Individualized hand motor training under suitable brain states to improve performance and learning after stroke

• Applicants: Dr. Michael Tangermann; Professor Dr. Cornelius Weiller, since 10/2019
• Subject Area: Clinical Neurology; Neurosurgery and Neuroradiology ; Image and Language Processing, Computer Graphics and Visualisation, Human Computer Interaction, Ubiquitous and Wearable Computing ; Biomedical Systems Technology ; Human Cognitive and Systems Neuroscience
• Term: from 2017 to 2023
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 387670982

Rapid progress in neurotechnology and machine learning enables closed-loop interaction with the brain and unlocks the door to a new class of clinical applications, e.g. for stroke patients. This group exhibits strong performance fluctuations in repeated trials of hand motor training. Utilizing single-trial data analysis from the field of EEG-based brain-computer interfaces, we found evidence, that performance variability can partly be explained and even temporally predicted by exploiting individual oscillatory brain activity. It represents an objective feature (neural marker) for decoding task-specific suitable brain states. Current decoding algorithms, however, reveal unsolved challenges: as features need to be optimized individually, substantial training data is required. Furthermore, the achievable decoding quality varies over time due to non-stationarities and brain states often lack clinical interpretability, which impedes cross-subject comparisons. The project SuitAble addresses these challenges by dictionary learning to represent oscillatory subspaces in combination with transfer learning between data sets.In a clinical proof-of-concept study, the envisioned algorithmic advances will be utilized to enable a novel brain state-dependent closed-loop training, which will be evaluated for chronic stroke patients. The training extends an established hand motor training protocol by equipping it with a novel single-trial strategy exploiting the ongoing brain state. As explicit feedback about the neural markers will be provided within the training, we expect patients to learn how to actively adopt suitable brain states and thus accelerate their motor skill learning.Within the runtime of SuitAble, we will employ the novel brain state-dependent training approach: we investigate its feasibility for stroke patients, test its efficacy and compare its efficiency to a hand motor training disregarding brain states. Furthermore, dictionary-based brain state decoding will permit to describe clinically interpretable training-induced changes of functionally relevant oscillatory processes, e.g. in the sensorimotor and visual attention domain, and allow for comparison with healthy controls.The expected merits of the interdisciplinary project are threefold. First, the improvement in decoding methods will allow to shorten calibration periods, run closed-loop protocols with stable decoding performance and yield data-driven characterization of oscillatory processes, which accompany successful motor learning in clinical and non-clinical scenarios. Second, the evaluation of the decoding methods in a clinical setting will deliver a large amount of data, which will support further algorithmic development for closed-loop interaction protocols in the field of neurotechnology. Third, the novel brain state-dependent training approach will pave the road to more individualized motor rehabilitation paradigms and may spark novel training strategies in sports science.

• DFG Programme: Research Grants
• Ehemalige Antragstellerin: Dr. Janine Reis, until 9/2019