Decoding Of Movement Intention Research Articles

High Performance Wearable Ultrasound as a Human-Machine Interface for Wrist and Hand Kinematic Tracking.

Non-invasive human machine interfaces (HMIs) have high potential in medical, entertainment, and industrial applications. Traditionally, surface electromyography (sEMG) has been used to track muscular activity and infer motor intention. Ultrasound (US) has received increasing attention as an alternative to sEMG-based HMIs. Here, we developed a portable US armband system with 24 channels and a multiple receiver approach, and compared it with existing sEMG- and US-based HMIs on movement intention decoding. US and motion capture data was recorded while participants performed wrist and hand movements of four degrees of freedom (DoFs) and their combinations. A linear regression model was used to offline predict hand kinematics from the US (or sEMG, for comparison) features. The method was further validated in real-time for a 3-DoF target reaching task. In the offline analysis, the wearable US system achieved an average [Formula: see text] of 0.94 in the prediction of four DoFs of the wrist and hand while sEMG reached a performance of [Formula: see text]= 0.60. In online control, the participants achieved an average 93% completion rate of the targets. When tailored for HMIs, the proposed US A-mode system and processing pipeline can successfully regress hand kinematics both in offline and online settings with performances comparable or superior to previously published interfaces. Wearable US technology may provide a new generation of HMIs that use muscular deformation to estimate limb movements. The wearable US system allowed for robust proportional and simultaneous control over multiple DoFs in both offline and online settings.

Measurement, Evaluation, and Control of Active Intelligent Gait Training Systems—Analysis of the Current State of the Art

Gait recognition and rehabilitation has been a research hotspot in recent years due to its importance to medical care and elderly care. Active intelligent rehabilitation and assistance systems for lower limbs integrates mechanical design, sensing technology, intelligent control, and robotics technology, and is one of the effective ways to resolve the above problems. In this review, crucial technologies and typical prototypes of active intelligent rehabilitation and assistance systems for gait training are introduced. The limitations, challenges, and future directions in terms of gait measurement and intention recognition, gait rehabilitation evaluation, and gait training control strategies are discussed. To address the core problems of the sensing, evaluation and control technology of the active intelligent gait training systems, the possible future research directions are proposed. Firstly, different sensing methods need to be proposed for the decoding of human movement intention. Secondly, the human walking ability evaluation models will be developed by integrating the clinical knowledge and lower limb movement data. Lastly, the personalized gait training strategy for collaborative control of human–machine systems needs to be implemented in the clinical applications.

The quantitative application of channel importance in movement intention decoding

The complex brain network consists of multiple collaborative regions, which can be activated to varying degrees by motor imagery (MI) and the induced electroencephalogram (EEG) recorded by an array of scalp electrodes is usually decoded for driving rehabilitation system. Either all channels or partially selected channels are equally applied to recognize movement intention, which may be incompatible with the individual differences of channels from different locations. In this paper, a channel importance based imaging method is proposed, denoted as CIBI. For each electrode of MI-EEG, the power over 8–30 Hz band is calculated from discrete Fourier spectrum and input to random forest algorithm (RF) to quantify its contribution, namely channel importance (CI); Then, CI is used for weighting the powers of α and β rhythms, which are interpolated to a 32×32 grid by using Clough-Tocher method respectively, generating two main band images with time-frequency-space information. In addition, a dual branch fusion convolutional neural network (DBFCNN) is developed to match with the characteristic of two MI images, realizing the extraction, fusion and classification of comprehensive features. Extensive experiments are conducted based on two public datasets with four classes of MI-EEG, the relatively higher average accuracies are obtained, and the improvements achieve 23.95% and 25.14% respectively when using channel importance, their statistical analysis are also performed by Kappa value, confusion matrix and receiver operating characteristic. Experiment results show that the personalized channel importance is helpful to enhance inter-class separability as well as the proposed method has the outstanding decoding ability for multiple MI tasks.

EEG-Based Continuous Hand Movement Decoding Using Improved Center-Out Paradigm.

The continuous decoding of human movement intention based on electroencephalogram (EEG) signals is valuable for developing a more natural motor augmented or assistive system instead of its discrete classifications. The classic center-out paradigm has been widely used to study discrete and continuous hand movement parameter decoding. However, when applying it in studying continuous movement decoding, the classic paradigm needs to be improved to increase the decoding performance, especially generalization performance. In this paper, we first discuss the limitations of the classic center-out paradigm in exploring the hand movement's continuous decoding. Then, an improved paradigm is proposed to enhance the continuous decoding performance. Besides, an adaptive decoder-ensemble framework is developed for continuous kinematic parameter decoding. Finally, with the improved center-out paradigm and the ensemble decoding framework, the average Pearson's correlation coefficients between the predicted and recorded movement kinematic parameters improve significantly by about 75 percent for the directional parameters and about 10 percent for the non-directional parameters. Furthermore, its generalization performance improves significantly by about 20 percent for the directional parameters. This study indicates the advantage of the improved paradigm in predicting the hand movement's kinematic information from low-frequency scalp EEG signals. It can advance the applications of the noninvasive motor brain-computer interface (BCI) in rehabilitation, daily assistance, and human augmentation areas.

Homology Characteristics of EEG and EMG for Lower Limb Voluntary Movement Intention.

In the field of lower limb exoskeletons, besides its electromechanical system design and control, attention has been paid to realizing the linkage of exoskeleton robots to humans via electroencephalography (EEG) and electromyography (EMG). However, even the state of the art performance of lower limb voluntary movement intention decoding still faces many obstacles. In the following work, focusing on the perspective of the inner mechanism, a homology characteristic of EEG and EMG for lower limb voluntary movement intention was conducted. A mathematical model of EEG and EMG was built based on its mechanism, which consists of a neural mass model (NMM), neuromuscular junction model, EMG generation model, decoding model, and musculoskeletal biomechanical model. The mechanism analysis and simulation results demonstrated that EEG and EMG signals were both excited by the same movement intention with a response time difference. To assess the efficiency of the proposed model, a synchronous acquisition system for EEG and EMG was constructed to analyze the homology and response time difference from EEG and EMG signals in the limb movement intention. An effective method of wavelet coherence was used to analyze the internal correlation between EEG and EMG signals in the same limb movement intention. To further prove the effectiveness of the hypothesis in this paper, six subjects were involved in the experiments. The experimental results demonstrated that there was a strong EEG-EMG coherence at 1 Hz around movement onset, and the phase of EEG was leading the EMG. Both the simulation and experimental results revealed that EEG and EMG are homologous, and the response time of the EEG signals are earlier than EMG signals during the limb movement intention. This work can provide a theoretical basis for the feasibility of EEG-based pre-perception and fusion perception of EEG and EMG in human movement detection.

Shared Prosthetic Control Based on Multiple Movement Intent Decoders.

A number of movement intent decoders exist in the literature that typically differ in the algorithms used and the nature of the outputs generated. Each approach comes with its own advantages and disadvantages. Combining the estimates of multiple algorithms may have better performance than any of the individual methods. This paper presents and evaluates a shared controller framework for prosthetic limbs based on multiple decoders of volitional movement intent. An algorithm to combine multiple estimates to control the prosthesis is developed in this paper. The capabilities of the approach are validated using a system that combines a Kalman filter-based decoder with a multilayer perceptron classifier-based decoder. The shared controller's performance is validated in online experiments where a virtual limb is controlled in real-time by amputee and intact-arm subjects. During the testing phase subjects controlled a virtual hand in real time to move digits to instructed positions using either a Kalman filter decoder, a multilayer perceptron decoder, or a linear combination of the two. The shared controller results in statistically significant improvements over the component decoders. Specifically, certain degrees of shared control result in increases in the time-in-target metric and decreases in unintended movements. The shared controller of this paper combines the good qualities of component decoders tested in this paper. Herein, combining a Kalman filter decoder with a classifier-based decoder inherits the flexibility of the Kalman filter decoder and the limited unwanted movements from the classifier-based decoder, resulting in a system that may be able to perform the tasks of everyday life more naturally and reliably.

Decoding movement intent patterns based on spatiotemporal and adaptive filtering method towards active motor training in stroke rehabilitation systems

Upper extremity (UE) neuromuscular dysfunction critically affects post-stroke patients from performing activities of daily life. In this regard, various rehabilitation robotics have been developed for providing assistive and/or resistive forces that allow stroke survivors to train their arms towards regaining the lost arm function. However, most of the rehabilitation systems function in a passively such that they only allow patients navigate already-defined trajectories that often does not align with their UE movement intention, thus hindering adequate motor function recovery. One possible way to address this problem is to use a decoded UE motion intent to trigger active and intuitive motor training for the patients, which would help restore their UE arm functions. In this study, a new approach based on spatiotemporal neuromuscular descriptor and adaptive filtering technique (STD-AFT) is proposed to optimally characterize multiple patterns of UE movements in post-stroke patients towards providing inputs for intelligently driven motor training in the rehabilitation robotic systems. The proposed STD-AFT performance was systematically investigated and assessed in comparison with commonly adopted methods via high-density surface electromyogram recordings obtained from post-stroke survivors who performed 21 distinct classes of pre-defined limb movements. Furthermore, the movement intent decoding was done using four different classification algorithms. The experimental results showed that the proposed STD-AFT achieved significant improvement of up to 13.36% (p < 0.05) in characterizing the multiple patterns of movement intents with relatively lower standard-error value even in the presence of the external interference in form of noise compared to the existing benchmark methods. Also, the STD-AFT showed obvious pattern seperability for individual movement class in a two-dimensional space. The outcomes of this study suggest that the proposed STD-AFT could provide potential inputs for active and intuitive motor training in robotic systems targeted towards stroke-rehabilitation.

Hierarchical Decoding Model of Upper Limb Movement Intention From EEG Signals Based on Attention State Estimation

Decoding the motion intention of the human upper limb from electroencephalography (EEG) signals has important practical values. However, existing decoding models are built under the attended state while subjects perform motion tasks. In practice, people are often distracted by other tasks or environmental factors, which may impair decoding performance. To address this problem, in this paper, we propose a hierarchical decoding model of human upper limb motion intention from EEG signals based on attention state estimation. The proposed decoding model includes two components. First, the attention state detection (ASD) component estimates the attention state during the upper limb movement. Next, the motion intention recognition (MIR) component decodes the motion intention by using the decoding models built under the attended and distracted states. The experimental results show that the proposed hierarchical decoding model performs well under the attended and distracted states. This work can advance the application of human movement intention decoding and provides new insights into the study of brain-machine interfaces.

A Nonlinear Latching Filter to Remove Jitter From Movement Estimates for Prostheses.

Continuous movement intent decoders are critical for precise control of hand and wrist prostheses. Noise in biological signals (e.g., myoelectric or neural signals) can lead to undesirable jitter in the output of these types of decoders. A low-pass filter (LPF) at the output of the decoder effectively reduces jitter, but also substantially slows intended movements. This paper introduces an alternative, the latching filter (LF), a recursive, nonlinear filter that provides smoothing of small-amplitude jitter but allows quick changes to its output in response to large input changes. The performance of a Kalman filter (KF) decoder smoothed with an LF is compared with that of both an KF decoder without an additional smoother and a KF decoder smoothed with a LPF. These three algorithms were tested in real-time on target holding and target reaching tasks using surface electromyographic signals recorded from 5 non-amputee subjects, and intramuscular electromyographic and peripheral neural signals recorded from an amputee subject. When compared with the LPF, the LF provided a statistically significant improvement in amputee and non-amputee subjects' ability to hold the hand steady at requested positions and achieve movement goals faster. The KF decoder with LF provided a statistically significant improvement in all subjects' ability to hold the prosthetic hand steady, with only slightly lower speeds, when compared to the unsmoothed KF.

Appropriate Feature Set and Window Parameters Selection for Efficient Motion Intent Characterization towards Intelligently Smart EMG-PR System

The constantly rising number of limb stroke survivors and amputees has motivated the development of intelligent prosthetic/rehabilitation devices for their arm function restoration. The device often integrates a pattern recognition (PR) algorithm that decodes amputees’ limb movement intent from electromyogram (EMG) signals, characterized by neural information and symmetric distribution. However, the control performance of the prostheses mostly rely on the interrelations among multiple dynamic factors of feature set, windowing parameters, and signal conditioning that have rarely been jointly investigated to date. This study systematically investigated the interaction effects of these dynamic factors on the performance of EMG-PR system towards constructing optimal parameters for accurately robust movement intent decoding in the context of prosthetic control. In this regard, the interaction effects of various features across window lengths (50 ms~300 ms), increments (50 ms~125 ms), robustness to external interferences and sensor channels (2 ch~6 ch), were examined using EMG signals obtained from twelve subjects through a symmetrical movement elicitation protocol. Compared to single features, multiple features consistently achieved minimum decoding error below 10% across optimal windowing parameters of 250 ms/100 ms. Also, the multiple features showed high robustness to additive noise with obvious trade-offs between accuracy and computation time. Consequently, our findings may provide proper insight for appropriate parameter selection in the context of robust PR-based control strategy for intelligent rehabilitation device.

Effect of User Practice on Prosthetic Finger Control With an Intuitive Myoelectric Decoder.

Machine learning-based myoelectric control is regarded as an intuitive paradigm, because of the mapping it creates between muscle co-activation patterns and prosthesis movements that aims to simulate the physiological pathways found in the human arm. Despite that, there has been evidence that closed-loop interaction with a classification-based interface results in user adaptation, which leads to performance improvement with experience. Recently, there has been a focus shift toward continuous prosthesis control, yet little is known about whether and how user adaptation affects myoelectric control performance in dexterous, intuitive tasks. We investigate the effect of short-term adaptation with independent finger position control by conducting real-time experiments with 10 able-bodied and two transradial amputee subjects. We demonstrate that despite using an intuitive decoder, experience leads to significant improvements in performance. We argue that this is due to the lack of an utterly natural control scheme, which is mainly caused by differences in the anatomy of human and artificial hands, movement intent decoding inaccuracies, and lack of proprioception. Finally, we extend previous work in classification-based and wrist continuous control by verifying that offline analyses cannot reliably predict real-time performance, thereby reiterating the importance of validating myoelectric control algorithms with real-time experiments.

Using High-Frequency Local Field Potentials From Multicortex to Decode Reaching and Grasping Movements in Monkey

Intracortical brain–machine interfaces (iBMIs) hold the promise to restore communication and movement ability of paralyzed people. Recent studies showed that local field potentials (LFPs) could be a reliable neural signal for movement intention decoding in iBMIs. However, previous studies investigated primarily on low-frequency LFPs, while the LFPs used were recorded from only one or two cortices. The aim of this paper is to investigate the potential of high-frequency LFPs (200–300 Hz) from multicortex in movement intention decoding. In this paper, LFPs were recorded via microelectrode arrays chronically implanted into the primary motor cortex (M1), somatosensory cortex (S1), and posterior parietal cortex of two monkeys while they were trained to perform 3-D reaching and grasping movements. The wavelet packet transform (WPT) method was used to extract the time and frequency information of the high-frequency LFP signals and the node energy of the WPT coefficients was selected as the features. After feature reduction by the principal component analysis, a support vector machine decoder was used to classify discrete reaching positions and grasping postures. Our results indicate that high decoding accuracy can be achieved by the high-frequency LFPs with WPT method and this kind of LFPs could serve as useful signals in iBMIs for movement intention decoding. Moreover, better and more robust decoding performance can be achieved by LFPs from multicortex than single-cortex.

Deep Learning Movement Intent Decoders Trained With Dataset Aggregation for Prosthetic Limb Control.

The performance of traditional approaches to decoding movement intent from electromyograms (EMGs) and other biological signals commonly degrade over time. Furthermore, conventional algorithms for training neural network based decoders may not perform well outside the domain of the state transitions observed during training. The work presented in this paper mitigates both these problems, resulting in an approach that has the potential to substantially improve the quality of life of the people with limb loss. This paper presents and evaluates the performance of four decoding methods for volitional movement intent from intramuscular EMG signals. The decoders are trained using the dataset aggregation (DAgger) algorithm, in which the training dataset is augmented during each training iteration based on the decoded estimates from previous iterations. Four competing decoding methods, namely polynomial Kalman filters (KFs), multilayer perceptron (MLP) networks, convolutional neural networks (CNN), and long short-term memory (LSTM) networks, were developed. The performances of the four decoding methods were evaluated using EMG datasets recorded from two human volunteers with transradial amputation. Short-term analyses, in which the training and cross-validation data came from the same dataset, and long-term analyses, in which the training and testing were done in different datasets, were performed. Short-term analyses of the decoders demonstrated that CNN and MLP decoders performed significantly better than KF and LSTM decoders, showing an improvement of up to 60% in the normalized mean-square decoding error in cross-validation tests. Long-term analyses indicated that the CNN, MLP, and LSTM decoders performed significantly better than a KF-based decoder at most analyzed cases of temporal separations (0-150 days) between the acquisition of the training and testing datasets. The short-term and long-term performances of MLP- and CNN-based decoders trained with DAgger demonstrated their potential to provide more accurate and naturalistic control of prosthetic hands than alternate approaches.

Discrete Hand Motion Intention Decoding Based on Transient Myoelectric Signals

In the application of human-robot interaction (HRI) rehabilitation exercise controlled by Electromyogram (EMG), if the discrete motion intention decoded by EMG signals is within the range of electromechanical delay (EMD), through the rehabilitation training, it will largely enhance the user's feeling feedback and fully activate the brain plasticity. So the decoding of hand movement intention by transient EMG is investigated to reach ideal characteristics needed in the HRI. The high-density EMG signal (HDEMG) database CapgMyo was used to decode the motion intention based on the transient EMG signals within the EMD rather than the whole recorded signals. We investigate the impact of the different decoding window lengths (WL) and training sets constructing methods when using several machine learning algorithms. In addition, the transient EMG signals decoding performance of the sparse multi-electrode EMG signal database NinaPro performing the same hand movement was compared. The visual inspection of the EMG map was used to determine the onset of HDEMG. The proposed approach was tested on EMG decoding window length of 150 ms, demonstrating a mean±SD testing performance of 94.21%±4.84% after voting. However, it is worth noting that sparse EMG signal did not achieve the desired decoding accuracy. The result showed that the high-density EMG signal could be used to decode the motion intention within EMD by simple machine learning algorithms, and extending the window length of training set could improve the decoding accuracy.

On the design of EEG-based movement decoders for completely paralyzed stroke patients

BackgroundBrain machine interface (BMI) technology has demonstrated its efficacy for rehabilitation of paralyzed chronic stroke patients. The critical component in BMI-training consists of the associative connection (contingency) between the intention and the feedback provided. However, the relationship between the BMI design and its performance in stroke patients is still an open question.MethodsIn this study we compare different methodologies to design a BMI for rehabilitation and evaluate their effects on movement intention decoding performance. We analyze the data of 37 chronic stroke patients who underwent 4 weeks of BMI intervention with different types of association between their brain activity and the proprioceptive feedback. We simulate the pseudo-online performance that a BMI would have under different conditions, varying: (1) the cortical source of activity (i.e., ipsilesional, contralesional, bihemispheric), (2) the type of spatial filter applied, (3) the EEG frequency band, (4) the type of classifier; and also evaluated the use of residual EMG activity to decode the movement intentions.ResultsWe observed a significant influence of the different BMI designs on the obtained performances. Our results revealed that using bihemispheric beta activity with a common average reference and an adaptive support vector machine led to the best classification results. Furthermore, the decoding results based on brain activity were significantly higher than those based on muscle activity.ConclusionsThis paper underscores the relevance of the different parameters used to decode movement, using EEG in severely paralyzed stroke patients. We demonstrated significant differences in performance for the different designs, which supports further research that should elucidate if those approaches leading to higher accuracies also induce higher motor recovery in paralyzed stroke patients.

A Novel Flexible Sensor for Muscle Shape Change Monitoring in Limb Motion Recognition.

Human limb movement intent recognition fundamentally provides the control mechanism for assistive devices such as exoskeleton and limb prosthesis. While different biopotential signals have been utilized for limb movement intent decoding, they seldom could account for spatial information associated with changes in muscle shape that could also be used to characterize the limb motor intent. Therefore, this study developed a novel nano gold stretchable-flexible sensor that captures spatial information associated with the muscle shape change signal (MSCS) during different muscle activation patterns. The novel sensor consists of 2-channels to acquire MSCS at a sampling rate of 125 Hz, corresponding to multiple classes of upper limb movements acquired across six able-bodied subjects. By utilizing the linear discriminant analysis algorithm on the acquired data with a single extracted feature, an overall average motion decoding accuracy of 90.9% was achieved. In addition, the waveform analysis results show that the novel sensor's recordings were less affected by external interferences, thus yielding high quality signals. This study is the first to utilize nano gold stretchable-flexible material for sensor fabrication in pattern recognition of upper limb movement intent, which may facilitate the development of effective assistive devices.

Utilizing sensory prediction errors for movement intention decoding: A new methodology.

We propose a new methodology for decoding movement intentions of humans. This methodology is motivated by the well-documented ability of the brain to predict sensory outcomes of self-generated and imagined actions using so-called forward models. We propose to subliminally stimulate the sensory modality corresponding to a user's intended movement, and decode a user's movement intention from his electroencephalography (EEG), by decoding for prediction errors-whether the sensory prediction corresponding to a user's intended movement matches the subliminal sensory stimulation we induce. We tested our proposal in a binary wheelchair turning task in which users thought of turning their wheelchair either left or right. We stimulated their vestibular system subliminally, toward either the left or the right direction, using a galvanic vestibular stimulator and show that the decoding for prediction errors from the EEG can radically improve movement intention decoding performance. We observed an 87.2% median single-trial decoding accuracy across tested participants, with zero user training, within 96 ms of the stimulation, and with no additional cognitive load on the users because the stimulation was subliminal.

Decoding of Self-paced Lower-Limb Movement Intention: A Case Study on the Influence Factors.

Brain-machine interfaces (BMIs) have been applied as new rehabilitation tools for motor disabled individuals. Active involvement of cerebral activity has been shown to enhance neuroplasticity and thus to restore mobility. Various studies have focused on the detection of upper-limb movement intention, while the fewer study has investigated the lower-limb movement intention decoding. This study presents a BMI to decode the self-paced lower-limb movement intention, with 10 healthy subjects participating in the experiment. We varied four influence factors including the movement type (dorsiflexion or plantar flexion), the limb side (left or right leg), the processing method (time-series analysis based on MRCP, i.e., movement-related cortical potential or frequency-domain estimation based on SMR, i.e., sensory motor rhythm) and the frequency band (e.g., delta, theta, mu, beta and MRCP band at [0.1 1] Hz), to estimate both single-trial and sample-based performance. Feature analysis was then conducted to show the discriminant power (DP) and brain modulations. The average detection latency was −0.334 ± 0.216 s in single-trial basis across all conditions. An average area under the curve (AUC) of 91.0 ± 3.5% and 68.2 ± 4.6% was obtained for the MRCP-based and SMR-based method in the classification, respectively. The best performance was yielded from plantar flexion with left leg using time-series analysis on the MRCP band. The feature analysis indicated a cross-subject consistency of DP with the MRCP-based method and subject-specific variance of DP with the SMR-based method. The results presented here might be further exploited in a rehabilitation scenario. The comprehensive factor analysis might be used to shed light on the design of an effective brain switch to trigger external robotic devices.

Influence of artifacts on movement intention decoding from EEG activity in severely paralyzed stroke patients.

Brain-machine interfaces (BMI) can be used to control robotic and prosthetic devices for rehabilitation of motor disorders, such as stroke. The calibration of these BMI systems is of paramount importance in order to establish a precise contingent link between the brain activity related to movement intention and the peripheral feedback. However, electroencephalographic (EEG) activity, commonly used to build non-invasive BMIs, can be easily contaminated by artifacts of electrical or physiological origin. The way these interferences can affect the performance of movement intention decoders has not been deeply studied, especially when dealing with severely paralyzed patients, which often generate more artifacts by compensatory movements. This paper evaluates the effects of removing artifacts from the data used to train a BMI decoder on a dataset of 28 severely paralyzed stroke patients. We show that cleaning the training datasets reduces the global BMI performance for decoding attempts of movement. Further, we demonstrate that this performance drop especially affects the test trials contaminated by artifacts (i.e., trials that might not reflect cortical activity but noise), but not the clean test trials (i.e., trials representing correct cortical activity). This paper underlines the importance of cleaning the datasets used to train BMI systems to improve their efficacy for decoding movement intention and maximize their neurorehabilitative potential.

Stroke lesion location influences the decoding of movement intention from EEG.

Recent studies have demonstrated the efficacy of brain-machine interfaces (BMI) for motor rehabilitation after stroke, especially for those patients with severe paralysis. However, a cerebro-vascular accident can affect the brain in many different manners, and lesions in diverse areas, even from significantly different volumes, can lead to similar or equal motor deficits. The location of the insult influences the way the brain activates when moving or attempting to move a paralyzed limb. Since the essence of a rehabilitative BMI is to precisely decode motor commands from the brain, it is crucial to characterize how lesion location affects the measured signals and if and how it influences BMI performance. This paper compares the performances of an electroencephalography (EEG)-based movement intention decoder in two groups of severely paralyzed chronic stroke patients: 14 with subcortical lesions and 14 with mixed (i.e., cortical and subcortical) lesions. We show that the lesion location influences the performance of the BMI when decoding the movement attempts of the paretic arm. The obtained results underline the need for further developments for a better individualization of BMI-based rehabilitative therapies for stroke patients.