Human disabilities could develop through cerebral palsy (CP), stroke, spinal cord injury (SCI), traumatic brain injury (TBI), and humerus fracture. Rehabilitating these disabled patients can be achieved using traditional and modern treatments. Virtual reality, robotics, simulation, and exergame are commonly used in modern treatments. Exergame is a video game that provides physical exercise using technology to detect movement and body reaction. Exergame games are classified as an augmented reality which allows users to feel real sensations when using the application.
Virtual Reality Therapy (VRT) systems involve the use of virtual reality as an assistant tool for the rehabilitation process. Exergame is a VRT system that is used by patients who suffer from movement disability in their idle limbs. Through several training procedures, exergame helps patients to improve the physical movements of their muscles.
Virtual reality is utilized in exergames to help patients with disabilities improve the movement of their limbs. Exergame settings, such as the game difficulty, play important roles in the rehabilitation outcome. Similarly, suboptimal exergames’ settings may adversely affect the accuracy of the results obtained. As such, the improvement in patients’ movement performances falls below the desired expectations.
MIRA is a non-immersive type of VRT application developed to make physiotherapy entertaining and enjoyable for patients. The platform transforms prevailing physical therapy exercises into clinically-designed video games. The MIRA platform is an effective system that allows patients to play their way toward recovery. The platform will be featured with K-Nearest Neighbors (KNN), Collaborative Filtering (CF), and Bacterial Foraging Optimization Algorithm (BFOA).
Physiotherapists use default settings which invariably lowers the accuracy in playing the exergames and reduces the patients’ performances. To overcome this problem, a recommender system (RS) is needed in the MIRA platform. ReComS approaches are proposed to recommend preferences for the MIRA platform. These preferences are used to determine the input settings of the exergames by learning the precise behaviors of patients. ReComS integrates the K-NN and CF methods to classify the predicted values by reducing the error value. The error value is obtained using the projected and actual values of the previous session of the exergame. ReComS (KNN and CF), ReComS+ (K-Means, KNN, and CF), and ReComS++ (BFOA, K-Means, KNN, and CF).
The ReComS approach is proposed to predict the variables of the input setting according to the data history of the patient. As ReComS is expected to provide low prediction accuracy, it is integrated with a clustering method (as used in similar experimental works, the ReComS+ approach. ReComS+ provides good prediction accuracy. ReComS++ is developed by further integrating ReComS+ with the BFOA algorithm to learn the latent features of the patients and to lower the RMSE value throughout the learning iteration process. The experimental results of ReComS and ReComS+ are utilized to benchmark the prediction performance of the ReComS++ approach.
This study was carried out in the rehabilitation center of Melaka, Malaysia to analyze the generated data using MIRA platform with ethical approval no PRPTAR.600-5 by Pusat Rehabilitasi Perkeso Sdn. Bhd., Lot PT 7263, Bandar Hijau, Hang Tuah Jaya, 75450 Bemban, Melaka, Malaysia. The MIRA platform patient data file in this study contains patients’ personal information such as first and last names, patient ID, and birth date.
Each selected game and movement acts as one exergame with its unique input variables in the item settings dialogue. The settings include the sides used (left or right), duration, difficulty, tolerance, minimum, and maximum ranges. The values of these variables could be fixed based on the default values or adjusted by the physiotherapist after evaluating the performance of the patient. The MIRA platform could generate 26 variables based on the exergame or cognigame (a game that trains cognitive function), including time (duration), still time, moving time, moving time in exercise, average acceleration, average deceleration, average accuracy, average congruent correct answer reaction time, average congruent incorrect answer reaction time, average percentage, average speed, average variation, distance, maximum percentage, minimum percentage, repetition, and points. The experimental data contained 3,553 records generated by 61 patients with different types of diagnoses: 41 patients had a stroke, 14 patients had TBI, seven patients had SCI, one patient had CP, and two patients had humerus.
Physiotherapists who deal with this application need to predict the values of the input variables of the item settings for each patient manually, which is the main challenge in this domain. Therefore, in this study, we utilize a recommender system to suggest the most suitable settings for patients’ movements based on their movement history. Physiotherapist tracks the patient who practices exergame and records his/her performance as ‘positive’ (P) or ‘negative’ (N). The result showed that the positive preference was higher than the negative preference.
Since the exergames generate various features, automated analysis is required to provide a summary of the patient’s (movement) performance. To address these challenges, three experimental approaches: (1) ReComS with the CF and K-NN approach;(2) ReComS+ with the CF, K-NN; and (3) k-means approach, in addition to ReComS++with the CF, K-NN, k-means and the BFOA approach; were proposed and their shortcomings were tested by learning procedures. The experimental results demonstrated that ReComS+ yields more accurate predictions when compared with ReComS while ReComS++ achieves a higher accuracy as compared to ReComS+. Overall, ReComS++ performs best for MIRA exergames as it provides MIRA with the most accurate predictions for the input setting dialogue box. It thus assists patients to perform MIRA exergames correctly
Author: Rimuljo Hendradi
