AIM-FM: Advancements In Medical Foundation Models: Explainability, Robustness, Security, and Beyond

Towards next-generation medical analysis: Unlock the potential of medical foundation models for more explainable, robust, secure diagnosis solutions

Co-located with NeurIPS'24

December 14, 2024

Vancouver Convention Center

Overview

There have been notable advancements in large foundation models (FMs), which exhibit generalizable language understanding, visual recognition, and audio comprehension capabilities. These advancements highlight the potential of personalized AI assistants in efficiently assisting with daily tasks, ultimately enhancing human life.

Healthcare is one of the most crucial industries touching every individual. Yet, due to large populations and limited medical professionals, it faces significant challenges, including the high cost and low doctor-to-population ratio. This shortage is more pronounced in rural and developing regions, where access to qualified doctors is severely limited, exacerbating health disparities and preventing timely treatment for common and complex conditions alike. Hence, there is a critical need to develop effective, affordable, and professional AI-driven medical assistants.

Despite the great success in general domains, FMs struggle in specific domains requiring strict professional qualifications, such as healthcare, which has high sensitivity and security risk. In light of the growing healthcare demands, this workshop aims to explore the potential of Medical Foundation Models (MFMs) in smart medical assistance, thereby improving patient outcomes and streamlining clinical workflows. Considering the primary clinical needs, we emphasize the explainability, robustness, and security of the large-scale multimodal medical assistant, pushing forward its reliability and trustworthiness. By bringing together expertise in diverse fields, we hope to bridge the gap between industry and academia regarding precision medicine, highlighting clinical requirements, inherent concerns, and AI solutions. Through this cooperative endeavor, we aim to unlock the potential of MFMs, striving for groundbreaking advancements in healthcare.

Topics of Interest

Key topics of interest for the workshop may cover, but are not limited to, the following aspects.

MFMs at Scale Develop large-scale medical foundation models applicable for hospital use, including diagnosis, prognosis, treatment, and surgical assistance.
Explainable MFMs Open the black box of MFMs in medical decision-making, ensuring transparency and interpretability.
Robust Diagnosis Enhance the robustness of MFMs in diverse medical scenarios: scarcity/misalignment of medical data, parameter-efficient tuning, and validation techniques.
Patient Privacy Ensure data/model privacy in tuning and testing MFMs: federated learning, data encryption, and machine unlearning.
MFMs with Resource Constraint Research on optimizing MFMs with constrained resources, e.g., constrained computation, limited data and annotations, etc.
Human-AI Interaction Study the interaction dynamics to enhance the collaboration between healthcare professionals/patients and AI: prompt engineering, feedback refining, and system designing.
Multimodal Learning Effectively use heterogeneous medical data by addressing multimodal challenges: modality misalignment and missing.
Generative Model for Healthcare Develop generative models for producing multimodal data for healthcare: generative medical images, videos, reports, and biology structures.
Efficient MFMs Develop efficient MFMs in medical assistants: data efficiency, annotation efficiency, and small foundation model.
Agent for Healthcare Towards the applications of AI agent systems in healthcare: diagnosis, prognosis, surgical assistance, telehealth.
Fairness in MFMs Develop fair multimodal models in healthcare: addressing bias from data, model, annotation, and evaluation.

Schedule

Morning Session
8:20 - 8:30 AM: Introduction

8:30 - 9:10 AM: Speaker: Pearse Keane

9:10 - 9:50 AM: Speaker: Faisal Mahmood

9:50 - 10:00 AM: Break (10 minutes)

10:00 - 10:40 AM: Speaker: Sheng Wang

10:40 - 11:20 AM: Panel Discussion: Pearse Keane, Faisal Mahmood, Tianming Liu, James Zou, Shekoofeh Azizi

11:20 AM - 12:00 PM: Poster Session
Lunch Break
12:00 - 1:00 PM Afternoon Session
1:00 - 1:40 PM: Speaker: Tianming Liu

1:40 - 2:20 PM: Speaker: James Zou

2:20 - 2:30 PM: Break (10 minutes)

2:30 - 3:30 PM: Oral Session

3:30 - 4:10 PM: Speaker: Animashree Anandkumar

4:10 - 4:50 PM: Speaker: Karan Singhal

4:50 - 5:00 PM: Break (10 minutes)

5:00 - 5:10 PM: Award Presentation and Summary

5:10 - 5:30 PM: Poster Session

Submissions

The main text of a submitted paper is limited to nine content pages, including all figures and tables. Authors are encouraged to submit a separate ZIP file that contains any supplementary material, such as data or source code, when applicable. The submission must be formatted using the NeurIPS'24 template and will be collected on the OpenReview website, where the files must be in PDF format as a single PDF file. The reviewing process will be double-blind, and all submissions must be anonymized. This means that you should not include any author names, author affiliations, acknowledgments, or any other identifying information in your submission.

Submission Deadline: Extended to Sept 26, 2024, 23:59 UTC-0.
Workshop Accept/Reject Notification: Oct 09 '24 (Anywhere on Earth).

Invited Speakers and Panelists

Prof. Animashree Anandkumar

Bren Professor at California Institute of Technology, Pasadena, CA; Senior Director of AI Research at NVIDIA, Santa Clara, CA

Dr. Shekoofeh Azizi

Staff research scientist at Google DeepMind (formerly Google Brain) and a research lead for Health AI at Google DeepMind, Canada

Prof. Pearse Keane

Professor of Artificial Medical Intelligence at UCL Institute of Ophthalmology; Consultant ophthalmologist at Moorfields Eye Hospital, London

Prof. Tianming Liu

Distinguished Research Professor at University of Georgia (UGA), USA; Director of UGA Bioimaging Research Center, UGA Institute of Bioinformatics, and UGA Institute of Artificial Intelligence