Predictive simulation of coil mechanics uncovers shape-driven determinants of endovascular coiling efficacy in intracranial aneurysms

Endovascular coiling is a minimally invasive therapeutic option for intracranial aneurysm (IA). This technique involves occluding the aneurysmal sac with one or more metallic coils to decrease perfusion of the dilated arterial wall. Although coil deployment has proven to be generally effective, clinical outcomes are difficult to predict, with aneurysms reopening occurring in approximately 20% of cases. The purpose of this study is to use in silico modeling combined with statistical analysis and machine learning to predict the biomechanical outcomes of coiling by studying the relationship between IA morphology and post-coiling quantitative parameters. First, a clinical database of IAs was parametrized based on a set of geometrical features. and 500 synthetic sac geometries were virtually generated. Simulations of coil deployment were then performed for each of these geometries, and the resulting reaction force, elastic energy, and contact pressure were used to train a predictive machine learning algorithm. The surrogate model demonstrated strong predictive performance for the reaction force (R2 = 0.74 and MAPE = 4.56%) and moderate performance for the elastic energy stored by the deformed coil (R2 = 0.68 and MAPE = 26.95%). Morphological factors such as aneurysm sac volume, surface, and height showed a good correlation with simulation parameters using Spearman analysis. As a preliminary proof-of-concept study, this methodology represents a first step toward the development of an in silico tool aimed at enhancing pre-operative planning for endovascular coil procedures. This has the potential to improve the stratification of patients at higher risk of complications and the identification of borderline IAs that may not be suitable for endovascular coiling.