Model-based Bayesian inference of the ventilation distribution in patients with Cystic Fibrosis from multiple breath washout, with comparison to ventilation MRI

Abstract

AbstractBackgroundIndices of ventilation heterogeneity (VH) from multiple breath washout (MBW) have been shown to correlate well with VH indices derived from hyperpolarised gas ventilation MRI. Here we report the prediction of ventilation distributions from MBW data using a mathematical model, and the comparison of these predictions with imaging data.MethodsWe developed computer simulations of the ventilation distribution in the lungs to model MBW measurement with 3 parameters: σV, determining the extent of VH; V0, the lung volume; and VD, the dead-space volume. These were inferred for each individual from supine MBW data recorded from 25 patients with cystic fibrosis (CF) using approximate Bayesian computation. The fitted models were used to predict the distribution of gas imaged by 3He ventilation MRI measurements collected from the same visit.ResultsThe MRI indices measured (I1/3, the fraction of pixels below one-third of the mean intensity and ICV, the coefficient of variation of pixel intensity) correlated strongly with those predicted by the MBW model fits (r = 0.93, 0.87 respectively). There was also good agreement between predicted and measured MRI indices (mean bias ± limits of agreement: I1/3: 0.002 ± 0.112 and ICV: −0.001 ± 0.293). Fitted model parameters were robust to truncation of MBW data.ConclusionWe have shown that the ventilation distribution in the lung can be inferred from an MBW signal, and verified this using ventilation MRI. The Bayesian method employed extracts this information with fewer breath cycles than required for LCI, reducing acquisition time required, and gives uncertainty bounds, which are important for clinical decision making.New and NoteworthyThis paper demonstrates that the ventilation distribution observed by ventilation MRI in cystic fibrosis patients can be inferred using multiple breath washout data. The Bayesian method used quantifies prediction uncertainty. This has the potential to be used in the analysis of washout data in the clinic to give greater physiological insight more efficiently. The predictions also remained robust to truncation of the washout dataset, meaning that data-capture time can be significantly reduced using this method.