Uncertainty Quantification of Cardiac Properties | 97 to follow-up. It has previously been shown that heterogeneity in deformation patterns has prognostic value for disease progression38 and life-threatening arrhythmia39. Although no further follow-up of these subjects was available, we can hypothesize our model might identify abnormal tissue substrates before this is clearly visible in deformation patterns. Further studies should investigate whether our approach is able to detect AC in an early stage and whether it has added prognostic value. In this study, we estimated model parameters to predict tissue mechanics under mechanical loading similar to loading during measurement. To achieve this, we included CO in the parameter subset and EDV and EF in the likelihood function. The model could be used for predicting the behaviour of the heart under different loading conditions. This could facilitate the study of loading effects of drug interventions in the digital twin. Besides, the effect of exercise, which is an important modulator of phenotypic expression of AC47, could be studied in the digital twin. For the latter, a virtual cardiac exercise performance test as proposed by Van Loon et al.48 could be used to give more insight in the severity of the substrate and possible triggers for disease progressions. To allow the CircAdapt model to extrapolate its state to other loading conditions such as exercise, more information should be included. Limitations Uncertainties are assumed statistically independent and additive, however, this is in fact more complicated. Measurements have multiple sources for uncertainty. We have only included inter- and intra-observer variability of the speckle tracking imaging in our study. Global longitudinal strain has proven to be reproducible, however, it has been shown that beat-to-beat variability affects segmental peak strain, end systolic strain and post-systolic strain.49 More research should elucidate the origin of this uncertainty, its effect on normalized strain morphology as included in our study, and how to optimally include uncertainty in defining the likelihood function. This could also facilitate inclusion of realistic noise on virtual patient datasets, which was outside the scope of this study. AC is not only characterized by structural disease manifestation, but electrophysiologic substrates play an important role as well.50 Currently, the CircAdapt model only contains the lumped effect of electrophysiology to describe the mechanical behaviour. Future studies could extend the model with a more detailed electromechanical coupling, such as proposed by Lyon et al.51, to be able to describe the electrophysiologic substrate. CONCLUSION We presented a patient-specific modelling approach taking into account uncertainties. With this approach, we were able to reproduce regional ventricular deformation patterns and estimate the underlying tissue properties in AC mutation carriers with an acceptable level of uncertainty. Virtual estimations were precise and real-world estimations were highly reproducible. Two subjects in our case study revealed the evolution of early-stage AC disease over time using longitudinal follow-up datasets. Future studies should apply our method on a larger cohort and investigate the course of early stage RV disease development at individual as well as patient population levels. 5
RkJQdWJsaXNoZXIy MTk4NDMw