Hoifung Poon: Advancing Precision Health with AI at Microsoft

Hoifung Poon: Pioneering AI in Precision Health

Hoifung Poon is a leading figure in the field of Artificial Intelligence, particularly in its application to precision health. As the General Manager of Microsoft Health Futures, he spearheads research focused on leveraging Real-World Evidence (RWE) to advance AI in healthcare. His work aims to transform medicine from a reactive, imprecise practice into a proactive, data-driven system.

The Vision of Precision Health

Traditional medicine often struggles with patient variability, where a significant percentage of patients do not respond to common treatments. Precision health seeks to overcome this by developing a learning system that instantly incorporates new health information to optimize care and accelerate biomedical discovery. However, the healthcare ecosystem is hampered by unstructured data and manual processing, creating significant bottlenecks.

Key Research Areas and Contributions

Poon's research focuses on developing next-generation AI technologies to improve access, safety, and preventative care in precision health. His key areas of work include:

Biomedical Large Language Models (LLMs): Microsoft Health Futures is at the forefront of developing and applying LLMs in biomedicine, building on foundational work like PubMedBERT and BioGPT. The team is actively working on methods to enhance LLM fact-checking, provide fine-grained provenance, and facilitate efficient human-in-the-loop verification to address limitations such as incorrect generation.
Biomedical Multimodal Learning: Recognizing the richness of data beyond text, the research extends to integrating information from radiology images, digital pathology slides, and genomics. The goal is to develop multimodal learning and fusion methods for end-to-end applications like predicting disease progression and drug response.
Causal Learning for Real-World Evidence: Addressing the challenges of confounders in observational data, the team is developing advanced causal methods to correct implicit biases and scale biomedical discovery. This is crucial for generating reliable real-world evidence.

Notable Publications and Achievements

Hoifung Poon has an extensive publication record, with significant contributions to top-tier conferences and journals. Some of his recent work highlights include:

Whole-slide Foundation Model for Digital Pathology: A Nature 2024 publication introducing GigaPath, a foundation model for digital pathology trained on real-world data, with associated code and model releases.
Biomedical Multimodal Foundation Model: A NEJM AI 2024 publication detailing a foundation model trained on fifteen million image-text pairs, advancing multimodal understanding in biomedicine.
BiomedParse: A Nature Methods 2024 publication on a foundation model for joint segmentation, detection, and recognition across nine biomedical modalities, with code and demo available.
X-Reasoner: Research on a model for generalizable reasoning across modalities and domains.
Med-RLVR: Work on emerging medical reasoning from a 3B base model via reinforcement learning.
Universal Abstraction: Developing methods to structure real-world data at scale using frontier models.
MuirBench: A comprehensive benchmark for robust multi-image understanding.
Widespread Adoption of Precision Anticancer Therapies: A study published in ASCP, highlighting the impact of pathologist-directed comprehensive genomic profiling.
CheXprompt: Research on clinically accessible radiology foundation models.
GPT-4 in Radiology: Exploring the capabilities of GPT-4 for medical imaging tasks.
LLaVA-Med: Training a large language-and-vision assistant for biomedicine.
UniversalNER: Advancing named entity recognition for biomedical applications.
TRIALSCOPE: A causal framework for scaling real-world evidence generation with LLMs.
BiomedJourney: Generating counterfactual biomedical images from patient journeys.
BioGPT: A generative pre-trained transformer for biomedical text generation.
PubMedBERT and BioMedBERT: Early contributions to domain-specific language models for biomedical NLP.

Academic and Professional Background

Hoifung Poon holds a B.S. with Distinction in Computer Science from Sun Yat-Sen University and a Ph.D. in Computer Science and Engineering from the University of Washington. He is an affiliated faculty member at the University of Washington Medical School and has served as a co-PI on various academic projects, including DARPA Big Mechanisms. His earlier work has been recognized with Best Paper Awards at top conferences like NAACL, EMNLP, and UAI.

Recognition and Impact

In 2024, Hoifung Poon was honored as the "Technology Champion" by the Puget Sound Business Journal at the annual Health Care Leadership Awards, recognizing his significant contributions to the field. His research has been featured in numerous press outlets, including Bloomberg Technology, Microsoft News, The Verge, and Nature News, highlighting the transformative potential of AI in healthcare and cancer research.

Tutorials and Resources

Poon has also contributed to the field through tutorials on topics such as Markov Logic in Natural Language Processing, Natural Language Processing for Precision Medicine, Machine Reading for Precision Medicine, and Precision Health in the Age of Large Language Models.

Contact and Further Information

For more details on Hoifung Poon's work, one can refer to his publications on Google Scholar, his LinkedIn profile, and the Microsoft Health Futures lab page. He can be reached at hoifung@microsoft.com.

Microsoft Health Futures

Located at Microsoft Corporation, One Microsoft Way, Redmond, Washington, the Microsoft Health Futures lab is dedicated to advancing AI in healthcare. They actively collaborate with stakeholders in healthcare and life sciences to translate research into real-world applications.

Social Media Presence

Microsoft Research maintains an active presence on social media platforms including X (formerly Twitter), Facebook, LinkedIn, YouTube, and Instagram, sharing updates on their latest research and innovations.