Section: New Results

Handling non-rigid deformations

Participants : Marie-Odile Berger, Jaime Garcia Guevara, Daryna Panicheva, Pierre-Frédéric Villard.

Elastic multi-modal registration

Image-guided hepatic surgery is progressively becoming a standard for certain interventions. However, requirements on limited radiation dosis result in lower quality images, making it difficult to localize tumors and other structures of interest. Within J. Guevara's PhD thesis, we have proposed an automatic registration method exploiting the matching of the vascular trees, visible in both pre- and intra-operative images. The graphs are automatically matched using an algorithm combining Gaussian Process Regression and biomechanical model [20]. Indeed, Gaussian Process regression allows for a rigorous and fast error propagation but is extremely versatile. On the contrary, using biomecanical transformations is slower but provides physically correct hypotheses. Integrating the two approaches allows us to dramatically improve the quality of the matching for moderate or large organ deformations while reducing significantly the computational cost.

Individual-specific heart valve modeling

In this work, we focused on the segmentation of the valve cords. As dataset, we used 8 CT images of porcine hearts. Those data were acquired during various times with a microCT scan machine.

Within D. Panicheva's Master thesis, we worked on modeling the mitral valve chordae by applying a RANSAC-based method designed to extract cylinders with elliptical basis from a set of 3D contour points. To limit the search area, the results of segmentation obtained with classical methods for tubular structures extraction were used as initial assumptions of cords location.

The proposed method allows us to significantly improve cords segmentation results compared with classical methods, in particular, the section size and the endpoints of the cords are accurately defined which is important for future mechanical modeling of the mitral valve.

INVIVE: The Individual Virtual Ventilator: Image-based biomechanical simulation of the diaphragm during mechanical ventilation

The motivation for the project is the serious medical condition, called ventilator induced diaphragmatic dysfunction (VIDD). During mechanical ventilation, air is pushed into the lungs resulting in a passive displacement of the diaphragm. This unnatural forcing results in loss of function in the muscle tissue. Our goal is to develop a simulator that allows for an in-silico exploration of the respiratory function with and without mechanical ventilation in combination with intervention measures that can reduce or prevent the risk for VIDD in the patients.

In the first year of the project, we worked on extracting a mesh from the segmented medical data that includes the boundary conditions. This work relies on analyzing the physiological constrains (rib motions, thoracic, abdominal and lung pressures) that can be measured.

We also worked on a method to solve differential equations on a complex geometric domain using the Radial Basis Function Partition of Unity Collocation Method (RBF-PUM). To use RBF-PUM for solving differential equations a covering of the domain has to be formed. The test implemented Poisson's diffusion equation on a domain defined by the diaphragm geometry. The diaphragm is not an easy case due to its thickness and shape.

This work is funded by the Swedish Research Council and realized within a collaboration with Uppsala University.