INDIVIDUALIZED TRAJECTORY PREDICTION OF EARLY DEVELOPING FUNCTIONAL CONNECTIVITY.

Area of Science:

• Neuroscience
• Developmental Neuroscience
• Computational Neuroscience

Background:

• Predicting functional connectivity (FC) development from resting-state functional MRI (rs-fMRI) is crucial for understanding infant brain organization.
• Current deep learning methods struggle with longitudinal dependencies and data scarcity, leading to inconsistent FC predictions.
• Accurately modeling early brain functional development trajectories is challenging due to irregular data distribution.

Purpose of the Study:

• To propose a novel deep learning approach for accurate trajectory prediction of infant FC development.
• To address limitations of existing methods in handling longitudinal dependencies and data irregularities.
• To enable consistent prediction of FC at any infant age, capturing individual developmental characteristics.

Main Methods:

• Developed a Triplet Cycle-Consistent Masked Autoencoder (TC-MAE) model.
• TC-MAE is designed to process extended periods of FC data, identify individual traits, and predict future FC.
• The model ensures longitudinal consistency in FC trajectory predictions.

Main Results:

• TC-MAE demonstrated superior performance in longitudinal FC prediction compared to existing state-of-the-art methods.
• Experiments were conducted on 368 longitudinal infant rs-fMRI scans.
• The proposed method achieved more accurate and consistent predictions of infant brain functional development.

Conclusions:

• The TC-MAE model offers a significant advancement in predicting infant brain functional connectivity trajectories.
• This method effectively overcomes challenges posed by scarce and irregularly sampled longitudinal rs-fMRI data.
• TC-MAE provides a robust tool for elucidating intrinsic brain functional organization and its dynamic development during infancy.