Section: New Results

Handling non-rigid deformations

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

Elastic multi-modal registration

Our previous works about 3D tracking for deformable objects [1] are template-based methods and thus need to carefully register the model onto the image in the first image. In practice this task is performed manually and is especially difficult for deformable organs. Within J. Guevara’s PhD thesis, we have proposed an automatic method for registering pre and per-operative imagery which exploits the matching of the vascular trees, visible in most pre and intra-operative images. Although methods dedicated to non-rigid graph registration exist, they are not efficient when large intra-operative deformations of tissues occur. Our contribution is an extension of the graph-matching algorithm based on Gaussian process regression (GPR) proposed in [28]. Our idea is to combine GPR with a biomechanical model of the organ. Indeed, GPR allows for rigorous and fast error propagation but is extremely versatile and may thus produce non physically coherent registration, while biomechanical transformations are slower to compute but are capable of handling non linear deformation while preserving their physical nature. They thus allow earlier incoherent matching hypotheses to be removed. Integrating the two approaches allows us to significantly improve the quality of the registration while reducing computation times. These contributions have been published in the IPCAI conference [20] and in the International Journal of Computer Assisted Radiology and Surgery [13].

Individual-specific heart valve modeling

We first finished up a feasibility study aiming at providing fast image-based mitral valve mechanical simulation from individualized geometry [19]. The method was demonstrated on one dataset which was interactively segmented. In order to extend the pipeline to any data, robust methods to segment the valve components are required. Within the context of D. Panicheva's PhD thesis, we are currently working on means allowing to automatically segment the chordae. Valve chordae are generalized cylinders: Instead of being limited to a line, the central axis is a continuous curve. Instead of a constant radius, the radius varies along the axis. Most of the time, chordae sections are flattened ellipses and classical model-based methods commonly used for vessel enhancement or vessel segmentation fail. In this contribution, we exploit the fact that there are no other generalized cylinders than the chordae in the CT scan and we propose a topology-based method for chordae extraction. This approach is flexible and only requires the knowledge of an upper bound of the maximum radius of the chordae. The method has been tested on three CT scans. Overall, non-chordae structures are correctly identified and detected chordae ending points match up with actual chordae attachment points.

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

When intensive care patients are subjected to mechanical ventilation, the ventilator causes damage to the muscles that govern the normal breathing, leading to Ventilator Induced Diaphragmatic Dysfunction (VIDD). The INVIVE project aims to study the mechanics of respiration through numerical simulation in order to learn more about the onset of VIDD. We propose to use a meshfree RBF method. During this year, we have worked on building an implicit representation of the surface of the diaphragm based on the topology coming from last year researches. It has been associated with a linear elasticity model and the boundary conditions have been measured on landmark points extracted by medical experts.

3D catheter navigation from monocular images

In interventional radiology, the 3D shape of the micro-tool (guidewire, micro-catheter or micro-coil) can be very difficult, if not impossible to infer from fluoroscopy images, which may have an impact on the clinical outcome of the procedure. This question is considered as a single view 3D curve reconstruction problem. We follow a constrained non-rigid shape from motion approach, using a physics-based model as a prior for the object shape. The navigation model is implemented through interactive simulation that provides a prediction of the device, taking into account non-linear effects such as friction during contacts.

Raffaella Trivisonne started her PhD thesis in November 2015 (co-supervised by Stéphane Cotin, from MIMESIS team in Strasbourg) to address this research topic. An unscented kalman filter is used as a fusion mechanism of the model with image data (opaque markers placed along the device). Progress has been made this year towards an effective formulation of the filter. In particular, a good estimate is recovered for the device shape in the case of ambiguous views (overlapping anatomy, bifurcations).

The method has been implemented in Sofa simulation software platform. Validation has been performed on porcine in-vivo data, acquired in accordance with UE norms, in collaboration with Pr. Mario Gimenez and Dr. Alain Garcia from IHU-Strasbourg.

In-vivo procedures will be performed under ethical approval of MSER (reference to ethic protocol APAFIS #15433-2018060815283960).

Markerless similarity metrics were investigated during Juan Rocha's Master's thesis. The update equations of the filter were generalized to tackle curve to image similarity metrics, traditionally used in multi-view reconstruction methods.