〔Abstract〕 Objective To explore the value of deep learning reconstruction (DLR) in liver magnetic resonance imaging (MRI) by comparing the single-shot half-fourier rapid spin-echo sequence with DLR (HASTE DLR) and diffusion-weighted imaging sequence with DLR (DWI DLR) against the conventional BLADE and conventional DWI sequences. Methods70 patients underwent MRI examinations. Two observers independently evaluated the image quality of each sequence (including liver edge, blood vessels, lesion clarity, etc.). Additionally, quantitative evaluation was conducted by measuring the signal to noise ratio (SNR) of liver parenchyma and lesions, contrast to noise ratio (CNR) of lesions, as well as apparent diffusion coefficient (ADC) values from conventional DWI and DWI DLR. Intraclass correlation coefficient (ICC) was used to evaluate the consistency between the two observers. ResultsThe inter-observer consistency was high (ICC: 0.84~0.97). The scanning time was reduced by 92.63% for HASTE DLR and 50% for DWI DLR sequences, respectively. The lesion clarity score of the HASTE DLR group was significantly better than that of the BLADE group (P < 0.001), with artifacts reduced in both DLR sequences ( P < 0.05). The HASTE DLR group demonstrated higher SNR and CNR, while the DWI DLR group showed higher SNR and ADC values (all P < 0.05). ConclusionThe DLR technology can enhance the efficiency of liver MRI scans, improve image quality, and reduce artifacts, demonstrating promising application prospects.