Technical aspects of virtual augmented reality-based rehabilitation systems for musculoskeletal disorders of the lower limbs: a systematic review | BMC Musculoskeletal Disorders | Full Text
The initial search yielded 2343 studies from databases. After evaluating the title and abstract of the studies, based on our inclusion criteria, 294 articles were selected for evaluation of the full text. Finally, 19 journal articles met our inclusion criteria. A second search was conducted on 17 December 2021 to find new studies. Two recent studies were found for full-text analysis. Then, one new study was selected. The manual search did not add any new articles to the study. Figure 1 presents the PRISMA flowchart of the study selection.
Literature search flow diagram
Of the selected studies, 10 (50%) were conference studies. According to Table 2, in 15 (75%) studies, in addition to implementation, the VR/AR evaluated the system, and the sample size was reported. The sample size ranged from 4 to 287. The age of the participants was reported in 9 (45%) studies, and the participants were over 18 years old. As shown in Fig. 2, 6 (30%) studies focused on total knee replacement (TKR) rehabilitation, 5 (25%) rehabilitation for the lower limbs, 4 (20%) total hip replacement (THR) rehabilitation, 1 (5%) after lower limb fracture surgery rehabilitation, 2 (10%) ankle injury rehabilitation, 1 (5%) anterior cruciate ligament (ACL) rehabilitation, and 1 (5%) knee osteoarthritis rehabilitation.
Frequency of the studies due to lower limb disorders
Kinect, made by Microsoft for games, is a low-cost motion camera that can provide information about the 20 major human joints in three-dimensional (3D) coordinates. This information can develop various rehabilitation systems with Kinect [46, 53].
Among the studies, 11 (55%) used Kinect technology as input tools [33, 34, 37, 38, 41, 43,44,45,46, 49, 52].
Inertial measurement unit (IMU)
One approach to evaluating rehabilitation exercises is to use inertial sensors, which include IMU and magnetic sensors, accelerometers, and gyroscopes, which measure an object’s linear acceleration and angular velocity [54].
According to our findings, 2 (10%) studies used the IMU technology as an input tool [35, 36, 50].
Guggenberger et al. [31] used the built-in inertia measurement unit and the integrated front camera of the smartphone and head-mounted displays (HMD) to track the movements and produce the corresponding AR images. Furthermore, in 2021, Zhao et al. [32] used a combination of Kinect and IMU technologies for real-time rehabilitation and motion-tracking exercises. They boosted the system with the IMU sensor because the hip and knee angles can be significantly tracked with the Kinect, but tracking ankle movements is difficult with the Kinect.
Surface Electromyography (sEMG)
We found 2 (10%) studies used sEMG technology as input tools [12, 39]. In Günaydin’s study, a serious concept of computer games for physiotherapy and lower limb rehabilitation using sEMG signals and a feedback module for remote tracking of patients is presented. When the patient plays a game, the sEMG signals are recorded and then analyzed. Measuring sEMG during rehabilitation provides information about the progress of related muscles [12].
Other input tools
Three studies did not use the above input tools [40, 47, 48]. Pruna et al. [40] implemented a 3D virtual lower limb rehabilitation system using three space mocap sensors. Gonzalez-Franco et al. [47] used an accelerometer to empower patients in physiotherapy at home. Garcia and Felix Navarro [48] aimed at rehabilitating people with ankle sprains; they implemented an augmented reality application for mobile devices using an AR marker.
Movement recognition and assessment
Providing a rehabilitation program through an interface that detects human movement can help to perform the correct movements [37].
Some studies have described the techniques used to analyze human movements. Here, recognition and assessment techniques were classified according to the sensors used.
Movement recognition with Kinect
Kinect can provide real-time, in-depth skeleton tracking information of 20 joints and red, green, and blue (RGB) images for movement recognition [46]. Among the studies, 2 (10%) used the dynamic time warping (DTW) algorithm to distinguish between right and wrong movements [34, 45]. This algorithm processes the skeletal data [34]. Another successful method for achieving movement recognition is the discriminative approach. The main classifiers that use this method to identify movements are k-nearest neighbor (KNN), support vector machines (SVM), naïve Bayes, and the C4.5 decision tree. The last two algorithms are the most popular because they allow high classification accuracy [55]. In addition to DTW, Perez Medina et al. [34] used an SVM algorithm to recognize and process faces.
In Tannous et al.’s study on the avatar scaling process, a linear rigid trans-formation was applied, and body height, computed from the Kinect, was used as a scaling factor [41]. Since the Kinect skeleton model did not provide a reasonable estimate of ground positions, Zhiyu et al. [44] used depth images Kinect and multiple leg angle estimators for different angle regions to recognize. Choi et al. [46] also used common Kinect data to detect leg-strengthening exercises. Rybarczyk et al. [38] used an evaluation module based on the Hidden-Markov model approach to assess the quality of real-time movements. Su et al. [43] used a combination of a principal component analysis (PCA) and an adaptive network-based fuzzy inference system (ANFIS). They provided a predictive emotion model-based artificial emotion model with a Plutchik emotional wheel.
One approach to recognizing human movements is the rule-based method. In this method, movements are first described based on a set of rules and then classified according to the rules set for each movement [56]. In two studies, the rule-based method was used to recognize the movements. These studies define all rules based on the extensible markup language (XML) [32, 37].
Movement recognition whit IMU
Kontadakis et al. [36] used an automated exercise classification algorithm using data from the IMU sensor to recognize the movements. The input data of the algorithm were filtered using a Complementary filter. The algorithm’s output was a computational decision for the correctness or otherwise of the exercises. In 2018, they also used a similar algorithm to recognize the movements [35].
EMG signal analysis
In 2 (10%) studies, when the patient was playing, EMG signals were stored and then analyzed for quantitative evaluation of rehabilitation, and feature extraction methods were used to analyze the EMG signal [12, 39]. Feng et al. [39] propose a rehabilitation assessment method based on the multi-characteristic fusion of kinematic signals and EMG. This method consists of three indicators: mean square root EMG, joint activity, and joint smoothness.
VR/AR systems development tools
In 10 (50%) studies, including the AR study, the Unity game engine was used for visualization [12, 32, 35, 37, 40, 43, 44, 47, 48, 50]. Unity is a powerful and stable tool for designing and developing games, which has received much attention in the game industry [43]. We found that the non-commercial Kinect windows software development kit (SDK) and C# as the programming language for Kinect capabilities were the most common tools for developing rehabilitation systems [32, 33, 35, 41, 46, 48].
Evaluation of VR/AR-base systems
Examining the studies, we found that six studies (30%) did not express the system evaluation method, and AR-based research was one of these studies [32, 44, 48,49,50, 52].
System effectiveness evaluation
Most studies did not perform clinical trials on developed systems and only examined the systems with an initial evaluation; this initial evaluation of the rehabilitation systems developed showed the promising impact of these systems on lower limb rehabilitation. In 4 (20%) studies, it was stated that VR-based rehabilitation had an important role in motivating patients, potentially leading to greater participation and better outcomes in rehabilitation [12, 35, 48, 50].
According to Table 2, some studies used rehabilitation assessment methods, including measuring the amount of ROM [35, 37], one-leg standing test (OLST)[46], knee injury and osteoarthritis outcome score (KOOS) [33], patient-reported outcomes measurement information system (PROMIS) [33], American knee society score (AKSS) [43], and a primary health status questionnaire [34].
System usability evaluation
Acceptance of the system by the user is critical. The two vital factors in adopting a system are examining the usability of the system and considering the principles of user-centered design [43].
The principles of user-centered design in developing VR/AR systems have been considered in 3(15%) studies [12, 34, 45].
The usability of rehabilitation systems was assessed in 8 studies (40%) using a usability questionnaire. The results of these studies showed an acceptable level of usability [12, 37, 38, 40, 41, 43, 47, 53]. Two studies used the system usability scale (SUS) questionnaire[38, 43], and one applied the single ease question (SEQ) test to record the user comments [40].
Quality evaluation of studies
The quality evaluation results of studies based on the StaRI statement are presented shown in Table 3, and the results are summarized as follows:
In the title and summary section, 85% of the studies had followed at least one of the two checklist items. In the introduction section, 75% of the study had completed at least two of the three items mentioned in the checklist. In the method section (description), 50% of the studies had at least 4 out of 7 checklist items. In the method (evaluation) part, 80% of the survey had observed at least half of the items mentioned in this part of the checklist. But in the economic sector, no study had fully observed this case. In the results section, only 30% of the studies presented at least 5 of the ten items mentioned in the checklist. And in the discussion section, 15% of the studies presented at least two of the three items mentioned in the checklist.
Based on the results of the qualitative analysis of studies according to the StaRI statement, two studies evaluated the VR/AR rehabilitation system through RCT. Yeh et al.’s study included an experimental group with a computer game and a control group with traditional rehabilitation. For the experimental group, during the experiment, individuals can check the rehabilitation status in real-time through the system and learn the next movement mode. For the control group, rehabilitation dominated the rehabilitation process, including bending the knee and raising the thigh [51]. Prvu Bettger et al. [33] conducted an RCT to evaluate the effect of a virtual PT program on total costs at 12 weeks post-TKA and to evaluate the clinical efficacy and safety of virtual PT over conventional care with traditional PT. Patients in the intervention group used the VERA system, and patients in the routine care group followed the recommendations of their care team for all medical and pre and post-operative rehabilitation care.
According to Table 3, the StaRI checklist items for the studies are interpreted as follows:
Title and abstract
In the method section, most studies were well-designed in design and content. Still, as shown in Table 3, the studies were weak regarding targeted ‘sites’ and the description of the intervention and the implementation strategy. In addition, the evaluation section of the studies was presented acceptably, but none of the studies performed an economic evaluation.
This content was originally published here.
