Evaluation of vendor-neutral deep learning-based image reconstruction software for preclinical mouse brain MRI

Motivation: Is deep learning-based reconstruction (DLR) software useful for preclinical MRI? Goal(s): To investigate whether DLR software trained on clinical MR images of humans is applicable to preclinical MRI of mouse brain. Approach: To evaluate the original T1WI and T2WI of mouse brain obtained with varying averaging, as well as images at mild, moderate, and strong denoising levels using DLR software by the following quantitative metrics; signal-to-noise ratio, contrast-to-noise ratio, structural similarity index, and sharpness index. Results: Almost all of the quantitative metrics showed a simple increase with the rise in denoising level. Impact: The vendor-neutral deep learning-based reconstruction software trained only using clinical human MR images demonstrated excellent image quality improvement in mouse brain MR images which are preclinical images not included in the training dataset.