A framework for modelling whole-lung and regional transfer factor of the lung for carbon monoxide using hyperpolarised xenon-129 lung magnetic resonance imaging

Abstract

Background: Pulmonary gas exchange is assessed by the transfer factor of the lungs (T _L) for carbon monoxide (T _LCO), and can also be measured with inhaled xenon-129 (¹²⁹Xe) magnetic resonance imaging (MRI). A model has been proposed to estimate T _L from ¹²⁹Xe MRI metrics, but this approach has not been fully validated and does not utilise the spatial information provided by three-dimensional ¹²⁹Xe MRI.

Methods: Three models for predicting T _L from ¹²⁹Xe MRI metrics were compared: 1) a previously-published physiology-based model, 2) multivariable linear regression and 3) random forest regression. Models were trained on data from 150 patients with asthma and/or COPD. The random forest model was applied voxel-wise to ¹²⁹Xe images to yield regional T _L maps.

Results: Coefficients of the physiological model were found to differ from previously reported values. All models had good prediction accuracy with small mean absolute error (MAE): 1) 1.24±0.15 mmol·min^-1·kPa^-1, 2) 1.01±0.06 mmol·min^-1·kPa^-1, 3) 0.995±0.129 mmol·min^-1·kPa^-1. The random forest model performed well when applied to a validation group of post-COVID-19 patients and healthy volunteers (MAE=0.840 mmol·min^-1·kPa^-1), suggesting good generalisability. The feasibility of producing regional maps of predicted T _L was demonstrated and the whole-lung sum of the T _L maps agreed with measured T _LCO (MAE=1.18 mmol·min^-1·kPa^-1).

Conclusion: The best prediction of T _LCO from ¹²⁹Xe MRI metrics was with a random forest regression framework. Applying this model on a voxel-wise level to create parametric T _L maps provides a useful tool for regional visualisation and clinical interpretation of ¹²⁹Xe gas exchange MRI.

FIGURE 1

Schematic of the parallels between the underlying physiology measured by the transfer factor of the lung for carbon monoxide (T_LCO) lung function test and xenon-129 (¹²⁹Xe) magnetic resonance imaging (MRI). Like the membrane conductance (D_M), the ¹²⁹Xe membrane signal is dependent on the surface area and thickness of the alveolar membrane. The ¹²⁹Xe red blood cell (RBC) signal is influenced by both the gas exchange across the alveolar membrane and the capillary perfusion, so can be linked to the capillary blood volume (V_C). RBC:gas measures the transfer of gas from the alveolus, across the alveolar membrane and into the RBCs, so is analogous to the transfer coefficient (K_CO). VV is the volume of the lung where ¹²⁹Xe signal is detected, which is alike to the alveolar volume (V_A). θ is the reaction rate of CO with the RBCs. This figure was partly generated using Servier Medical Art, provided by Servier, licensed under a Creative Commons Attribution 3.0 unported license.

FIGURE 2

Regional random forest model: information on regional red blood cell (RBC) uptake and gas signal distribution from dissolved xenon-129 (¹²⁹Xe) imaging was utilised to produce regional maps of transfer factor (T_L). The transfer coefficient (K) prediction pipeline was applied to every voxel of the RBC:gas map, along with patient age and sex, to output a map of predicted K. For the prediction of accessible alveolar volume, an extra step (equation 7) was required to convert the gas signal map into a map of ventilation distribution (V_r), which had the required units of litres and order of magnitude. This involved finding the signal intensity of each pixel (I_r), dividing this by the mean signal intensity (I_total/N_voxel) and multiplying this fraction by the ventilated volume (VV) from ¹²⁹Xe ventilation imaging. The V_A random forest prediction was then applied to each voxel of this map, along with patient age and sex. The resulting map was renormalised by N_voxel so that it represented the ventilation per voxel and summed to give predicted V_A. This was then multiplied with the K map to obtain a map of T_L.

FIGURE 3

Random forest regression model. a) Linear regression (top) and Bland–Altman (bottom) plots of the measured and random forest-predicted transfer factor values for the training data. b) ranked importance of the prediction variables. c) Linear regression (top) and Bland–Altman (bottom) plots for the measured and random forest-predicted transfer factor values for the validation group. T_L: transfer factor of the lungs; MSE: mean squared error; T_LCO: T_L for carbon monoxide; RBC: red blood cells; V_A: alveolar ventilation; V: ventilated volume; PCH: post-COVID-19 hospitalisation; K: transfer coefficient.

FIGURE 4

a) Ultrashort echo time (UTE) lung structure images and b) random forest regression-predicted transfer factor of the lung (T_L) maps for five participants and their diagnosis, sex and age, and c) T_L for carbon monoxide (T_LCO) z-score at visit one. The fourth patient shown in a) had a lack of xenon-129 signal in the upper right lung due to underlying structural changes (yellow arrows). d) The measured and estimated T_L values for each participant are given, where “WL-RF” refers to the value from the whole-lung random forest model and “R-RF” refers to the value from the sum of T_Lr from the regional random forest model over all voxels. M: male; F: female; PCH: post-COVID-19 hospitalisation.

FIGURE 5

Random forest regression modelled a) transfer factor of the lung (T_L), b) transfer coefficient (K) and c) alveolar volume (V_A) maps for a 57-year-old female patient with both asthma and COPD for six lung slices, plus the modelled and measured whole-lung values at visit one. “WL-RF” refers to the value from the whole-lung random forest model and “R-RF” refers to the value from the regional random forest model. Although this patient had a normal T_L for carbon monoxide (T_LCO) (as measured by pulmonary function testing), their T_L map indicates reduced gas exchange. The K map shows a heterogeneous gas transfer rate which does not match the ventilation distribution.

FIGURE 6

Predicted maps of a) transfer factor of the lung (T_L), b) transfer coefficient (K) and c) alveolar volume (V_A) for a single lung slice for a 61-year-old female with both asthma and COPD pre- and post-bronchodilator (BD). The sum (mean for K_r) for each slice (posterior to anterior) is plotted, showing the change in distribution following BD administration. The lung slice is indicated with an arrow. These plots show that V_A,_r and T_Lr increase in the posterior lung following BD, while slightly decreasing in the central and anterior slices.