4D-FED-GNN+, a federated brain graph evolution trajectory prediction framework that learns from brain connectivity data with missing timepoints, coded up in Python by Zeynep Gürler. Please contact [email protected] for inquiries. Thanks.

Federated Brain Graph Evolution Prediction using Decentralized Connectivity Datasets with Temporally-varying Acquisitions

Zeynep Gürler¹, Islem Rekik¹

¹BASIRA Lab, Faculty of Computer and Informatics, Istanbul Technical University, Istanbul, Turkey

Abstract: Foreseeing the evolution of brain connectivity between anatomical regions from a baseline observation can propel early disease diagnosis and clinical decision making. Such task becomes challenging when learning from multiple decentralized datasets with missing timepoints (e.g., datasets collected from different hospitals with a varying sequence of acquisitions). Federated learning (FL) is an emerging paradigm that enables collaborative learning among multiple clients (i.e., hospitals) in a fully privacy-preserving fashion. However, to the best of our knowledge, there is no FL work that foresees the time-dependent brain connectivity evolution from a single timepoint --let alone learning from non-iid decentralized longitudinal datasets with \emph{varying acquisition timepoints}. In this paper, we propose the first FL framework to significantly boost the predictive performance of local hospitals with missing acquisition timepoints while benefiting from other hospitals with available data at those timepoints without sharing data. Specifically, we introduce 4D-FED-GNN+, a novel longitudinal federated GNN framework that acts as (i) a graph self-encoder if the next timepoint is locally missing or (ii) a graph generator if the local follow-up data is available. Further, we propose a dual federation strategy, where (i) GNN layer-wise weight aggregation and (ii) pairwise GNN weight exchange between hospitals in a random order. To improve the performance of the poorly-conditioned hospitals (e.g., consecutive missing timepoints, intermediate missing timepoint), we further propose a second variant, namely 4D-FED-GNN++, which federates based on an ordering of the local hospitals computed using their incomplete sequential patterns. Our comprehensive experiments on real longitudinal datasets show that overall 4D-FED-GNN+ and 4D-FED-GNN++ significantly outperform benchmark methods.

Our framework is a brain graph evolution trajectory prediction framework that learns from decentralized datasets with different acquisition timepoints resulting with missing timepoints for some data owners. Our Federated learning-based framework comprises two key steps. (1) GNN-based time-dependent prediction, (2) Mixed federation strategy. Experimental results against comparison methods demonstrate that our framework can achieve the best results in terms of Mean Absolute Error (MAE). We used the OASIS-2 longitudinal dataset collected by (https://www.oasis-brains.org/) to evaluate our 4D-FED-GNN+. Further, we also evaluated our proposed framework using a simulated dataset.

More details can be found at: (link to the paper) and our research paper video on the BASIRA Lab YouTube channel (link).

• Python 3.8
• PyTorch 1.7.0
• Torch-geometric
• Torch-sparse
• Torch-scatter
• Scikit-learn 0.23.0+
• Matplotlib 3.1.3+
• Numpy 1.20.1+

4D-FED-GNN+ is coded in Python 3.8 on Windows 10. GPU is not needed to run the code. demo.py is the implementation of 4D-FED-GNN+ proposed by Federated Brain Graph Evolution Prediction using Decentralized Connectivity Datasets with Temporally-varying Acquisitions paper. In this repo, we release the 4D-FED-GNN+ source code trained and tested on a simulated data as shown below:

Data preparation

We provide a demo code for the usage of 4D-FED-GNN+ for brain graph evolution trajectory prediction for decentralized datasets with missing timepoints. In demo.py, we train 4D-FED-GNN+ on a simulated dataset with 120 subjects with 6 timepoints which are pre-obtained using dataset.py. If you need to generate new simulated data with different distribution, you can change the mean correlation matrix and connectivity mean values in dataset.py.

If you want to train 4D-FED-GNN+ on the pre-obtained simulated data, you can set the model parameters in arg_parse function of demo.py and execute,

python demo.py

If you want to train on simulated dataset using 4D-GNN and 4D-FED-GNN techniques, you can change the mode parameter in arg_parse function of demo.py

demo.py generates the following outputs:

Network Normalization for Integrating Multi-view Networks (netNorm): Dhifallah, S., Rekik, I., 2020, Estimation of connectional brain templates using selective multi-view network normalization

To install and run 4D-FED-GNN+, check the following YouTube video:

https://bit.ly/3t74vga

To learn about how 4D-FED-GNN+ works, check the following YouTube video:

https://bit.ly/3WHGGJp

@inproceedings{gurler2022federated,
  title={Federated Time-Dependent GNN Learning from Brain Connectivity Data with Missing Timepoints},
  author={G{\"u}rler, Zeynep and Rekik, Islem},
  booktitle={International Workshop on PRedictive Intelligence In MEdicine},
  pages={1--12},
  year={2022},
  organization={Springer}
}