For upper limb rehabilitation after stroke, two categories of rehabilitation systems will be described: robotic training systems and sensor-based training systems.
Robotic training systems
Therapeutic robotics development started about 15 years ago at which time scientific evidence supporting rehabilitation approaches was much sparser. This has been a difficulty for development of technological rehabilitation systems in the past [
125].
The upper limb robotic systems that exist until today can be classified roughly in passive systems (stabilising limb), active systems (actuators moving limb) and interactive systems [
21]. Interactive systems are equipped with actuators as well as with impedance and control strategies to allow reacting on patient actions [
21]. The interactive systems can be classified by the degrees of freedom (DOF) in which they allow movement to occur.
Existing interactive one-degree of freedom systems are e.g. Hesse's Bi-Manu-Track, Rolling Pin, Push & Pull [
126,
127], BATRAC [
65] & the Cozens arm robot [
128]. These systems are useful for stroke patients with lower functional levels (= proficiency level for skill related movement). Multi-degrees of freedom interactive robotic systems may be useful for patients with lower as well as higher functional levels.
One of the first robotic rehabilitation systems for upper limb training after stroke is
MIT-MANUS developed by Krebs et al [
12,
129]. It allows for training wrist, elbow and shoulder movements by moving to targets, tracing figures and virtual reality task-oriented training. The robot allows two degrees of freedom. This enables training at patient function level, improving e.g. movement range and strength. The patient can train in passive, active and interactive (movement triggered or EMG-triggered) training modes. Patients with all levels of muscle strength can use the system. Visual, tactile and auditory feedback during movement is provided [
12,
125,
130‐
134]. MIT-MANUS has been shown to improve motor function in the hemiparetic upper extremity of acute, subacute and chronic stroke patients in 5 clinical trials (CTs)[
131,
135‐
138] and 5 randomized clinical trials (RCTs) [
139‐
143]. In total 372 persons were tested. This is close to half of the total number of stroke patients tested in technology-supported arm training trials until the end of 2007.
MIME (Mirror Image Movement Enhancer) [
132,
144‐
146] consists of a six degrees of freedom robot manipulator, which applies forces (assistance or resistance as needed) to a patient's hand through a handle that is connected to the end-effector of the robot. This robot treatment focuses on shoulder and elbow function. The MIME system can work in preprogrammed position and orientation trajectories. It can also be used in a configuration where the affected arm is to perform a mirror movement of the movement defined by the intact arm. The forearm can be positioned in a large range of positions and has therefore the possibility to let the patient exercise in complex movement patterns. Four modes of robot-assisted movement are available: passive, active-assisted, active-constrained and bimanual mode. The MIME system has been validated through 1 CT [
147] and 3 RCTs [
145,
146,
148], involving 76 chronic stroke patients.
BI-MANU-TRACK is a one degree of freedom system, designed by Hesse et al [
126,
127,
149] to train forearm pro-/supination and wrist flexion/extension. Training is done bilaterally in a passive or active training mode. No feedback is given to the patient. BI-MANU-TRACK has been validated for subacute and chronic stroke patients in two CTs [
149,
126] and one RCT [
127]. In total 66 persons after stroke were tested.
BATRAC [
65] is an apparatus comprising of 2 independent T-bar handles that can be moved by the patient's hands (through shoulder and elbow flexion/extension) on a horizontal plane. Repetitive bilateral arm training is supported by rhythmic cueing and, where necessary, by assistance of movement. No patient feedback is provided. BATRAC has been tested for chronic stroke patients in one CT [
65] and one RCT [
67]. In total 37 patients were involved.
ARMin [
150‐
153] is a semi-exoskeleton for movement in shoulder (3DOF), elbow (1DOF), forearm (1DOF) and wrist (1DOF). Position, force and torque sensors deliver patient-cooperative arm therapy supporting the patient when his/her abilities to move are inadequate. The combination of a haptic system with an audiovisual display is used to present the movement task to the patient. One small-scale CT [
154] tested the clinical outcome of arm hand function in 3 chronic stroke patients after training with ARMin.
NeReBot [
155,
156] is a 3-degree of freedom robot, comprising of an easy to transport aluminum frame and motor controlled nylon wires. The end of each wire is linked to the patient's arm by means of a rigid orthosis, supporting the forearm. The desired movement is first stored into the system, by moving the patient's arm in a "learning phase" mode. Visual feedback comprises of graphical interface providing a 3D-image of a virtual upper limb on which 3 arrows show desired movement direction during movement. Auditory feedback accompanies the start and end of the exercise. NeReBot has been clinically tested in a RCT [
156] involving 35 acute stroke patients.
AJB or Active Joint Brace [
157] is a light-weight exoskeletal robotic brace that is controlled by means of surface EMG from affected elbow flexor and extensor muscles. It allows for assistance of movement in the elbow joint (1DOF). No feedback about exercise performance is provided. AJB has been tested in a small clinical study, involving 6 chronic stroke patients [
157].
T-WREX is based on
Java Therapy, that was developed by Reinkensmeyer et al [
133]. T-WREX can train increased range of movement and more degrees of freedom, allowing for more functional exercising than Java Therapy does [
19]. An additional orthosis can be used to assist in arm movement across a large, although not fully functional, workspace, with elastic bands to counterbalance arm weight. This makes it suitable for usage by patients with low muscle strength. Position sensors and grip sensors allow feedback on movement [
133] and grip force [
19]. T-Wrex aims to offer training of e.g. following activities: shopping, washing the stove, cracking eggs, washing the arm, eating, making lemonade. Limitations in movement of the shoulder (especially rotations) and forearm (no pro- or supination) cause a discrepancy between functional relevance of the exercise that is instructed and the actual movement that is performed.
Patients and therapists are presented with three types of progress charts: 1) frequency of system usage; 2) performed activity in comparison with customisable target score, average past performance and previous score; and 3) progress overview, which displays a graphical history of the user's scores on a particular activity [
19,
130,
133]. T-Wrex has been validated through a clinical trial, involving 9 chronic stroke patients [
19].
UniTherapy [
158,
159] is a computer-assisted neurorehabilitation tool for teleassessment and telerehabilitation of the upper extremity function in stroke patients. It makes use of a force-feedback joystick, a modified joystick therapy platform (TheraJoy) and a force-feedback steering wheel (TheraDrive).
Four operational modes are used: assessment mode; passive training mode; interactive mode (interaction with telepractitioner) and bi-manual mode (use of two force devices simultaneously).
UniTherapy provides visual and auditive cues in response to success/failure.
Although very engaging, UniTherapy offers movement therapy that is not task-oriented. Apart from moving a car steering wheel, as practised in TheraDrive (Driver's SEAT) [
160,
161], one can question transfer to skilled performance that is needed in everyday life. UniTherapy has been validated for chronic stroke patients in one CT [
161] and one CCT [
14], involving a total of 23 patients.
Haptic Master [
144] is a three degrees of freedom robot, equipped with force and position sensors, that has been used for training arm movements of stroke patients [
162‐
164]. A robotic wrist joint that provides one additional active and two passive degrees of freedom can extend it. All exercises happen in a virtual environment. Performance feedback is provided. The therapist can create virtual tasks. Three different therapy modes are implemented: the Patient Passive mode, the Patient Active Assisted mode and the Patient Active Mode. Therapy is, amongst others, focussing on task-oriented training in a 3D virtual environment as in the GENTLE/S project (reaching to a supermarket shelf, pouring a drink) [
164] or focussing on task-oriented training with real object manipulation as done with ADLER (Activity of Daily Living Exercise Robot)[
163]. A limiting factor for task-oriented training is the device's small range of motion. Two clinical trials provide evidence for improvement of arm hand function after use of haptic master training in subacute and chronic stroke patients [
162,
164]. In total 46 patients have been tested.
Assisted Rehabilitation and Measurement Guide (ArmGuide) is a 4 degrees of freedom robotic device, developed by Kahn et al. [
165‐
168] to provide arm reaching therapy for patients with chronic hemiparesis. An actuator controls the position of the subject's arm, which is coupled to the device through a handpiece. This handpiece slides along a linear track in the reaching direction. Real time visual feedback of the location of the arm (along the track, elevation angles of track, target location) is given to the patient. ArmGuide has been tested in three clinical studies, involving in total 41 chronic stroke patients [
165,
167,
169].
Virtual reality-based hand training systems that have been developed by Burdea et al. are
Rutgers Master II glove and Cyber Glove [
170,
15,
171]. Patients practise by doing one to four hand exercise programs in form of computer games. Each program focuses on different aspects of hand movement: range of movement, speed of movement, individual finger movement or finger strengthening. The exercises are aiming to have a task-oriented component (e.g. grasp virtual ball, piano) but are mostly analytic. Patients receive concurrent haptic feedback, visual feedback and auditory feedback on exercise performance. Also feedback about speed, range, and strength are provided real-time. In total, seven patients were included in two small-scale clinical trials [
15,
171].