Recently, novel MRI reconstruction techniques based on deep learning (deep learning reconstruction: DLR) have been introduced, revolutionizing the trade-off between reducing scan time and maintaining image quality [1,2]. The first stage of super-resolution deep learning-based reconstruction (SR-DLR) acquires high-SNR images with denoising. The second stage of super-resolution processing generates high-resolution images by zero-fill interpolation processing (ZIP) of the acquired high-SNR images. Super-resolution processing is performed by learning and constructing a preset of high-resolution input images after ZIP processing and high-resolution teacher images that are fully sampled by actual collection (Figure 1). This process produces images with improved spatial resolution and SNR.

The purpose of this study is as follows:

• To compare SR-DLR images with conventional reconstructed (C-R) images and to

              demonstrate the usefulness of SR-DLR.

• To evaluate whether SR-DLR can reduce scan time while ensuring the image quality of C-R.