AIOZ AI Research: Federated Endovascular Foundation Model with Unseen Data

We are proud to announce that our AI research paper, titled "Federated Endovascular Foundation Model with Unseen Data," has been accepted and published at the IEEE International Conference on Robotics and Automation (ICRA) 2025 @ieee_ras_icra!

This research introduces a Foundation AI Model that achieves state-of-the-art performance while preserving data security, setting a new benchmark for robotic-assisted endovascular surgery.

Let's dive into the key aspects of this groundbreaking research below:

Introducing FedEFM: The Future of Federated Endovascular Foundation Models

Endovascular surgery requires precise catheter and guidewire detection in X-ray images to ensure patient safety. However, training deep learning models for this task is challenging due to limited labeled data and strict privacy regulations.

FedEFM addresses these challenges by training a foundation model using federated learning, enabling collaborative model improvement without sharing sensitive patient data.

Key innovations of FedEFM include:

• Federated Learning Setup: Enables collaborative model training across multiple hospital silos to ensure data privacy.
• New Multishot Distillation Technique: Uses Earth Mover’s Distance (EMD) to address challenges posed by unseen data distributions.
• Diverse Endovascular Dataset: Combines data from different sources, including human, animal, phantom, and simulated X-ray images.
• Optimization for Downstream Tasks: Enhances model performance for catheter and guidewire segmentation tasks.

Building a Dataset for FedEFM: Privacy-Compliant and High-Fidelity

Developing a robust foundation model for endovascular intervention requires diverse and high-quality datasets. However, the data privacy requirement in this federated learning setup complicates data collection and management.

Here’s how FedEFM tackles these challenges:

• Robotic Data Collection Setup: A robotic system captures large-scale X-ray images using a master-follower control setup, generating the EIPhantom dataset with 4,700 labeled samples.
• Real & Simulated Data: Merges real-world X-ray images with data from the CathSim simulator, ensuring a diverse dataset for model training.
• Federated Learning with Unseen Data: Employs a multishot federated distillation algorithm, enhanced by EMD, to enable privacy-preserving learning across isolated datasets in different hospitals.

Why this matters:

• FedEFM is the first federated foundation model tailored for endovascular interventions.
• It provides pre-trained weights that can be applied directly to downstream medical imaging tasks—boosting performance without compromising privacy.

Training FedEFM: Techniques Behind the Model

FedEFM leverages a decentralized learning approach known as federated knowledge distillation alongside EMD-based optimization to ensure effective and privacy-preserving learning across distributed environments.

Federated Knowledge Distillation for Unseen Data

• Each hospital independently trains a local model on its private dataset.
• Weights are transferred to neighboring silos for shared learning.
• EMD quantifies inter-silo similarities and optimizes weight fusion.
• A custom loss function balances local training and knowledge from other silos.

EMD-Based Optimization For Knowledge Transfer

• Feature representations are compared between local and transferred models.
• The optimal matching flow minimizes the distance for knowledge adaptation.
• The gradients are computed efficiently for backpropagation without modifying the optimization path.

Knowledge Aggregation

• Each silo refines its local model using EMD-weighted objectives, learning from both its own data and insights transferred from others.
• The result is a global aggregated model that encapsulates collective knowledge from all participating silos!

FedEFM in Action: Evaluation & Results

FedEFM sets a new benchmark for privacy-preserving medical AI with its strong performance on unseen data and real-world tasks:

• State-of-the-art validation across Centralized Local Learning (CLL), Client-server Federated Learning (CFL), and Decentralized Federated Learning (DFL)—outperforming existing methods with up to 98.2% accuracy.
• Robust fine-tuning for endovascular classification & segmentation tasks, leveraging models like CLIP, SAM, and LVM-Med.
• Unseen data resilience—FedEFM maintains 84.9% accuracy in extreme cases where all data labels are unseen.
• Backbone adaptability—integrates with UNet, TransUNet, SwinUNet, and ViT to improve segmentation and classification tasks.

FedEFM demonstrates significant potential for federated learning in medical AI, setting new standards in privacy-preserving endovascular intervention models.

Challenges remain in hardware variability and training time, but the approach paves the way for future applications, including robotic-assisted surgery and pathology.

If you would like to dive deeper into this research work, you can refer to the links below:

• Full paper: arXiv:2501.16992
• Project page: aioz-ai.github.io/FedEFM
• Source code: GitHub – aioz-ai/FedEFM

About the AIOZ Network

AIOZ Network is a DePIN for Web3 AI, Storage, and Streaming.

AIOZ empowers a fast, secure, and decentralized future.

Powered by a global community of AIOZ DePIN, AIOZ rewards you for sharing your computational resources for storing, transcoding, and streaming digital media content and powering decentralized AI computation.

Find Us

AIOZ All Links | Website | X | Telegram