Icon Legend

This session is not in your schedule.

This session is in your schedule. Click again to remove it.

Biomedical Imaging and Instrumentation

Biomedical Imaging and Instrumentation - Poster Session C & D

U46 - Deformable Image registration using high-order finite element geometry: application to tracking heart contours

Friday, October 10, 2025

10:00 AM - 11:00 AM PDT

Location: Exhibit Hall F, G & H

Presenting Author(s)

Walter O'Dell, PhD

Associate Professor
University of Florida
Gainesville, Florida, United States

Primary Investigator(s)

Walter O'Dell, PhD

Associate Professor
University of Florida
Gainesville, Florida, United States

Co-Author(s)

MP

Minjun Park

High School Student
University of Florida Student Science training Program, United States
AT

Alexander Tamayo

High School Student
University of Florida Student Science training Program, United States
Carli Smith

Undergraduate Student Researcher
University of Florida, United States

Introduction: :

Background: Deformable image registration (DIR) has many uses in medicine and engineering; one being automatic tracking of heart left ventricular boundaries during contraction which can be used to achieve automatic heart segmentation. The most common DIR spatial interpolation approach is 2D and 3D B-spline grids. The grid is defined on the deforming image. Each control point in the grid is shifted in space, warping the underlying image, with the goal to maximize a similarity metric between the warped and reference images. A limitation with grids is that each rectangular element can contain tissues with wildly different volume changes. In the heart, these are the volume-conserving myocardium and blood pool which typically undergoes a 60-70% volume decrease.

Hypothesis: Performing DIR using elements whose shapes match the left ventricular (LV) myocardium and cavity geometries will result in improved tracking of the LV wall.

Materials and

Methods: : A representative MRI 4-chamber view of human heart was acquired using standard procedures. The endocardial and epicardial borders were manually segmented in the end-diastolic (ED: filled) and end-systolic (ES: contracted) states. A high-order finite element (FE) mesh was generated from the segmented borders in the ED state. At each node in the FEM, the 2D location and slope of the border through that point are specified, thus there are 3 degrees of freedom per node. Cubic-Hermit basis functions were used for spatial interpolation within each element. A single central point in the LV cavity was used to define 3-sided elements that extended to the endocardial nodes. An outer ring of elements was defined to complete the mesh. During DIR, we adjusted just the endocardial and epicardial nodes. To assess wellness, upon completion, the manually drawn contour points in the ED state were warped onto the ES state. A high-density of ES points were computed from the manually-drawn ES points. The root-mean-square (RMS) miss between the tracked ED points and the high-density ES point set. For comparison, B-spline grids of 6x6, 8x8, and 10x10 elements were applied in the same manner. In a second experiment, binary masks of the myocardium were generated from the manually drawn contours and DIR performed on the masks using the same FEM and B-spline grids. The Dice coefficient and Jaccard Index was computed for each run.

Results, Conclusions, and Discussions::

Results: The FE-based DIR achieved an RMS error in tracking the endocardial border points of 1.24 mm. That for the 6x6 B-spline grid was 3.65 mm and for the 8x8 grid was 1.36 mm. The degrees of freedom for the 6x6 grid was 98; for the 8x8 grid was 162; and for the FEM was 60. The FE-based DIR gave lower RMS even though the B-spline grids had a larger number of degrees of freedom. The binary results were also superior for the FE mesh.

The FE-based Dir was advantaged by having elements that contain only one type of tissue, being either the nearly incompressible myocardium, or the cavity blood pool. The contracted appearance of the deformed cavity elements exemplifies this. Note that because the myocardial tissue seen in the ED image is not the same tissue seen in the ES image because of heart translation out of the imaging slice, we are not tracking myocardial points in space. Future extensions are to bring this into 3D, where true tracking of points would be possible. The ultimate goal is to facilitate fully automatic heart segmentation across times and across patients, to bring research methods like cardiac tagging into routine clinical use.

Acknowledgements and/or References (Optional): : Funding support for MP and AT provided by the UF Student Science Training Program (SSTP). The MR images were acquired under a project funded by the Florida Breast Cancer Foundation.