Automatic collimation adjustment for dynamic x-ray imaging

Through AI-based models and optical image data, dynamically adjusting the collimated area of ​​X-ray imaging solves the challenge of collimation adjustment in fluoroscopy imaging, achieving equalization of radiation exposure and improving image quality, and reducing artificial errors.

CN119014892APending Publication Date: 2024-11-26SIEMENS HEALTHINEERS AG
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410652443.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2023-05-26
Filing Date
2024-05-24
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

The prior art is difficult to effectively adjust collimation in sequence X-ray images, especially during fluoroscopy imaging, resulting in uneven radiation exposure and image quality degradation, and artificial errors often lead to collimation failures.

Method used

Using AI-based model and optical image data, the adjustment collimation is generated by receiving standard data from the library, and automatically checks whether the X-ray frame meets the exact criteria, and dynamically adjusts the collimation area for automation and accuracy.

Benefits of technology

Automatic adjustment of collimation for dynamic X-ray imaging is achieved, reducing radiation exposure, improving image quality, reducing artificial errors, and ensuring the accuracy and robustness of collimated areas.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119014892A_ABST
    Figure CN119014892A_ABST
Patent Text Reader

Abstract

The invention relates to a method for automatically adjusting collimation for dynamic X-ray imaging, comprising the following steps: i) receiving standard data for adjusting collimation from a library, the standard data being assigned to a selected examination scheme and comprising a first standard and a second standard, ii) acquiring optical image data from the examination object using an optical image recording device, iii) generating an adjusted collimation based on the acquired optical image data and a first criterion, iv) acquiring an X-ray frame from the examination object using the adjusted collimation using a dynamic X-ray imaging device, v) generating an X-ray frame from the examination object using the adjusted collimation based on the acquired X-ray frame, performing an automatic check on the adjusted collimation of the acquired X-ray frame by checking whether the acquired X-ray frame satisfies a second criterion, vi) if the automatic check is not passed, generating a readjusted collimation such that the next acquired X-ray frame is most likely to satisfy the second criterion; or if the automatic check is passed, the adjusted collimation is maintained.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to a method for automatically adjusting collimation for dynamic X-ray imaging. In the method according to the present invention, a first standard of standard data for adjusting collimation is received from a library, wherein the first standard is assigned to a selected inspection scheme. In addition, optical image data is acquired from an inspection object using an optical image recording device. Adjusted collimation is generated based on the acquired optical image data and the first standard. In addition, the present invention relates to a method for generating a trained AI-based model to perform automatic inspection of the adjusted collimation of acquired X-ray frames. The present invention also relates to an adjustment device. In addition, the present invention also relates to a medical imaging system. Background Art

[0002] Automatic collimation is an essential part of acquiring X-ray images. A collimator is a device used to limit and narrow the beam, which defines the boundaries of the image radiated by the X-rays. Achieving the correct collimator setting is one of the key aspects of improving radiographic imaging techniques. The correct collimator setting prevents unnecessary exposure of anatomical structures outside the area of ​​interest and improves image quality by producing less scattered radiation from these areas.

[0003] However, in the image, some necessary parts of the organ that should be seen may not be visible. This is due to the following reasons: the collimator area is not wide enough to capture all necessary parts. In other scenarios, a very wide collimator area may also be produced, which causes unnecessary exposure, thus violating the ALARA principle.

[0004] ALARA is an abbreviation for "As Low As Reasonably Achievable" and refers to the principle of radiation protection. When dealing with ionizing radiation, the ALARA principle requires that the exposure of humans, animals and materials to radiation (even if below the limit value) be as low as practicable, taking into account practical reasons and weighing the pros and cons. The word "reasonable" is interpreted as "as low as possible" due to the risk of cancer. The precautionary principle, which is also associated with radiation, can be offset in this way.

[0005] Human error is the most common cause of initial alignment failure and identification of incorrect alignment.

[0006] One known technique for defining the boundaries of the collimator is to use an RGB camera together with a depth camera. In this case, the camera in the acquisition room will first capture an image from the patient, and then (by using an AI model) will identify several key points on the image that will then be used to define the boundaries of the collimator. Figure 1 shows the operating steps of this technique for the specific case where the desired body part is the chest.

[0007] It has to be considered that each body part has specific key points and requirements for correct alignment, therefore having several accurate models for detecting the correct key points for each body part is necessary. Furthermore, good human knowledge is also required to identify possible faults in case the output of the camera model is incorrect.

[0008] The mentioned methods have been developed and frequently used for static (radiographic) X-ray imaging requiring only one acquisition at a time. However, no known techniques have been developed for sequential X-ray images similar to images in fluoroscopy, so new approaches are needed to address the main challenges in fluoroscopy images.

[0009] Fluoroscopy is an examination method in which an X-ray camera is connected to a monitor, allowing the doctor to view the organ to be examined directly on the monitor. In contrast to static X-rays, fluoroscopy is usually performed by a doctor rather than a radiologist. Small mobile fluoroscopy units on casters are used by surgeons and surgical staff to visualize fractures or dislocations, and for examinations during surgical settings.

[0010] In addition, fluoroscopy is suitable for recording the representation of blood vessels, bile ducts and gastrointestinal tract sections with appropriate contrast agents. In addition, fluoroscopy is used to monitor the placement of the probe in the body under X-ray control and for better localization of pathological processes in the body (e.g., lung nodules) by rotating or changing the position of the patient. Fluoroscopy is also used to observe dynamic processes, for example, to exclude vesicoureteral reflux, to monitor cardiac motion, to detect valve calcifications, to monitor swallowing motion (esophageal display) and for excluding leaks (fistulas) after surgical interventions.

[0011] In order to minimize the X-ray burden on the patient, it is important to avoid errors in planning the region of interest and in the collimator size.

[0012] The general problem on which the present invention is based is therefore to improve the collimation adjustment for dynamic medical imaging processing involving sequential X-ray images, in particular for fluoroscopic imaging processing. Summary of the invention

[0013] The aforementioned problems are solved by: a method for automatically adjusting collimation for dynamic X-ray imaging according to claim 1, a method for generating a trained AI-based model to perform automatic checking of adjusted collimation of acquired X-ray frames according to claim 11, an adjustment device according to claim 12, and a medical imaging system according to claim 13.

[0014] As described at the beginning, in a method for automatically (preferably with the aid of AI) adjusting collimation for dynamic X-ray imaging (preferably fluoroscopy), standard data for adjusting collimation are received from a library. The criteria of the standard data are selected based on the selected examination scheme and are related to the required information to be reflected by the X-ray image and the above-mentioned ALARA principle. The library preferably includes a data base containing the above-mentioned standard data.

[0015] The method of the present invention can be applied to any dynamic X-ray imaging method that includes the continuous generation of frames of X-ray images. As an example, the method of the present invention can also be applied to data acquired by cone beam CT. A "frame" includes a single image of a dynamic picture sequence. It represents the basic unit of dynamic picture media, similar to a letter in writing. A frame in a static photo is called a still image.

[0016] The standard data comprises: a first standard, which is related to generating an adjusted collimation based on optical image data from the examination object; and a second standard, which is related to generating a readjusted collimation based on an X-ray frame from the examination object. Since the X-ray frame reveals the interior of the examination object, the second standard is usually more detailed and accurate than the first standard. Preferably, the first standard is related to the outer contour of the examination object or the outer contour of a part of the examination object. The second standard is preferably related to internal features, in particular internal landmarks, such as special types of bones or organs, which often enable a more accurate determination of the area to be included by the collimation. The examination object preferably comprises a patient, in particular a human or an animal.

[0017] In other words, the criterion is based on keeping the radiation exposure as low as possible, but depicting all body areas relevant for the specific examination related to the selected protocol. In addition, optical image data are acquired from the examination object, preferably a human patient or an animal patient, using an optical image recording device, preferably a camera device. In addition, an adjusted collimation is generated based on the acquired optical image data and based on the first criterion. Collimation adjustment means that the position and size of the area to be irradiated with X-rays are adjusted based on the acquired optical image data. The adjustment can be achieved by modifying the parameter values ​​of the adjusted collimation, in particular by modifying the position and / or height and / or width of the collimation area.

[0018] To start dynamic X-ray imaging including a series of acquisitions of X-ray frames, first, an X-ray frame is acquired from the examination object with the adjusted collimation using the dynamic X-ray imaging apparatus.

[0019] Compared to the prior art, after acquiring the X-ray frame, an automatic check of the adjusted collimation of the acquired X-ray frame is performed, wherein it is checked whether the acquired X-ray frame meets the second criterion. The acquired X-ray frame comprises a 2D image.

[0020] If the automatic check is not passed, a readjusted collimation is generated so that the next acquired X-ray frame is most likely to meet the second criterion of the standard data. If the automatic check is passed, the adjusted collimation remains unchanged. "Most likely" means that the content of the next acquired X-ray frame is predicted based on the previously acquired X-ray frame or the acquired X-ray frame sequence, and the collimation is adjusted so that the predicted X-ray frame meets the second criterion. In the simplest variant, it is assumed that the content of the next acquired X-ray frame is very similar to the last acquired X-ray frame, and the collimation is adjusted so that the last acquired X-ray frame, which is enlarged or reduced by the adjusted collimation, meets the second criterion. The information about the "enlarged" last acquired X-ray frame is preferably acquired based on the previously acquired X-ray frame (if available) or based on available anatomical knowledge or medical knowledge.

[0021] Preferably, after readjusting the collimation or adjusting the collimation, dynamic X-ray imaging is continued by repeating the following steps: starting dynamic X-ray imaging; performing automatic inspection; and readjusting the collimation based on the result of the automatic inspection, if a readjusted collimation is generated, using the readjusted collimation as the adjusted collimation, or if the automatic inspection is passed, using the adjusted collimation; until the dynamic X-ray imaging is completed.

[0022] Advantageously, the image chain may be automatically and dynamically adjusted based on the most recently recorded X-ray frames.Advantageously, a feedback loop is achieved by using acquired X-ray frames of the dynamic X-ray imaging for readjusting the collimation of the dynamic X-ray imaging.

[0023] In a method for generating a trained AI-based model to perform automatic inspection of adjusted collimation of acquired X-ray frames, labeled input data including input data and verification result data is generated. The labeled input data includes an X-ray frame of an inspection object as input data, and information related to a second criterion that satisfies the standard data as verification result data. The second criterion is related to generating a readjusted collimation based on an X-ray frame from the inspection object. As described in detail later, the information related to the second criterion that satisfies the standard data includes information for analyzing the acquired X-ray frames with respect to satisfying these criteria.

[0024] A possibility to create training data based on fluoroscopy sequences is to acquire X-ray frames from a 4D-CT volume based X-ray fluoroscopy simulation, where "4D" means 3D and time as a fourth dimension.

[0025] After this, the AI-based model to be trained is applied to the labeled input data, wherein result data is generated, and the training of the AI-based model is performed based on the result data generated by the AI-based model and the verification result data of the labeled input data. The training can be implemented using a cost function and a back-propagation algorithm to generate an adjusted AI-based model. Finally, the trained AI-based model is provided to a user to apply the trained AI-based model to a method for automatically (preferably, with the assistance of AI) adjusting collimation for dynamic X-ray imaging according to the present invention. Advantageously, the AI-based model can be adapted to a separate training database depending on the specific type of application.

[0026] The adjustment device according to the present invention comprises an input interface for receiving standard data for adjusting collimation from a library, for receiving optical image data from an examination object, and for repeatedly receiving X-ray frames from the examination object from an X-ray imaging device.

[0027] The standard data includes: a first standard related to generating an adjusted collimation based on optical image data from the examination object; and a second standard related to generating a readjusted collimation based on an X-ray frame from the examination object.

[0028] The adjustment device according to the present invention further comprises an adjustment unit for generating an adjusted collimation based on the acquired optical image data and the standard data.

[0029] The adjustment device according to the present invention further comprises an output interface, which is used to output the adjusted collimation to the X-ray imaging device to adjust the collimation of the X-ray imaging device.

[0030] The adjustment device according to the present invention further comprises a checking device for performing an automatic check by checking whether the acquired X-ray frame satisfies a second standard of the standard data.

[0031] The adjustment unit of the adjustment device according to the present invention is also configured to: if the automatic inspection fails, generate a readjusted collimation based on the result of the automatic inspection of the inspection unit so that the next acquired X-ray frame is most likely to meet the second criterion; and if the automatic inspection passes, maintain the adjusted collimation.

[0032] The output interface of the adjustment device according to the present invention is configured to output the readjusted collimation to the X-ray imaging device to readjust the collimation of the X-ray imaging device. The adjustment device shares the advantages of the method for automatically adjusting collimation for dynamic X-ray imaging according to the present invention.

[0033] The medical imaging system according to the present invention comprises: an X-ray imaging device, preferably a dynamic X-ray imaging device; a library comprising standard data for adjusting the collimation of the X-ray imaging device; an optical imaging device for recording optical image data from an object under examination; and an adjustment device according to the present invention for controlling the collimation of the X-ray imaging device. The dynamic X-ray imaging device records a series of multiple X-ray frames in sequence. The medical imaging system shares the advantages of the adjustment device according to the present invention.

[0034] Some units or modules of the above-mentioned adjustment device, in particular the adjustment unit, the inspection unit, and the input interface and the output interface, can be fully or partially implemented as a software module, which runs on a processor of a corresponding computing system, for example, on a processor of a control device of a medical imaging system. Realization in the form of software modules to a large extent can have the following advantages: applications already installed on existing medical imaging systems can be updated with relatively little effort to install and run these units of the present application. The object of the present invention is also achieved by a computer program product with a computer program or by a computer program, which can be directly loaded into a memory of a computing system, preferably a medical imaging system. The computer program includes a program unit, which is used to execute the steps of the inventive method for adjusting the collimation for dynamic X-ray imaging with the assistance of AI, in particular, the step of receiving standard data for adjusting the collimation from a library, the step of generating the adjusted collimation based on the acquired optical image data and based on the first standard of the standard data, the step of performing an automatic check on the adjusted collimation, and the step of generating the readjusted collimation. The computer program, when the program is executed by a computing system, further comprises steps of a method for generating a trained AI-based model to perform an automatic check of the adjusted collimation of the acquired X-ray frames, in particular, the steps of generating labeled input data, the steps of applying the AI-based model to be trained to the labeled input data, and the steps of training the AI-based model. In addition to the computer program, such a computer program product may also include other parts, such as documentation and / or additional components, and may also include hardware components, such as a hardware key (dongle, etc.) to facilitate access to the software.

[0035] Computer readable media such as memory sticks, hard disks or other removable or permanently installed carriers can be used to transport and / or store executable portions of a computer program product so that these executable portions can be read from a processor unit of a computing system. The processor unit may include one or more microprocessors or their equivalents.

[0036] The dependent claims and the following description all contain particularly advantageous embodiments and improvements of the invention. In particular, a claim in one claim category can also be further improved similarly to a dependent claim in another claim category. In addition, within the scope of the invention, various features of different exemplary embodiments and claims can also be combined to form new exemplary embodiments.

[0037] In a variant of the method for automatically adjusting collimation for dynamic X-ray imaging according to the invention, the automatic check comprises an explicit check. The explicit check comprises explicit positioning and identification of image features of the acquired X-ray frame to determine whether the second criterion is met based on these image features. The explicit check means that positioning and identification of image features and determining whether the second criterion is met are performed by applying separate steps. Furthermore, the explicit check means that the second criterion is checked using human expert knowledge based on fixed rules.

[0038] Based on the acquired X-ray frame, an explicit check of the adjusted collimation of the acquired X-ray frame is performed by checking whether the adjusted collimation meets a second criterion of the criterion data.

[0039] In addition to the explicit check, or optionally, or alternatively, an implicit check is performed on the adjusted collimation of the acquired X-ray frame by applying a trained AI-based model to the acquired X-ray frame to check whether it meets the second standard of the standard data. The implicit check includes using an end-to-end artificial neural network to determine whether the acquired X-ray frame meets the second standard. The end-to-end artificial neural network means that the input data of the AI-based model is the acquired X-ray frame, and the result includes a determination of whether the second standard is met. In other words, in addition to applying the AI-based model, no other steps need to be performed to complete the implicit check. Therefore, in the implicit check, there are no rules that must be met to pass the check, but the check is performed based on an end-to-end trained AI-based model.

[0040] In a variant of the method according to the invention for automatically adjusting the collimation for dynamic X-ray imaging, a readjustment of the collimation is performed if at least one of the implicit check and the explicit check is not passed. Advantageously, due to the redundancy of the different types of checks, a particularly robust and reliable monitoring and adjustment of the collimation for dynamic X-ray imaging is achieved.

[0041] In a variant of the method according to the invention for automatically adjusting collimation for dynamic X-ray imaging, the second criterion comprises a criterion of the following type:

[0042] - a predetermined anatomical structure must be visible in the optical image data, or in an acquired X-ray frame, or in a sequence of X-ray frames,

[0043] - The size of the possible re-adjusted collimation area should be as small as possible.

[0044] Advantageously, the anatomical structure to be examined is completely recorded by the acquired X-ray frames and the collimation area is limited to a minimum, so that the X-ray exposure of the patient is also limited to a minimum.

[0045] In other variants of the method for automatically adjusting collimation for dynamic X-ray imaging according to the invention, the explicit check comprises the step of attempting to detect landmarks related to a predetermined anatomical structure in the acquired X-ray frame to determine whether the predetermined anatomical structure is visible in the acquired X-ray frame. The explicit check further comprises the step of determining the size and / or position of a possible re-adjustment of the collimation region based on the detection of the landmark and the absence of the detection of the landmark. Advantageously, the size and position of the collimation region can be determined based on anatomical knowledge and knowledge of the particular details of the examination to be performed by the X-ray imaging sequence.

[0046] Preferably, in the method for automatically adjusting collimation for dynamic X-ray imaging, the step of detecting landmarks is performed automatically, preferably according to an AI-based model. Advantageously, landmark detection can be performed automatically and no human expert is required for the detection task. Advantageously, the amount of training data can be reduced compared to the training of an end-to-end AI-based model for implicit inspection, because the localization and identification of landmarks is assigned to an artificial neural network structure having a depth that is less than the depth of the artificial neural network structure assigned to the end-to-end AI-based model.

[0047] Particularly preferably, the size of the possible re-adjustment collimation area is determined such that it includes predefined landmarks. The landmarks usually mark the boundaries of the area to be examined in the scheduled examination. If these landmarks are visible on the acquired X-ray frames, it can be ensured that the collimation area includes the complete area for the scheduled examination.

[0048] In a variant of the method for automatically adjusting collimation for dynamic X-ray imaging according to the invention, the implicit check comprises determining whether to make the collimation area larger or smaller. Advantageously, a recommendation is generated which can be implemented directly or automatically to readjust the collimation.

[0049] Preferably, in the method for automatically adjusting collimation for dynamic X-ray imaging according to the present invention, the check is performed for each X-ray frame, or for the first few X-ray frames and then for every N X-ray frames, wherein N is a natural number greater than 1. Advantageously, the number of checks can be adapted to the volatility of the imaging scene and the computing power available for the check. For example, if the rate of change of the content of the X-ray frames is low, it is not necessary to perform a check on every single X-ray frame. In contrast, if the rate of change of the content of the X-ray frames is high, the check must be performed at a higher frequency to ensure that the ALARA principle is met.

[0050] In a variant of the method for automatically adjusting collimation for dynamic X-ray imaging according to the invention, the explicit examination comprises: extrapolating the positions of landmarks required for passing the explicit examination in subsequent X-ray frames of the dynamic X-ray imaging, and estimating the length and width of the collimation area in the subsequent X-ray frames based on the extrapolated positions. Advantageously, the collimation area can be determined prospectively for X-ray frames acquired in the future. If the X-ray frames indicate movement of the X-ray imaging device or movement of a region of interest to be recorded, the direction and speed of the movement can be determined based on a set of recorded X-ray frames, and the extrapolation is determined based on the determined direction and speed. This advantageous variant can be used for one of the following types of fluoroscopic sequences: contrast agent flow; walking examination; bending examination, in particular spinal bending examination; examination involving the application of barium volume, such as recording of swallowing movements, recording of speech disorders or recording of structural swallowing tests.

[0051] Preferably, the step of generating a readjusted collimation of the method for automatically adjusting collimation for dynamic X-ray imaging comprises: making the collimation region larger if there are missed structures in the acquired X-ray frame; and making the collimation region smaller if there are too many unwanted structures in the acquired X-ray frame. Advantageously, the collimation region can be minimized to meet the ALARA principle and at the same time, the X-ray frame shows all structures acquired for the predetermined examination.

[0052] In a further preferred variant of the method according to the invention for automatically adjusting the collimation for dynamic X-ray imaging, the step of generating the readjusted collimation is performed using one of the following modes:

[0053] - Fully automated execution,

[0054] - Manually by the user based on automatically generated suggestions.

[0055] In case of fully automatic alignment readjustment, human resources can be saved. In case of manual readjustment, the automatically determined alignment parameter values ​​can be checked by a human, in particular a human expert.

[0056] In a variant of the method for automatically adjusting the collimation for dynamic X-ray imaging, the step of generating an adjusted collimation based on the acquired (optical) image data and a first criterion is also performed according to an AI-based model. In such a variant, preferably, information of an implicit check and / or an explicit check of the adjusted collimation is also used to retrain the AI-based model for generating the adjusted collimation. Advantageously, the initial adjustment of the collimation is improved. Thus, an online cross-modal learning is achieved, which enables the improvement of the initial collimation for dynamic X-ray imaging.

[0057] In a variation of the method for generating a trained AI-based model to perform automatic inspection of adjusted collimation of acquired X-ray frames according to the present invention, the labeling input data comprises one of the following data sets:

[0058] - For explicit checks:

[0059] - X-ray frames of the examination object as input data, and detected anatomical landmarks as validation result data,

[0060] - a fluoroscopic sequence of X-ray frames of the examination object as input data, and detected anatomical landmarks as validation result data,

[0061] -For implicit checks:

[0062] - an X-ray frame of the examination object as input data, and the result of the implicit examination as verification result data,

[0063] - an X-ray frame of the examination object and / or a fluoroscopic sequence of X-ray frames of the examination object as input data, and an adjustment proposal as validation result data,

[0064] - X-ray frames of the examination subject and / or a fluoroscopic sequence of X-ray frames of the examination subject as input data and the realigned collimation as validation result data.

[0065] Advantageously, specific training data is generated to train an AI-based model for performing explicit inspection or implicit inspection for highly robust and reliable monitoring of collimation for dynamic X-ray imaging. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] The present invention will be described below with reference to the accompanying drawings again. In each of the accompanying drawings, the same components are provided with the same reference numerals.

[0067] The drawings are generally not to scale.

[0068] FIG. 1 shows a schematic diagram of a method for adjusting collimation of static X-ray imaging according to the prior art,

[0069] Figure 2 A flow chart illustrating a method for automatically adjusting collimation for dynamic X-ray imaging according to an embodiment of the present invention is shown,

[0070] Figure 3 A flow chart illustrating a method for generating a trained AI-based model to perform implicit inspection of adjusted collimation of acquired X-ray frames is shown,

[0071] Figure 4 A schematic diagram showing a medical imaging system according to an embodiment of the present invention is shown, the medical imaging system comprising an adjustment device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0072] In Fig. 1 , a schematic diagram 10 of a method for adjusting the collimation of static X-ray imaging according to the prior art is depicted. In Fig. 1 , an optical imaging process is performed, wherein an object under examination O (in this particular case, a patient) is visualized using an optical image recording device 1 (in this particular case, a combination of an RGB camera and a depth camera). A known technique for defining a collimator boundary CB is to use an RGB camera together with a depth camera. In this case, the optical image recording device 1 in the acquisition room first captures an image ID from the patient O, and then uses a trained AI-based model AI-M1 to identify several key points, i.e., landmarks LM, on the image ID. Fig. 1 shows the steps of the prior art for the specific case where the desired body part is the chest.

[0073] It has to be considered that each body part has specific keypoints (i.e., landmarks LM) and the requirement for correct alignment, therefore, having several accurate models for detecting the correct keypoints for each body part is necessary. Furthermore, good human knowledge is also required to identify possible faults in case the output of the camera model is incorrect.

[0074] As symbolized by the small figure in the middle of FIG1 , after the landmark LM is detected, a collimation area CA having a collimation boundary CB is calculated (“calculated” is symbolized by “C” in FIG1 ) based on criteria including that the detected landmark LM must be part of the collimation area CA. The mentioned method has been developed and is frequently used for static X-ray imaging requiring only one acquisition at a time.

[0075] exist Figure 2 In FIG. 2 , a flow chart 200 is shown illustrating a method for automatically adjusting collimation for dynamic X-ray imaging according to an embodiment of the present invention.

[0076] In step 2.I, standard data CD is received from the library LB, the standard data CD including a first standard C1 and a second standard C2 for adjusting collimation. The first standard C1 is related to collimation based on the optical image data ID from the object O under examination, and the second standard C2 is related to collimation based on the X-ray frame XF from the object O under examination.

[0077] In step 2.II, an optical image recording device 1 (in Figure 2 In the illustrated embodiment, an imaging device (100) acquires optical image data ID from an inspection object (O).

[0078] In step 2.III, based on the acquired optical image data ID and the first standard C1 of the standard data CD, an adjusted collimation A-CL is generated. "Adjusted collimation" means that the parameter value characterizing the size of the collimation area is adjusted.

[0079] In step 2.IV, an X-ray frame XF is acquired from the object under examination O using the adjusted collimation A-CL.

[0080] In step 2.V, an explicit check EC of the adjusted collimation A-CL of the acquired X-ray frame XF is performed by checking whether the acquired X-ray frame XF meets the second criterion C2 of the criterion data CD.

[0081] Furthermore, in step 2.V, an implicit check IC of the adjusted collimation A-CL of the acquired X-ray frame XF is also performed by applying the trained AI-based model AI-M2 to the acquired X-ray frame XF.

[0082] In step 2.VI, it is determined whether the result RS of the checks EC, IC performed in step 2.V is positive or negative. If both checks EC, IC are successfully passed, the result RS is positive; if at least one of the checks EC, IC is not passed, the result RS is negative. Figure 2 If the result RS is negative (indicated by "yes" in Figure 2 If "No" is used in the above analysis, the method continues with step 2.VII.

[0083] In step 2.VII, a re-adjusted collimation RA-CL is generated so that the second criterion C2 of the criterion data CD is most likely met. After this, the method continues with step 2.IV, in which the re-adjusted collimation RA-CL is used as the adjusted collimation A-CL to acquire another X-ray frame XF. The method continues until the dynamic X-ray imaging is completed.

[0084] exist Figure 3, a flowchart 300 illustrating a method for generating trained AI-based models AI-M2, AI-M3 to perform implicit inspection IC or explicit inspection EC of an adjusted collimation A-CL of an acquired X-ray frame XF is depicted.

[0085] In step 3.I, the labeled input data L-ID is generated, and the labeled input data L-ID includes the input data IND and the verification result data V-RS. The labeled input data L-ID includes one of the following data sets:

[0086] - an X-ray frame XF of the examination object O as input data and the detected anatomical landmarks LM as verification result data V-RS; or

[0087] - an X-ray frame XF of the object under examination O as input data and the result RS of the implicit examination IC as verification result data V-RS; or

[0088] - X-ray frames XF of the examination object O as input data and adjustment suggestions or automatic readjustment of measurement results according to the result RS of the implicit examination IC as verification result data V-RS.

[0089] In step 3.II, the AI-based models M2, M3 to be trained are applied to the labeled input data L-ID, wherein result data RS are generated. The second model M2 is trained to perform end-to-end determination of collimation realignment in implicit inspection IC, and the third model M3 is trained to perform localization and identification of landmarks LM in X-ray frames XF.

[0090] In step 3.III, the AI-based models M2, M3 are trained based on the result data RS and the verification result data V-RS.

[0091] In step 3.IV, the trained AI-based models AI-M2, AI-M3 are provided for use in a method for AI-assisted adjustment of the collimation CL for dynamic X-ray imaging.

[0092] exist Figure 4 , a schematic diagram of a medical imaging system 50 according to an embodiment of the present invention is shown. The medical imaging system 50 includes an adjustment device 40 according to an embodiment of the present invention.

[0093] In addition to the adjustment device 40, the medical imaging system 50 also includes a library LB depicted in the upper left corner of the medical imaging system 50. In addition, the medical imaging system 50 includes: Figure 4 The optical imaging device 1 is depicted at the upper middle position of and includes an RGB camera device and a depth camera device; and Figure 4 The upper right side depicts an X-ray imaging device 2 .

[0094] exist Figure 4 The adjustment device 40 depicted in the lower part of includes an input interface 41 for receiving standard data CD for adjusting the collimation of the X-ray imaging device 2 from the library LB. The input interface 41 is also configured to receive optical image data ID of the inspection object O, and to repeatedly receive X-ray frames XF of the inspection object O from the X-ray imaging device 2. Figure 4 In FIG. 4 , the input interface 41 is shown in two parts.

[0095] Furthermore, the adjustment device 40 comprises an adjustment unit 42 for generating an adjusted collimation A-CL based on the acquired optical image data ID and the standard data CD.

[0096] The adjustment device 40 further includes an output interface 43 , which is used to output the adjusted collimation A-CL to the X-ray imaging device 2 to adjust the collimation of the X-ray imaging device 2 .

[0097] The adjustment device 40 also includes a checking unit 44, which is used to: perform an explicit check EC on the acquired X-ray frame XF by checking whether it meets the second standard of the standard data CD, or perform an implicit check IC on the acquired X-ray frame XF by applying the trained AI-based model AI-M2.

[0098] The adjustment unit 42 is further configured to generate a re-adjusted alignment RA-CL based on the result RS of the explicit check EC and the implicit check IC of the check unit 44. If an implicit end-to-end check IC is performed, the re-adjusted alignment RA-CL may also be generated by the check unit 44. Figure 3 As mentioned in the context of step 3.I, the result RS may also include automatic rescaling of measurement results.

[0099] The output interface 43 is further configured to output the readjusted collimation RA-CL to the X-ray imaging device 2 to readjust the collimation CL of the X-ray imaging device 2 .

[0100] The above description is only a preferred embodiment of the present disclosure and is not intended to limit the present disclosure. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present disclosure should be included in the protection scope of the present disclosure.

[0101] Furthermore, the use of the indefinite article "a" or "an" does not exclude that the mentioned feature may also occur several times. Likewise, the term "unit" or "means" does not exclude that it consists of several parts which may be spatially distributed. Independently of the grammatical usage of the term, individuals with masculine or feminine identities are also included in the term.

Claims

1. A method for automatically adjusting collimation (CL) for dynamic X-ray imaging, comprising the following steps: i) receiving standard data (CD) for adjusting the collimation (CL) from a library (LB), wherein the standard data (CD) is assigned to a selected inspection scheme and comprises: a first criterion (C1) relating to the generation of an adjusted collimation (A-CL) based on optical image data (ID) from the object under examination (O), a second criterion (C2) relating to the generation of a realigned collimation (RA-CL) based on an X-ray frame (XF) from the object under examination (O), ii) acquiring optical image data (ID) from the inspection object (O) using the optical image recording device (1), iii) generating an adjusted collimation (A-CL) based on the acquired optical image data (ID) and said first criterion (C1), iv) using a dynamic X-ray imaging device (2), acquiring an X-ray frame (XF) from the examination object (O) with the adjusted collimation (A-CL), v) based on the acquired X-ray frame (XF), performing an automatic check (EC, IC) of the adjusted collimation (A-CL) of the acquired X-ray frame (XF) by checking whether the acquired X-ray frame (XF) meets said second criterion (C2), vi) if said automatic check (EC, IC) is not passed, generating a re-adjusted collimation (RA-CL) such that the next acquired X-ray frame (XF) most likely meets said second criterion (C2); or if said automatic check (EC, IC) is passed, maintaining said adjusted collimation (A-CL).

2. The method according to claim 1, wherein: If the automatic inspection (EC, IC) is passed in step vi), steps iv) to vi) are repeated with the adjusted collimation (A-CL) unchanged, or if a readjusted collimation (RA-CL) is generated in the last step vi), steps iv) to vi) are repeated with the readjusted collimation (RA-CL) as the adjusted collimation (A-CL) until the dynamic X-ray imaging is completed.

3. The method according to claim 1 or 2, wherein: The automatic inspection (EC, IC) comprises at least one or both of the following steps: - based on the acquired X-ray frame (XF), performing an explicit check (EC) of the adjusted collimation (A-CL) of the acquired X-ray frame (XF) by checking whether the acquired X-ray frame (XF) meets said second criterion (C2), -Performing an implicit check (IC) on the adjusted collimation (A-CL) of the acquired X-ray frame (XF) by applying the trained AI-based model (AI-M2) to the acquired X-ray frame (XF) to check whether the acquired X-ray frame (XF) meets the second criterion (C2).

4. A method according to any one of the preceding claims, wherein: The standard data (CD) includes the following standards: - in said optical image data (ID) or in said X-ray frame (XF) a predetermined anatomical structure must be visible, - The size of the collimation area (CA) allocated for possible realignment of the collimation (RA-CL) should be as small as possible.

5. The method according to claim 3 or 4, wherein: An explicit check (EC) is performed, and the explicit check (EC) comprises the following steps: - attempting to detect a landmark (LM) associated with a predetermined anatomical structure in the acquired X-ray frame (XF) to determine whether the predetermined anatomical structure is visible in the acquired X-ray frame (XF), - Determining the size of the collimation area (CA) allocated for possible realignment (RA-CL) based on the detection of a landmark (LM) and the absence of a landmark (LM).

6. The method according to claim 5, wherein: The step of detecting landmarks (LM) is performed automatically, preferably according to an AI based model (AI-M3).

7. The method according to claim 5 or 6, wherein: A collimation area (CA) allocated to a possible realignment (RA-CL) is sized such that the collimation area (CA) includes a predefined landmark (LM).

8. The method according to any one of claims 3 to 7, wherein: An explicit check (EC) is performed, and the explicit check (EC) comprises: - extrapolating the positions of required landmarks (LM) in subsequent X-ray frames (XF) of said dynamic X-ray imaging, - Estimate the length and width of the collimation area (CA) in subsequent X-ray frames (XF) based on the extrapolated position.

9. The method according to any one of claims 3 to 8, wherein: An implicit check (IC) is performed and includes determining whether to make the collimation area (CA) larger or smaller.

10. The method according to any one of claims 2 to 9, wherein: The automatic checks (EC, IC) are performed on: - performed for each X-ray frame (XF), or - Executed on the first few X-ray frames (XF), and then executed every N X-ray frames (XF), where N is a natural number greater than 1.

11. A method for generating a trained AI-based model (AI-M2, AI-M3) to perform automatic inspection (EC, IC) of an adjusted collimation (A-CL) of an acquired X-ray frame (XF), comprising the following steps: - generating label input data (L-ID), the label input data (L-ID) comprising input data and verification result data (V-RS), wherein the label input data (L-ID) comprises: an X-ray frame (XF) of the examination object (O) as input data, and information related to satisfying a second criterion (C2) of the criterion data (CD) as verification result data (V-RS), the second criterion (C2) being related to generating a realigned collimation (RA-CL) based on the X-ray frame (XF) from the examination object (O), - applying the AI-based model (M2, M3) to be trained to said labeled input data (L-ID), wherein result data (RS) are generated, - training said AI-based model (M2, M3) based on said result data (RS) and said verification result data (V-RS), -Provide the trained AI-based models (AI-M2, AI-M3).

12. An adjustment device (40), comprising: - an input interface (41), - the input interface (41) is used to receive standard data (CD) for adjusting the collimation (CL) from a library (LB), wherein the standard data (CD) is assigned to a selected inspection scheme, wherein the standard data (CD) comprises: a first criterion (C1) relating to the generation of an adjusted collimation (A-CL) based on optical image data (ID) from the object under examination (O), a second criterion (C2) relating to the generation of a realigned collimation (RA-CL) based on an X-ray frame (XF) from the object under examination (O), - the input interface (41) is used to receive optical image data (ID) from the inspection object (O), and - the input interface (41) is used to repeatedly receive X-ray frames (XF) from the examination object (O) from the X-ray imaging device (2), - an adjustment unit (42) for generating an adjusted collimation (A-CL) based on the acquired optical image data (ID) and the first criterion (C1), - an output interface (43), the output interface (43) being used to output the adjusted collimation (A-CL) to the X-ray imaging device (2) so as to adjust the collimation (CL) of the X-ray imaging device (2), - a checking unit (44) for performing an automatic check (EC, IC) on the acquired X-ray frame (XF) by checking whether the acquired X-ray frame (XF) meets a second criterion (C2) of the criterion data (CD), in, - the adjustment unit (42) is further configured to: if the automatic check (EC, IC) is not passed, generate a re-adjusted collimation (RA-CL) such that the second criterion (C2) is most likely met; and / or if the automatic check (EC, IC) is passed, maintain the adjusted collimation (A-CL), - the output interface (43) is configured to: output the re-adjusted collimation (RA-CL) to the X-ray imaging device (2) to re-adjust the collimation (CL) of the X-ray imaging device (2), or output the maintained adjusted collimation (A-CL).

13. A medical imaging system (50), comprising: - X-ray imaging device (2), - a library (LB) comprising standard data (CD) for adjusting the collimation (CL) of the X-ray imaging device (2), - an optical image device (1) for recording optical image data (ID) from an object to be examined (O), - The adjustment device (40) according to claim 12, wherein the adjustment device (40) is used to control the collimation (CL) of the X-ray imaging device (2).

14. A computer program product comprising instructions which, when executed by a computer, cause the computer to perform steps i) to vi) of the method for automatically adjusting collimation (CL) for dynamic X-ray imaging according to any one of claims 1 to 10; or cause the computer to perform the following steps of the method for generating a trained AI-based model (AI-M2, AI-M3) to perform automatic inspection (EC, IC) of the adjusted collimation (A-CL) of the acquired X-ray frame (XF) according to claim 11: a step of generating labeled input data (L-ID), a step of applying the AI-based model (M2, M3) to be trained to the labeled input data (L-ID), and a step of training the AI-based model (M2, M3).

15. A computer program comprising instructions which, when executed by a computer, cause the computer to perform steps i) to vi) of the method for automatically adjusting collimation (CL) for dynamic X-ray imaging according to any one of claims 1 to 10; or cause the computer to perform the following steps of the method for generating a trained AI-based model (AI-M2, AI-M3) to perform automatic inspection (EC, IC) of the adjusted collimation (A-CL) of the acquired X-ray frame (XF) according to claim 11: a step of generating labeled input data (L-ID), a step of applying the AI-based model (M2, M3) to be trained to the labeled input data (L-ID), and a step of training the AI-based model (M2, M3).