Motion-Informed 3D Deep Learning Reconstruction in Patients with Cognitive Impairment

Area of Science:

• Neuroimaging
• Artificial Intelligence in Medical Imaging
• Radiology

Background:

• Motion artifacts are a major challenge in 3D brain MRI, particularly for patients with cognitive impairments.
• Existing deep learning (DL) methods often enhance signal-to-noise ratio but lack explicit motion correction capabilities.
• This study addresses the need for improved MRI reconstruction in motion-prone individuals.

Purpose of the Study:

• To validate a novel DL reconstruction method incorporating retrospective motion correction for 3D T1-weighted brain MRI.
• To assess the impact of motion-informed DL reconstruction on morphometric accuracy and image quality.
• To evaluate the potential of this technique in clinical settings with patients experiencing memory loss.

Main Methods:

• Prospective, intraindividual comparison study involving healthy volunteers and patients with memory loss.
• Acquisition of 4-fold undersampled 3D MPRAGE with integrated scout accelerated motion estimation and reduction (SAMER).
• Reconstruction using standard methods versus the proposed DL approach; quantitative morphometric accuracy and qualitative image quality assessments by neuroradiologists.

Main Results:

• The DL method significantly reduced segmentation error under moderate (12.4% to 3.5%) and severe (44.2% to 12.5%) motion (P < .001).
• Visual image quality scores were significantly improved with the DL approach (4.26 vs. 3.59, P < .001).
• Motion artifact severity was reduced in 47% of cases, with moderate to substantial interreader agreement.

Conclusions:

• Motion-informed DL reconstruction enhances both morphometric accuracy and perceived image quality in 3D T1-weighted brain MRI.
• This technique shows promise for improving diagnostic utility in motion-prone patients, including those with cognitive impairment.
• The method may lead to reduced scan failure rates in challenging patient populations.