EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

A single-camera video-based assessment of locomotive syndrome using pose-silhouette fusion model

Yu Moriguchi iD, Junichi Kushioka iD, Noriko Takemura, Hiroki Hagizawa, Masafumi Kashii, Ichiro Nakahara, Taichi Kimura, Keisuke Akiyama, Taku Fujimoto, Hajime Nagahara, Keiko Yamada, Rodrigo Navarro-Ramirez und 5 weitere

PLOS Digital Health · 26.06.2026 · 5 (6) · e0001530

DOI: 10.1371/journal.pdig.0001530 ↗

Kurz gefasst

Fachartikel in der Zeitschrift PLOS Digital Health, online erschienen am 26.06.2026. Verfasst von Yu Moriguchi und 16 weiteren Autor:innen. Beteiligte Einrichtung: Department of Orthopaedic Surgery, National Hospital Organization, Osaka Minami Medical Center, Kawachinagano, Osaka, Japan. Die Arbeit nennt 23 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Japan Science and Technology Agency, Nakatomi Foundation.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Locomotive syndrome (LS), characterized by declining mobility due to musculoskeletal disorders, significantly affects older adults and is highly prevalent in Japan. Early detection and intervention are crucial to mitigate disease progression and preserve quality of life. Current diagnostic methods rely on subjective self-reports and labor-intensive clinical assessments, underscoring the need for automated, objective, and scalable screening tools. Building on our previous work, this study presents an updated deep learning-based computer vision model for LS screening using single-camera walking videos. We collected 511 walking videos from 178 participants clinically classified into four LS stages (non-LS, stage 1, stage 2, and stage 3). We used the MMPose framework to extract 2D body keypoints and generated silhouette-based Gait Energy Images (GEIs) from seven frames automatically sampled from fixed 40-frame walking windows to capture complementary gait-shape information. Model performance was evaluated using 10-fold cross-validation based on held-out evaluation-window predictions, while ensuring that windows from the same participant and the same walking video were not split across training, validation, and test sets. We also evaluated an independent external validation cohort of 33 cases (LS0–1, n = 15; LS2–3, n = 18). In internal validation, the pose-estimation model achieved sensitivity of 0.9440 and an AUC of 0.9668, whereas the silhouette model showed specificity of 0.9085 and an AUC of 0.9191. Fusion models outperformed the individual approaches, with the highest F1-score (0.908) at a 60:40 pose-to-silhouette ratio and the highest AUC (0.971) at a 70:30 ratio. In external validation, the score-fusion model achieved 90.9% accuracy, 88.9% sensitivity, 93.3% specificity, a macro-F1 score of 0.909, and an AUC of 0.967. In conclusion, integrating pose-estimation and silhouette-based features substantially improves the accuracy and clinical utility of automated LS screening. This low-burden approach provides a practical bridge between population-level screening and definitive clinical assessment.

Autor:innen und Institutionen

  • Department of Orthopaedic Surgery, National Hospital Organization, Osaka Minami Medical Center, Kawachinagano, Osaka, Japan.
  • Department of Orthopaedic Surgery, Osaka University Graduate School of Medicine, Suita, Osaka, Japan.
  • Department of Medical Innovation, Osaka University Graduate School of Medicine, Suita, Osaka, Japan.
  • ayumo, Inc., Osaka, Osaka, Japan.
  • Osaka University Institute for Sports and Global Health, Osaka University, Suita, Osaka, Japan.
  • Department of Clinical Research, National Hospital Organization Osaka Minami Medical Center, Kawachinagano, Osaka, Japan.

ORCID-Profile: Yu Moriguchi, Junichi Kushioka

Förderung

  • Japan Science and Technology Agency
  • Nakatomi Foundation

Zitieren

APA

Moriguchi, Y., Kushioka, J., Takemura, N., Hagizawa, H., Kashii, M., Nakahara, I., Kimura, T., Akiyama, K., Fujimoto, T., Nagahara, H., Yamada, K., Navarro-Ramirez, R., Oda, T., Mochizuki, H., Nakata, K., Okada, S., & Tada, S. (2026). A single-camera video-based assessment of locomotive syndrome using pose-silhouette fusion model. PLOS Digital Health, 5(6), e0001530. https://doi.org/10.1371/journal.pdig.0001530

Vancouver

Moriguchi Y, Kushioka J, Takemura N, Hagizawa H, Kashii M, Nakahara I, et al. A single-camera video-based assessment of locomotive syndrome using pose-silhouette fusion model. PLOS Digital Health. 2026;5(6):e0001530. doi:10.1371/journal.pdig.0001530

BibTeX
@article{Moriguchi2026,
  author = {Moriguchi, Yu and Kushioka, Junichi and Takemura, Noriko and Hagizawa, Hiroki and Kashii, Masafumi and Nakahara, Ichiro and Kimura, Taichi and Akiyama, Keisuke and Fujimoto, Taku and Nagahara, Hajime and Yamada, Keiko and Navarro-Ramirez, Rodrigo and Oda, Takenori and Mochizuki, Hideki and Nakata, Ken and Okada, Seiji and Tada, Satoru},
  title = {A single-camera video-based assessment of locomotive syndrome using pose-silhouette fusion model},
  journal = {PLOS Digital Health},
  year = {2026},
  volume = {5},
  number = {6},
  pages = {e0001530},
  doi = {10.1371/journal.pdig.0001530}
}
Kostenlos starten →