ABSTRACT
Adding prior knowledge to compressed sensing reconstruction can improve image reconstruction. In this work, two approaches are investigated to improve reconstruction of two-dimensional hyperpolarized (3)He lung ventilation images using prior knowledge. When compared against a standard compressed sensing reconstruction, the proposed methods allowed acquisition of images with higher under-sampling factors and reduction of the blurring effects that increase with higher reduction factors when fixed flip angles are used. These methods incorporate the prior knowledge of polarization decay of hyperpolarized (3)He and the mutual anatomical information from a registered (1)H image acquired in the same breath. Three times accelerated two-dimensional images reconstructed with compressed sensing and prior knowledge gave lower root-mean square error, than images reconstructed without introduction of any prior information. When introducing the polarization decay as prior knowledge, a significant improvement was achieved in the lung region, the root mean square value decreased by 45% and from the whole image by 36%. When introducing the mutual anatomical information as prior knowledge, the root mean square decreased by 21% over the lung region and by 15% over the whole image.
Subject(s)
Algorithms , Data Compression/methods , Helium , Image Interpretation, Computer-Assisted/methods , Lung/anatomy & histology , Magnetic Resonance Imaging/methods , Adult , Contrast Media , Female , Gases , Humans , Image Enhancement/methods , Isotopes , Male , Radiopharmaceuticals , Reproducibility of Results , Sensitivity and SpecificityABSTRACT
In imaging of human lungs with hyperpolarised noble gases, measurements of apparent diffusion coefficient (ADC) and relaxation time provide valuable information for the assessment of lung microstructure. In this work, a sequence was developed for interleaved acquisition of ventilation images, ADC, T(2)* and flip angle maps in a single scan from the human lungs with a single dose of inhaled (3)He at 3 T. Spatially registered ventilation images with parametric maps were obtained. The total acquisition time was reduced by random undersampling of the k-space and reconstruction using compressed sensing (CS). The gain in speed was used for an increase in spatial resolution. Mean ADC values from the fully sampled and undersampled CS data exhibit no statistically significant difference in a given subject. The mean T(2)* values, however, were found to differ significantly, which is attributed to the combined effect of low signal-to-noise ratio (SNR) of the fully sampled data and the smoothing effect inherent in CS reconstruction.
Subject(s)
Helium , Image Processing, Computer-Assisted/methods , Magnetic Resonance Imaging/methods , Respiration , Adult , Anisotropy , Diffusion , Female , Health , Humans , MaleABSTRACT
The (3)He MR diffusion signal is sensitive to lung microstructure, but it is also affected by the presence of background field inhomogeneities induced by the magnetic susceptibility difference at the air-tissue interface. These susceptibility-induced gradients, which are dependent on field strength, have been assumed negligible in theoretical models used to extract airway morphometric information from (3)He MR diffusion data at field strengths up to 4.7 T. In this work, the effect of susceptibility gradients on (3)He apparent diffusion coefficient is demonstrated with experiments in healthy volunteers at two B(0) field strengths: 1.5 and 3 T. Apparent diffusion coefficient values obtained at 3 T were systematically larger than at 1.5 T, demonstrating that susceptibility effects are statistically significant even at clinical field strengths (B(0) ≤ 3 T) and introduce biases in the estimates of airway dimensions (e.g., mean linear intercept up to 17% larger at 3 T than 1.5 T). Susceptibility effects should be taken into account in the development of theoretical models of lung (3)He MR diffusion and considered when interpreting (3)He apparent diffusion coefficients obtained at different B(0).
Subject(s)
Diffusion Magnetic Resonance Imaging/methods , Image Processing, Computer-Assisted/methods , Lung/physiology , Administration, Inhalation , Adult , Algorithms , Artifacts , Computer Simulation , Gases , Helium , Humans , Isotopes , Lung Volume Measurements , Models, Theoretical , Reference Values , Signal-To-Noise RatioABSTRACT
Models of lung acinar geometry have been proposed to analytically describe the diffusion of (3)He in the lung (as measured with pulsed gradient spin echo (PGSE) methods) as a possible means of characterizing lung microstructure from measurement of the (3)He ADC. In this work, major limitations in these analytical models are highlighted in simple diffusion weighted experiments with (3)He in cylindrical models of known geometry. The findings are substantiated with numerical simulations based on the same geometry using finite difference representation of the Bloch-Torrey equation. The validity of the existing "cylinder model" is discussed in terms of the physical diffusion regimes experienced and the basic reliance of the cylinder model and other ADC-based approaches on a Gaussian diffusion behaviour is highlighted. The results presented here demonstrate that physical assumptions of the cylinder model are not valid for large diffusion gradient strengths (above approximately 15 mT/m), which are commonly used for (3)He ADC measurements in human lungs.