Skip to main content Skip to secondary navigation

Patient Specific Modeling

Main content start


Cardiovascular simulations need to be run on patient-specific geometries to achieve meaningful results. However, the process of generating three-dimensional models from medical images is time-consuming. We develop methods to generate patient-specific models with minimal user interaction. Moreover, we are interested in interactive ways of deforming already built models to explore surgical scenarios efficiently.

Data-driven methods for segmentation

Blood vessel segmentation. One of the steps to the generation of 3D blood vessel geometries is segmentation. In this often time-consuming process, human operators must manually identify the vessel lumen at numerous locations along its centerline. In our lab, we develop neural networks to identify the lumen of the vessel and automate the segmentation step.

Heart segmentation. Creating accurate patient-specific models of the heart from image data often requires significant time and human efforts and is one of the major bottlenecks limiting large-scale validations and clinical applications of personalized computational modeling of the heart function. We have been developing novel deep-learning algorithms to construct simulation-ready meshes from patient medical image data automatically. Our recent work can directly reconstruct computer meshes of the heart from 3D patient image data by training a neural network to deform a mesh template to match the patient image. We are currently developing methods to create meshes from images of patients with congenital heart diseases. 

Lab members involved in the project: Fanwei Kong, Luca Pegolotti, Karthik Menon, Noah Fields

Surgical Planning

As mentioned above, traditional methods for developing patient-specific cardiovascular models are tedious. Researchers must manually edit the pathlines and segmentations to model different vascular diseases or surgical options, which inherently change the geometry of the vasculature (e.g., stenoses in coronary artery disease or grafting in the Fontan procedure) of their respective model. To overcome these challenges, we are developing methods to enable the direct morphing of meshes. These methods will allow users to quickly change the geometry of their model to explore the geometric implications of cardiovascular diseases and surgical procedures.

Lab members involved in the project: Jonathan Pham