Introduction

Cardiac amyloidosis (CA) is characterized by the accumulation of misfolded amyloid proteins in the myocardium that impair cardiac function. Quantitative T₁ and T₂ values can be used to detect diseased tissue, potentially expediting diagnosis and treatment. Cardiac Magnetic Resonance Fingerprinting (cMRF) simultaneously quantifies myocardial T₁ and T₂ in a single breath-hold and has the potential to streamline cardiac MRI protocols while providing improved tissue characterization through analysis of signal evolution data. The use of cMRF for characterization of CA was evaluated as part of an ongoing study.

Methods

Six patients with CA (age: 62±12) and four healthy controls (age: 33±3) were scanned on a 3T Siemens Prisma scanner. Three cardiac short axis slices were acquired with cMRF for assessment of native left ventricular myocardium T₁ and T₂ values. The cMRF sequence has been reported previously¹ and consists of a 15-heartbeat, ECG gated (end diastole), single breath-held acquisition with spiral readout and spatial resolution of 1.6x1.6x8 mm³ at 300 mm FOV. Nonuniform fast Fourier transform² was used for gridding of undersampled images for T₁ and T₂ quantification and signal analysis. A dictionary of T₁ values ranging from 10 to 2000 ms in 10 ms increments, 2050 to 3000 in 50 ms increments, and T₂ values ranging from 4 to 80 in 2 ms increments, 85 to 120 in 4 ms increments, and 130 to 300 in 10 ms increments was generated. The dictionary included corrections for slice profile and preparation pulse efficiency.³ T₁ and T₂ maps were computed by inner product matching between the measured signal evolutions and the dictionary.⁴ Average segmental values were obtained in manually drawn regions of interest according to the American Heart Association 16-segment model. Segmental values were averaged to obtain a surrogate measure of global myocardial T₁ and T₂. Segmental and global values were compared between healthy controls and patients using unpaired, two-sided t-tests. In order to investigate the potential for additional information present in cMRF signal evolutions as compared to T₁ and T₂ values, data were analyzed using linear discriminant analysis (LDA) and calculation of the Fisher coefficient. The Fisher coefficient has similarly been used to compare feature sets for classification tasks such as in MRI texture analysis.⁵ For this analysis, large regions of interest were drawn in the mid-ventricular septal myocardium to obtain T₁ and T₂ values as well as signal evolutions. Complex-valued signal evolutions were averaged within ROIs to reduce noise, and the absolute-valued signal evolution was recorded for each subject. Regularized LDA was used to account for small sample size in the signal evolution analysis. Separability of the amyloid and control classes was evaluated for both relaxometric data-based LDA and signal-based LDA using the Fisher coefficient, and the corresponding LDA scores were analyzed. LDA scores were compared between healthy controls and patients using unpaired, two-sided t-tests.

Results

Representative T₁ and T₂ maps are shown in Figure 1. Global native T₁ and T₂ values are shown in Figure 2. Native myocardial T₁ was elevated in patients relative to healthy subjects by cMRF (1445±91 vs 1315±47 ms, p<0.05). Native myocardial T₂ was also significantly elevated in patients versus healthy subjects with cMRF (40.4±4.1 vs 34.7±2.7 ms, p<0.05). At a segmental level (Figure 3), elevated T₁ and T₂ was observed in all segments, but with differing degrees of significance. Averaged cMRF signal evolutions for the patient group and control group are shown in Figure 4 with observable differences at various points in the time course. LDA performed on relaxometric data (T₁, T₂) and signal evolutions both showed a statistically significant separation in amyloid patients as compared to controls in terms of LDA scores (p<0.01). The Fisher coefficient was larger for the signal-based analysis as compared to the relaxometric analysis (24.1 vs 3.1, respectively), suggesting greater separability between the two groups.

Discussion

Elevated T₁ and T₂ in the patient group were consistently observed, suggesting a significant difference in the tissue properties of the patient group relative to the controls. Although analysis of T₁ and T₂ values independently provided discrimination of patients from controls, some overlap of these groups was observed. Analysis of signal evolutions with LDA suggests improved discrimination of the groups, as the Fisher coefficient was greater in the signal-based LDA and a greater spread in LDA scores between groups was observed. These observations suggest that there may be additional information in cMRF signal evolutions that is not captured by T₁ and T₂ quantification alone. However, given the limited sample size, age difference in populations, and potential overfitting or confounding of the signal-based LDA despite the use of regularized LDA, continued study of cMRF for characterization of CA is needed.

Conclusion

From these preliminary results, cMRF was found to detect elevated native myocardial T₁ and T₂ in patients with diagnosed CA. Analysis of magnetic signal evolutions obtained from cMRF may improve discrimination of diseased myocardium from healthy tissue. Continued evaluation of cMRF for characterization of cardiac amyloidosis is warranted.

Acknowledgements

This work was funded in part by the following sources: 2R01HL094557 and the Cleveland Clinic Imaging Institute Pilot Projects Program. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH or the Cleveland Clinic.