GTC 2023 Healthcare Special Address

Category: Healthcare Special Address · Year: 2023 · ▶ Watch

Switch language → 中文

Segments (9)

  • 00:00 · Introduction to Generative AI in Healthcare
    • Kimberly Powell introduces the impact of generative AI across the healthcare ecosystem, from imaging to drug discovery.
  • 02:30 · NVIDIA Clara and Ecosystem
    • Overview of the NVIDIA Clara computing platform and its broad partner ecosystem.
  • 04:05 · MONAI: The AI Factory for Imaging
    • Deep dive into MONAI, its 1 million downloads milestone, and new generative AI models for medical imaging.
  • 08:50 · NVIDIA Holoscan for Medical Devices
    • Introduction of Holoscan and IGX Orin for real-time, software-defined AI processing at the edge.
  • 12:12 · Medtronic Partnership
    • Interview with Medtronic CEO Geoff Martha on building an AI platform for medical devices using Holoscan.
  • 18:49 · Accelerating Genomics with Parabricks
    • Updates on NVIDIA Parabricks 4.1, accelerating whole genome sequencing and supporting new sequencing platforms.
  • 23:32 · Digital Biology and Drug Discovery
    • How generative AI is turning biology into engineering, and the introduction of the BioNeMo cloud service.
  • 27:38 · BioNeMo Service Demo
    • A demonstration of the BioNeMo web interface for protein structure prediction and molecule generation.
  • 33:30 · Tokyo-1 Supercomputer Announcement
    • Announcement of Japan’s first pharmaceutical supercomputer in partnership with Mitsui & Co.

Product Announcements (6)

  • [03:08] NVIDIA DGX Cloud for Healthcare
    • Cloud-based AI supercomputing access
    • specs: Available through Microsoft Azure, Google GCP, and Oracle OCI
    • availability: Available now
  • [11:11] Holoscan IGX Orin Developer Kits
    • All-in-one industrial computer for medical edge AI
    • specs: High-speed sensor input, ConnectX NIC, Orin SoC, RTX GPU
    • availability: Available for order now
  • [18:05] Medtronic GI Genius with Holoscan
    • AI-assisted colonoscopy system
    • specs: Adopts NVIDIA Holoscan and IGX for real-time AI enhanced imaging
    • availability: Shipping at the end of the year
  • [20:43] NVIDIA Parabricks 4.1
    • Genomic analysis software suite
    • specs: Whole genome sequencing in 16 mins, PacBio pipeline, DeepVariant fine-tuning
    • availability: Available Q2 2023
  • [27:00] NVIDIA BioNeMo Service
    • Cloud service for generative AI in drug discovery
    • specs: Pre-trained models for structure prediction, molecule generation, and docking
    • availability: Early access now, general availability 2H 2023
  • [33:34] Tokyo-1
    • Pharmaceutical supercomputer in Japan
    • specs: 16 NVIDIA DGX H100 nodes
    • availability: Live in the second half of 2023

Specific Numbers (9)

Timestamp Metric Value Context
05:25 Downloads 1,000,000 Number of downloads for the MONAI framework.
08:19 Dollars $500 billion Annual cost of clinical trials.
08:23 Percentage 90% Failure rate of drug candidates.
13:13 Patients 76 million Number of patients served globally by Medtronic.
15:46 Percentage 50% Reduction in miss rate of polyps using Medtronic’s GI Genius.
20:48 Minutes 16 Time to complete whole genome sequencing analysis with Parabricks 4.1.
26:18 Dollars $2 trillion Approaching size of the pharmaceutical industry.
26:34 Years 10 Average time to get a new drug to market.
31:02 Time 4 weeks Time Amgen took to develop generative AI models using BioNeMo, down from 3 months.

Benchmark Claims (3)

  • [15:46] Polyp Miss Rate: Up to 50% reduction
    • vs: Human-only colonoscopy
    • gain: Significant improvement in early detection of potential colorectal cancer.
  • [20:48] Whole Genome Sequencing Analysis Time: 16 minutes
    • vs: 21 minutes (previous version)
    • gain: 23% faster processing time.
  • [31:27] Protein Structure Prediction Speed: 5x to 100x faster
    • vs: Standard unoptimized models
    • gain: Massive acceleration in biologics discovery workflows.

Customer Stories (5)

  • [06:50] King’s College London
    • Developed a latent diffusion model for generating synthetic brain imaging data.
    • outcome: Enabled researchers to control global and regional brain volumes and adjust for age, gender, and disease.
  • [08:13] Bristol Myers Squibb
    • Used MONAI to build large foundation models for clinical trial outcomes.
    • outcome: Created a semantic segmentation model to detect, classify, and measure patient data for consistent response assessment.
  • [15:04] Medtronic
    • Developed GI Genius, an AI-assisted colonoscopy module.
    • outcome: Reduced the miss rate of polyps by up to 50% compared to human-only colonoscopy.
  • [30:28] Amgen
    • Used BioNeMo to customize property prediction models and generate proteins.
    • outcome: Reduced model development time from 3 months to 4 weeks and ran structure prediction models up to 100x faster.
  • [31:34] Deloitte
    • Developed ProMAP, an intelligent structural drug discovery pipeline using BioNeMo APIs.
    • outcome: Created an automated workflow that selects the best prediction model and refines structures for drug discovery.

Key Technologies (5)

  • NVIDIA Clara: A suite of computing platforms, software, and services for healthcare applications.
  • MONAI: Medical Open Network for AI, an open-source framework for building and deploying medical imaging AI.
  • NVIDIA Holoscan: A real-time AI sensor processing platform for building software-defined medical devices.
  • NVIDIA Parabricks: AI-accelerated genomic analysis software suite.
  • NVIDIA BioNeMo: A cloud service providing generative AI models for drug discovery, including protein structure prediction and molecule generation.

Demos Shown (3)

  • [06:34] Video demonstrating MONAI generative AI models, including latent diffusion for brain imaging and RadImageGAN.
    • True
  • [10:00] Video showing Holoscan processing real-time sensor data in surgical environments (endoscopy, ultrasound).
    • True
  • [27:38] Video demonstration of the BioNeMo web interface, showing protein structure prediction and molecule generation workflows.
    • True

Predictions / Commitments (5)

  • [13:31, Future] AI will help move upstream to less sick, high-risk people, increasing reach from 80 million to 800 million patients a year.
  • [18:18, End of the year (2023)] Medtronic GI Genius with Holoscan will be shipping.
  • [20:51, Q2 2023] Parabricks 4.1 will be available.
  • [32:21, Second half of 2023] BioNeMo general availability.
  • [34:22, Second half of 2023] Tokyo-1 supercomputer will be live.

Companies Mentioned (6)

Microsoft Azure, Google GCP, Oracle OCI · Amazon SageMaker · Arrow, PNY, Leadtek, Edom, Ryoyo, Macnica · Medtronic · Oxford Nanopore, PacBio · Mitsui & Co.

Notable Quotes (4)

Generative AI has become a human co-creation engine. — Kimberly Powell @ 01:33

It feels like it’s the Super Bowl for AI, and I couldn’t be more excited to be here. — Geoff Martha @ 12:59

For us to do this at scale and speed, to meaningfully impact society, we need AI. It’s that simple. — Geoff Martha @ 14:11

Generative AI is able to learn and represent biology, and it’s transforming biology from science to engineering. — Kimberly Powell @ 24:33

Key Topics

Generative AI · Healthcare · Medical Imaging · MONAI · Edge Computing · NVIDIA Holoscan · Medical Devices · Genomics · NVIDIA Parabricks · Drug Discovery · NVIDIA BioNeMo · Digital Biology · Supercomputing

Takeaways

  • Generative AI is fundamentally transforming the healthcare industry, acting as a co-creation engine across imaging, genomics, and drug discovery.
  • NVIDIA Clara provides a comprehensive, end-to-end AI computing platform tailored for the specific needs of healthcare and life sciences.
  • MONAI has become an industry standard for medical imaging AI, surpassing 1 million downloads and introducing advanced generative models for synthetic data.
  • NVIDIA Holoscan and IGX Orin enable the creation of software-defined medical devices capable of real-time AI processing at the edge.
  • Medtronic is partnering with NVIDIA to integrate Holoscan into its medical devices, starting with the GI Genius system to improve colonoscopy outcomes.
  • NVIDIA Parabricks continues to accelerate genomic analysis, reducing whole genome sequencing times and supporting new long-read technologies.
  • The new NVIDIA BioNeMo cloud service democratizes access to state-of-the-art generative AI models for drug discovery, significantly speeding up the R&D process.
  • Major pharmaceutical companies like Amgen are already leveraging BioNeMo to reduce model development time from months to weeks.
  • The Tokyo-1 supercomputer initiative will provide dedicated AI computing power to accelerate pharmaceutical research and development in Japan.