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

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Computational Anatomy

Background:

• Liver motion during respiration poses challenges for accurate medical imaging.
• Previous liver motion models relied on spatial registration of dynamic MRI data.
• Enhancements are needed to improve predictive accuracy and robustness.

Purpose of the Study:

• To extend an existing liver motion model with temporal prediction and respiratory signal integration.
• To evaluate the performance improvements and limitations of these extensions.
• To assess the model's clinical applicability for dynamic MRI-guided interventions.

Main Methods:

• Dynamic MRI data from four volunteers were acquired with respiratory monitoring.
• The liver motion model was implemented using spatial prediction and temporal forward prediction (300-1200 ms) with an extended Kalman filter.
• Model performance was evaluated based on liver overlap (Dice coefficient), surface distance, and vessel misalignment.

Main Results:

• Temporal prediction led to a slight decrease in liver overlap (0.85%) and increased surface distance (20.6%) and vessel misalignment (20%).
• Mean vessel misalignment ranged from 2.9 mm (original) to 4.01 mm (1200 ms prediction).
• The respiratory signal input was feasible, and prediction errors from sudden breathing changes were transient and auto-correcting.

Conclusions:

• The extended liver motion model demonstrates acceptable accuracy despite minor prediction decreases.
• Integration of respiratory signals is feasible and beneficial for liver motion modeling.
• The model's ability to self-correct transient errors suggests potential clinical utility in dynamic MRI applications.