3D deep-learning image reconstruction for fast spin-echo triple-echo Dixon images acquired with flexible echo-spacing (FTED-Flex)

Motivation: The fast spin-echo triple-echo Dixon images acquired with flexible echo-spacing (FTED-Flex) can be used to generate separated water and fat images with enhanced T2-weighted contrast. However, their performance and clinical applications are limited by its long image reconstruction time. Goal(s): Our goal is to develop a fast and accurate FTED-Flex image reconstruction method. Approach: The time-consuming phase estimation was replaced by a 3D deep-learning neural network. Results: The FTED-Flex integrated with a 3D deep-learning network was highly accurate (average Dice coefficient in volume-of-interest=0.989) and reduced the processing time for phase-estimation to a few seconds, compared to tens of minutes by conventional methods. Impact: The developed FTED-Flex integrated with a 3D deep-learning network is highly accurate and reduces the processing time for phase-estimation to a few seconds, thus it has a great potential to expand clinical applications of FTED-Flex imaging.