Subject-specific liver motion modeling in MRI: a feasibility study on spatiotemporal prediction

• Published in: Physics in medicine & biology Vol. 62; no. 7; pp. 2581 - 2597
• Main Authors: Noorda, Yolanda H, Bartels, Lambertus W, Viergever, Max A, Pluim, Josien P W
• Format: Journal Article
• Language: English
• Published: England IOP Publishing 07.04.2017
• Subjects: dynamic data; Feasibility Studies; Humans; Kalman filtering; Liver - diagnostic imaging; liver motion model; Magnetic Resonance Imaging - methods; Models, Theoretical; Movement - physiology; MR-HIFU; registration; Respiration; Spatio-Temporal Analysis; temporal prediction
• ISSN: 0031-9155; 1361-6560; 1361-6560
• DOI: 10.1088/1361-6560/aa5e96

A liver motion model based on registration of dynamic MRI data, as previously proposed by the authors, was extended with temporal prediction and respiratory signal data. The potential improvements of these extensions with respect to the original model were investigated. Additional evaluations were performed to investigate the limitations of the model regarding temporal prediction and extreme breathing motion. Data were acquired of four volunteers, with breathing instructions and a respiratory belt. The model was built from these data using spatial prediction only and using temporal forward prediction of 300 ms to 1200 ms, using the extended Kalman filter. From temporal prediction of 0 ms to 1200 ms ahead, the Dice coefficient of liver overlap decreased with 0.85%, the median liver surface distance increased with 20.6% and the vessel misalignment increased with 20%. The mean vessel misalignment was 2.9 mm for the original method, 3.42 mm for spatial prediction with a respiratory signal and 4.01 mm for prediction of 1200 ms ahead with a respiratory signal. Although the extension of the model to temporal prediction yields a decreased prediction accuracy, the results are still acceptable. The use of the breathing signal as input to the model is feasible. Sudden changes in the breathing pattern can yield large errors. However, these errors only persist during a short time interval, after which they can be corrected automatically. Therefore, this model could be useful in a clinical setting.