Quantitative Ischemic Lesions of Portable Low-Field Strength MRI Using Deep Learning-Based Super-Resolution.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Neurology

Background:

• Low-field strength magnetic resonance imaging (LF-MRI) offers portability but has limitations in quantitative lesion assessment.
• Deep learning-based synthetic super-resolution magnetic resonance imaging (SynthMRI) is a novel approach to enhance LF-MRI image quality.
• Improving diagnostic performance in ischemic stroke detection is crucial for timely treatment.

Purpose of the Study:

• To evaluate the diagnostic performance of SynthMRI in assessing ischemic lesions.
• To compare SynthMRI's accuracy against standard LF-MRI and high-field MRI (HF-MRI).
• To assess SynthMRI's effectiveness in quantifying lesion volume and apparent diffusion coefficient (ADC).

Main Methods:

• Retrospective analysis of 178 stroke patients and 104 controls with LF-MRI and HF-MRI data.
• Utilized a deep learning framework (SCUNet) pretrained on large datasets to generate SynthMRI from LF-MRI.
• Evaluated sensitivity, specificity, Alberta Stroke Program Early Computed Tomography Score (ASPECTS) agreement, lesion volume, and ADC agreement with HF-MRI as ground truth.

Main Results:

• SynthMRI demonstrated significantly higher sensitivity (89.0%) and specificity (91.3%) than LF-MRI (77.1% and 71.0%, respectively).
• SynthMRI showed superior agreement with HF-MRI for ASPECTS, lesion volume, and ADC values (ICCs > 0.85, r > 0.78).
• Improved agreement with HF-MRI was noted for SynthMRI, particularly for lesions in the early hyperacute and subacute stroke phases.

Conclusions:

• SynthMRI significantly enhances the diagnostic performance of LF-MRI for ischemic lesions.
• The method provides high agreement with HF-MRI in lesion detection and quantification.
• SynthMRI is particularly valuable for assessing early-stage ischemic lesions.