ORION: A Hierarchical Surgical World Model for Real-Time, Multi-Agent Operating-Room Intelligence
Journal of Ubiquitous Computing and Communication Technologies · 12.08.2026 · 8 (3) · 276-298
DOI: 10.36548/jucct.2026.3.008 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Ubiquitous Computing and Communication Technologies, online erschienen am 12.08.2026. Verfasst von Dev Mehta und 1 weiteren Autor:innen. Die Arbeit nennt 25 Literaturangaben.
Abstract (Originalsprache)
Modern operating suites generate dense, heterogeneous data streams endoscopic video, multi-parameter vitals, anesthesia registers, device telemetry, intraoperative imaging, and robotic kinematics yet the overwhelming majority of these signals are discarded, with the complete clinical reality of a procedure routinely collapsed into a single postoperative note. This documentation bottleneck loses roughly ninety-nine percent of the granular information produced during surgery, preventing systematic audit, objective skill assessment, and longitudinal research. We present ORION (Operating Room Intelligence and Optimization Network), an AI-native platform that transforms multi-sensory input into a continuously updated digital twin of the patient and operating environment. ORION is organised as a twelve-layer Hierarchical Surgical World Model (HSWM) spanning real-time perception, relational scene-graph construction, a Gaussian-splatting digital twin, temporal workflow modelling, predictive risk forecasting, knowledge grounding, and multi-agent clinical reasoning. A generative component, SurgWorld, mitigates kinematic-data scarcity by synthesising physically plausible surgical video and recovering pseudo-kinematics through an inverse-dynamics model. Because ORION is presented as a proposed architectural framework, the reported Cholec80 and AutoLaparo figures up to 94.6% and 89.5% accuracy at 238.7 FPS for the underlying state-space model are drawn from the constituent methods' published evaluations, not a new experiment; prospective validation of the integrated system is future work. The reporting layer is designed to compress postoperative documentation to under thirty seconds. We further describe the low-latency execution architecture (NVIDIA Holoscan and Triton) required to stay below the 50 ms perceptual threshold, and map the platform onto India's CDSCO Software-as-a-Medical-Device regulatory pathway
Zitieren
APA
Mehta, D., & R.T., C. (2026). ORION: A Hierarchical Surgical World Model for Real-Time, Multi-Agent Operating-Room Intelligence. Journal of Ubiquitous Computing and Communication Technologies, 8(3), 276-298. https://doi.org/10.36548/jucct.2026.3.008
Vancouver
Mehta D, R.T. C. ORION: A Hierarchical Surgical World Model for Real-Time, Multi-Agent Operating-Room Intelligence. Journal of Ubiquitous Computing and Communication Technologies. 2026;8(3):276-298. doi:10.36548/jucct.2026.3.008
BibTeX
@article{Mehta2026,
author = {Mehta, Dev and R.T., Charulatha},
title = {ORION: A Hierarchical Surgical World Model for Real-Time, Multi-Agent Operating-Room Intelligence},
journal = {Journal of Ubiquitous Computing and Communication Technologies},
year = {2026},
volume = {8},
number = {3},
pages = {276--298},
doi = {10.36548/jucct.2026.3.008}
}
EUVIMED