Image processing method and medical system

Abstract

For a subjectively pleasant representation of a medical image sequence without loss of structures, a method is provided for processing an x-th image of a medical image sequence from at least two consecutive images, in particular X-ray images, of an imaging unit, comprising the following steps: capturing the x-th image of the image sequence from at least two consecutive images, performing a variance-stabilizing transformation of the x-th image, performing spatial denoising of the variance-stabilized x-th image, subtracting the denoised variance-stabilized x-th image from the variance-stabilized x-th image and thereby obtaining an x-th noise component of the x-th image, performing inverse variance stabilization of the denoised variance-stabilized x-th image and thereby obtaining a denoised x-th signal component of the x-th image,Performing a temporal filtering of the x-th noise component and thereby obtaining a temporally filtered x-th noise component of the x-th image, adding the temporally filtered x-th noise component with the denoised x-th signal component of the x-th image to an x-th result image, and displaying the x-th result image.

Description

The invention relates to a method for processing an x-th image of a medical image sequence according to patent claim 1, to a method for processing an x-th image of a medical image sequence according to patent claim 2, and to a medical system for carrying out such a method according to patent claim 8.In medical imaging, particularly in x-ray images, noise reduces the quality of the images. In particular, in the recording and display of image sequences which are to be recorded with the smallest possible X-ray dose in order to conserve the patient and the medical personnel, noise is a great problem. When displaying such image sequences, for example during interventional interventions, the entire image impression can be distorted by noise for the viewer, so that in the worst case errors can occur during the intervention.The noise is in a first approximation time uncorrelated. However, it has been found that by suitable temporal filtering of the images (e.g. using filtering based on a moving average or moving exponential weighting), the noise is correlated in time, which allows the viewer to present a subjectively pleasing representation. However, temporal filtering also has the disadvantage that contrasts of structures imaged on the images can be lost.It is an object of the present invention to provide an image processing method which enables a representation that is pleasant for the viewer with the least possible loss of contrast; furthermore, it is an object of the invention to provide a medical system suitable for carrying out the method.The object is achieved according to the invention by a method for processing an x-th image of a medical image sequence according to patent claim 1, by a method for processing an x-th image of a medical image sequence according to patent claim 2 and by a medical system according to patent claim 8.The method according to the invention for processing an x-th image of a medical image sequence of at least two successive images, in particular X-ray images, of an imaging unit comprises the following steps: recording or providing the x-th image of the image sequence of at least two successive images, carrying out a variance-stabilizing transformation of the x-th image, carrying out a spatial denoiseing of the variance-stabilized x-th image, subtracting the denoised variance-stabilized x-th image from the variance-stabilized x-th image and thereby obtaining an x-th noise component of the x-th image, carrying out an inverse variance stabilization of the denoised variance-stabilized x-th image and thereby obtaining a denoised x-th signal component of the x-th image, performing temporal filtering of the x-th noise component and thereby obtaining a temporal filtered x-th noise component of the x-th image, adding the temporal filtered x-th noise component with the denoised x-th signal component of the x-th image to an x-th result image, and displaying the x-th result image.A further method according to the invention for processing an x-th image of a medical image sequence of at least two successive images, in particular X-ray images, of an imaging unit comprises the following steps: recording or providing the x-th image of the image sequence of at least two successive images, dividing the x-th image into an x-th noise component and an x-th signal component, temporal filtering of the x-th noise component of the image using the noise component of at least one other image, assembling the temporal filtered x-th noise component and the x-th signal component of the x-th image to form an x-th result image, and displaying the x-th result image.Both methods are based on the realization that during the temporal filtering of the respective entire x-th image of an image sequence, a loss of contrast or artifacts occur because the imaged structures are filtered in time together with the noise, but the improvement of the representation is achieved only by the temporal filtering of the noise. Both methods use this knowledge in order to separate the respective noise component and the signal component of the respective image from one another as far as possible and process them differently before the two components are reassembled. By applying the temporal filtering only to the noise component, the representation is improved without losing contrast and without generating artifacts.According to one embodiment of the invention, at least two, in particular a plurality, successive (r) images of the image sequence are provided and processed as described above and subsequently displayed together as a result image sequence.The result is a qualitatively and subjectively improved representation of an image sequence. A viewer is enabled to view the image sequence in an optimum subjective manner and to obtain more information therefrom, which leads to an improved OP result for example for carrying out interventional interventions. The X-ray dose can also be kept low and even reduced, as a result of which both patients and medical personnel are less encumbered than in known methods.It should be noted that a complete separation of the noise component and the signal component is generally not possible. For this reason, in the context of the present method, a division into a noise component and a signal component always means a division that is technically possible to such an extent.The two methods prevent both a loss of contrast and image artifacts in the signal image with a simultaneously temporally correlated noise representation due to the splitting into a temporally filtered noise image and a spatially filtered signal image.According to a further embodiment of the invention, the temporal filtering is carried out by averaging, in particular moving average or moving exponential weighting. Such filtering methods are known, easy to use and have proven useful in temporal filtering. The averaging involves, for example, the noise component of the previous image of the x-th image or of a plurality or many other images.According to a further embodiment of the invention, the images of the image sequence are formed by X-ray images, in particular fluoroscopic images or angiography images. X-ray images generally have a significant noise component by using x-ray radiation to generate it, which cannot be completely removed. For this reason, the method is particularly suitable for the representation of image sequences of X-ray imaging. A method which is very frequently used in interventional interventions under X-ray monitoring is fluoroscopy, since it allows continuous recording of projection images at a low dose. The low dose particularly often leads to high noise, which can be displayed in an improved manner by the methods described.For a reasonable temporal filtering, in which the noise components of other images of the image sequence are generally included, the image sequence is formed from at least three successive images.According to a further embodiment of the invention, at least one machine learning (in particular also deep learning) algorithm is used to carry out the method. Such an algorithm can be used, for example, in the context of performing the spatial denoising of the variance-stabilized x-th image.The invention also comprises a medical system for carrying out a method for processing an x-th image of a medical image sequence from at least two successive images, in particular X-ray images, of an imaging unit, having an X-ray device with an X-ray source and an X-ray detector, designed to record an image sequence from at least two successive images, an image processing system for processing the images in accordance with the image processing method, a display unit for displaying images, and a system controller for automatically controlling the method.The invention and further advantageous embodiments according to features of the dependent claims are explained in more detail below with reference to schematically represented exemplary embodiments in the drawing, without the invention thereby being limited to these exemplary embodiments. The following are shown: FIG. 1 shows a sequence of steps of an image processing method according to the invention; FIG. 2 shows a further sequence of steps of a further image processing method according to the invention; FIG. 3 shows a view of a medical system for carrying out the method; and FIG. 4 shows a further sequence of steps of a further image processing method.FIGS. 1 and 2 show steps of a variant of a method for processing an x-th image, in particular an x-ray image, of a medical image sequence, in particular an x-ray image sequence, in each case. The x-ray image sequence can be recorded, for example, in the context of monitoring an interventional medical method; it can therefore be, for example, fluoroscopic or angiography x-ray images. The x-ray image sequence can be recorded, for example, when monitoring a navigation of an object in a hollow organ of the patient, for example catheter navigation, ablation, stent implantation or the like. The x-ray image sequence can also be recorded within the scope of a DSA (digital subtraction angiography) with and / or without contrast agent application. The x-ray image sequence generally comprises a plurality of individual x-ray images recorded successively, but at least two x-ray images.In a first step 10, at least one x-th x-ray image of the x-ray image sequence is recorded, e.g. by means of an x-ray apparatus, or provided from a memory or a database, see FIG. 1.In a second step 11 (FIG. 1 ), the x-th x-ray image is subsequently subjected to a variance-stabilizing transformation after it has been recorded / provided. This is to ensure that the image chain is always applied to an image normalized with respect to the noise. The noise in X-ray images generally has Poisson statistics (quantum noise) with possible superposition of low electrical noise by the X-ray detector (electronic noise). The noise may be stabilized or normalized using variance stabilization for Poisson noise over the image signal. A frequently used example of such a variance stabilization is a so-called Anscombe transformation (see, for example, in https: / / en.wikipedia.org / wiki / Anscombe transform). Ideal variance stabilization may also stabilize the differential chromaticity of electrical and quantum noise.The variance-stabilized x-th x-ray image is subsequently subjected to spatial denoise or noise suppression (spatial decoloring) in a third step 12. Spatial noise suppression generally means noise reduction based on data within an image plane. Spatial noise suppression may include, for example, application of one or more linear or non-linear filters, for example bilateral filters, wavelet transforms, non-local averaging, or variation approaches (e.g., total variation denoising). Methods for the spatial noise suppression of images / X-ray images are generally known, for example from the patent specification EP 2169592 B1. For spatial noise suppression, e.g. a classic approach (bilateral filters, wavelet transformations) can be carried out or a pre-trained machine learning algorithm (machine learning or deep learning, also CNNs or GANs) can be used. Approaches are also known in which data from a plurality of images is used.In a fourth step 13, a subtraction of the denoised variance-stabilized x-th x-ray image from the (non-denoised) variance-stabilized x-th x-ray image is carried out. The result of this subtraction is the x-th noise component of the x-th x-ray image.The x-th x-ray image denoised in the third step 12 can then be transformed back in a fifth step 14 by an inverse variance stabilization. This produces the denoised x-th signal component of the x-th x-ray image.The x-th noise component obtained in the fourth step 13 is then subjected to temporal filtering in a sixth step 15. In this case, noise components of temporally adjacent x-ray images are generally used, for example of the x-ray image of the x-th x-ray image arranged directly upstream in time, or else of a plurality of x-ray images of the x-th x-ray image arranged upstream in time. Such a temporal filter may be e.g. a moving average or an exponentially weighted average. Averaging is formed e.g. simply by averaging over the respective same pixels of different X-ray images. The temporal filtered x-th noise component of the x-th x-ray image is obtained by the temporal filtering.In a seventh step 16, the temporally filtered x-th noise component of the x-th x-ray image is then added to the denoised x-th signal component of the x-th x-ray image, resulting in a noise-corrected x-th x-ray image, which is then displayed in an eighth step 17. In particular, after a plurality of x-ray images of the x-ray image sequence have been processed in this way, the processed x-ray image sequence is displayed in a tenth step 19 (see FIG. 4 ).For purposes of illustration, additional processing steps can also be used after denoised and before addition to the denoised variance-stabilized x-th x-ray image (or to all such x-ray images of the x-ray image sequence), such as e.g. LUT (look-upable) functions (in particular a logarithm function, since this is inverse to the exponential attenuation law and can thus be a preferred representation of the signal component) or further spatial filters.Some steps can also be carried out in a different order, for example the sixth step 15 can be carried out before the fifth step 14.FIG. 2 shows a further embodiment of the method. Here, after the first step 10, i.e. the acquisition or provision of the x-th x-ray image of the x-ray image sequence, in an eleventh step 21, the x-th x-ray image is divided into an x-th noise component and an x-th signal component. It should be noted here that a complete division is generally not possible, but rather a division is technically possible to the extent possible. In a twelfth step 22, temporal filtering of the x-th noise component of the x-th x-ray image is subsequently carried out using the noise component of at least one other x-ray image, for example of an x-ray image arranged directly upstream in time or else of a plurality of x-ray images arranged upstream in time. The temporal filter may be, for example, a moving average or an exponentially weighted average. In a thirteenth step 23, the x-th signal component of the x-th x-ray image is subsequently reassembled with the temporally filtered x-th noise component of the x-th x-ray image. Further filtering or image processing steps can be carried out before or after this. Subsequently, the x-th x-ray image and / or the entire x-ray image sequence is displayed.This method prevents a loss of contrast or artifacts in the signal image in the case of a simultaneously temporally correlated noise representation by splitting into a temporally filtered noise image and a spatially filtered signal image. In contrast to the prior art, in which a temporal filtering is applied to the entire image and thus the moved signals are also averaged, so that the loss of contrast of structures occurs, only the noise components are filtered in time. This produces an image impression that is subjectively pleasant for the observer without loss of structures. The method makes it possible to save X-ray dose.FIG. 4 shows a medical system 30 for carrying out the method. The medical system 30 comprises an X-ray device with an X-ray source 31 and an X-ray detector 32, which is designed to record an X-ray image sequence from at least two successive X-ray images. The X-ray source 31 and the X-ray detector can be arranged on a movable C-arm in order to record X-ray images from different algorithms. The X-ray device can be designed to record angiography X-ray images and / or fluoroscopy X-ray images. In addition, the medical system 30 has an image processing unit 34 for processing the images according to the method. For this purpose, the image processing unit can have a processor which operates one or more or a plurality of software algorithms. Machine learning algorithms can also be used to perform one or more processing steps. In addition, the medical system has a display unit 35 for displaying edited or unprocessed images. The medical system 30 is controlled by a system controller 33. The system control is also designed for automatically controlling the method.Regardless of the grammatical sex of a certain term, individuals with male, female or other sex identity are included.The invention can be summarized briefly in the following manner: For a subjectively pleasing representation of a medical image sequence without loss of structures, a method for processing an x-th image of a medical image sequence from at least two successive images, in particular X-ray images, of an imaging unit is provided, having the following steps: recording the x-th image of the image sequence from at least two successive images, carrying out a variance-stabilizing transformation of the x-th image, carrying out a spatial denoiseation of the variance-stabilized x-th image, subtracting the denoised variance-stabilized x-th image from the variance-stabilized x-th image and thereby obtaining an x-th noise portion of the x-th image, performing inverse variance stabilization of the denoised variance stabilized x-th image and thereby obtaining a denoised x-th signal portion of the x-th image, performing temporal filtering of the x-th noise portion and thereby obtaining a temporal filtered x-th noise portion of the x-th image, adding the temporal filtered x-th noise portion with the denoised x-th signal portion of the x-th image to an x-th result image, and displaying the x-th result image.References included in the specificationThis list of documents cited by the applicant has been produced in an automated manner and is only included for the better information of the reader. The list is not part of the German patent application or utility model application. The DPMA does not take any adhesion for any faults or omissions.Patent Literature citedEP 2169592 B1

[0022]

Claims

Method for processing an x-th image of a medical image sequence from at least two successive images, in particular X-ray images, of an imaging unit, having the following steps: • recording or providing the x-th image of the image sequence from at least two successive images, • carrying out a variance-stabilizing transformation of the x-th image, • carrying out a spatial denoisement of the variance-stabilized x-th image, • subtracting the denoised variance-stabilized x-th image from the variance-stabilized x-th image and thereby obtaining an x-th noise component of the x-th image, • carrying out an inverse variance stabilization of the denoised variance-stabilized x-th image and thereby obtaining a denoised x-th signal component of the x-th image, • performing temporal filtering of the x-th noise portion and thereby obtaining a temporal filtered x-th noise portion of the x-th image, • adding the temporal filtered x-th noise portion to the denoised x-th signal portion of the x-th image to an x-th result image, and • displaying the x-th result image.Method for processing an x-th image of a medical image sequence of at least two successive images, in particular X-ray images, of an imaging unit, having the following steps: • recording or providing the x-th image of the image sequence of at least two successive images, • dividing the x-th image into an x-th noise component and an x-th signal component, • temporal filtering of the x-th noise component of the image using the noise component of at least one other image, • assembling the temporal filtered x-th noise component and the x-th signal component of the x-th image to form an x-th result image, and • displaying the x-th result image.A method according to claim 1 or claim 2, wherein at least two successive images of the image sequence are processed in such a manner and subsequently displayed together as a result image sequence.Method according to one of the preceding claims, wherein the temporal filtering is carried out by averaging, in particular moving average or moving exponential weighting.Method according to one of the preceding claims, wherein the images are formed from X-ray images, in particular fluoroscopic images or angiography images.Method according to any of the preceding claims, wherein the image sequence is formed from at least three successive images.Method according to one of the preceding claims, wherein at least one machine learning algorithm is used to carry out the method.Medical system (30) for carrying out a method for processing an x-th image of a medical image sequence comprising at least two successive images, in particular X-ray images, of an imaging unit, having: • an X-ray device having an X-ray source (31) and an X-ray detector (32), designed to record an image sequence comprising at least two successive images, • an image processing unit (34) for processing the images according to the method according to one of Claims 1 to 6, • a display unit (35) for displaying images, and • a system controller (33) for automatically controlling the method.

Images