Most references developed or used sensor gloves in their studies. Sensor-based devices is a very broad category since many sensors can be used to monitor hand function. The different types are broken down in Figure 7. Bend/flex sensors and inertial measurement unit (IMU) sensors are the standard for monitoring movement and relative location because they are inexpensive and easily available. They also produce measurements that can be monitored in real time.

Resistive bend sensors are employed in the commercially available “Cyberglove” device (CyberGlove Systems LLC, San Jose, CA) (Merians et al., 2009; Lee, 2014; Thielbar et al., 2014; Dimbwadyo-Terrer et al., 2016; Ranganathan, 2017; Jarque-Bou et al., 2020; Padilla-Magana et al., 2022). The CyberGlove monitors all individual finger joints. Other commercially available gloves are the P5 Glove (Sivak et al., 2009) (Mindflux, Australia) and 5DT Sensor Glove (5DT Technologies) (Golomb et al., 2010). While these gloves are effective at monitoring hand function, they can be relatively expensive (USD 190 to 2,500 or more) compared to standard video monitoring systems, which may limit their use in at-home rehabilitation (Merians et al., 2009; Sivak et al., 2009; Golomb et al., 2010; Lee, 2014; Thielbar et al., 2014; Dimbwadyo-Terrer et al., 2016; Ranganathan, 2017; Jarque-Bou et al., 2020; Padilla-Magana et al., 2022). Besides commercial bend/flex sensing devices, there are custom devices that address limitations of commercial gloves (Gentner and Classen, 2009; Knutson et al., 2009; Wille et al., 2009; Oess et al., 2012; Jung et al., 2017; Fu et al., 2019; Huynh et al., 2019; Burns et al., 2021; Kim et al., 2022; Syeda et al., 2022). For example, Oess et al. (2010, 2012) developed the NeuroAssess Glove, which has four bend sensors (Flexpoint Sensor Systems, Inc., Draper, UT) to monitor the individual phalanges of the index finger and thumb. The sensors were accurate to within 3 degrees (95% confidence interval) (Oess et al., 2010). Some groups are combining other sensors with bend/flex sensors (Wille et al., 2009; Burns et al., 2021). For example, Wille et al. (2009) custom-designed gloves with bend sensors, accelerometers, and magnetometers placed on the forearm and individual fingers. The glove was meant to be integrated into a virtual reality game to treat children with neuromuscular disorders. A limitation is that the glove is designed for simplicity rather than accurate tracking of movement (Wille et al., 2009). From the reviewed articles, the bend/flex sensors were found widely in research and commercial products. The commercially available gloves are costly, while the gloves designed by research groups have not been fully validated and ready for clinical use.

IMU sensors are also very common in research studies and in commercial gloves (Cavallo et al., 2013; Li et al., 2016; Lemos et al., 2017; Mohan et al., 2018; Salchow-Hommen et al., 2019; Bhagubai et al., 2021; Fei et al., 2021; Kamockij and Fedorov, 2021; Pan et al., 2021; Schwarz et al., 2021; Hwang et al., 2022). They incorporate gyroscopes, accelerometers, and magnetometers to track the wearer's orientation. In regard to commercial systems, Schwarz et al. (2021) used a motion capture system called Xsens MVN Awinda (Xsens Technologies, Netherlands). Custom-designed IMU gloves have also been developed to improve on commercial gloves (Cavallo et al., 2013; Li et al., 2016; Lemos et al., 2017; Mohan et al., 2018; Salchow-Hommen et al., 2019; Bhagubai et al., 2021; Fei et al., 2021; Kamockij and Fedorov, 2021; Pan et al., 2021; Hwang et al., 2022). For example, the i-Glove has embedded IMU sensors on the wrist, index finger, middle finger, and thumb (Mohan et al., 2018). Fei et al. (2021) developed a data glove system with IMU sensors on the individual finger joints. However, they found that IMU sensors can cause potential errors in sensor-to-segment calibration of location and defining the global frame of reference for the sensors (i.e., the sensors' origin in space) (Fei et al., 2021). While IMU sensors are most commonly used, there are cases where a glove has only one of the three measuring devices, the accelerometer, for instance (Gerber et al., 2016). Additionally, some kinematic sensors track spatial and temporal variables (i.e., position, velocity, etc.) of finger and hand movement (Wille et al., 2009; Avanzino et al., 2011; King et al., 2011; Bonzano et al., 2013; Gerber et al., 2016; Bisio et al., 2017; Jung et al., 2017; Adams et al., 2019; Bonassi et al., 2020). The most common system used was the Glove Analyzer System (GAS) (Avanzino et al., 2011; Bonzano et al., 2013; Bisio et al., 2017; Bonassi et al., 2020). GAS attaches these sensors onto the individual finger joints. All the articles that used the device were focused on intervention or hand rehabilitation at home in which this glove was used to monitor physical movement. No specific limitations of the glove were noted, but they are not one-size-fits-all, which means that the end user must select the closest size to their hand. In general, these kinematic sensor measurements are meant to track the velocity of finger movement during a given task (Avanzino et al., 2011; Bonzano et al., 2013; Bisio et al., 2017; Bonassi et al., 2020). From the reviewed references, the IMU sensors were found to be commonly used in research and commercial systems, specifically for motion capture systems. IMU sensors were found to have errors during calibration due to external magnetic fields, therefore they are not very reliable. Similar to the flex/bend sensors, the commercially available gloves are very costly, and the ones currently being developed in research have not undergone extensive testing for validation.

Another noteworthy sensor is the force sensor, which measures the cumulative normal force applied to its surface (Friedman et al., 2014; Signori et al., 2017; Wachter et al., 2018; Wolbrecht et al., 2018; Mawase et al., 2020; Burns et al., 2021; Patane et al., 2022; Urone et al., 2022). In the ergonomic device designed by Mawase et al. (2020), isometric force sensors are placed on the fingertips including the thumb. This was used to test the importance of two different finger functions—maximal voluntary contraction force and individuation (i.e., applied pressure at each finger). They found that rehabilitation of the flexor synergies (i.e., spasticity) could be meaningful in stroke patients' hand function recovery (Mawase et al., 2020). Force sensors are important for determining grip force of the fingers, a key component of overall hand function. However, they have not been widely incorporated into hand worn devices in research and especially in commercially available devices.

The last type of sensor, the electromyography (EMG) electrode, which measures electrical potential changes associated with muscle contraction, was only mentioned in two references per our search criteria as part of a hand-worn sensor glove (Linderman and Rupasov, 2012; Pan et al., 2021). One study used EMG electrodes in a sensor glove to monitor activity in three muscles of the hand. The glove was designed for measurements to be recorded, synced, and stored with EEG for biometric assessments. These assessments focus on showing the correlation between predicted effort (motor imagery (MI) or volitional effort) and physical movement. The importance of this type of system is to show the wearer that even though they may have minimal physical motor function, their effort is measurable and can be harnessed. This EMG sensor glove was the only such device found to be integrated with a BCI system (Linderman and Rupasov, 2012). The limitation with using EMG as a sensor is the difficulty of predicting fine movements of the fingers. This review found that EMG sensors are rarely applied to the hand muscles for monitoring finger movement during muscle contraction.

Overall, sensor-based gloves were used to monitor hand function and provide feedback on the physical movement or applied pressure of the hand. These devices can be used for functional assessment to determine the severity of the impairment or to track a patient's progress through rehabilitation. Due to the simplicity of these devices, they can be integrated into a game (e.g., VR rehabilitation games) to give positive feedback to the patient and motivate them to continue their rehabilitation as well.

Assistive devices were the second most investigated type of device and fell into five categories: pneumatic actuators, hydraulic actuators, electrical actuators, elastic effectors, and end-effectors (refer to Figure 8). Electrical actuators appeared to be the most used component of assistive devices (de Araujo et al., 2011; Triandafilou et al., 2011; Iwamuro et al., 2015; Sallum et al., 2015; Fischer et al., 2016; Thielbar et al., 2017; Bernocchi et al., 2018; Chen et al., 2020; Leuthardt et al., 2020; Osuagwu et al., 2020; Yurkewich et al., 2020; Crema et al., 2022; Guo et al., 2022; Luhmann and Randazzo, 2022). A couple of references used a version of the commercially available GloReha glove (Gloreha IDROGENET, Italy), which uses electrical actuators to move individual finger joints. These gloves were used in interventions for at-home hand rehabilitation (Bernocchi et al., 2018) and FES treatment (Crema et al., 2022). Although these studies showed that the interventions improved impaired hand function, there are key limitations. The GloReha glove cannot be worn by patients with severe hand spasticity; and the motor tended to fail if used for long periods (Bernocchi et al., 2018; Crema et al., 2022). An example with non-commercial electrical actuators is the device made by Leuthardt et al. (2020), meant to be used with a BCI to track rehabilitation progress. However, the fingers were actuated in pairs tied together (i.e., index and middle, ring and pinky) and not individually (Leuthardt et al., 2020). Although electrical actuators are commonly used, they are limited in their ability to assist with natural finger movements.

Pneumatic actuators were also commonly used in assistive devices for the hand (Coffey et al., 2014; Thielbar et al., 2014; Yap et al., 2016; Khallaf et al., 2017; Cheng et al., 2020; Jiryaei et al., 2021). A common example is the pneumatic exercise glove (PneuGlove), which induces movement of the whole hand (Coffey et al., 2014; Thielbar et al., 2014). Inflatable bladders hyperextend the hand and induce dorsiflexion of the wrist. Although the glove is comfortable, portable, adjustable and does not obstruct movement, it cannot control individual fingers, only all of them together (Coffey et al., 2014; Thielbar et al., 2014). Other pneumatic actuator devices have been developed for rehabilitation (Cheng et al., 2020; Jiryaei et al., 2021). Cheng et al. (2020) used air pressure actuators in a BCI-based soft robotic glove that assists individual finger movements excluding the thumb. Their protocol used BCI-assisted motor imagery in robotic rehabilitation. They compared the use of the soft robotic glove to assist movement with and without a BCI in the loop. However, neurotypical individuals without hand impairment were not assessed to provide a performance baseline (Cheng et al., 2020). While pneumatic actuators were found to be useful in rehabilitation strategies, they have not been thoroughly validated with a healthy control group or generalized to a range of stroke acuities.

Other assistive devices include hydraulic actuators (Gallo et al., 2017; Yeow et al., 2019; Wijesundara et al., 2021) and elastic effectors (Biggar and Yao, 2016; Lieber et al., 2022). The hydraulic actuators can mimic the articulation of the human hand (Wijesundara et al., 2021). They have also been used in feedback loops. For example, Gallo et al. (2017) developed a multimodal haptic device that uses hydraulic actuators on the individual fingers and feedback sensors on the fingertips. The hydraulic actuators would react to the input modalities (e.g., specific hand postures) and move the fingers accordingly; but the device is bulky, which can restrict movement (Gallo et al., 2017). The elastic assistive devices use elastic bands (Biggar and Yao, 2016) to extend the fingers. The main issue with these devices is that they alter natural grasp configurations. The main finding was that the hydraulic actuators were found to be similar to the electrical and pneumatic actuators but restrict natural hand movement more. The elastic assistive devices were even more constricting to gross hand movement and therefore, are limited in the amount of assistance they can provide to the individual with impaired hand function.

Lastly, there are end-effector devices, of which only one example was found. Zbytniewska et al. (2021) used an end-effector robot for the index finger alone, which is only one degree of freedom (DOF). As mentioned previously, the human hand and wrist have 21 DOF, which means that rehabilitation of all finger movements will be very time-consuming with this one-DOF end-effector robot (Jones and Lederman, 2006; Zbytniewska et al., 2021). This review found confirmation that the end-effector devices are restricted on how much assistance they can provide to an individual's hand/finger movement.

In summary, the assistive devices reviewed were used for rehabilitation to aid in execution of movement. They can provide physical feedback to the patient based on induced and self-initiated movement. Induced movement involves a device physically moving the fingers rather than by stimulation to produce a muscle contraction. These devices were helpful in improving finger and/or hand function.

As shown in Figure 6, FES devices make up 8% of the references (Golaszewski et al., 2012; Krukowska et al., 2014; Lin et al., 2014; Sullivan et al., 2015; Kattenstroth et al., 2018; Friedenberg et al., 2022). They were commonly used in rehabilitation and several clinical studies tested their capabilities in treatments for hand impairments. A common hand-worn FES device is the Mesh Glove (MG), which stimulates muscles over the whole hand (Golaszewski et al., 2012; Lin et al., 2014). The MG was used in interventions with mirror therapy (Lin et al., 2014) and one group aimed to test the glove's utility in rehabilitation (Golaszewski et al., 2012). Another version of FES glove only targeted hand extensor muscles (i.e., three muscles on the hand). This group, Krukowska et al. (2014) compared their glove with standard bipolar surface electrodes placed on the forearm. They found that their device helped restore motor function in the forearm as well as the hand (Krukowska et al., 2014). Another group tested the usability of their FES glove, which delivered stimulation to the hand via embedded electrodes, on chronic stroke patients. It was found difficult to use when completing the intervention tasks (Sullivan et al., 2015). Another group developed a glove that used 20 Hz electrical stimulation from built-in electrodes to target individual fingers rather than the whole hand. Kattenstroth et al.'s (2018) study tested the improvements in hand motor function after a three-and-a-half-week intervention with and without this glove and found it to be more effective than standard therapies. These studies were tested only on mild to moderate, but not severe hand impairments. Therefore, they cannot be used to help the severely impaired individuals.

Overall, FES devices were found to be beneficial in rehabilitation for individuals who have impaired hand function. A lot of these devices used triggers (e.g., unimpaired hand, protocol cues) to stimulate the impaired hand at the appropriate time. FES devices can be paired with conventional therapies (e.g., mirror therapy) or customized rehabilitation protocols.

Combinations of the three major device types were as common as FES devices alone (refer to Figure 6) (Merians et al., 2009; de Araujo et al., 2011; Thielbar et al., 2014; Gallo et al., 2017; Hoffman et al., 2017; Wolbrecht et al., 2018; Huynh et al., 2019; Chen et al., 2020; Osuagwu et al., 2020; Kim et al., 2022). Most of these combine sensors and assistive devices because they typically use the former in a glove to control the latter. For example, Kim et al. (2022) used a glove with bend sensors and a 4 DOF soft robotic glove focused on the thumb and index finger. A sensor glove worn on the unimpaired hand was used to help rehabilitate the impaired hand wearing a soft robotic assistive glove (Kim et al., 2022). Another combination involved FES and a sensor-based device, which provided visual feedback to the participant on FES-triggered hand function. Fu et al. (2019) used a kinematic sensor glove for the unimpaired hand, biphasic current-controlled transcutaneous stimulator for the FES, and a fingerless mitten with bend sensors for the impaired hand (Handana Corp., Austin, TX). The sensor glove was used to provide input movement to the FES stimulator. The fingerless mitten sent the impaired hand's movement to a virtual game for rehabilitation. In this way, self-administered FES treatment and video-gaming was possible for a less impaired individual (Fu et al., 2019).

Additionally, Knutson et al. (2009) developed a custom-built command glove with bend sensors on the thumb, index, and middle fingers. The glove was designed for monitoring movement during an FES intervention. The limitation to this device was that it only monitored three fingers and not the whole hand. Therefore, this glove does not provide a complete characterization of hand function (Knutson et al., 2009). The last combination involves FES and passive devices. Passive devices are only used to support the upper extremity but do not facilitate or monitor hand function in any way. In this review, passive devices appeared in one reference: Uswatte et al. (2006) described two different devices—hand splint and fingerless gloves. Their study investigated the effects of different types of training and restraints on recovery of the impaired hand. They found that using or not using these supportive devices made no difference (Uswatte et al., 2006). In the context of this device combination, passive devices provide support to the hand, while the individual is subject to FES intervention. For example, Hoffman et al. (2017) developed an orthotic device with support sections and electrodes on the forearm, hand, and fingers. The limitation of this combination is that the glove cannot monitor physical responses to the electrical stimulation.

In summary, these device combinations were proven useful in their targeted goals (i.e., assisting with movement, etc.). Majority of these combinations used a sensor glove as an input and either a FES or an assistive device as the movement initiator. These system can be very useful in motivating and providing proprioceptive feedback to the user since their unimpaired hand is used to trigger the assistive or FES device on their unimpaired hand.

In addition to combinations of the main type of devices (e.g., sensor and assistive) presented in this review, there were three instances in the literature of combinations of devices integrated into a BCI system (Linderman and Rupasov, 2012; Leuthardt et al., 2020; Guo et al., 2022). Guo et al. (2022) used an assistive device with pneumatic actuators to move the fingers while the subject's EEG was monitored. Particular features detected on the scalp over the occipital lobe when an image was shown to the patient would trigger closing and opening of fingers of the hand by pneumatic actuators. After training with the system, the patients' clinical assessment scores were found to improve significantly (Guo et al., 2022). This BCI-assistive system focused on extracting EEG features over the occipital lobe and not the premotor and primary motor cortex. This is an unconventional choice because it focuses on visual attention—specifically steady-state visual evoked potentials (SSVEPs)—rather than motor pathways associated with hand function. Guo et al.'s (2022) system requires little training and targets different neurological mechanisms from MI or volitional controlled BCIs. As mentioned in the introduction, the premotor and primary motor cortex are responsible for volitional control of hand movement. The only reference that used a sensor glove with EEG was by Linderman and Rupasov (2012). They developed an EMG sensor glove that monitored activity in three muscles of the hand. They synced the EMG and EEG signals to detect potential biomarkers of Parkinson's Disease (PD). Using only three channels of EMG can limit the ability to detect the PD biomarkers in physical finger movement (Linderman and Rupasov, 2012). These references found that using brain activity to be integrated with a hand-worn device was beneficial to the patient by improving motor function and gain vital information on the relationship between finer hand movements and the brain. These types of studies lay the foundation for BCI development.

Overall, these device combinations proved effective in rehabilitation interventions. The sensor-based gloves can be used to provide input (i.e., the trigger or controller) to an assistive device, an FES device, or a monitoring device, and to confirm movement or applied pressure. This dual utility of sensor-based gloves allows them to aid in both rehabilitation and assessment of hand function. The FES and assistive devices can only be used for rehabilitation when not paired with a sensor-based device. The integration of these devices with EEG was also found to be helpful in both assessment—when paired with a sensor-based device—and rehabilitation. All of these devices have been used in motivational therapies (i.e., games) to provide feedback to the wearer in either assisting with activities or confirming that the user completed specific tasks.