US-12626337-B2 - Computer-implemented method for adjusting the noise of an x-ray image, x-ray facility, computer program and electronically-readable data medium

US12626337B2US 12626337 B2US12626337 B2US 12626337B2US-12626337-B2

Abstract

A computer-implemented method for noise adjustment of an x-ray image recorded with an x-ray facility by an x-ray detector with image points is disclosed. In the method, image values are assigned, measured according to an incident x-ray dose, wherein the image values of the x-ray image include a first detector noise component arising from detector-internal noise and a second dose-dependent signal component arising from the imaging including quantum noise. A local, dose-dependent filter, adjusting the correlation between image points, evaluating a subarea of the x-ray image around an image point currently being processed, is applied to the image values of all image points of the x-ray image, which brings about a change of at least a part of the initial statistics of image values of the subareas to common target statistics of all subareas.

Inventors

Niko Köster

Assignees

SIEMENS HEALTHCARE GMBH

Dates

Publication Date: 20260512
Application Date: 20230630
Priority Date: 20220715

Claims (14)

1 . A computer-implemented method for noise adjustment of an x-ray image recorded with an x-ray facility by an x-ray detector with image points, the method comprising: assigning image values measured according to an incident x-ray dose, wherein the image values of the x-ray image comprise a first detector noise component arising from detector-internal noise and a second dose-dependent signal component comprising quantum noise arising from an x-ray imaging, wherein a local, dose-dependent filter adjusting a correlation between the image points, evaluating a subarea of the x-ray image around an image point currently being processed is applied to the image values of all of the image points of the x-ray image, which brings about a change of at least one part of initial statistics of image values of the subareas to a common target statistics of all subareas, wherein the local, dose-dependent filter is established from a transformation of the image values of an environment extending around a respective image point, comprising the respective image point, and wherein, by the transformation, a covariance matrix of the initial statistics is adjusted to a target covariance matrix of target statistics.
2 . The method of claim 1 , wherein the incident x-ray dose for parameterization of the local, dose-dependent filter is established for each image point depending on an image value of at least one image point of the subarea.
3 . The method of claim 1 , wherein the incident x-ray dose for parameterization of the local, dose-dependent filter is established from a result value of a lowpass-filtered filter result of the x-ray image at the respective image point to be filtered.
4 . The method of claim 1 , wherein, for establishing the local, dose-dependent filter, dose-independent components of the initial statistics are established in a calibration measurement specifically for a layout of the x-ray detector and an operating mode of the x-ray detector during recording of the x-ray image.
5 . The method of claim 4 , wherein, in a first calibration measurement, at least one unexposed calibration image for establishing the initial statistics related to the first detector noise component is recorded and evaluated, and/or wherein, in a second calibration measurement, at least one exposed calibration image with a measurement dose of more than a noise equivalent dose of the x-ray detector is recorded and evaluated.
6 . The method of claim 5 , wherein at least half of a saturation dose of the x-ray detector for establishing the initial statistics related to the quantum noise are recorded and evaluated.
7 . The method of claim 1 , wherein the target statistics are chosen: (1) tailored to an image processing algorithm and/or evaluation algorithm using the x-ray image as initial data, (2) corresponding to the quantum noise or describing noise amounts uncorrelated to the first detector noise component, (3) describing noise amounts mapped to the first detector noise component, or (4) a combination thereof, and/or wherein a normalization of a dose dependence of the local, dose-dependent filter is chosen so that an average value of the image values of the respective environment remains the same or that a variance stabilization is undertaken or that a variance is linear in the incident x-ray dose without offset.
8 . The method of claim 7 , wherein a skew tensor of the initial statistics is additionally adjusted to a target skew tensor of the target statistics through the transformation.
9 . The method of claim 7 , wherein a first noise matrix of the covariance matrix for the first detector noise component and a second noise matrix of the covariance matrix, which multiplied by the incident x-ray dose describes a covariance of the signal component, is established by a calibration measurement, and wherein the transformation is established as a solution of an equation system for the noise adjustment to the target covariance matrix.
10 . The method of claim 9 , wherein, for resolving the equation system, a Cholesky decomposition of covariance matrices is undertaken.
11 . The method of claim 7 , wherein, for establishing the local, dose-dependent filter from the transformation, a quadrant linkage established for an edge point of the subarea is expanded with the image values of other image points of the subarea while assuming a symmetry to all four quadrants around the edge point and is used for the respective image point to be filtered.
12 . The method of claim 1 , wherein filter cores of the local, dose-dependent filter for x-ray dose values covering a dynamic range of the x-ray detector, each filter core representing an x-ray dose interval, are pre-calculated and are stored in a memory, and wherein, for application of the local, dose-dependent filter, a filter core assigned in each case to a corresponding x-ray dose interval is retrieved from the memory and used.
13 . An x-ray facility comprising: an x-ray emitter; an x-ray detector; and a control facility configured to: assign image values measured according to an incident x-ray dose, wherein the image values of an x-ray image comprise a first detector noise component arising from detector-internal noise of the x-ray detector and a second dose-dependent signal component comprising quantum noise arising from an x-ray imaging using the x-ray emitter and the x-ray detector, wherein a local, dose-dependent filter adjusting a correlation between image points, evaluating a subarea of the x-ray image around an image point currently being processed is applied to the image values of all of the image points of the x-ray image, which brings about a change of at least one part of initial statistics of image values of the subareas to a common target statistics of all subareas, wherein the local, dose-dependent filter is established from a transformation of the image values of an environment extending around a respective image point, comprising the respective image point, and wherein, by the transformation, a covariance matrix of the initial statistics is adjusted to a target covariance matrix of target statistics.
14 . A non-transitory computer readable medium comprising a computer program, which, when executed on a control facility of an x-ray facility, is configured to cause the x-ray facility to: assign image values measured according to an incident x-ray dose, wherein the image values of an x-ray image comprise a first detector noise component arising from detector-internal noise and a second dose-dependent signal component comprising quantum noise arising from an x-ray imaging, wherein a local, dose-dependent filter adjusting a correlation between image points, evaluating a subarea of the x-ray image around an image point currently being processed is applied to the image values of all of the image points of the x-ray image, which brings about a change of at least one part of initial statistics of image values of the subareas to a common target statistics of all subareas, wherein the local, dose-dependent filter is established from a transformation of the image values of an environment extending around a respective image point, comprising the respective image point, and wherein, by the transformation, a covariance matrix of the initial statistics is adjusted to a target covariance matrix of target statistics.

Description

The present patent document claims the benefit of German Patent Application No. 10 2022 207 239.1, filed Jul. 15, 2022, which is hereby incorporated by reference in its entirety. TECHNICAL FIELD The disclosure relates to a computer-implemented method for noise adjustment of an x-ray image recorded with an x-ray facility by an x-ray detector with image points, to each of which image values measured according to an incident x-ray dose are assigned, wherein the image values of the x-ray image include a first detector noise component arising from detector-internal noise and a second dose-dependent signal component including quantum noise arising from the imaging. Additionally, the disclosure relates to an x-ray facility, a computer program, and an electronically readable data medium. BACKGROUND In x-ray imaging, not only the pure signals but also noise effects are contained in the image values of a recorded x-ray image. In this case, noise effects from different sources exist, which manifest themselves differently in the x-ray image and, depending on dominance, may lead to a different image impression, which may be undesirable both in respect of the optical display of the x-ray image but also poses additional challenges to image processing algorithms and evaluation algorithms. An image value at an image point of an x-ray image may be broken down into two independent random variables. A first component of these random variables relates to noise arising from the x-ray detector, in particular from the detector electronics, which may be referred to as an offset. The second component of the image value is the actual x-ray signal, (i.e., the signal component), which however also includes quantum noise. The statistics of the detector noise component may be seen in good approximation as Gaussian-distributed (normal distribution), while the underlying statistics of the signal component and of its quantum noise are given by the Poisson-distributed x-ray field. These statistics are further modified by the modulation transfer function of the recording arrangement and also by additional noise sources during the conversion process in the scintillator, for example, by Swank noise and detector gain instabilities. Since the signal component is dose-dependent, it is dependent on the incident x-ray dose at the x-ray detector where which type of noise dominates. However, it is precisely in respect of the further processing of x-ray images that knowledge about the noise level in x-ray images is extremely important. Examples in this regard include the correct setting of threshold values in noise reduction algorithms, the design of Look-Up Tables (LUTs) for image presentation, but also the creation of robust learning prerequisites for artificial intelligence algorithms. Accordingly, it has already been proposed that x-ray images be pre-processed for noise adjustment before they are supplied to such image processing algorithms and/or evaluation algorithms, wherein a variance stabilization may be undertaken. As has already been explained, the variance of the quantum noise scales linearly with the x-ray dose, so that for pre-processing, for example, Look-Up Tables are proposed for the image values that take account of this dependence and stabilize the standard deviation of the noise. These types of approaches deliver good results for x-ray doses for which the quantum noise is markedly greater than the electronic noise. For lower doses these approaches are not efficient. Other proposed approaches for variance stabilization include the use of a generalized Anscombe transformation for example and attempt to take account of the Poisson Gaussian distribution of the underlying statistics. What are known as Gamma Look-Up Tables have also been proposed, which apply an assignment with a steep edge on the basis of empirical values. All these pre-processing measures aim to stabilize the variance of the x-ray image over the dynamic range, but cannot, or may only approximately take account of the massive differences between the noise spectra, especially the noise color, of the detector noise component and of the quantum noise. While the detector noise component may be at least essentially white noise, the quantum noise is moreover filtered by the modulation transfer function of the recording arrangement, which acts on this as a low-frequency filter. Known from DE 10 2007 046 941 A1 is a method for presentation of medical images by a reproduction facility of a diagnostic facility with a suppression of the noise. The method includes: a) one-time calibration of the signal-dependent noise, b) separation of the signal and noise components in the image, c) adjustment of the two components according to parameters set, and d) composition of the signals. Known from DE 10 2009 010 873 A1 is a method for noise reduction of images, in which, during a rotational movement of a radiation source of a CT system around an examination object, acquired data assigned to