Peter Rogelj

Assessment of 3D DCE-MRI of the kidneys using non-rigid image registration and segmentation of voxel time courses.

We have applied automated image analysis methods in the assessment of human kidney perfusion based on 3D dynamic contrast-enhanced MRI data. This approach consists of non-rigid 3D image registration of the moving kidney followed by k-means clustering of the voxel time courses with split between left and right kidney. This method was applied to four data sets acquired from healthy volunteers, using 1.5 T (2 exams) and 3 T scanners (2 exams). The proposed registration method reduced motion artifacts in the image time series and improved further analysis of the DCE-MRI data. The subsequent clustering to segment the kidney compartments was in agreement with manually delineations (similarity score of 0.96) in the same motion corrected images. The resulting mean intensity time curves clearly show the successive transition of contrast agent through kidney compartments (cortex, medulla, and pelvis). The proposed method for motion correction and kidney compartment segmentation might improve the validity and usefulness of further model-based pharmacokinetic analysis of kidney function in patients.

Symmetric image registration.

Abstract This paper presents an original non-rigid image registration approach, which tends to improve the registration by establishing a symmetric image interdependence. In order to gather more information about the image transformation it measures the image similarity in both registration directions. The presented solution is based on the interaction between the images involved in the registration process. Images interact through forces, which according to Newton’s action–reaction law form a symmetric relationship. These forces may transform both of the images, although in our implementation one of the images remains fixed. The experiments performed to demonstrate the advantages of the symmetric registration approach involve the registration of simple objects, the recovery of synthetic deformation, and the inter-patient registration of real images of the head. The results show that the symmetric approach improves both the registration consistency and the registration correctness.

Point similarity measures for non-rigid registration of multi-modal data

High-dimensional non-rigid registration of multi-modal data requires similarity measures with two important properties: multi-modality and locality. Unfortunately all commonly used multi-modal similarity measures are inherently global and cannot operate on small image regions. In this paper, we propose a new class of multi-modal similarity measures, which are constructed from information of the whole images but can be applied pointwise. Due to their capability of measuring correspondence for individual image points we call them point similarity measures. Point similarity measures can be derived from global measures and enable detailed relative comparison of local image correspondence. We present a set of multimodal point similarity measures based on joint intensity distribution and test them as an integral part of non-rigid multi-modal registration system. The comparison results show that segmentation-based measure, which models the joint distribution as a sum of intensity classes, performs best. When intensity classes do not exist or cannot be accurately modeled, each intensity pair can be treated as a separate class, which results in a more general measure, suitable for various non-rigid registration tasks.

Validation of a nonrigid registration algorithm for multimodal data

We describe the evaluation of a non-rigid image registration method for multi-modal data. The evaluation is made difficult by the absence of gold standard test data, for which the true transformation from one image to another is known. Different approaches have been used to deal with this deficiency, e.g., by using synthetically warped data, by comparison of anatomic regions of interest identified either manually or automatically, and by direct comparison of the registered data. Each of these approaches are limited and in this paper, we illustrate some of the problems that arise based on their application to the evaluation of our multi-modal non-rigid registration method.

Symmetric image registration

A quality of image match is usually estimated by measuring image similarity. Unfortunately, similarity measures assess only such transformations that change appearance of the deformed image, and in the case of non-rigid registration the results of the similarity measurement depend on the registration direction. This asymmetric relation leads to registration inconsistency and reduces the quality of registration. In this work we propose a symmetric registration approach, which improves the registration by measuring similarity in both registration directions. The solution presented in this paper is based on the interaction of both images involved in the registration process. Images interact with forces, which are according to the Newtons action-reaction law forming a symmetric relationship. These forces may transform both of the images, although in our implementation one of the images remains fixed. The experiments performed to demonstrate the advantages of the symmetric registration approach involve registration of simple objects, recovering synthetic deformation, and interpatient registration of real images of head. The results show improvements of registration consistency and also indicate the improvement of registration correctness.

Comparison of 3D MRI with high sampling efficiency and 2D multiplanar MRI for contouring in cervix cancer brachytherapy

Comparison of 3D MRI with high sampling efficiency and 2D multiplanar MRI for contouring in cervix cancer brachytherapy Background. MRI sequences with short scanning times may improve accessibility of image guided adaptive brachytherapy (IGABT) of cervix cancer. We assessed the value of 3D MRI for contouring by comparing it to 2D multi-planar MRI. Patients and methods. In 14 patients, 2D and 3D pelvic MRI were obtained at IGABT. High risk clinical target volume (HR CTV) was delineated by 2 experienced radiation oncologists, using the conventional (2D MRI-based) and test (3D MRI-based) approach. The value of 3D MRI for contouring was evaluated by using the inter-approach and inter-observer analysis of volumetric and topographic contouring uncertainties. To assess the magnitude of deviation from the conventional approach when using the test approach, the inter-approach analysis of contouring uncertainties was carried out for both observers. In addition, to assess reliability of 3D MRI for contouring, the impact of contouring approach on the magnitude of inter-observer delineation uncertainties was analysed. Results. No approach- or observer - specific differences in HR CTV sizes, volume overlap, or distances between contours were identified. When averaged over all delineated slices, the distances between contours in the inter-approach analysis were 2.6 (Standard deviation (SD) 0.4) mm and 2.8 (0.7) mm for observers 1 and 2, respectively. The magnitude of topographic and volumetric inter-observer contouring uncertainties, as obtained on the conventional approach, was maintained on the test approach. This variation was comparable to the inter-approach uncertainties with distances between contours of 3.1 (SD 0.8) and 3.0 (SD 0.7) mm on conventional and test approach, respectively. Variation was most pronounced at caudal HR CTV levels in both approaches and observers. Conclusions. 3D MRI could potentially replace multiplanar 2D MRI in cervix cancer IGABT, shortening the overall MRI scanning time and facilitating the contouring process, thus making this treatment method more widely employed.

Local similarity measures for multimodal image matching

In this paper we focus on local similarity measures based on Shannon entropy which can be used for multimodal image matching employing deformations. The advantage of our approach is that global similarity or similarity of a larger image region can be computed from the similarities of its constitutive parts or individual voxels. We also discuss the interpolation artefacts in entropy based similarity measures caused by linear and partial volume interpolation.

Point Similarity Measure Based on Mutual Information

Registration of multi-modality images requires similarity measures that can deal with complex and unknown image intensity dependencies. Such measures have to rely on statistics, and consequently, they require relatively large image regions to operate. This makes the detection of localized image discrepancies difficult. As a solution we propose point similarity measures, which can measure similarity of arbitrarily small image regions, including similarity of individual image points. In this paper we present a point similarity measure derived from the mutual information. In addition to its extreme locality it can also avoid the interpolation artifacts and improve the spatial regularization to better suit the spatial deformation model.

TEXTURE FEATURE BASED IMAGE REGISTRATION

The aim of our research is to analyse the importance of texture information for registration of a DRR (digital reconstructed radiograph) and EPI (electronic portal image) medical images. In our research, texture features are extracted by Laws texture coefficients and used for computing registration criterion functions. The proposed feature based approach is compared to the commonly used approach, where a registration criterion function is computed directly from intensity features, i.e. grey values. For this purpose we observed accuracy of registration, the distinctiveness of local extrema and the distinctiveness of a global extremum of the criterion functions. These parameters are essential to achieve a correct image alignment. Our results show that for the given image modalities we can expect more robust and more correct registration when texture based criterion function instead of intensity based one is used

Similarity measures for nonrigid registration

Non-rigid multimodal registration requires similarity measure with two important properties: locality and multi- modality. Unfortunately all commonly used multimodal similarity measures are inherently global and cannot be directly used to estimate local image properties. We have derived a local similarity measure based on joint entropy, which can operate on extremely small image regions, e.g. individual voxels. Using such small image regions reflects in higher sensitivity to noise and partial volume voxels, consequently reducing registration speed and accuracy. To cope with these problems we enhance the similarity measure with image segmentation. Image registration and image segmentation are related tasks, as segmentation can be performed by registering an image to a pre-segmented reference image, while on the other hand registration yields better results when the images are pre-segmented. Because of these interdependences it was anticipated that simultaneous application of registration and segmentation should improve registration as well as segmentation results. Several experiments based on synthetic images were performed to test this assumption. The results obtained show that our method can improve the registration accuracy and reduce the required number of registration steps.