Figure 1

The placement of phantom reference image and SNR measurement in experiments A: The placement of phantoms on the scan table; B: The representative high-resolution reference image used for assessment; C: Signal intensity of the component; D: were quantitative method of component signal strength for the computation of signal-to-noise ratio.

Table 1

Scanning parameters

Figure 2

Schematic of subjective scores from 1 to 4 Sharpness, distortion, and detail conspicuity was scored as 1 (A), 2 (B), 3 (C), and 4 (D). The minute component enclosed within the circular box in figure A principally serves the purpose of evaluating detail conspicuity and distortion, while the component in the rectangular box primarily focused on evaluating sharpness.

Figure 3

Scanning parameters for subjective images of satisfactory quality Images underwent reconstruction with varying levels of noise reduction (DLR_H, DLR_M, DLR_L) and conventional reconstruction (ConR). The findings revealed that DLR, especially DLR_H, demonstrated a higher SNR, sharpness, and improved detail conspicuity with a reduced NEX and resolution thereby decreasing scan time. A: Images were obtained with a constant resolution (matrix size: 512×512). Notably, the maximum detail conspicuity score remained at 3, when the matrix size held at 512×512. B: Images were obtained employing a constant NEX of 6. To achieve optimal detail conspicuity (Score=4), minimum in-plane resolution was set at 0.38 mm×0.38 mm. DLR_H,DLR_M,DLR_L: deep learning reconstruction of high/medium/low-level of noise reduction, ConR: conventionally reconstructed images; Res: resolution; NEX: number of excitations; SNR: signal-to-noise ratio.

Figure 4

Correlation of SNR, NEX with scan time under a fixed resolution Correlation between SNR and NEX with fixed resolution (A, B), as well as correlation between scan time and NEX (C, D). The solid lines denoted the measured SNR and actual scan time across various reconstruction methods, while the dashed lines delineated the fitted curves, with the corresponding R-squared values shown in the upper right corner (A, C). B and D exhibited the corresponding fitting curve only. DLR_H,DLR_M,DLR_L: deep learning reconstruction of high/medium/low-level of noise reduction, ConR: conventionally reconstructed images; Res: resolution; NEX: number of excitations; SNR: signal-to-noise ratio.

Figure 5

Impact of NEX on the subjective image quality under different reconstruction Subjective evaluation results of images with different NEX and reconstruction methods, including sharpness (A), distortion (B), and detail conspicuity (C). DLR_H,DLR_M,DLR_L: deep learning reconstruction of high/medium/low-level of noise reduction, ConR: conventionally reconstructed images; Res=resolution; NEX: number of excitations; SNR: signal-to-noise ratio.

Figure 6

Correlation of SNR, resolution with scan time under a fixed NEX Correlation between SNR and resolution under NEX (A, B), as well as correlation between scan time and imaging resolution (C, D). The solid lines denoted the measured SNR and actual scan time across various reconstruction methods, while the dashed lines delineated the fitted curves, with the corresponding R-squared values shown in the right corner (A, C). B and D exhibit the corresponding fitting curve only. DLR_H,DLR_M,DLR_L: deep learning reconstruction of high/medium/low-level of noise reduction, ConR: conventionally reconstructed images; Res: resolution; NEX: number of excitations; SNR: signal-to-noise ratio.

Figure 7

Impact of resolution on the subjective image quality under different reconstruction Subjective evaluation results of images with different imaging resolution and reconstruction methods, including sharpness (A), distortion (B), and detail conspicuity (C). DLR_H,DLR_M,DLR_L: deep learning reconstruction of high/medium/low-level of noise reduction, ConR: conventionally reconstructed images; Res: resolution; NEX: number of excitations; SNR: signal-to-noise ratio.