Development and Evaluation of Deep Learning-Accelerated Single-Breath-Hold Abdominal HASTE at 3 T Using Variable Refocusing Flip Angles

Objective Deep learning (DL) reconstruction enables substantial acceleration of image acquisition while maintaining diagnostic image quality. The aims of this study were to overcome the drawback of specific absorption rate (SAR)-related limitations at 3 T and to develop a DL-accelerated single-breath-hold half-Fourier acquisition single-shot turbo spin echo (HASTE) sequence for 2-dimesional T2-weighted fat-suppressed magnetic resonance imaging of the abdomen at 3 T using a variable flip angle (FA) evolution for the refocusing radiofrequency pulses, as well as to evaluate its feasibility and image quality in comparison to state-of-the-art T2-weighted fat-suppressed imaging technique (BLADE). Materials and Methods First, a suitable FA evolution with low cardiac motion-related signal loss (CRSL) and low SAR was determined through a prospective volunteer study with 11 participants. Image quality and diagnostic confidence with 5 different FA evolutions of a HASTE(DL) were assessed to identify the most suitable FA evolution. Second, the identified FA evolution was implemented clinically and evaluated in 51 patients undergoing a clinically indicated liver magnetic resonance imaging at 3 T. Two radiologists assessed the HASTE(DL) and standard sequences regarding overall image quality, noise, contrast, sharpness, artifacts, CRSL, and diagnostic confidence using a Likert scale ranging from 1 to 4, with 4 being the best. Comparative analyses were conducted to assess the differences between HASTE(DL) (acquisition time, 21 seconds; single breath-hold) and the routinely used T2-weighted BLADE sequence (acquisition time, 4 minutes; respiratory triggering). Results From the volunteer study, the FA evolution characterized by the control points 130-90-110-130 degrees (HASTE(DL)) was identified as optimal among the 5 evolutions evaluated and was implemented in our clinical protocol. In all 51 patients, HASTE(DL) was successfully acquired at 3 T and showed excellent image quality (median, 4; interquartile range, 3-4). Although BLADE was rated significantly higher for overall image quality, noise, contrast, sharpness, artifacts, CRSL, and diagnostic confidence than HASTE(DL), no differences were found concerning the number (n = 102) and measured diameter of the detected hepatic lesions between the 2 sequences BLADE and HASTE(DL). Conclusions The proposed single-breath-hold abdominal HASTE(DL) with variable refocusing FAs is feasible at 3 T within SAR limits and yields high image quality and diagnostic confidence as compared with a standard T2-weighted acquisition technique, at a 10th of the acquisition time.