Characterization of Cardiac Amyloidosis using Cardiac Magnetic Resonance Fingerprinting

Abstract

Background:

Cardiac amyloidosis (CA) is an infiltrative cardiomyopathy with poor prognosis absent appropriate treatment. Elevated native myocardial T₁ and T₂ have been reported for CA, and tissue characterization by cardiac MRI may expedite diagnosis and treatment. Cardiac Magnetic Resonance Fingerprinting (cMRF) has the potential to enable tissue characterization for CA through rapid, simultaneous T₁ and T₂ mapping. Furthermore, cMRF signal timecourses may provide additional information beyond myocardial T₁ and T₂.

Methods:

Nine CA patients and five controls were scanned at 3T using a prospectively gated cMRF acquisition. Two cMRF-based analysis approaches were examined: (1) relaxometric-based linear discriminant analysis (LDA) using native T₁ and T₂, and (2) signal timecourse-based LDA. The Fisher coefficient was used to compare the separability of patient and control groups from both approaches. Leave-two-out cross-validation was employed to evaluate the classification error rates of both approaches.

Results:

Elevated myocardial T₁ and T₂ was observed in patients vs controls (T₁ 1395±121 vs 1240±36.4 ms, p<0.05; T₂: 36.8±3.3 vs 31.8±2.6 ms, p<0.05). LDA scores were elevated in patients for relaxometric-based LDA (0.56±0.28 vs 0.18±0.13, p<0.05) and timecourse-based LDA (0.97±0.02 vs 0.02±0.02, p<0.05). The Fisher coefficient was greater for timecourse-based LDA (60.8) vs relaxometric-based LDA (1.6). Classification error rates were lower for timecourse-based LDA vs relaxometric-based LDA (12.6±24.3 vs 22.5±30.1 %, p<0.05).

Conclusions:

These findings suggest that cMRF may be a valuable technique for the detection and characterization of CA. Analysis of cMRF signal timecourse data may improve tissue characterization as compared to analysis of native T₁ and T₂ alone.

Introduction

Cardiac amyloidosis (CA) is characterized by the accumulation of misfolded amyloid proteins in the myocardium that impair cardiac function. Myocardial T₁ and T₂ mapping can be used to characterize diseased tissue, potentially expediting diagnosis and treatment; native, noncontrast myocardial T₁ is elevated in CA¹, and native myocardial T₂ has been reported to provide discrimination of CA subtypes². Cardiac Magnetic Resonance Fingerprinting (cMRF) enables simultaneous myocardial T₁ and T₂ quantification in a single breath-held scan^3,4 and has potential to streamline protocols. Furthermore, recent “MRF residual” analysis suggests that signal timecourses may be influenced by tissue microstructural variations beyond relaxometric parameters with an example of magnetization transfer⁵, a property known to differ between healthy and fibrotic myocardium⁶. Similarly, diffusion can also influence MRF signals⁷ and diffusion tensor cardiac MRI has been reported to characterize amyloid infiltration in the myocardium⁸. Thus, a number of underlying tissue properties may affect cMRF signal timecourses beyond native myocardial T₁ and T₂ that could support disease assessment. The proof-of-principle use of cMRF for CA characterization was evaluated as part of an ongoing study.

Methods

Nine CA patients (age: 67±13) and five healthy controls (age: 37±11) were scanned on a 3T Prisma scanner (Siemens Healthcare, Erlangen, Germany) under an IRB-approved protocol that conforms to the ethical guidelines of the 1975 Declaration of Helsinki and includes written informed consent for each subject. Three cardiac short axis slices were acquired with cMRF to assess native left ventricular myocardium T₁ and T₂. The cMRF sequence consisted of a 15-heartbeat, gated (ECG or pulse, end-diastole), breath-held acquisition with spiral readout, 1.6x1.6x8 mm³ resolution, and 300x300 mm² field-of-view.³ T₁ and T₂ maps were computed by inner product matching of measured signal timecourses to scan-specific dictionaries of modeled signal timecourses for unique T₁ and T₂ combinations.³ Average heart rates during scans were recorded.

Relaxometric (T₁, T₂) and measured signal timecourse data were analyzed. The myocardium was manually contoured and divided following the American Heart Association 16-segment model. Average segmental relaxometric values were recorded, then averaged together to obtain surrogate measures of global myocardial T₁ and T₂. In order to investigate whether additional information is present in cMRF timecourses as compared to relaxometric data, linear discriminant analysis (LDA) was employed. Mid-ventricular septal segments were processed to obtain timecourse data. Complex-valued timecourses were averaged within segments to reduce noise and absolute-valued timecourses were recorded. Regularized LDA was used to account for small sample size. Class separability of CA and control groups was assessed by the Fisher coefficient (larger coefficient value meaning greater separability), akin to previous work that compared the class separability of textural feature sets.⁹ Leave-2-out cross-validation was employed to evaluate classification error rates. Statistical analyses of T₁, T₂, LDA scores, error rates, and heart rates were performed using unpaired, two-sided Student’s t-tests.

Results

Significant elevation of myocardial T₁ and T₂ was observed in patients compared to controls (T₁: 1395±121 vs 1240±36.4 ms, p<0.05; T₂: 36.8±3.3 vs 31.8±2.6 ms, p<0.05). Significant differences in LDA scores between patients and controls were observed for both relaxometric-based LDA (0.56±0.28 vs 0.18±0.13, p<0.05) as well as timecourse-based LDA (0.97±0.02 vs 0.02±0.02, p<0.05). The Fisher coefficient was greater for timecourse-based LDA (60.8) compared to relaxometric-based LDA (1.6). Leave-2-out cross-validation indicated lower classification error rates for timecourse-based LDA than relaxometric-based LDA (12.6±24.3 vs 22.5±30.1 %, p<0.05). Heart rates were similar between groups (patients: 87±23 bpm; controls 87±20 bpm; p=0.95).

Discussion

In this proof-of-principle report of cMRF for characterization of CA, differences between CA patients and healthy controls were observed for both relaxometric and signal timecourse analyses. Global measures of myocardial T₁ and T₂ derived from cMRF were elevated in CA patients compared to controls, suggesting the ability for cMRF to detect amyloid deposition in the myocardium. Analysis of cMRF signal timecourses by LDA also showed significant differences between CA patients and controls. When comparing the ability of relaxometric-based LDA and signal timecourse-based LDA to discriminate CA patients from healthy controls, timecourse-based LDA had greater class separability, according to the Fisher Coefficient, and a statistically significant reduction in cross-validation error rate. The improvement in discriminating CA from controls may be due to additional information in signal timecourses beyond that captured by T₁ and T₂ relaxation times alone. Such differences in signal timecourses may be influenced by a number of factors, although average heart rate appears to be an unlikely driver of such observed differences as heart rates were not significantly different between groups. Two potential sources of the observed differences are magnetization transfer and diffusion, both of which are known to affect MRF signal timecourses^7,10 as well as have been reported in the context of identifying myocardial fibrosis⁶ and myocardial infiltration of amyloid proteins⁸. However, continued study of cMRF is needed to isolate and identify specific sources of the observed differences that may eventually be incorporated into the cMRF technique or added to CA assessments. These observations from cMRF characterization of CA highlight the potential utility of cMRF as well as opportunities for further investigation and development.

Limitations of this study include small sample size, age differences between the CA group and control group, and the usage of peripheral pulse gating usage for cases with inadequate ECG signal or inconsistent triggering (four patients, one control). Given these limitations, continued evaluation of cMRF in a larger patient population is warranted. Additionally, evaluation of cMRF-based analysis to discriminate CA patients from clinically comparable patients without amyloid disease (e.g. hypertrophic cardiomyopathy) is needed.

In conclusion, cardiac Magnetic Resonance Fingerprinting may enable the characterization of cardiac amyloidosis by simultaneous T₁ and T₂ mapping as well as signal timecourse analysis. Analysis of cMRF signal timecourse data may improve tissue characterization as compared to native myocardial T₁ and T₂ alone.

Figure 1. CA characterization by cardiac MR Fingerprinting (cMRF).

(a) Example T₁ and T₂ maps derived from cMRF in an immunoglobulin light-chain CA patient and a control. Myocardial T₁ and T₂ are elevated in the CA patient. (b) Global native myocardial T₁ and T₂ values. (c) Scatterplot of global T₁ and T₂ values. (d) Averaged cMRF timecourses for patients and controls, smoothed with 25-point temporal filtration. Differences are observed (red arrows). Horizontal bars along the x-axis indicate magnetization preparation during the sequence: inversion (black), none (medium-grey), T₂-preparation (light-grey). (e) Normalized LDA scores. (f) Fisher coefficients. (g) Error rates from 91 unique iterations of leave-2-out cross-validation. *Indicates statistical significance at p<0.05.

Highlights.

• Cardiac amyloidosis (CA) is an infiltrative cardiomyopathy

• Myocardial T1 and T2 mapping can support CA tissue characterization, aid diagnosis

• Cardiac Magnetic Resonance Fingerprinting (cMRF) simultaneously maps T1 and T2

• Signal timecourse data from cMRF may enable improved CA tissue characterization

• cMRF-derived T1 and T2 were elevated in CA patients vs controls

• cMRF signal timecourse data showed improved CA discrimination vs T1 and T2 alone