1349296 - High-Resolution MR Fingerprinting for T1 and T2 Mapping of Both Left and Right Ventricles

CMR T₁ and T₂ mapping is often limited to low spatial resolutions not suitable for imaging the thin right ventricular (RV) wall. While Magnetic Resonance Fingerprinting (MRF) (1) allows for simultaneous T₁-T₂ mapping, prior work has focused on the left ventricle (LV). The purpose of this study is to enable T₁-T₂ mapping of both ventricles in a breathhold by combining a high-resolution MRF acquisition with deep learning.


Methods:

Data were collected using a spiral FISP-MRF sequence in a 15-heartbeat breathhold with ECG triggering (2). Two variants were tested with lower (1.6x1.6x8mm³) and higher (1.2x1.2x5mm³) resolutions (Table 1). Spectral fat saturation was applied before each scan window to improve delineation of the RV. To mitigate the decreased SNR at higher resolution, two reconstructions were compared—a low-rank subspace method with locally-low rank regularization (SLLR) (3) and a deep image prior (DIP) reconstruction (4), which uses a u-net to output maps consistent with the acquired k-space data without additional training data. Six healthy subjects were scanned at 1.5T (Sola, MAGNETOM Siemens) using lower and higher resolution MRF in a medial slice. MOLLI and T₂-prepared bSSFP were collected at the same resolutions for comparison. The LV septum and RV myocardium were segmented, and mean T₁ and T₂ values were compared among methods.

Results:

Figure 1 shows maps from one subject. High-resolution MRF provided improved delineation of the RV, pericardium, papillary muscles, and trabeculations compared to the lower resolution scan, with the DIP reconstruction yielding less noise than SLLR. Relaxation time measurements are given in Figure 2. Lower-resolution MRF yielded similar T₁ in both ventricles (LV 1051±17 / RV 1051±33ms) but higher T₂ in the RV (LV 38.4±3.0 / RV 42.0±4.0ms). Both ventricles had similar T₁ (LV 1082±32 / RV 1083±28ms) and T₂ (42.5±3.2 / RV 43.0±3.9) using high-resolution MRF. MOLLI had significantly lower T₁ than MRF and T₂-prep bSSFP had significantly higher T₂ than MRF at both resolutions and in both ventricles. MOLLI and T₂-prep bSSFP yielded higher T₁ and T₂ at higher resolution compared to lower resolution, likely due to motion and partial volume artifacts arising from the longer (320 vs 215ms) scan window. Among higher resolution scans, MRF had lower intrasubject SD in both ventricles (LV/RV T₁ 56/90ms, T₂ 3.8/6.4ms) than conventional mapping (LV/RV T₁ 94/141ms, T₂ 8.7/12.3ms).

Conclusion:

Simultaneous T₁ and T₂ mapping of the LV and RV was demonstrated using high-resolution MRF with deep learning. While Look-Locker RV T₁ mapping with respiratory navigation has been developed (5), the proposed approach maps both T₁ and T₂ and can be performed in a breathhold. This technique has clinical implications for cardiomyopathies with RV involvement. Furthermore, reduced partial volume effects may facilitate easier segmentation and improve measurement precision. Future work will include incorporation of off-resonance deblurring and validation in patients.