To develop our VR-based CIT program, Kinect for Windows SDK was used to provide an Application Programmer’s Interface (API) to the Kinect 2 sensor. The Kinect 2 sensor includes an RGB camera and an infrared emitter with an infrared depth sensor, which allows measurement of the movements of players in three dimensions and provides real-time feedback on movements while the game is played. The Kinect 2 API’s skeletal tracking functions provide position estimates for 25 anatomical landmarks on the head, neck, trunk, shoulders, elbows, hands, thumbs, hips, knees, ankles and feet with a frequency of 30 Hz. The game was developed using Unity, with computer graphics created in 3ds Max for three-dimensional (3D) animation and Photoshop for the user interface.

This game was designed by a research team of occupational therapists, engineers and designers. The theme of this game was based on a feasibility survey of ten children with CP to determine the preferred game type prior to the game design. Based on the children’s favorite type of game, a game called “Adventure Island” was developed. In this game, a child acts as a warrior to defeat all the monsters and maintain peace on the island. The missions of the warrior are to collect cannonballs and then place the cannonballs in a fort to defeat the monsters (Figure 1).

This game was developed for the UL rehabilitation of children with unilateral CP. Previous studies revealed that children with CP have difficulty in shoulder flexion, elbow extension, forearm supination, and voluntary movement of the fingers of the affected side (Bhardwaj and Sabapathy, 2011). In accordance with CIT protocol, the categories of arm-reaching, manipulation, and arm-hand tasks were addressed (Gordon et al., 2005; Abd El-Kafy et al., 2014). Therefore, the motor training goals of “Adventure Island” were to train the following movements of the affected UL: reaching, grasping, releasing, holding and aiming (Figure 2). In the game, before defeating the monsters, the player has to reach for and grasp the cannonballs. After collecting sufficient cannonballs, the player has to hold and/or aim the cannon to attack the monsters. Moreover, a pause mechanism during gameplay prevents inadequate compensatory movements (e.g., trunk flexion instead of upper limb movement) and ensures the player’s safety. The Kinect system can detect abnormal postures and movement patterns of the player during gameplay. If the player exhibits compensatory movement of the trunk, the game pauses and a warning appears on the screen to instruct the user to assume the correct position.

Two key components of CIT are restraint of the less-affected UL and intensive structured training of the affected UL. A contextual restraint is used instead of a physical restraint. The less-affected limb is gently restrained (i.e., the less-affected hand is placed on one knee) to accomplish in-game actions such as placing a hand on a box for support or pressing a button on a shield to defend against attacks by monsters (Figure 1). If the player uses the less-affected hand to play the game, the game pauses and a pop-up warning appears on the screen to instruct the player to put the less-affected hand on a knee.

To incorporate the concept of intensive structured training of the affected UL required for CIT, five strategies from motor learning theory were applied. First, the intensity of training is achieved by repetitive practice of the tasks in the game. For example, before defeating the monsters, the children have to collect cannonballs by repeating the tasks of reaching, grasping and releasing. Second, visual and auditory feedback are incorporated into the game design to encourage the children to participate (Figure 1). Knowledge of results (KR, information about the outcome of performing an action) and knowledge of performance (KP, information about the movement characteristics leading to the performance outcome) are two categories of the feedback. Both KR and KP are valuable for facilitating motor learning (Chen et al., 2016) and thus were adopted in our game design. In the game, the score indicating the defeated monsters is KR, while the performance of grasping cannonballs is KP. Since too much feedback may interfere with motor learning and retention (Hemayattalab and Rostami, 2010), 50% feedback (providing feedback for half of the trials) was chosen for the game. Third, a shaping skill was used for game design. It emphasizes progressively increasing the task difficulties in temporal and spatial dimensions, as well as accuracy of movement (Abd El-Kafy et al., 2014). During the cannonball gathering stage, cannonballs randomly appear in four directions relative to the player: inward, outward, upward and downward. There are three levels of game difficulty. At the game with higher difficulties, the movements are more complex and more challenging in temporal, spatial and accuracy (Figure 2). For example, in level 1, the player defeats the monster by putting cannonballs into the fort. When the child’s ability increase, the difficulty in spatial is increased in terms of moving the positions of the cannonballs more far away from the warrior. In level 3, the player must aim at the monster and hold that position for 3 s to inflict the greatest amount of damage to the monster. The temporal, spatial and accuracy difficulties of the required movements are higher than those of level 1. Moreover, to ensure smooth game progress and to improve the child’s sense of accomplishment, if the Kinect system cannot detect movement during the required time due to the child’s poor motor capacity, the task is automatically accomplished by the system. Fourth, personalized game difficulty was adopted to provide just the right amount of challenge for each child with CP according to his/her motor ability. At the beginning of the game, the maximal reaching distances of both hands are measured by the Kinect system. The children with CP are guided by bubbles to reach in five different directions, namely, outward, inward, upward, downward and forward, for 15 s in each direction. When the child reaches the bubble, it moves farther and farther away to induce the child’s maximal reaching distance without trunk movement. After the evaluation, the personalized game difficulty is decided. Fifth, user interfaces for both the therapists and the children were designed. In the therapist’s interface, a therapist can choose the difficulty of the game level to fit the child’s ability. In addition, the system outputs data on the child’s performance for the therapist to understand the progress of the intervention. For the children’s interface, a leaderboard is provided so that the children can see their own performances and compare them with those of other children playing the game, adding an element of competition that might help to inspire them. Moreover, a weapon collection book is provided to enhance the motivation to play the game. Upon finishing the game, child is awarded a new style of cannonball, and this cannonball appears in the subsequent gameplay session.

The demographic data of the recruited participants are listed in Table 1A. No adverse events occurred during the gameplay. During the 20 min gameplay sessions, the children performed a mean of 72 grasping actions (range 54–108 repetitions). Two children completed all three levels of the game in 20 min. Six children finished the first 2 levels. Two children completed only the first level.

Range of motion (ROM) of shoulder flexion, abduction, adduction, and elbow flexion, as well as wrist flexion and extension, are reported in Table 2A. Compensatory movements (i.e., trunk flexion) were also recorded (Table 2A). According to the results of the questionnaire, all of the children reported a positive experience (Table 2B), and all felt safe during the gameplay. Most of the children preferred the Kinect-based upper limb motor training system (75%) over the regular intervention (25%).

The demographic data of the recruited participants are listed in Table 1B. For the performance in the game, the mean performance score at baseline was 60.30% (SD = 18.17%). The performance score increased significantly at the third (p = 0.002) and fifth (p = 0.002) weeks from baseline, and progression was stable from the fifth week to the last (week 6: p = 0.042; week 7: p = 0.001; week 8: p = 0.001). For the BBT, the mean score at baseline was 22.33 blocks (SD = 20.93). The mean score of the BBT increased significantly at the third (p = 0.001) and the fifth (p = 0.004) weeks from baseline, and progression was stable from the fifth week to the last, which was consistent with the performance scores (week 6: p = 0.019; week 7: p = 0.003; week 8: p = 0.005) (Figure 3) .

Restraint of the less-affected UL without negative emotion was successfully achieved in our program. The Kinect, a motion capture sensor, allows the design of a limb-specific game wherein the player must play the game using specific movements of the limb specified by the game designer. In our program, children with unilateral CP were asked to play the game using only their affected UL. Moreover, to prevent negative emotion resulting from the restraint, contextual restraint methods were adopted. In our program, when the children with CP put their less-affected hand on one knee, the box is held stable during bomb collection or the shield is raised to defend against attacks by monsters. According to the results of the questionnaire, all of the children enjoyed the gameplay and thought highly of it (>8). We found that during the gameplay, all of the children were involved in the context of the game and actively used their affected UL to play it. No children complained about the game requirements or refused to use their affected ULs to play the game.

Intensive training of the affected UL, one of the goals of CIT, was also achieved by our program. Multi-directional reaching and grasping were the two motor training tasks in our program. In view of this, we wanted to ensure that this game could induce children to perform the correct movements instead of compensatory movements. Recently, many commercially-available VR systems have been broadly used for rehabilitation in children with CP. However, some limits of the system itself may reduce the sensitivity of detection. For example, the Wii console only detects the acceleration of the remotes, so children can accomplish tasks by performing compensatory movements (Pastor et al., 2012). Thus, a Wii-based game might not be able to train ideal movements because the children only need to perform wrist or finger movements to accomplish all the tasks. A VR-based game using the Kinect system can eliminate this disadvantage. Based on the motion analysis data in Table 2A, our Kinect-based program induced active ROM of all UL joints. This result was in line with our training goal, namely, that the children should reach for and grasp cannonballs in multiple directions. In addition, the maximal angles of trunk flexion in the children with CP were greater than 30°. Trunk movement is a common compensatory movement during forward reach in children with CP (Jaspers et al., 2011). Thus, it is necessary to monitor trunk movement to prevent compensatory movement during gameplay, and a corresponding pausing mechanism triggered by excessive trunk movement is indispensable for postural correction. Moreover, according to the Kinect-based program, the children with unilateral CP could achieve an average of 72 repetitive grasps within 20 min of gameplay. This intensive amount of training is difficult to achieve in conventional training. Since intensive training is crucial to causing neural reorganization in the brain and improving the learning of motor and functional skills (Pereira et al., 2014), our Kinect-based CIT program can provide the opportunity for intensive and repetitive practice.

Strategies from motor learning theory, including the structured training required for CIT, were adopted, making the Kinect-based CIT program motivational for children with unilateral CP. One primary consideration in the game design was the pace of the game. In general, the speed requirements of commercial games are too fast for children with CP, and they may feel frustration and anxiety during gameplay. Moreover, a game with excessively fast movements may induce abnormal muscle tone in the affected UL during gameplay (Scholtes et al., 2006). To overcome the disadvantages of commercial games, our Kinect-based CIT game was designed based on the motor abilities of children with unilateral CP, and all levels of the game were graded by the difficulty of the motor requirements. All of the children responded that the most difficult part of the game was defeating the third monster, which required the children to perform aiming and holding tasks. This result supported our game setting of level of difficulty; the third level presented the greatest challenge. In addition, according to the results of the questionnaire, questions such as “Was this game difficult to you?” and “Did you feel you were good at this game?” suggested that the difficulty of the game may be moderate for children with CP. Our Kinect-based CIT program can provide just the right amount of challenge for children with unilateral CP.

To evaluate the feasibility of our Kinect-based CIT program, a feasibility study for evaluating the potential effect in our program (i.e. Phase 2 study) was also performed. Based on the results of the performance scores recorded by the Kinect system and the outcomes of the BBT, both performances increased stably after 5 weeks of intervention, revealing the potential contribution to the affected UL motor function for the children with unilateral CP. However, the results showed decreases in the performance score and the outcomes of the BBT from the third week to the fourth week. For the performance score in our research, the positive impact of motor function was indicated in the third week (p = 0.002). Therefore, the therapist adjusted the game difficulty in the fourth week. The participants faced a new challenge and had not yet adapted, which might have caused the decline in the raw scores of participants in the game. In addition, a similar drop in BBT outcome could have been caused by motor reorganization in response to an unadaptable and challenging task. At the time of reorganization, the participants may have tried to integrate their skills to improve the quality of movement, which would have led to an increase in variability and more errors in motor performance (Thibaut and Toussaint, 2010; Jongbloed-Pereboom et al., 2013b). This decrease in motor performance during reorganization processes has also been found in the development of motor planning skills and handwriting (Meulenbroek and van Galen, 1988; Jongbloed-Pereboom et al., 2013b). Limited to our preliminary findings, however, more robust studies were needed to establish the effectiveness of the Kinect-based CIT training for children with unilateral CP.

Regarding the clinical implications for intervention in the future, our findings reveal the potential benefits of using this Kinect-based CIT program for telerehabilitation for children with unilateral CP. The use of telerehabilitation, or using telecommunication devices to provide treatments and evaluations for clients, has been increasing recently (Peretti et al., 2017). One advantage of telerehabilitation is that medical services can be provided anywhere and anytime, and even directly in the client’s own home. It is individualized, effective and flexible, and it might reduce the risk of spreading diseases (e.g., COVID-19) and other infections (Anton et al., 2018; Molinaro et al., 2020; da Silva et al., 2021). Several studies have reported that game-based telerehabilitation tends to contribute to the improvement of UL motor function in children with CP (Golomb et al., 2010; da Silva et al., 2021). The Kinect system is an inexpensive and widely-used sensor for treatment, and the therapist can refer to the Kinect records to give advice on the movements and levels of game difficulty to clients and their parents periodically. Thus, future studies could evaluate the effectiveness of this Kinect-based CIT program delivered through telerehabilitation on motor function of the affected UL in children with unilateral CP.

This study had some limitations. First, the sample size was small, and relatively high-functioning participants were recruited for our feasibility study. The results of this study thus might not be generalizable to other types of CP. Second, limited to the present preliminary study to evaluate the potential effects of our VR-based CIT program, further larger and controlled studies are needed to confirm the more robust effectiveness of our Kinect-based CIT program, which may help to improve the clinical utility of the Kinect-based CIT program.