NVIDIA 展示 AI 如何加速医疗护理与患者匹配
AI Is Helping Care Teams Get Patients the Right Care, Sooner
Proximie 利用 NVIDIA Cosmos 3 优化手术室物流,使病例处理量提升 8% 且周转速度加快 45%。ConcertAI 基于 NVIDIA AI 构建的引擎将癌症患者临床试验匹配速度提升至人工方法的 3 倍,并支持高达 50% 的试验入组率。这些应用旨在减少医护人员在常规任务上的时间,使其能更专注于患者护理。
From pediatric research to smart hospitals, AI is helping care teams spend less time on routine tasks and more time with patients.
Health systems and startups are putting it to work across healthcare, from matching patients to clinical trials to supporting radiologists, hospital teams and patients at home. Clinicians stay at the center, reviewing AI outputs and guiding decisions, while NVIDIA technologies provide the compute behind the applications.
Read on to learn how healthcare teams are using AI to improve and speed patient care.
Tuesday, Nov. 17, 8:00 a.m. PT 
Proximie Helps Surgical Teams Spend Less Time on Logistics and More Time With Patients
As AI advances healthcare, surgery is a high-stakes, highly regulated setting for its use — and one with significant opportunity for efficiency gains.
Only about 10% of time spent in the operating room is clinical. The rest is devoted to scheduling, turnover and documentation.
With the help of NVIDIA technology, Proximie is introducing AI into operating room (OR) logistics so that clinicians can focus on taking care of patients.
A New Vantage Point
Proximie is using NVIDIA Cosmos 3, the frontier open world foundation model family for physical AI, to power a platform used in more than 700 ORs today.
Trained on a decade of data, it captures real-time audio, video and device signals from a close-up view of the surgery as well as a wide view of the whole room. The comprehensive view offers context held by no single person in the room, whether the surgeon, scrub nurse or coordinator.
Agentic notifications and voice AI tools can call the next patient when a room is ready, document alerts and capture practitioner notes. To protect privacy, Proximie’s models blur identifying details, such as faces, tattoos and onscreen text before footage leaves the OR. Surgeons control the blurring to ensure that recordings used for training still show what trainees need to see.
Across health systems where Proximie’s technology has been applied, the platform has enabled clinical teams to handle 8% more cases, turn over the OR 45% faster and start procedures on time 26% more often.
Next, the company hopes to turn its technology into an OR copilot that assists with staging equipment ahead of time, estimating time remaining and running implant safety checks. Proximie will also use NVIDIA Cosmos 3 to generate realistic synthetic scenes grounded in real ORs, so models can learn from the rarest, most critical scenarios.
The goal is to create a room that runs itself, so practitioners can spend more time with patients and dedicate less time to documentation and preparation.
Proximie is working with UC San Diego Health to build an AI model that learns from many kinds of surgical data at once using NVIDIA open models.
Before being trained on Proximie’s data, Cosmos 3 could already identify 80% of key equipment in the OR with high accuracy. Training it further on Proximie’s footage made the model more accurate and smaller, so several models can run at once on a single computer in the OR alongside tools to protect patient privacy.
Proximie is also among the first participants in Open-H-Hospital, an NVIDIA-led, community-driven open initiative that brings real hospital operation data together with simulation data to train the next generation of hospital AI and robotics.
Learn more about Proximie.
Tuesday, Nov. 17, 8:00 a.m. PT 
ConcertAI Matches Cancer Patients to Clinical Trials Faster
Interest in clinical trials is high among people with cancer. More than half of those offered a trial say yes, but only about 7% of U.S. patients with cancer take part in one.
The issue isn’t the availability of trials, but the time it takes to match patients to them. Care teams have to manually review months or years of medical records and pull together information like physician notes, pathology reports and genomic results from different systems.
Built on NVIDIA AI, ConcertAI’s Cara AI engine and TriaLinQ module are speeding up that process, identifying eligible patients roughly 3x faster than manual methods. They assess trial eligibility across the CancerLinQ oncology network, ConcertAI’s data network that works across nearly 1,000 sites of care and over 13 million patient records.
From Scattered Records to One View
Cara pulls all relevant patient data into a single view for each person and updates it weekly with disease stage, biomarker status, treatment response and line of therapy. TriaLinQ, a module of the CancerLinQ suite, uses that information to assess a patient’s eligibility for a trial.
“We use a chain of AI models and agents to parse all the information, understand really challenging concepts like progression or timelines within a diagnosis journey, and summarize it into a structured data model that can be queried from our software stack,” said Eron Kelly, CEO of ConcertAI.
ConcertAI uses healthcare-specific NVIDIA microservices, software libraries and AI agent tools to accelerate its work.
Clinicians Stay in Charge
ConcertAI brings new technology to care teams, but treatment decisions stay with clinicians.
“The promise of technology in oncology isn’t measured by how sophisticated the platform is, but instead by whether it meaningfully supports the clinicians delivering care and positively impacts patient outcomes,” said Dr. Shaalan Beg, chief medical officer of oncology at ConcertAI. “That’s the standard we hold ourselves to: AI that earns its place at the bedside.”
Every eligibility finding links back to its source in the patient’s record, so oncologists and care teams can make fully informed treatment decisions.
By integrating into the way care teams and clinicians already work, ConcertAI’s systems support up to 50% higher trial enrollment, helping more patients find treatment options when they need them.
Learn more about ConcertAI.
Artisight Helps Nurses Prevent Patient Falls With In-Room AI Device
Roughly 700,000 to 1 million patients fall in U.S. hospitals each year. A fall with injury costs over $14,000 on average and can add days to a patient’s hospital stay.
Artisight’s All-in-One device alerts nurses when a patient at high risk of falling starts to get out of bed. The early warning gives nurses time to reach the patient before they stand — meaning fewer injuries, shorter hospital stays and lower bills for patients, while saving nurses about five hours per unit per day to focus on other care tasks.
The device uses audio and video to capture data about a patient’s movement that can be used to recognize when a patient moves towards the edge of the bed. When the patient starts to get out of bed, it alerts staff and automatically timestamps and charts the alert.
Each device runs an NVIDIA Jetson Orin NX, a compact edge computer, to process audio, video and movement in the room. Several AI models run at the same time to recognize the movements and to send alerts quickly. AI model performance is optimized with NVIDIA CUDA and NVIDIA TensorRT, with NVIDIA Dynamo-Triton managing how the models run.
That data stays on the device and in the hospital, which keeps alerts quick and protects patient privacy.
Learn more about Artisight.
Sword Brings Dependable AI Guidance to Nearly a Million Members

Many patients have faced the challenge of deciding whether a new symptom warrants a clinician visit or can be safely managed with the right adjustments at home. Sword, a leading AI health company, built its Phoenix AI health partner to help make that call for nearly a million people and over 2,500 organizations.
After Sword trained an NVIDIA Nemotron open model to its clinical standards for musculoskeletal (MSK) care, Phoenix’s performance improved significantly — making the right call in 10 test runs nearly 80% of the time, up from under 50%. Sword estimates that using an open model costs 5-6x less than frontier models, saving hospitals and patients valuable resources.
“Clinicians don’t trust an AI teammate because it’s right once,” said Megan Hill, head of clinical specialists at Sword. “We trust it because it makes the same call on Tuesday at 1 a.m. that it made on Saturday at 7 p.m.”
An AI Teammate That Knows Its Role
Phoenix works alongside each person’s dedicated clinician. For every message, it chooses to answer or bring the clinician in — and decides whether to speak or stay quiet to respect the clinician and patient’s conversation.

Using an NVIDIA HGX B300 system, Sword trained Phoenix with reinforcement learning — a technique where the model practices, gets graded and repeats what earns high marks — on de-identified care conversations.
An AI judge scores every conversation on seven criteria defined by physical therapists, nurses and dietitians on Sword’s team, including tone, clinical boundaries and respect for the clinician’s role. If Phoenix crosses a clinical boundary, the safety score automatically drops to the minimum.
Consistent, Repeatable Results
To test the model, Sword runs Phoenix on situations where the right call is predetermined, running each situation 10 times. With NVIDIA Nemotron, first-try accuracy for MSK scenarios rose from 78% to 93%, while in women’s health, the model’s accuracy rose from 79% to 87% correct on the first try and from 50% to 72% correct on all 10 runs.
Sword can retrain Phoenix whenever it identifies a problem, and member data is kept private.
This work builds on Dawn 1, Sword’s previous model focused on mental health. By pairing open models with rigorous clinical oversight, Sword is proving that AI can expand access to care, extend clinician time for the patients who need it most, and support safe, reliable care at global scale.
Learn more about Sword.
Dyania Health Uses AI to Turn Decades of Patient Records Into Better Care
Hospitals hold decades of patient history, most of it written in free text, as physicians’ notes, operative reports, imaging and pathology reads and discharge summaries. To understand one patient’s full story, a clinician would need to read through years of charts.
Dyania Health, a New Jersey-based clinical AI company, built Synapsis AI to do just that. Synapsis AI is a clinical reasoning engine: It works across every note, report and lab result in a patient’s record, connecting findings that may be years apart to answer clinical questions the way a specialist reviewing the chart would.
The output is a set of clinical signals, each traceable to its source in the record, that tells care teams which patients need attention and why. Clinicians review the evidence and make every decision.
Keeping Patient Data Inside the Hospital Walls
Synapsis AI runs on NVIDIA-accelerated computing inside a health system’s own firewall. The entire health system’s patient records can be processed onsite, so patient data is never shared and never leaves the hospital.
Processing records this way takes chart review off the desks of clinical teams and lets them spend that time with patients. But Synapsis AI still requires clinician input — it’s an assistive tool that surfaces evidence from medical records that a clinician must still confirm and act on.
Dyania builds and trains its clinical models with NVIDIA NeMo open libraries, then uses NVIDIA TensorRT and TensorRT-LLM to optimize them to run faster and more efficiently on the hospital’s own GPU servers.
Synapsis AI in the Real World
Cleveland Clinic used Synapsis AI to support a study of more than 7,800 aortic valve replacement patients between 2005 and 2021.
Synapsis AI reasoned across each patient’s record, including every echocardiogram added in the years after surgery, and used this information to build a predictive model of valve deterioration and reintervention risk at the three-, five- and ten-year mark after a patient undergoes an aortic valve replacement procedure. The AI platform completed chart review across the full cohort that would have taken clinicians years to complete by hand.
“Using technology to bring together and analyze information that can help us predict the duration of implanted heart valves is a promising step towards enhanced outcomes and more personalized care for patients,” said Dr. Lars Svensson, chief of the Heart, Vascular & Thoracic Institute at Cleveland Clinic.
The same approach can be applied to any healthcare challenge where the answer sits inside the medical record, such as identifying patients eligible for clinical trials, closing care gaps in cardiology and oncology, and populating clinical registries.
Learn more about Dyania Health.
来源:NVIDIA Blog · blogs.nvidia.com
