Design and Development of Rehabi, a mHealth Telerehabilitation Platform with Markerless Motion Analysis
, Hipólito Aguilar-Sierra 2, Rafael Zepeda-Mora 1, Aldo Alessi-Montero 1, Gerardo RodrÃguez-Reyes 1, Lidia Núñez Carrera 1, Ivett Quiñones-Uriostegui 1, Paola Ayala-Cadena 1 and Adriana Gomez-Verdad 1
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsDear authors,
I have reviewed the manuscript bioengineering-4165208 and can confirm that the Introduction has a solid and up-to-date contextualization (telemedicine, mHealth, population aging), with an attempt to integrate into the relevant literature, with the presentation of the gap (offline-first, accessibility, low-resource populations) and with the pursuit of the study objective. I found the methodology well-structured, with a detailed description of the software architecture. I appreciate that the authors have specified defined inclusion/exclusion criteria for the clinical study, that they have clearly presented the data and that they have compared the results with the existing literature and that the authors have not ignored the limitations (capture variability, contracts). I consider the suggestion of coherent future directions and the cautious conclusions a plus.
Please, however, analyze the following observations:
0) The purpose of the Abstract, although essentially the same as the one in the Introduction, however, the wording may provide confusion to the readers, because in the Abstract it highlights Rehabi and the validation of markerless motion analysis, and at the end of the Introduction, it presents Rehabi and the integration of pose detection plus support for low-resource populations.!!! The Introduction emphasizes design and accessibility, and the Abstract puts more emphasis on clinical validation. I ask the authors to check this aspect and to standardize, to clarify the purpose everywhere in the manuscript with the same wording, in order to be understood clearly and without doubts by the journal readers. Thank you
1) The Abstract does not clearly specify the study design (pilot study) and does not summarize the small sample size as a limitation. Maybe it would be ok if it were reformulated more carefully. Thanks
2) The introduction is excessively extensive (it includes technical descriptions of other systems that are too detailed), and some sections seem like a narrative review, not a purpose-focused introduction. I also recommend including the approach on the integration of digital technologies, including AI and ICT, in the field of rehabilitation and body movement technologies, which is related to the general topic of markerless motion capture and mHealth technologies: https://doi.org/10.3390/app15179826.
3) In Materials and Methods, it is not specified: the exact number of respondents, the distribution of patients vs. clinicians, the selection method. The validation of the instrument (questionnaire) is not reported and NLP metrics are missing (model used, thematic validation, reproducibility). I ask the authors to clarify and provide details on these aspects. Thanks
System Design is extremely technically detailed, with descriptive redundancy.
Clinical Validation presents a very small sample (n=14), no statistical power calculation, no standard error analysis, and no confidence intervals for ICC are mentioned. The experience of the raters and the control of inter-rater variability are not described. I ask the authors to clarify and provide details on these aspects. Thank you
Replicability for the software part is real because this part is very well described, with an explicit mathematical formula. The clinical part is ok, inclusion/exclusion criteria defined and the exercise protocol described. However, I miss the camera setup (exact distance, height, lighting), hardware specifications, frame rate, exact MediaPipe version, ICC calculation method (ICC(2,1)? ICC(3,1)? model) and statistical software used. I ask the authors to clarify and provide details on these aspects. Thank you
4) Results does not present ICC confidence intervals, separate analysis THA vs TKA and analysis of the influence of postoperative time. For Bland–Altman, the limits are reported but not critically discussed. I ask the authors to clarify and provide details on these aspects. Thank you
Missing Statistical analyses: ICC confidence interval, power analysis and proportional systematic error analysis. I ask the authors to clarify and provide details on these aspects. Thank you
5) Discussion does not sufficiently address the limitations given by the sample size, the lack of test-retest, the lack of inter-rater evaluation, and the statements about scalability are prospective, not demonstrated. I ask the authors to clarify these aspects. Thank you
6) Conclusions introduce concepts not validated in the study (“hybrid sensor fusion”, “clinician-in-the-loop workflows”), which have not been evaluated experimentally.
7) Exaggerations in Discussion / Conclusions
Moderate exaggerations for “clinically useful” (acceptable for the knee, questionable for the hip, ICC 0.687 = moderate). Slight exaggerations in the conclusions for “scalable deployment” (because these are not experimentally validated).
8) The conclusions are ok, but there is a small problem: they extend the scope to unvalidated future directions.
9) References. I ask the authors to check the relevance of references #12, #13, #14#15. These are statistical report type sources, not peer-reviewed. I don't think they are irrelevant, but they are gray sources, not academic. I recommend https://doi.org/10.3390/app15179826 (In Introduction/ Discussions in the area where you discuss AI, MediaPipe, markerless tracking and movement digitization, after the discussion on global trends in AI integration in human movement or in the subsection where state-of-the-art solutions are presented, to provide depth to the concepts that support the idea proposed by the authors. Thank you.
12) Table 2 does not have defined units in the legend (e.g. kg, m – although they appear in the header). “bilateral hip” is not explained in the legend.
The manuscript is technically sound, methodologically correct for a pilot study, statistically adequate, I would even like to offer a Minor revision, but the authors still need to provide information and clarify the previously mentioned aspects, for replicability, eliminating aspects that are not validated by data and results (assuming them perhaps as a limit) and verifying missing statistical aspects.
Thus, although it is a solid and promising study, I will offer a Major revision, with the hope that the authors will provide details, adapt and argue and clarify the targeted aspects.
I trust that they can do it.
Thank you
Author Response
We sincerely thank the reviewers for their thorough evaluation of our manuscript and for their constructive feedback. Their comments significantly strengthened the clarity, methodological transparency, and scientific rigor of the revised version. Below, we provide a detailed, point‑by‑point response outlining how each comment was addressed. All line numbers refer to the revised manuscript. We have highlighted many of the applied changes in yellow in the article.
|
Section |
Comment / Observation (Original) |
Corrections applied |
Lines |
|
Abstract |
“The purpose of the Abstract… wording may provide confusion… Abstract highlights Rehabi and validation… Introduction presents Rehabi and integration of pose detection plus support for low-resource populations… standardize wording.” |
The abstract has been revised for clarity and consistency. We now explicitly state that this is a pilot study evaluating a markerless motion algorithm and acknowledge the small sample size as a limitation |
|
|
Abstract |
“The Abstract does not clearly specify the study design (pilot study) and does not summarize the small sample size as a limitation.” |
The introduction has been shortened and refocused. We added the recommended AI/ICT integration reference and improved transitions to maintain conceptual coherence |
17-22 |
|
Introduction |
“The introduction is excessively extensive… some sections seem like a narrative review… recommend including AI and ICT integration reference (doi:10.3390/app15179826).” |
The introduction has shortened, and a reference to AI/ICT integration has been added. |
46-60 |
|
Methods |
“Not specified: exact number of respondents, distribution of patients vs. clinicians, selection method. |
Further details on this topic are provided in Section 3.1 |
304-308 |
|
|
The validation of the instrument (questionnaire) is not reported |
In section 2.1 Because the questionnaire was conceived as an exploratory tool for requirements elicitation rather than as a standardized measurement scale, a full psychometric validation was not undertaken. Instead, its adequacy was supported by operational validation processes inherent to software development, including iterative prototyping. In the discussion, we propose that the questionnaire could go to a full psychometric validation |
108-111 |
|
|
NLP metrics are missing.” |
Further details on this topic are provided in Section 2.1 |
129-138 |
|
System Design |
“System Design is extremely technically detailed, with descriptive redundancy.” |
We have worked on the system design section |
|
|
Clinical Validation |
“Very small sample (n=14), no statistical power calculation, no standard error analysis, no confidence intervals for ICC… rater experience and inter-rater variability not described.” |
We have acknowledged the pilot nature, and section 2.2 was improved, taking into account the observation
Clinical goniometric assessments were performed by two board-certified physicians specializing in Physical and Rehabilitation Medicine, each with advanced subspecialty training in Orthopedic Rehabilitation and more than five years of experience managing joint arthroplasty patients. Patients with total knee arthroplasty (TKA) were evaluated by one specialist with extensive clinical experience in knee arthroplasty rehabilitation, while patients with total hip arthroplasty (THA) were evaluated by a different specialist with equivalent expertise in hip arthroplasty rehabilitation. To minimize measurement bias, both evaluators were blinded to the algorithm-derived measurements. Clinical goniometric assessment was performed first, followed by markerless motion analysis, ensuring that the reference measurements were not influenced by the algorithm results. |
412-446 |
|
Technical Details |
“Camera setup (distance, height, lighting), hardware specifications, frame rate, exact MediaPipe version, |
We have added technical details
Video recordings were acquired using a Lenovo Tab M11 tablet (MediaTek Helio G88, 8 GB RAM, 128 GB storage) equipped with a 13‑megapixel rear camera. The device was hand‑held by an operator at a height of around 1.50 m and a distance of 2.5 m from the participant, ensuring that the entire body remained visible within the frame. Recordings were captured at 1080p resolution with a frame rate of 15–20 Hz, under controlled lighting conditions with diffuse illumination and a neutral background to minimize landmark detection errors. Pose estimation was implemented using MediaPipe (com.google.mediapipe:tasks‑vision:0.10.21) with the Pose Landmarker full model, integrated through OpenCV and Chaquopy. This configuration provided robust anatomical landmark detection and enabled reliable estimation of hip, knee, and ankle joint angles during the rehabilitation exercises.
|
269-279 |
|
Results |
ICC calculation method, statistical software not reported.” |
we have expanded the statistical analysis to include 95% confidence intervals for all intraclass correlation coefficients (ICC), calculated using a two-way mixed-effects model for absolute agreement (ICC(A,1)), consistent with established methodological recommendations. For hip flexion, the ICC was 0.686 (95% CI: 0.362–0.863, p < 0.001), and for knee flexion, the ICC was 0.801 (95% CI: 0.571–0.916, p < 0.001). These confidence intervals have now been added to the Results section and provide a measure of precision and reliability of the estimated agreement between the markerless motion analysis system and clinical goniometry. |
412-446 |
|
Results |
“Results does not present ICC confidence intervals, separate analysis THA vs TKA, analysis of postoperative time. |
In the THA subgroup, postoperative time was not associated with absolute measurement error for hip flexion (β = −0.049° per month, p = 0.637; R² = 0.021). Similarly, postoperative time was not associated with signed differences (β = 0.055° per month, p = 0.715). Spearman's correlation indicated no significant association between postoperative time and absolute error (ρ = 0.212, p = 0.487). A non-significant trend was observed for signed differences (ρ = 0.540, p = 0.056). In the TKA subgroup, postoperative time showed a non-significant trend toward reduced absolute measurement error for knee flexion (β = −2.047° per month, p = 0.0837; R² = 0.482). No association was observed for signed differences (β = −1.049° per month, p = 0.648). Spearman correlation similarly suggested a non-significant negative association between postoperative time and absolute error (ρ = −0.692, p = 0.0847). Given the limited sample size in the TKA subgroup (n=7), these findings should be interpreted as exploratory. |
412-446 |
|
|
Bland–Altman limits reported but not critically discussed.” |
Bland–Altman limits discussed in the discussion section |
457-509 |
|
Statistics |
“Missing Statistical analyses: ICC confidence interval, power analysis, and proportional systematic error analysis.” |
Icc confidence interval has been added and discussed in the discussion section |
457-509 |
|
Discussion |
“Does not sufficiently address limitations… scalability statements are prospective, not demonstrated.” |
We have added limitations and scalability statements to the discussion section |
457-509 |
|
Conclusions |
“Introduce concepts not validated (‘hybrid sensor fusion’, ‘clinician-in-the-loop workflows’).” |
We have worked over conclusion section |
512-522 |
|
Conclusions |
“Moderate exaggerations for ‘clinically useful’… slight exaggerations for ‘scalable deployment.’” |
We have worked on conclusion section |
512-522 |
|
References |
“Check relevance of references #12–15… recommend doi:10.3390/app15179826.” |
Statistical references have been removed, and text has been rephrased to give support to claims |
|
|
Tables |
“Table 2 does not have defined units… ‘bilateral hip’ not explained.” |
Units have been checked, bilateral hip has been rephrased for better understanding |
447-448 |
Reviewer 2 Report
Comments and Suggestions for AuthorsDear authors,
I reviewed your manuscript. The subject is a significant focus in current rehabilitation sciences. This app has the potential to greatly enhance rehabilitation efforts. You can find my comments below:
In the introduction,
the transition from general telemedicine to telerehabilitation lacks a signposting sentence that narrows the focus to motion-based rehabilitation.
The introduction does not define the clinical problem (for example, specific rehabilitation contexts, conditions, or impairments) that Rehabi primarily targets.
By reading this section, I can't understand whether the primary contribution is clinical validation, system architecture, user-centered design methodology, or markerless motion analysis.
Claims regarding the benefits of telerehabilitation (for example, increased frequency, empowerment) are presented without discussion of mixed or negative evidence.
The introduction does not specify how biomechanical models differ from or improve upon existing markerless kinematic approaches in the literature.
In your method,
software versions, library versions, and hardware dependencies are not specified.
The absence of pseudocode, workflow diagrams, or algorithmic descriptions limits technical reproducibility.
For Software Architecture and System Design,
the specific layers implemented (e.g., entities, use cases, interface adapters, frameworks) are not explicitly mapped.
The interfaces used to abstract MediaPipe, OpenCV, and Chaquopy are not technically described.
It is unclear how dependency inversion was enforced at the code level.
No architectural diagram is referenced in the Methods section.
For Markerless Motion Analysis Methodology,
the camera specifications (resolution, frame rate, device models) used for motion capture are not reported.
The body model or skeletal representation used for joint angle computation is not explicitly defined.
It is unclear whether filtering or smoothing was applied to raw pose data.
Calibration procedures (if any) for camera placement or subject positioning are not described. The handling of occlusions, missing key points, or noisy detections is not explained.
In the discussion,
you don't explain why the reported accuracy is higher or lower than in comparable studies.
Differences in protocols, camera setups, or validation standards across studies are not analyzed.
Best regards.
Author Response
We sincerely thank the reviewers for their thorough evaluation of our manuscript and for their constructive feedback. Their comments significantly strengthened the clarity, methodological transparency, and scientific rigor of the revised version. Below, we provide a detailed, point‑by‑point response outlining how each comment was addressed. All line numbers refer to the revised manuscript. We have highlighted many of the applied changes in yellow in the article.
|
Section |
Comment / Observation (Original) |
Corrections applied |
Lines |
|
Introduction |
“Transition from general telemedicine to telerehabilitation lacks a signposting sentence that narrows the focus to motion-based rehabilitation.” |
We have worked on the introduction section and tried to improve it to address this matter.
|
36-60 |
|
Introduction |
“The introduction does not define the clinical problem… specific rehabilitation contexts, conditions, or impairments are not clear.” |
The initial clinical focus is on postoperative care following total hip arthroplasty (THA) and total knee arthroplasty (TKA), procedures commonly performed for osteoarthritis and related conditions, where objective, repeatable measures of range of motion and compensatory movement are critical for tailoring rehabilitation |
36-41, 89-93 |
|
Introduction |
“By reading this section, I can't understand whether the primary contribution is clinical validation, system architecture, user-centered design methodology, or markerless motion analysis.” |
The primary contribution is a practical, user-friendly application that integrates markerless motion capture and demonstrates preliminary clinical feasibility for objectively monitoring hip and knee range of motion |
80-82 |
|
Introduction |
“Claims regarding benefits of telerehabilitation… presented without discussion of mixed or negative evidence.” |
Further details are addressed in the introduction and discussion sections |
457-510 |
|
Introduction |
“The introduction does not specify how biomechanical models differ from or improve upon existing markerless kinematic approaches.” |
We tried to emphasize about markeless motion analysis system advantages |
55-61
|
|
Methods |
“Software versions, library versions, and hardware dependencies are not specified.” |
We have added version of software.
Software library version were specified on section 2.1. Android Studio (Ladybug Feature Drop 2024.2.2), MediaPipe (com.google.mediapipe:tasks‑vision:0.10.21), [29], OpenCV (org.opencv:opencv:4.9.0) [30], and Chaquopy (15.0.1) |
151-172 |
|
Methods |
“The absence of pseudocode, workflow diagrams, or algorithmic descriptions limits technical reproducibility.” |
We have tried to be clearer in system design and added a new figure to illustrate the workflow of the markerless motion algorithm. |
Section 2.2 |
|
Markerless Motion Analysis |
“Camera specifications not reported… skeletal representation not defined… filtering/smoothing not explained… calibration procedures missing… handling of occlusions not described.” |
The Domain Layer then applies the biomechanical logic encapsulated in the AngleCalculator class. This component constructs a simplified skeletal model of the lower limb using hip, knee, ankle, and toe landmarks from MediaPipe, which already provide normalized coordinates. In addition, missing landmarks were handled by using linear interpolation, in which the last valid coordinate was connected to the next available coordinate by a straight line. Intermediate frames were filled by proportionally distributing values along this line, ensuring smooth, continuous motion without abrupt jumps. On top of this skeletal representation, the AngleCalculator implements reusable mathematical routines for joint angle estimation, defined entirely in terms of vectors and dot products as defined
to avoid numerical errors a 4th‑order Butterworth low‑pass filter with a 6 Hz cutoff frequency is applied to the calculated joint angles, reducing frame‑to‑frame jitter and attenuating high‑frequency noise while preserving physiologically meaningful motion [32]. |
Section 2.2 |
|
Discussion |
“You don't explain why the reported accuracy is higher or lower than in comparable studies… differences in protocols, camera setups, or validation standards not analyzed.” |
We have addressed the observation in the discussion section.
The pilot clinical validation of the markerless motion algorithm demonstrated preliminary feasibility for mobile markerless motion analysis in post‑arthroplasty patients. In a cohort of 14 participants, hip flexion showed moderate agreement (ICC = 0.687), while knee flexion demonstrated good agreement (ICC = 0.803). These results support the feasibility of mobile markerless assessment, yet they should be interpreted as early‑stage findings requiring confirmation in larger, controlled studies. Differences from the higher ICCs reported in the literature (0.81–0.98 for spatiotemporal and goniometric metrics [22–24]) can be attributed to methodological heterogeneity, including variations in capture hardware, sampling frequency, camera geometry, operator technique, pose‑model conventions, signal‑processing pipelines, and the clinical reference standard. Importantly, these results align with trends highlighted in the Introduction: the shift toward mHealth delivery, the growing use of AI‑driven motion‑tracking tools, and the emergence of markerless systems as practical, low‑cost, and easy-to-use alternatives to traditional motion‑capture technologies [16,17,21].
|
477-490 |
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsDear Authors,
Thank you for your responses and for the adjustments made in response to your comments. I believe the changes made are appropriate and well-reasoned.
I have no further comments at this time.
Good luck
