C-R image and the SR-DLR image are shown in Figures 3 and 4. The results of image evaluation are shown in Figure 5. For all sequences, overall image quality, noise, and sharpness scores were significantly higher for the SR-DLR image than for the C-R image (p<0.05). Even for clinical images, a high-resolution, low-noise image can be obtained from a low-resolution, high-noise image, demonstrating the usefulness of SR-DLR. There was no significant difference in scores for T2 and FLAIR for contrast. High-resolution, low-noise images can be acquired without contrast change. There was no significant difference in scores for DWI, T2, and T2* for artifact. The FLAIR score for artifacts was significantly higher for the SR-DLR image than for the C-R image (p<0.05). This was due to the reduction of truncation artifacts by the SR-DLR (Figure 6). This is because the truncation artifact, which is a problem of ZIP, can be reduced by learning using a high-resolution teacher image fully sampled from the actual collection and a ZIP image created from the teacher image during the construction of the SR-DLR (Figure 7).

The scan time with C-R and SR-DLR is shown in Figure 8. Images were acquired in a short time by lowering the matrix in the phase encoding direction and increasing the parallel imaging reduction factor. Image quality was ensured by SR-DLR in each sequence. The total scan time for brain MRI (DWI, T2, FLAIR, T2*) was 2:25 sec (Figure 9). The total scan time could be reduced by 61%. Compared to the same month last year, the number of examinations increased by up to 23% (Figure 10). The increase in the number of exams also has significant management benefit.