A variational approach to multi-modal image matching

Abstract

Nous considérons le problème du recalage non-rigide entre images de modalités différentes. Nous proposons un cadre général qui repose sur une formulation variationnelle, que nous appliquons sous la forme de trois algorithmes de recalage multimodal : recalage supervisé par apprentissage de la loi jointe, maximisation de l'information mutuelle, et maximisation du rapport de corrélation. Pour permettre une régularisation de la solution, nous utilisons un opérateur issu de la théorie de l'élasticité. Nous considérons aussi une méthode de régularisation avec préservation des contours. Des résultats expérimentaux préliminaires sur des images synthétiques et des données IRM sont présentés.

Feasibility of free-breathing dynamic contrast-enhanced MRI of the abdomen: a comparison between CAIPIRINHA-VIBE, Radial-VIBE with KWIC reconstruction and conventional VIBE

“…5 Motion correction of the DCE-MRI scan was performed by using a nonrigid registration technique. 8,9 In this study, we did not use T 1 mapping for the reason that T 1 values obtained from T 1 mapping may be inaccurate or some errors my occur in the process of coregistration of T 1 value to the dynamic data. We aimed to focus on the effect of free-breathing dynamic series on image quality.…”

“…Perfusion maps were computed based on Tofts model using a population-averaged arterial input function with two exponentials. [8][9][10] Image analysis Qualitative analysis Two abdominal radiologists (NS and KWK) with 7 and 13 years' experience in liver imaging independently analyzed the quality of perfusion maps by using a 5-point scale as follows: 1 5 uninterpretable; 2 5 poor; 3 5 fair; 4 5 good; and 5 5 very good. The rating was based on the registration quality of the perfusion maps, the presence of misregistration artefacts and any architectural distortion of the organ.…”

Correspondence Transfer for the Registration of Multimodal Images

“…The joint histograms of I 1 •w and I 2 within their region of overlap are constructed by binning the corresponding intensity pairs (I 1 (w(x)), I 2 (x)), and the marginal histograms are obtained by integrating over rows and columns, respectively. Substituting (12) into (11) and rearranging following [4,9], yields the first variation of Φ M I :…”

A minimum description length objective function for groupwise non-rigid image registration

“…There are a myriad of methods of automatic non-rigid image registration of pairs of images (e.g., [1][2][3][4]); see [5] for a review of general image registration algorithms, not limited to medical images. Such algorithms typically involve two independent choices: the objective function, the extremum of which defines what is meant by the 'best' correspondence between the images, and the representation of the deformation field that defines the dense correspondence between the images.…”