Evaluation of linear diaphragm-chest expansion models for magnetic resonance imaging motion artifact correction

The efficacy of Fourier analysis, autoregressive with exogenous input (ARX) and adaptive models to estimate diaphragm position from respiratory belt signal (a measure of chest expansion) was evaluated for the purpose of correcting respiratory motion artifacts in magnetic resonance imaging (MRI). Respiratory belt signal and diaphragm position data were obtained simultaneously during one-dimensional MRI scans with sampling intervals of 100 ms for 128 s (1280 samples). The models were trained using the first 512 data samples for the Fourier method and the first 640 samples for the ARX and adaptive methods. The remaining samples were used as a test set for evaluating the models. Both ARX and adaptive methods produced more accurate results than the Fourier method as reflected by the normalized mean square error (NMSE) and correlation coefficient (R) between the estimated and actual diaphragm position during normal breathing (P < 0.05). However, all three models had difficulty modeling diaphragm positions during breathing plateaus.