We propose to fuse two currently separate research lines on novel therapies for stroke rehabilitation: brain-computer interface (BCI) training and transcranial electrical stimulation (TES). Speciﬁcally, we show that BCI technology can be used to learn personalized decoding models that relate the global conﬁguration of brain rhythms in individual subjects (as measured by EEG) to their motor performance during 3D reaching movements. We demonstrate that our models capture substantial across-subject heterogeneity, and argue that this heterogeneity is a likely cause of limited effect sizes observed in TES for enhancing motor performance. We conclude by discussing how our personalized models can be used to derive optimal TES parameters, e.g., stimulation site and frequency, for individual patients.
Motor deﬁcits are one of the most common outcomes of stroke. According to the World Health Organization, 15 million people worldwide suffer a stroke each year. Of these, ﬁve million are permanently disabled. For this third, upper limb weakness and loss of hand function are among the most devastating types of disabilities, which affect the quality of their daily life . Despite a wide range of rehabilitation therapies, including medication treatment , conventional physiotherapy , and robot physiotherapy , only approximately 20% of patients achieve some form of functional recovery in the ﬁrst six months , .
Current research on novel therapies includes neurofeedback training based on brain-computer interface (BCI) technology and transcranial electrical stimulation (TES). The former approach attempts to support cortical reorganization by providing haptic feedback with a robotic exoskeleton that is congruent to movement attempts, as decoded in real-time from neuroimaging data , . The latter type of research aims to reorganize cortical networks in a way that supports motor performance, because post-stroke alterations of cortical networks have been found to correlate with the severity of motor deﬁcits , . While initial evidence suggested that both approaches, BCIbased training  and TES , have a positive impact, the signiﬁcance of these results over conventional physiotherapy was not always achieved by different studies , , .
One potential explanation for the difﬁculty to replicate the initially promising ﬁndings is the heterogeneity of stroke patients. Different locations of stroke-induced structural changes
are likely to result in substantial across-patient variance in the functional reorganization of cortical networks. As a result, not all patients may beneﬁt from the same neurofeedback or stimulation protocol. We thus propose to fuse these two research themes and use BCI technology to learn personalized models that relate the conﬁguration of cortical networks to each patient’s motor deﬁcits. These personalized models may then be used to predict which TES parameters, e.g., spatial location and frequency band, optimally support rehabilitation in each individual patient.
In this study, we address the ﬁrst step towards personalized TES for stroke rehabilitation. Using a transfer learning framework developed in our group , we show how to create personalized decoding models that relate the EEG of healthy subjects during a 3D reaching task to their motor performance in individual trials. We further demonstrate that the resulting decoding models capture substantial acrosssubject heterogeneity, thereby providing empirical support for the need to personalize models. We conclude by reviewing our ﬁndings in the light of TES studies to improve motor performance in healthy subjects, and discuss how personalized TES parameters may be derived from our models.[…]