Restoring movement and independence for people with paralysis remains a challenging clinical problem, currently with no viable solution. Recent demonstrations of intracortical brain-computer interfaces, neuroprosthetic devices that create a link between a person and a computer based on invasive recordings of a person's brain activity, have brought hope for their potential to restore movement and communication (Ajiboye et al. 2017; Pandarinath et al. 2017; Gilja et al. 2015; Jarosiewicz et al. 2015; Hochberg et al. 2012; Wodlinger et al. 2015; Collinger et al. 2013; Bouton et al. 2016; Aflalo et al. 2015). While the intracortical brain-computer interfaces have steadily improved over the last decade, the recent success in linking brain activity with the newly developed techniques for spinal cord stimulation look to revolutionize locomotor rehabilitation (Moraud et al. 2016; Wenger et al. 2016; Wenger et al. 2014; van den Brand et al. 2012; Rejc et al. 2016; Angeli et al. 2014; Harkema et al. 2011). Specifically, in a recent study a brain-spine interface — a neuroprostheses using gait states decoded from intracortically recorded neuronal activity to control spinal cord stimulation — restored weight-bearing locomotion of the paralyzed leg as early as six days post-injury in rhesus macaques (Capogrosso et al. 2016). The talk will discuss our progress towards enhancing the capabilities of brain-spine interfaces and demonstrating their use to alleviate motor deficits in other neurological disorders. In parallel, there is an ongoing search for identifying neural features and designing decoding algorithms with the aim to deliver both stable and accurate brain-computer interface control over clinically relevant periods of several months (Jarosiewicz et al. 2015; Vansteensel et al. 2016). The talk will also present our progress in developing techniques to identify stable neural features from intracortical neural recordings of people with tetraplegia and locked-in syndrome. The talk will show the use of these techniques to deliver stable long-term control of neural interfaces. This abstract is based on join work with Flavio Raschella, Giuseppe Schiavone, Matthew Perich, Marco Capogrosso, David Borton, Anish A. Sarma, Fabien Wagner, Eduardo Martin Moraud, Christopher Hitz, Jean-Baptiste Mignardot, Daniel Bacher, John D. Simeral, Jad Saab, Chethan Pandarinath, Brittany L. Sorice, Christine Blabe, Erin M. Oakley, Kathryn R. Tringale, Nicolas Buse, Jerome Gandar, Quentin Barraud, David Xing, Elodie Rey, Simone Duis, Yang Jianzhong, Wai Kin D. Ko, Qin Li, Chuan Qin, Emad Eskandar, Sydney S. Cash, Jaimie M. Henderson, Peter Detemple,","PeriodicalId":6537,"journal":{"name":"2018 6th International Conference on Brain-Computer Interface (BCI)","volume":"24 1","pages":"1-2"},"PeriodicalIF":0.0,"publicationDate":"2018-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"74806844","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 : 2017-11-21DOI: 10.1109/IWW-BCI.2018.8311493
K. Hartmann, R. Schirrmeister, T. Ball
Recently, there is increasing interest and research on the interpretability of machine learning models, for example how they transform and internally represent EEG signals in Brain-Computer Interface (BCI) applications. This can help to understand the limits of the model and how it may be improved, in addition to possibly provide insight about the data itself. Schirrmeister et al. (2017) have recently reported promising results for EEG decoding with deep convolutional neural networks (ConvNets) trained in an end-to-end manner and, with a causal visualization approach, showed that they learn to use spectral amplitude changes in the input. In this study, we investigate how ConvNets represent spectral features through the sequence of intermediate stages of the network. We show higher sensitivity to EEG phase features at earlier stages and higher sensitivity to EEG amplitude features at later stages. Intriguingly, we observed a specialization of individual stages of the network to the classical EEG frequency bands alpha, beta, and high gamma. Furthermore, we find first evidence that particularly in the last convolutional layer, the network learns to detect more complex oscillatory patterns beyond spectral phase and amplitude, reminiscent of the representation of complex visual features in later layers of ConvNets in computer vision tasks. Our findings thus provide insights into how ConvNets hierarchically represent spectral EEG features in their intermediate layers and suggest that ConvNets can exploit and might help to better understand the compositional structure of EEG time series.
{"title":"Hierarchical internal representation of spectral features in deep convolutional networks trained for EEG decoding","authors":"K. Hartmann, R. Schirrmeister, T. Ball","doi":"10.1109/IWW-BCI.2018.8311493","DOIUrl":"https://doi.org/10.1109/IWW-BCI.2018.8311493","url":null,"abstract":"Recently, there is increasing interest and research on the interpretability of machine learning models, for example how they transform and internally represent EEG signals in Brain-Computer Interface (BCI) applications. This can help to understand the limits of the model and how it may be improved, in addition to possibly provide insight about the data itself. Schirrmeister et al. (2017) have recently reported promising results for EEG decoding with deep convolutional neural networks (ConvNets) trained in an end-to-end manner and, with a causal visualization approach, showed that they learn to use spectral amplitude changes in the input. In this study, we investigate how ConvNets represent spectral features through the sequence of intermediate stages of the network. We show higher sensitivity to EEG phase features at earlier stages and higher sensitivity to EEG amplitude features at later stages. Intriguingly, we observed a specialization of individual stages of the network to the classical EEG frequency bands alpha, beta, and high gamma. Furthermore, we find first evidence that particularly in the last convolutional layer, the network learns to detect more complex oscillatory patterns beyond spectral phase and amplitude, reminiscent of the representation of complex visual features in later layers of ConvNets in computer vision tasks. Our findings thus provide insights into how ConvNets hierarchically represent spectral EEG features in their intermediate layers and suggest that ConvNets can exploit and might help to better understand the compositional structure of EEG time series.","PeriodicalId":6537,"journal":{"name":"2018 6th International Conference on Brain-Computer Interface (BCI)","volume":"446 1","pages":"1-6"},"PeriodicalIF":0.0,"publicationDate":"2017-11-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"75184555","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 : 2017-11-16DOI: 10.1109/IWW-BCI.2018.8311531
Joos Behncke, R. Schirrmeister, Wolfram Burgard, T. Ball
The importance of robotic assistive devices grows in our work and everyday life. Cooperative scenarios involving both robots and humans require safe human-robot interaction. One important aspect here is the management of robot errors, including fast and accurate online robot-error detection and correction. Analysis of brain signals from a human interacting with a robot may help identifying robot errors, but accuracies of such analyses have still substantial space for improvement. In this paper we evaluate whether a novel framework based on deep convolutional neural networks (deep ConvNets) could improve the accuracy of decoding robot errors from the EEG of a human observer, both during an object grasping and a pouring task. We show that deep ConvNets reached significantly higher accuracies than both regularized Linear Discriminant Analysis (rLDA) and filter bank common spatial patterns (FB-CSP) combined with rLDA, both widely used EEG classifiers. Deep ConvNets reached mean accuracies of 75% ± 9 %, rLDA 65% ± 10% and FB-CSP + rLDA 63% ± 6% for decoding of erroneous vs. correct trials. Visualization of the time-domain EEG features learned by the ConvNets to decode errors revealed spatiotemporal patterns that reflected differences between the two experimental paradigms. Across subjects, ConvNet decoding accuracies were significantly correlated with those obtained with rLDA, but not CSP, indicating that in the present context ConvNets behaved more “rLDA-like” (but consistently better), while in a previous decoding study with another task but the same ConvNet architecture, it was found to behave more “CSP-like”. Our findings thus provide further support for the assumption that deep ConvNets are a versatile addition to the existing toolbox of EEG decoding techniques, and we discuss steps how ConvNet EEG decoding performance could be further optimized.
{"title":"The signature of robot action success in EEG signals of a human observer: Decoding and visualization using deep convolutional neural networks","authors":"Joos Behncke, R. Schirrmeister, Wolfram Burgard, T. Ball","doi":"10.1109/IWW-BCI.2018.8311531","DOIUrl":"https://doi.org/10.1109/IWW-BCI.2018.8311531","url":null,"abstract":"The importance of robotic assistive devices grows in our work and everyday life. Cooperative scenarios involving both robots and humans require safe human-robot interaction. One important aspect here is the management of robot errors, including fast and accurate online robot-error detection and correction. Analysis of brain signals from a human interacting with a robot may help identifying robot errors, but accuracies of such analyses have still substantial space for improvement. In this paper we evaluate whether a novel framework based on deep convolutional neural networks (deep ConvNets) could improve the accuracy of decoding robot errors from the EEG of a human observer, both during an object grasping and a pouring task. We show that deep ConvNets reached significantly higher accuracies than both regularized Linear Discriminant Analysis (rLDA) and filter bank common spatial patterns (FB-CSP) combined with rLDA, both widely used EEG classifiers. Deep ConvNets reached mean accuracies of 75% ± 9 %, rLDA 65% ± 10% and FB-CSP + rLDA 63% ± 6% for decoding of erroneous vs. correct trials. Visualization of the time-domain EEG features learned by the ConvNets to decode errors revealed spatiotemporal patterns that reflected differences between the two experimental paradigms. Across subjects, ConvNet decoding accuracies were significantly correlated with those obtained with rLDA, but not CSP, indicating that in the present context ConvNets behaved more “rLDA-like” (but consistently better), while in a previous decoding study with another task but the same ConvNet architecture, it was found to behave more “CSP-like”. Our findings thus provide further support for the assumption that deep ConvNets are a versatile addition to the existing toolbox of EEG decoding techniques, and we discuss steps how ConvNet EEG decoding performance could be further optimized.","PeriodicalId":6537,"journal":{"name":"2018 6th International Conference on Brain-Computer Interface (BCI)","volume":"114 1","pages":"1-6"},"PeriodicalIF":0.0,"publicationDate":"2017-11-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"85878730","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 : 2017-10-25DOI: 10.1109/IWW-BCI.2018.8311491
M. Völker, R. Schirrmeister, L. Fiederer, Wolfram Burgard, T. Ball
We recorded high-density EEG in a flanker task experiment (31 subjects) and an online BCI control paradigm (4 subjects). On these datasets, we evaluated the use of transfer learning for error decoding with deep convolutional neural networks (deep ConvNets). In comparison with a regularized linear discriminant analysis (rLDA) classifier, ConvNets were significantly better in both intra- and inter-subject decoding, achieving an average accuracy of 84.1 % within subject and 81.7 % on unknown subjects (flanker task). Neither method was, however, able to generalize reliably between paradigms. Visualization of features the ConvNets learned from the data showed plausible patterns of brain activity, revealing both similarities and differences between the different kinds of errors. Our findings indicate that deep learning techniques are useful to infer information about the correctness of action in BCI applications, particularly for the transfer of pre-trained classifiers to new recording sessions or subjects.
{"title":"Deep transfer learning for error decoding from non-invasive EEG","authors":"M. Völker, R. Schirrmeister, L. Fiederer, Wolfram Burgard, T. Ball","doi":"10.1109/IWW-BCI.2018.8311491","DOIUrl":"https://doi.org/10.1109/IWW-BCI.2018.8311491","url":null,"abstract":"We recorded high-density EEG in a flanker task experiment (31 subjects) and an online BCI control paradigm (4 subjects). On these datasets, we evaluated the use of transfer learning for error decoding with deep convolutional neural networks (deep ConvNets). In comparison with a regularized linear discriminant analysis (rLDA) classifier, ConvNets were significantly better in both intra- and inter-subject decoding, achieving an average accuracy of 84.1 % within subject and 81.7 % on unknown subjects (flanker task). Neither method was, however, able to generalize reliably between paradigms. Visualization of features the ConvNets learned from the data showed plausible patterns of brain activity, revealing both similarities and differences between the different kinds of errors. Our findings indicate that deep learning techniques are useful to infer information about the correctness of action in BCI applications, particularly for the transfer of pre-trained classifiers to new recording sessions or subjects.","PeriodicalId":6537,"journal":{"name":"2018 6th International Conference on Brain-Computer Interface (BCI)","volume":"1 1","pages":"1-6"},"PeriodicalIF":0.0,"publicationDate":"2017-10-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"89213704","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：