Frequency and Phase Correction of MRS Data by Leveraging Sequential Patterns along the Transient Dimension

Motivation: Freqeuncy and phase correction of MRS data typically works on one transient at a time. However, by processing all transients simultaneously, the FPC algorithm can account for systematic variations in the transients that occurs due to certain factors like scanner B0 drift. Goal(s): The goal is to propose a deep learning method can directly on in vivo data without the need for pre-training on simulated data. Approach: 2D-FPC is unsupervised learning approach based on a CNN-LSTM that takes in all transients at once. Results: 2D-FPC showed strong performance on simulated and in vivo data. The method is robust against addtional noise and offsets. Impact: The 2D-FPC method proposed follows an unsupervised learning approach that allows it to be directly applied in vivo dataset without the need for pre-training. The simultaneous processing of all transients allows the algorithm to capture long-range patterns in the data.