Data Correction in X-Ray Imaging

Abstract

Embodiments of the present invention relate to data correction in X-ray imaging. A method for correcting an input data set comprises the following steps: obtaining an input data set including at least one data error; determining a correction function; generating a corrected output data set by applying the correction function to the input data set; outputting the corrected output data set; wherein the correction function is designed to reduce at least two mutually affecting data errors in the input data set.

Description

Technical Field

[0001] The present invention relates to data correction in X-ray imaging. Background Art

[0002] A new technology in the field of X-ray imaging, especially in the field of computed tomography, is a detector that directly converts or directly converts or (photon) counts, which can realize spectral resolution imaging or energy selective imaging. Counting X-ray detectors do not use a scintillator (Szintillator) that first converts X-ray photons into visible light, but use semiconductors that directly detect X-ray photons. For example, in spectral computed tomography, direct conversion detectors can advantageously distinguish the energy of incident photons in addition to position. Here, the X-ray quanta incident on the detector surface are assigned to specific energy bands or energy boxes according to the quantum energy, that is, energy is selectively classified, and the events of each pixel in each energy band are counted separately. Therefore, the counting X-ray detector is suitable for simultaneously generating at least two projection data sets, and the X-ray quantum energy distributions of these projection data sets are different from each other when only one X-ray radiation source is used. There are usually two to four energy bands. At least two projection data sets are formed by all pixel signals of the detector in the energy band. The semiconductors cadmium telluride, cadmium zinc telluride or gallium arsenide are particularly suitable as detector materials for quantum counting detectors or, in the case of flat detectors, amorphous selenium or the like. The thresholds of the energy bands are predetermined by the manufacturer and the electronics in the detector module are set accordingly. Fixed, unchangeable thresholds can be preset or the thresholds can be selected by the user in accordance with or together with the desired scanning protocol, for example for head or chest recordings. That is, in order to optimize the medically relevant information content of the acquired image recordings, the thresholds can be adjusted to suit the preliminary guess, the expected diagnosis or the medical question.

[0003] In particular, photon counting X-ray detectors have the problem that strong nonlinearities exist due to pile-up effects in the detector, which in turn are affected by factors such as beam hardening. The influence of spectral effects such as

[0004] Pile-up describes the following situation, in which at least two detection events occur at an X-ray detector within a short time such as the resolution time below, which is used for the duration of the electrical pulse generated by a single incident X-ray quantum. If the detection events are very close to each other in time, the detection events will be combined into pulses with cumulative amplitudes according to their quantum energies (peak pile-up). If the detection events now occur at a larger time interval, although these detection events can be detected as independent events, at least the (multiple) X-ray quanta incident later in time will be acquired with quantum energies distorted to higher values ​​(tail pile-up). Therefore, in the case of pile-up, count losses of detection events may result, which are similar to the count losses caused by dead time. In addition, the X-ray quantum spectrum will be caused to move to higher quantum energies. According to the predetermined energy box, the detection events superimposed due to pile-up may be grouped into the wrong box, which may distort the count rate of multiple projection data sets. The pile-up effect depends largely on the count rate of the system, that is, the higher the count rate, the greater the impact of higher-order pile-up effects (superposition of more than two detection events).

[0005] The beam hardening effect is the shift of the effective energy of the X-ray beam towards higher values ​​when penetrating a material. This is due to the energy dependence of the linear X-ray attenuation coefficient. In other words, the average energy of the X-ray quanta increases with increasing penetration depth, since lower energy X-ray quanta are preferentially absorbed. The relative proportion of low energy X-ray quanta decreases with increasing material penetration. Therefore, the thicker the material, the greater the hardening error in the acquired detector signal.

[0006] Nonlinearities caused by pile-up and beam hardening both manifest themselves as a shift of the effective X-ray energy to higher values. They cannot therefore be easily separated.

[0007] So far, this problem has been solved at least approximately by performing repeated correction steps iteratively, i.e. first a nonlinear correction, then a beam hardening correction, then a nonlinear correction, etc. However, the described procedure is lengthy and computationally intensive. Summary of the invention

[0008] In contrast, the present invention aims to provide alternative means which enable reliable, automatic and fast error correction of energy-selective measurement data. In particular, the present invention aims to provide error correction for at least two mutually influencing data errors simultaneously.

[0009] This object is achieved by a method for correcting an input data set, a corresponding processing unit, a corresponding computer program and a corresponding computer-readable data carrier according to the independent claims. Preferred and / or alternative advantageous design variants are the subject matter of the dependent claims.

[0010] The solution according to the invention for this purpose is described below with reference to the claimed method and also with reference to the claimed device. The features, advantages or alternative embodiments mentioned here are also applicable to the other claimed subjects and vice versa. In other words, the objective claim (e.g. for a method) can also be improved by combining a device description or claimed features. Here, the corresponding functional features of the method are composed of corresponding entity modules or units.

[0011] In a first aspect, the invention relates to a method for correcting an input data set. The process comprises a number of steps.

[0012] A first step involves acquisition of an input data set comprising at least one data error. Another step involves determination of a correction function. Another step involves generating a corrected output data set by applying the correction function to the input data set. Another step involves outputting the corrected output data set.

[0013] The invention is characterized in that a correction function is designed which reduces at least two mutually influencing data errors in the input data set. In other words, two data errors which cause distortion of the input data set can be minimized or eliminated in parallel (i.e., in one operation step) by means of the correction function proposed here.

[0014] The at least two data errors influence each other at least in that the different data errors have similar error effects on the input data set, so that a part of the data distortion can no longer be clearly attributed to one of the two data errors.

[0015] The method according to the invention transforms an input data set into an output data set by using a correction function according to the invention. In other words, the correction function corresponds to a mathematical rule that transforms an input data set into an output data set.

[0016] Although the input data set has data errors, the output data set will be free of these data errors. The data errors in the output data set are at least significantly reduced.

[0017] The input data set is characterized in that the input data set comprises error effects which are based on at least one, preferably at least two different data errors.

[0018] The input data set according to the invention corresponds to projection data of an X-ray detector, such as projection data generated in a medical examination of a patient, such as a computed tomography examination. In this respect, the input data set is in particular a medical input data set. The input data set in particular comprises at least two projection data sets which differ spectrally, i.e. in their (average) X-ray quantum energy. Here, the projection data set represents, for a quantum energy spectrum mapped by it, a spatial intensity distribution of X-ray quanta incident on the entire sensitive detector surface, wherein the detected X-ray quanta emitted from the X-ray source have passed through the examination object. Thus, each pixel entry (i.e., a projection measurement value of the projection data set) corresponds to a plurality of X-ray quanta which are incident in this pixel with a quantum energy corresponding to a corresponding energy range, energy band or energy box. Thus, the input data set corresponds to an erroneous, detected attenuation distribution of the imaged examination object.

[0019] In other words, the step of acquiring the input data set includes recording (medical) measurement data by means of an X-ray source and an X-ray detector and transmitting the measurement data to a computing unit for further processing there.

[0020] In the sense of the present invention, an examination subject is a patient, usually a human. In principle, a patient can also be an animal. However, the examination subject can also be a non-biological, in particular non-medical object, such as a pathological specimen, a mechanical component or a historical artifact. Therefore, the two terms "examination subject" and "patient" are used as synonyms hereinafter.

[0021] The output data set according to the invention corresponds to the input data set from which contained data errors have been removed. That is, the output data set is also designed in particular as a medical output data set, which can be understood, for example, as a projection data set of an X-ray detector, which, like the input data set, can in particular include at least two spectrally different projection measurement data sets. The output data set thus corresponds to the corrected, detected attenuation distribution of the imaged examination object.

[0022] The step of outputting the output data set preferably comprises transmitting the output data set to a processing unit or a reconstruction unit, which is configured to generate essentially artifact-free X-ray image data by means of reconstruction rules known per se, for example by means of (iterative or weighted) filtered back-projection or algebraic reconstruction.

[0023] The order of the steps of the method according to the invention is not limited to the order of the steps listed here. The step of determining the correction function can in particular completely or at least partially precede the remaining steps.

[0024] In a first embodiment of the method according to the invention, the correction function is designed to at least reduce data errors in the input data set that are based on pile-up effects and beam hardening effects that occurred during the acquisition of the input data set. These two effects that have been explained above and that cause data errors influence each other, since they in particular each lead to a shift of the detected X-ray quantum energy spectrum to higher values, so that it is no longer possible to determine which data error component can be attributed to which of the two effects. The correction function according to the invention advantageously dispenses with the assignment of data error components to the effects that cause data errors and corrects the data errors together, so that computing time can be saved and the work process can be made more efficient.

[0025] In another embodiment of the method according to the invention, the correction function is also designed to reduce another data error in the input data set based on the scattered radiation effect. The scattered radiation effect describes the generation of scattered X-ray quanta, which change their radiation direction when interacting with the material being transmitted. If there are scattered X-ray quanta, the scattered X-ray quanta will be incident on the detector pixels that are inconsistent with their initial radiation direction, thereby causing distortion of the spatial X-ray intensity distribution. In addition, due to the scattering phenomenon, the X-ray quanta will lose part of their quantum energy, so that the scattered X-ray quanta may be assigned to the wrong energy band. The scattered radiation intensity depends on the thickness of the material and the type of material. The scattered radiation can especially include radiation generated by multiple scattering, that is, higher-order scattered radiation. In addition, the scattered radiation can also be formed as cross scattering of a dual-energy X-ray imaging system. In other words, the scattered radiation corresponds to the following radiation, which is emitted by the X-ray source of the first radiator-detector pair, but is scattered so that it is detected by the X-ray detector of the second radiator-detector pair in the system, and / or vice versa. The correction function according to the invention advantageously dispenses with the need for further distribution of the data error components to the effects causing the data error components and effectively corrects the data error components together.

[0026] The following embodiments of the method according to the invention may be suitable, alone or in combination, for explaining the method for determining a correction function, wherein the correction function according to the invention reduces at least two mutually influencing data errors in an input data set.

[0027] In another embodiment of the method according to the invention, the correction function is designed as a function that depends on a monitoring value about the X-ray source used to acquire the input data set. The monitoring value is designed to provide a measure of the applied X-ray radiation. The monitoring value is preferably designed as an X-ray tube current set at the X-ray source, which determines decisively the intensity of the X-ray beam emitted by means of the X-ray tube. The monitoring value can also be a dose value. In other words, the correction function is designed to implement a correction related to the tube current or the X-ray intensity on the input data set. This is particularly advantageous because this takes into account the intensity dependence or count rate dependence of the data errors.

[0028] In a further embodiment of the method according to the invention, the determination of the correction function comprises a calibration measurement with a spectrally resolving X-ray detector.

[0029] The spectrally resolved X-ray detector is designed to generate a measurement data set comprising at least two projection data sets that are spectrally different. In other words, the spectrally resolved X-ray detector is an energy selective X-ray detector. As described above, the spectrally resolved X-ray detector can be designed as a quantum counting X-ray detector. However, the spectrally resolved X-ray detector can also be designed as a double-layer detector.

[0030] A dual layer detector (Dual Layer Detektor or Double Layer Detektor) is designed to decompose the incident X-ray tube spectrum into a low energy component and a high energy component. To this end, the dual layer detector consists of two layers. The detector layer facing the X-ray source measures the X-ray photons of the incident X-rays at low energy and assigns the measured signal to a first projection measurement data set. The first detector layer is penetrated by high energy X-rays. In the detector layer below or behind it, i.e., away from the X-ray radiation source, X-ray photons with higher quantum energy are measured, and the X-ray photons with higher quantum energy are assigned to the second projection measurement data set. Typically, both detector layers include scintillators, so dual layer detectors are typically indirect conversion detectors. Crystals such as cesium iodide, cadmium tungstate, or ceramic materials such as gadolinium oxysulfide are used as scintillating materials.

[0031] In other words, in this embodiment, spectrally resolved calibration data comprising at least two spectrally different calibration data sets will be generated, which are included in the determination of the correction function.

[0032] In a preferred embodiment of the method according to the invention, the calibration measurement comprises the acquisition of calibration data of a test object, the test object comprising at least two basic materials. In other words, in this embodiment, a calibration data set comprising at least two calibration data sets that differ in spectra will be generated, wherein each calibration data set describes the X-ray attenuation distribution of the test object with respect to its energy bin. The test object is essentially a phantom, which is made of at least two basic materials. Alternatively, the test object may consist of two test objects each comprising only one basic material.

[0033] Therefore, the calibration data is advantageously applicable to the base material decomposition.

[0034] In the case of base material decomposition, it is assumed that the X-ray attenuation values ​​measured by means of an X-ray detector can be described as a linear combination or line integral of the X-ray quantum energy distribution for each energy band along the beam path of the X-ray attenuation values ​​of the base material. The measured X-ray attenuation values ​​or distributions are derived from at least two calibration data sets for the respective different X-ray quantum energy distributions. The material or base material can be any substance or any tissue, in particular water, contrast media such as iodine, soft tissue, bone, etc. can be used. Within the scope of the invention, more than two base materials can also be combined.

[0035] In this regard, in a preferred embodiment of the method according to the invention, the base material comprises one of the following material combinations:

[0036] – Water-iodine,

[0037] – Water-Bone, or

[0038] – Aluminum-Iron.

[0039] The calibration measurement according to the invention advantageously comprises a plurality of measurements of a plurality of test objects which differ in the thickness of the included base materials. For example, a plurality of test objects with a material combination of, for example, water and bone can be measured, wherein the thickness of one base material is further increased and the thickness of the second base material is also increased or decreased to the same extent.

[0040] Furthermore, for each thickness ratio of the material combination, a plurality of measurements are carried out with different monitoring values, in particular with different tube currents of the X-ray source.

[0041] In this way, a series of calibration data sets are generated for selected material combinations and different thickness ratios of the materials, which calibration data sets represent the X-ray attenuation distribution of the base material depending on energy, thickness and tube current. In other words, a base material decomposition can be performed for calibration data set pairs / calibration data set groups each comprising two calibration data sets from two or more energy bands of the X-ray detector.

[0042] The X-ray attenuation properties of the base materials are theoretically known and provided, for example, in the form of a table for retrieval. In this regard, the expected (ie, ideal) attenuation data set is also known or determinable for any combination of base materials and any thickness ratios thereof and different monitoring values.

[0043] The invention is now based on the consideration that the calibration data generated as described above with respect to the test object already have the data errors described, so that the calibration data set deviates from the theoretically known or ideal X-ray attenuation characteristics. According to the invention, a correction function is now determined as a function depending on the monitoring values, which function maps the calibration data set to the ideal projection data set.

[0044] The correction function can be designed as a polynomial, a linear combination of multiple functions or a nonlinear function. In particular, the advantage of a linear combination is that it can be optimized in an analytical manner very simply and directly in mathematics, i.e., the coefficients can be directly calculated here instead of iteratively. According to the design of the correction function, the parameterization of the correction function is performed based on the calibration data set and the ideal projection data set.

[0045] In another embodiment of the method according to the invention, the calibration measurement of the test object includes at least two integrated effective cross sections taking into account the Compton effect (Compton-Effekt) and the photoeffect (Photoeffekt). This embodiment is an alternative to the previous embodiment of the present invention. In addition to decomposing the X-ray attenuation coefficient into individual attenuation coefficients of two or more basic materials, the X-ray attenuation coefficient can also be decomposed into a Compton attenuation coefficient and a photoattenuation coefficient for the interaction, which can be obtained by converting two calibration data sets with different spectra similar to the basic material decomposition. The theoretical / ideal attenuation coefficient is known for different materials and different monitoring values. In this embodiment, the test object can also include only one material. In the Compton effect, the X-ray quantum releases part of its kinetic energy by changing direction when it collides elastically with an electron. In the photoeffect, the X-ray quantum is completely absorbed by the electron.

[0046] In another embodiment of the method according to the invention, the determination of the correction function includes a calibration measurement of the test object taking into account at least two monochromatic effective cross sections. This embodiment is another alternative to the aforementioned embodiment of the invention. In addition to decomposing the X-ray attenuation coefficient into individual attenuation coefficients of two or more base materials, the X-ray attenuation coefficient can also be decomposed in an energy-selective manner into attenuation coefficients for specific quantum energy values, which can also be obtained by converting two spectrally different calibration data sets. The theoretical / ideal attenuation coefficient is again known for different materials and different monitoring values. In this embodiment, the test object can include only one material.

[0047] In a further alternative embodiment of the invention, the determination of the correction function comprises a simulation of the calibration measurement in one of the above-described embodiments. In this case, the calibration data set (i.e. the calibration measurement values ​​for each pixel element of the X-ray detector) is determined in a purely mathematical manner by simulating the detection characteristics of each individual energy band of each individual detector pixel for the defined test object according to the monitoring values ​​with the aid of a simulation rule. The simulation preferably generates data errors in the simulated calibration measurement values, such as those caused by pile-up effects or beam hardening. Particularly preferably, the simulation also additionally generates data errors, such as those caused by scattered radiation.

[0048] In another embodiment of the method according to the invention, the correction function is designed as a function trained by means of machine learning. In the sense of the present invention, machine learning includes computer-implemented techniques in which an algorithm recognizes patterns or regularities based on existing data and independently derives solutions to unknown new data by using these patterns or regularities. A prerequisite for independently finding a solution is a training phase in which a machine learning algorithm is applied to a known, defined and usually very large database in order to find rules or predictions that achieve the desired output or desired result. The training can be designed as supervised training or unsupervised training, in which in a first variant the algorithm presents value pairs (Werte-Paare) in the form of input values ​​and corresponding correct output values, while in a second variant the algorithm must be independently adapted based on the input values ​​to provide the correct output values.

[0049] The machine learning algorithm is particularly advantageously designed as an artificial neural network. The artificial neural network preferably includes a plurality of other layers between the input layer and the output layer, each of which includes at least one node. Here, each node corresponds to a processing unit. The nodes in one layer of the network can be connected to the nodes of other layers by directional connections (edge). These connections define the data flow within the network. Therefore, each node represents an operation applied to the input data. In addition, each node or each of its connections has a weighting parameter (weight). The influence or importance of the output of the node is defined as the input value of the receiving node by the weighting parameter. In the training phase, which is preferably implemented as supervised learning, the artificial neural network "learns" the weighting parameters of all nodes or connections with the help of training data, and adjusts these weighting parameters until the output layer of the network provides the correct output value.

[0050] Thus, during its training, the trained machine learning function establishes a fixed correlation between input values ​​(input) (here in the form of a calibration data set and corresponding monitoring values) and output values ​​(output) in the form of an ideal projection data set. In this embodiment, the trained function replaces the correction function.

[0051] In another aspect, the invention relates to a computing unit for correcting an input data set, the computing unit having means for performing the method according to the invention. The computing unit comprises:

[0052] – an interface configured to obtain an input data set comprising at least one data error,

[0053] - a determination unit configured to determine a correction function,

[0054] a correction unit configured to generate a corrected output data set by applying a correction function to the input data set, and

[0055] – An interface configured to output the corrected output data set.

[0056] The correction function is designed such that at least two mutually influencing data errors in the input data set are reduced.

[0057] In particular, the computing unit can be designed as a computing unit of a medical imaging device, which includes an X-ray source, an energy selective X-ray detector and a patient couch. The computing unit can also be designed as a computing unit of an energy selective X-ray detector. In a preferred embodiment, the computing unit is connected to a medical imaging device including an energy selective X-ray detector. Alternatively, the computing unit is directly connected to the energy selective X-ray detector. The computing unit is advantageously integrated into the medical imaging device. Alternatively, the computing unit can also be arranged remotely or away from the medical imaging device. The computing unit can be designed to specifically perform the step of determining the correction function, but the entire method according to the present invention can also be performed for one medical imaging device or multiple devices, for example in a radiology center or hospital including multiple medical imaging devices.

[0058] A further aspect of the invention relates to an X-ray detector in the form of an energy-selective X-ray detector, which comprises a computing unit according to the invention.

[0059] The invention further relates to a computer program having a program code in order to carry out the method according to the invention for correcting an input data set when the computer program is run on a computer.

[0060] The invention further relates to a computer-readable data carrier having a program code of a computer program in order to carry out the method according to the invention for correcting an input data set when the computer program is run on a computer. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] The above characteristics, features and advantages of the present invention and their implementation methods can be more clearly and specifically understood in conjunction with the following embodiments. These embodiments are described in more detail below in conjunction with the accompanying drawings. This description does not limit the present invention to these embodiments. In different drawings, the same components have the same reference numerals. These drawings are generally not drawn to scale. Among them:

[0062] Figure 1 shows a schematic diagram of a method according to the invention in one embodiment of the invention,

[0063] Figure 2 shows a schematic diagram of a neural network for use in the method according to the invention,

[0064] Figure 3 An X-ray detector according to an embodiment of the present invention is shown, and

[0065] Figure 4 A medical imaging device in the form of a computed tomography scanner according to one exemplary embodiment of the present invention is shown, which comprises a computing unit according to the present invention. DETAILED DESCRIPTION

[0066] Figure 1 A schematic diagram of the method according to the invention in one embodiment is shown. In a first step S1, medical input data or a medical input data set EG is acquired. The acquisition may include data acquisition by means of an imaging device, which in particular includes a Figure 3 or Figure 4 X-ray detector 9. In other words, the acquisition according to step S1 may include the generation or creation of the medical input data set EG. Alternatively, the medical input data set EG may be a data set that has already been acquired and already exists, which data set can be read from an external or internal storage unit 22, for example, from a hospital information system including an electronic medical record. The acquisition of the medical input data set may generally also include acquiring the input data set EG by means of the interface 31 of the imaging device 1 or the data processing unit 16 or the correction unit 23, that is, including the step of transmitting or reading in the input data from one of the mentioned data sources 9, 22, etc. The acquisition step S1 generally includes the acquisition of the input data set EG.

[0067] The acquisition according to step S1 may also include the acquisition of the following monitoring value m about the X-ray source 8, which is set for acquiring the input data set EG at the X-ray source 8. The monitoring value m can, for example, be included in the input data set EG itself as meta-information and transmitted together with the input data set EG. Alternatively, the monitoring value m is (intermediately) stored and transmitted as an independent variable, which however includes a relationship or association with the input data set EG. According to the present invention, the monitoring value m represents a measure of the applied X-ray radiation. In this embodiment, the monitoring value m corresponds to the tube current of the X-ray source 8. However, the monitoring value m can also be an X-ray dose value.

[0068] The medical input data set is, for example, projection data that can be acquired using an X-ray detector 9. The projection data describes the spatial distribution of the radiation intensity of an X-ray beam, which has passed through the object of examination from the X-ray source and has been attenuated here depending on the nature, composition and thickness of the tissue passed through. In order to resolve the spatial X-ray radiation distribution, the X-ray detector acquires the number of X-ray quanta incident in the pixel element for a large number of planarly arranged pixel elements. According to the present invention, the projection data is obtained by X-ray measurement using an energy selective X-ray detector, which is, for example, in the form of a double-layer detector or a quantum counting detector. The energy selective detector is designed to provide one, two or more than two energy selective measurement values ​​for each pixel element. The incident X-ray quanta are classified into one of the preset energy ranges according to their quantum energy. The projection data set includes all of all pixel measurement values ​​of an energy range. Therefore, the projection data of the input data set can include one, two or more projection data sets corresponding to the number of energy ranges. As long as the data acquisition provides an examination of the object of examination in multiple projection directions, the projection data can include multiple projection data sets for each energy range according to the number of projection directions.

[0069] The medical input data set EG contains data errors. The data errors are based on at least two imaging errors, namely pile-up and beam hardening. In addition, the medical input data set EG may also have data errors in the form of scattered radiation errors. All data errors have in common that they influence each other so that the data error components cannot be separated according to the effect causing the data errors, or can only be separated by extensive calculations.

[0070] In a second step S2, the correction function KF is determined. The correction function KF is determined in the following way: the correction function KF is set to correct the mentioned data errors in the input data set EG. In addition, the correction function KF is designed as a function that depends on the monitoring value m. Step S2 is preferably performed by the determination unit 21 of the operation unit 12. Step S2 can be performed before or after step S1. Step S2 can be designed as one of a plurality of alternative implementations S21, S22. According to step S21, the correction function is determined by calibration measurements of the test object. According to step S22, the correction function is determined by simulation calibration measurements. Particularly preferably, the determination of the correction function includes the training of a machine learning algorithm.

[0071] Step S21 comprises a calibration measurement performed with the spectrally resolving X-ray detector 9. For example, when materials with K-edge absorption are to be quantified, a calibration measurement is performed with a test object comprising two base materials or comprising more base materials. The following base material combinations are preferred:

[0072] – Water-iodine,

[0073] – Water-Bone,

[0074] – Aluminum-Iron.

[0075] Test objects with a selected combination of base materials are measured for a number of different thickness ratios, for example at the following ratios:

[0076] Material 1 Thickness (cm) Material 2 thickness (cm) 0 1 / 2 / 3 5 0 / 1 / 2 / 3 10 0 / 1 / 2 / 3 20 0 / 1 / 2 / 3

[0077] and / or

[0078] Material 2 thickness (cm) Material 1 Thickness (cm) 0 5 / 10 / 20 1 0 / 5 / 10 / 20 2 0 / 5 / 10 / 20 3 0 / 5 / 10 / 20

[0079] The test objects can each consist of a test object comprising only one base material having a certain thickness.

[0080] Calibration measurements can be performed for a plurality of monitoring values ​​m. The result of the calibration measurements is calibration data KD comprising a plurality of calibration data sets for a plurality of thickness ratios. Each calibration data set comprises calibration measurement values ​​si=[s1, ..., sx] for each pixel element, which calibration measurement values ​​are respectively dedicated to one energy range in the energy range of the X-ray detector. For each monitoring value m and each thickness ratio, there are at least two calibration data sets corresponding to the energy range of the X-ray detector and having different spectra.

[0081] Now, the invention is based on the consideration that the calibration measurement values ​​si of each pixel element, like the measurement values ​​of the projection data set, have errors due to error effects such as pile-up, beam hardening, etc. However, for the base materials, the energy-dependent ideal attenuation characteristics are known. Therefore, for each thickness ratio of the base material combination, the ideal attenuation data set ji = [j1, ..., jx] is known or can be calculated for all possible tube currents m.

[0082] Now, the present invention makes use of the generally known assumption that the measured (X-ray attenuation) measurement values ​​can be decomposed into the attenuation components of the base material which cause the X-ray attenuation and of the fact that for all cases of monitored values ​​- base material composition / thickness ratios, there are at least two calibration data sets with different spectra for the X-ray detector.

[0083] Now, for each calibration data set of the calibration data, a basic material decomposition is performed on a calibration measurement value-by-calibration value basis (ie, pixel element-by-pixel basis). In other words, each calibration measurement value si is represented as a basic material line integral L = [L1, ..., L p ](p≤n, n=number of basic materials), that is, s i =[m, L1, ..., L p ].

[0084] Ideal attenuation measurement value j i It can also be expressed as a function of the basic material line integral L, that is, j i =[m, L1, ..., L p ].

[0085] For each energy range, the correction function KF is now determined such that for each energy range, the calibration measurement value s i The monitored value m is mapped to the (ideal) line integral, ie L = KF(s, m).

[0086] In one embodiment, the correction function KF may be a variable s i and m, which best reflects the relationship between variables s, m and L. Alternatively, the correction function KF can also be expressed as s i and a linear combination of other functions of m. In both cases, the problem formulated as a "least squares" problem leads to a system of linear equations. Alternatively, the correction function can also be designed as a nonlinear function and adapted so that it optimally reflects the calibration measurement values.

[0087] Instead of calibration measurements i , we can first calculate the logarithmic variable The variables marked with the number 0 are calibration measurement values ​​obtained in a calibration measurement of the X-ray detector (e.g. free air, i.e. without a test object). The advantage of this method is that after a new calibration measurement the operating point of the correction function KF is restored, so that, for example, an air measurement with an air calibration flow (Luftkalibrierstrom) always has the value As the input value of the correction function KF.

[0088] Replace the logarithmic variable It is also possible to use an antilogarithmic variable in the input of the correction function KF with the same effect Alternatively, the antilog variable l can be used at the output of the correction function KF a = =exp(-L a ):a=1,...,p.

[0089] In another embodiment, the correction function KF can be designed as a function trained by means of machine learning, in particular the neural network 400. In this embodiment, the calibration measurement values ​​si and the monitoring values ​​m can be used as training input data. Alternative designs of the calibration measurement values ​​(logarithmic or antilogarithmic) can facilitate the learning of the correction function KF by the neural network 400. According to the present invention, the trained neural network can output the line integral L as output value, but it can also directly output the ideal attenuation measurement value ji .

[0090] Instead of a basic material decomposition, a calibration measurement of the test object can take into account at least two integrated effective cross sections of the Compton effect and the optical effect to determine the correction function KF. The method is based on the assumption that the X-ray attenuation of all materials can be expressed as a linear combination of the effective cross sections of the optical effect (absorption effect) and the Compton effect (scattering effect). This applies at least approximately and in the absence of K-edge absorption. Polychromatic X-ray attenuation distributions can also be expressed similarly as a linear combination of different monochromatic effective cross sections.

[0091] In an alternative step S22, a simulated calibration data set SIMKD is generated by simulating calibration measurements with the aid of an X-ray detector. Here, the calibration data acquisition of the X-ray detector is simulated in the different cases of the basic material composition, thickness ratio and monitoring values ​​described with respect to step S21. In this case, a scattered quantum flux Φ with a known spectrum can also be additionally included in the simulation, and the calibration measurement value s i It can also be simulated as a function of the scattered quantum flux. Then, the correction function KF will have the scattered quantum flux Φ as an additional input parameter. In this way, the correction function can advantageously take into account data errors based on scattered radiation (which in turn changes the spectral situation or pile-up situation of the X-ray detector).

[0092] Then, the correction function KF in the form of a neural network must also be trained with different values ​​of Φ.

[0093] In step S3, the input data set EG is corrected by applying the correction function KF to the input data set EG. Step S3 can preferably be performed by the correction unit 23 of the computing unit 12. Thus, an output data set AG is generated. The correction function makes the medical input set free of data errors contained in the medical input set or reduces data errors by correcting each individual projection measurement value.

[0094] In another preferred embodiment of the method, step S3 may include a check step as to whether there is metal in the beam path that greatly attenuates the X-rays. In this case, the correction function KF may provide unusable values. For this purpose, an X-ray image can first be reconstructed from the uncorrected input data set EG. With the aid of this X-ray image, metal detection, which is known per se, can be performed. The detected metal parts of the image can be projected forward. Information about which measuring beam contains metal is obtained from this. This additional information can preferably flow into the correction function KF.

[0095] In a further step S4 , the corrected output data set AG is output, for example, to a reconstruction unit 32 for the purpose of reconstructing a substantially artifact-free X-ray image.

[0096] Figure 2 A correction function designed as an artificial neural network 400 is shown, which can be used to Figure 1 The neural network 400 responds to inputs to a plurality of input nodes x i 410 to generate one or more outputs o j In this embodiment, the neural network 400 adjusts the weight coefficient w of each node based on the training data. i (weights) to learn. Input node x i Possible input values ​​for 410 are energy-selective calibration measurements s that are dependent on the X-ray tube current and that are subject to errors in the test object due to pile-up, beam hardening, scattered radiation, etc. i The output value 440 of the neural network 400 is preferably the corresponding ideal attenuation measurement data j without data errors caused by imaging errors. i or line integral L. It can be done through single or multiple output nodes o j Output 440 is performed. For example, for each tube current value m, the input value may be continuously input in the following manner: each input node x i 410 receives the measured values ​​of the detector pixels corresponding to the projection data set. Accordingly, the corrected projection data set can be output pixel by pixel at each output node.

[0097] The artificial neural network 400 preferably includes a hidden layer 430, which includes a plurality of nodes h j . You can set multiple hidden layers h jn , where one hidden layer 430 uses the output value of another hidden layer 430 as an input value. The nodes of the hidden layer 430 perform mathematical operations. Here, the node h j The output value corresponds to its input value x i and weighting factor w i The nonlinear function f of node h j When getting the input value x i Then each input value x i Multiply by the weighting factor w i The products of are added, as determined by the following function:

[0098] h j =f(∑ i x i .W ij )

[0099] Node h j The output value of is particularly formed as a function f of the node activation, for example a sigmoidal function or a linear ramp function.j is transmitted to one or more output nodes o j . Calculate each output value h again j The sum of weighted products of as a function of node activation f:

[0100] o j =f(∑ i h i ·w' ij )

[0101] The neural network 400 shown here is a feedforward network, in which all nodes 430 process the output value of the previous layer in the form of its weighted sum as input value. Of course, according to the present invention, other neural network types can also be used, such as a feedback network, in which the node h j The input value of the node h j The output value of .

[0102] The neural network 400 is trained to recognize patterns by means of supervised learning methods. A known process is back propagation, which can be used for all embodiments of the present invention. During training, the neural network 400 is applied to training input values ​​and must produce corresponding, previously known training output values. According to the present invention, the training input values ​​are calibration data sets, preferably for a plurality of test objects, whose theoretical X-ray attenuation characteristics are known. The training output values ​​are corresponding ideal projection measurement data sets, which can be called up as needed, for example from a training database. The mean square error ("MSE") between the calculated output values ​​and the expected output values ​​is iteratively calculated and the individual weighting factors 420 are adjusted until the deviation between the calculated output values ​​and the expected output values ​​is below a predetermined threshold.

[0103] Figure 3The schematic structure of a counting or direct conversion X-ray detector 9 in one embodiment is shown. The X-ray detector 9 is composed of a plurality of detector modules 35. The detector module 35 includes a plurality of pixel elements 15, an ASIC 27, a segment of a direct conversion material or a direct converter 24, and a coupling portion (e.g., a bump bond 36) between the direct converter 24 and the ASIC 27. The ASIC 27 is connected to a peripheral electronic system 38 via a substrate 37. The incident X-rays are converted in the direct converter 24 (e.g., CdTe or CZT), and the generated charge carrier pairs are separated by an electric field, which is generated by a common top electrode 26 and a pixel electrode 25 specific to each pixel element 15. The charge generates a charge pulse in one of the pixel electrodes 25 of the ASIC 27, and the level of the charge pulse corresponds to the X-ray quantum energy. Therefore, events are counted in the pixel element 125 by adding one to the digital storage unit. If the generated electrical signal is higher than one of the at least one adjustable threshold, the memory cells of the energy band higher than the threshold are counted up. Otherwise, the storage cells of the energy bands lower than it are counted cumulatively. By forming the difference between the counter contents of the two corresponding counters, the X-ray quanta counted in a specific energy range or energy band can be obtained. The threshold is determined by means of a discriminator. In principle, the threshold can be fixedly preset in analog form on the one hand, but is usually set by a digital-to-analog converter (DAC) so that it can be variably adjusted within a certain range. In principle, the threshold can be set pixel by pixel or globally for all pixel elements 15. The X-ray detector according to the present invention has at least one global threshold.

[0104] Figure 4 A medical imaging device 1 in the form of a computed tomography scanner 1 is shown. Other X-ray imaging devices are also conceivable. The imaging device 1 is designed to carry out the correction method according to the invention. The imaging device 1 is particularly designed to generate a medical input data set EG and a medical output data set AG. In addition, the imaging device 1 is also designed to determine the correction function KF according to the invention.

[0105] The computer tomography scanner 1 has a recording unit 17, which includes an X-ray (radiation) source 8 and an energy selective X-ray (radiation) detector 9. The X-ray detector 9 is designed as a direct conversion X-ray detector 9 and, for example, is substantially in the same manner as a reference. Figure 3 The same method is used. Other embodiments are also conceivable. During the recording of the spectrally separated X-ray projections, the recording unit 17 rotates around the system axis 5, and the X-ray source 8 emits X-rays 2 during the recording, which pass through the patient 3 and are attenuated at the patient 3 and are incident on the X-ray detector 9 as projection data.

[0106] When recording spectrally separated X-ray projection data, the patient 3 lies on a patient couch 6. The patient couch 6 is connected to a couch base 4 so that the couch base 4 supports the patient couch 6 and the patient 3. The patient couch 6 is designed to move the patient 3 along a recording direction through an opening 10 of a recording unit 17. The recording direction is usually given by a system axis 5, about which the recording unit 17 rotates when recording the X-ray projection data. In this example, the body axis of the patient 3 is identical to the system axis 5. In the case of a spiral recording, the patient couch 6 is continuously moved through the opening 10, while the recording unit 17 rotates around the patient 3 and records the X-ray projections. As a result, the X-rays 2 appear in a spiral shape on the surface of the patient 3. The patient 3 can, but does not have to, move during the examination.

[0107] The computed tomography scanner 1 has a computing unit 12 in the form of a computer, which is connected to a display unit 11, for example for a graphic display of a medical image recording, in this case in the form of a control menu of the computed tomography recording or the imaging device 1, and to an input unit 7, for example for inputting control commands.

[0108] The computing unit 12 is connected to the rotatable recording unit 17 via an interface 31 for data exchange. On the other hand, the projection data recorded for the patient 3 can be transmitted to the computer 12 as medical input data for the method according to the invention or image reconstruction by means of conventional reconstruction methods. The connection 14 is realized in a known manner in a wired or wireless manner.

[0109] According to this embodiment, the computing unit 12 in the form of a computer includes a data processing unit 16. The data processing unit 16 includes means for executing the steps of the method according to the present invention. Therefore, the data processing unit 16 includes a determination unit 21 for determining the correction function KF. In addition, the data processing unit 16 also includes a correction unit 23, which is configured to apply the correction function KF to the medical input data set EG. Both units 21 and 23 can be designed as separate processing units, but can also be designed together in one unit. In the first case, units 21 and 23 are at least in a data connection that can be a network connection 14, so that the correction function KF obtained by unit 21 is transmitted to unit 23 for use.

[0110] Furthermore, the operation unit 12 may also include a reconstruction unit 32, which receives the output data AG from the correction unit 23, which has been corrected for data errors, and generates medical image data of the patient 3 by using a reconstruction method known per se. The reconstruction unit 32 may also be designed as a subcomponent of the data processing unit 16. In any case, there is a data connection in the form of a network connection 14 between the correction unit 23 and the reconstruction unit 32.

[0111] Interface 31 can be designed as a hardware interface or a software interface, such as a PCI bus, USB or Firewire. Preferably, data exchange is performed by means of a network connection 14. The network can be designed as a local area network (LAN) or a wide area network (WAN) such as an intranet. Network connection 14 is preferably designed as wireless, such as a wireless LAN (WLAN or WiFi). The network can include a combination of different network examples. Data transmission can be based on data query or actively performed. Data transmission between two units or system components can be performed bidirectionally or unidirectionally.

[0112] The determination unit 21 can be in data exchange with an external storage unit (not shown) or an internal storage unit 22, for example, in order to call up tables of ideal attenuation data or ideal attenuation measurement values ​​for basic material combinations, optical effective cross sections and Compton effective cross sections or monochromatic effective cross sections. The external storage unit can be, for example, a hospital server or a radiology information system server (HIS or RIS) or a cloud storage, which in a manner known per se transmits, in particular automatically, to the computing unit 12 or provides, for example, table entries when calling up.

[0113] The operation unit 12 can cooperate with the computer-readable data carrier 13, in particular, in order to perform the method according to the present invention by a computer program with program code. In addition, the computer program can be stored on a machine-readable data carrier in a callable manner. In particular, the machine-readable carrier can be a CD, a DVD, a blue-ray disc, a memory stick or a hard disk. The operation unit 12 and its subcomponents can be designed as hardware form or software form. For example, the operation unit 12 is designed as a so-called FPGA (abbreviation of "Field Programmable Gate Array" in English), or includes an arithmetic logic unit. Alternatively, the operation unit 12, a single or all subcomponents of the operation unit 12 can be arranged in a decentralized manner, for example, the various calculation steps of the method can be performed in a central data center of a medical service structure (such as a hospital) or in the cloud. Here, when exchanging data, data protection and patient protection should be considered in particular.

[0114] Further alternatively, the computing unit 12 can also be designed as a subcomponent of the X-ray detector 9 or as a (separate) cloud-based computer, wherein the data exchange with the imaging device 1 and / or the X-ray detector 9 is carried out via a secure Internet connection. In a preferred embodiment, the DICOM standard is used for communication, but other standards or data formats are also feasible.

[0115] In the embodiment shown here, at least one computer program is stored in the memory 22 of the computing unit 12, and when the computer program is run on the computer 12, the computer program executes the method steps of the method according to the present invention. The computer program for executing the method steps of the method according to the present invention includes a program code. In addition, the computer program can be designed as an executable file and / or can be stored on another computing system other than the computer 12. For example, the computer tomography scanner 1 can be designed so that the computer 12 loads the computer program for executing the method according to the present invention into its internal working memory via an intranet or the Internet.

[0116] In summary, the advantage of the present invention lies primarily in that the correction of pile-up nonlinearity and the correction of beam hardening no longer have to be performed independently of one another. The solution according to the invention performs the above-mentioned two corrections in a correction step by means of an adapted correction function and, if necessary, also a scattered radiation correction. In order to parameterize the correction function or the correction algorithm, a number of calibration measurements and / or simulations are combined. The result of the correction method according to the invention is error-free corrected X-ray attenuation values, which can flow directly (i.e. without further preprocessing) into the image reconstruction. Since the X-ray attenuation values ​​of the projection data set are usually used for at least two energy channels respectively, the corrected X-ray attenuation values ​​are suitable for creating a monochromatic image corresponding to the required X-ray energy by means of a linear combination.

[0117] In the case that it has not been explicitly realized but is advantageous and within the meaning of the invention, the various embodiments, their various aspects or features can be combined or interchanged with each other without departing from the scope of the invention. Where applicable, the advantages of the invention described with reference to one embodiment also apply to other embodiments without explicit mention.

Claims

1. A method for correcting an input data set, wherein the input data set is projection data of an X-ray detector, the method comprising the following steps: - obtaining said input data set comprising at least one data error, – determine a correction function, - generating a corrected output data set by applying the correction function to the input data set, - outputting the corrected output data set, The correction function is designed to reduce at least two mutually influencing data errors in the input data set, wherein the correction function is designed to at least reduce data errors in the input data set based on pile-up effects and beam hardening effects occurring during acquisition of the input data set. 2 . The method according to claim 1 , wherein the correction function is further designed to reduce another data error in the input data set based on scattered radiation effects.

3. A method according to claim 1 or 2, wherein the correction function is a function that depends on a monitoring value regarding an X-ray source (8) used to acquire the input data set, wherein the monitoring value represents a measure of the applied X-ray radiation.

4. The method according to claim 1 or 2, wherein the determination of the correction function comprises calibration measurements using a spectrally resolving X-ray detector (9).

5. The method of claim 4, wherein the calibration measurements comprise acquiring calibration data of a test object comprising at least two base materials.

6. The method according to claim 5, wherein the base material comprises one of the following material combinations: – Water-iodine, – Water-Bone, – Aluminum-Iron.

7. The method of claim 4, wherein the determining of the correction function comprises: Calibration measurements are performed on a test object taking into account at least two integrated effective cross sections of the Compton effect and the optical effect.

8. The method of claim 4, wherein the determining of the correction function comprises: A calibration measurement is performed on a test object taking into account at least two monochromatic effective cross sections.

9. The method according to claim 1 or 2, wherein the determination of the correction function comprises a simulation of a calibration measurement.

10. The method according to claim 1 or 2, wherein the correction function is designed as a function trained by means of machine learning.

11. A computing unit (12) for correcting an input data set, wherein the input data set is projection data of an X-ray detector, the computing unit comprising: - an interface (31) configured to acquire said input data set comprising at least one data error, - a determination unit (21) configured to determine a correction function, - a correction unit (23) configured to generate a corrected output data set by applying the correction function to the input data set, and - an interface configured to output said corrected output data set, The correction function is designed to reduce at least two mutually influencing data errors in the input data set, wherein the correction function is designed to at least reduce data errors in the input data set based on pile-up effects and beam hardening effects occurring during acquisition of the input data set.

12. An X-ray detector (9) in the form of an energy selective X-ray detector, comprising a computing unit (12) according to claim 11.

13. A computer program product that can be directly loaded into a storage unit (22) of a computing unit (12), the computer program having program segments so that when the computer program is run in the computing unit, the steps of the method according to any one of claims 1 to 10 are executed.

14. A computer-readable medium on which a program segment is stored that can be read and executed by a computing unit (12), so that when the program segment is run by the computing unit, the steps of the method according to any one of claims 1 to 10 are performed.

Images