Providing radiation load distribution for imaging system with radiation source

Abstract

The invention relates to a computer-implemented method for providing load distribution data of radiation of a radiation source of an imaging system, comprising the steps of: receiving setting data of the imaging system; receiving a load characteristic value of the radiation corresponding to the setting data; receiving intensity distribution data of the radiation at the radiation detector corresponding to the setting data; applying the trained first function to the setting data, the load characteristic value and the intensity distribution data as first input data of the first function, generating load distribution data as first output data of the trained first function; providing load distribution data; wherein the intensity distribution data is provided by applying the trained second function to the setting data as the second input data of the second function, generating the intensity distribution data as the second output data of the trained second function. In addition, the invention also relates to a data processing device, a computer-executed training method and an imaging system.

Description

Technical Field

[0001] The present invention relates to a computer-executed method for providing load distribution data of the radiation of a radiation source of an imaging system, a computer-executed training method for training a function, a corresponding data processing device, a computer program product, and an imaging system. Background Art

[0002] In an imaging system having a radiation source for generating ionizing radiation, it is desirable, for example for regulatory reasons, to record the radiation load on patients and medical staff as precisely as possible. For this purpose, corresponding radiation load characteristic values can be measured during the operation of the imaging system.

[0003] For this purpose, in an X-ray-based imaging system, such as an X-ray-based angiography system, as a radiation load characteristic value, for example, the dose area product, DAP (dose area product in English), in the optical path of the imaging system can be measured by means of a measuring device called a DAP chamber. The measured DAP is used for the regulation of the radiation source in addition to regulatory requirements.

[0004] However, this DAP can only represent the total load on the irradiated surface and cannot indicate the distribution of the load on this surface. In particular, it cannot be assumed that the load is the same, i.e., uniform, on this surface. On the contrary, the radiation load may be unevenly distributed on this surface. In particular, the radiation source emits unevenly distributed rays onto this surface, which may result in an uneven load distribution of the rays on the irradiated surface. Thus, it is also inaccurate to estimate the local load of the radiation based only on this DAP. This radiation may act on the skin, vascular organs, etc. in an X-ray-based angiography system, for example. On the contrary, the local load may be higher than the load that is virtually uniformly distributed based on this DAP. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to more accurately determine the load distribution of the radiation of the radiation source of the imaging system.

[0006] The technical problem is solved according to the present invention by a computer-executed method for providing load distribution data of the radiation of a radiation source of an imaging system, the method having the following steps:

[0007] - Receiving the setting data of the imaging system;

[0008] - Receiving the load characteristic value of the radiation corresponding to the setting data;

[0009] - Receiving the intensity distribution data of the radiation corresponding to the setting data at the radiation detector of the imaging system;

[0010] - Apply the trained first function to setup data, payload eigenvalue, and intensity distribution data that are the first input data of the trained first function, wherein payload distribution data that is the first output data of the trained first function is generated;

[0011] - Provide the payload distribution data;

[0012] wherein the intensity distribution data is provided by applying a trained second function to setup data that is the second input data of the trained second function, wherein intensity distribution data that is the second output data of the trained second function is generated.

[0013] The technical problem is also solved according to the present invention by a computer-executed training method for providing a trained first function that generates payload distribution data of radiation from a radiation source of an imaging system, and the training method is also used to provide a trained second function that generates intensity distribution data of radiation from the radiation source of the imaging system, and the training method includes the following steps:

[0014] - Receive first input training data, the first input training data including

[0015] - Setup data of the imaging system,

[0016] - Payload eigenvalue of radiation corresponding to the setup data, and

[0017] - Intensity distribution data of radiation at a radiation detector of the imaging system corresponding to the setup data;

[0018] - Receive first output training data, which includes intensity distribution data, wherein the first output training data is associated with the first input training data;

[0019] - Train the first function based on the first input training data and the first output training data by generating first output data by applying the first function to the first input training data and adjusting parameters of the first function based on a comparison between the first output data and the first output training data;

[0020] - Provide the trained first function;

[0021] - Receive second input training data, which includes the setup data of the imaging system;

[0022] - Receive second output training data, which includes intensity distribution data, wherein the second output training data is associated with the second input training data;

[0023] - training the second function based on the second input training data and the second output training data by generating second output data by applying the second function to the second input training data, and adjusting a parameter of the second function based on a comparison of the second output data with the second output training data;

[0024] - Provide a trained second function.

[0025] According to the invention, this object is also achieved by a data processing system having at least one computing unit which is suitable for carrying out a computer-implemented method of the type described above.

[0026] This object is also achieved according to the invention by a computer program product having instructions which, when the computer program product is run in a data processing system, cause the data processing system to carry out a method of the type described above.

[0027] According to the invention, this object is also achieved by an imaging system having a data processing device of the aforementioned type, a radiation source for generating radiation, and a radiation detector for detecting a portion of the ionizing radiation, in particular a portion of the radiation that has passed through an object to be imaged.

[0028] The invention is based on the idea of ​​generating load distribution data as first output data at least using a trained function based on setting data of the imaging system, characteristic values ​​of the radiation and load distribution data on the radiation detector.

[0029] According to one aspect of the invention, a computer-implemented method for providing load distribution data of radiation of a radiation source of an imaging system, in particular a radiation source for forming ionizing radiation, is provided.

[0030] The load distribution may also be referred to as dose distribution. For example, the load distribution data may exist as image data, which shows the distribution of the load on a two-dimensional surface. The pixel value of the image may indicate the local load at the pixel through the load.

[0031] In one step of the computer-implemented method, configuration data of an imaging system is received. In another step of the computer-implemented method, a load characteristic value or a dose characteristic value of radiation corresponding to the configuration data is received. In another step, load distribution data of radiation on a radiation detector of the imaging system corresponding to the configuration data is received.

[0032] In another step, the trained first function is applied to the received setting data, the received load characteristic values, and the received load distribution data that are the first input data of the first function. Based on the input data, load distribution data is generated as the first output data of the trained first function. The generated load distribution data can be provided in a subsequent step.

[0033] The provided load distribution data can be estimated especially by the trained function without performing a measurement for this purpose to determine the load distribution. Accordingly, such a measurement process is advantageously omitted.

[0034] The load distribution data can be stored or at least cached in a corresponding storage medium especially in the form of a corresponding computer-readable file, preferably in the form of a two-dimensional image file, which image file includes a two-dimensional intensity distribution. The load distribution data can be provided, for example, by storing the computer-readable file in the storage medium.

[0035] For example, the data of this document can be used, for example, in an electronic medical record. It can also be provided that the load distribution data can be provided to an electronic system having a computing unit, which can further process the data.

[0036] Unless otherwise specified, all steps of the method performed by a computer can be implemented by a data processing device having at least one computing unit. The at least one computing unit is especially configured or adapted to implement the steps of the method performed by a computer. For this purpose, the at least one computing unit can store, for example, one or more computer programs, and the running of the computer programs causes the at least one computing unit to execute the method performed by a computer.

[0037] Advantageously, improved data of the load distribution can be provided by the method. In particular, the improved data does not underestimate the local load peaks that may act on local sites, such as on the skin of a patient. For example, due to the improved data, the skin redness that appears on local sites of the irradiated surface weeks after radiotherapy can be attributed to radiation peaks. In addition, due to the recognition of the load peaks on local sites of the irradiated surface, the radiotherapy can be adjusted so as to be able to reduce consequences, such as skin redness.

[0038] The setting data (also referred to as the settings of the imaging system) can preferably be set personalized by the user of the imaging system, such as by an attending physician or an administrator. In particular, the setting data can be stored or at least cached in a corresponding storage medium in the form of a corresponding computer-readable file, which computer-readable file includes the setting data. The reception of the setting data can be performed, for example, by reading the computer-readable file from the storage medium.

[0039] The load characteristic value is a quantity or value of the dosimeter and is the basis for calculating the radiation load on the irradiated body or irradiated surface during irradiation with a radiation source (e.g., during X-ray imaging using an X-ray machine, such as fluoroscopy or angiography). The load characteristic value corresponds to the sum of the radiation loads reaching the irradiated surface in the optical path of the radiation source. The unit of measurement for the value of the load characteristic value is cGy x cm 2 or Gy x m 2 . The radiation load characteristic value can be, for example, the dose area product, DFP (English: dose area product, DAP).

[0040] In an X-ray based imaging system, such as an X-ray based angiography system, for this purpose, the dose area product in the optical path of the imaging system can be measured as the load characteristic value by means of a measuring device called a DAP chamber, and this dose area product adds up the loads in the irradiated surface of the radiation to obtain a value.

[0041] The load characteristic value depends in particular on the respective imaging system and its setting data. If the setting data changes, the load characteristic value also changes accordingly. In this regard, the received load characteristic value corresponds to the load characteristic value measured or alternatively determined in the imaging system set with the received setting data. The load characteristic value can in particular be stored or at least cached in a corresponding storage medium in the form of a corresponding computer-readable file, which computer-readable file includes the load characteristic value. The reception of the load characteristic value can be carried out, for example, by reading the computer-readable file from the storage medium.

[0042] The received intensity distribution corresponds to the intensity distribution measured at the radiation detector or alternatively determined at the radiation detector, i.e., without measurement, in the imaging system set with the received setting data.

[0043] The load distribution data can correspond to the detector image of the radiation detector, which in particular graphically shows the intensity distribution on the detector surface of the radiation detector irradiated by the radiation. In particular, the intensity distribution can be stored or at least cached in a corresponding storage medium in the form of a corresponding computer-readable file, preferably a two-dimensional image file, which image file includes the two-dimensional intensity distribution. The reception of the intensity distribution can be carried out, for example, by reading the computer-readable file from the storage medium.

[0044] In a second preferred embodiment, an intensity distribution can be provided by irradiating the radiation detector with the radiation of a radiation source under the setting data of the imaging system, where there is no object or a homogeneous object in the optical path, and where the radiation detector provides an intensity distribution as a two-dimensional detector image. A "homogeneous object" can be understood as an object that does not affect the intensity distribution. For example, acrylic plexiglass with a corresponding thickness can be used for this purpose.

[0045] The setting data, the load characteristic values, and the load distribution data can be used as input data for a trained first function. Preferably, the trained first function can be based on an artificial neural network (KNN), in particular with at least one convolutional layer. In particular, the trained function can be based on a general inverse algorithm. In particular, a computing unit can be designed to generate load data as first output data using the trained function.

[0046] KNN can be understood as an assembly of software code or multiple software code components, where the software code can include multiple software modules for different functions, such as one or more encoder modules and one or more decoder modules.

[0047] KNN can be understood as a non-linear model or algorithm that maps an input to an output, where the input is an input feature vector or an input sequence, and the output can be the output class of a classification task or a prediction sequence.

[0048] KNN can be provided, for example, in a computer-readable form.

[0049] A neural network can include multiple modules, which include encoder modules and decoder modules. The modules can be understood as software modules or corresponding parts of a neural network. A software module can be understood as software code that is functionally connected or combined with a unit. A software module can include or execute multiple processing steps and / or data structures.

[0050] In particular, a module itself is a neural network or a sub-network. Unless otherwise specified, the modules of a neural network can be understood as trainable modules of the neural network, especially trained modules. For example, before executing a program, the neural network and all its trainable modules can be trained end-to-end. However, in other embodiments, different modules can also be trained separately or pre-trained. In other words, the method according to the invention can correspond to the usage phase of a trained first function, especially KNN.

[0051] The trained first function can be trained in particular to be suitable for forming real and meaningful data of the load distribution as output data. Preferably, the first function can be trained by a training method executed by a computer to train and provide the trained first function.

[0052] By using the trained first function, the load distribution can be predicted extremely precisely and personalized. In addition, the computing unit can also generate load distribution data extremely quickly, robustly and with less computational effort when using the trained first function.

[0053] The trained first function can in particular be provided as a general function for a series of imaging systems that are identical or at least similar in construction, or be used for each specific imaging system in this series.

[0054] The proposed method can be used in particular during the irradiation of the body with the imaging system, for example during a medical operation, but can also be used for material inspection.

[0055] According to the invention, load distribution data is provided by applying a trained second function to setting data that is the second input data of the trained second function, wherein the load distribution data is generated as the second output data of the second function.

[0056] This is at least advantageous because there is no need to perform an actual irradiation to provide the intensity distribution, nor to perform actual measurements based on the setting data to determine the intensity distribution. Instead, the intensity distribution can be provided purely in a computer-executed manner.

[0057] Thus, in particular, the second output data of the trained second function can form part of the first input data of the trained first function. The trained second function provides the intensity distribution as a digital two-dimensional image, in particular as a corresponding image file.

[0058] The trained second function can in particular be provided as a specific function for each specific imaging system of a series of imaging systems that are identical or at least similar in construction, or be used only for this specific imaging system in this series. By associating the trained first function that can be generally used for this series of imaging systems with the trained second function, the trained first function can be to a certain extent specialized for this specific imaging system.

[0059] In particular, the trained first function and the trained second function can be combined with each other in an application, for example in a medical operation for irradiating a patient. Another advantage resulting therefrom is that the computing unit only needs to receive the setting data and the load characteristic value when using this combination in the application, and thus the computing unit can form the load distribution. This significantly accelerates and simplifies the method.

[0060] The trained second function can be trained in particular so that it is suitable for forming real and meaningful data of the intensity distribution as output data. Preferably, the second function can be trained by a training method executed by a computer to train and provide the trained second function. Preferably, the trained second function can be based on an artificial neural network (KNN), in particular having at least one convolutional layer. In particular, the trained second function can be based on a general inverse algorithm. In particular, a computing unit can be designed to generate intensity distribution data as the second output data using the trained second function.

[0061] According to at least one embodiment, the trained second function is provided by a second training method executed by a computer. The second training method executed by a computer particularly includes the steps of:

[0062] - Receiving second input training data, which includes setting data of the imaging system;

[0063] - Receiving second output training data, which includes load distribution data, wherein the second output training data is associated with the second input training data;

[0064] - Training the second function based on the second input training data and the second output training data by generating second output data by applying the second function to the second input training data and adjusting the parameters of the second function based on the comparison between the second output data and the second output training data;

[0065] - Providing the trained second function.

[0066] According to at least one embodiment, load characteristic values are provided by applying a trained third function to setting data that is the third input data of the trained third function, wherein the load characteristic values are generated as the third output data of the third function.

[0067] This is at least advantageous because there is no need to perform actual irradiation to provide load characteristic values, nor is it necessary to perform actual measurements based on the setting data to determine the load characteristic values. Instead, the load characteristic values can be provided purely in a computer-executed manner.

[0068] Thereby, in particular, the third output data of the trained third function can form part of the first input data of the trained first function. The trained third function provides the load characteristic values as numerical values for this purpose.

[0069] The trained third function can in particular be provided as a specific function for each specific imaging system of a series of imaging systems having the same or at least similar construction, or is only used for that specific imaging system in the series.

[0070] Thereby, other advantages are obtained from the combination of the trained third function with the trained second function and the trained first function. When using this combination, the computing unit only needs to receive the setting data, thereby being able to generate the load distribution. This significantly accelerates and simplifies the method.

[0071] The trained third function can be trained in particular so that it is suitable for forming real and meaningful load characteristic values as output data. Preferably, the third function can be trained by a training method executed by a computer to train and provide the trained third function. Preferably, the trained third function can be based on an artificial neural network (KNN), especially having at least one convolutional layer. In particular, the trained third function can be based on a general inverse algorithm. In particular, a computing unit can be designed to generate load characteristic values as the third output data using the trained function.

[0072] According to at least one embodiment, the trained third function is provided by a third training method executed by a computer. The third training method executed by a computer especially includes the steps of:

[0073] - Receiving third input training data, which includes the setting data of the imaging system;

[0074] - Receiving third output training data, which includes load characteristic values, wherein the third output training data is associated with the third input training data;

[0075] - Training the third function based on the third input training data and the third output training data by generating third output data by applying the third function to the third input training data and adjusting the parameters of the third function based on the comparison between the third output data and the third output training data;

[0076] - Providing the trained third function.

[0077] According to at least one embodiment, the load characteristic value is the dose area product or a quantity related to this mass area product.

[0078] For example, the radiation source can be an X-ray tube in this case. However, similar quantities can also be obtained for other types of ionizing radiation.

[0079] According to at least one embodiment, the load distribution data is a two-dimensional X-ray projection image. The intensity distribution can be displayed, for example, according to the color intensity of the individual pixels of this X-ray projection image.

[0080] The X-ray projection image, also called the detector image, can especially graphically display a two-dimensional array of pixels, wherein the pixel values can reflect the measurement values of the individual radiation sensors of the radiation detector.

[0081] According to at least one embodiment, the load distribution data is a two-dimensional image of the load distribution at the intervention reference point. Here, the intervention reference point is the position where the body to be irradiated is specified to be located in the optical path. Thus, through this load distribution, it is also possible to accurately indicate the load distribution acting on or supposed to act on the body to be radiotherapy.

[0082] According to at least one embodiment, the setting data of the imaging system at least has the operating parameters of the collimator and / or the radiation source of the imaging system.

[0083] For example, if the radiation source is an X-ray radiation source, especially an X-ray tube, the operating parameters of the radiation source can especially include the peak kilovoltage, kVp (English: peak kilovoltage), that is, the maximum tube voltage applied when generating X-ray radiation as ionizing radiation at the X-ray tube. The operating parameters of the radiation source can also include the tube current of the X-ray tube and / or the focal spot size (English: focal spot size), etc.

[0084] The collimator is a physical barrier that focuses or parallelizes high-energy electromagnetic waves, such as X-ray radiation or gamma radiation. For example, it allows the spatial limit of the radiation of the radiation source to be given or set according to the operating parameters of the collimator.

[0085] According to at least one embodiment, the imaging system is an X-ray based imaging system, and the radiation source is an X-ray radiation source, especially an X-ray tube.

[0086] Another aspect of the present invention relates to a first computer-executed training method for providing a trained first function that generates load distribution data of the radiation of the radiation source of the imaging system. This method can especially be implemented by a computing unit and especially has the following steps.

[0087] In the first step of the method, first input training data is received. The first input training data includes the setting data of the imaging system, the load characteristic values of the radiation corresponding to the setting data, and the load distribution data of the radiation at the radiation detector of the imaging system corresponding to the setting data.

[0088] In the second step of the method, output training data is received, and the output training data includes load distribution data. The output training data is associated with the input training data here.

[0089] In the third step, the first function is trained based on the first input training data and the first output training data by generating first output data by applying the first function to the first input training data and adjusting the parameters of the first function based on the comparison between the first output data and the first output training data.

[0090] In a fourth step, a trained first function is provided.

[0091] The input training data can be received using a first training interface of an electronic training system, and the output training data can be received via a second training interface of the electronic training system. The trained first function can be provided via a third training interface of the electronic training system.

[0092] Preferably, the first function can be based on a k-nearest neighbors (KNN) artificial neural network, especially having at least one convolutional layer. In particular, the function can be based on a general inverse algorithm. The electronic training system can in particular be designed for training the first function. The first function can in particular be trained such that it is suitable for forming or predicting, as the trained first function, real and meaningful data of the load distribution of radiation as output data.

[0093] By comparing the output data with the output training data, the error of the function can in particular be determined. For example, a so-called cost function can be calculated to quantify the error. For example, the parameters of the function can be adjusted by known algorithms such that the cost function is minimized, especially by iteration.

[0094] According to at least one embodiment of a first training method performed by a computer, the output training data is provided by dose measurement and / or dose simulation.

[0095] Here, the dose measurement as output training data should be carried out with the corresponding set data as input training data. The dose measurement can in particular be carried out with the aid of a radiochromic film. The film contains a pigment that changes color when irradiated with ionizing radiation, enabling the degree of irradiation, the radiation characteristics of the radiation, and the load distribution to be characterized. The film can in particular be digitized by scanning and digitally processed accordingly to form a two-dimensional digital image.

[0096] As an alternative, load distribution data can be provided as output training data based on the corresponding set data, load characteristic values, and intensity distribution by dose simulation. For this purpose, for example, a Monte Carlo simulation can be carried out.

[0097] The intensity distribution and / or the load characteristic values as input training data can likewise be based on measurement and / or simulation and / or provided with the aid of a trained function.

[0098] According to at least one embodiment of a first training method performed by a computer, the trained first function is provided for a series of imaging systems.

[0099] The trained first function can in particular be provided as a general function for a series of imaging systems that are identical or at least similar in construction, or be used for each specific imaging system in the series.

[0100] According to at least one embodiment of a method implemented by a computer for providing load distribution data, a trained first function is provided by a first training method implemented by the computer.

[0101] According to the present invention, a second training method implemented by a computer for providing a trained second function is provided. The trained second function generates load distribution data of radiation at a radiation detector of an imaging system having a radiation source. This method can in particular be implemented by a computing unit and in particular has the following steps.

[0102] In a first step of the method, second input training data is received, which second input training data includes setting data of the imaging system.

[0103] In a second step of the method, second output training data can be received, which second output training data includes load distribution data. The second output training data is associated with the second input training data here.

[0104] In a third step, a second function is trained based on the second input training data and the second output training data by generating second output data by applying the second function to the second input training data and adjusting the parameters of the second function based on a comparison of the second output data with the second output training data.

[0105] In a fourth step, a trained fourth function is provided.

[0106] The input training data can be received using a first training interface of an electronic training system, and the output training data can be received through a second training interface of the electronic training system. The trained second function can be provided through a third training interface of the electronic training system.

[0107] Preferably, the second function can be based on a k-nearest neighbors (KNN) artificial neural network, in particular having at least one convolutional layer. In particular, this function can be based on a general inverse algorithm. The electronic training system can in particular be designed to train the second function. The second function can in particular be trained so as to be suitable for forming or predicting, as the trained second function, real and meaningful data of the intensity distribution of radiation as output data.

[0108] By comparing the output data with the output training data, in particular the error of the function can be determined. For example, a so-called cost function can be calculated to quantify the error. For example, the parameters of the function can be adjusted by a known algorithm such that the cost function is minimized in particular by iteration.

[0109] According to at least one embodiment of a second training method performed by a computer, the trained second function is provided for a specific imaging system dedicated to a series of imaging systems. This series of imaging systems can in particular have imaging systems with the same structure or at least a similar structure.

[0110] The second output training data for training a specific second function can in particular be provided by means of the corresponding specific system. As the second output training data, the load distribution data of the specific system can in particular be completed by means of measurements of the specific system, where there is no object or there is a uniform object in the optical path. Such measurements can in particular be provided by regular, for example annual, maintenance measurements.

[0111] In particular for training the second function, it is sufficient to train the second function with a relatively small number of training data. The reason for this may be that the trained second function is associated with the trained first function when applied in a method performed by a computer for providing load distribution data. Here, the trained second function is only specialized for the specific system of the trained first function for the corresponding trained second function, where the trained first function is trained with a relatively small amount of training data. Therefore, a relatively small amount of training data in this combination is sufficient to train the second function.

[0112] According to at least one embodiment of a method performed by a computer for providing load distribution data, a trained second function is provided by a second training method performed by a computer.

[0113] Another aspect of the invention relates to a third training method performed by a computer for providing a trained third function, the trained third function generating data of the load characteristics of the radiation of a radiation source of an imaging method. This method can in particular be implemented by a computing unit and in particular has the following steps.

[0114] In a first step of the method, third input training data is received, the third input training data including the setting data of the imaging system.

[0115] In a second step of the method, second output training data can be received, the second output training data including load characteristics. The third output training data is associated with the third input training data here.

[0116] In a third step, the third function is trained based on the third input training data and the third output training data, in such a way that the third output data is generated by applying the third function to the third input training data, and the parameters of the third function are adjusted based on the comparison of the third output data with the third output training data.

[0117] In a fourth step, the trained third function is provided.

[0118] The input training data can be received using the first training interface of an electronic training system, and the output training data can be received through the second training interface of the electronic training system. The trained third function can be provided through the third training interface of the electronic training system.

[0119] Preferably, the third function can be based on a k-nearest neighbors (KNN) artificial neural network and particularly has at least one convolutional layer. In particular, the function can be based on a general inverse algorithm. The electronic training system can in particular be designed for training the second function. The second function can in particular be trained so as to be suitable for forming or predicting, as the trained second function, real and meaningful data of the intensity distribution of radiation as output data.

[0120] By comparing the output data with the output training data, the error of the function can in particular be determined. For example, a so-called cost function can be calculated to quantify the error. For example, the parameters of the function can be adjusted by a known algorithm such that the cost function is minimized in particular by iteration.

[0121] According to at least one embodiment of a third training method executed by a computer, the trained third function is provided specifically for a specific imaging system of a series of imaging systems. The series of imaging systems can in particular have imaging systems with the same or at least similar structures.

[0122] The third output training data for training a specific third function can in particular be provided with the aid of the corresponding specific system. The load characteristic value of the specific system can in particular be determined by measurement with the aid of the specific system as the third output training data.

[0123] According to at least one embodiment of a method executed by a computer for providing load distribution data, the trained third function is provided by a third training method executed by a computer.

[0124] Another aspect of the invention relates to a data processing device having at least one computing unit, the computing unit being adapted to execute a method executed by a computer according to the invention to provide load distribution data of the radiation of a radiation source of an imaging system.

[0125] In particular, the data processing device is used to provide load distribution data of the radiation of a radiation source of an imaging system, and the data processing device has:

[0126] - a first interface for receiving setting data of the imaging system;

[0127] - a second interface for receiving load characteristic values of the radiation corresponding to the setting data;

[0128] - A third interface for receiving load distribution data of radiation corresponding to the setting data at a radiation detector of an imaging system;

[0129] - A calculation unit for applying a trained first function to the setting data, load characteristic values, and load distribution data that are the first input data of the first function, wherein the load distribution data that is the first output data of the first function is generated;

[0130] - A fourth interface for providing the load distribution data.

[0131] The calculation unit can particularly be understood as a data processing device including a processing circuit. The calculation unit can particularly process data for performing calculation operations. The calculation operations may also include operations for performing retrieval access to data structures, such as a lookup table LUT (English: “look - up table”).

[0132] The calculation unit can particularly include one or more computers, one or more microcontrollers, and / or one or more integrated circuits, such as one or more application - specific integrated circuits, i.e., ASICs (English: “application - specific integrated circuit”), and / or one or more field - programmable gate arrays, i.e., FPGAs, and / or one or more systems - on - a - chip, i.e., SOCs (English: “system on a chip”). The calculation unit can also include one or more processors, such as one or more microprocessors, one or more central processing unit CPUs (English: “central processing unit”), one or more graphics processing unit GPUs (English: “graphics processing unit”), and / or one or more signal processors, particularly one or more digital signal processors DSPs. The calculation unit can also include an entity or virtual complex of a computer or other mentioned units.

[0133] In different embodiments, the calculation unit includes one or more hardware and / or software interfaces and / or one or more storage units.

[0134] The storage unit can be designed as a volatile data memory, such as a dynamic random access memory (DRAM) or a static random access memory (SRAM), or as a non-volatile data memory, such as a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a flash memory or a flash EEPROM, a ferroelectric random access memory (FRAM), a magnetoresistive random access memory (MRAM), or a phase-change random access memory (PCRAM).

[0135] Other embodiments of the data processing device according to the present invention directly result from different design solutions of the corresponding method according to the present invention, and vice versa. The respective features and corresponding explanations and advantages related to the various embodiments of the method according to the present invention can in particular be transferred analogously to the corresponding embodiments of the data processing device according to the present invention.

[0136] According to another aspect of the present invention, a computer program with instructions is provided. When the instructions are executed by at least one data processing device, in particular the data processing device according to the present invention, the instructions allow the data processing device to implement the method according to the present invention to provide load distribution data of the radiation of the radiation source of the imaging system.

[0137] The instructions can exist, for example, as program code. The program code can be provided, for example, in the form of binary code or an assembly program and / or the source code of a programming language, such as the C language, and / or a program script, such as Python.

[0138] Another aspect of the invention relates to an electronically readable data carrier comprising instructions which, when executed in at least one data processing device, in particular a data processing device according to the invention, allow the data processing device to carry out the method according to the invention to provide load distribution data of the radiation of a radiation source of an imaging system.

[0139] A computer program and a computer-readable storage medium are computer program products having instructions.

[0140] According to another aspect of the invention, there is provided an imaging system. The imaging system has a data processing device according to the invention, a radiation source for generating in particular ionizing radiation, and a radiation detector for detecting a part of the radiation, in particular ionizing radiation, penetrating a part of the object to be imaged.

[0141] Another aspect of the invention relates to a first electronic training system for providing a trained first function which generates load distribution data of the radiation of a radiation source of an imaging system, the first electronic training system comprising:

[0142] - A first training interface for receiving first input training data, including

[0143] - Setup data of the imaging system,

[0144] - A load characteristic value of the radiation corresponding to the setup data, and

[0145] - Load distribution data of the radiation on the radiation detector of the imaging system corresponding to the setup data, which is associated with the setup data;

[0146] - A second training interface for receiving first output training data, which includes load distribution data, wherein the output training data is associated with the input training data;

[0147] - A calculation unit for training the first function based on the first input training data and the first output training data, in such a way that first output data is generated by applying the first function to the first input training data, and the parameters of the first function are adjusted based on the comparison between the first output data and the first output training data;

[0148] - A third training interface for providing the trained first function.

[0149] The first electronic training system is in particular configured to carry out a first training method executed by a computer.

[0150] Other embodiments of the training system according to the invention directly result from different design concepts of the training system according to the invention and vice versa. The individual features and the corresponding explanations and advantages relating to the various embodiments of the training method according to the invention can in particular be transferred analogously to the corresponding embodiments of the training system according to the invention.

[0151] Another aspect of the invention relates to a second electronic training system for providing a trained second function which generates load distribution data of radiation at a radiation detector of an imaging system having a radiation source, the second electronic training system comprising:

[0152] - a fourth training interface for receiving second input training data comprising setting data of the imaging system;

[0153] - a fifth training interface for receiving second output training data comprising load distribution data, wherein the output training data is associated with the input training data;

[0154] - a computing unit for training the second function based on the second input training data and the second output training data by generating second output data by applying the second function to the second input training data and adjusting parameters of the second function based on a comparison of the second output data with the second output training data;

[0155] - a sixth training interface for providing the trained second function.

[0156] The second electronic training system is in particular configured to carry out a second training method executed by a computer.

[0157] Other embodiments of the training system according to the invention directly result from different design concepts of the training system according to the invention and vice versa. The individual features and the corresponding explanations and advantages relating to the various embodiments of the training method according to the invention can in particular be transferred analogously to the corresponding embodiments of the training system according to the invention.

[0158] Another aspect of the invention relates to a third electronic training system for providing a trained third function which generates load characteristic values of radiation of a radiation source of an imaging system, the third electronic training system comprising:

[0159] - a seventh training interface for receiving third input training data comprising setting data of the imaging system;

[0160] - an eighth training interface for receiving first output training data comprising load characteristic values, wherein the output training data is associated with the input training data;

[0161] - A computing unit for training a third function based on third input training data and third output training data by applying the third function to the third input training data to generate third output data and adjusting the parameters of the third function based on a comparison between the third output data and the third output training data;

[0162] - A ninth training interface for providing the trained third function.

[0163] The third electronic training system is particularly configured to implement a computer-executed third training method.

[0164] Other embodiments of the training system according to the invention directly result from different design options of the training system according to the invention, and vice versa. The respective features, corresponding explanations, and advantages related to the various embodiments of the training method according to the invention can in particular be transferred analogously to the corresponding embodiments of the training system according to the invention.

[0165] Another aspect of the invention relates to a computer training program comprising instructions which, when the corresponding computer program runs in an electronic training system, cause the electronic training system to implement the corresponding training method according to the invention.

[0166] Another aspect of the invention relates to an electronically readable training data carrier comprising instructions which, when the corresponding data carrier is executed in an electronic training system, cause the electronic training system to implement the corresponding training method according to the invention.

[0167] For application scenarios or application cases that can be derived in the method or training method and are not described in detail here, it can be stipulated that a fault report and / or a request for inputting user feedback and / or an adjustment of a standard setting and / or a predetermined initial state are output according to the method.

[0168] Other features and feature combinations of the invention result from the drawings, the description of the drawings, and the embodiments. In particular, other embodiments of the invention do not necessarily include all features. Other embodiments of the invention can have features or feature combinations not mentioned in the description. Description of the Drawings

[0169] The invention will be explained in more detail below with the aid of specific embodiments and the associated schematic diagrams. In the drawings, identical or functionally identical elements can be provided with the same reference signs. Where necessary, the same or functionally identical elements do not need to be described repeatedly with respect to different drawings.

[0170] In the drawings:

[0171] Figure 1 A schematic diagram showing an exemplary embodiment of an imaging system according to the invention;

[0172] Figure 2 Flow schematic diagram showing an exemplary embodiment of a computer-executed method for providing load distribution data according to the present invention;

[0173] Figure 3 Schematic diagram showing an exemplary embodiment of a computer-executed method for providing load distribution data according to the present invention;

[0174] Figure 4 Flow schematic diagram showing an exemplary embodiment of a first training method executed by a computer for providing a trained first function according to the present invention;

[0175] Figure 5 Schematic flowchart showing an exemplary embodiment of a second training method executed by a computer for providing a trained second function according to the present invention;

[0176] Figure 6 Flow schematic diagram showing an exemplary embodiment of a method executed by a computer for providing a trained third function according to the present invention;

[0177] Figure 7 Schematic diagram showing an embodiment of a trained first function;

[0178] Figure 8 Schematic diagram showing an embodiment of a trained second function;

[0179] Figure 9 Schematic diagram showing a first embodiment of an artificial neural network (KNN) of a trained function;

[0180] Figure 10 Schematic diagram showing a first embodiment of a convolutional artificial neural network (KNN) of a trained function. Detailed Description

[0181] In Figure 1 An exemplary embodiment of an imaging system 1 according to the present invention is schematically shown. The imaging system 1 has at least one data processing device 2 and an imaging module, the imaging module having a radiation source 4 for forming ionizing radiation 6 and a radiation detector 3 for detecting partial current radiation 6. For example, the imaging module is designed as an X-ray based imaging module, for example, designed as an X-ray based angiography system, and has an X-ray source as the radiation source 4 and an X-ray detector as the radiation detector 3. In addition, a patient is shown as an object 5 to be imaged, which is at the intervention reference point 9 of the imaging system 1.

[0182] The data processing device 2 has at least one computing unit 7, 17. At least one computing unit 7 is designed to implement the computer-executed method M1 according to the invention to provide load distribution data of the radiation 6 of the radiation source 4 of the imaging system 1. Figure 2 A flow schematic diagram showing an exemplary embodiment of such a method M1 is shown. For this purpose, the corresponding computer program product 8 can be executed on the data processing device 2.

[0183] In Figure 2 in a flow schematic diagram and in Figure 3 in a schematic diagram, a schematic diagram of an exemplary embodiment of the computer-executed method M1 according to the invention for providing load distribution 16 data is shown, which method can in particular be executed by the computing unit 7, in particular the computing unit of the data processing device 2. The following is described together Figure 2 and Figure 3 .

[0184] In a first step S1 of the method M1, the setting data 10 of the imaging system 1 can be received by the computing unit 7. In an embodiment, the setting data 10 of the imaging system 1 can at least have the operating parameters of the radiation source 4 and / or the collimator of the imaging system 1.

[0185] In a second step S2 of the method M1, the intensity distribution 15 data of the radiation 6 corresponding to the setting data 10 at the radiation detector 3 of the imaging system 1 can be received by the computing unit 7. For example, the intensity distribution 15 data can be a detector image, in particular a two-dimensional X-ray projection image.

[0186] Preferably, the intensity distribution 15 data is provided by applying a trained second function 12 to the setting data 10 as the second input data of the trained second function 12, wherein the intensity distribution 15 data is generated as the second output data of the trained second function 12. The trained second function 12 can in particular be used by other computing units 17, wherein the other computing units 17 and this computing unit 7 can be configured as a common computing unit 7.

[0187] In a third step S3 of the method M1, the load characteristic value 14 of the radiation 6 corresponding to the setting data 10 can be received by the computing unit 7. For example, the load characteristic value 14 can be a dose area product (English: dose areaproduct, DAP) or a quantity related to this dose area product.

[0188] Preferably, by applying the trained third function 13 to the setting data 10 that is the third input data of the trained third function 13, a load characteristic value 14 is provided, where the load characteristic value 14 is generated as the third output data of the trained third function 13. The trained third function 13 can be used in particular by other computing units 17, where the other computing units 17 and this computing unit 7 can be configured as a common computing unit 7.

[0189] Alternatively, the load characteristic value can be measured by means of a DAP chamber in the optical path of the imaging system.

[0190] In the fourth step S4 of the method M1, the trained first function 11 can be applied to the setting data 10, the load characteristic value 14, and the intensity distribution 15 data that are the first input data of the trained first function 11, where the load distribution 16 data is formed as the first output data of the trained first function.

[0191] The load distribution 16 data can in particular be a two-dimensional image of the load distribution 16 at the intervention reference point 9 of the imaging system 1.

[0192] In the fifth step S5, the load distribution 16 data is provided.

[0193] Figure 4 A flowchart showing an exemplary embodiment of a first training method M2 executed by a computer according to the present invention, which is used to provide a trained first function 11, and the first training method can be implemented by a first electronic training system 18 having at least one computing unit.

[0194] In the first step S21, first input training data is received, and the first input training data includes the setting data 10' of the imaging system 1, the load characteristic value 14' of the radiation 6 corresponding to the setting data (10'), and the intensity distribution 15' data of the radiation 6 corresponding to the setting data (10') at the radiation detector 3 of the imaging system 1.

[0195] In the second step S22, first output training data is received, including the load distribution (16') data, where the first output training data is associated with the first input training data;

[0196] In the third step S23, the first function is trained based on the first input training data and the first output training data, in such a way that the first output data is generated by applying the first function to the first input training data, and the parameters of the first function are adjusted based on the comparison between the first output data and the first output training data.

[0197] In the fourth step S24, the trained first function 11 is provided.

[0198] Figure 5 A flowchart showing an exemplary embodiment of a second computer-executed training method M3 according to the present invention, which is used to provide a trained second function 12, and the second training method can be implemented by a second electronic training system 19 having at least one computing unit.

[0199] In a first step S31, second input training data is received, and the second input training data includes the setting data 10' of the imaging system 1.

[0200] In a second step S32, second output training data is received, including intensity distribution (15') data, wherein the second output training data is associated with the second input training data.

[0201] In a third step S33, the second function is trained based on the second input training data and the second output training data, in such a way that the second output data is generated by applying the second function to the second input training data, and the parameters of the second function are adjusted based on the comparison between the second output data and the second output training data.

[0202] In a fourth step S34, the trained second function 12 is provided.

[0203] Figure 6 A flowchart showing an exemplary embodiment of a third computer-executed training method M4 according to the present invention, which is used to provide a trained third function 13, and the third training method can be implemented by a third electronic training system 20 having at least one computing unit.

[0204] In a first step S41, third input training data is received, and the third input training data includes the setting data 10' of the imaging system 1.

[0205] In a second step S42, third output training data is received, including load characteristic values (14'), wherein the third output training data is associated with the third input training data.

[0206] In a third step S43, the third function is trained based on the third input training data and the third output training data, in such a way that the third output data is generated by applying the third function to the third input training data, and the parameters of the third function are adjusted based on the comparison between the third output data and the third output training data.

[0207] In a fourth step S44, the trained third function 13 is provided.

[0208] Figure 7Schematic diagram showing an embodiment of a trained first function 11. In this embodiment, the trained first function 11 may at least have an encoder 21 and a decoder 22.

[0209] In particular, the intensity distribution 15 data is shown here, for example, as a graph in which the radiation intensity in the two-dimensional XY plane is plotted on the Z-axis. The intensity distribution 15 data can be used as input data for the encoder, where the intensity distribution 15 data with variable length can be encoded into a sequence with a fixed length. Preferably, the setting data 10 and the load eigenvalue 14 already have sequences with fixed lengths and can thus be directly used as input data for the decoder 22. The decoder 22 is then configured to form load distribution 16 data based on the input data. The load distribution data is shown here, for example, as a graph in which the load intensity in the two-dimensional XY plane is plotted on the Z-axis.

[0210] Figure 8 Schematic diagram showing an embodiment of a trained second function 12. In this embodiment, the trained second function 12 may at least have a decoder 22. Preferably, the setting data 10 already has a sequence with a fixed length and can thus be directly used as input data for the decoder 22. The decoder 22 is then configured to form intensity distribution 16 data based on the input data. The load distribution data is shown here, for example, as a graph in which the radiation intensity in the two-dimensional XY plane is plotted on the Z-axis.

[0211] Figure 9 Schematic showing an embodiment of an artificial neural network (KNN) 100 for a function or a trained first function and / or a trained second function 12 and / or a trained third function 13. Alternative terms for "artificial neural network" are "neural network", "artificial neural network", or "neural network".

[0212] The artificial neural network 100 includes nodes 120,......, 132 and edges 140,......, 142, where each edge 140,......, 142 is a directed connection from a first node 120,......, 132 to a second node 120,......, 132. Generally, the first node 120,..., 132 and the second node 120,..., 132 are different nodes 120,..., 132, and the first node 120,..., 132 and the second node 120,..., 132 can also be the same. For example, in Figure 6In this case, edge 140 is a directed connection from node 120 to node 123, and edge 142 is a directed connection from node 130 to node 132. The edges 140,......, 142 from the first nodes 120,......, 132 to the second nodes 120,......, 132 are also referred to as the "incoming edges" of the second nodes 120,......, 132 and the "outgoing edges" of the first nodes 120,......, 132.

[0213] In this embodiment, the nodes 120,......, 132 of the artificial neural network 100 can be arranged as layers 110,......, 113, where these layers can have an inherent hierarchy introduced by the edges 140,......, 142 between the nodes 120,......, 132. In particular, the edges 140,......, 142 may only exist between adjacent node layers. In the illustrated embodiment, there is an input layer 110, an output layer 113, and hidden layers 111, 112 between the input layer 110 and the output layer 113. The input layer only includes nodes 120,......, 122 that have no incoming edges, and the output layer only includes nodes 131, 132 that have no outgoing edges. Generally, the number of hidden layers 111, 112 can be arbitrarily selected. The number of nodes 120,..., 122 in the input layer 110 generally refers to the number of input values of the neural network, and the number of nodes 131, 132 in the output layer 113 generally refers to the number of output values of the neural network.

[0214] In particular, a (real) number can be assigned to each node 120,..., 132 of the neural network 100 as a value. Here, x(n)i represents the value of the i-th node 120,..., 132 in the n-th layer 110,..., 113. The values of the nodes 120,..., 122 in the input layer 110 correspond to the input values of the neural network 100, and the values of the nodes 131, 132 in the output layer 113 correspond to the output values of the neural network 100. In addition, each edge 140,......, 142 can have a weight, which is a real number, and in particular, a real number within the interval [-1, 1] or within the interval [0, 1]. Here, w(m,n)i,j represents the weight of the edge between the i-th node 120,......, 132 in the m-th layer 110,......, 113 and the j-th node 120,......, 132 in the n-th layer 110,......, 113. In addition, the abbreviation w(n)i,j refers to the weight w(n,n + 1)i,j.

[0215] To calculate the output value of the neural network 100, the input value is propagated through the neural network in particular. In particular, the values of the nodes 120, ..., 132 of the (n + 1)-th layer 110, ..., 113 can be calculated based on the values of the nodes 120, ..., 132 of the n-th layer 110, ..., 113 by the following formula

[0216]

[0217] Here, the function f is a transfer function (an alternative term is "activation function"). Known transfer functions are step functions, sigmoid functions (such as logistic functions, generalized logistic functions, hyperbolic tangent functions, arctangent functions, error functions, smoothstep functions), or rectifier functions. Transfer functions are mainly used for normalization.

[0218] In particular, the value is propagated layer by layer through the neural network, where the value of the input layer 110 is given by the input of the neural network 100, where the value of the first hidden layer 111 can be calculated based on the value of the input layer 110 of the neural network, where the value of the second hidden layer 112 can be calculated based on the value of the first hidden layer 111, and so on.

[0219] To determine the value of the edge w(m,n)i,j, the neural network 100 must be trained with training data. The training data particularly includes training input data and training output data (denoted as ti). In the training step, the neural network 100 is applied to the training input data to generate the calculated output data. In particular, the training data and the calculated output data include a certain number of values, and this number corresponds to the number of nodes in the output layer.

[0220] In particular, the comparison between the calculated output data and the training data is used to recursively adjust the weights of the neural network 100 (backpropagation algorithm). In particular, the weights are changed according to the following formula

[0221]

[0222] where γ is the learning rate, and δ(n)j can be recursively calculated as

[0223]

[0224] When the (n + 1)-th layer is not the output layer, based on δ (n+1) j , and

[0225]

[0226] If the output layer is the (n + 1)-th layer 113, where f’ is the first derivative of the activation function and y(n + 1)j is the comparison training value of the j-th node 17, 18 of the output layer 113.

[0227] Figure 10 An embodiment of a convolutional neural network 200 for a function or a trained function is shown. In the illustrated embodiment, the convolutional neural network 200 includes an input layer 210, a convolutional layer 211, a pooling layer 212, a fully connected layer 213, and an output layer 214. Alternatively, the convolutional neural network 200 may also include multiple convolutional layers 211, multiple pooling layers 212, multiple fully connected layers 213, and other types of layers. The order of these layers can be arbitrarily selected, but usually the fully connected layer 213 serves as the last layer before the output layer 214.

[0228] In particular, in the convolutional neural network 200, the nodes 220,..., 224 of the layers 210,..., 214 can be regarded as d-dimensional matrices or d-dimensional images. In particular, in the two-dimensional case, the values of the nodes 220,..., 224 labeled by i and j in the n-th layer 210,..., 214 can be expressed as x(n)[i, j]. However, the layout of the nodes 220,..., 224 of the layers 210,..., 214 has no influence on the calculations performed in the convolutional neural network 200, because these calculations are only given by the structure of the edges and the weights.

[0229] In particular, the convolutional layer 211 is characterized in that the structure and weights of the incoming edges constitute a convolutional operation based on a specific number of kernels. In particular, the structure and weights of the incoming edges are selected such that the value x of the node 221 of the convolutional layer 211 (n) k is calculated based on the value x of the node 220 of the previous layer 210 (n-1) as the convolution x (n) k = K k * x (n-1) where the convolution is defined in the two-dimensional case as

[0230]

[0231] Here, the k-th kernel K kis a d-dimensional matrix (in this case a two-dimensional matrix), which is typically small compared to the number of nodes 220, ..., 224 (e.g., a 3×3 matrix or a 5×5 matrix). This particularly means that the weights of the incoming edges are not independent, but are chosen such that the said convolution equation is obtained. In particular for a 3×3 matrix kernel, there are only 9 independent weights (each entry of the kernel matrix corresponds to an independent weight), regardless of the number of nodes 220, ..., 224 in the corresponding layers 210, ..., 214. In particular, for the convolutional layer 211, the number of nodes 221 in the convolutional layer is equal to the number of nodes 220 in the previous layer 210 multiplied by the number of kernels.

[0232] When the nodes 220 of the previous layer 210 are arranged as a d-dimensional matrix, the use of multiple kernels can be interpreted as adding an additional dimension (referred to as the "depth" dimension), such that the nodes 221 of the convolutional layer 221 are arranged as a (d + 1)-dimensional matrix. If the nodes 220 of the previous layer 210 are already arranged as a (d + 1)-dimensional matrix with a depth dimension, then the use of multiple kernels can be interpreted as an expansion along the depth dimension, such that the nodes 221 of the convolutional layer 221 are also arranged as a (d + 1)-dimensional matrix, where the size of the (d + 1)-dimensional matrix in terms of the depth dimension is a factor of the number of kernels greater than in the previous layer 210.

[0233] The advantage of using the convolutional layer 211 is the ability to exploit the spatial local correlation of the input data by forcing a local connection pattern between the nodes of adjacent layers, in particular by connecting each node only to a small region of the nodes of the previous layer.

[0234] In the illustrated embodiment, the input layer 210 includes 36 nodes 220, which are arranged as a two-dimensional 6×6 matrix. The convolutional layer 211 includes 72 nodes 221, which are arranged as two two-dimensional 6×6 matrices, where each of these two matrices is the result of the convolution of the values of the input layer with the kernel. Equivalently, the nodes 221 of the convolutional layer 211 can be interpreted as a three-dimensional 6×6×2 matrix, where the last dimension is the depth dimension.

[0235] The pooling layer 212 can be characterized by the structure and weights of the incoming edges and the activation function of its nodes 222, which together constitute a pooling operation based on the non-linear pooling function f. In the two-dimensional case, the value x of the nodes 222 of the pooling layer 212 (n) can be calculated, for example, based on the values x of the nodes 221 of the previous layer 211 (n-1) as follows:

[0236]

[0237] In other words, by using the pooling layer 212, the number of nodes 221, 222 can be reduced in such a way that the adjacent nodes 221 with quantities d1, d2 in the previous layer 211 are replaced by a single node 222, and the single node is calculated in the pooling layer as a function of the values of a certain number of adjacent nodes. The pooling function f can in particular be a maximum function, an average value or an L2 norm. In particular, for the pooling layer 212, the weights of the incoming edges are fixed and not changed due to training.

[0238] The advantage of using the pooling layer 212 is to reduce the number of nodes 221, 222 and the number of parameters. This reduces the amount of computation in the network and overfitting is monitored.

[0239] In the illustrated embodiment, the pooling layer 212 is a max pooling, in which four adjacent nodes are replaced by only one node, and the value of the node is the maximum value among the values of the four adjacent nodes. Max pooling is applied to each d-dimensional matrix of the previous layer; in this embodiment, max pooling is applied to each of the two two-dimensional matrices, thereby reducing the number of nodes from 72 to 18.

[0240] The characteristics of the fully connected layer 213 can be that there are most, in particular all, edges between the nodes 222 of the previous layer 212 and the nodes 223 of the fully connected layer 213, and the weight of each edge can be set individually.

[0241] In this embodiment, the nodes 222 in the previous layer 212 of the fully connected layer 213 are represented both as a two-dimensional matrix and additionally as non-continuous nodes (represented as a row of nodes, where the number of nodes is reduced for improved representability). In this embodiment, the number of nodes 223 in the fully connected layer 213 is equal to the number of nodes 222 in the previous layer 212. Alternatively, the numbers of nodes 222 and 223 can also be different.

[0242] Furthermore, in this embodiment, the value of the node 224 of the output layer 214 is determined by applying the softmax function to the values of the nodes 223 of the previous layer 213. By applying the softmax function, the sum of the values of all the nodes 224 of the output layer is 1, and all the values of all the nodes 224 of the output layer are real numbers between 0 and 1. In particular, when using the convolutional neural network 200 to classify input data, the values of the output layer can be interpreted as the probabilities that the input data belong to one of the different classes.

[0243] The convolutional neural network 200 may also include a rectified linear units (ReLU) layer. In particular, the number of nodes and the node structure in the ReLU layer match the number of nodes and the node structure in the previous layer. In particular, the value of each node in the ReLU layer is calculated by applying a rectification function to the value of the corresponding node in the previous layer. Examples of rectification functions are f(x) = max(0, x), the hyperbolic tangent function, or the sigmoid function.

[0244] The convolutional neural network 200 may in particular be trained based on the backpropagation algorithm. To prevent overfitting, regularization methods may be used, such as omitting nodes 220, ..., 224, random pooling, using artificial data, weight removal based on L1 or L2 norms or the maximum norm constraint.

[0245] The embodiments generally show how a machine learning method can be provided to estimate the X-ray load distribution.

[0246] Non-uniform X-ray radiation results in a non-uniform load distribution. The resulting estimate of the skin (organ...) dose is incorrect because the local dose is higher / lower than the dose determined using the DAP chamber.

[0247] An important aspect of the embodiments of the present invention is to predict non-uniform radiation (dose distribution) with the aid of a machine learning module, in particular a trained function, which has been trained based on the measured load and detector intensity distribution.

[0248] Non-uniform radiation results in a non-uniform load distribution in the radiation and a non-uniform intensity distribution at the detector.

[0249] Therefore, an important aspect of the embodiments of the present invention is to use the intensity distribution of the detectors of a specific system in order to "personalize" the general load distribution model.

[0250] An important aspect of the embodiments of the present invention is to combine two machine learning models, in particular a trained first function and a trained second function. The trained first function may predict the dose distribution in the optical path based on the intensity distribution of the detector (no object or uniform object in the optical path), the tube settings and the collimator settings, and the DAP (dose area product). This function may be trained based on dose measurements (e.g., using a film) or simulations (e.g., Monte Carlo simulations) for multiple systems and detector images, as well as tube settings and collimator settings.

[0251] The trained second function predicts a system-specific detector intensity distribution (no object or homogeneous object) based on tube settings and collimator settings. The model can be trained based on data that has been collected, for example, during annual maintenance measurements (detector images as well as tube settings and collimator settings).

[0252] During use, for example, during a medical procedure in which a patient is in the optical path, the two models, i.e., the trained first function and the trained second function, can be combined so that the dose distribution in the optical path can be predicted based on tube settings and collimator settings.

[0253] This enables a more accurate calculation of the X-ray dose to be advantageously ensured. In particular, it can be ensured that the local peak dose (skin dose) is not underestimated.

Claims

1. A computer-implemented method (M1) for providing data on a load distribution (16) of radiation (6) of a radiation source (4) of an imaging system (1), the method comprising the following steps: - receiving setting data (10) of the imaging system (1); - receiving a load characteristic value (14) of the radiation (6) corresponding to the setting data (10); - receiving data of an intensity distribution (15) of radiation (6) at a radiation detector (3) of the imaging system (1) corresponding to the setting data (10); - applying the trained first function (11) to the setting data (10), the load characteristic value (14) and the intensity distribution (15) data as first input data of the trained first function (11), wherein load distribution (16) data are generated as first output data of the trained first function (11); - providing said load distribution (16) data; The intensity distribution (15) data are provided by applying a trained second function (12) to setting data (10) as second input data of the trained second function (12), wherein the intensity distribution (15) data are generated as second output data of the trained second function (12).

2. The computer-implemented method (M1) according to claim 1, wherein: A trained second function (12) is provided by a second training method (M3) executed by a computer, the second training method comprising the following steps: - receiving second input training data comprising setting data (10') of said imaging system (1); - receiving second output training data comprising intensity distribution (15') data, wherein said second output training data is associated with said second input training data; - training the second function based on the second input training data and the second output training data by generating the second output data by applying the second function to the second input training data, and adjusting a parameter of the second function based on a comparison of the second output data with the second output training data; - providing a trained second function (12).

3. A computer-implemented method (M1) according to any one of the preceding claims, wherein: The load characteristic value (14) is provided by applying a trained third function (13) to setting data (10) as third input data of the trained third function (13), wherein the load characteristic value (14) is generated as third output data of the trained third function (13).

4. The computer-implemented method (M1) according to claim 3, wherein: A trained third function (13) is provided by a third training method (M4) executed by a computer, the third training method comprising the following steps: - receiving third input training data comprising setting data (10') of said imaging system (1); - receiving third output training data, which comprises a load characteristic value (14'), wherein the third output training data is associated with the third input training data; - training the third function based on the third input training data and third output training data by generating third output data by applying the third function to the third input training data, and adjusting parameters of the third function based on a comparison of the third output data with the third output training data; - providing a trained third function (13).

5. A computer-implemented method (M1) according to any one of the preceding claims, wherein: The load characteristic value (14) is a dose-area product or a quantity related to the dose-area product.

6. A computer-implemented method (M1) according to any one of the preceding claims, wherein: The intensity distribution (15) data is a two-dimensional X-ray projection image.

7. A computer-implemented method (M1) according to any one of the preceding claims, wherein: The load distribution (16) data is a two-dimensional image of the load distribution (16) at the intervention reference point (9) of the imaging system (1).

8. A computer-implemented method (M1) according to any one of the preceding claims, wherein: The setting data (10) of the imaging system (1) include at least operating parameters of a radiation source (4) and / or a collimator of the imaging system (1).

9. A computer-implemented method (M1) according to any one of the preceding claims, wherein: The imaging system (1) is an X-ray based imaging system and the radiation source (4) is an X-ray radiation source.

10. A computer-implemented training method (M2), the training method being used to provide a trained first function (11), the first function generating load distribution (16) data of radiation (6) of a radiation source (4) of an imaging system (1), and the training method being used to provide a trained second function (12), the second function generating intensity distribution (15) data of radiation (6) of the radiation source (4) of the imaging system (1), the training method comprising the following steps: - receiving first input training data, the first input training data comprising - setting data (10) of the imaging system (1), - a load characteristic value (14) of the radiation (6) corresponding to the setting data (10), and - data of intensity distribution (15) of radiation (6) at a radiation detector (3) of the imaging system (1) corresponding to the setting data (10); - receiving first output training data comprising intensity distribution (16') data, wherein the first output training data is associated with the first input training data; - training the first function based on the first input training data and first output training data by generating first output data by applying the first function to the first input training data, and adjusting a parameter of the first function based on a comparison of the first output data with the first output training data; - providing a trained first function (11); - receiving second input training data comprising setting data (10') of said imaging system (1); - receiving second output training data comprising intensity distribution (15') data, wherein said second output training data is associated with said second input training data; - training the second function based on the second input training data and the second output training data by generating second output data by applying the second function to the second input training data, and adjusting a parameter of the second function based on a comparison of the second output data with the second output training data; - providing a trained second function (12).

11. The computer-implemented training method (M2) according to claim 10, wherein: These output training data are provided by dose measurements and / or dose simulations.

12. A data processing device (2) having at least one computing unit (7) which is suitable for carrying out a computer-implemented method (M1) according to any one of claims 1 to 9.

13. A computer program product (8) comprising instructions which, when the computer program product (8) is run in a data processing device (2), cause the data processing device (2) to carry out the method (M1) according to any one of claims 1 to 9.

14. An imaging system (1) comprising a data processing device (2) according to claim 12, a radiation source (4) for generating radiation (6), and a radiation detector (3), wherein the radiation detector is used to detect a portion of the ionizing radiation (6), in particular a portion of the radiation (6) that has passed through an object (5) to be imaged.

Images