Summary of Main Findings

Introduction: Inhaled-gas (¹²⁹Xe/¹⁹F) MRI has been proven to be useful for diffusion-weighted and dynamic-lung-imaging.^1-3 Diffusion-weighted-pulmonary ¹²⁹Xe MRI^4,5 provides physiologically relevant biomarkers (specific diffusion length scale estimate (L_mD) and mean-linear-intercept-estimate (L_m)) of obstructive lung-disease including emphysema, bronchopulmonary dysplasia, congenital lobar emphysema and alpha-1-antitrypsin-deficiency.6-8 MRI-based techniques permit the acquisition of regional-fractional-ventilation (r)^9-11 which can be very useful as a radiation-free alternative to CT for detecting gas-trapping in lung-diseases such as lung, fibrosis and COPD.¹ Thus, free-breathing ¹⁹F dynamic-lung-imaging has been recently demonstrated in human lungs ¹². Clearly, each new wash-out breath of air replaced some volume of the ¹⁹F gas in lung, so the signal intensity of the resulting images was gradually attenuated. Similarly, the large b-values cause diffusion-decay, so the noisy (SNR<5) large b-value or low ¹²⁹Xe/¹⁹F density images “accelerates” the decay that leads to over-estimation of the diffusion/ventilation decay-rate-constants. We hypothesize that denoising of the diffusion/ventilation data will permit extracting accurate diffusion/ventilation regional-biomarkers. In this proof-of-concept study, our objective was to demonstrate the feasibility of this method using the diffusion-weighted and dynamic-ventilation data in a small animal cohort.
Methods: Four normal-animals (Sprague Dawley rats) were studied (ethics board approved protocol). 129Xe MR was performed at 3.0T (MR750, GEHC, WI) using a high-performance insert-gradient-coil (G=500mT/m, slew-rate=2000T/m/s)13 and a commercial, xenon-quadrature-rat-sized ¹²⁹Xe RF coils¹⁴ (Morris Instruments, Canada). Pulse-sequence details and the breathing protocol were previously described.¹¹ In all xenon measurements the diffusion-sensitization gradient pulse ramp up/down time was 500μs, constant time=2ms, Δ=5.0ms, providing eight b-values 0, 2.35, 9.39, 21.1, 37.6, 58.7, 84.5 and 115s/cm2. A single breath-hold, interleaved (eight interleaves) centric 2D FGRE diffusion-weighted sequence was acquired for a coronal slice (TE/TR=10.3ms/14.5ms, BW=8kHz, matrix size=64x64, and FOV=50x50mm2, variable flip angle, 10sec breath-hold).¹⁵ Hyperpolarized ¹²⁹Xe (83%) gas was provided by turn-key, spin-exchange-polarizer systems (Polarean-9800, polarization range: 7-15%; Xemed XeBox-E1016, polarization range: 20-35%). ¹⁹F MRI ventilation measurements were performed using a 3.0T Philips Achieva scanner with maximum gradient strengths of 4G/cm and a home-built rat-sized quadrature transmit/receive coil tuned to the ¹⁹F resonance frequency of 120.15MHz. Pulse sequence details and the breathing protocol were previously described.¹⁷ Two denoising methods (Method-1: MP-PCA18 and Method-2: MP-PCA preceded by hard-thresholding over a nuclear norm in k-space, in order to suppress noise-correlations from the reconstruction) were applied to the ¹²⁹Xe diffusion-attenuation-data (Fig.1, top-panel), ¹⁹F dynamic ventilation data (Fig.1, middle-panel), and ¹²⁹Xe dynamic-ventilation data (Fig.1, bottom-panel) as previously described.¹⁸ The signal intensity of the original and denoised diffusion-data were represented as a functions of the b-value (Fig.2a) and then fitted following Abascal et al method.¹⁹ L_mD and L_m were generated using the stretched-exponential-model^20,21 as previously described.²² The signal intensity of the original and denoised ventilation data were represented as a function of the number of wash-out-breaths (Fig.2b and c) and then fitted following. ²³ n=0, 1, 2, 3, 4, 5, 6, 8 was used to fit ¹²⁹Xe/¹⁹F original and denoised data and generate the fractional-ventilation maps as previously described. ²³ SNR was calculated as previously described.²⁴
Results: Fig.2 shows the improved SNR of the denoised images over the original images for the diffusion/ventilation data. Tables 1a and b shows the Stretched Exponential Model based morphometry estimates for all rats and all type of data. Mean L_mD and L_m estimates for original were significantly larger compared to corresponding mean L_mD and L_m estimates obtained from denoising method I data (p<.03). The mean differences of 14.0%/(4.0%) and 17.0%/(7.0%,), were observed between original and denoising-1/(2) for the overall L_mD and L_m values, respectively. Fig.3 shows representative L_mD and L_m maps obtained from the original and denoised data. Table 2 shows the ¹⁹F MRI-based mean fractional-ventilation-estimates for all rats and all type of data. Original mean r estimates were significantly larger compared to corresponding mean r estimates for the denoising-2 data (p<.08) and not significantly different from corresponding mean r estimates for the denoising-1 data (p>.05). The mean differences of 5.0% and 10.0%, were observed between original and denoising-1 and 2 for the r values, respectively. Fig.4 (top-panel) shows representative r maps obtained from the original and denoised data. Table 3 shows the ¹²⁹Xe MRI-based mean fractional-ventilation estimates for all rats and all type of data. The mean differences of 24.0% and 23.0%, were observed between original and denoising-1/2 for the r values, respectively. Fig.4 (bottom-panel) shows representative r maps obtained from the original and denoised data.
Discussion and Conclusion: In this proof-of-concept study, we showed that the denoising-1 method improves SNR of diffusion data (Fig.2a) and therefore L_mD and L_m (Tables 1a and b) matched with the previously reported values. Denoising-2 led to significant over-estimation of L_mD and L_m and cannot be recommended for improving the diffusion-weighted ¹²⁹Xe MRI data. In turn, the ¹²⁹Xe/¹⁹F dynamic-ventilation data denoised using the method-2 permitted to extract the reasonable mean r estimates (Tables 2 and 3), while the denoising method-1 did not change the mean r estimate compared the mean-fractional-ventilation-estimates extracted from the original data (Table 2, Fig.2b). For the first-time we have demonstrated that the ¹²⁹Xe/¹⁹F MRI images can be denoised using two different methods and it does not lead to compromising the accuracy of the regional-biomarkers such as regional L_mD and L_m and fractional-ventilation estimates.

Acknowledgements

We acknowledge the support of the Natural Sciences and Engineering Research Council of Canada, R5942A04