The Effect of Hemodynamics on Mass Transport for Modelling Carotid Artery Stenosis

 

Abdu B. Yearwood1, Boppana V. Chowdary2

Mechanical and Manufacturing Engineering, University of the West Indies, St. Augustine Campus, Trinidad and Tobago

abdu.yearwood@my.uwi.edu1; boppana.chowdary@sta.uwi.edu2

INTRODUCTION

Surgical procedures have confirmed that plaque build-up in blood vessels does not randomly occur. Instead, it forms and progresses in specific regions where blood flow is highly disturbed (VanderLaan et al., 2004). In this work, we studied the effect of geometrically-induce hemodynamics on mass transport in carotid arteries; and to determine its use in plaque progression modelling. Our results show that the flow is able to move mass particles from main flow towards outer walls of the external and internal carotid arteries via vortex recirculation and shedding at the close of peak flow. These phenomena will play a role, but not alone, in development of plaque that can lead to stenosis and increased workload on major organs through reflected pressure waves.

 

  • METHODOLOGY

    First, we apply level set segmentation to the acquired MRI dataset  to obtain the vessel of interest. Next, we convert the 3D surface model to a solid step file and export it to ANSYS Workbench for remeshing, modelling and simulation. 2D models (as shown in figure 1 right) were obtained from the contour of the 3D model and exported with 0.5mm thickness. We discretise all geometries with a body sizing of 0.05mm, 15 layers of inflation and total thickness of 0.64mm. Finally, we use a periodic inflow velocity of 0.5 m/s at peak amplitude and 0.1 m/s at minimum from the work of Sinnott et al. (2015). The fluid was modelled as non-Newtonian using a user-defined function in four simulated models as shown in figure (2).

  •  
  • RESULTS

    • The results show an increase in total pressure of 1.34x104 Pascals at the bifurcation stagnant region. WSS ranged from 0.022x10-3 to 1.5 Pascals except at inner walls of the ICA and ECA as expected based on curvature flow with corresponding velocities in ranges of 0.001 to 1ms-1. The results obtained in original geometry agreed with the work of others (Groen et al., 2007; Bahrami and Norouzi 2018) and similar case studies of wall shear stress and pressure distributions. Table (1) shows a sample data frame of time-series data from results taken from all case studies,. Four attributes (flow velocity, plaque thickness, class labels and response variables) were developed based on feature engineering to reduce multicollinearity among predictor variables. Velocities were measured between 0.001-0.26 (normal), 0.44 (mild), 0.64 (moderate) and 1ms-1 (severe). In total, 361 datapoints were measured, randomised and labelled; while a test sample of 57 were removed as a test set for model scoring and selection using evaluation metrics.


    Table 1:     CFD Measures attributes

     

    Velocity (m/s)

    Plaque Thickness

    Class_n

    Response

    0.199019

    1.055547

    0

    Normal

    0.282149

    1.352204

    1

    Mild

    0.187053

    0.765789

    0

    Normal

    0.621982

    3.509114

    2

    Moderate

    0.295608

    1.476386

    1

    Mild

    0.978112

    4.94

    3

    Severe

    0.97964

    5.15

    3

    Severe



    Figure 1:     (left) Vortex shedding in ICA. (right) Four modelled modes for plaque


    CONCLUSION

    In  any  blood vessel  for  which  a  uniform  flow  is  caused to become highly  disturbed,  the  resulting  forces  will  play a role in  plaque  development.  This means that  although  vascular  topology  is  a  driver  of  highly  disturbed  flows, plaque  formation  and  progression  that  leads  to  luminal stenosis  may  occur only  in  the  presence  of  prolong  factors (in  controlled  or  uncontrollable  forms).In future work, we

     

  • FUTURE WORK

    • In future work, we will combine CFD results with clinical features to train and test three classifiers: a deep learning model, a gradient boosting machine and a random forest with cross-validation on a dataset of 303 data points using 75/25 split. Several metrics, including logloss error of the cost function, Mean Square Error, Root Mean Square Error and confusion matrix will be used to determine a model’s performance for selection when applied to a hold out test set of 57 samples.

     

    REFERENCES

    1. Bahrami, S., & Norouzi, M. (2018). A numerical study on hemodynamics in the left coronary bifurcation with normal and hypertension conditions. Biomechanics and modeling in mechanobiology, 1-12.
    2. Groen, H. C., Gijsen, F. J., van der Lugt, A., Ferguson, M. S., Hatsukami, T. S., van der Steen, A. F., . . . Wentzel, J. J. (2007). Plaque rupture in the carotid artery is localized at the high shear stress region: a case report. Stroke, 38(8), 2379-2381.
    3. Sinnott, M. D., Cleary, P. W., Harrison, S. M., Cummins, S. J., Beare, R., Srikanth, V., & Phan, T. G. (2015). How arterial pressures affect the consideration of internal carotid artery angle as a risk factor for carotid artherosclerotic disease. Progress in Computational Fluid Dynamics, an International Journal, 15(2), 87-101.
    4. VanderLaan, P. A., Reardon, C. A., & Getz, G. S. (2004). Site specificity of atherosclerosis: site-selective responses to atherosclerotic modulators. Arteriosclerosis, thrombosis and vascular biology 24(1), 12-22.