|Title:||Shape-based modeling of aneurysmal disease status|
|Authors :||Juchler, Norman|
|Conference details:||ECCOMAS Congress 2016, Minisimposium on Aneurysms: Solid Mechanics, fluid mechanics and mechanobiology, Heraklion, Greece, 5-10 June 2016|
|License (according to publishing contract) :||Licence according to publishing contract|
|Type of review:||Peer review (Abstract)|
|Subjects :||Shape-based risk assessment; Intracranial aneurysms|
|Subject (DDC) :||003: Systems |
616: Internal medicine and diseases
|Abstract:||Introduction. To date, it is difficult for clinicians to judge the associated risks of intracranial aneurysms reliably. Because the 3D-shape of an aneurysm is strongly linked to the underlying formation processes it is very likely that the presence or absence of certain shape features indicate the disease status of the aneurysm. Shape (extracted from medical imaging data) already plays a significant role in the qualitative assessment of the aneurysm by the clinician and has been associated with risk prediction. Still, no consensus exists about which shape features reliably predict instability or whether there exist any that qualify as biomarkers at all. In an effort to find support for the assumption that aneurysm shape carries information about the aneurysm status, we developed a flexible classification pipeline that extracts shape features and tests their applicability. Methods. 3D models of aneurysms are extracted from medical imaging data (mostly 3DRA) by a standardized vessel segmentation method. The aneurysm is cut from its parent vessels according to a simple and reproducible cut protocol. Different representations of the 3D shape that have been suggested by literature are calculated for the extracted aneurysm. So far we have looked at Zernike moments (ZM), their invariants (ZMI) and simple geometry indices such as undulation or non-sphericity. Different feature reduction techniques (for ZMI) and machine-learning methods are applied to find linking patterns between shape features and aneurysm stability. This processing pipeline is applied to a relatively large clinical dataset (ca. 400 cases), whereas the collection of new cases is an on-going effort. We will present the findings on the latest state of the database. Results. Initial results indicate that the ZMI alone are insufficient to classify aneurysms reliably in terms of rupture or stability status. Because of their global support, ZMI bare a strong dependency on the placement of the cuts. In a preliminary study, ZMI were only slightly superior regarding rupture prediction compared to much simpler geometry indices. Conclusions. We implemented a framework to test the performance of different combinations of shape features and machine learning techniques. While the shape-only representation of an aneurysm does carry information about its disease status, the classification results will benefit from the stratification of the aneurysms in terms of location, size and clinical factors.|
|Departement:||Life Sciences and Facility Management|
|Organisational Unit:||Institute of Applied Simulation (IAS)|
|Publication type:||Conference Other|
|Appears in Collections:||Publikationen Life Sciences und Facility Management|
Files in This Item:
There are no files associated with this item.
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.