64 | Chapter 4 modelled segmental strain is shifted in time to best match strain around . Modelled strain is obtained from the modelled sarcomere length , using ϵmodel,seg = ls,seg ls,seg t=t0 −1 ⋅ 100% Using the objective function, goodness of the simulation is quantified relative to the measurements. It is unknown which strain indices identify the strain. Therefore, the used objective function is based on the full strain curve. To reduce the effect of drift, and reduce the effect of the atria on the estimate, only strain is used from start of QRS complex up to 50% of relaxation. The modelled onset of shortening is matched with the measured onset of shortening. The quadratic difference of each segment is defined as A2 seg = t50relax ∫ tonsetQRS (ϵmodel,seg −ϵmeas,seg) 2 dt t50relax −tonsetQRS The used segments are the apical, mid, and basal segment of the RVfw, and the LVfw and IVS strain. To prevent the parameter estimation protocol from exploring non-physiological area in the input space, a maximum mean left atrial pressure is added as a penalty function. There are signs of increased atrial volumes in this cohort18, but no invasive pressure measurements are available. Therefore, we cannot rule out increased diastolic pressures. No pulmonary hypertension was observed in this cohort, so it can be assumed that the mean left atrial pressure (mLAP) not exceeds 15 mmHg.19 To constrain our simulation results to physiological values without having too much effect on the estimation protocol, we use a threshold mLAP of pthres =25 mmHg. This is implemented as a penalty function epressure = ¯pla −pthres 0 if ¯pla >pthres else The final objective function results in e2 = 1 3 ⋅ (1/3 A2 RVapex α2 +1/3 A2 RVmid α2 +1/3 A2 RVbase α2 + A2 IVS α2 + A2 LVfw α2 )+ e2 pressure β2 With α =1% and β =1mmHg Parameter Estimation protocol Parameters were individually estimated using a parameter estimation framework previously described in more detail.8 This framework estimates model parameters using the clinical data as described above and results in a virtual subject that reproduces the clinical data. In brief, the parameter estimation framework consisted of two steps. First, with CO and HR set to the measured values, 5000 quasi-random Monte Carlo simulations were performed from which the best 60 simulations were used as initial candidate solutions. Second, these candidate (1) (2) (3) (4) t =t0 ls
RkJQdWJsaXNoZXIy MTk4NDMw