Reduced-Order Learning of Cerebral Aneurysm Hemodynamics: Comparing POD, PCA, and Autoencoder Models Coupled with TCN
محل انتشار: مجله مکانیک کاربردی محاسباتی، دوره: 57، شماره: 4
سال انتشار: 1405
نوع سند: مقاله ژورنالی
زبان: انگلیسی
مشاهده: 27
فایل این مقاله در 21 صفحه با فرمت PDF قابل دریافت می باشد
- صدور گواهی نمایه سازی
- من نویسنده این مقاله هستم
استخراج به نرم افزارهای پژوهشی:
شناسه ملی سند علمی:
JR_JCAM-57-4_009
تاریخ نمایه سازی: 22 آبان 1405
چکیده مقاله:
Accurate characterization of cerebral aneurysm hemodynamics is important for understanding flow-related mechanisms associated with aneurysm progression and for quantifying the hemodynamic modifications induced by endovascular treatment. However, high-fidelity transient computational fluid dynamics (CFD) simulations are computationally demanding, particularly when time-resolved three-dimensional hemodynamic fields are required. This study presents a comparative evaluation of three reduced-order modeling (ROM) strategies—proper orthogonal decomposition (POD), principal component analysis (PCA), and Autoencoder—coupled with a temporal convolutional network (TCN) for reconstruction and temporal prediction of hemodynamics in a patient-specific internal carotid artery saccular aneurysm before and after endovascular coiling. Pulsatile, non-Newtonian blood flow was simulated using the Casson model, while the coiled aneurysm region was represented through a porous-medium formulation. Wall shear stress (WSS/AWSS), oscillatory shear index (OSI), and pressure fields were extracted from CFD simulations and compressed into low-dimensional representations using the three ROM approaches. A common TCN architecture was subsequently employed to learn the temporal evolution of the reduced coefficients and predict future hemodynamic states. The results demonstrate that coiling substantially suppresses intra-aneurysmal flow activity and modifies the spatial distributions of WSS, OSI, and pressure. The modal analysis further shows that approximately ۹۹% of the dominant information can be retained using a relatively small number of modes or latent variables, although the required dimensionality varies among hemodynamic quantities and ROM techniques. All three hybrid frameworks successfully reconstructed the major spatial characteristics of the CFD fields, while prediction accuracy decreased with increasing forecast horizon. Among the investigated approaches, PCA–TCN showed the most stable predictive behavior for the present dataset and effectively reproduced the principal distributions of WSS, OSI, and pressure in both original and coiled configurations. These findings demonstrate the potential of ROM–TCN frameworks as computationally efficient surrogate models for time-resolved assessment of cerebral aneurysm hemodynamics and treatment-induced flow alterations.
کلیدواژه ها:
Cerebral saccular aneurysm ، Endovascular coiling ، Computational fluid dynamics (CFD) ، Reduced-order modeling (ROM) ، Machine learning
نویسندگان
مراجع و منابع این مقاله:
لیست زیر مراجع و منابع استفاده شده در این مقاله را نمایش می دهد. این مراجع به صورت کاملا ماشینی و بر اساس هوش مصنوعی استخراج شده اند و لذا ممکن است دارای اشکالاتی باشند که به مرور زمان دقت استخراج این محتوا افزایش می یابد. مراجعی که مقالات مربوط به آنها در سیویلیکا نمایه شده و پیدا شده اند، به خود مقاله لینک شده اند :