Reaching Training Based on Robotic Hybrid Assistance for Stroke Patients

This is multi-center randomized controlled trial designed according to the CONSORT Statement recommendations. A total of 68 patients will be recruited in the two centers. This sample size was a-priori calculated as capable to detect a clinically important between-group difference of 5.7 points in the primary endpoint Action Research Arm Test, considering a standard deviation of 12.5, a type I error of 5%, and a power of 80%.

More technical details on the RETRAINER platform for the rehabilitation of the arm are here reported.

The experimental setup consists of a lightweight passive arm exoskeleton for weight compensation, a current-controlled stimulator with 2 channels of stimulation and 2 channels of EMG recordings developed by Hasomed GmbH, and interactive objects, which are daily life objects equipped with RFID (Radio Frequency Identification) tags used to identify the target positions so as to drive the execution of the rehabilitation exercises. A suitable reader is embedded in the exoskeleton with the antenna on the wrist joint. The control system is shared between an Embedded Control System (ECS), running on a BeagleBoneBlackTM, for real-time operation, and a Windows-based table (Microsoft Surface 3 running Windows 8), which provides a graphical user interface (GUI) for the therapist and the patient.

The exoskeleton is characterized by four degrees of freedom (DOFs): three of them, e.g. shoulder elevation, shoulder rotation in the transversal plane and elbow flex-extension, are equipped with angle sensors (Vert-X 13 E, ConTelec AG) to measure the position and electromagnetic brakes to avoid the fatiguing and unnecessary use of FES to hold a target position once reached. The additional DOF is provided by an inclination module, which enables the patient to move the trunk 20° forward without constriction. In addition to the 4 DOFs, the humeral rotation, the prono-supination as well as the length of the forearm and the upper arm can be adjusted at the beginning of the training session at subject-specific positions. The gravity compensation modules for upper arm and forearm consist of a carbon fiber-tube with springs inside whose pre-tension can be adjusted at the beginning of the training session in order to change the level of compensation. Thanks to the adjustability of the lengths and the level of compensation, the exoskeleton can fit and support patients within 5th and 95th female/male percentile. The exoskeleton can be mounted on the user's wheelchair or on a normal chair by means of a universal clamping mechanism which assures easy and stable mounting. The exoskeleton weights about 4kg plus 2kg for the clamping mechanism.

In addition to the support provided by the exoskeleton, EMG-triggered FES is delivered to two muscles, selected by the therapist based on the subject-specific needs. For each stimulated muscle, the residual volitional EMG signal is detected and used to trigger the onset of a predetermined stimulation sequence applied to the muscle itself. In case the muscle does not reach the pre-defined threshold, the stimulation sequence is automatically started after a time-out. EMG signals are acquired at 4kHz, the stimulation frequency is set at 25Hz, the pulse width is fixed at 300µs, while the stimulation intensity is set at the beginning of the training session on each muscle individually at a value tolerated by the subject and able to induce a functional movement. Separate EMG and stimulation (Pals® electrodes, Axelgaard Manufacturing Ltd) are placed over each muscle belly. When the stimulation starts, EMG signals are continuously measured in order to provide a visual feedback about the patient's volitional involvement at the end of the execution of each task. An adaptive linear prediction filter is used to estimate the volitional EMG during hybrid muscle contractions. If the mean value of the volitional EMG estimate during the stimulation phase is over a pre-defined threshold, a happy emoji is shown to the patient through the GUI; conversely, if it is below the pre-defined threshold a sad emoji is shown in order to promote the active involvement of the subject. A fast and automatic calibration procedure is required before the beginning of each session. This procedure aims at setting the current amplitude and the EMG threshold values. During the procedure the subject is asked to be relaxed. Specifically, three thresholds are set on each muscle: two of them are used to trigger the stimulation, one in case the muscle is activated as first and one in case the muscle is activated as second one; the third threshold is used to define the subject's active involvement in the task. The thresholds are defined as twice the mean volitional EMG during a phase of no stimulation (first threshold), during a phase of stimulation of the other muscle (second threshold), and during a phase of simultaneous stimulation of the two muscles (third threshold).

The control interface of the system, implemented in .Net 4.6, provides a GUI including multiple software tools to organize rehabilitation exercises and monitor rehabilitation progress. The heart of the control interface is a State Machine, which drives both the parameterization and the execution of the exercises. Each exercise is divided into single tasks: the State Machine drives the exercise execution throughout the tasks, while the execution of each single task is controlled by the ECS. The ECS controls all the modules requiring real time constraints, such as the stimulator, the FES controller and the exoskeleton sensors. To keep the control interface and the ESC synchronized, a strict master slave concept using a custom made communication protocol was implemented, meaning that the ECS must not act independently, but only reacts to commands sent by the high level control. Transitions between states of the state machine and thus tasks of the exercise are triggered by angle sensors data, RFID data or a timer (depending on the task). Transitions have to fulfill certain conditions, so called guards. These guards are predefined for each task and have to be parameterized as described in the Section D. The GUI guides the user through the training by providing visual instructions and feedback.

The workflow of a typical training session consists of four main phases: the setting, donning and parameterization of the system, and the training following a pre-defined sequence of exercises. The control interface supports the therapist and the patient throughout all the phases via the GUI.

The setting starts with the therapist creating a new user, or selecting an existent one, and selecting the exercises. Afterwards, the donning phase starts with the placement of the EMG and stimulation electrodes. Once the electrodes placement is checked, the therapist should adjust the exoskeleton lengths to fit with the patient and let the patient don the exoskeleton. The following step is the calibration of the FES controller by means of the automatic procedure previously described. The therapist sets the gravity compensation both at the arm and forearm level and saves the final exoskeleton settings. On the following training days, the setting and donning procedure is partly simplified since the therapist can load the settings of the previous day and eventually adjust them.

The parameterization step is designed to set the guards of the State Machine. In this process the GUI guides the patient and the therapist through each task of the selected exercises without stimulation. The patient-specific parameters for each task, such as the target positions, the desired time for the execution of each task, and the time of the relax phases, are determined. At the end of the parameterization phase, all the parameters are stored and the training session can start.

The training consists of the execution of a series of exercises involving the arm during daily life activities. Typical exercises are anterior reaching on a plane or in the space, moving an object on a plane or in the space, moving the hand to the mouth, with or without an object in the hand, and lateral elevation of the shoulder. The execution of the exercises is controlled by the control interface which leads the patient throughout the single tasks by means of both visual and audio messages via the GUI.