Automated detection of motion artifacts in brain MR images using deep learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Quality assessment is crucial for MRI data integrity.
• Motion artifacts significantly degrade image quality and analysis.
• Manual quality assessment is time-consuming and requires expertise.

Purpose of the Study:

• To develop and validate a deep learning model for automated detection of rigid motion in T1-weighted brain MRI.
• To assess the model's performance on retrospective and prospective datasets.
• To interpret model failure modes and its correlation with image quality metrics.

Main Methods:

• A 2D convolutional neural network (CNN) was trained on synthesized motion data for three-class classification.
• The model was evaluated on six retrospective, motion-simulated datasets and one prospective dataset.
• Grad-CAM heatmaps were used for model interpretability and failure mode analysis.

Main Results:

• The model achieved 85% average precision and 80% recall on retrospective datasets.
• Classifications on the prospective dataset showed 93% agreement with radiologist labeling.
• A strong inverse correlation (-0.84) was found between model output and average edge strength, an image quality metric.

Conclusions:

• The deep learning model effectively detects rigid motion artifacts in brain MRI.
• This automated approach can accelerate quality assessment and support on-site expertise.
• The model is particularly valuable for low-resource settings with limited local MR expertise.