Complementing sparse vascular imaging data by physiological adaptation rules

Mathematical modeling of pressure and flow waveforms in blood vessels using pulse wave propagation (PWP) models has tremendous potential to support clinical decision making. For a personalized model outcome, measurements of all modeled vessel radii and wall thicknesses are required. In clinical practice, however, data sets are often incomplete. To overcome this problem, we hypothesized that the adaptive capacity of vessels in response to mechanical load could be utilized to fill in the gaps of incomplete patient-specific data sets. We implemented homeostatic feedback loops in a validated PWP model to allow adaptation of vessel geometry to maintain physiological values of wall stress and wall shear stress. To evaluate our approach, we gathered vascular MRI and ultrasound data sets of wall thicknesses and radii of central and arm arterial segments of 10 healthy subjects. Reference models (i.e., termed RefModel, n = 10) were simulated using complete data, whereas adapted models (AdaptModel, n = 10) used data of one carotid artery segment only, and the remaining geometries in this model were estimated using adaptation. We evaluated agreement between RefModel and AdaptModel geometries, as well as that between pressure and flow waveforms of both models. Limits of agreement (bias ± 2 SD of difference) between AdaptModel and RefModel radii and wall thicknesses were 0.2 ± 2.6 mm and -140 ± 557 µm, respectively. Pressure and flow waveform characteristics of the AdaptModel better resembled those of the RefModels as compared with the model in which the vessels were not adapted. Our adaptation-based PWP model enables personalization of vascular geometries even when not all required data are available.NEW & NOTEWORTHY To benefit personalized pulse wave propagation (PWP) modeling, we propose a novel method that, instead of relying on extensive data sets on vascular geometries, incorporates physiological adaptation rules. The developed vascular adaptation model adequately predicted arterial radius and wall thickness compared with ultrasound and MRI estimates, obtained in humans. Our approach could be used as a tool to facilitate personalized modeling, notably in case of missing data, as routinely found in clinical settings.