From intracranial signals decoding to FES: a first approach using simulation of electrical pulses for objects grasping

Nowadays, there is a growing interest in those people with a lack of limb movement ability, with the aim of giving them the capacity to perform voluntary movements, for instance, through supporting exoskeletons or through electrical muscle stimulation. In some cases, these are people who have undergone amputations of one or more limbs, in other cases they are patients with SCI. The MMG group at Uppsala University, which hosted the development of this project, aims to open the door to a new way of developing BCIs, as suggested by their work in B-CRATOS, an ambitious interdisciplinary project of which several partners are part. The present work falls within this context, with the objective of simulating the muscular stimulation through Functional Electrical Stimulation (FES) technique starting from decoding intracranial signals with Artificial Intelligence (AI). The methodology of this project starts from the processing of spiking activity and hand movement data collected from a macaque monkey while performing the task of grasping objects with different shapes and sizes. The data were collected by the German Primate Center in the context of the B-CRATOS project and made available for this work. Following in the methodology is the analysis of the muscles of the hand and arm involved in the grasping and release task of the objects used in the experiment and then the simulation of the FES signals with the definition of the electrical parameters (waveform, amplitude, frequency, pulse duration, relative delay of the specific muscle in relation to the other muscles). The last stage of the methodology is the search and choice of a proper architecture to decode intracranial input signals and give as output electrical stimulation parameters. After an in-depth literature search, eleven hand and arm muscles were selected, and simulations of the FES signals were performed. Then, electrical parameters to achieve the correct combinations of stimuli were generated for each type of grasping present in the dataset. The preliminary investigation of a suitable Neural Network architecture suggested using a decoder architecture mixing Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) as in previous work in the MMG group. This work, based on the simulation of FES signals, seems to be optimistic in the context of muscle stimulation, which for those who still have undamaged limbs and muscles but not the capacity for voluntary movement is a viable option to be able to perform basic movements.