STRENGTHS AND LIMITATIONS OF THIS STUDY

Our system employs a combination of computer vision and augmented reality technologies to enable patients to conduct remote sports rehabilitation training in real time using a cell phone equipped with a camera. This technology holds a high potential for widespread usage as it uses electronic devices commonly owned by the patient population, such as cell phones, and eliminates the need for additional sensors. This ensures that patients receive real-time training guidance and feedback during rehabilitation.

Our study also has some limitations, for example, our study is a single-centre study, which may have affected the applicability of the results to other regions and ethnicities. Our study will include only patients aged 40–85 years, a common population favoured for knee osteoarthritis (OA), but will also limit the application of the results to patients in other age classes. Since exercise therapy is only one of the adjunctive therapies for patients with advanced OA, our study will include only patients with Kellgren-Lawrence classifications I–III, which will limit the generalisation of the results.

Background

Over the last two decades, osteoarthritis (OA) has emerged as a significant cause of increased medical and economic burden globally, with research indicating that the overall prevalence of knee OA among middle-aged and older adults in China is 21.5%.1 This prevalence increases with age, and studies have shown that, in individuals over 80 years old, prevalence reaches 36.4%, contributing significantly to disability in the middle-aged and elderly population.2 In the USA, studies have shown that approximately 9.29% of subjects over the age of 60 have symptomatic knee OA, while the lifetime risk of knee OA is approximately 13.83%.3 The overall global prevalence of hip and knee OA is estimated to be about 300 million, making it the 11th largest contributor to global disability.4 5 According to the diagnostic criteria for OA, the presence of clinical symptoms is one of the conditions for a patient to be diagnosed with OA of the knee.6 OA of the knee is known to cause lower limb pain, stiffness, muscle weakness and joint instability, all of which can significantly impact patients’ daily activities, such as walking and climbing stairs, and even increase their risk of falls. The decreased physical activity due to knee OA contributes to a sedentary lifestyle, which in turn increases the risk of cardiovascular disease and significantly lowers the patient’s quality of life.7–9 As the population of China continues to age, the management of knee OA will become one of the most urgent healthcare concerns in the country.

Treatment options for knee OA are diverse, and guidelines strongly recommend non-pharmacological interventions such as exercise, along with weight loss, activity modification, bracing, and oral and topical anti-inflammatories. As a step in the treatment algorithm, exercise and physical therapy can offer significant benefits in pain management and joint stabilisation without the risk of intra-articular injections or surgical procedures, particularly in patients with early-stage arthritis.

Exercise therapy is considered a core treatment for OA according to the 2014 recommendations of the OA Research Society International.10 Studies have shown that exercise therapy for knee OA has a small to moderate effect compared with no exercise and that aerobic and muscle strengthening exercises may help reduce OA symptoms and improve joint function.11 12 Recent studies have demonstrated the efficacy of exercise therapy regardless of its content, but discontinuation of the exercise programme may lead to decreased efficacy after 6 months. Hence, ensuring patient adherence to an exercise programme is crucial for the success of treatment.13 A prior review has demonstrated that therapeutic exercise under professional supervision can have significant effects. However, this approach could be resource-intensive and create a socioeconomic burden.14 Therefore, it would be clinically valuable to develop a remote rehabilitation approach that could enhance patient adherence.

An emerging area of research in rehabilitation is the use of technology and wearable sensors to assist in OA management. However, the cost of these devices and their potential limitations for use in less developed economic areas remain as major concerns.15 Computer vision and augmented reality (AR) technologies have recently gained popularity in healthcare due to their ease of use and ability to provide an interactive experience for users. Currently, these technologies are primarily used in intraoperative navigation and physician education with limited application in rehabilitation.16 17 By combining computer vision and AR technology, patients can engage in real-time remote rehabilitation training using only a cell phone equipped with a camera. Thus, this technology has the potential to provide widespread access to patients of many socioeconomic levels. This development eliminates the need for additional sensors and ensures that patients receive real-time training, guidance and feedback, serving as a promising solution for remote rehabilitation training. Currently, there have been studies reporting the use of AR technology for rehabilitation of upper and lower extremities in stroke patients and amputation patients and showed significant improvements in balance, gait, muscle and physical performance.18–20 Several recent studies have used AR technology for telerehabilitation of patients with adhesive arthritis and post-total knee replacement, using monitors to display the patient’s movements and provide feedback, resulting in significant improvements in patient functioning.21 22 However, this technique is still in its infancy and requires further research and lacks reports of its use for OA of the knee. Therefore, we intend to conduct a pilot study to compare the clinical efficacy of an AR-based remote rehabilitation system that uses cell phone camera with traditional outpatient rehabilitation for patients diagnosed with knee OA.

Research question

For adult patients with knee OA, compare whether remote rehabilitation using a cell phone camera–based AR artificial intelligence rehabilitation system can improve clinical symptoms of knee OA and knee function after 12 weeks, and compare the clinical outcomes with those patients undergoing outpatient rehabilitation.

Research objectivesMain objective

The main objective of this study was to compare the change in clinical function of patients with knee OA after the implementation of a telerehabilitation using an artificial intelligence system based on AR technology to guide patients with knee OA and a rehabilitation outpatient intervention programme.

Secondary objectives

To assess the maintenance of clinical outcomes 24 weeks after completion of the intervention.

To assess whether the intervention programme resulted in a significant clinical benefit (minimal clinically important difference (MCID), patient accepted symptom status (PASS) and significant clinical benefit (SCB)).

To gain a better understanding of the user experience of both patients and providers.

Research design

This evaluator-blinded, prospective, randomised controlled study was approved by the Institutional Review Board for the Protection of Human Subjects. The study was registered with the Chinese Clinical Trials Registry (ChiCTR2300070319). Written informed consent will be obtained from all participants after a detailed explanation of the study. Figure 1 is the overview of participant-related study processes.

Figure 1

Overview of participant-related study processes. EQ-5D, European Five-Dimensional Health Scale; KOOS, Knee injury and Osteoarthritis Outcome Score; MCID, minimal clinically important difference; NSAID, non-steroidal anti-inflammatory drug; PASS, patient accepted symptom status; SCB, significant clinical benefit; SF-12, Short Form-12; VAS, visual analogue scale; WOMAC, University of Western Ontario and McMaster University Osteoarthritis Index.

Blinding. This study will be blinded by dividing investigators and subjects into pairs, so only researchers set the blind and analysts were blinded to randomisation and allocation.

Subject selection

Conservatively treated, nonoperative patients with OA of the knee whose ages meet the indications for conservative treatment of OA.

Inclusion criteria:

Between the ages of 40 and 85 years.

Meet the clinical and imaging criteria for OA of the knee established by the American College of Rheumatology, with Kellgren and Lawrence stages I to III.

Be able to receive a total of 36 weeks of medical care, including 12 weeks of treatment and 24 weeks of follow-up.

Exclusion criteria:

Knee infection, inflammation, autoimmune disease or fracture.

History of underlying disease affecting posture and balance, such as malignancy, dizziness, vertigo or stroke.

Being pregnant or planning to become pregnant.

History of knee trauma within 12 weeks.

Ligamentous knee injury.

Sample size

This pilot study will recruit 30 participants per group.23 We will then select the University of Western Ontario and McMaster University Osteoarthritis Index (WOMAC) score as the primary outcome and use the variance of the measurements from the pilot study to calculate the power of the change in the WOMAC score and the MCID. all results from the pilot study will be used in the sample size calculations for the subsequent large study.

Patient recruitment

Patients will be recruited through the Sports Medicine Clinic of West China Hospital of Sichuan University and will be invited to participate in the study if the attending physician, who is also the investigator, thinks that the patient is eligible for the study; recruitment information will be posted in the hospital clinic, and all patients who are interested in the study will be evaluated to see if they meet the inclusion and exclusion criteria for participation in the study.

Randomisation and patient allocation

Randomisation and patient assignment will be performed by third-party researchers who will be not involved in this clinical study. After patients will be determined to be eligible for enrolment, researchers will use a computer to automatically generate a random sequence using Microsoft Excel to generate random integers, with odd numbers being the experimental group and even numbers being the control group. The grouping information will be stored in a separate server. Patients will be informed of their exercise programme by the researcher on the day of treatment initiation.

Blinding

To ensure the integrity of the study, investigators and subjects will be paired and blinded. Only researchers will establish the blinding, while analysts will be blinding to randomisation and allocation. Blinding may occur when all subjects have completed the trial and follow-up; in the event of emergencies such as serious adverse events, suspected unintended serious adverse reactions, etc., blinding may occur after consulting the principal investigator.

Study intervention

Control group. Patients will receive face-to-face rehabilitation under the guidance of clinicians.

Intervention group. Patients will receive rehabilitation guidance using the cell phone camera–based AR artificial intelligence recognition system.

Augmented reality (AR) system

The remote computer vision rehabilitation system comprises four main components, including camera systems, human pose models(figure 2), image features for motion capture, and algorithms for determining the pose, motion and torque of the body. Figure 3 is the equipment operation workflow, and figure 4 is the demonstration of the artificial intelligence system.

Figure 2

Human pose model.

Figure 3

Equipment operation workflow.

Figure 4

Demonstration of the artificial intelligence system.

Camera system

A cell phone camera or a depth-sensing red, green, blue-depth (RGB-D) camera can be used. The RGB-D camera used in the remote machine vision rehabilitation system is a depth-sensing camera that operates on the principle of optical triangulation. This involves projecting a structured light pattern onto the surface of an object, which forms a three-dimensional image of the surface that is changed by the shape of the object being measured. By analysing the distortion of the light image coordinates, the three-dimensional shape of the object surface contours can be accurately calculated.

Human pose model and image features for motion capture

The pose model used for motion capture without markers is similar to the pose model used in traditional motion capture methods. A skeleton is composed of a set of joints connected by bones of varying lengths and orientations. The skeleton is characterised by the lengths and orientations of the bones and the angles of the joints.

Algorithm to determine the posture, movement and torque of the human model

Each person in the scene is modelled as an articulated skeleton consisting of 24 bones and 25 joints. Rigid three-dimensional posture can be completely specified by 6 df, three related to translation and three to define orientation. Joint angles are computed from the model using trigonometric functions. Combining the body part’s inertia parameters, the body’s centre of mass position can be inferred. Combined with kinematic analysis and inverse dynamics, joint torques and powers can be computed.

Outcomes

Patient-reported outcomes will be documented at the following intervals: pre-intervention baseline, intervention weeks 4, 8 and 12; and 4, 8, 12 and 24 weeks after the end of the intervention. All patient questionnaires were collected remotely via the internet. Specific patient–reported outcomes assessed will be knee function scores (Knee injury and Osteoarthritis Outcome Score (KOOS), WOMAC), pain scale (VAS) and overall quality of life (European Five-Dimensional Health Scale (EQ-5D), Short Form-12 scale (SF-12)) as described below, and WOMAC is most dominant outcome:

WOMAC: A scale with good validity and reliability that assesses pain, stiffness and motor function in knee OA. Higher scores indicate worse symptoms and more functional limitations.24

KOOS was used to assess patient outcomes in the short and long term. This scoring system has sufficient internal consistency, retest reliability, content validity and construct validity for patients with knee OA. The KOOS score includes five components: pain, symptoms, daily functioning, sports and recreation and knee-related quality of life. Higher scores indicate more normal joint function.25 26

VAS: Scores range from 0 to 10, with 0 indicating no pain and 10 indicating the most severe pain.27

Overall quality of life score: The EQ-5D and the SF-12 were selected for assessment. The EQ-5D is a tool developed and widely used in Europe to assess general quality of life. Patients can report their perceived health status on a scale from 0 (worst health status) to 100 (best health status).28 The SF-12 scores are grouped into two scales: the physical component score (PCS) and the mental component score (MCS). Both scores range from 0 to 100, with higher scores indicating better health status.29 And an objective physical measure based on the participant’s 6-min walk test.30 31

Function analysis: The artificial intelligence visual assessment system recorded data on movement angles, persistence duration and stability during remote rehabilitation were recorded from patients in the experimental group to assess the completion of their movements, and gait analysis as well as knee mobility assessment were performed (figure 5).

Clinical benefit assessment: Based on the improvement in KOOS score, the MCID, PASS and SCB were assessed.32–34

Figure 5

Function analysis.

Study execute time

10 April 10 2023–10 April 2025.

Exercise regimen

Both groups of patients will follow a standardised rehabilitation programme.

Both the AR rehabilitation group and the control group will receive a 12-week intervention consisting of two sessions per week, with each session lasting 60 min. The rehabilitation programme will be developed by a sports medicine physician.

Patients in the AR rehabilitation group will be trained by the researcher in the use of the system prior to rehabilitation to ensure that the patients were proficient in the use of the AR rehabilitation system. The researcher will be able to check the patient’s recovery record on the terminal and will contact the patient via the internet to supervise patients with poor compliance.

Patients will be allowed to use drugs such as non-steroidal anti-inflammatory drugs for systemic medication for osteoarthritis of the knee during the course of the study. All patients who withdrew will have their exit contact information archived, allowing for follow-up data collection and preventing data loss.

The specific training regimen consisted of joint mobility training, muscle strength training and aerobic exercise training, which will be divided into three phases10 35–40.

Phase 1: 1–4 weeks (the primary objective of the joint mobility and muscle stretching exercises will be to improve range of motion and flexibility.)

Seated knee extension: The patient sits in a chair and extends their leg upward while keeping the leg parallel to the ground. This exercise is repeated 20 times on each side and lasts for 10 min.

Knee flexion: Stand up straight with your legs slightly apart, and bend your knee to a 90-degree angle, and then straighten your leg again; repeat 20 times on each side for a total of 10 min.

Standing posterior hip extension: Stand straight with your hands on a chair for support and lift one leg straight back while keeping your back and other leg straight; repeat 20 times on each side for 10 min.

Standing hip abduction: Stand up straight, hold the chair back with your hands and abduct your leg upwards to the side while keeping the leg straight; repeat 20 times on each side for 10 min.

Hamstring stretch: Lie on your back, grasp your thighs with both hands, bend your hip and knee to a 90-degree angle, slowly extend your knee until you feel a stretch in the back of your thigh, hold for 30 s and then relax. Repeat this stretch two times on each leg for a total of 5 min.

Quadriceps isometric exercise: Lie on your back with a mat under your feet, and tighten the muscles at the front of your thigh to feel the contraction while keeping your knee straight. Repeat this 20 times on each side for 5 min and then rest for 2–3 min before performing two sets.

Lateral quadriceps stretch: Lie on your side, bend your knees, grab your ankles with your hands and hold for 30 s; repeat two times on each side for 5 min.

Phase 2: 5–8 weeks (increase the intensity of muscle strength training)

Quadriceps isometric exercise: Lie on your back with a mat under your feet, and tighten the muscles at the front of your thigh to feel the contraction while keeping your knee straight. Repeat this 20 times on each side for 5 min, and then rest for 2–3 min before performing two sets.

Resistance seated knee extension: Sit on a chair and extend your leg forward, keeping it parallel to the ground, while holding a resistance band under your foot; repeat 10 times on each side for 5 min, and then rest for 2–3 min; perform two sets.

Knee flexion: Stand up straight with your legs slightly apart and bend your knee to a 90 degree angle, and then straighten your leg again; repeat 20 times on each side for a total of 10 min.

Standing posterior hip extension: Stand straight with your hands on a chair for support, and lift one leg straight back while keeping your back and other leg straight; repeat 20 times on each side for 10 min.

Standing resistance hip abduction: Stand straight, and hold the chair with your hands. Keep your leg straight, and lift it to the side while using a resistance band tied around your legs. Repeat this movement 20 times on each side for a total of 10 min.

Side lying hip abduction with resistance band: Lie on your side on the yoga mat, with a resistance band around your thighs, above your knees. Keep your legs bent. Slowly lift your top thigh and then lower your top thigh back down and repeat 20 times on each side for 10 min.

Phase 3: 9–12 weeks (increased aerobic exercise)

Resistance seated knee extension: Sit on a chair and extend your leg forward, keeping it parallel to the ground, while holding a resistance band under your foot; repeat 10 times on each side for 5 min, and then rest for 2–3 min; perform two sets.

Side lying hip abduction with resistance band: Lie on your side on the yoga mat, with a resistance band around your thighs, above your knees. Keep your legs bent. Slowly lift your top thigh and then lower your top thigh back down, repeat 20 times on each side for 10 min, then rest for 2–3 min, and perform two sets in total.

Resistance alternate walking: Stand straight with arms crossed and resistance bands around the ankles, knees slightly bent into a small squatting position, shift weight over one leg and step sideways with the other foot, and then slowly bring the feet together; repeat 10 times on each side for 4 min, and then rest 2–3 min; perform two sets in total.

Aerobic training: Walk for 10 min at a heart rate reserve of approximately 40% to 60%.

Data management and quality control

Inclusion of subjects will be in strict accordance with the study plan.

Fully communicated with subjects to improve informed consent.

In case of adverse events during the study, immediately removed subject from the study and provided symptomatic treatment.

All data will be entered by two people and saved on a special computer, and the entered data will be organised and verified at each follow-up time point.

Statistical analysisMain objective

The data will be analysed using SPSS 22.0 statistical software, and the descriptive epidemiological method will be used to analyse the data. The Shapiro‒Wilk test for age compliance with normal distribution and homogeneous variance will be chosen first for the measurement data. Repeated measures analysis of variance (ANOVA) will be used to assess the between-group differences (intervention vs control), time effects (baseline, postintervention, and 24-week follow-up) and group-by-time interactions, with statistical significance set at p<0.05.

If the sphericity test will be not satisfied, the Greenhouse‒Geisser method will be used for correction, the Bonferroni method will be used for comparison between different time points in the same group, and multifactor ANOVA will be used for comparison between different groups at the same time point.

Secondary objectives

The MCID will be calculated using a stepwise approach, with the value of the MCID equal to half the SD of the improvement in the overall patient functioning score at 24 weeks after the end of the intervention, and patients will be categorised as having achieved the MCID if they achieved the MCID on ≥1 of the included outcome measures.41 42 The PASS will be calculated using an anchor-point-based approach assess using a 2-point scale (satisfactory or unsatisfactory) and will be assessed by receiver operating characteristic (ROC) curve analysis and the Youden Index to detect thresholds in order to maximise the sensitivity and specificity of the thresholds. For the purposes of this study, predictive models with area under the curve values greater than 0.7 will be considered acceptable, whereas predictive models greater than 0.8 were considered excellent. Patients will be categorised as achieving PASS if they achieved PASS on ≥1 included outcome measure.41 43 SCB will agree to use the anchor-based ROC methodology for the calculations; we will compare those who report ‘more improvement than I thought’ and ‘a lot of improvement’ with those who report ‘a little improvement’ or ‘no improvement’.44

Missing data

For missing data, a general linear mixed effects regression model was fitted to statistically analyse the data to determine if the data could be missing completely at random, missing at random or missing at non-random. Multiple interpolation was used to fill in the missing data.

End of study

The study will end when all subjects have completed 12 weeks of follow-up or when one of the following occurs: (1) the ethics committee of the participating study unit requests that the study be discontinued and (2) the study is no longer receiving funding.

Adverse event

Adverse event monitoring will begin when participants are randomly assigned and will continue until the end of follow-up. We will record serious adverse events (eg, death, severe disability or incapacity) and adverse events that may be relevant to the intervention. We will assess the probability and severity of occurrence and relevance to the intervention for all reported adverse events. Interventions will be stopped immediately in all patients who experience adverse events, and relevant investigations will be completed and symptomatic treatment will be administered.

Ethics

This research project was approved by the Biomedical Ethics Review Committee of West China Hospital of Sichuan University and will be conducted in strict accordance with ethical guidelines. In this study, we will respect and protect the rights and privacy of the participants and ensure the confidentiality of their personal information.

Participants’ informed consent: We will explain the purpose, process, risks and benefits of the study to all individuals who will participate in the study either orally or in writing and obtain their informed consent. Participants will participate voluntarily and could withdraw from the study at any time.

Data confidentiality and privacy protection: We will strictly protect the privacy of participants’ personal information. We will not publicise or disclose any personal information that may lead to the identification of participants. In the study, we will anonymise participants’ information. We will replace the subject’s recognisable information with irrelevant character sequences instead.

Potential risk assessment and management: We will assess the potential risks that may be involved in the study during the project design phase, and we will ensure that participants will not suffer any physical or psychological harm as a result of participating in the study.

Use of research data: We will strictly adhere to the principles of legality and transparency in the use of data to ensure the correct use and interpretation of research data. We will avoid misinterpretation and misuse of the data as much as possible and will only use the data for research purposes.

Dissemination policy

The results of our study will be published through articles in open access journal articles and conference presentations. We will notify participants of the results of the study through a web-based newsletter. We will not use professional writers for writing. We fully support the data sharing programme, and both the public and professionals can request access to the full protocol, study data and statistical codes.

Protocol version

3 March 2024 V.2.0.

Patient and public involvement

During the study design process, we solicited the opinions of osteoarthritis patients with previous outpatient follow-up, including those who regularly participated in training and those who did not participate in exercise rehabilitation, and defined our research questions based on this by learning about their preference for exercise, how they would like to be intervened in and their dissatisfaction with existing traditional rehabilitation treatment measures. Patients were not involved in the recruitment and implementation of the study, and the results of the study will be pushed to the study participants via social media.

Ethics statementsPatient consent for publication

Not applicable.

Acknowledgments

We acknowledge all the patients who participated in this study.