Denoising preclinical MRI with vendor-neutral deep learning-based image reconstruction

Area of Science:

• Biomedical Imaging
• Machine Learning in Radiology
• Preclinical Research

Background:

• Deep learning-based image reconstruction (DLR) is a key technology in clinical MRI.
• Vendor-neutral DLR software development is advancing rapidly.
• Assessing DLR applicability to animal experimental MR images is valuable.

Purpose of the Study:

• Evaluate a vendor-neutral DLR method for denoising preclinical mouse brain MR images.
• Determine the effectiveness of DLR trained on human data for non-human animal data.
• Quantify the impact of DLR on image quality metrics.

Main Methods:

• A vendor-neutral DLR method, trained on human MR images, was applied to preclinical mouse brain MR images.
• Six FVB mice were scanned using a 4.7T MRI system.
• Image quality was assessed using signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), and sharpness index at varying denoising levels.

Main Results:

• DLR significantly reduced noise, improving the recognizability of anatomical structures.
• Noise reduction did not compromise anatomical details; fine structures showed enhanced clarity.
• SNR and CNR increased with higher denoising levels, and sharpness improved, especially at mild denoising.

Conclusions:

• Vendor-neutral DLR substantially denoised preclinical mouse brain MR images.
• The DLR method demonstrated effectiveness in enhancing preclinical MRI quality.
• This technology shows promise for improving animal experimental MR imaging.