The IMU-based sensors were comprised of 6-axis microelectromechanical systems that included accelerometers and gyroscopes. These devices collect information about the angular motion of the shoulder of interest. Three sensors are required for angular measurements of the shoulder. In this study, they were strapped to the sternum, distal third of the lateral upper arm approximately 5 cm above the lateral epicondylar of the humerus bone, and dorsal wrist (Figure 1). Three sensors were calibrated in position A once the measurement began (Figure 1A). The initial calibration process involved placing the sensor on a horizontal fixture to measure the offset for each axis, to align the initial difference for each sensor. The calibration process involved obtaining the initial value of the sensor for relative offset correction instead of correcting the deviation caused by incorrect sensor placement. In addition, the process entailed identifying the original offset of the sensors to each other and applying the zero setting using an algorithm.

Dual-sensor systems were used in the motion sensor device to accurately record the shoulder ROM angle. Using the sensors on the sternum as a reference datum point, the raw data from each sensor can be converted into a quaternion algorithm to convert the relative angle changes of the sensors on the upper arm and wrist into a 3-dimensional (3D) motion of the shoulder structure. Although one sensor can provide the angular change with the gyroscopes on one plane of motion, dual-sensor systems can provide relative angular changes in shoulder movements between the arm and axial body (by compensating for the angle of the body tilting relative to the ground), which was superior for eliminating the error in the shoulder ROM measurement attributed to body tilting from a one-sensor system, especially for patients with AC and painful shoulder movement (Figure 2). Additionally, repeated angular measurements with the dual-sensor system on simulated shoulder motion was validated as highly accurate. This is referred to in Multimedia Appendix 2.

To design the mobile app for patients with AC for use on mobile phones, a cocreation process was used. The app has modes for measuring shoulder mobility, generating historical records of angular measurements and exercise completion rates, and providing daily shoulder exercises with detailed instructions (Figure 3). The mobile app determines shoulder ROM in all directions by calculating the relevant information regarding the angle measurement reported by the 3 wearable sensors (Figures 2B and 3B). The Patient App provides 7 sets of home-based shoulder exercises for training, including a forward wall walking stretch, lateral wall walking stretch, and cane stretch for shoulder flexion, extension, abduction, internal rotation, and external rotation (Figure 4). The mobile app assigns each set of exercises as a “daily task” assigned by the supervising physiotherapist or physicians and provides a demonstration using a 3D avatar (Figure 3E). The app mirrors the user’s shoulder movement during the exercise using angular information collected from the worn motion sensors. Each user has an account in the Patient App to access his or her records, which includes daily progress and the completion rate of the daily exercises (Figures 3C and 3D).

The Doctor App was designed for physiotherapists and physicians for use on a mobile tablet. This app provides each patient’s information, including the latest shoulder ROM measurements and exercise completion rates for the previous week (Figure 5A). Physiotherapists and physicians can assign personalized daily home-based exercises with adjustable targeted angles, numbers of repetitions, and holding times for each patient based on the angular status of the affected shoulder (Figure 5B). Physiotherapists/physicians can directly communicate with the patients by sending text messages through the Doctor App.

From January 2017 to August 2017, two groups of participants — 10 healthy participants and 15 patients with a confirmed diagnosis of AC — were prospectively enrolled for measurement of shoulder angular motion.

Volunteers aged 20 to 70 years were included in the group of healthy participants. Participants who reported discomfort or limited ROM of the shoulder within the preceding year were excluded. Patients were included in the AC group who (1) were aged between 40 and 70 years, (2) had spontaneous onset of a painful stiff shoulder and marked loss of active and passive global shoulder motion, (3) had reported local shoulder pain that was frequently present either over the anteromedial aspect of the shoulder extending distally into the biceps region or over the lateral aspect of the shoulder extending into the lateral deltoid region, and (4) had symptoms for at least 3 months, with normal findings on anteroposterior and axillary radiographs of the glenohumeral joint. Patients were excluded if they (1) had pathological findings of rotator cuff tear on physical examination (eg, abduction weakness and positive dropped arm test); (2) had glenohumeral osteoarthritis upon radiographic evaluation; (3) had clinical evidence of severe cervical spine disease; (4) had a history of severe trauma to the shoulder; (5) received a local corticosteroid injection or any physiotherapeutic intervention to the affected shoulder in the 3 months preceding the study start date; (6) had a history of inflammatory joint disease, infection, stroke, or thyroid diseases that affected the shoulder; or (7) were unwilling to undergo an intervention or participate in the trial.

For angle measurements, aROM and pROM for flexion, extension, abduction, and internal and external rotation at a neutral position for both shoulders of healthy participants and for the affected shoulder of the patients with AC were assessed. The shoulder ROM in all directions for each participant was first assessed using the motion sensor device. Two examiners (one well-trained physiotherapist and one highly experienced orthopedic surgeon) who were blinded to the angle measurements by the motion sensor device measured shoulder ROM in all directions using a universal goniometer.

Each participant stood straight, and scapular rotation was permitted to measure abduction, flexion, and extension (Figures 1B, 1C, 1D). External rotation was measured in a neutral position with the shoulder abducted, the elbow flexed at 90°, and the forearm in a neutral position; the angle between the long axis of the forearm and the sagittal plane of the trunk was measured as external rotation (Figure 1E). Internal rotation was determined by having the participants reach up the back to the highest point along the midline (Figure 1F). Six levels of internal rotation were defined according to the areas that the participant’s hand reached, as follows: interscapular region (T7), T12 vertebra, waist (L3), lumbosacral junction, buttock, and lateral thigh. A high reach corresponded to better internal rotation. The aROM was measured by instructing participants to move their arm as far as they could, and the pROM was measured as the examiner moved the subject’s arm to its mechanical limit or a limit imposed by pain.

The 15 patients with confirmed diagnoses of AC were allocated to 2 intervention groups: 7 patients to the home-based exercise group and 8 patients to the motion sensor–assisted rehabilitation group. The patients were allocated according to their willingness to use, and familiarity with, mobile apps. The patients in both groups received 3 months of rehabilitation.

Patients with AC in the home-based exercise group received comprehensive instructions on the daily shoulder exercises from an experienced orthopedic surgeon before beginning the exercise program. The information included instructions for home-based exercises; a description of frozen shoulder; and advice on sleep, posture, and pain relief. Home-based exercises included Codman’s or pendulum exercises (circumduction) and passive stretching exercises, including “wall walking stretch exercises” and “cane stretch exercises” (Figure 4). Further information regarding the standard protocol of the home-based exercises is available in Multimedia Appendix 3. The patients were instructed to perform these exercises within the painless range, maintain the maximal achievable angle for at least 10 seconds in each exercise, and repeat each exercise at least 10 times daily. All patients were instructed to follow the standard home-based exercise protocol, but individualized modification was permitted based on health professionals’ judgment. In practice, an orthopedic surgeon reviewed the progress of each patient every month. They gave advice and modified the exercise program based on the patient’s compliance and AC status. Therefore, if a patient had poor performance on a specific direction of shoulder ROM after the exercise program of the previous month (eg, on shoulder abduction), the orthopedic surgeon could make adjustments to the next month’s exercise protocol by increasing the daily exercise frequency to 15-20 times and the holding time to 15-20 seconds per session for the “lateral wall walking stretch” and “cane stretch for shoulder abduction” exercises.

All patients in the motion sensor–assisted rehabilitation group received detailed instructions on operating the motion sensor device and Patient App on their own mobile phones before they began the rehabilitation program. The aforementioned orthopedic surgeon assigned the standard protocol of daily home-based shoulder exercises to each patient via the Patient App with individualized modifications according to the patient’s performance. The completion rate of these daily tasks was calculated using the app for each patient and sent to the Doctor App over the internet, enabling the supervisor to immediately discuss the progress of each patient with them. In addition, patients in the motion sensor–assisted rehabilitation group were individually assigned each set of stretching exercises with the target angle on the mobile app by the orthopedic surgeon based on the patients’ previous shoulder angular performance (Figure 5B). For example, if the patient had 100° shoulder abduction before starting the shoulder exercise program, the target angle for the “lateral wall walking stretch” and “cane stretch for shoulder abduction” would be assigned as 120° with an acceptable range of 30°. Achievement of the target angle of the assigned exercise could be recorded as a successful count for calculating the daily exercise completion rate on the mobile app (Figure 3D). In this manner, the patient could therefore stretch the involved shoulder to the maximal limit of the angle on each attempt, thereby effectively maximizing the training effect of stretching exercises.

The self-reported qDASH, consisting of 11 items, was developed to measure the relevant physical function and symptoms of patients with upper limb disorders [19]. The Chinese version of qDASH has excellent reliability (Cronbach α=.818, intraclass correlation coefficient [ICC]=0.907) [20]. A higher qDASH score indicates poorer shoulder function.

A VAS is an instrument that is regularly used to measure the intensity of pain [21]. A researcher asked the patients, “On a scale of 0 to 10, how severe has been the worst pain that you have experienced within the last week?” This question was repeated in the follow-up sessions.

The exercise completion rate reported by the motion sensor device was calculated by dividing the number of exercises completed daily by the number of assigned exercise tasks daily. The monthly exercise completion rate was the average of the daily exercise completion rates over an entire month. The patient-reported exercise completion rate in each month was obtained by asking each patient, “On a scale of 0%-100%, what was your average exercise completion rate in the previous month?”