Neural Engineering

Transformative Technologies

Work Package: WP7
Programme: P17
Deliverable: 17.2 EEG decoders to detect motor anticipatory potentials in stroke patients without inter-session recalibration

Deliverable due date: month 36

This document reports on the progress of work on Deliverable 17.2. All the planned tasks related to this Deliverable have been accomplished. Preliminary results with healthy subjects were submitted to “Journal of Neural Engineering” peer-reviewed journal in December 2014 and published in April 2015 together with results of Deliverable 17.1. Results for stroke patients were published here. Final results are currently under review in “Frontiers of Neuroscience”. The paper was submitted in July 2015. It is currently under discussion since December 2015.

These publications are a scientific part of the Deliverable 10.1 and this “Project Deliverable Report” document focuses on its relation to the NETT proposal.
In the NETT 289146 Grant Annex, we stated in the context of Deliverable 17.1 and 17.2: “P17 is related to robot-mediated motor rehabilitation therapies for patients after stroke, where one of the key points is to proceed with the rehabilitation task only when the patient is engaged and thus using their (damaged) motor cortex. In this context the objective is the development of signal processing and machine learning techniques able to detect in real-time neural processes preceding movement and motor imagery, i.e. intention of movement.”
In the aforementioned publications we study the need for subject- and session-specific calibration prior to the use of motor intention detectors in gait. This common, long and tedious procedure for brain-computer interfaces hinders the applicability of the developed technology (i.e. intention of motion detectors) in rehabilitation scenarios since it limits effective rehabilitation time with a proper closed-loop control. Our first step consisted in evaluating and characterizing the loss of performance between sessions using the MRCP and ERD features proposed in D17.1. These features are not robust enough for session transfer, especially in stroke subjects. In a second step, we have proposed a new type of feature for MRCPs based on phase patterns. This novel representation improves the general performance of the decoder and its latency. More importantly, it allows to reuse a single calibration across multiple subjects or sessions. Interestingly, these features are robust enough to remove subject-specific calibration by using a joint calibration from previous subjects and apply it in new patients.
In such a way we accomplish Deliverable 17.2 and proposed free calibration schemes for rehabilitation therapies that use gait intention detectors. The work also covers milestone 19.

Contributors: Andreea Sburlea, Luis Montesano, Javier Minguez