GTC 2025 GE Healthcare Partnership
Category: Healthcare Special Address · Year: 2025 · ▶ Watch
Speakers: Parminder Bhatia - Chief AI Officer / GE HealthCare · Roland Rott - President & CEO, Imaging / GE HealthCare
Segments (9)
- 00:00 · Introduction
- Introduction of the speakers and the session’s focus on AI in healthcare.
- 01:15 · Challenges in Healthcare
- Discussion on global healthcare challenges including aging populations, data overload, and staff burnout.
- 08:11 · GE HealthCare’s D3 Strategy
- Introduction of the D3 strategy focusing on Smart Devices, Disease States, and Digital Health.
- 12:31 · Smart Devices and AI in Imaging
- Examples of AI accelerating MRI scans, prioritizing critical X-rays, and guiding ultrasound.
- 20:45 · CareIntellect for Oncology and Tumor Boards
- Using GenAI to synthesize multimodal patient data to streamline oncology care and tumor boards.
- 25:35 · The Future: Agentic AI and Project Health Companion
- Vision for multi-agent AI systems that proactively reason and recommend treatment plans.
- 30:15 · NVIDIA Partnership and Technology Stack
- Overview of the 17-year partnership with NVIDIA spanning edge devices to cloud data centers.
- 33:44 · Foundation Models and SonoSAM
- Development of medical imaging foundation models for X-ray, MRI, and ultrasound.
- 38:36 · Autonomous X-ray and Conclusion
- Vision for autonomous medical devices to automate exam setup and alleviate staff shortages.
Product Announcements (9)
- [09:25] AIR Recon DL
- Deep learning-based image reconstruction for MRI
- specs: Reduces scan time by 50% while improving image quality
- availability: Launched in 2020
- [11:25] Command Center
- AI-enabled hospital operations software
- specs: Reduces length of stay and improves capacity utilization
- availability: Available
- [13:40] Sonic DL
- Deep learning technology for cardiac MRI
- specs: Reduces scan times from minutes to seconds (12x faster)
- availability: Available
- [16:18] Critical Care Suite
- On-device AI for mobile X-ray systems
- specs: Automatically detects and prioritizes critical conditions like Pneumothorax
- availability: Available
- [18:48] Vscan Air SL with Caption AI
- Handheld ultrasound with AI guidance
- specs: Provides real-time guidance and quality metrics for novice users
- availability: Available
- [24:37] CareIntellect for Oncology
- Cloud-based oncology care assistant
- specs: Aggregates multimodal data and uses GenAI to summarize patient history
- availability: In development/Available
- [26:39] Project Health Companion
- Multi-agent AI system for clinical decision support
- specs: Uses specialized AI agents coordinated by a supervisory agent to propose treatment plans
- availability: Concept/Research
- [33:44] SonoSAM
- Ultrasound foundation model
- specs: Tracks tumors and anatomy in real-time with minimal user input
- availability: In development
- [38:36] Cosmos (Autonomous X-ray)
- Vision for autonomous medical imaging devices
- specs: Robots handle patient positioning and scanner alignment automatically
- availability: Concept/In development
Specific Numbers (16)
| Timestamp | Metric | Value | Context |
|---|---|---|---|
| 02:10 | Cancer diagnosis rate | 1 in 3 | People globally who will be diagnosed with cancer. |
| 02:20 | Lack of healthcare access | 4.5 Billion | People globally without access to proper care. |
| 04:25 | Data generation | 50 petabytes | Amount of data generated per hospital per year. |
| 04:28 | Unused data | 97% | Percentage of healthcare data that goes unused. |
| 05:25 | Physician time allocation | 60% | Percentage of doctors who spend more time with EMRs than patients. |
| 06:29 | Staff turnover | 100% | Staff turnover rate in major healthcare systems every five years. |
| 09:38 | Scan time reduction | 50% | Reduction in MRI scan time using AIR Recon DL. |
| 11:17 | Clinical trial enrollment | 7% | Percentage of eligible patients actually enrolled in clinical trials. |
| 11:48 | Capacity increase | 2,000 | More beds annually served at Deaconess Health using Command Center. |
| 11:58 | Virtual bed capacity | 35 | New bed capacity created at Humber River without adding physical beds. |
| 14:13 | Scan speed improvement | 12x | Speed increase for cardiac MRI using Sonic DL. |
| 14:24 | Exam time reduction | 83% | Overall reduction in cardiac MRI exam time using Sonic DL. |
| 35:01 | Data size | 5 to 10 gigabytes | Size of a single 3D MR scan. |
| 38:35 | Global X-ray volume | 2.6 Billion | Number of X-ray exams performed per year globally. |
| 38:35 | Image rejection rate | 25% | Percentage of X-ray images rejected due to poor positioning. |
| 38:35 | Staffing shortages | 80% | Percentage of healthcare organizations that are shorthanded. |
Benchmark Claims (2)
- [14:13] Sonic DL Scan Speed: 12x faster
- vs: Conventional cardiac MRI
- gain: Reduces scan times from minutes to seconds, eliminating the need for long breath holds.
- [36:56] SonoSAM Segmentation Accuracy: 90%+ accuracy
- vs: Previous models
- gain: Maintains high accuracy even on unseen anatomies like breast lesions.
Customer Stories (3)
- [11:38] Deaconess Health
- Implemented GE HealthCare’s Command Center to manage hospital operations.
- outcome: Improved capacity utilization, resulting in 2,000 more beds annually served.
- [11:52] Humber River
- Used Command Center predictive analytics to reduce length of stay.
- outcome: Created the capacity equivalent of 35 new beds without adding physical infrastructure.
- [17:45] Dr. Keigo Yasukawa
- Used the Vscan Air handheld ultrasound while traveling by jet ski to remote islands in Japan.
- outcome: Expanded access to critical diagnostic care for elderly patients in remote locations.
Key Technologies (5)
- Deep Learning Reconstruction: Improves MRI image quality while significantly reducing the time required to acquire the scan.
- On-device AI: Runs AI algorithms directly on edge devices like mobile X-rays for immediate triage without cloud dependency.
- Generative AI: Summarizes and synthesizes unstructured, multimodal patient data into actionable clinical insights.
- Agentic AI: Utilizes multiple specialized AI agents that reason, coordinate, and propose comprehensive treatment plans.
- Foundation Models: Large-scale models trained on vast amounts of multimodal medical data to enable zero-shot or few-shot downstream tasks.
Demos Shown (7)
- [13:40] Comparison of Sonic DL vs Conventional MRI scan speed.
- True
- [16:55] Critical Care Suite interface showing Pneumothorax detection and confidence scoring.
- True
- [18:48] Vscan Air SL with Caption AI providing real-time guidance for capturing an ultrasound image.
- True
- [24:40] CareIntellect for Oncology interface showing a patient timeline and GenAI-generated summaries.
- True
- [27:35] Project Health Companion concept showing multiple AI agents interacting to answer a clinical query.
- True
- [36:10] SonoSAM tracking a lesion in an ultrasound video with a single click.
- True
- [39:40] Cosmos Autonomous X-ray concept animation showing automated patient positioning.
- True
Predictions / Commitments (3)
- [26:18, Future] Agentic AI will transition healthcare from reactive systems to proactive systems that suggest outputs and explain their logic.
- [33:38, Future] Adopting NVIDIA DGX systems will position GE HealthCare at the forefront of AI innovation for building foundation models.
- [39:38, Future] Autonomous medical devices will automate exam setup and patient positioning to address severe staffing shortages.
Companies Mentioned (3)
Bill & Melinda Gates Foundation · NVIDIA · AWS
Notable Quotes (3)
97% of that data is actually just generated, it’s stored, it’s never really retrieved. — Roland Rott @ 04:56
This marks a significant difference from being a reactive system to a proactive system where the system is able to not just assist but being able to provide inputs, suggest outputs, as well as explain its logic. — Parminder Bhatia @ 26:18
Creating a world where healthcare has no limits. — Roland Rott @ 40:50
Key Topics
AI in Healthcare · Medical Imaging · Deep Learning Reconstruction · Agentic AI · Foundation Models · Multimodal Data Synthesis · Oncology Care · Autonomous Medical Devices · Healthcare Burnout · Access to Care · Edge Computing · Generative AI
Takeaways
- Healthcare faces massive global challenges, including aging populations, severe staff burnout, and 97% of generated data going unused.
- GE HealthCare’s D3 strategy focuses on Smart Devices, Disease States, and Digital Health & AI to address these systemic challenges.
- AI embedded directly into devices (like Sonic DL and AIR Recon DL) drastically reduces scan times and improves image quality, enhancing patient comfort and throughput.
- CareIntellect for Oncology uses Generative AI to synthesize fragmented, multimodal patient data, saving clinicians hours of manual review.
- The future of healthcare AI involves Agentic AI, where multiple specialized agents collaborate to reason and propose personalized treatment plans.
- Foundation models for medical imaging (like SonoSAM) are being developed to enable zero-shot or few-shot segmentation and tracking across various anatomies.
- GE HealthCare is partnering closely with NVIDIA to leverage their full stack, from edge RTX GPUs to DGX data center systems, to power these innovations.
- Autonomous medical devices, such as the Cosmos X-ray concept, aim to automate patient positioning and exam setup to alleviate critical staff shortages.