Joint Sparse and Low-Rank Regularized MultiTask Multi-Linear Regression for Prediction of Infant Brain Development...

Area of Science:

• Neuroscience
• Developmental Biology
• Medical Imaging

Background:

• Dynamic infant brain development research requires complete longitudinal datasets for accurate trajectory charting.
• Missing data in longitudinal studies poses a significant challenge for precise developmental analysis.

Purpose of the Study:

• To propose a novel method for predicting future infant brain development scores using early longitudinal imaging data, accommodating missing data.
• To address the challenge of incomplete longitudinal datasets in charting infant brain development.

Main Methods:

• A multi-dimensional regression approach is employed, treating the prediction as a multi-task, multi-linear problem.
• An objective function with joint ℓ1 and low-rank regularization on the mapping weight tensor is proposed for feature selection and structural information preservation.
• Features are extracted from longitudinal imaging data using a bag-of-words model.

Main Results:

• The proposed method effectively predicts infant brain development scores at four years of age.
• Accurate predictions are achieved using imaging data from as early as two years of age.
• The method demonstrates efficacy in handling missing data within longitudinal datasets.

Conclusions:

• The developed method offers a robust solution for predicting infant brain development trajectories despite missing longitudinal data.
• This approach enhances the ability to chart early brain development using neuroimaging, with implications for early identification and intervention.
• The findings highlight the potential of advanced regression techniques in pediatric neuroscience research.