Executed and imagined grasping movements can be decoded from lower dimensional representation of distributed non-motor brain areas

Abstract

Using brain activity directly as input for assistive tool control can circumvent muscular dysfunction and increase functional independence for physically impaired people. Most invasive motor decoding studies focus on decoding neural signals from the primary motor cortex, which provides a rich but superficial and spatially local signal. Initial non-primary motor cortex decoding endeavors have used distributed recordings to demonstrate decoding of motor activity by grouping electrodes in mesoscale brain regions. While these studies show that there is relevant and decodable movement related information outside the primary motor cortex, these methods are still exclusionary to other mesoscale areas, and do not capture the full informational content of the motor system. In this work, we recorded intracranial EEG of 8 epilepsy patients, including all electrode contacts except those contacts in or adjacent to the central sulcus. We show that executed and imagined movements can be decoded from non-motor areas; combining all non-motor contacts into a lower dimensional representation provides enough information for a Riemannian decoder to reach an area under the curve of 0.83 ± 0.11. Additionally, by training our decoder on executed and testing on imagined movements, we demonstrate that between these two conditions there exists shared distributed information in the beta frequency range. By combining relevant information from all areas into a lower dimensional representation, the decoder was able to achieve high decoding results without information from the primary motor cortex. This representation makes the decoder more robust to perturbations, signal non-stationarities and neural tissue degradation. Our results indicate to look beyond the motor cortex and open up the way towards more robust and more versatile brain-computer interfaces.

The Easy and Versatile Neural Recording Platform (T-REX): Design and Development Study

“…Subsequently, experiment initializes a recorder instance and adds all streams to the list of streams it should record. For a movement experiment [19][20][21][22][23], the streams recorded could be the neural amplifier and experimental triggers. Additionally, a movement tracker [24][25][26] or a force sensor [27] could be added.…”

“…We see no indications of different data quality in our neural decoding endeavors. We can decode speech [44,45] and movement trajectories [46] with performance equal to that using our previous setup.…”

The Easy and Versatile Neural Recording Platform (T-REX): Design and Development Study (Preprint)

“…We see no indications of different data quality in our neural decoding endeavors. We can decode speech [44,45] and movement trajectories [46] with performance equal to that using our previous setup.…”

T-Rex: sTandalone Recorder of EXperiments; An easy and versatile neural recording platform

“…Subsequently, initializes a Recorder instance and adds all streams to the list of streams it should record. For a movement experiment ([12], [13], [14], [15], [16]), the streams recorded could be the neural amplifier, experimental triggers, and a movement tracker could be potentially added ([17], [18], [19]) or a force sensor ([20]. For speech perception ([21], [22], [23]) or auditory perception ([24], [25]) the audio stream, experiment triggers and neural data.…”

“…This experiment is a simple text-based instruction for a grasping task (Figure 4A). The participant is prompted by text in a Python tkinter 5 window to continuously open and close either their left or right hand, as used in [13]. The experiment requires neural data as the input device and generates a StreamOutlet to send markers that inform about the start and end of the experiment and of the trials.…”