In silico modelling to advance arterial spin labelling magnetic resonance imaging

Abstract

Rationale Arterial spin labelling (ASL) magnetic resonance imaging (MRI) of the lungs is an imaging methodology used for the measurement of pulmonary perfusion. Unlike other pulmonary perfusion imaging techniques such as Positron emission tomography (PET), Single-photon emission computed tomography and Computed tomography (CT), ASL MRI does not involve the use of ionising radiations; thereby enabling repetitive measurements. Perfusion is quantified by obtaining a difference image whereby blood is used as an endogenous tracer and alternately tagged as ‘bright’ and ‘dark’. The difference between the bright and dark image gives the ASL magnetic resonance (MR) image. Although pulmonary ASL MRI aims to quantify perfusion, these measurements are often contaminated with artefacts such as signals from non-capillary blood. Intensity thresholding is one approach that has been proposed for minimising the noncapillary blood signal within an ASL MR image. This technique involves eliminating voxels with signal intensity above a set threshold as these encompass large vessels feeding capillary beds outside the imaging plane. This method of minimising non-capillary blood signal has been tested in a previous in silico human modelling study; however, it has only been tested under a restricted set of physiological conditions (supine posture and a cardiac output of 5 L/min). This thesis presents an in silico approach that extends previous intensity thresholding analysis to estimate the optimal ‘per-slice’ intensity threshold value under various conditions and across two species (human and pig). The intensity thresholding was examined using the individual components of the simulated ASL signal (signal arising independently from capillary blood as well as pulmonary arterial and pulmonary venous blood). This analysis aimed to assess whether the threshold value as applied in human ASL MRI should vary with slice location, posture or cardiac output. Pigs are often used for translational research and as preclinical models due to the anatomical and physiological similarities to humans. However, there are structural differences evident in the lungs of pigs having greater asymmetry in their airway and vascular branching structures than humans. The differences in the geometry of the branching structures create variations in the distribution of pulmonary perfusion within both human and pig lungs. The weight of a developing animal (such as a pig) has been shown to influence cardiovascular and pulmonary parameters such as cardiac output, heart rate, lung size and pulmonary vessel diameters and this has the potential of affecting blood flow distribution. This thesis examined whether the differences in pulmonary blood flow distribution between a pig and human impact on both the ASL MRI distribution and the choice of thresholding. Also, the robustness of optimum thresholding was determined for varying slice location and cardiac output. A 35% thresholding was predicted within an adult human lung, and it is not yet known how changes in lung size affect the choice of thresholding. The use of microspheres has been considered the ‘gold standard’ for pulmonary perfusion estimation since this technique provides a direct assessment of capillary blood when the appropriate size and number are used. This technique involves the injection of radio- or fluorescence-labelled microspheres into the bloodstream of an animal. The injected microspheres are assumed to be distributed in proportion to blood flow. The animal is then euthanised and the lungs excised from the chest. The lungs are allowed to dry over a positive airway pressure of 25 cmH2O, embedded in foam and uniformly sliced into cubes of 2.0 cm3 in volume within iso-gravitational planes. Each lung piece is then weighed and the perfusion within each piece quantified by radioactivity or fluorescence. Currently, no studies validate or compare pulmonary ASL MRI with microsphere estimation of pulmonary perfusion; in this thesis, a theoretical comparison is provided for pulmonary capillary blood flow quantified using both an in silico model of ASL MRI and microsphere estimation of flow. Method A one dimensional (1D) Poiseuille-based flow model was used to predict the distribution of pulmonary blood flow through the arteries, capillaries and veins within CT derived human and porcine lung models. Solutions of the blood flow model together with the profile of magnetisation defined by the inversion pulse were used to replicate the ASL MRI protocol across multiple sagittal slices of both the human and porcine lung models at varying conditions of posture and cardiac output. The total ASL signals, median ASL signals and heterogeneity of blood flow obtained from the in silico porcine ASL model were compared with measured pig ASL dataset. The sensitivity of an intensity thresholding cost function developed in this thesis was tested within both the human and porcine lung models for varying conditions of slice location, posture and cardiac output. In modelling the microsphere estimation of flow, the microspheres were distributed in proportion to flow. Varying number of microspheres and degrees of stochasticity in microspheres distribution at vessel bifurcations, were incorporated to assess how these affect flow estimations. The lung model was then cut into cubes of 1.9 cm3 and each cube normalised by the weight of the cube. For similar lung regions (using an undeformed lung), the ASL MRI voxels were re-sampled into cubes of volume 1.9 cm3 and compared to microsphere estimation of flow. Results The model predictions of porcine ASL MRI signals agreed with that from the measured ASL MRI pig dataset to within 10%. For both datasets, ASL MRI signals increased from the lateral to the medial region of the lungs due to the predominance of large vessel signals around the pulmonary trunk. From the ASL MRI simulations, when posture was changed from the supine to prone position, both the gradient and heterogeneity of blood flow decreased. When ASL simulations were performed within a human lung model derived from a 70 kg man, an optimised threshold value of 35% was predicted, and this was determined to be independent of slice location, posture and cardiac output. In a developing pig, there are significant changes in the lung size, cardiac output, pulmonary vascular dimensions and heart rate. When the ASL MRI modelling was repeated within the porcine lung model of a developing pig, the choice of thresholding was observed to be highly dependent on the weight of the developing pig. From the microsphere flow simulations, microsphere numbers greater than 100, 000/kg body-weight ( 4 * 106 microspheres for use within a 40 kg porcine model) and a stochasticity level less than or equal to 3% were determined to be the most representative (when compared experimental analysis) in estimating pulmonary capillary blood flow as these reduced regional Poisson noise to within 10%. The incorporation of tissue deformations (due to gravity) into the model increased both the gradient and heterogeneity of capillary blood flow estimation using microspheres. For the unfiltered ASL MRI signals, a better correlation with microsphere estimation of flow was observed for the lateral slice (R = 0.52) compared to the medial slice (R = 0.24) due to the predominance of large vessel signals in the medial region of the lungs. Thresholding improved the correlation across the slices; however, this was still relatively weak for the medial slice (R = 0.61) when compared to the correlation at the lateral region (R = 0.82) of the lungs. Conclusions The optimised intensity threshold value as applied within a matured human lung model or porcine lung model (derived from a developing pig) was independent of slice location, posture and cardiac output. For lung models derived from a developing pig, the optimised intensity thresholding value was highly dependent on the weight of the pig. Microsphere estimation of capillary blood flow was highly reliant on the number of microspheres injected. Both tissue deformations and the geometry of pulmonary blood vessels were found to influence the distribution of microsphere estimation of capillary flow. ASL MRI was observed to better correlate microsphere estimation of capillary blood flow in regions of the lungs devoid of large vessel signals. The relatively weak correlation between filtered ASL MRI and microsphere estimation of flow at the medial region of the lungs highlighted the possible imperfections (inability to completely eliminate the unwanted conduit vessel signals) associated with the method of thresholding (as used in ASL MRI).