Dose measurement for monitoring medical radiation device

Abstract

A first aspect of the present disclosure relates to a computer-implemented method for monitoring dose output of a medical radiation device, comprising the steps of: obtaining training dose data from a reference device; training a machine learning algorithm (MLA) with the training dose data to instruct the MLA to identify the reference device based on dose data from the reference device that is different from the training dose data; obtaining operation dose data from the medical radiation equipment to be monitored; providing the operational dose data to the trained MLA; output information related to a difference between the reference device and the medical radiation device is provided based on MLA.

Description

Technical Field

[0001] The present disclosure relates to a method for monitoring the dose output of a medical radiation device, a device for training a machine learning algorithm for monitoring the dose output of a medical radiation device, a device for monitoring a medical radiation device, and a medical radiation device to be monitored by a method with a machine learning algorithm. Background Art

[0002] Medical linear accelerators produce high-energy photon radiation used to treat oncological diseases. These devices are not the same. That is, the radiation characteristics of different devices vary from device to device. In order to characterize and configure the appropriate beam for medical treatment, the radiation characteristics of the device must be analyzed.

[0003] Radiation characteristics can be analyzed by dosimetry. Dosimetry is usually performed in a water tank and identifies the emitted particles (e.g. photons, electrons, positrons) based on the energy released when they collide with other matter. Such dosimetry is inherently noisy. Measurements can also be obtained in solids using single point detectors or detector arrays.

[0004] Furthermore, dose measurements are often subject to errors. Measurement errors can lead to malfunctions, for example due to the detectors used being not ideal. Furthermore, configuration errors and operating errors can occur during the measurement process. Furthermore, radiation equipment is often not perfectly tuned. Therefore, in practice, dose measurements are subject to common and individual errors.

[0005] Although there are numerous guidelines for measuring dose data, there are fewer methods for ensuring their quality. In particular, plausibility checks or standard statistical analyses are themselves often subject to errors and are inaccurate.

[0006] The measurement data often constitutes an input to, or is used to configure, an algorithm that calculates the dose to be applied to a patient, for example for cancer treatment. Thus, the measurement data often constitutes an input to an algorithm that is used to calculate the dose to be applied to a patient for the treatment of cancer. Measurement errors can lead to incorrect configuration of these algorithms.

[0007] Inaccurate dose calculations for patient treatment are a problem with linear accelerator therapy. Improvements in this area are desirable. Summary of the invention

[0008] An object of the present disclosure is to improve medical radiation therapy.

[0009] This object is achieved by the disclosed embodiments, which are particularly defined by the subject matter of the independent claims. Dependent claims provide for further embodiments. Various aspects and embodiments of these aspects providing additional features and advantages are also disclosed in the following summary and description.

[0010] A first aspect of the present disclosure relates to a computer-implemented method for monitoring the dose output of a medical radiation device,

[0011] Includes steps:

[0012] - Obtain training dose data from a reference device;

[0013] - training a machine learning algorithm MLA with the training dose data to teach the MLA to recognize the reference device based on dose data from the reference device that is different from the training dose data; - obtaining operating dose data from the medical radiation device to be monitored;

[0014] - providing said operational dose data to said trained MLA;

[0015] - providing output information related to differences between the reference device and the medical radiation device based on the MLA.

[0016] Typically, dose data from medical radiation devices is derived from a complex quantum mechanical process whereby high-energy particles travel through matter and dissipate energy in a random manner. The dose data includes randomness and / or probability.

[0017] Dose data can be provided as an output factor (OF), which provides the dose of a medical radiation device normalized to a calibrated dose. OF can also be called "relative dose" or "relative dose factor". For a given (photon) beam at a given source-to-surface distance, the dose output at a certain depth in a medical radiation device generally depends on the size of the field. OF can be defined as the ratio of the dose for any field size (e.g., length × width in cm^2) in a water tank to the dose for a reference field of the same source-to-surface distance and the same depth. The reference field can be a square field, for example, a field of size 10 × 10 cm^2 with a source-to-surface distance of 100 cm. OF can be considered an intrinsic characteristic of the accelerator and can be measured regularly to ensure that the accelerator is operating correctly to treat patients. Other defined output factors or output quantities can also be used.

[0018] Dose data can also be provided as a depth-dose-curve (DDC). The DDC provides the dose distribution of a medical radiation device along the beam direction. For a given (photon) beam at a given source-to-surface distance, the DDC is obtained by performing a dose scan along the beam between the surface of a water phantom and a maximum depth determined by the phantom dimensions. The DDC also depends on the field size and, like the OF, can be considered an intrinsic property of the accelerator and can be measured regularly to ensure that the accelerator is operating correctly to treat patients. Other defined depth dose curves and / or quantities can also be used.

[0019] Dose data can also be provided as dose cross profiles (CP). CP provides the dose distribution of a medical radiation device perpendicular to the direction of the beam. For a given (photon) beam at a given source-to-surface distance, CP is obtained by scanning the dose perpendicular to the beam at various depths (e.g. 5 cm, 10 cm and 20 cm). CP also depends on the field size and, like DDC and OF, can be considered an intrinsic property of the accelerator and can be measured regularly to ensure that the accelerator is operating correctly to treat patients. Other defined cross profiles or quantities can also be used.

[0020] The radiation beam of a medical radiation device is usually characterized by a number of OFs, DDCs and CPs. These data provide information to properly tune the beam and configure the dose calculation algorithm to represent the specific medical radiation device.

[0021] Additionally or alternatively, dose data may be provided in absolute units such as Gray (Gy). Gray (Gy) is a unit of absorbed dose that quantifies the amount of radiation energy absorbed by a particular material such as a patient's body tissue. It is a measure of the actual deposited energy per unit mass of material.

[0022] Training dose data can be obtained in order to have information about the radiation process of a medical radiation device that is not affected by faults. Therefore, the training dose data represents a reference truth value to be compared with the dose data of the medical device to be monitored. Additionally or alternatively, the training dose data can exclude only predefined faults, such as a fault of a measuring amplifier. Such training dose data can provide a reference truth value relative to the predefined faults. The training dose data from a reference device can include the device itself or also include other devices that are important for the analysis. Such devices can be one or more measuring devices.

[0023] Training dose data may be obtained for different field sizes (eg, from a size of 1 cm x 1 cm to a field size of 50 cm x 50 cm). Additionally or alternatively, training dose data may be obtained for different intensities and / or depths (eg, in a water tank).

[0024] In order to capture all aspects of the radiation process of a medical radiation device, a machine learning algorithm is trained with training dose data. The output of the learning process is then compared with the output of the medical radiation device to be monitored. Or, in other words, the machine learning algorithm checks whether the output of the medical radiation device may also come from the flawless reference device being learned.

[0025] The output of the medical radiation device to be monitored is referred to as "operational dose data". Operational dose data can be obtained by measuring the output of the medical radiation device during actual use and / or test use. For example, operational dose data can be obtained under various parameters such as different intensities, different field sizes, different hardware components (e.g., different amplifiers), different collimator sizes, etc. In particular, operational data can be used in conjunction with certified hardware to rule out certain hardware failures, such as failures of beam control circuits and / or amplifiers attached to the medical dosimetry equipment.

[0026] In a subsequent step, the obtained operational dose data is provided to a trained machine learning algorithm. The trained machine learning algorithm compares the operational dose data with a reference truth value or with relative reference truth value information used to train the machine learning algorithm. This comparison can, for example, produce a value representing the probability that the medical radiation device to be monitored includes a fault. This will be explained in detail below.

[0027] According to the method of the first aspect, it is possible to effectively identify defective or at least unnecessarily noisy dose data and / or defective or unnecessarily noisy medical radiation equipment.

[0028] Embodiments of the first aspect relate to a method for monitoring dose output of a medical radiation device, wherein training dose data comprises measured dose data from one or more real medical radiation devices.

[0029] Obtaining training dose data from one or more reference devices may include obtaining training dose data from a manufacturer. Additionally or alternatively, the training dose data may come from one or more separate reference devices. In particular, the reference devices have been inspected and tested to operate without any faults and / or under the required environmental conditions, thereby providing flawless dose data. Such dose data excludes machine faults, human faults of machine operators and / or faults of measuring detectors and equipment. Error-free training dose data may be particularly useful for unsupervised learning.

[0030] Additionally or alternatively, the training dose data may also be obtained from multiple devices of the same type. This can make the training dose data more robust to variations within a group of medical radiation devices of the same type. Additionally or alternatively, the training dose data may also be obtained from multiple devices of different types. This can make the training dose data more robust to variations between different types of medical radiation devices. Additionally or alternatively, the training dose data may also be obtained for different configurations. This can make the training more robust to variations regarding the configuration (e.g., radiation intensity, radiation field size, attached hardware components) or settings of the dose measurement equipment of the medical radiation device.

[0031] The training dose data may also include data from one or more faulty devices. Additionally or alternatively, the training dose data may also include one or more predefined operating errors. Additionally or alternatively, the training dose data may also include one or more predefined detector and / or measurement equipment errors. Training a machine algorithm with defective training dose data may be done in the form of supervised learning. Thus, defective training dose data may be marked as defective, in particular including predefined errors. Additionally or alternatively, direct feedback may also be provided so that the machine learning learns whether the input is error-free and / or defective.

[0032] The training data may also include data acquired with various types of detectors and / or other measurement equipment. Additionally or alternatively, the training dose data can be acquired in different measurement bodies. Additionally or alternatively, the training dose data can be acquired with different combinations of detectors, equipment and / or measurement bodies. Training a machine learning algorithm with variable acquired dose data can be accomplished in the form of supervised learning. Thus, the training dose data can be labeled with the type of detector, equipment or measurement body. Additionally or alternatively, direct feedback may also be provided so that the machine learning algorithm knows that the input is affected by the specific characteristics of the type of detector, equipment or measurement body.

[0033] Training of machine learning algorithms can include supervised learning, unsupervised learning, and / or reinforcement learning.

[0034] Based on training dose data from one or more real devices, the radiation process of a medical radiation device can be learned with sufficient accuracy to identify errors from dose measurement equipment and / or operating errors hidden in the dose data.

[0035] Embodiments of the first aspect relate to a method for monitoring dose output of a medical radiation device, wherein training dose data comprises simulated dose data from one or more virtual reference devices.

[0036] The simulated training dose data can come from a model of a medical radiation device or from a model of a group of medical radiation devices, for example, the model can be associated with predefined types of medical radiation devices. Error-free simulated training dose data can be used in particular for unsupervised learning of machine learning algorithms.

[0037] Additionally or alternatively, the training dose data can be corrupted, in particular with known fault patterns. For example, a defective measuring amplifier provides an output that decreases with depth. In this case, error-free and defective training data can be marked as such. As another example, a detector type can show a sensitivity that depends on the measurement depth or the quality of the measured radiation. This can also be simulated.

[0038] Based on the training dose data from one or more virtual devices (ie, models), the radiation process of a medical radiation device can be learned by focusing on certain parameters and / or identifying certain faults.

[0039] An embodiment of the first aspect relates to a method for monitoring the dose output of a medical radiation device, comprising the steps of:

[0040] - providing data of one or more real devices, in particular the device to be monitored, to the virtual reference device, in particular configuration data and / or input data.

[0041] Information from real devices can be related to field size, energy intensity, ambient temperature, humidity, etc. Configuring the model of the medical radiation device with information from real devices can improve the accuracy of the model.

[0042] Embodiments of the first aspect relate to a method for monitoring dose output of a medical radiation device, wherein operational dose data is obtained for one or more different configurations of one or more medical radiation devices.

[0043] Thus, faults can be identified with respect to different configurations. For example, a fault due to a radiation field that is too small relative to the detector area (no ideal point measurement) may be better identified based on a small radiation field configuration, and / or based on a comparison between a medical radiation device with a small radiation field configuration and the same medical radiation device with a large radiation field configuration.

[0044] An embodiment of the first aspect relates to a method for monitoring the dose output of a medical radiation device, comprising the steps of:

[0045] -Adapting the configuration of the medical radiation device to be monitored;

[0046] - measuring operational dose data by said monitored medical radiation device; and - providing said generated operational data to said trained MLA.

[0047] For example, if certain operational dose data from a medical radiation device to be monitored is identified by a machine learning algorithm as being defective, the configuration (e.g., hardware configuration) can be changed and a second instance of the operational dose data provided to the machine learning algorithm to check whether the dose data is still evaluated as defective in the same way. If it is still defective, the defect is not caused by the changed parameters. In this way, the medical radiation device to be monitored can be adapted until the machine learning algorithm considers it to be error-free.

[0048] In another example, if the operational dose data is identified as error-free, modifications can be made to the configuration of the medical radiation device providing the operational dose data to check whether the medical radiation device is still considered error-free. Thus, alternative configurations of the medical radiation device can be analyzed for flawlessness.

[0049] Embodiments of the first aspect relate to a method for monitoring a dose output of a medical radiation device, wherein operational dose data are provided to the MLA until a predefined quantity threshold of operational dose data and / or a predefined accuracy threshold of the output information is reached.

[0050] Based on thresholds, machine learning algorithms can provide fast results with very short computation time. Based on this result, a more detailed analysis can be performed. This detailed analysis can be performed using different machine learning algorithms. For example, a detailed analysis can be performed to confirm predefined faults, particularly based on a machine learning algorithm that has been trained to identify predefined faults.

[0051] An embodiment of the first aspect relates to a method for monitoring the dose output of a medical radiation device, wherein the machine learning algorithm comprises a plurality of different machine learning algorithms trained with the training dose data, and the operational dose data is provided to the different machine learning algorithms.

[0052] In order to efficiently validate multiple fault hypotheses, the operational dose data can be provided to multiple machine learning algorithms that are trained to identify different faults. For example, a first neural network can be configured to identify a fault in an amplifier used in a medical radiation device. A second deep learning neural network can be configured to identify a fault associated with the field size used relative to the detector size. A third neural network can be configured to identify predefined operational errors caused by a human operator.

[0053] In particular, multiple differently configured machine learning algorithms can be used based on a previous threshold-based analysis that identified the provided operational dose data as defective. In another embodiment, the operational dose data of a medical radiation device can be directly provided to multiple differently trained machine learning algorithms to check the dose data for different faults without performing a previous general analysis. The operational dose data (or data based on the operational dose data) can be provided to different machine learning algorithms at least partially simultaneously to reduce monitoring time.

[0054] Embodiments of the first aspect relate to a method for monitoring the dose output of a medical radiation device, wherein the MLA comprises a neural network, in particular a fully connected neural network and / or a convolutional neural network.

[0055] According to a first aspect, the neural network used within the machine learning algorithm can be a neural network with multiple hidden layers, also called a deep neural network. In particular, the deep neural network can be trained by unsupervised learning. Thus, the network can learn the characteristics of error-free and / or defective dose data by itself.

[0056] Based on neural networks, especially deep neural networks, monitoring of medical radiation equipment can be performed in an effective and robust manner.

[0057] Embodiments of the first aspect relate to a method for monitoring the dose output of a medical radiation device, wherein the MLA comprises a classifier and / or a regressor.

[0058] Once the classifier has learned from the training data, it can be used to classify operational dose data. In particular, a classification can be performed between 'error-free' type and 'defective' type of dose data / medical radiation device. Additionally or alternatively, a classification can be performed in defective dose data regarding the type of defect. Additionally or alternatively, the classifier can be used for decisions related to the processing of the operational dose data. For example, if the dose data is defective, the classifier can decide to provide the operational dose data, or at least data based on the operational dose data, to one or more additional machine learning algorithms, in particular one or more machine learning algorithms, which are fine-tuned based on the provided data and / or dose errors introduced by a certain type of detector to identify predefined defects and / or predefined operational errors of the medical radiation device.

[0059] Additionally or alternatively, the regressor can predict a continuous variable, such as the probability or number of defects identified based on the analyzed operational dose data. In conjunction with predefined thresholds, the regressor can be used in a decision-making process, such as described above. In one embodiment, the regressor can predict the maintenance time of the monitored medical radiation equipment.

[0060] Embodiments of the first aspect relate to a method for monitoring dose output of a medical radiation device, wherein a response from the MLA comprises a classification associated with one or more predetermined fault types of the medical device to be monitored.

[0061] Based on the classifier, the machine learning algorithm can provide a classification statement about the provided operational dose data. For example, the classification statement can be: whether the operational dose data comes from an error-free device, and / or whether it includes operational errors.

[0062] An embodiment of the first aspect relates to a method for monitoring the dose output of a medical radiation device, wherein the output information includes at least one of the following:

[0063] - the probability that the operational dose data is defective and / or error-free;

[0064] - the probability that the operational dose data are from the reference device and / or from the medical radiation device to be monitored; and

[0065] -Maintenance time of the medical radiation equipment to be monitored.

[0066] The probability that the operational dose data is defective includes: the probability that the operational dose data includes predefined errors, such as predefined machine-related errors and / or predefined operator-related errors. Such analysis can be provided by a regressor as part of a machine learning algorithm.

[0067] Embodiments of the first aspect relate to a method for monitoring a dose output of a medical radiation device, measuring said operating data from said medical device to be monitored and providing said operating data to said MLA being performed at least partially simultaneously.

[0068] By performing the measuring process and the analyzing process at least partially simultaneously, the medical radiation device can be monitored during operation and / or in soft real time or even hard real time. This allows errors to be detected in the radiation procedure and thus the patient can be protected from malfunctions occurring during radiation therapy.

[0069] A second aspect of the present disclosure relates to a computer-implemented method for training a machine learning algorithm for monitoring dose output of a medical radiation device,

[0070] is configured as:

[0071] - Obtain training dose data from a reference device;

[0072] - training a machine learning algorithm (MLA) with the training dose data to teach the MLA to recognize the reference device based on dose data from the reference device that is different from the training dose data; and

[0073] - providing said trained MLA to a device for monitoring dose output.

[0074] If the machine learning algorithm comprises a plurality of algorithms, such as a plurality of deep neural networks, training can be performed on each of the machine learning algorithms simultaneously or separately. In particular, if different machine learning algorithms are associated with identifying different faults, each machine learning algorithm can be trained separately. The training method can include all features necessary for providing a trained machine learning algorithm according to the method of the first aspect.

[0075] A third aspect of the present disclosure relates to a device for monitoring a medical radiation device, configured to:

[0076] - Obtaining operational dose data from the medical radiation equipment to be monitored;

[0077] - providing the operational dose data to a trained machine learning algorithm (MLA) configured to output output information related to a difference between a reference device and the medical radiation device based on the operational dose data; and

[0078] - providing said output information to a user.

[0079] The device according to the third aspect may include one or more functions described with respect to the first aspect. In particular, the machine learning algorithm may include multiple machine learning algorithms, in particular deep neural networks, wherein a single machine learning algorithm is trained to generally identify faults of the medical radiation device, and / or multiple machine learning algorithms are trained to identify certain predefined errors.

[0080] The device according to the third aspect may in particular be a software-implemented device. Additionally or alternatively, the device according to the third aspect may include an application user interface (API) configured to obtain information and monitor a specific medical radiation device. Additionally or alternatively, the device may be configured to evaluate the API of the medical radiation device. Thus, multiple medical radiation devices may be monitored.

[0081] A fourth aspect of the present disclosure relates to medical radiation equipment,

[0082] is configured as:

[0083] - performing a method according to any of the preceding claims; and

[0084] - interacting with an application user interface of a device executing a method according to one of the preceding claims and / or with an application user interface of a device according to one of the preceding claims.

[0085] This aspect is related to the previous aspect, and embodiments may include one or more of the previous features. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] Further advantages and features are derived from the following embodiments, some of which are referred to in the accompanying drawings. The drawings do not always show the embodiments to scale. The dimensions of the various features may be increased or reduced accordingly, in particular for the sake of clarity of description. For this purpose, the drawings partially schematically show:

[0087] Figure 1 A method for monitoring a medical radiation device is shown.

[0088] Figure 2 A block diagram of an apparatus for monitoring a medical radiation device is shown.

[0089] Figure 3 A block diagram of an apparatus for monitoring a medical radiation device is shown.

[0090] Figure 4 A block diagram of an apparatus for monitoring a medical radiation device is shown.

[0091] Figure 5a and Figure 5b A monitoring diagram for medical radiation equipment is shown.

[0092] Figure 6a and Figure 6b A monitoring diagram for medical radiation equipment is shown.

[0093] Figure 7 The workflow of a system for monitoring medical radiation equipment is shown.

[0094] Figure 8 The principle of radiation dose measurement is shown.

[0095] Fig. 9 Shown is a medical radiation device with measurement and IT equipment.

[0096] In the following description, reference is made to the accompanying drawings which form a part of the present disclosure and which illustrate specific aspects through which the present disclosure can be understood. Like reference numerals refer to features which are the same or at least functionally or structurally similar.

[0097] Typically, the disclosure of the described method also applies to a corresponding device (or apparatus) for performing the method or a corresponding system including one or more devices, and vice versa. For example, if a specific method step is described, the corresponding device may include features for performing the described method step, even if the feature is not explicitly described or represented in the drawings. On the other hand, if, for example, a specific device is described based on a functional unit, the corresponding method may include one or more steps for performing the described function, even if these steps are not explicitly described or represented in the drawings. Similarly, a system may be provided with corresponding device features or have features for performing specific method steps. Unless otherwise explicitly stated, the features of the various exemplary aspects and embodiments described above or below may be combined. DETAILED DESCRIPTION

[0098] Figure 1 A method 100 for monitoring a medical radiation device according to an embodiment of the first aspect of the present disclosure is shown.

[0099] Dose measurements that characterize medical linear accelerator beams are often imperfect, which can lead to inaccurate dose calculations, for example, in patient cancer treatments. This has been identified as the leading cause of suboptimal patient treatment.

[0100] Although there are numerous guidelines for obtaining such dose data, there is no method to ensure their quality. Sanity checks on other data may also be flawed, or even if they are not, they may be imprecise because of high data variability. Vendors also disseminate “golden” standardized beam current data, which have also been shown to be affected by measurement error. Statistical descriptions of measured dose data have been published, but these are too crude to be effective in evaluating operational dose data from medical radiation equipment.

[0101] In a first step 110, training dose data from one or more real or virtual reference devices are acquired. These training dose data are used to train a machine learning algorithm. Thus, the machine learning algorithm is taught to recognize reference devices based on error-free input dose data and to distinguish reference devices from non-error-free devices, i.e., devices that provide defective data (e.g., the data is flawed and / or the hardware and / or software of the device is defective). This is achieved by learning features, in particular deterministic and / or non-deterministic features, of an error-free medical radiation device based on its ideal error-free dose data. Furthermore, the machine learning algorithm can learn deterministic features and / or non-deterministic features of a defective medical radiation device.

[0102] In a second step 120, operational dose data of the medical radiation device (106) to be monitored is acquired. The operational dose data is acquired during a test procedure and / or actual operation.

[0103] In a third step 130, the obtained operational dose data is provided to a trained machine learning algorithm to analyze whether it can distinguish between the medical radiation device and the reference device. If the machine learning algorithm cannot distinguish the medical device to be monitored from the reference device, the operational dose data is considered to be error-free. Otherwise, it is considered to be defective.

[0104] In a fourth step 140, the analysis results are provided as output information for further processing. Alternatively, the results may be provided to a user in order to decide whether the medical radiation device can be used. Additionally or alternatively, the output may also provide corrected dose data and / or corrected configuration data, which may be used as input information for the medical radiation device to be monitored and / or similar devices. In particular, dose data correction and / or configuration data correction may be used in a real-time hardware-in-the-loop program.

[0105] Based on the described method, flawed dose measurements and / or defective medical radiation equipment can be identified and their errors can be qualitatively, quantitatively and corrected. In particular, errors in dose measurements can be eliminated if the data will be fed into dose calculation software that creates beam parameters for patient treatment. Since the method can be operated in real time, measurements can be verified while they are being performed, making the measurement process more efficient. In addition, the operator can be prompted with possible sources of error.

[0106] Figure 2 A block diagram structure 200 of a device according to the present disclosure is shown. Input data x_R is provided to a reference device 202. The input data x_R may be, for example, configuration data. The reference device 202 may be a real reference device or a virtual reference device. The virtual reference device may also include features of a real (reference) device. The reference device 202 may also include features of a set of real medical radiation devices. The reference device 202 is an error-free reference device. The reference device 202 generates an output y_tr,R. This data is used as training data to train a machine learning algorithm 204 (e.g., a neural network) to identify features of an error-free reference device. The machine learning algorithm may also receive training data from defective real and / or virtual reference devices to improve its ability to distinguish between error-free and defective medical radiation devices. The machine learning algorithm also receives operational dose data y_D from the medical radiation device 206. This information is analyzed based on the learned flawless reference device and / or flawless operational data. The machine learning algorithm provides an output p, including information on whether the medical radiation device to be monitored is error-free or defective. Additionally, the output may include information on the probability of the monitored medical radiation device being defective or error-free.

[0107] Figure 3 A block diagram 300 is shown, which is Figure 2 The system depicted in block diagram 300 is the same as the system of block diagram 200. In addition, input data x_D of the medical radiation device to be monitored is also input into the reference device 202. In this case, training data y_tr,R can be generated for the machine learning algorithm 204, which includes real information about the medical device to be monitored. These input data x_D can be specific configuration information (such as intensity, field size, hardware components, measurement equipment, etc.) or environmental information (such as device temperature).

[0108] Figure 4 A block diagram 400 according to another embodiment of the present disclosure is shown. The system of the block diagram 400 is based on Figure 2The system of the block diagram 200 shown. In addition to the neural network 204, the machine learning algorithm also includes three error-sensitive neural networks 402a, 402b, and 402c. The functions of the additional network layers are as follows. These error-sensitive neural networks 402a, 402b, and 402c are trained using training data y_tr,def1, y_tr,def2, and y_tr,def3, respectively, to identify predefined errors of medical devices. Therefore, dose data including corresponding errors are used as training data for each network. Defective dose data can be provided by one or more real medical radiation devices and / or models (i.e., virtual medical devices). If the neural network 204 identifies the operating dose data as defective, the operating dose data y_D (or data based on the operating dose data) will be provided to the trained error-sensitive neural networks 402a, 402b, and 402c at the same time. Then, each of these neural networks will analyze the operating dose data y_D according to its trained defects and provide classifications p_1, p_2, and p_3, indicating whether the operating dose data includes corresponding errors. Additionally or alternatively, p_1, p_2, p_3 may also comprise the probability that the operating dose data y_D comprise a corresponding error.

[0109] Figure 5a and 5b Information generated when a medical radiation device is monitored using a method and system of the present disclosure is shown. In the two figures 500 and 510, the x-axis represents the radiation penetration depth (ranging from 0 to 400 mm), the left y-axis represents the dose output, and the right y-axis represents the relative error between the virtual output and the measured operating depth dose. The solid lines 502 and 512 represent the virtual dose output generated using the error-free beam model (only as a direction reference). The dotted lines 504 and 514 represent the customer measured operating dose data of the medical radiation device to be monitored. The beam model is a general beam model and is adjusted for the output factors measured for a single medical radiation device. The depth dose curve is provided for a source-to-surface distance of 950 mm.

[0110] exist Figure 5a In the graph 500 , it can be seen that the predicted dose data and the measured operational dose data are almost consistent, with only a small prediction error 506 being different. Figure 5b The predicted error 516 in the diagram 510 of FIG. 5 is similar. In particular, the information included in the predicted error is used by a machine learning algorithm (not shown) to analyze whether the provided operational data 502 is error-free or defective. In this case, the fault of greater depth is caused by a fault in the instrumentation amplifier.

[0111] Figure 6a and 6bInformation generated when a medical radiation device is monitored using a method and system of the present disclosure is shown. In graphs 600 and 610, the x-axis represents the radiation penetration depth (ranging from 0 to 250 mm), the left y-axis represents the dose intensity, and the right y-axis represents the relative error between the virtual output and the measured operational dose data. The solid lines 602, 612 represent the virtual dose output generated using the error-free beam model. The dashed lines 604, 614 represent the customer measured operational dose data of the medical radiation device to be monitored. The beam model is a general beam model and is adjusted for the output factors measured for a single medical radiation device. The dose curve is provided for a source-to-surface distance of 450 mm.

[0112] exist Figure 6a In the graph 600 of FIG. 600 , it can be seen that the predicted dose data and the measured operational dose data are almost identical, differing only by a small prediction error 606. However, the prediction error 616 in graph 600 is even smaller. The difference in error is due to the size of the radiation field, which is different from Figure 6b Compared with the configuration of Figure 6a In the configuration of Figure 6b In the configuration, the detector is smaller than the field size. By providing both measurements to the trained neural network, the neural network can understand that both measurements are defects due to the non-point-like detector size. This error is more severe for smaller radiation fields because the ratio of detector size to radiation field size is closer to 1. Therefore, in order to facilitate fault detection, operational data with different field sizes can be provided to the trained machine learning algorithm and / or used to train the machine learning algorithm.

[0113] Figure 7 The workflows of different systems for monitoring medical radiation equipment are shown. Alternatively, these workflows can also be implemented by a single system. The first workflow / system 700a describes the measurement process in a water tank. A linear accelerator 702 emits radiation 704 into a water tank 706 that includes a measurement device. Subsequently, the measurement data 708 is provided to a device for monitoring the dose output.

[0114] In the device, information 708 is analyzed by a first machine learning algorithm 710 (MLA1). MLA1 is trained and configured to analyze whether the provided measurement data 708 is defective 720 or whether it is error-free 730. In the first case, if MLA1 identifies that the measurement data 708 includes a fault, the measurement data and / or other data generated by MLA1 will be passed to a second machine learning algorithm 722 (MLA2).

[0115] MLA2 is trained to identify, based on the acquired data, whether the fault in the measurement data is caused by the operator 724 or the equipment 726. In the second case, if MLA1 identifies that the measurement data 708 is not faulty, then the measurement data and / or other data from MLA1 will be passed to a third machine learning algorithm 732 (MLA3).

[0116] MLA3 is trained to recognize whether the beam used for measurement 706 is tuned correctly 734. In addition, MLA3 is trained to recognize whether the type of beam used for measurement 706 needs to be adjusted 736. In addition, MLA3 is trained to recognize whether there is a fault in the linear accelerator used 738. For this three-way analysis, MLA3 can also include other machine learning algorithms that have been trained differently. The use of specialized machine learning algorithms can speed up the generation of results 724, 726, 734, 736, 738 and increase the accuracy of different result categories. This analysis of medical irradiators can be used to investigate whether the beam characteristics and operator inputs of the linear accelerator are within specifications and to propose remedial measures for measurement errors and beam tuning deviations.

[0117] The second workflow / system 700b is similar to the first system, i.e., data 708 measured about the linear accelerator 702 is analyzed by a first machine learning algorithm 710 to determine whether the data includes a fault 720. If the data includes a fault 720, a second machine learning algorithm 722 is used to determine whether the fault is due to an operator 724 or a device 726. However, if the data does not include a fault 730, the data and / or other information provided by the first machine learning algorithm 710 is provided to a fourth machine learning algorithm 740 (MLA4).

[0118] The MLA 4 is trained and configured to analyze whether the linear accelerator 702 used is not faulty or faulty 742 even if the measurement data is error-free. In addition, the MLA 4 is trained and configured to analyze whether the provided measurement data 708 contains information that requires preemptive maintenance 744. In addition, the MLA 4 is also configured to identify the cause of the defect that caused the linear accelerator failure 746. In summary, by arrangement 700b, it is possible to verify by measurement whether the beam characteristics of a medical radiation device meet specifications.

[0119] The third workflow / system 700c shows another embodiment based on the same measurement workflow 702-708 and a first machine learning algorithm 710 for identifying whether the measurement data includes errors 720 or is error-free 730. In the first case 720, the measurement data and / or other information of MLA1 is provided to two different machine learning algorithms in parallel.

[0120] MLA2 is used to identify whether the provided data 708, 720 is based on operator error 724 or device error 726, and is used in the same manner as the workflow / system 700a, 700b. In addition, the data 708, 720 are input into the fifth machine learning algorithm 750 (MLA5). MLA5 is trained and configured to provide corrected data 751, that is, it is able to synthesize new data that does not contain the identified errors. Therefore, even in the presence of errors, the correct input / measurement data can be identified. In the second case, that is, if the measurement data is error-free 730, the data is approved. Synthetic data can be created by using a virtual reference device, which may include a beam model adapted to a specific device, or may include a general beam model and configure the virtual reference device to multiple settings, such as different field sizes, measurement settings or additional accessories. Synthetic data can also be generated for different measurement volumes and arbitrary positions and orientations that may not be accessible experimentally.

[0121] Figure 8 The principle of the radiation dose measurement 800 in the water tank is shown. After the radiation source in the treatment machine emits radiation, the irradiation field 801 presents a first shape. Due to the diffusion of the radiation, the shape of the irradiation field changes and when the radiation beam enters the water 830 in the water tank, it presents a second shape 810. In the water tank, the radiation beam further diffuses so that the radiation field presents a third shape 820 in the plane of measurement 860. The coordinate system indicates three lateral dimensions. The scanning direction of the depth dose curve is represented by dimension 840, while the scanning direction of the profile distribution is represented by dimension 850.

[0122] Fig. 9 Shown with Figure 2 , 3 or a medical radiation device 910 connected to the system described in claim 4. System 900 includes a medical radiation device 910 and a computer system 920. The medical radiation device 910 is configured to provide a radiation dose. The measuring device 930 is configured to observe the radiation dose and transmit the measurement information to the computer system 920. The computer system 920 is configured to perform at least a portion of the method according to the first aspect of the present disclosure. The computer system 920 may be configured to execute a machine learning algorithm and / or communicate with a machine learning algorithm. The computer 920 and the medical radiation device 910 may be independent entities, but may also be integrated in a common housing. The computer 920 may be part of a central processing system of the medical radiation device 910 and / or the computer 920 may be part of a subcomponent of the medical radiation device 910.

[0123] The computer system 920 can be a local computer device (e.g., a personal computer, laptop, tablet, or mobile phone) with one or more processors and one or more storage devices, or a distributed computer system (e.g., with one or more processors and one or more storage devices, distributed in various locations, such as a local client and / or one or more remote servers and / or data centers). The computer system 920 may include any circuit or combination of circuits. In one embodiment, the computer system 920 may include one or more processors, which may be of any type. As used herein, a processor may refer to any type of computing circuit, such as, but not limited to, a microprocessor, a microcontroller, a complex instruction set computing (CISC) microprocessor, a reduced instruction set computing (RISC) microprocessor, a very long instruction word (VLIW) microprocessor, a graphics processor, a digital signal processor (DSP), a multi-core processor, a field programmable gate array (FPGA), such as a microscope or a microscope component (e.g., a camera), or any other type of processor or processing circuit. Other types of circuits that may be included in the computer system 920 include custom circuits, application-specific integrated circuits (ASICs), and the like, such as one or more circuits (e.g., communication circuits) for wireless devices (e.g., mobile phones, tablets, laptops, two-way radios, and similar electronic systems). The computer system 920 may include one or more storage devices, which may include one or more storage elements suitable for a particular application, such as main memory in the form of random-access memory (RAM), one or more hard disk drives, and / or one or more drives for handling removable media (e.g., compact disks (CDs), flash memory cards, digital video disks (DVDs), etc.). The computer system 920 may also include a display device, one or more speakers, and a keyboard and / or controller, which may include a mouse, a trackball, a touch screen, a voice recognition device, or any other device that allows a system user to input information to and receive information from the computer system 920.

[0124] Some or all of the method steps may be performed by (or using) a hardware device (such as a processor, a microprocessor, a programmable computer or an electronic circuit). In some embodiments, some of the most important method steps may be performed by such a device.

[0125] Depending on specific implementation requirements, embodiments of the present disclosure may be implemented in the form of hardware or software. The implementation may be performed using a non-temporary storage medium (e.g., a digital storage medium), such as a floppy disk, DVD, Blu-ray disc, CD, ROM, PROM, EPROM, EEPROM, or FLASH memory, on which electronically readable control signals are stored that can work in conjunction with a programmable computer system (or can work in conjunction with it) to perform the corresponding method. Therefore, the digital storage medium may be computer readable.

[0126] According to the present disclosure, some embodiments comprise a data carrier having electronically readable control signals, which are capable of cooperating with a programmable computer system, such that one of the methods described herein is performed.

[0127] Typically, embodiments of the present disclosure may be implemented as a computer program product having a program code, which is used to perform a corresponding method when the computer program product is run on a computer. For example, the program code may be stored on a machine-readable carrier. Other embodiments include a computer program stored on a machine-readable carrier for performing one of the methods described herein. In other words, therefore, one embodiment of the present disclosure is a computer program having a program code, which is used to perform one of the methods described herein when the computer program is run on a computer.

[0128] Therefore, another embodiment of the present disclosure is a storage medium (or a data carrier, or a computer-readable medium) having stored thereon a computer program for performing one of the methods described herein when executed by a processor. The data carrier, the digital storage medium or the recorded medium is typically tangible and / or non-transitory. Another embodiment of the present disclosure is an apparatus comprising a processor and a storage medium as described herein.

[0129] Therefore, another embodiment of the disclosure is a data stream or a sequence of signals representing the computer program for performing one of the methods described herein.The data stream or the sequence of signals may, for example, be configured to be transmitted via a data communication connection (eg via the Internet).

[0130] Embodiments may be based on the use of machine learning models or machine learning algorithms. Machine learning may refer to algorithms and statistical models in which a computer system relies on models and reasoning to perform specific tasks without the use of explicit instructions. For example, in machine learning, data transformations may not be based on rules, but rather inferred through analysis of historical and / or training data. For example, a machine learning model or machine learning algorithm may be used to analyze the content of an image. In order to enable a machine learning model to analyze the content of an image, the machine learning model may be trained using training images as input and training content information as output. By training the machine learning model using a large number of training images and / or training sequences (e.g., words or sentences) and related training content information (e.g., labels or annotations), the machine learning model "learns" to recognize the content of the image, so the machine learning model can be used to recognize image content that is not included in the training data. The same principle can also be used for other types of sensor data: by training the machine learning model using training sensor data and expected outputs, the machine learning model "learns" the conversion between sensor data and output, which can be used to provide output based on non-training sensor data provided to the machine learning model. The provided data (e.g., sensor data, metadata, and / or image data) can be preprocessed to obtain a feature vector, which is used as an input to the machine learning model.

[0131] Machine learning models can be trained using training input data. The above examples use a training method called "supervised learning". In supervised learning, a machine learning model is trained using multiple training samples, each of which may include multiple input data values ​​and multiple expected output values, i.e., each training sample is associated with an expected output value. By specifying the training samples and the expected output values, the machine learning model "learns" to provide output values ​​based on input samples similar to the samples provided during training. In addition to supervised learning, semi-supervised learning can also be used. In semi-supervised learning, some of the training samples lack corresponding expected output values. Supervised learning may be based on supervised learning algorithms (such as classification algorithms, regression algorithms, or similarity learning algorithms). When the output is restricted to a finite set of values ​​(categorical variables), i.e., the input is classified as one of this finite set of values, a classification algorithm can be used. When the output may have any value (within a certain range), a regression algorithm can be used. Similarity learning algorithms may be similar to classification and regression algorithms, but are based on the principle of learning from examples using a similarity function, which is used to measure the similarity or correlation between two objects. In addition to supervised or semi-supervised learning, unsupervised learning can also be used to train machine learning models. In unsupervised learning, one may (only) provide input data and use an unsupervised learning algorithm to discover the structure of the input data (e.g., find commonalities in the data by grouping or clustering the input data). Clustering is the assignment of input data, including multiple input values, into subsets (clusters) such that input values ​​in the same cluster are similar according to one or more (predefined) similarity criteria, but different from input values ​​included in other clusters.

[0132] Reinforcement learning is the third group of machine learning algorithms. In other words, reinforcement learning can be used to train machine learning models. In reinforcement learning, one or more software actors (called "software agents") are trained to take actions in an environment. Based on the actions taken, rewards are calculated. Reinforcement learning is based on training one or more software agents to select actions that increase the cumulative reward, so that the software agent becomes better at a given task (as evidenced by increasing rewards).

[0133] In addition, some techniques can be applied to some machine learning algorithms. For example, feature learning can be used. In other words, a machine learning model can be trained at least in part using feature learning, and / or a machine learning algorithm can include a feature learning component. Feature learning algorithms, also known as representation learning algorithms, not only retain input information, but also transform it in a way that makes it more useful, usually as a preprocessing step before performing classification or prediction. For example, feature learning may be based on principal component analysis or cluster analysis.

[0134] In some examples, anomaly detection (i.e., outlier detection) may be used, the goal of which is to identify suspicious input values ​​that are significantly different from the majority of inputs or training data. In other words, the machine learning model may be trained at least in part using anomaly detection, and / or the machine learning algorithm may include an anomaly detection component.

[0135] In some examples, a machine learning algorithm may use a decision tree as a prediction model. In other words, a machine learning model may be based on a decision tree. In a decision tree, observations about an item (such as a set of input values) may be represented by branches of the decision tree, and output values ​​corresponding to that item may be represented by leaves of the decision tree. Decision trees can support both discrete and continuous values ​​as output values. If discrete values ​​are used, the decision tree may be called a classification tree; if continuous values ​​are used, the decision tree may be called a regression tree.

[0136] Association rules are another technique that can be used in machine learning algorithms. In other words, a machine learning model may be based on one or more association rules. Association rules are created by identifying relationships between variables in a large amount of data. Machine learning algorithms can identify and / or utilize one or more relationship rules that represent knowledge learned from the data. These rules can be used to store, manipulate, or apply knowledge.

[0137] Machine learning algorithms are often based on machine learning models. In other words, the term "machine learning algorithm" may refer to a set of instructions for creating, training, or using a machine learning model. The term "machine learning model" may refer to a data structure and / or a set of rules that represent learned knowledge (e.g., based on training performed by a machine learning algorithm). In some embodiments, using a machine learning algorithm may mean using one or more underlying machine learning models. Using a machine learning model may mean that the machine learning model and / or the data structure / rule set that is the machine learning model is trained by a machine learning algorithm.

[0138] For example, a machine learning model can be an artificial neural network (ANN). ANNs are systems inspired by biological neural networks, such as those in the retina or brain. ANNs consist of a number of interconnected nodes and a large number of connections between the nodes, the so-called edges. There are usually three types of nodes: input nodes that receive input values, hidden nodes that are only connected to other nodes, and output nodes that provide output values. Each node can represent an artificial neuron. Each edge can transfer information from one node to another. The output of a node can be defined as a (non-linear) function of the input (e.g., the sum of the inputs). The input of a node can be used in a function depending on the "weight" of the edge or node that provides the input. During the learning process, the weights of the nodes and / or edges can be adjusted. In other words, the training of an artificial neural network may include adjusting the weights of the nodes and / or edges of the artificial neural network, i.e., to achieve the desired output for a given input.

[0139] Alternatively, the machine learning model can be a support vector machine, a random forest model, or a gradient boosting model. A support vector machine (i.e., a support vector network) is a supervised learning model with an associated learning algorithm that can be used to analyze data (e.g., in classification or regression analysis). A support vector machine can be trained by providing an input with multiple training input values ​​belonging to one of two categories. A support vector machine can be trained to assign new input values ​​to one of two categories. Alternatively, the machine learning model can be a Bayesian network, which is a probabilistic directed acyclic graph model. A Bayesian network can use a directed acyclic graph to represent a set of random variables and their conditional dependencies. Alternatively, the machine learning model can be based on a genetic algorithm, which is a search algorithm and heuristic technique that mimics the process of natural selection.

[0140] Reference numerals list

[0141] 100 Methods for monitoring medical radiation equipment

[0142] 110 First Step

[0143] 120 Second Step

[0144] 130 The Third Step

[0145] 140 Step 4

[0146] 200 Block Diagram

[0147] 202 Reference Equipment

[0148] 204 Neural Network

[0149] 206 Medical radiation equipment

[0150] 300 Block Diagram

[0151] 400 Block Diagram

[0152] 402a Error Sensitive Neural Network

[0153] 402b Error Sensitive Neural Network

[0154] 402c Error Sensitive Neural Network

[0155] 500 verification images

[0156] 502 Virtual dose output

[0157] 504 Operational dose data

[0158] 506 prediction error

[0159] 510 verification diagram

[0160] 512 virtual dose output

[0161] 514 Operational dose data

[0162] 516 prediction error

[0163] 600 images

[0164] 602 Error-free virtual dose output

[0165] 604 Operational dose data

[0166] 606 prediction error

[0167] 610 images

[0168] 612 Error-free virtual dose output

[0169] 614 Operational dose data

[0170] 616 prediction error

[0171] 700a First workflow or system

[0172] 700b Secondary Workflow or System

[0173] 700c Third Workflow or System

[0174] 702 Linear Accelerator

[0175] 704 irradiation

[0176] 706 water tank

[0177] 708 measurement data

[0178] 710MLA1

[0179] 720 Flawed Measurement Data

[0180] 722 MLA2

[0181] 724 Operator Error

[0182] 726 Equipment Error

[0183] 730 trouble-free measurement data

[0184] 732 MLA3

[0185] 734 Beam tuning correct

[0186] 736 Linear Accelerator Failure

[0187] 740 MLA4

[0188] 742 Linear Accelerator OK

[0189] 744 Preemptive maintenance type

[0190] 746 Linear Accelerator Defect Causes

[0191] 750 MLA5

[0192] 751 Approved Measurement Data

[0193] 800 Radiation Dose Measurement

[0194] 801 Irradiation Field First Form

[0195] 810 Irradiation Field Second Form

[0196] 820 Underwater irradiation field

[0197] 830 Water Surface

[0198] 840 Scanning direction of depth dose curve

[0199] 850 Scanning direction of profile distribution

[0200] 860 Measurement point of output factor

[0201] 900 Medical Radiation Systems

[0202] 910 Medical radiation equipment

[0203] 920 Computer

[0204] 930 Measuring equipment

Claims

1. A method for monitoring the dose output of a medical radiation device, comprising the steps of: - obtaining training dose data (y_tr, R) from a reference device (202); - training a machine learning algorithm MLA (204) with the training dose data (y_tr, R) to teach the MLA to recognize the reference device based on dose data (y_R) from the reference device that is different from the training dose data (y_tr, R); - obtaining operational dose data (y_D) from the medical radiation device (206) to be monitored; - providing the operational dose data (y_D) to the trained MLA; - providing output information (p) related to differences between the reference device and the medical radiation device based on the MLA.

2. The method according to the preceding claim, in, The training dose data (y_tr, R) includes measured dose data (y_R1, y_R2) from one or more real medical radiation devices.

3. The method according to claim 1, in, The training dose data (y_tr, R) includes simulated dose data (y_tr, Rv) from one or more virtual reference devices.

4. The method according to the preceding claim, comprising the steps of: - Providing data of one or more real devices, in particular the device to be monitored, to the virtual reference device, in particular configuration data (p_R1, p_R2, P_D) and / or input data (x_R1, y_R2, X_D).

5. The method according to any one of the preceding claims, in, Operational dose data (y_d) is obtained for one or more different configurations of one or more medical radiation devices.

6. The method according to any one of the preceding claims, comprising the steps of: - adapting the configuration of the medical radiation device (206) to be monitored; - measuring operational dose data (y_D,2) by the monitored medical radiation device; and -Provide the generated operation data (y_D, 2) to the trained MLA.

7. The method according to any one of the preceding claims, in, Operational dose data (y_d) are provided to the MLA until a predefined quantity threshold of operational dose data and / or a predefined accuracy threshold (a) of the output information (p) is reached.

8. The method according to any one of the preceding claims, in, The machine learning algorithm includes a plurality of different machine learning algorithms trained with the training dose data (y_tr, R), and the operational dose data (y_d) is provided to the different machine learning algorithms.

9. The method according to any one of the preceding claims, in, The MLA comprises a neural network, in particular a fully connected neural network and / or a convolutional neural network.

10. The method according to any one of the preceding claims, in, The MLA includes a classifier and / or a regressor.

11. The method according to any one of the preceding claims, in, The response from the MLA includes a classification associated with one or more predetermined fault types of the medical device to be monitored.

12. The method according to any one of the preceding claims, in, The output information includes at least one of the following: - the probability (p) that the operational dose data (y_d) are defective and / or error-free; - the probability (p) that the operational dose data (y_d) are from the reference device (202) and / or from the medical radiation device to be monitored; and -Maintenance time of the medical radiation equipment to be monitored.

13. The method according to any one of the preceding claims, in, At least partially simultaneously performing: measuring operational data (y_d) from the medical device to be monitored; and providing the operational data to the MLA.

14. A device for training a machine learning algorithm (MLA) for monitoring the dose output of a medical radiation device, configured to: - obtaining training dose data (y_tr, R) from a reference device (202); - training a machine learning algorithm MLA (204) with the training dose data (y_tr, R) to teach the MLA to recognize the reference device based on dose data (y_R) from the reference device that is different from the training dose data (y_tr, R); and - providing said trained MLA to a device for monitoring dose output.

15. A device for monitoring medical radiation equipment, configured to: - obtaining operational dose data (y_D) from the medical radiation device (206) to be monitored; - providing the operational dose data (y_D) to a trained machine learning algorithm MLA (204), the MLA (204) being configured to: output output information (p) related to the difference between a reference device (202) and the medical radiation device (206) based on the operational dose data (y_D); and - providing said output information (p) to a user.

16. A medical radiation device, configured to: - performing a method according to any of the preceding claims; and - interacting with an application user interface of a device executing a method according to one of the preceding claims and / or with an application user interface of a device according to one of the preceding claims.

Images