Applications

Capture and edge AI for operating rooms.

YUAN provides capture and edge-computing systems for remote 3D surgery and AI-assisted ultrasound. This work includes projects with Foxconn, National Taiwan University Hospital, and NVIDIA's medical-AI ecosystem.

Robotic surgical arms positioned above an operating table

<50 ms: YUAN-reported glass-to-glass latency in its tested 3D pipeline
15 µs: YUAN-reported HDMI/SDI DMA latency in the tested configuration
87%: YUAN-reported accuracy on the tested ultrasound data
50: iGood Liver AI ultrasound systems deployed across rural Taiwan

Remote surgery

A tested 3D pipeline measured under 50 milliseconds.

Remote robotic surgery requires a validated latency budget across the camera, capture card, encoder, network, decoder, display, control system, and robot. One YUAN project used 70 milliseconds as its glass-to-glass target; that target is not a general clinical threshold.

For its tested NVIDIA Jetson Orin, HDMI/SDI DMA, AV1, and SkyLink X configuration, YUAN reported capture latency as low as 15 microseconds and glass-to-glass latency below 50 milliseconds. Results depend on the complete equipment and network path.

The platform provides up to 275 TOPS for image-depth calculation and medical-image analysis. Any surgical use requires clinical and system validation, regulatory approval, network redundancy, cybersecurity controls, and operating procedures beyond the compute platform.

Robotic surgical system arranged around an operating table

Diagnostics

AI-assisted ultrasound screening at the edge.

YUAN's ultrasound-assistance system combines a VPP6N0-S NX, a 4K60 HDMI 2.0 M.2 capture card, and 70 or 100 TOPS of compute. YUAN reports that the model was trained on more than 6,000 expert-annotated images and measured up to 87 percent accuracy when distinguishing normal from abnormal tissue in the tested data.

Developed with National Taiwan University Hospital and the Good Liver Foundation, the NVIDIA Jetson-based iGood Liver AI program deployed 50 systems to rural and underserved clinics in Taiwan. The model flags suspicious liver lesions for clinician review.

Reported performance does not establish results for every patient, probe, operator, or clinical setting. The system assists trained clinicians and requires institutional validation, regulatory review, and operating procedures.

Clinician holding an ultrasound probe beside a portable scanner

With Foxconn

CoDoctor AI, from colonoscopy to the smart hospital.

CoDoctor AI Endovia, developed with Foxconn, is intended to assist polyp and lesion detection during colonoscopy. It runs on YUAN's NVIDIA IGX Thor-based NEAR T7X platform and was evaluated with endoscopy data at Kaohsiung Medical University Chung-Ho Memorial Hospital.

The system includes bowel-preparation scoring, withdrawal-time monitoring, and structured report generation with NVIDIA Nemotron. These functions support clinician review; they do not replace clinical judgment or establish results for every procedure.

The collaboration reaches further into the operating room: a scrub nurse collaborative robot, jointly developed by Foxconn, Kawasaki Heavy Industries, Taichung Veterans General Hospital, and YUAN, uses the NVIDIA Isaac for Healthcare Rheo Blueprint and Isaac GR00T VLA architecture for multimodal perception and surgical scene understanding.

Endoscopy equipment and monitors beside a patient bed

The architecture

Synchronize OR video, device, and patient data.

YUAN's Smart Surgical Video Platform brings together 4K endoscopes, PACS, ultrasound, and remote medical devices. It supports 4K30 2D and 3D processing, Single View, PIP, and Multi-View output to multiple OR monitors. The platform can record to USB, NAS, or FTP and stream over RTSP, RTMP, or WebRTC for cross-hospital consultation.

NVIDIA Holoscan Sensor Bridge synchronizes HD and 4K endoscopic video, physiological signals such as heart rate and SpO2, surgical-device telemetry, and external camera feeds in one pipeline. IGX Thor provides security and functional-safety capabilities at the compute layer; the complete clinical system still requires validation and regulatory review.