Heart Finite Element Method (FEM) approach for automatic heart segmentation and deformation modeling
Background and Significance
My aim is to develop a FEM that can be used to first automatically segment the heart in a 3D set of MRI images, and then to extend this to estimate elastic deformation at any point in the ventricles. These results allow quantitative analysis of regional wall dynamics and wall stress.
Breast cancer treatment usually involves chemotherapy and radiation therapy to kill any small, unseen islands of cancer that may have spread from the primary tumor to the chest wall or adjacent lymph nodes. Although this improves overall survival, it can also inadvertently damage the heart. Recent studies have shown that breast cancer patients who receive radiation have higher rates of cardiotoxicity and that patients with left-sided breast cancer suffer more heart-related deaths than patients with right-sided breast cancer. Our lab has developed methods to accurately compute changes global and regional heart function at early time points after radiation treatment, but the most prominent obstacle to bringing such technology to the clinic is the laborious and time-consuming steps of accurately segmenting the heart in each view and at each time point.
Our hypothesis is that we can combine the 3D geometrical modeling of FEM with deformable image registration to fully automatically segment the heart from a stack of 3D MR images in multiple views. A simple binary mask of the heart generated from the FEM model will be simulated on each imaging slice, and warped by perturbing the FEM nodes and nodal parameters. Each nodal parameter will be iteratively adjsuted to obtain an optimal match to the patient's MRI data. Having achieved this, then I propose to apply the FEM geometry to model the deformation of the myocardial tissue by deformable image registration to 3D tagged MR images of the left ventricle.
We are acquiring cardiac magnetic resonance (CMR) images in patients with left-sided breast cancer undergoing breast or chest-wall photon or proton therapy.
As a first step, our specific aims are to:
i. Generate a 3D FEM geometry for a given patient's heart based on manually defined contour points on the LV.
ii. Create a set of simulated myocardial mask images for each MR slice and view.
iii. Perturb the FEM nodal parameters and have the simulated mask images be warped likewise, and compare the similarity of the mask images to the masked images of the patient's heart.
iv. Apply this method to automatically generate a FEM geomatric model for each subsequent time point in the same patient.
v. Apply the first patient's heart FEM model and generated mask images at the referecne time point to fit to the heart MRI of a second patient at their reference time point.