Key influencers in an aneurysmal thrombosis model: A sensitivity analysis and validation study

Figures

• 

  FIG. 1.. The 3D model and elementary effect (EE) method results using FiPi as the output metric indicate that the resting platelet concentration is the unique most important parameter that affects the final formed thrombus composition.

• 

  FIG. 2.. We manually labeled the partial spontaneous thrombosis region as pretreatment clinical ground truth. High residence (RT) and low shear rate (SR) are widely used in computational models to characterize the flow stasis that triggers the thrombosis process in the aneurysm sac. Using our previously calibrated trigger thresholds,20 RT 1.9 s and SR 11 s−1, our flow simulation model successfully predicted the main thrombosis area before treatment.

• 

  FIG. 3.. Comparison between the patient-specific and non-patient-specific simulation results using the clinical record as ground truth. In the clinical ground truth figure, the red part was the manually labeled partial thrombosis region. In the simulation results, the thrombus were assumed to be formed in regions where the fibrin concentration, CFI, is greater than 600 nM.

• 

  FIG. 4.. Comparison between the immediate post-treatment angiographic result and our simulation result. (a) The clinical immediate post-treatment digital subtraction angiography (DSA) result shows minimal residual flow in the neck of the aneurysm. (b) The post-treatment simulation result also shows residual blood flow in the neck region of the aneurysm. (c) The thrombus formation result predicted by our thrombosis model after virtual patient-specific treatment. The thrombi are assumed to be formed in regions where the fibrin concentration, CFI, is greater than 600 nM. The white patches in the slice shown are caused by the visualization of the virtual coils.

• 

  FIG. 5.. The thrombosis model by Sarrami-Foroushani et al.,8 including four main biochemically coupled events: thrombin generation, fibrin generation, platelet activation, and platelet aggregation.

• 

  FIG. 6.. The sensitivity analysis (SA) workflow. There are 31 parameters in our thrombosis model with 12 parameters that can be assessed with a lumped 0D model, 14 that can be investigated with the 3D full model, and 5 others that do not need to be assessed. We first screened 4 influential parameters from the 12 kinetic-associated parameters using the lumped 0D model, then performed SA with the elementary effect (EE) method using the 3D full model for 18 (4 + 14) parameters to identify the most influential model parameter. Finally, we further demonstrated the necessity of SA with a validation case by comparing the simulation results under patient-specific and non-patient-specific settings with the clinical ground truth.

• 

  FIG. 7.. A partially thrombosed clinical case with manual label of the thrombosis region before treatment. (a) The raw image from Leeds General Infirmary. (b) The manually labeled partial thrombosis region.