This application claims priority of German application No. 10 2011 080 905.8 filed Aug. 12, 2011, which is incorporated by reference herein in its entirety.
The present application relates to a method for visualizing the registration quality of medical image datasets. Additionally the present application relates to a corresponding device for visualizing the registration quality of medical image datasets.
In interventional radiology, in the evaluation of a case, image datasets of different modalities, such as for example 2D and 3D x-ray recordings (XA), computed tomography (CT), magnetic resonance tomography (MR), positron emission tomography (PET), single-photon emission computed tomography (SPECT) or sonography (US), are often displayed merged, i.e. overlaid. In this way a physician can combine the respective advantages of the various methods for a specific case. Ideally only elementary digital image processing operations, such as rotations or displacements, are necessary to overlay two images, in order to bring the images correctly into congruence. This is known as rigid registration of two recordings or image datasets. In practice the recordings to be overlaid are often made at different times or with different camera positions and hence may display displaceable or flexible objects, such as blood vessels, organs or muscles, in a different position or shape. For example, respiratory artifacts or different arm positions during two recordings may be the reason why a blood vessel or an organ, e.g. the liver, assumes a different position in a first image than in a second image. Rigid registration would in this case provide an unsatisfactory overlay quality. In such cases a non-rigid, flexible or elastic registration is often used, which transfers objects in an image to the corresponding objects in a different image. When using a non-rigid registration algorithm the fact that it is difficult for someone examining the registration result, i.e. the merged image, to evaluate which part of an image has been more intensely and which less intensely “elastically” transformed can be problematic. If for example a region of an image which is important for a diagnosis is intensely changed by a non-rigid registration algorithm, this may result in an incorrect evaluation with far-reaching consequences.
The object of the present application is thus to specify a method for visualizing the registration quality of medical image datasets.
The application achieves this object with a method for visualizing the registration quality of medical image datasets with the features of the first independent claim and a device for visualizing the registration quality of medical image datasets with the features of the second independent claim.
The basic idea underlying the application is a method for visualizing a registration quality of medical image datasets, which comprises the following method steps:
In the first method step a reference image and an object image are acquired. For this purpose an imaging method referred to in the introduction can be used. Both images can be obtained using the same or different imaging methods. The images may be spatial, i.e. three-dimensional, images, two-dimensional images or two-dimensional sectional images of a 3D image dataset. Image and image dataset are used as synonyms below. In the case of a spatial image the image points are referred to as voxels, and in two dimensions as pixels. The reference image and the object image generally comprise different image sections of an examination object, e.g. of a person or an animal, but the intersecting set is not empty.
In the second method step the reference image is registered with the object image with the aid of a registration method to form a merged image. Image registration of two images refers to a method in digital image processing, with the aid of which two images of an at least similar scene are made to match one another as closely as possible. The reference image is not changed. For the object image a transformation is determined which adjusts the object image as closely as possible to the reference image. What is involved here is thus a problem of optimization. Image registration is a common task in medical image processing for which there are numerous proposed solutions. Examples of usable optimization methods for registration methods that can be cited are gradient descent methods, downhill simplex methods, hillclimb methods and simulated annealing. In general non-elastic and elastic registration methods can be distinguished. A non-elastic registration method means an image registration method in which rigid transformations such as translation and rotation, affine transformations such as scaling and shearing, and projective transformations can be used. An elastic registration method means an image registration method in which elastic transformations (also called “non-rigid transformations”) such as spline-based or polynomial-based transformations, can be applied. The disclosed second method step uses a registration method which comprises a non-elastic and an elastic registration method section. This means that rigid, affine and projective transformations, where these are applied in the registration method, can be assigned to the non-elastic registration method section. It is also conceivable that displacement, rotation and size adjustment are performed by an operator. Elastic transformations can be assigned to the elastic registration method section. Instead of one registration method containing a non-elastic and an elastic registration method section a non-elastic registration method and subsequently an elastic registration method can be applied.
In the third method step a deformation field is determined, wherein the deformation field comprises displacement vectors from image points of the object image to image points of the reference image which are conditioned by the elastic registration method section. The deformation field can be a matrix of vectors which relocates each image point of the object image such that optimum conformity with the reference image is achieved.
In the fourth method step at least one region of the merged image is overlaid with an overlay, into which parameters of the deformation field are fed. Parameters, e.g. distance dimensions or derivations which determine the overlay or the elements thereof, are thus calculated from the deformation field. The overlay is then overlaid coincidently over a region of the merged image or the whole merged image.
In the fifth and last method step the at least one region of the merged image and the overlaid overlay are visualized. Visualization of the at least one region of the merged image can refer to the display of the region of the merged image e.g. on a monitor. The overlay, the elements of which, as described, are determined by parameters, such as distance dimensions, of the deformation field, is graphically overlaid over this region of the merged image.
The overlay comprises a matrix, the elements of which are determined from the Jacobi matrix of the deformation field. The derivations or gradients of the deformation field are thus fed into the overlay, so that the elements of the overlay are a measure of the intensity of the adjustment or of the deformation of objects of the object image to the reference image.
In another embodiment the overlay comprises a matrix, the elements of which are determined from the length of assignment vectors which describe the assignment of the image points of the reference image from the image points of the object image. The displacement vectors from the deformation field and the length parameter of the displacement vectors from the deformation field are fed into the overlay. The lengths or the absolute values of the assignment vectors are a measure of the intensity of the adjustment or of the deformation of objects of the object image to the reference image.
An embodiment of the application provides that the visualization of the at least one region of the merged image and of the overlaid overlay comprises a color coding of the elements of the overlay, wherein predefinable value ranges are assigned to a predefinable color palette or wherein the value range of the elements of the overlay is assigned to a predefinable color palette. Assuming an overlay matrix, the elements of which indicate to what extent the image points of the object image have been displaced by the elastic registration method section, in order to bring them into conformity with the corresponding image points of the reference image, a display of the intensity of the deformation which is intuitively understandable for a human observer is enabled thanks to color coding. Known color schemes are used, such as transparency for no deformation, green for a slight deformation, through yellow to red for a large deformation. The assignment of the degree of deformation and color can be predefined by a user. In this case the assignment can be predefined absolutely or a predefinable color palette extends in each case across the value range of the overlay, i.e. for example a deformation intensity of 0 to 10% of the maximum intensity of deformation is assigned no color or transparency, 10% to 20% the color green, etc., up to the color red for 90% to 100% of the maximum intensity of deformation.
In another embodiment at least one region of interest in the reference image can be predefined, by an operator, which in accordance with the non-elastic registration method section of the registration method of reference image and object image is taken into account by the elastic registration method section of the registration method. In diagnostics it can be advantageous if only one region of interest (ROI) is taken into account by the elastic registration method section of the registration method. The method step is thus represented as follows: one or more regions of interest are marked in the reference image, e.g. by an operator. A non-elastic registration of reference image and object image is followed by an elastic registration, which is however limited to the region or regions of interest. The overlay and the visualization thereof with the merged image consequently comprise only the regions of interest. This embodiment can be referred to as a punching operation, as the regions of interest are as it were “punched” out of the reference image and only these undergo elastic registration.
It is conceivable for the reference image and the object image to comprise 3D image datasets and for the registered image to be visualized as a 2D sectional image with the overlaid overlay.
Another basic idea underlying the application relates to a device for visualizing the registration quality of medical image datasets. The device comprises a reception means for receiving an acquired reference image and an acquired object image. The device further comprises at least one calculation means for registering the reference image with the object image to form a merged image with the aid of a registration method, which comprises a non-elastic and an elastic registration method section, and for determining a deformation field, wherein the deformation field comprises displacement vectors from image points of the object image to image points of the reference image, which are conditioned by the elastic registration method section, and for overlaying at least one region of the merged image with an overlay, into which parameters of the deformation field are fed. The device further comprises a display means for visualizing the at least one region of the merged image and of the overlaid overlay.
An embodiment of the application provides for the calculation means and the display means of the device to be designed to execute a previously described method.
Advantages which may be achieved from using one of the disclosed methods or one of the disclosed devices may be:
The embodiments depicted below represent embodiments of the present application. Other developments emerge from the following figures plus description, in which:
Finally,
Number | Date | Country | Kind |
---|---|---|---|
102011080905.8 | Aug 2011 | DE | national |