Optimizing Curve-Fitting Models for T1-Mapping Via Inversion Recovery with 3 Tesla MRI
Abstract
Purpose
To optimize a curve-fitting model for quantitative T1-mapping via inversion recovery (IR) sequences considering varying numbers of data points. For clinical applications, higher resolution and shorter scan times are desired.
Methods
An International Society of Magnetic Resonance in Medicine/National Institute of Standards and Technology (ISMRM/NIST) phantom was scanned on a 3T MRI scanner. Three sets of scans were analyzed: (1) a high-resolution sequence with five inversion times (TIs) ranging from 500-1700 ms, (2) the calibration sequence provided by the manufacturer with ten TIs ranging from 35-3000 ms, (3) a subset of five TIs from the calibration sequence. All sets of scans were analyzed using an in-house Python workflow to calculate the T1 values in eight vials in the phantom according to a two-parameter model and a three-parameter model. These values were compared to the true values provided by the manufacturer.
Results
For set (1), the mean absolute percent differences (MAPDs) were 16.4% and 1.90% for the three-parameter and two-parameter models respectively. For set (2), the MAPDs were 0.129% and 2.94% for the three-parameter and two-parameter models respectively. For set (3), the MAPDs were 7.78% and 2.40% for the three-parameter and two-parameter models respectively.
Conclusion
Sets (1) and (3) show similar patterns, indicating that the differences in accuracy are not due to differences in resolution. While the two-parameter model assumes perfect inversion, it performs reasonably well even with limited data points. The three-parameter model is more flexible as it does not assume perfect inversion and performs exceptionally well when more TIs are used. However, it is prone to overfitting when there are too few data points. Hence, if fewer TIs are used to reduce the scan time, this two-parameter is most appropriate for quantitative T1-mapping. This variability in accuracy highlights the importance of appropriate pulse sequence optimization for clinical T1-mapping.