3DStrain ANR Project

Multimodal quantification and validation of 3D regional myocardial function

This ANR project started in 2011 for 48 months. I’m working on the data normalization part. Here is a presentation I did, before we have any real data, giving hints on what we could do (in french).

Normalisation non linéaire des déplacements et du strain par les méthodes à noyaux

Recall of the project description (from the ANR website).

The aim of the project is to design a generic formalism for parietal and regional tracking of the left ventricle (LV) and to adapt it to the 3D+t cardiac imaging modalities used in clinical routine (echocardiography, gated-SPECT, cine-MRI). The estimated displacement field must be reliable enough and insensitive to various artifacts to assess regional myocardial function in 3D from the accurate and precise computation of strain.
The strain has recently proved to be of great interest for diagnosis and prognostic in cardiology, but its interpretation remains delicate because of the relative nature of the indices; the clinical objective of the 3DStrain project is to bring answers about the knowledge of normality.
A continuous variational approach will allow to integrate spatial properties for segmentation and to take into account the temporal cues provided in image sequences. The problem of dynamic tracking is considered both in a regional way where we estimate a dense displacement field inside myocardium over successive frames and using matched endocardial and epicardial walls.
Each imaging modality will be beforehand decomposed in several components thanks to a Morphological Component Analysis. They will be integrated to the variational framework of segmentation thanks to classical boundary or regional Chan and Vese terms, but also by making the most of new approaches, either of kernel type in order to get a better separation between the moving object and the background, or of local type to take into account spatial heterogeneities.
Motion estimation for regional tracking includes several interesting and novel approaches. The problem will be expressed as a transport problem that does not assume gray level to be invariant as in optical flow, and allows us to take into account the contraction-dilation-torsion movement characteristics. The standard transport problem is ill-posed and requires an additional regularization term. However the proposed optimal transport problem is well-conditioned and equivalent to the minimal kinetic energy solution. Finally motion estimation will be coupled to segmentation.
The partnership with the Keosys company enables the design of specialized plugins on a collaborative multimodal platform to make the three medical fields using cardiac imaging (cardiology, nuclear medicine and radiology) cooperate more efficiently. Each plugin is centered around one medical field and one imaging modality (resp. echography, SPECT and MRI) with variants in terms of user interface. 3D interactive visualization of the results is a priority to insure transfer of 3D strain towards clinical practice, that must be visually assessed by the expert, and corrected if needed by virtual interaction with the endocardial and epicardial detected surfaces.
Clinical experts of the three imaging modalities of interest are involved in the project and will participate to the definition of specifications and to validation. Results will be checked against the method considered as a reference (MRI tagging) and to the more recent echocardiographic imaging technique for deformation, the 3D speckle tracking.