Multi-Regression based supervised sample selection for predicting baby connectome evolution trajectory from neonatal...

Area of Science:

• Neuroscience
• Developmental Biology
• Computational Biology

Background:

• The infant brain's connectome undergoes rapid changes, making prediction difficult.
• Lack of longitudinal data hinders development of models for predicting brain network trajectories.
• Predictive models are crucial for early identification of atypical neurodevelopmental disorders.

Purpose of the Study:

• To propose the first approach for predicting longitudinal brain network development from a single baseline connectome.
• To address data scarcity and incompleteness in infant neuroimaging studies.
• To enable early detection of neurodevelopmental disorders by forecasting brain connectome evolution.

Main Methods:

• Developed a supervised multi-regression sample selection strategy to identify relevant neighboring connectomes.
• Employed low-rank tensor completion with robust principal component analysis to impute missing connectome data.
• Utilized an ensemble of bidirectional regressors to leverage temporal dependencies for progressive prediction.

Main Results:

• The proposed method achieved superior prediction accuracy compared to ablated versions.
• The approach effectively captures dynamic changes in brain connectomes over time.
• Leave-one-out cross-validation confirmed the robustness and effectiveness of the predictive model.

Conclusions:

• This novel method offers a promising solution for predicting infant brain network development from limited baseline data.
• The approach has significant potential for early identification and monitoring of neurodevelopmental trajectories.
• Future work can refine the model for more precise and personalized predictions of brain development.