Table 1 The aim, methodology, and key findings of the included studies.

Gupta and Kohli⁶³

To evaluate the effectiveness of VRehab on hospital readmission rate.

Using AI and computer vision, the TheraNow smartphone application provided exercise-based rehabilitation plans for patients and assessed the quality of their exercises.

Using AI-driven VRehab resulted in lower hospital readmission rates.

Kohli and Gupta⁶²

To evaluate patients' level of satisfaction and likelihood of recommending the VRehab platform to others.

Using AI and computer vision, the TheraNow smartphone application provided exercise-based rehabilitation plans for patients and assessed the quality of their exercises.

Patients reported high levels of satisfaction with the VRehab platform.

Fabio et al.⁶¹

To compare the performance of patients in non-AI-driven VRehab and AI-driven VRehab.

While non-AI-driven VRehab was simple video communication between patients and clinicians, AI-driven VRehab was equipped with eye gaze and body skeleton acquisition. The eye gaze and skeleton data were observed by the clinician to understand patients' interaction, attention, and movements.

AI-driven VRehab resulted in improvements in a few neuropsychological measurements.

Bo et al.²⁵

To establish a progressive framework for predicting rehabilitation outcomes.

Patients' motion data was collected using the built-in sensors in their smartphones. AI algorithms were used to analyze patients' motion data and demographic information to predict rehabilitation outcomes.

Combining clinical and demographic data with movement data significantly improved the performance of predictive AI algorithms.

Bouteraa et al.²⁶

To develop predictive models to estimate pain using features extracted from various sensors and use the estimated pain in the control loop for generating safe robot actions.

By using a computer vision system, the physiotherapist’s gestures were translated into commands for the robot. As a measure of safety, if the pain level exceeded a certain threshold, the robot would stop the action, even if the desired angle had not yet been reached.

The developed human-robot interface was able to provide a control and monitoring interface for home-based VRehab.

Tsvyakh et al.⁶⁰

To implement an AI-driven VRehab platform and compare it with traditional rehabilitation.

Different sensors were used to collect data from patients, including exercise time, local temperature, and the biomechanics of active movements of the injured limb. The collected data was accessible to clinician surgeons to monitor patients.

Compared to traditional rehabilitation, VRehab reduced the time that surgeons spent consulting with their patients and resulted in higher levels of patient satisfaction.

Fang et al.⁵⁹

To longitudinally examine the efficacy of VRehab.

Wearable sensors collected accelerometer data from patients while they performed rehabilitation exercises, which was transferred to and analyzed in the cloud. Using the results of the analysis, clinicians were able to monitor the progress of their patients remotely.

Compared to in-person (and phone-based) rehabilitation, VRehab resulted in a steady increase in Mobility Index and at least one stage improvement in Brunnstrom Stage.

Ghorbel et al.⁵⁸

To examine the impact of color-based 3D skeletal feedback to guide patients in completing rehabilitation exercises.

In a desktop application, color-based 3D skeletal feedback was superimposed on the videos of patients to guide them in completing exercises. Additionally, the movements of the patients were automatically analyzed and reported to clinicians.

The visual feedback improved the posture of the patients and enhanced the motion in the case of simple exercises. The VRehab platform was reliable, simple to use, and positively impacted patients' psychology measures. Clinicians and patients both found the measurement and feedback to be accurate, reliable, and safe.

Qiu et al.⁵⁷

To evaluate the feasibility of VRehab platform to prepare for a future efficacy study.

Patients controlled the rehabilitation game with the Leap Motion controller on their hand, and the difficulty of the game was determined adaptively according to the movements of patients. The movement data was transferred to the cloud, where clinicians could view it.

Patients were able to use the VRehab platform resulting in improvements in Upper Extremity Fugl-Meyer and hand kinematics.

Sobrino et al.⁵⁶

To evaluate the perceived usefulness and ease of use of VRehab platform.

The movements of patients were analyzed, and accordingly, on-screen textual and visual feedback was provided to patients regarding the quality of their exercises. Movement analysis results were also reported to clinicians.

The collected questionnaire data regarding the perceived usefulness and ease of use of the platform indicated a positive view of patients.

Triantafyllidis et al.⁶⁶

To evaluate the feasibility of VRehab platform.

In response to the real-time sensor data collected, a virtual coach was animated to provide patients with safe and personalized exercise feedback within their beneficial heart rate zones.

With the assistance of the virtual coach, patients were able to exercise within or above their beneficial heart rate zones for the majority of the exercise duration.

Yu et al.⁶⁵

To develop a remote quantitative Fugl-Meyer assessment framework,

The collected data from a wearable sensor network was used to automatically measure the Fugl-Meyer score.

The proposed quantitative models could precisely predict the Fugl-Meyer assessment based on wearable sensor data.

Zhang et al.⁶⁴

To develop and evaluate a wearable exoskeleton rehabilitation robot for clinic and home-based rehabilitation.

A wearable exoskeleton rehabilitation robot, along with a 3D animation, was used to perform task-based repetitive therapy.

Significant improvements in both the Wolf Motor Function Test and Fugl–Meyer Assessment scores were reported for some patients in both clinical and home settings.