Longitudinal Prediction of Infant Diffusion MRI Data via Graph Convolutional Adversarial Networks

Area of Science:

• Medical Imaging
• Neuroscience
• Artificial Intelligence

Background:

• Missing data is a significant challenge in longitudinal studies, particularly in diffusion MRI (dMRI) due to subject dropouts and scan failures.
• Accurate imputation of missing dMRI data is crucial for reliable longitudinal analysis of brain development and disease progression.

Purpose of the Study:

• To develop and evaluate a novel graph-based convolutional neural network (GCNN) for predicting missing dMRI data in longitudinal infant brain studies.
• To improve the accuracy and perceptual quality of data imputation compared to existing methods.

Main Methods:

• Constructed a graph by considering relationships between spatial sampling points and diffusion wave-vector domains.
• Employed a GCNN with a multi-scale residual architecture and adversarial learning for non-linear data mapping.
• Validated the method on longitudinal infant brain dMRI datasets.

Main Results:

• The proposed GCNN method demonstrated high accuracy and robustness in predicting missing dMRI data.
• The approach achieved superior perceptual quality in the imputed data.
• Experimental results confirmed the method's effectiveness for longitudinal infant brain data.

Conclusions:

• The graph-based convolutional neural network offers a powerful solution for handling missing data in longitudinal dMRI studies.
• This method can significantly enhance the reliability and quality of neuroimaging research, especially in vulnerable populations like infants.
• The GCNN approach provides a promising tool for advancing the analysis of brain development and neurological conditions.