The DOI matches its Crossref record; 37 of the 37 references it cites were found in a registry or an index; 30 of the 36 sources its claims cite were checked for support.
DS-MS-TCN: Otago Exercises Recognition With a Dual-Scale Multi-Stage Temporal Convolutional Network
Open accessAbstract
The Otago Exercise Program (OEP) represents a crucial rehabilitation initiative tailored for older adults, aimed at enhancing balance and strength. Despite previous efforts utilizing wearable sensors for OEP recognition, existing studies have exhibited limitations in terms of accuracy and robustness. This study addresses these limitations by employing a single waist-mounted Inertial Measurement Unit (IMU) to recognize OEP exercises among community-dwelling older adults in their daily lives. A cohort of 36 older adults participated in laboratory settings, supplemented by an additional 7 older adults recruited for at-home assessments. The study proposes a Dual-Scale Multi-Stage Temporal Convolutional Network (DS-MS-TCN) designed for two-level sequence-to-sequence classification, incorporating them in one loss function. In the first stage, the model focuses on recognizing each repetition of the exercises (micro labels). Subsequent stages extend the recognition to encompass the complete range of exercises (macro labels). The DS-MS-TCN model surpasses existing state-of-the-art deep learning models, achieving f1-scores exceeding 80% and Intersection over Union (IoU) f1-scores surpassing 60% for all four exercises evaluated. Notably, the model outperforms the prior study utilizing the sliding window technique, eliminating the need for post-processing stages and window size tuning. To our knowledge, we are the first to present a novel perspective on enhancing Human Activity Recognition (HAR) systems through the recognition of each repetition of activities.
Overview
CoChat Summary
TL;DR
The paper proposes DS-MS-TCN, a dual-scale multi-stage temporal convolutional network with micro labeling, to recognize Otago Exercise Program exercises from a single waist-mounted IMU in community-dwelling older adults. In lab data it achieved f1-scores over 85% and IoU f1-scores over 60% for all four exercises, but performance dropped substantially when generalizing to home data.
Key findings
- The proposed DS-MS-TCN achieved lab LOSOCV f1-scores of 89.777% for ankle plantarflexors, 85.363% for abdominal muscles, 88.298% for knee bends, and 90.028% for chair rising; IoU f1-scores were 66.667%, 68.852%, 63.333%, and 92.308%, respectively.
- Baseline models did not surpass 50% IoU f1-score for ankle plantarflexors, abdominal muscles, or knee bends, despite some reporting f1-scores above 80%; the MS-TCN without micro labels reached the highest baseline IoU f1-score of 70.27% for chair rising.
- Using one stage for micro-label classification outperformed two stages; for example, chair rising f1/IoU f1 was 90.028%/92.308% with one stage versus 84.483%/73.446% with two stages.
- Generalization to home data reduced performance: ankle plantarflexors f1 dropped from 89.777% to 48.897% and IoU f1 from 66.667% to 9.524%; chair rising IoU f1 dropped from 92.308% to 15.385%.
Methods
- Community-dwelling older adults aged 65 and older performed modified OEP and ADLs while wearing a single waist-mounted McRoberts MoveMonitor+ 9-axis IMU. The lab dataset included 36 older adults; the home dataset included 7 additional older adults.
- The DS-MS-TCN used four SS-TCN stages: stage 1 classified micro labels, stage 2 classified macro labels from micro-label probabilities, and stages 3–4 refined macro labels. Models were trained seq-to-seq on 40-second slices with 50% overlap, batch size 32, Adam, PyTorch, and an NVIDIA P100 SXM2 GPU.
- Evaluation used Leave-One-Subject-Out Cross-Validation on lab data, then trained on lab data and generalized to home data. Metrics were sample-wise f1-score and segment-wise IoU f1-score with an overlap threshold of 0.5; baselines included a hierarchical sliding-window system, CNN, Transformer, CNN-LSTM, and MS-TCN without micro labels.
Limitations
- The paper states that generalization to home data was lower because home participants followed only a booklet without researcher instruction, performed exercises less standardly, and could perform other activities out of camera view; future work needs more daily-life data and improved generalization.
- The paper states that micro-label annotation requires a huge workload and is time-consuming, and that micro-label recognition itself was not expected to be perfect.
- The paper states that with only one waist IMU, only activities involving trunk movement could be recognized. Evident from the text: the home validation used only 7 participants and different subjects from the lab set, limiting the generalization estimate.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
14 Valid · 14 Not valid · 3 Not checked · 59 Own results and interpretations
At least one cited source doesn’t support this claim or points to a different paper, and none confirmed it — see the verdicts below.
In the paper
“The previous study was collected in a highly controlled environment (young healthy adults in the lab).”
What the sources say
- Disputedcites [7] sourceExperimental — accuracy not yet measured
Section 3.1 says this study collected data from community-dwelling adults aged 65 and older, not young healthy adults. Sections 3.1.1 and 3.1.2 describe both a lab dataset and a home dataset. The paper’s reference to four healthy young subjects in Section 5.3 concerns a different, earlier study. (LLM, conf=0.90)
Checked by a reasoning model — may be wrong.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Disputed
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“On the other hand, in the field of action recognition by videos, it has been proven efficient to train the models with a time series of frames [11].”
What the sources say
- Partly supportedcites [11] sourceExperimental — accuracy not yet measured
The Introduction describes activity recognition as a time-series classification problem, and §3.8 reports improved accuracy from modeling temporal dependencies in sensor data. But the paper studies inertial-sensor signals, not training video-action models on sequences of frames; its Introduction mentions video methods only in general terms. The claimed efficiency for video frames is therefore not established by this source. (LLM, conf=0.84)
boostedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources that could be checked couldn’t confirm or dispute this claim.
In the paper
“The previous study [29] reported lower f1-scores for chair rising because it was confused with knee bends .”
What the sources say
- Could not verifycites [29] sourceExperimental — accuracy not yet measured
The ROI prediction section reports relatively low intersection-over-union scores for Arising From Chair, but those are not F1-scores. The Video-level classification section identifies confusion among hand tasks and between Toe Tapping and Leg Agility; the passages found for F1, chair rising, and knee bends do not report chair rising being confused with knee bends. (LLM, conf=0.76)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Could not verify
Every source the paper cites for this claim was checked and supports it.
In the paper
“Despite previous efforts utilizing wearable sensors for OEP recognition, existing studies have exhibited limitations in terms of accuracy and robustness.”
What the sources say
- Supportedcites [7] source
The Abstract (page unavailable) describes limitations in both earlier wearable-sensor OEP studies: one tested only healthy young adults and did not distinguish OEP from daily activities, while the other achieved f1-scores no higher than 0.8. These are limitations in real-world applicability and recognition accuracy, respectively. (LLM, conf=0.87)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“Also, the determination of sliding window sizes requires a lot of prior knowledge and experiments for optimized results, as in the previous study for OEP recognition [7].”
What the sources say
- Supportedcites [7] source
Section 2.1 says window size depends on the characteristics of the activities being recognized. Section 3.3.1 shows the study identified an optimal size and overlap by comparing multiple settings, supporting the citing sentence’s description of activity knowledge and experimentation. No page numbers are available in our copy. (LLM, conf=0.86)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
40 more claims · 4 with their cited sources checked
Create a free account to see all 90 claims and every source verdict.
- Probably not a substantial claim. 22
- It may state two separate claims as one. 30
Citations
Cites
Sarcopenia: European consensus on definition and diagnosis
Cruz-Jentoft, A. J., Baeyens, J. P., Bauer, J. M., Boirie, Y., Cederholm, T., Landi, F., Martin, F. C., Michel, J.-P., Rolland, Y., Schneider, S. M., et al. (2010) · Age and Ageing · doi:10.1093/ageing/afq034 · Cited by 11,601
Temporal Convolutional Networks for Action Segmentation and Detection
Lea, C., Flynn, M. D., Vidal, R., Reiter, A., & Hager, G. D. (2017) · 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2017.113 · Cited by 2,238 · cited 1×
“As an evaluation metric, segmental F1-scores based on Intersection over Union (IoU) were computed for each class based on Lea et al. [27] and Sarapata et al. In case 7, if a true segment is predicted as some smaller segments, FP numbers increase.”
MethodSensor-based and vision-based human activity recognition: A comprehensive survey
Minh Dang, L., Min, K., Wang, H., Jalil Piran, M., Hee Lee, C., & Moon, H. (2020) · Pattern Recognition · doi:10.1016/j.patcog.2020.107561 · Cited by 730 · cited 1×
“To date, much of the research up to now has annotated the daily activities and repetitive exercises as a whole [1], [8].”
BackgroundMS-TCN: Multi-Stage Temporal Convolutional Network for Action Segmentation
Farha, Y. A., & Gall, J. (2019) · 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2019.00369 · Cited by 673 · cited 5×
“Recently, multi-stage temporal convolutional networks (MS-TCNs) were applied for action recognition by videos [13].”
BackgroundMethodInfluentialBianchi, V., Bassoli, M., Lombardo, G., Fornacciari, P., Mordonini, M., & De Munari, I. (2019) · IEEE Internet of Things Journal · doi:10.1109/jiot.2019.2920283 · Cited by 466 · cited 1×
“H U man activity recognition (HAR) has emerged as a significant research area in recent years due to its wide range of applications in various domains, including healthcare [1], sports monitoring [2], and smart environments [3].”
BackgroundA Wavenet for Speech Denoising
Rethage, D., Pons, J., & Serra, X. (2018) · 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) · doi:10.1109/icassp.2018.8462417 · Cited by 457
A CNN-LSTM Approach to Human Activity Recognition
Mutegeki, R., & Han, D. S. (2020) · 2020 International Conference on Artificial Intelligence in Information and Communication (ICAIIC) · doi:10.1109/icaiic48513.2020.9065078 · Cited by 452 · cited 1×
“Recently, researchers have applied the combination of CNN and Long short-term memory (LSTM) cells (CNN-LSTM) [21], [22], which outperformed single CNN or LSTM network on many datasets.”
BackgroundA survey on wearable sensor modality centred human activity recognition in health care
Wang, Y., Cang, S., & Yu, H. (2019) · Expert Systems with Applications · doi:10.1016/j.eswa.2019.04.057 · Cited by 409 · cited 4×
“Most existing studies for HAR systems annotate the full range of each activity (macro labels), i.e. from the start of the first movement to the end of the last movement [1], [7].”
BackgroundInfluentialDeep Residual Bidir-LSTM for Human Activity Recognition Using Wearable Sensors
Zhao, Y., Yang, R., Chevalier, G., Xu, X., & Zhang, Z. (2018) · Mathematical Problems in Engineering · doi:10.1155/2018/7316954 · Cited by 348 · cited 1×
“By leveraging architectures such as recurrent neural networks (RNNs) [16] and convolutional neural networks (CNNs) [17]–[20], deep learning models can effectively learn hierarchical representations, automatically extract discriminative features, and model complex temporal dependencies in activity…”
BackgroundThomas, S., Mackintosh, S., & Halbert, J. (2010) · Age and Ageing · doi:10.1093/ageing/afq102 · Cited by 335 · cited 1×
“The Otago Exercise Program (OEP) comprises a set of activities designed to enhance balance, strength, and walking abilities, thereby mitigating the risk of falls in older adults [4].”
Background
Cited by
DFF-Net: Dynamic feature fusion network for time series prediction
Xiao, B., Chen, Z., Wu, Y., Wang, M., Hu, S., & Zhang, X. (2025) · International Journal of Approximate Reasoning · doi:10.1016/j.ijar.2025.109436 · Cited by 5
AI-assisted Automatic Jump Detection and Height Estimation in Volleyball Using a Waist-worn IMU
Xu, W., Wang, C., Shang, M., Bleecker, C. D., Vega, M. T., Vanrenterghem, J., & Vanrumste, B. (2025) · Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) · doi:10.1109/embc58623.2025.11251662 · Cited by 6
“Ten volleyball players were recruited for the study.”
Self-citationModeling epithelial wound closure dynamics with AI: A comparative study across cell types
(2025) · doi:10.1016/j.reth.2025.10.002 · Cited by 2
“However, they underperformed with the more stable healing patterns of MCF10A, likely due to overfitting or excessive smoothing from their sliding window mechanism when applied to gradually evolving processes [35].”
Audio-Driven Multi-Modal Unobtrusive Health Monitoring and Inference for Smart Eldercare at Home
Fan, X., Li, Y., Huang, Z., & He, Z. (2026) · IEEE Journal of Biomedical and Health Informatics · doi:10.1109/jbhi.2026.3651261 · Cited by 1
“Wearablesensors havebeenwidelyusedtomonitorthebehaviorofelderlyindi-viduals[27],[28],[29]andcanalsocapturephysiologicalsignals usefulfordiseasedetection[30].”
Metadata
- Authors
- Shang, M., Dedeyne, L., Dupont, J., Vercauteren, L., Amini, N., Lapauw, L., Gielen, E., Verschueren, S., Varon, C., De Raedt, W., & Vanrumste, B.
- Affiliations
- KU Leuven (BE)
- Universitair Ziekenhuis Leuven (BE)
- Imec, Leuven, Belgium
- ORCID
- Venue
- IEEE Journal of Biomedical and Health Informatics
- Published
- 2024-12
- Volume
- 28
- Issue
- 12
- Pages
- 7138-7150
- ISSN
- 2168-2194 (ISSN-L)
- 2168-2208
- OpenAlex
- W4402301684
- Cited by
- 5
- References
- 35
- Citations by year
- 2026: 2 · 2025: 3
- FWCI
- 0.70
- Type
- journal article
- Language
- en
- Open access
- Licence
- Locations
- IEEE Journal of Biomedical and Health Informatics (publishedVersion)
- PubMed (publishedVersion)
- Lirias (KU Leuven) (submittedVersion, other-oa)
- Funders
- China Scholarship Council (202007650007)
- Province Flemish Brabant Smart Region (3E230623)
- ENHANce (S60763)
- Fonds Wetenschappelijk Onderzoek (G099721N)
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Context-Aware Activity Recognition Systems
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Human Pose and Action Recognition
- Health Sciences › Medicine › Public Health, Environmental and Occupational Health › Physical Activity and Health
- Keywords
- Otago exercise program, exercise recognition, older adults, human activity recognition, inertial measurement unit, temporal convolutional network, community-dwelling older adults, rehabilitation, wearable sensors
- MeSH terms
- Wearable Electronic Devices, Deep Learning, Aged, Aged, 80 and over, Exercise Therapy, Female, Humans, Male, Middle Aged, Signal Processing, Computer-Assisted, Exercise, Neural Networks, Computer
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: DS-MS-TCN: Otago Exercises Recognition With a Dual-Scale Multi-Stage Temporal Convolutional Network. https://app.cochat.ai/papers/10.1109/jbhi.2024.3455426
- The paper
- Shang, M., Dedeyne, L., Dupont, J., Vercauteren, L., Amini, N., Lapauw, L., Gielen, E., Verschueren, S., Varon, C., De Raedt, W., & Vanrumste, B. (2024). DS-MS-TCN: Otago Exercises Recognition With a Dual-Scale Multi-Stage Temporal Convolutional Network. IEEE Journal of Biomedical and Health Informatics. https://doi.org/10.1109/jbhi.2024.3455426
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.