90 | Chapter 5 Figure 4. Convergence of estimated model parameters. The distributions on the right show the final estimated posterior distribution. Figure 3 (right) shows the estimated regional RV model parameters and the RV tissue properties contractility, activation delay, compliance, and work density. The RV tissue properties were estimated with distributions with a smaller variance compared to the estimated model parameters. A decrease in basal contractility, compliance, and work density with respect to the apical and mid segment was found which is in line with the abnormal basal deformation pattern. Figure 5 (bottom) shows the correlation between posterior model parameter distributions with the RV tissue properties contractility, compliance, and work density. Contractility was mostly correlated with SfAct, AmRef, and CO. In the RVapex and RVmid, contractility was not only dependent on the parameters prescribing its own segmental mechanics, but also on the parameters prescribing other segmental mechanics. Similar results were observed for compliance, which was correlated with SfAct, k1, and dT. Compliance showed no correlation with AmRef, RSD, and CO. Work density was mostly correlated with CO. Estimated model parameters were highly reproducible. Computational reproducibility was found with an MI of 89.9% (95% CI (60.1 – 95.9)). The reproducibility error given inter- and intraobserver variability were estimated with an MI of 86.5% (95% CI (46.0 – 95.2)) and 85.9% (95% CI (43.7 – 95.3)), respectively. Uncertainty Quantification of Virtual Patient datasets Nine virtual patients were created based on the nine real-patient estimations. As an example, Figure 6 shows the virtual patient based on the patient results described above. Regional deformation characteristics were simulated close to the virtual patients deformation characteristics (Χ2 opt = 2.0 (95% CI = (1.2 − 3.0)) =2.0 (95% CI=(1.2 —3.0)). The true parameter values were well captured by the estimated distributions. The HDI of the true parameter values was 9% (95% CI (0 – 79)). Heterogeneity in model parameters was well preserved. The width of the distribution in virtual fits was similar to that in the original patient estimation.
RkJQdWJsaXNoZXIy MTk4NDMw