Abstract:Brain network analysis is vital for understanding the neural interactions regarding brain structures and functions, and identifying potential biomarkers for clinical phenotypes. However, widely used brain signals such as Blood Oxygen Level Dependent (BOLD) time series generated from functional Magnetic Resonance Imaging (fMRI) often manifest three challenges: (1) missing values, (2) irregular samples, and (3) sampling misalignment, due to instrumental limitations, impacting downstream brain network analysis and clinical outcome predictions. In this work, we propose a novel model called BrainODE to achieve continuous modeling of dynamic brain signals using Ordinary Differential Equations (ODE). By learning latent initial values and neural ODE functions from irregular time series, BrainODE effectively reconstructs brain signals at any time point, mitigating the aforementioned three data challenges of brain signals altogether. Comprehensive experimental results on real-world neuroimaging datasets demonstrate the superior performance of BrainODE and its capability of addressing the three data challenges.

What problem does this paper attempt to address?

The paper aims to address three major challenges in dynamic brain signal analysis: missing values, irregular sampling, and sampling misalignment, and proposes a new model, BrainODE, to tackle these issues. Specifically: - **Missing values**: Due to equipment limitations or mechanical noise, neuroimaging data may have missing values at certain time points. - **Irregular sampling**: Equipment may not be able to collect data at the expected time points accurately, for example, data that should be collected at 1 second is actually collected at 1.2 seconds. - **Sampling misalignment**: Different samples may be collected at different frequencies, for instance, one device collects data once per second, while another device collects data once every 2 seconds. These issues lead to deficiencies in existing brain network analysis methods based on blood-oxygen-level-dependent (BOLD) time series generated from functional magnetic resonance imaging (fMRI), especially in handling dynamic properties and data quality. To address these problems, the paper proposes the BrainODE model, which uses ordinary differential equations (ODE) to continuously model dynamic brain signals, thereby enabling the reconstruction of brain signals at any given time point, effectively mitigating the aforementioned three data challenges. Experimental results show that BrainODE outperforms existing methods on real-world neuroimaging datasets, particularly in terms of handling missing values, irregular sampling, and sampling misalignment.

BrainODE: Dynamic Brain Signal Analysis via Graph-Aided Neural Ordinary Differential Equations