IPMN Risk Assessment under Federated Learning Paradigm

Hongyi Pan,Ziliang Hong,Gorkem Durak,Elif Keles,Halil Ertugrul Aktas,Yavuz Taktak,Alpay Medetalibeyoglu,Zheyuan Zhang,Yury Velichko,Concetto Spampinato,Ivo Schoots,Marco J. Bruno,Pallavi Tiwari,Candice Bolan,Tamas Gonda,Frank Miller,Rajesh N. Keswani,Michael B. Wallace,Ziyue Xu,Ulas Bagci

2024-11-09

Abstract:Accurate classification of Intraductal Papillary Mucinous Neoplasms (IPMN) is essential for identifying high-risk cases that require timely intervention. In this study, we develop a federated learning framework for multi-center IPMN classification utilizing a comprehensive pancreas MRI dataset. This dataset includes 653 T1-weighted and 656 T2-weighted MRI images, accompanied by corresponding IPMN risk scores from 7 leading medical institutions, making it the largest and most diverse dataset for IPMN classification to date. We assess the performance of DenseNet-121 in both centralized and federated settings for training on distributed data. Our results demonstrate that the federated learning approach achieves high classification accuracy comparable to centralized learning while ensuring data privacy across institutions. This work marks a significant advancement in collaborative IPMN classification, facilitating secure and high-accuracy model training across multiple centers.

Image and Video Processing,Distributed, Parallel, and Cluster Computing,Machine Learning

What problem does this paper attempt to address?

The paper attempts to address the problem of accurately classifying Intraductal Papillary Mucinous Neoplasms (IPMN) in a multi-center environment to identify high-risk cases that require timely intervention. Specifically:

1. **Challenges in IPMN Classification**:
- **Diversity and Scale of Datasets**: Due to differences in imaging protocols between different medical institutions and the rarity of IPMN, developing robust and generalizable machine learning models is challenging.
- **Data Privacy**: Centralized data aggregation is difficult to achieve due to privacy regulations on medical imaging.

2. **Solution**:
- **Federated Learning Framework**: Through the Federated Learning (FL) framework, multiple medical institutions can collaboratively train models without sharing raw data, thus protecting patient data privacy.
- **Multi-center Dataset**: A large multi-center dataset was constructed, including 653 T1-weighted and 656 T2-weighted MRI images from 7 leading medical institutions, making it the largest IPMN classification dataset to date.

3. **Research Methods**:
- **Model Selection**: DenseNet-121 was used as the backbone network to evaluate its performance in both centralized and federated learning environments.
- **Federated Learning Algorithms**: Federated Averaging (FedAvg) and Federated Proximal (FedProx) optimization algorithms were employed to address data heterogeneity issues.

4. **Experimental Results**:
- **Three-class Classification Experiment**: In preliminary experiments, DenseNet-121 achieved AUC values of 0.7632 and 0.8092 on T1 and T2 modalities, respectively.
- **Binary Classification Experiment**: In a multi-center environment, binary classification experiments of high-risk vs. non-high-risk were conducted using both centralized and federated learning methods. The results showed that federated learning methods could achieve classification accuracy comparable to centralized learning while maintaining data privacy.

Through these methods, the paper aims to provide a scalable and privacy-preserving IPMN classification solution, laying the foundation for future clinical applications.