Automated motion artifact detection in early pediatric diffusion MRI using a convolutional neural network.

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Pediatric Neuroscience

Background:

• Diffusion MRI (dMRI) is crucial for studying brain microstructure.
• Quality control (QC) is vital for dMRI data, especially in infants prone to motion.
• Manual QC is time-consuming, costly, and subjective.

Purpose of the Study:

• To develop and validate a deep learning-based tool for automated motion artifact detection in pediatric dMRI.
• To improve the reliability of dMRI analysis in infants and toddlers.
• To assess the impact of motion artifact removal on DTI and NODDI model fitting.

Main Methods:

• A 3D convolutional neural network (3DCNN) was trained on a pediatric dMRI dataset (1-24 months).
• The model was tested for accuracy (95%) and generalizability on a separate dataset (2-36 months, 98% accuracy).
• dMRI data were processed with and without motion artifact removal for DTI and NODDI analysis.

Main Results:

• The 3DCNN achieved high classification accuracy for motion artifacts in pediatric dMRI.
• The tool demonstrated excellent generalizability across different acquisition parameters and age ranges.
• Removing motion artifacts improved the quality and reliability of DTI and NODDI parameter maps.

Conclusions:

• Deep learning offers an effective solution for automated dMRI quality control in pediatric populations.
• Automated motion artifact detection is essential for accurate brain microstructure analysis in infants.
• This tool can enhance the clinical utility of dMRI in early childhood research and diagnostics.