|Title:||Big Data : machine learning to identify shape biomarkers in intracranial aneurysm|
|Authors :||Hirsch, Sven|
|Conference details:||Swiss Society for Biomedical Engineering, Wintertthur, 30. August 2017|
|License (according to publishing contract) :||Licence according to publishing contract|
|Type of review:||Not specified|
|Subject (DDC) :||004: Computer science |
616: Internal medicine and diseases
|Abstract:||An intracranial aneurysm is a disease of the cerebral blood vessel wall resulting in the deformation and enlargement of the vascular lumen. If the process of deformation remains active, the vessel wall may either rupture and produce a hemorrhage, or thrombosis and ischemia may occur. Fig. 1 A typical small bifurcation aneurysm. Of all IAs 70 % are bifurcation type. Although statistically safe, some IAs do rupture. The physician needs to decide what to do for the individual patient. Obviously, an aneurysm should only be treated if the treatment is less risky than the aneurysm itself. So far, no accepted criteria exist for individual assessment of aneurysm stability and there are no clear treatment guidelines. Consequently, there is currently no validated tool to help predict development or treatment outcomes for an individual aneurysm and physicians rely solely on their personal judgment. Aneurysm Database The AneuX consortium collects a comprehensive number of patient data sets to estimate the disease status of intracranial aneurysms. The starting point is the hypothesis that vessel 3D-shape can be used as an image biomarker. Research and development in the field require massive information integration realized by a diverse community of scientists, physicians and engineers involved in better understanding the biological processes, and the development of new tools to manage and treat patients. The available cases are collected in a unified database containing imaging and clinical patient information on intracranial aneurysms. The AneurysmDataBase hosted by the Swiss Neuro Foundation  aspires to establish the standard for collecting and characterizing information about intracranial aneurysms. We develop web-based applications to inspect, analyze and display data for various users: clinicians, patients and industry. Machine Learning on 3D shapes As an aneurysm’s 3D-shape is strongly linked to the underlying formation processes, it is believed that the presence or absence of certain shape features mirror the disease status of the aneurysm wall. The shape of the aneurysm and its circumjacent arterial lumen already plays a significant role in the qualitative assessment of the aneurysm. Currently, clinicians associate irregularity with wall instability. However, no consensus yet exists about which shape features reliably predict instability or whether there exist any that qualify as biomarkers of the disease status at all. Here we present a classification pipeline that allows us to identify shape features with the highest predictive power of aneurysm instability. 3D models of aneurysms are extracted from medical imaging data (mostly 3D rotational angiography) using a standardized vessel segmentation protocol. Subsequently, the aneurysm and adjacent segments of parent vessels are cut from the lumen replica of the vascular tree. A variety of established representations of the 3D shape are calculated for the extracted aneurysm segment. These include the calculation of Zernike moments (ZM), their invariants (ZMI) and simple geometry indices such as undulation, ellipticity or non-sphericity. Different feature reduction techniques (for ZMI) and machine-learning classification methods are applied to find linking patterns between shape features and aneurysm disease status. Conclusions We will present a machine learning framework to identify imaging biomarkers for intracranial aneurysm. These biomarkers are used for assessing the stability of an aneurysm and finally weighed against an interventional risk to propose the best treatment strategy for a patient. The AneurysmDataBase is pivotal to realize studies on larger cohorts and we are presenting the current state and vision of this disease management platform. The database provides statistical knowledge about a lesion site and can e.g. serve design of new devices, by providing geometries for sizing of a device or for conducting Computational Fluid Dynamics.|
|Departement:||Life Sciences und 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.