Building a Federated Multimodal AI Network for Cancer Trial Matching: Co-Design Lessons from Four Cancer Centers

Presentation Type: Panel
Presentation Time: 02:00 PM - 03:15 PM

Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Clinical Decision Support, Natural Language Processing
Programmatic Theme: Clinical Research Informatics

Cancer clinical trials remain critically underenrolled in the U.S., and a major barrier is reliably identifying eligible patients from fragmented, unstructured documentation and multimodal clinical data. This panel describes a four-site collaboration across NCI-designated comprehensive cancer centers to co-design and implement a privacy-preserving, federated multimodal AI network for trial eligibility signal extraction and matching—without sharing protected health information. Using NV-FLARE orchestration and parameter-efficient fine-tuning (QLoRA adapters) on a shared MedGemma foundation model, each site fine-tunes locally while retaining data control, with audit logging and secure aggregation of encrypted updates to support compliance and reproducibility. Panelists will share implementation patterns and pitfalls spanning infrastructure packaging, security/IRB alignment, early feasibility benchmarks (text classification and structured extraction), and a prospective validation design for real-world trial matching workflows. The collaboration has demonstrated cross-site connectivity in simulation and completed initial local fine-tuning at participating sites.

Moderator:
Umit Topaloglu, PhD
National Cancer Institute

Speaker(s):
Wei Zhang, PhD
Wake Forest School of Medicine
Ghulam Rasool, PhD
Moffitt Cancer Center
Jihad Obeid, MD
Medical University of South Carolina
Kirk Roberts, PhD
University of Texas Health Science Center at Houston
Author(s):
Umit Topaloglu, PhD - National Cancer Institute; Ghulam Rasool, PhD - Moffitt Cancer Center; Jihad Obeid, MD - Medical University of South Carolina; Hongfang Liu, PhD - University of Texas Health Science Center at Houston; Wei Zhang, PhD - Wake Forest School of Medicine;

---