Influence of Segmentation on Geometrical Parameters and Hemodynamics in Cerebral Aneurysms

This study proposes to investigate the effects of various segmentations issued from three different intracranial aneurysm 3D datasets, on geometrical parameters and on blood flow patterns. The objective is to find the best segmentation protocol that reduces gray-level selection arbitrariness. Segmentation is an important source of uncertainty with a clinical impact on both morphological and hemodynamics numerical simulations that may largely influence therapeutic decisions taken by neuroradiologists and neurosurgeons. Medical imaging data was acquired in patient carrying 1) posterior communicating artery bi-lobed aneurysm, 2) aneurysm and 3) other aneurysm . For each patient, four different threedimensional segmentations (reconstructions) were further generated by the same technician. Geometrical parameters such as artery, aneurysm and neck sizes, as well as aspect ratio were studied. Numerical simulations were performed under physiological unsteady flow conditions (mean Re=194, maximal Re=447, α=2.71 and pulsatility index was equivalent to 1.87). Flow patterns changes and their sensitivity within the four reconstructed geometries are discussed. Geometrical parameters presenting the highest variability were the aneurysm lobe dimensions (20%). The neck length was the second parameter presenting the highest variability, which reached 21.1% (range 15.1%-27.8%). The neck width variability reached 13.8% (range 11.8%-16.7%) and the aspect ratio variability was calculated to reach 14.2% (range 6.8%-17.9%). The artery height and the artery width were presenting a variability of 13.7% (range 12.3%15.1%) and 10.8% (range 8.1%-12.4%), respectively. Finally, when it came to the aneurysm dimensions, the aneurysm depth, aneurysm height, and aneurysm width presented variabilities of 12.8% (range 10.1%-13.2%), 9.4% (range 7.6%-12.0%), and 7.3% (range 4.6%-12.1%), respectively. Differences in the flow path lines and magnitude, wall shear stress, pressure and vorticity are also reported and discussed. To be filled in with Philippe findings A segmentation process reconstructing 3D anatomies that is less sensitive to human intervention is of high added-value.