Iterative denoising accelerated 3D SPACE FLAIR sequence for brain MR imaging at 3T

Abstract

Listen

Translate article

PurposeThe purpose of this study was to prospectively evaluate image quality of three-dimensional fluid attenuated inversion recovery (3D-FLAIR) sequence acquired with a high acceleration factor and reconstructed with iterative denoising (ID) for brain magnetic resonance imaging (MRI) at 3-T. Material and methodsPatients with brain tumor who underwent brain MRI were consecutively included. Two 3D-FLAIR sequences were successively performed for each patient. A first conventional FLAIR acquisition (conv-FLAIR) was performed with an acceleration factor of 6. The second acquisition was performed with an increased acceleration factor of 9. Two series one without ID (acc-FLAIR) and one with ID (acc-FLAIR-ID) were reconstructed. Two neuroradiologists independently assessed image quality, deep brain nuclei visualization and white matter/gray matter (WM/GM) differentiation on a 4-point scale. ResultsThirty patients with brain tumor were consecutively included in this study. There were 16 women and 14 men with a mean age of 54 ± 17 (SD) years (range: 22–78 years). Scanning time of Acc-FLAIR-ID and Acc-FLAIR (4 min 40 sec) was 37% shorter than that of conv-FLAIR (2 min 50 sec) (P < 0.01). Improved image quality score was significantly different for both conv-FLAIR and acc-FLAIR-ID compared to acc-FLAIR (P < 0.01 for both). WM/GM differentiation score of conv-FLAIR was not significantly different compared to acc-FLAIR-ID (P = 0.10). Improved WM/GM differentiation score was different for both sequences compared to acc-FLAIR (P = 0.017 and P < 0.001). Deep brain nuclei visualization score was not different between conv-FLAIR and acc-FLAIR-ID (P = 0.71). However, the improved deep brain nuclei visualization score was significantly different for both sequences compared to acc-FLAIR (P < 0.001 for both). ConclusionScanning time of 3D-FLAIR sequence using a high acceleration factor reconstructed with ID algorithm can be reduced by 37% while preserving image quality for brain MRI.