Identifiability of spatiotemporal tissue perfusion models.

Area of Science:

• Medical Imaging
• Biophysics
• Pharmacokinetics

Background:

• Standard dynamic contrast-enhanced MRI (DCE-MRI) models for perfusion quantification introduce bias by treating voxels as isolated systems.
• Spatiotemporal models offer a potential solution to this bias, but their fundamental identifiability remains unproven.

Purpose of the Study:

• To investigate the identifiability of spatiotemporal perfusion models in DCE-MRI.
• To compare the identifiability of one-compartment and two-compartment models using in silico simulations.

Main Methods:

• Theoretical analysis and in silico simulations were performed on 1D toy systems.
• One-compartment and two-compartment models were simulated with varying noise and temporal undersampling.
• Model parameters were fitted using gradient descent, with iterative time window increases and repeated initial value testing.

Main Results:

• Theoretical analysis revealed flow and volume fraction identifiability up to a constant, resolvable by parameter choice.
• One-compartment models demonstrated accurate and robust reconstruction of tissue concentrations, blood velocities, and arterial input functions.
• Two-compartment models exhibited non-identifiability for perfusion parameters and venous velocities, with high sensitivity to measurement error.

Conclusions:

• One-compartment spatiotemporal DCE-MRI models are supported for perfusion quantification due to their identifiability.
• Two-compartment perfusion models require further investigation to resolve identifiability issues in more complex systems or with optimized parameters.