Skip to main content

Table 2 Overview of studies categorized under the section applications of Kinect in stroke rehabilitation

From: Systematic review of Kinect applications in elderly care and stroke rehabilitation

Author Year Population Significant findings
Stroke rehabilitation > Evaluation of Kinect’s spatial accuracy
Pedro et al. 2012 Study type: methodologyParticipants: 1 (robotic arm) Age: N/A A KUKA robotic arm (precision accuracy of up to 0.05 mm) was utilized for precise movements. The Kinect was attached to this arm and a target was positioned at a static position in the KUKA arm’s work space resulting in Kinect readings with a min error of 0.036 mm, max error of 12.25 mm, mean error of 4.95 mm and standard deviation of 2.09 mm in comparison to the KUKA as a ground truth.
Chang et al. 2012 Study type: methodologyParticipants: 2 (m = 1, f = 1) Age: (unspecified) Appraised the tracking performance of the kinect specifically for a set of six upper limb motor tasks in regards to a high fidelity OptiTrack optical tracking system consisting of an array of 16 ceiling-mounted cameras. The following motions were utilized: external rotation, shoulder abduction, shoulder adduction (diagonal pull down) scapular retraction, shoulder flexion, and shoulder extension. While a statistical analysis of data captured was not offered, a visual representation demonstrated that data trends for both systems, in regards to hand and elbow represent competitive movement tracking performance, whereas shoulder readings were widely inconsistent. The authors attribute these inconsistent shoulder readings as due to differing methods of motion capture and joint estimation between the OptiTrack and the Kinect. Furthermore, the participants were asked to utilize External Rotation of the shoulder 10 times each, with 5 correct movements and 5 incorrect movements. The Kinect-based game implemented successfully identified all the incorrect movements.
Clark et al. 2012 Study type: methodologyParticipants: 20 (healthy, m = 10,f = 10) Age: 27.1 yr (± 4.5) Height: 173.7cm (± 10.3) Mass: 71.7kg (± 11.0) Type 2,1 intra-class correlation coefficient difference between Kinect and Vicon Nexus (ICC) and ratio of coefficient of variation difference between systems (CV) was conducted using three postural control tests: a forward reach, a lateral reach, and a one leg standing balance test. The points of examination were of distance reached, trunk flexion angle (sagittal and coronal), and a balance test focused on spatio-temporal changes in the sternum, pelvis, knee and ankle as well as the angle of lateral and anterior trunk flexion. The results demonstrated a very high level of agreement between systems. The following is a sample of reported data (units in mm): Lateral reach: - Sternum: ICC = 0.03, CV = 0.1; Hand: ICC = 0.16, CV = 5.5; Trunk (deg): ICC = 0.01, CV = 0.7; Forward reach: - Sternum: ICC = 0.07, CV = 1.0; Hand: ICC = 0.05, CV = 1.2; Trunk (deg) ICC = 0.00, CV = 0.6; Single leg balance: - For a full-body joint-by-joint char of details see Table one and Table two on page 375 of the study
Obdrzalek et al. 2012 Study type: methodologyParticipants: 5 (unspecified gender) Age: unspecified Full-body comparison between the Kinect and PhaseSpace Recap for joint position readings of mean difference, standard deviation from mean, and right and left specific measurements. Overall error was typically within sub-centimeter accuracy; however, centimeter level accuracy was also noted on more difficult joint comparisons, such as the hip For detailed results of the comparison based on a front view see Table one on page 5 of the study For detailed results of the comparison based on a 30° view see Table two on page 5 of the study For detailed results of the comparison based on a 60° view see Table three on page 5 of the study
Loconsole et al. 2012 Study type: methodologyParticipants: 1 (healthy, male) Age: 25 This study utilized an L-Exos controller exoskeleton robot arm and a Kinect in order to track a patients upper extremities and objects and examined: 1) light variation: very intensive, medium and low illumination - no substantial differences; 2) occlusions: two objects moved to occlude each other - no adverse effect and both items were correctly recognized again post occlusion; 3) object roto-traslation: rotation and movement of two tracked objects - no substantial error introduced, and 4) accuracy: error was negligible (within 2 cm). Accuracy test starting distances: 500 mm, 700 mm, and 900 mm on the Z axes. The object was moved 10 mm, and then 20 mm, and finally 50 mm along the X and Z axes. The following shows the error introduced by the specified movements on the Z and X axes (all units in mm): 500 distance: +10 mm: Z = 0.1, X = 0.1; +20 mm: Z = 0.3, X = 0.1; +30 mm: Z = 0.5, X = 0.5 700 distance: +10 mm: Z = 0.5, X = 0.2; +20 mm: Z = 0.8, X = 0.2; +30 mm: Z = 1.2, X = 0.5 900 distance: +10 mm: Z = 0.6, X = 0.4; +20 mm: Z = 1.9, X = 0.4; +30 mm: Z = 2.1, X = 0.5
Fern et al. 2012 Study type: methodologyParticipants: 1 (healthy, male) Age: unspecified Accuracy comparison was done between Kinect (OpenNI and Primesense’s NITE) and a 24 camera Vicon (MX3) system. Movements included: 1) knee flexion and extension; 2) hip flexion and extension on the sagittal plane; 3) hip adduction and abduction on the coronal plane with knee extended; 4) shoulder flexion and extension on the sagittal plane with elbow extended; 5) shoulder adduction and abduction on the coronal plane with elbow extended, and 6) shoulder horizontal adduction and abduction on the transverse plane with elbow extended. Mean Error (ME) and mean error relative to Range of Motion (ROM) was calculated. All error readings for the knee and hip are lower than 10° ranging from 6.78° to 9.92°. Dynamic ranges of motion are between 89° and 115°. ME increases when ROM is higher mainly due to occlusion. Error readings for the shoulder range from 7° to 13°.
Stroke rehabilitation > Rehabilitation methods
Saposnik et al. 2011 Study type: review A meta-analysis to determine the benefit of VR technology for post stroke upper extremity recovery was conducted and reported improvement of Fugl-Meyer scores and measures of arm speed, range of motion, and force at the ‘Body Structure and Function’ level (of International Classification of Functioning (ICF) [3]). Improvements for the VR-trained experimental groups ranged from 13.7% to 20% vs 3.8% to 12.2% in the non-VR control groups. The ‘Activity’ level of the ICF tests (such as the Wolf Motor Function Test (WMFT), Jebson-Taylor Hand Function Test, and the Box and Block Test) also showed increased results within VR-trained experimental groups from 14% to 35.5% vs 0% to 49% for non-VR control groups. Randomized controlled trials (RCTs) were evaluated using the pooled treatment effect (Mantel-Haenszel (OR)) by using random-effect models to reduce the effects of heterogeneity between studies. The effect of VR-based rehabilitation on motor impairment level once the 5 RCTs were combined was OR = 4.89 (95% CI, 1.31 to 18.3; P <0.02). No significant improvement was noted on the Box and Block Test (2 RCTs; OR, 0.49; 95% CI, 0.09 to 2.65; P = 0.41) or WMFT (3 RCTs; OR, 1.29; 95% CI, 0.28 to 5.90; P = 0.74). When considering observational studies, VR-based intervention affected motor impairment percent improvement by 14.7% (95% CI, 8.7% to 23.6%; P <0.001). VR-based intervention on Jebson-Taylor Hand Function Test, WMFT, and Motor Activity Scale resulted in 20.1% improvement in motor function after VR-based intervention. (95% CI, 11.0% to 33.8%; P <0.001).
Hussain et al. 2012 Study type: methodology A prototype system SITAR (System for Independent Task-oriented Assessment and Rehabilitation) aimed at delivering controlled, task-oriented stroke therapy in an independent manner with minimal therapist supervision was presented. The SITAR is a tabletop system that has function as an assessment or rehabilitation system for upper extremities. SITAR has three parts 1) a set of intelligent objects for haptic-based patient interaction, 2) a marker-less tracking system using inertial measurement units and the Kinect to track the position of the intelligent objects and the movement kinematics of a subject extremities and trunk, and 3) Kinect-based games to engage and motivate patient participation.
Bo et al. 2011 Study type: methodologyParticipants: unspecified (healthy) Age: unspecified Study proposed a system which utilized a fusion of Kinect and inertial measurement units (IMU) of gyrometers and accelerometers. Using only IMU sensors, individual errors occur in both gyrometers (accumulated error due to bias) and accelerometers (noise and inertial acceleration peaks). Data was significantly more aligned when a fusion of Kinect and the IMU sensors was used via online calibration; however, the study did not provide quantitative results analysis. A video of the experiment can be found at
Shiratuddin et al. 2012 Study type: methodology A framework for utilizing non-contact natural user interfaces for an interactive visuotactile 3D virtual environment system was presented in this study. Utilizing the 3D environment of the Kinect may be an approach which could more accurately stimulate the visual cortex and enable more authentic rehabilitation feedback than the current 2D feedback paradigm, ultimately leading to better outcomes.
Yeh et al. 2012 Study type: methodology The main objective of the proposed system is to stimulate patient participation in upper limb rehabilitation activities. This is accomplished through various manipulations of a virtual ball that a patient interacts with through control of a Kinect-generated skeleton. In order to target the rehabilitation exercises for clinical purposes, a therapist can control parameters related to the ball (e.g. speed and size).
Da Gama et al. 2012 Study type: methodologyParticipants: 10 (3 physiotherapyprofessionals, 4 healthy adults, and 3 elderly subjects of unspecified sex.) Age: unspecified The system introduced in this study focused on the guidance and correction of participant movements during motor rehabilitation therapies. The study focused on shoulder abduction using the following requirements: 1) shoulder abduction (angle >= 90°); 2) elbow angle >= 160°; 3) angle between the arm and frontal vector plane of >= 80° and <= 100°; 4) right and left shoulder height (Y coordinate) must be similar (for trunk compensation detection); 5) actual shoulder abduction angle must be higher than it was before; 6) return to starting position. Study examined 50 ‘correct’ movements (e.g. fulfilling all the former requirements) with participant standing, seated, and positioned at different angles in respect to the Kinect sensor. All 50 of these ‘correct’ movements were recognized as correct to the system. 60 unspecified ‘incorrect’ exercises (e.g. not fulfilling all the former requirements) were also performed and recognized as incorrect by the system - including postural compensation. The participants also completed a Likert-scale questionnaire to assess the negative aspects of the system (5 = as strongly agree): size of letters (2.77), information clarity (3.75), and stimulus (3.47). The positive reported aspects were: user satisfaction (4.67), motivation (4.67), the system easiness (4.64).
Pastor et al. 2012 Study type: researchParticipants: 1 (stroke, female) Age: 46 Gameplay involves sliding the impaired limb on top of a transparent support in an attempt to hit various targets. The patients range of motion did not show any statistically significant change before and after system use: Fugl-Meyer score before = 16; after = 16. The patient’s score in game steadily increased during the study; however, the authors note that while the game’s score is proportional to the arm’s movement speed, it does not necessarily correspond to motor recovery.
Frisoli et al. 2012 Study type: researchParticipants: 7 (m = 6, f = 1, three healthy volunteers, 4 chronic stroke patients) Age: healthy = 27 (± 7), stroke = 64.5(±13) Thisstudy presented a Kinect-based, multimodal architecture for a brain-controlled interface-driven robotic upper-limb exoskeleton with a goal of providing active assistance during reaching tasks for stroke rehabilitation. The individual and aggregated performance of the SVM classifier in both trainings of visual condition only, and robot-assisted sessions were examined. The reported performance was based on the offline evaluation of the SVM classifier on the training set. Averaged Correct Classification Rate (%), Healthy subject (H), Stroke patient (P), All (A): Visual: H = 88.1(±5.9); P = 91.9(±9.3); A = 88.2(±10.4) Robot: H = 81.2(±13.6); P = 90.4(±4.9); A = 89.4(±5.0) All: H = 86.4(±8.3); P = 91.1(±6.9); A = 88.8(±7.9)
Stroke rehabilitation > Kinect gaming
Borghese et al. 2012 Study type: methodologyParticipants: unspecified Age: unspecified Authors state that the system enables quantitative and qualitative exercise evaluation and automatic game-play level adaptation. Presents two serious minigames: Animal Feeder and Fruit catcher. Animal Feeder offers training for dual tasks management (i.e. using both arms simultaneously for different purposes), and In Fruit Catcher the patient is required to utilize reaching and weight shift without movement of the feet. Also, inappropriate movements issue a warning to the player or, in extreme cases, abort the task when detected as unsafe.
Huang et al. 2011 Study type: methodology A prototype of a serious game based off Jewel Mine using a Smart Glove that would enable participants to actually reach out and grasp target gems, which are located in a semi-circle above a virtual avatar, and place the gems into a receptacle instead of just touching the gems for collection. This combination would enable concurrent hand and upper limb rehabilitation in one serious game.
Lange et al. 2011 Study type: methodologyParticipants: 23 (m = 19, f = 4)Participants consisted of those with balance issues related to Stroke(n = 10), TBI (n = 4) and SCI (n = 9) and 10 clinicians (m = 4, f = 6) The study presented and discussed three potential applications of the Kinect. 1) virtual environments, 2) gesture controlled PC games, and 3) a game developed to target specific movements for rehabilitation. A prototype balance-based reaching game was developed based on Jewel Mine; however, only anecdotal qualitative data was presented in that patients had reported that the games were challenging and fun, and they would be likely to use the technology within the clinic and home settings if given the option. Clinicians also expressed excitement about the use of this type of technology within the clinical setting.
Pirovano et al. 2012 Study type: methodology A low-cost game-oriented platform for patients who would benefit greatly from intensive rehabilitation at home. The system proposed would allow for the patient to continue beneficial physician-controlled rehabilitation exercises through remote monitoring and difficulty adjustments as well as a Bayesian-based adaptation schema for automatic game-based difficulty level adjustments.
Saini et al. 2012 Study type: methodology The study presented a low-cost game framework for stroke rehabilitation. This program’s goal is to increase patients’ motivation for therapy, and also to study the effects of Kinect-based gaming on hand and leg rehabilitation. Also, game design principles for hand and leg rehabilitation for improving the efficacy of stroke exercise was presented. The proposed framework provides angle based limb representation during exercise to ensure exercises are conducted in a correct biomechanical direction angle lessening the chance of reinjury.
Sadihov et al. 2013 Study type: methodologyParticipants: unspecified amount of therapists and stroke patients with slight impairment. Age: unspecified Based on the Kinect-based haptic glove algorithms discussed, three rehabilitation game applications were developed: 1) a table wiping game; 2) a meteor deflection game, and 3) a rope pulling game. The table wiping game consists of an avatar-hand used to wipe stains from a table with different vibration patterns being initiated in the worn haptic glove based on the participant’s movements. In the rope pulling game, the participant’s virtual hand is able to grab and pull a colorful rope which can be modified for various feels through different force thresholds and feedback types. The meteor Game allows the player to deflect falling meteors from smashing into a virtual village.
Lange et al. 2011 Study type: methodologyParticipants: 20 (m = 17, f = 4) (stroke, TBI, SCI) Age: unspecified This study presents a system prototype to assess an interactive game-based rehabilitation tool for balance training of adults with neurological injury and was based off the previously developed Jewel Mine game. A series of interviews with clinicians, researchers and patients suffering from neurological conditions impacting balance was used. Preliminary testing took place in an informal setting and reported results were limited to qualitative data about user perceptions of the technology, motivation to use the technology, and the enjoyment level of the system with no quantitative data presented. The authors note that in general participants found the system usable and enjoyable.
Jiang et al. 2012 Study type: researchParticipants: 3 (upper extremityimpairment, m = 2, f = 1) Age: unspecified This study presents the following heuristics on selecting gesture patterns for patients with upper extremity impairments based off interviews with subjects with upper extremity impairments and subsequent Borg scale rankings regarding potential movements. The guidelines for gestures selection reported is as follows and were derived using a human-based approach which constructs the gesture lexicon based on studying how potential users interact with each other rather than what would be easy for the system to recognize: (1) Select gestures that do not strain the muscles; (2) Select gestures that do not require much outward elbow joint extension; (3) Select gestures that do not require much outward shoulder joint extension; (4) Select gestures that avoid outer positions; (5) Select dynamic gestures instead of static gestures; (6) Select vertical plane gestures where hands’ extension is avoided; (7) Relaxed neutral position is in the middle between outer positions, and (8) Select gestures that do not require wrist joint extension caused by hand rotation.
Llorens et al. 2012 Study type: researchParticipants: 15 (m = 8, f = 7) Age: 51.87(±15.57) A Kinect-based stepping exercise game for clinical effectiveness. In this study an exergame was created with an objective of stepping on randomly rising objects that emerged from the floor surrounding the patient. Each participant underwent twenty 45-minute training sessions, which consisted of six 6-minute repetitions with a one minute resting time between repetitions. Each participant completed at least (max 5) sessions per week. Assessment was with the Berg balance scale (BBS) [4]; The balance subscale of the Tinetti performance oriented mobility assessment (POBMAb) [5], and the Brunnel balance assessment (BBA) [6]. Assessment was completed at the beginning with an initial assessment (IA), the end with a final assessment (FA), and 1 month after completion with a final update assessment (FUA). The experimental results demonstrated that virtual training significantly improved time scales in balance recovery for stroke patients. For detailed BBS, POBMAb, and BBA results please see Table two on page 111 of the study.