Objective :To explore the application value of the turbo spin echo (TSE) T2-weighted imaging based on deep learning reconstruction (DLR) technology in improving mediastinum image quality and reducing scanning time. Methods: 35 patients with mediastinal lesions were prospectively collected for mediastinal conventional T2WI and DLR-based T2WI scanning. The artifacts, image sharpness, display of the detail of the lesion and overall image quality of the two groups were qualitatively scored by two radiologists independently. The signal-to-noise ratio of the lesion (SNR_lesion) and the contrast noise ratio of the lesion to vertical muscle (CNR_{lesion/muscle}) of the two groups of images were measured and compared. Results Qualitative analysis: T2WI_DL was superior to conventional T2WI in the scores of artifact, image sharpness, display of the detail of the lesion and overall image quality (P<0.001). Quantitative analysis: The SNR and CNR of the image of T2WI_DL sequence were better than those of conventional T2WI (P<0.001). Conclusion Compared with conventional T2WI sequences, T2WI_DL sequence has fewer motion artifacts, better image resolution, focal detail display and overall image quality, and higher SNR and CNR, which has great potential in clinical application of mediastinal diseases.