Uncertainty Quantification of Cardiac Properties | 89 RESULTS Uncertainty Quantification of Real Patient datasets Regional deformation characteristics were accurately simulated close to the measured deformation and with reasonable uncertainty (Χ2 opt = 9.4 (95% CI [5.4−20.9])) =9.4 (95% CI [5.4-20.9])). Figure 3 (left) shows a representative example. The modelled strain followed the pattern of clinically measured strain during systole and heterogeneity between the segments was well captured. A 1D representation of the convergence of the proposal distribution, corresponding to the estimated model parameters is shown in Figure 4. In the first 50 iterations, the proposal distribution decreased, increased, and moved to the area of interest. From the 50th iteration, most proposal distributions stabilized. This behaviour was also observed in estimations in other datasets. Figure 3. Measured and estimated strain of real subject (left) and violin plots of estimated parameters (right) Deformation patterns and regional heterogeneity was well captured by the model. The best simulation in the sample set was in good agreement with to the patients dataset (Χ2 opt =8.9) . The estimated posterior distributions of the model parameters (Figure 4) of most parameters were estimated with small variances, except for parameters SfAct and k1, because they were unidentifiable in some segments. The posterior correlation matrix (Figure 5, top) shows the correlation between estimated posterior distributions. Notable are the correlations between model parameters SfAct, k1, dT, and AmRef describing mechanics in the same wall segment. Additionally, there was a high correlation between different segments for the model parameters dT and AmRef. From the two global parameters, only RSD seemed to correlate with dT. 5
RkJQdWJsaXNoZXIy MTk4NDMw