GTC 2022 Healthcare Special Address

Category: Healthcare Special Address · Year: 2022 · ▶ Watch

Switch language → 中文

Segments (16)

  • 00:00 · Introduction and GTC Overview
    • Kimberly Powell welcomes attendees to GTC 2022 and highlights the diverse range of speakers and topics in healthcare AI.
  • 01:53 · NVIDIA Clara Platform
    • Overview of NVIDIA Clara, an AI computational platform for healthcare, addressing the massive growth in healthcare data.
  • 03:34 · MONAI Framework
    • Update on MONAI, the open-source medical AI framework, highlighting its rapid adoption and new features like MONAI Label.
  • 05:28 · Digital Biology Revolution
    • Discussion on the digital biology revolution driven by advancements in genomic sequencing and the need for accelerated computing.
  • 07:18 · Breakthroughs in Nanopore Sequencing
    • Highlighting the Guinness World Record for ultra-rapid DNA sequencing achieved by Stanford, Oxford Nanopore, and NVIDIA.
  • 10:28 · Announcing UNAP
    • Introduction of the Ultra-Rapid Nanopore Analysis Pipeline (UNAP) for end-to-end acceleration of nanopore whole genome sequencing.
  • 11:23 · Digital Biology Meets AI
    • Exploring how AI and deep learning are transforming drug discovery and digital biology, moving beyond traditional methods.
  • 15:13 · NVIDIA Powers Transformer Neural Network Revolution
    • Overview of NVIDIA’s frameworks and pre-trained models for large language models in biology and chemistry.
  • 16:58 · Announcing Clara Discovery MegaMolBART Early Access
    • Introduction of MegaMolBART, a large chemical language model developed with AstraZeneca, available for early access.
  • 17:53 · Announcing NVIDIA Cambridge-1 Startups
    • Highlighting four startups using the Cambridge-1 supercomputer to accelerate AI drug discovery.
  • 20:08 · Janssen and NVIDIA Collaborate
    • Details on the collaboration between Janssen and NVIDIA to improve adverse event detection using BioMegatron.
  • 21:28 · U. Florida and NVIDIA Create GatorTron
    • Announcement of GatorTron and SynGatorTron, the world’s largest clinical generative NLP models developed with the University of Florida.
  • 22:48 · Software-Defined Medical Devices
    • Discussion on the future of medical devices becoming software-defined and robotic, driven by AI and continuous sensing.
  • 23:58 · NVIDIA Clara Holoscan
    • Overview of Clara Holoscan, an open, real-time AI computing platform for medical devices.
  • 27:33 · Announcing Clara Holoscan Developer Kit
    • Introduction of the Clara Holoscan Developer Kit, providing the hardware and software needed to build software-defined medical devices.
  • 29:38 · Announcing Clara Holoscan MGX
    • Introduction of Clara Holoscan MGX, a medical-grade, real-time AI computing platform for production deployment.

Product Announcements (5)

  • [10:28] Ultra-Rapid Nanopore Analysis Pipeline (UNAP)
    • End-to-end acceleration of nanopore whole genome sequencing.
    • specs: Runs on a single DGX A100, containerized on NGC, high accuracy AI whole genome variant calling, reduces compute cost from $568 to $183 per patient, achieves real-time AI base calling and alignment.
    • availability: Available Now
  • [16:58] Clara Discovery MegaMolBART
    • A large chemical language model for cheminformatics.
    • specs: Achieves molecule generation at supercomputing scale with high validity and uniqueness (98%), molecular optimization using latent space search, easy fine-tuning with NVIDIA NeMo.
    • availability: Early Access April
  • [21:28] SynGatorTron and GatorTron
    • The world’s largest clinical generative NLP models.
    • specs: Clinical language generation model, used to generate synthetic de-identified clinical notes and reports, trained GatorTron-S with synthetic data.
    • availability: Available April
  • [27:33] NVIDIA Clara Holoscan Developer Kit
    • A developer kit for building software-defined medical devices.
    • specs: Multi-modality sensor front ends, cloud-native SDK, reference applications, includes NVIDIA ConnectX-7, NVIDIA Orin, and RTX A6000.
    • availability: Available Now, Early Access May
  • [29:38] NVIDIA Clara Holoscan MGX
    • A real-time, medical-grade AI computing platform.
    • specs: Reference design for embedded edge computing, scalable AI architecture (250-600 TOPS), streaming IO up to 200 GbE with GPU Direct RDMA, safety, security, and manageability built-in, 10-year whole stack software support, built for medical IEC 60601, 62304.
    • availability: Early Access Q1 2023

Specific Numbers (8)

Timestamp Metric Value Context
01:53 Healthcare Data Growth 36% CAGR by 2025 Healthcare data is 30% of the world’s data and is growing rapidly.
03:34 MONAI Downloads 50,000 per month MONAI is approaching 50,000 downloads per month, a 7.5x increase in one year.
05:28 Genomic Data Generation 40 Exabytes by 2025 The global trend is to generate over 40 exabytes of genomic data by 2025.
07:18 Sequencing Time 7 hours and 18 minutes The Guinness World Record time for ultra-rapid DNA sequencing from sample to diagnosis.
10:28 Compute Cost Reduction $568 to $183 per patient UNAP reduces the compute cost for nanopore whole genome sequencing.
11:23 VC Funding in Drug Discovery $60 Billion Over $60 billion in VC funding has been invested in drug discovery over the last two years.
20:08 Adverse Event Detection Accuracy 12% increase to 88% Janssen improved adverse event detection accuracy using NVIDIA NeMo.
22:48 Aging Population 1 in 6 people over 60 by 2030 Highlighting the growing demand for medical care due to an aging population.

Benchmark Claims (2)

  • [10:28] Nanopore WGS Analysis Pipeline: 2h 46m (Basecalling & Alignment), 1h 24m (Multiple Variant Calling)
    • vs: Previous methods
    • gain: Real-time AI basecalling and alignment, significant reduction in compute cost.
  • [20:08] Adverse Event Detection: 88% accuracy
    • vs: Previous baseline
    • gain: 12% increase in accuracy, 2x improvement in model throughput.

Customer Stories (4)

  • [07:18] Stanford University and Oxford Nanopore
    • Developed an ultra-rapid DNA sequencing technique for critical care patients.
    • outcome: Achieved a Guinness World Record for the fastest DNA sequencing technique, reducing the time from sample to diagnosis to 7 hours and 18 minutes.
  • [16:58] AstraZeneca
    • Collaborated with NVIDIA to develop MegaMolBART, a large chemical language model.
    • outcome: Created a model capable of generating molecules at supercomputing scale with 98% validity and uniqueness.
  • [20:08] Janssen
    • Used NVIDIA NeMo and BioMegatron to improve adverse event detection.
    • outcome: Increased accuracy by 12% to achieve 88% and improved model throughput by 2x.
  • [21:28] University of Florida
    • Created GatorTron and SynGatorTron, the world’s largest clinical generative NLP models.
    • outcome: Developed models capable of generating synthetic, de-identified clinical notes and reports, improving performance on clinical NLP tasks.

Key Technologies (7)

  • NVIDIA Clara: AI computational platform for healthcare.
  • MONAI: Open-source medical AI framework.
  • UNAP: Ultra-Rapid Nanopore Analysis Pipeline for genomic sequencing.
  • NVIDIA NeMo Megatron: Framework for training large language models.
  • MegaMolBART: Large chemical language model for drug discovery.
  • GatorTron: Clinical generative NLP model.
  • Clara Holoscan: Real-time AI computing platform for medical devices.

Demos Shown (1)

  • [24:53] Light-sheet microscopy with NVIDIA Clara Holoscan, visualizing living cells in real-time.
    • Yes

Predictions / Commitments (4)

  • [01:53, 2025] By 2025, healthcare data will be growing at the highest compound annual growth rate of any other industry at 36%.
  • [05:28, 2025] A global trend to generate over 40 exabytes of genomic data by 2025.
  • [22:48, 2030] By 2030, one in six people will be over 60.
  • [22:48, Future] Every medical device will become robotic.

Companies Mentioned (5)

AWS, Google Cloud, Azure · Oxford Nanopore Technologies · AstraZeneca · Janssen · University of Florida

Notable Quotes (3)

Today, healthcare data is 30% of all the world’s data. And by 2025, healthcare data will be growing at the highest compound annual growth rate of any other industry at 36%. — Kimberly Powell @ 01:53

Digital biology has become an information science. — Kimberly Powell @ 11:23

Every medical device will become robotic. — Kimberly Powell @ 22:48

Key Topics

Healthcare AI · NVIDIA Clara · MONAI · Digital Biology · Genomic Sequencing · Drug Discovery · Large Language Models · NLP in Healthcare · Software-Defined Medical Devices · Clara Holoscan · Edge Computing · Real-time AI

Takeaways

  • Healthcare data is growing exponentially, requiring advanced AI and accelerated computing solutions.
  • NVIDIA Clara provides a comprehensive platform for healthcare AI, from training to deployment.
  • MONAI is rapidly becoming the standard open-source framework for medical imaging AI.
  • AI is revolutionizing digital biology and drug discovery, significantly reducing time and costs.
  • Large language models (Transformers) are being applied to chemistry and biology to accelerate research.
  • Medical devices are transitioning to software-defined, robotic systems powered by real-time AI.
  • NVIDIA Clara Holoscan offers a complete hardware and software platform for building the next generation of medical devices.