Automated x-ray collimation system

Abstract

A system for controlling collimation comprises an automated collimation controller that includes a processor configured to evaluate one or more images from an x-ray imaging system to identify features of interest. The x-ray imaging system comprises a collimator controlled by the automated collimation controller. The processor is further configured to obtain at least one feature weight parameter or dose weight parameter of the automated collimation controller; determine a weighted combination of the identified features of interest based on the at least one feature weight parameter or dose weight parameter; and tune collimation settings based on the weighted combinations of the identified features of interest.

Description

AUTOMATED X-RAY COLLIMATION SYSTEMFIELD

[0001] The present disclosure generally relates to the field of X-ray imaging technology, and more specifically, to systems, methods, and devices that allow for automated and tunable control of X- ray collimation.BACKGROUND

[0002] X-ray imaging is a widely used diagnostic tool in the medical field, providing clinicians with a non-invasive method to visualize the internal structures of a patient's body. This technology relies on the principle of differential absorption of X-ray radiation by different tissues in the body, resulting in a contrasted image that can be interpreted by medical professionals.

[0003] One of the components of an X-ray imaging system is the collimator, which is used to shape and direct the X-ray beam towards the region of interest. The collimator typically includes a shutter and a wedge filter, which can be adjusted to control the size and shape of the X-ray beam. This is a process known as collimation. Collimation is an integral part of X-ray imaging as it helps to improve image quality by reducing scatter from surrounding tissues, and also helps to reduce the radiation dose to the patient by blocking unnecessary X-ray radiation from the patient's body, as well as helps to reduce the radiation dose to the operator(s) due to scatter.

[0004] In the context of interventional procedures, X-ray collimation settings often require frequent adjustments to accommodate the varying phases of the procedure and the specific anatomical features of interest. For instance, during an endovascular aneurysm repair procedure, the field of view may initially include the patient's whole head for a diagnostic image, but may later be zoomed in to focus on the specific vasculature being navigated to reach the aneurysm.

[0005] However, manual adjustment of collimation settings can be challenging due to the complexity of the controls and the fast-paced nature of interventional procedures. This has led to underutilization of collimation in many cases, potentially compromising image quality and increasing radiation exposure to both patients and medical staff.

[0006] Artificial intelligence (Al) has been increasingly applied in various aspects of medical imaging, including image acquisition, processing, and interpretation. In the context of X-ray collimation, Al algorithms can potentially automate the adjustment of collimation settings, thereby reducing the manual workload for clinicians and improving the efficiency of interventional procedures. However, the implementation of such Al-driven collimation algorithms presents its own set of challenges, particularly in terms of ensuring that the algorithm outputs are clinically acceptable and can be tuned according to the preferences of different users.SUMMARY OF INVENTION

[0007] According to an aspect of the inventive concepts, an X-ray imaging system includes a data processing unit configured to collect and process images, and a processor configured to apply an automatic collimation controller to evaluate images from the X-ray imaging system to identify features of interest and to tune the automated collimation controller via weighted combinations of the identified features of interest.

[0008] The automatic collimation controller may include artificial intelligence (Al) models.

[0009] The X-ray imaging system may further include a user interface for adjusting parameters of the automatic collimation controller. The user interface include a set of interactable sliders, each slider corresponding to a parameter of the automated collimation controller. The X-ray imaging system may include a collimator including a shutter and a wedge, where the set of interactable sliders includes a slider for adjusting the shutter feature weight, a slider for adjusting the shutter dose weight, a slider for adjusting the wedge feature weight, and a slider for adjusting the wedge dose weight. The automatic collimation controller may be further configured to adapt its outputs based on user feedback provided via interaction with the user interface.

[0010] The automatic collimation controller may be further configured to adapt its outputs based on eye tracking data indicating a user's gaze on regions of the image.

[0011] The automatic collimation controller may be further configured to run in the background and provide collimation recommendations when the user's manual settings are within a predetermined range.

[0012] The automatic collimation controller may be further configured to adjust feature weights based on a known phase of a procedure being performed.

[0013] The X-ray imaging system of claim 1, further comprising a calibration mode for a user to customize the automated collimation controller to the user’s preferences by providing feedback on a set of representative images.

[0014] The automatic collimation controller may be further configured to adjust collimation settings based on the presence or absence of interventional devices in the image.

[0015] According to another aspect of the inventive concepts, a method for controlling X-ray collimation in an interventional procedure is provided. The method includes receiving an image from an interventional imaging system, the interventional imaging system including an X-ray collimator, extracting features from the image corresponding to the relevance of displaying each part of the image to a user, generating a pixel-level map indicating the relevance of revealing each pixel to the user, and adjusting collimation settings of the X-ray collimator based on the pixellevel map and user preferences.

[0016] The extraction of features from the image may include using a convolutional neural network to generate a heatmap indicating the relevance of revealing each pixel to the user. The convolutional neural network may be trained using a dataset of interventional images annotated to highlight the relevance of different anatomical features.

[0017] The method may further include adjusting the collimation settings based on eye tracking data indicating a user's gaze on regions of the image. The adjusting collimation settings may include optimizing an objective function that balances the relevance of revealing each pixel to the user and the reduction of radiation dose.

[0018] According to yet another aspect of the inventive concepts, method for calibrating an automated X-ray collimation system is provided. The method includes presenting a user with a set of representative interventional images, receiving a user input indicating preferred collimation settings for each image, determining parameters of an automated collimation controller that would result in the preferred collimation settings, and storing the determined parameters for use in subsequent interventional procedures.

[0019] The method may further include presenting the user with a set of representative interventional images from different procedure types.

[0020] The receiving of the user input indicating preferred collimation settings for each image may include includes drawing the preferred collimation settings on the image.

[0021] The parameters of the automated collimation controller are stored in a user profile for use in subsequent interventional procedures.

[0022] According to another aspect of the inventive concepts, a system for controlling collimation includes an automated collimation controller that includes at least one processor configured to: evaluate one or more images from an x-ray imaging system to identify features of interest, wherein the x-ray imaging system comprises a collimator controlled by the automated collimation controller, obtain at least one feature weight parameter or dose weight parameter of the automated collimation controller, determine a weighted combination of the identified features of interest based on the at least one feature weight parameter or dose weight parameter, and tune collimation settings based on the weighted combinations of the identified features of interest.

[0023] The automatic collimation controller of the system may include artificial intelligence (Al) models to tune the collimation settings. The system may further include a user interface for adjusting parameters of the automated collimation controller, and the user interface may comprise a set of interactable sliders, each slider corresponding to a parameter of the automated collimation controller. The system may further include the collimator that includes a shutter and a wedge, and the at least one feature weight parameter or dose weight parameter includes at least one of a shutter feature weight, a shutter dose weight, a wedge feature weight, or a wedge dose weight.

[0024] The automatic collimation controller may be further configured to adapt the collimation settings based on at least one of user feedback provided via interaction with the user interface or eye tracking data indicating a gaze of the user on regions of the image. The automatic collimation controller may be further configured to run in the background and provide collimation recommendations when manual settings of the user are within a predetermined range. The at least one processor may be further configured to adjust the at least one feature weight parameter or dose weight parameter based on a known phase of a procedure being performed. The automatic collimation controller may be further configured to tune the collimation settings based on the presence or absence of interventional devices in the image.

[0025] The at least one processor may be further configured to: extract a plurality of features from an image of the one or more images, the plurality of features corresponding to relevance of displaying each part of the image to a user; generate a pixel-level map indicating relevance of revealing each pixel of the image to the user based on the extracted features; and adjust thecollimation settings based on the pixel-level map and user preferences. The at least one processor may be further configured to use a convolutional neural network to extract the plurality of features from the image and generate a heatmap indicating the relevance of revealing each pixel to the user, wherein the convolutional neural network is trained using a dataset of interventional images annotated to highlight relevance of different anatomical features. To adjust the collimation settings, the at least one processor is further configured to optimize an objective function that balances the relevance of revealing each pixel to the user and the reduction of radiation dose.

[0026] The at least one processor may be further configured to calibrate the x-ray imaging system by: presenting a user with a set of representative interventional images; receiving a user input indicating preferred collimation settings for each representative interventional image; determining parameters of the automated collimation controller that would result in the preferred collimation settings; and storing the determined parameters for use in subsequent interventional procedures.

[0027] According to another aspect of the inventive concepts, a method for controlling collimation includes: evaluating one or more images from an x-ray imaging system to identify features of interest, wherein the x-ray imaging system comprises a collimator controlled by the automated collimation controller; obtaining at least one feature weight parameter or dose weight parameter of the automated collimation controller; determining a weighted combination of the identified features of interest based on the at least one feature weight parameter or dose weight parameter; and tuning collimation settings based on the weighted combinations of the identified features of interest.

[0028] The automatic collimation controller used in the method may include artificial intelligence (Al) models to tune the collimation settings. The parameters of the automated collimation controller may be adjustable by a user interface, wherein the user interface comprises a set of interactable sliders, each slider corresponding to a parameter of the automated collimation controller. The collimator used in the method may include a shutter and a wedge, and the at least one feature weight parameter or dose weight parameter may include at least one of a shutter feature weight, a shutter dose weight, a wedge feature weight, or a wedge dose weight.

[0029] According to another aspect of the inventive concepts, a non-transitory computer-readable storage medium has stored a computer program comprising instructions, which, when executed by a processor of an automated collimation controller, cause the processor to: evaluate one or moreimages from an x-ray imaging system to identify features of interest, wherein the x-ray imaging system comprises a collimator controlled by the automated collimation controller; obtain at least one feature weight parameter or dose weight parameter of the automated collimation controller; determine a weighted combination of the identified features of interest based on the at least one feature weight parameter or dose weight parameter; and tune collimation settings based on the weighted combinations of the identified features of interest.

[0030] The automatic collimation controller may include artificial intelligence (Al) models to tune the collimation settings. The parameters of the automated collimation controller may be adjustable by a user interface, wherein the user interface comprises a set of interactable sliders, each slider corresponding to a parameter of the automated collimation controller. The collimator used in the method may include a shutter and a wedge, and the at least one feature weight parameter or dose weight parameter may include at least one of a shutter feature weight, a shutter dose weight, a wedge feature weight, or a wedge dose weight.BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The above and other aspects and features of the inventive concepts will become readily apparent from the detailed description that follows, with reference to the accompanying drawings, in which:

[0032] FIG. 1 illustrates an X-ray imaging system that can be used to implement embodiments of the inventive concepts;

[0033] FIG. 2 is a perspective view for reference in describing a collimator that that can be used to implement embodiments of the inventive concepts;

[0034] FIG. 3 provides examples of potential desired collimation shutter placement for specific medical procedures;

[0035] FIG. 4 depicts a processing system that may be used to implement embodiments of the inventive concepts;

[0036] FIG. 5 shows a user interface displaying a user modification to a wedge filter position, according to embodiments of the inventive concepts;

[0037] FIG. 6 demonstrates an example of eye tracking to detect whether regions of interest in an image are covered by a wedge filter according to embodiments of the inventive concepts;

[0038] FIG. 7 presents an example of a user interacting with collimation settings while the automated collimation controller runs in the background according to embodiments of the inventive concepts; and

[0039] FIG. 8 illustrates an automated collimation controller output that positions collimation shutters and wedges to collimate out all air around the patient's head or collimate out all anatomy away from a device depending on the procedure phase according to embodiments of the inventive concepts.DETAILED DESCRIPTION

[0040] In the following detailed description, for purposes of explanation and not limitation, representative embodiments disclosing specific details are set forth in order to provide a thorough understanding of the present teachings. However, it will be apparent to one having ordinary skill in the art having had the benefit of the present disclosure that other embodiments according to the present teachings that depart from the specific details disclosed herein remain within the scope of the appended claims. Moreover, descriptions of well-known apparatuses and methods may be omitted to avoid obscuring the description of the example embodiments. Such methods and apparatuses are clearly within the scope of the present teachings. Further, throughout the drawings, like reference numbers refer to the same or similar elements.

[0041] The terminology used herein is for purposes of describing particular embodiments only, and is not intended to be limiting. The defined terms are in addition to the technical and scientific meanings of the defined terms as commonly understood and accepted in the technical field of the present teachings. As used in the specification and appended claims, the terms ‘a’, ‘an’ and ‘the’ include both singular and plural referents, unless the context clearly dictates otherwise. Thus, for example, ‘a device’ includes one device and plural devices. Further, for example, when one element is described as being “connected to” another element, the one element may be directly connected to the other element, or indirectly connected to the other element in an operative manner.

[0042] As is traditional in the field of the inventive concepts, embodiments may be described, and illustrated in the drawings, in terms of functional blocks, units and / or modules. Those skilled in the art will appreciate that these blocks, units and / or modules are physically implemented by electronic (or optical) circuits such as logic circuits, discrete components, microprocessors, hard-wired circuits, memory elements, wiring connections, and the like, which may be formed using semiconductor-based fabrication techniques or other manufacturing technologies. In the case of the blocks, units and / or modules being implemented by microprocessors or similar, they may be programmed using software (e.g., microcode) to perform various functions discussed herein and may optionally be driven by firmware and / or software. Alternatively, each block, unit and / or module may be implemented by dedicated hardware, or as a combination of dedicated hardware to perform some functions and a processor (e.g., one or more programmed microprocessors and associated circuitry) to perform other functions. Also, each block, unit and / or module of the embodiments may be physically separated into two or more interacting and discrete blocks, units and / or modules without departing from the scope of the inventive concepts. Further, the blocks, units and / or modules of the embodiments may be physically combined into more complex blocks, units and / or modules without departing from the scope of the inventive concepts.

[0043] The present disclosure pertains to the field of X-ray imaging technology, and more specifically, to systems, methods, and devices that leverage artificial intelligence (Al) to automatically adjust and tune X-ray collimation settings. Collimation, a process that involves adjusting the settings of a collimator to narrow the beam of X-ray radiation to a specific area of a patient's body, plays a pivotal role in X-ray imaging. By effectively controlling the size and shape of the X-ray beam, collimation can enhance the quality of the image and minimize the patient's exposure to radiation.

[0044] In some aspects, the present disclosure provides an X-ray imaging system equipped with a data processing unit, a user interface, and a collimator. The data processing unit, designed to collect and process anatomical images, may by way of example implement an automatic collimation controller. As an example, the controller may include Al models that evaluate images from the X-ray imaging system to identify features of interest and enable tunability of an automated collimation algorithm via weighted combinations of these identified features. Themodels may be further configured to adapt to user feedback and adjust the collimation settings based on this feedback, thereby offering a personalized and efficient approach to collimation.

[0045] In some cases, the user interface, connected to the data processing unit, displays information to the user and facilitates user interaction with the system. The collimator, comprising a shutter and a filter wedge, collimates X-ray radiation onto a region of interest. The automatic collimation algorithm, implemented in the data processing unit, evaluates images from the X-ray imaging system to identify features of interest. The algorithm enables tunability of the automated collimation algorithm via weighted combinations of these identified features, and is further configured to adapt to user feedback and adjust the collimation settings based on this feedback.

[0046] For example, in some embodiments, the automated collimation controller that includes at least one processor configured to: evaluate one or more images from an x-ray imaging system to identify features of interest, wherein the x-ray imaging system comprises a collimator controlled by the automated collimation controller, obtain at least one feature weight parameter or dose weight parameter of the automated collimation controller, determine a weighted combination of the identified features of interest based on the at least one feature weight parameter or dose weight parameter, and tune collimation settings based on the weighted combinations of the identified features of interest.

[0047] In other aspects, the present disclosure provides methods for controlling X-ray collimation in an X-ray imaging system. These methods involve receiving an image from the X- ray imaging system, extracting features from the image that correspond to the relevance of displaying each part of the image to a user, and generating a pixel-level map based on the extracted features. The value at each pixel indicates the relevance of revealing that pixel to the user. The collimation settings of the X-ray imaging system are then adjusted based on the pixellevel map, and this adjustment is tunable based on user feedback.

[0048] In yet other aspects, the present disclosure provides methods for tuning an automatic collimation controller in an X-ray imaging system. These methods involve receiving an image from the X-ray imaging system, extracting features from the image that correspond to the relevance of displaying each part of the image to a user, and generating a pixel-level map based on the extracted features. User feedback regarding the collimation settings is received, and thecollimation settings of the X-ray imaging system are adjusted based on the pixel-level map and the user feedback. The adjustment of the collimation settings is tunable based on the user feedback, thereby providing a personalized and efficient approach to collimation.

[0049] These systems and methods offer potential benefits in the field of X-ray imaging, including improved image quality, reduced radiation exposure, and enhanced user control over collimation settings. In some example embodiments, by leveraging Al to automate and tune collimation settings, these systems and methods address the challenges associated with manual adjustment of collimation settings and offer a more efficient and personalized approach to X-ray imaging.

[0050] FIG. 1 is perspective view illustrating an example of an X-ray medical scanning system10. The system 10 may be used, for example, during an interventional procedure to assist a surgeon in positioning and monitoring an interventional device within a patient.

[0051] Referring to FIG. 1, the X-ray medical scanning system 10 of this example includes an X-ray source 12 that produces X-ray radiation, and a collimator 13 that includes a shutter and a wedge (not shown) for collimating the X-ray radiation onto a region of interest. A table 14 may be provided for receiving a patient to be examined. Further, the system 10 includes X-ray image detection module 16 positioned opposite to the X-ray source 12. During the imaging procedure, the patient may be located on the table 14, i.e., between the X-ray source 12 and the detection module 16. X-rays may be emitted by the X-ray source 12 and pass through the patient before being detected by the detection module 16.

[0052] A data processing unit 18 may be connected to the detection module 16, the X-ray source 12 and the collimator 13. The data processing unit may transmit control signals to the X-ray source 12 and the collimator 13, and may receive detection data from the detection module 16. In the case of the collimator 13, the control signal may control the positioning and / or transparency of the shutter and filter wedge of the collimator 13. The data processing unit 18 may be configured to collect and process anatomical images based on the data received from the detection module 16.

[0053] In addition, the system 10 may include a display device 20 for displaying information to the person operating the X-ray imaging system 10, i.e., a clinician such as a cardiologist or cardiac surgeon. For example, the image representation showing the current position of the interventionaldevice in the patient may be displayed on the display device 20. Further, an interface unit 22 in the form of a keyboard, control knobs, etc. may be provided to enable an operator to input information.

[0054] In some aspects, the X-ray medical scanning system 10 may be a so-called C-type X-ray image acquisition device. However, the inventive concepts are not limited to any particular type of X-ray imaging system. Nor are the inventive concepts limited to the exemplary set-up of FIG. 1.

[0055] FIG. 2 is a perspective view for reference in describing a collimator 40.

[0056] Referring to FIG. 2, in this example a patient P is located below an X-ray generator 30. The X-ray generator is connected to the power and control cables 32 / 33, and includes a transmission window 31. As is well understood in the art, X-ray radiation is emitted through the window 31 in a direction towards the patient P.

[0057] Interposed between the X-ray generator 30 and the patient P is the collimator 40.

[0058] The collimator 40 may include an upper window 44 that is aligned with the window 31 of the X-ray generator 30, and a lower surface region that is transparent to X-ray radiation and light. In the examples herein, the collimator 40 has collimator components that include a shutter assembly 41 and a wedge filter 42. The shutter assembly 41 and the wedge filter 42 may be used for collimating X-ray radiation onto a region of interest. The shutter assembly 41 may be configured to control the size and shape of the X-ray beam, while the wedge filter 42 may be used to attenuate the X-ray beam, reducing its intensity. In some cases, the collimator 40 may include multiple shutters and wedges for collimating X-ray radiation, providing greater flexibility and control over the collimation process.

[0059] The collimator 40 may also include a mirror 46 and a light source 43. The mirror 46 may be used to reflect light from the light source 43 onto the region of interest, aiding in the visualization of the region of interest.

[0060] Also shown in FIG. 2 is a detection module 50 located below a patient P. The detection module 50 may be configured to detect X-ray radiation that has passed through the patient P and generate image data based on the detected radiation.

[0061] In some embodiments, the X-ray medical scanning system may include a method for automatically placing angular wedge filters as well as the rectangular shutters. This method may involve adjusting the position and orientation of the wedge filters and shutters based on the pixel-level map and user feedback, as described previously. This allows for a more precise and customizable control over the collimation settings, enabling the user to adjust the collimation settings according to their specific preferences and the requirements of the imaging procedure.

[0062] To provide some context, FIG. 3 illustrates potential desired collimation shutter placements for a variety of specific medical procedures, such as a right femoral access cerebral aneurysm case, as identified through discussions with interventionalists. The figure shows an X-ray image of a patient's body with areas of interest highlighted. The highlighted areas represent the regions that are considered relevant for displaying to the user during the specific medical procedure.

[0063] In some aspects, the highlighted areas may correspond to the regions of the body that are not collimated out by the shutters of the collimator. These regions may be determined based on the pixel-level map generated by the Al-based automatic collimation algorithm, as described previously. The pixel-level map may assign higher values to the pixels corresponding to the highlighted areas, indicating their relevance for displaying to the user.

[0064] In some cases, the placement of the collimation shutters may be adjusted based on the pixel -level map and user feedback. For instance, the user may interact with the user interface of the X-ray imaging system to adjust the position and orientation of the shutters, thereby controlling the size and shape of the X-ray beam and the regions of the body that are collimated out.

[0065] In other cases, the placement of the collimation shutters may be automatically adjusted by the Al-based automatic collimation algorithm. The algorithm may evaluate the pixel-level map and user feedback to determine the ideal placement of the shutters that would result in the desired collimation settings.

[0066] It is to be understood that the specific medical procedure and the highlighted areas shown in FIG. 3 are provided as an example, and the present disclosure may be applicable to other medical procedures and other regions of interest. Furthermore, the collimation settings may be adjusted based on other factors, such as the phase of the procedure, the user's preferences, and the requirements of the imaging procedure.

[0067] FIG. 4 is a block diagram for reference in describing an example of the processing system 200 that may be utilized in embodiments of the inventive concepts. In particular, the system 200 may be used to control the X-ray generator and the collimator, which are collectively represented by reference number 211 in the figure. Reference number 214 denotes collimated X-ray radiationtransmitted towards a patient, and reference number 218 denotes detected X-ray energy after passing through the patient.

[0068] In the example of FIG. 4, a system base 230 includes a bus system 226 having a number of system components connected thereto. Among these components are a processor module 223, a user interface module 236, a local memory module 229 and a control and acquisition module 240.

[0069] In at least some embodiments, the processor module 223 controls an overall operation of the system 200. The processor module 223 includes one or more central processing units (CPUs) and may operate in accordance with executable instructions 225, which may be originally stored in the local memory 229. In addition, the processor module 223 may include an Al engine 228, which is generically represented in FIG. 4 by a neural network having input and output layers. Further, the processing module 223 may include a separate display processor 234 for controlling the display of images on one or more display devices. In the figure, the Al engine 228 and the executable instructions may be considered to constitute an example of processor implemented automatic collimation controller 227. However, as will be noted later, the automatic collimation control of the inventive concepts is not limited to Al implementations.

[0070] The control and acquisition module 240 is the interface with the X-ray generator and collimator 211. This module 224 may include a controller module 224 for controlling operations of the X-ray generator and collimator 211 according to instructions received from the processing module 223. The module 224 may also include a signal processor 222 for pre-processing of X- ray image signals received from the X-ray apparatus. The X-ray & collimation and image reception module 220 is an interface that forwards control signals from the controller 224 to the X-ray generator and collimator 211, and image signals from the X-ray generator and collimator 211 to the signal processor 222.

[0071] The user interface 236 may include a display 238 for displaying the X-ray images 232 and in some cases, interactive graphical user interface (GUI) components. The user interface 236 may also include one or more user controls 237 for controlling operation(s) of the system 200. In some embodiments, the user control(s) 237 may include one or more hard controls (e.g., buttons, knobs, dials, encoders, mouse, trackball or others), which may be provided on a control panel of the system base 230. In some embodiments, the user control(s) 237 may additionally or alternativelyinclude soft controls (e.g., GUI control elements or simply, GUI controls) provided on a touch sensitive display.

[0072] As mentioned above, the system 200 may also include local memory 229. The local memory may be provided by one or more hard disk drives, solid-state drives, or any other type of suitable storage device comprising non-volatile memory. The local memory 229 may be configured to store image data, executable instructions, or any other information that is deemed to be beneficial for the operation of system 200. In some examples, the system 200 may also be communicatively connected (via wired or wireless connection) to external memory, for example a picture archiving and communication system (PACS) storage device for longer term storage of image data and other patient information.

[0073] In some embodiments, the method for tuning an artificial intelligence based automatic collimation algorithm in an X-ray imaging system includes adjusting the collimation settings based on a combination of the pixel -level map, the user feedback, and a set of predefined rules associated with the X-ray imaging system. This allows the tuning of collimation settings based on desired outcomes related to radiation dose or targeting features in the anatomy rather than the more abstract parameters controlling the collimation mechanism. This also allows for more precise and customizable control over the collimation settings, enabling the user to adjust the collimation settings according to their specific preferences and the requirements of the imaging procedure.

[0074] The X-ray medical scanning system may include a feature extraction algorithm that is a convolutional neural network whose input is an interventional image and whose output is a heatmap corresponding to the above criteria. This allows the system to identify and prioritize features of interest in the image, which can then be used to adjust the collimation settings. The use of a convolutional neural network for feature extraction provides a robust and efficient method for identifying relevant features in the image, and can be particularly beneficial in cases where the image contains complex or variable anatomical structures.

[0075] With the above in mind, one purpose of at least some embodiments of the inventive concepts is to produce an automatic collimation algorithm with a set of tunable and human- understandable parameters which can be exposed to the user and changed in real-time to modify the output of the algorithm. This is a necessary design due to the vast user-to-user variation in preferred collimation settings across different procedure types, procedure phases, and anatomies.Additionally, the algorithm needs to respond to these changes quickly in order to meet the requirements of an interventional imaging setting. We structure the following algorithm with this goal of enabling and empowering the user to influence the artificial intelligence algorithm to produce their desired collimation outcome.

[0076] The algorithm takes as input an image from an interventional imaging system and extracts features from the image corresponding to the importance of displaying each part of the image to the user. The output of the algorithm would take the form of a pixel-level map, where the value at each pixel indicates the importance of revealing that pixel to the user. In one embodiment, the algorithm could be trained to associate high values to important anatomical structures relevant to the interventional procedure at hand, as well as interventional devices such as guidewires, catheters, etc. For example, in a right femoral access cerebral aneurysm procedure, the algorithm would assign high importance to visible devices as well as the anatomy of relevance in each phase of the procedure: 1) the major vessels in the access phase such as the femoral artery and the descending aorta, 2) the aortic arch and superior vessels in the arch ascent phase (aortic arch, brachial arteries, carotid arteries), 3) the local area of the neck including the carotid arteries leading up to the aneurysm site during approach to cerebral vasculature, 4) the entire head during the diagnostic phase, and 5) the neighborhood of the aneurysm in the treatment phase.

[0077] In one embodiment, the feature extraction algorithm is a convolutional neural network whose input is an interventional image and whose output is a heatmap corresponding to the above criteria.

[0078] A neural network as described in relation to FIG. 4 can be trained using a dataset of interventional images annotated in such a way that the critical anatomy is circled in a free form contour in each image frame.

[0079] Feature extractor output in the format of a heatmap enables on-the-fly tunability of the algorithm based on user preferences or interaction. In order to translate the feature extraction output to collimation settings, we pose an optimization problem whose objective is to find an equilibrium between maximizing the saliency of features revealed by the eventual region of interest (hereinafter we refer to the region of the imaging field not obscured by collimation shutters or wedges as the “region of interest” or “ROI”) and minimization of the radiation dose associated with the given collimation settings. We separately address the placement of shutters (100% beamattenuation and locked into cardinal angles with no rotational component) and wedges (less than 100% beam attenuation and freely translating / rotating) in our embodiment, however this algorithm is generalizable to other mechanisms of collimation.

[0080] Here, described below is one method for selecting a tunable placement of collimation settings from the heatmap output of the feature extraction algorithm described above. First, the placement of the rectangular shutters parameterized as [x, y, h, w] is described where:

[0081] (x,y) = RO I center

[0082] (h, w) = RO I height, width

[0083] The objective function to be minimized can be described as:

[0084] E(x, y, h, w) = a(features included) + (3 (dose reduced)

[0085] For simplicity, the energy of the dose reduction is modeled as being proportional to the area occluded:

[0086] dose reduction = — / i2w2

[0087] However, in other embodiments the dose reduction term could take the form of any radiation dose analysis function.

[0088] The energy of the features included is modeled as the sum of the heatmap values inside of the current ROI. This sum may be referred to as the function f(x,y,h,w) and the integral image method may be used to quickly compute this sum and its gradients:

[0089] H(-) = heatmap imagew.2 ^

[0092] Then, the objective function is:

[0094] One can then solve for the gradients of the objective with respect to the parameters [x, y, h, w] and maximize the energy. An example for one parameter x follows:

[0096] where — — - is the integral image function of the spatial gradient of a heatmap H. Then, for ci ) a given a and ?, there will be an optimal set of x, y, h, w that can be solved by any gradient-based optimizer. The behavior of the algorithm can be modified by changing a and ?, for instance raising a will weight inclusion of image features more heavily than dose reduction, which in practice will result in a larger ROI centered around important anatomical structures. The collimation shutter system on some X-rays systems is always centered with (x,y) at the center of the image, in which case the same algorithm can be used by optimizing h and w only.

[0097] In addition to setting the shutters, the same algorithmic structure can be used to set angular wedge filters as well as the rectangular shutters. In some X-ray systems, the wedge filters are able to freely translate and rotate around the center of the image. As an example, the objective function E above can be reformulated as follows to place a single wedge:

[0098] E =

[0101] where J is a mask image whose values are related to the attenuation of the wedge, R is a rigid transform applied to the wedge, H is the feature heatmap. / ( / ? • x) then represents the amount of image that is not attenuated, and we use the same approximation for the dose reduction term. The gradients of E with respect to 0 and a can be calculated following the same derivative procedure from above. In order to place multiple wedges, the same function can be optimized for multiple J representing each wedge filter.

[0102] In other embodiments, the same structure could be extended to non-rectangular or non- polygonal collimator / wedge filter mechanisms. In general, herein the term “collimation” interchangeably to refers to any modes of filtering or collimating the image.

[0103] Additionally, in clinical practice, the presence or absence of interventional devices such as guidewires or catheters may have a strong impact on the ideal algorithm output. A method of extending this framework to additional inputs such as one or more tips of interventional devices may be derived. These device coordinates could be provided by hardware such as electromagnetic tracking, software such as image processing, or otherwise. In the example of collimation shutters from above, the addition of device tip information is implemented as a weighted sum of the objective function above with another objective function that penalizes shutter ROIs that are far away from including the device tip in the aperture.

[0104] E(x, y, h, w) = a(features included) + (3 (dose reduced) + / (device tip inclusion)

[0105] In a simplified embodiment, the device tip inclusion criteria could be formulated as a simple loss function whose value is constant if the device tip is sufficiently inside the ROI, and increases as the device tip gets further away from the ROI, where cxand cyrepresent the coordinates of the device tip. As an example in a single dimension x:

[0107] The gradient of this expression with respect to the collimation parameters is trivially computed.

[0108] In practice, all of the above calculations are very fast at inference time and allow the user to change these parameters and observe changes in the algorithmic output at a speed that is close to standard x-ray frame rates.

[0109] In formulating the optimization problem separably on the same feature extractor output, the user is enabled to precisely tune the style of collimation around several factors. In one embodiment, the minimal set of algorithm parameters that influence the output collimation settings are:

[0110] The shutter feature weight[OHl] The shutter dose weight

[0112] The wedge feature weight

[0113] The wedge dose weight

[0114] A sequence of device weights

[0115] These parameters can be exposed to in the user interface in order to tune the algorithm to one’s personal preferences. One possible mode of interaction is a set of interactable sliders, one for each parameter. Another possible mode of interaction is a single slider that combines several parameters onto one scale - for example, moving the slider to the right could increase the shutter and wedge dose weights, decrease the shutter and wedge feature weights, and increase the device weight, resulting in a tighter region of interest closely centered around the device tip and revealing only the immediate neighborhood of critical anatomy. Interacting with the slider(s) while the x- ray is not on could additionally display a preview of the proposed automatic collimation settings which would then go into effect the next time the x-ray pedal is depressed.

[0116] FIG. 5 illustrates an example user interface displaying a user modification to a wedge filter position. That is, in an embodiment, users can provide feedback directly via interaction with the shutter or wedge positions. This may be done directly via interaction with the UI (e.g., the touch screen module available with some systems). For example, in FIG. 5, the wedge outline 500 may have been moved to the right by the user in a touch screen manner. The neural network can ingest this input by the user to adjust feature weights and adapt its outputs to the user’s preferences.

[0117] FIG. 6 represents an example embodiment in which eye tracking is used to detect whether regions of interest in an image are covered by a wedge filter, and if so, repositioning the wedge filter to expose the relative regions. Referring to FIG. 6, user modifications may also be inferred via eye tracking. To the left of the example of FIG. 6, in an eye tracking mode the user focuses from the wedge 600 to the location 601. As a result, the wedge is repositioned as shown to the right of FIG. 6. Since wedges do not block all X-ray, some X-ray features are still visible behind the wedges. In some cases, users may not be able to manually reposition the wedges or adjust sliders to expose regions behind wedges. However, if eye tracking reveals that their gaze is on regions behind wedges, the wedge position may automatically be adjusted to expose those regions and the neural network can incorporate this information into its feature weights.

[0118] Referring now to FIG. 7, in another embodiment, the algorithm may not be actively running, but may be activated or recommendations from the algorithm may be communicated to the user as necessary. For instance, if the user has turned off active collimation recommendations, the algorithm may continue to run in the background and only display an output when the user’s manual settings are close to the algorithm output. For instance, as represented by the arrow in image on the left side of FIG. 7, if the user interacts with a wedge position to open the field of view, moving the wedge outwards to a position within a range of the predicted output of the automated collimation algorithm (running in the background), then the algorithm may kick in and ‘snap’ the wedge to the predicted position. This is shown on the image on the right side of FIG. 7. The user input can further serve as a validation of the algorithm suggestions and can be incorporated into the algorithms feature weights.

[0119] In yet another embodiment, as a supplement to enabling the user to change collimation algorithm settings in real-time, the algorithm would learn the user’s personal preferences associated to anatomy, procedure phase, or other factor. One method of learning the user’spreference is to incorporate a “calibration” mode that every user would enter prior to using the automatic collimation system. In this mode, the user would be presented with a set of representative images for the procedure types that they will be performing. The user would draw their ideal collimation settings (for the example of rectangular shutters, this is equivalent to providing x, y, h, and w parameters), and the calibration system would attempt to recover the user settings parameters (a, (3 , y from the minimal example equations in the main embodiment, where a represents the importance of including high value features in the ROI, / 3 represents the importance of reducing radiation, and y represents the importance of including the device tip in the ROI).

[0120] In the example of three parameters <z, ?, and y, because these parameters are simply coefficients of a weighted sum in the algorithm formulation, one can lock one of them at an arbitrary constant and vary the other two to identify the optimal user settings that produces an ROI closest to the desired ROI. Therefore, for each image presented to the user, a lookup table of the parameter space could be pre-computed before the sale of the system to the user and identifying the optimal parameters would amount to a simple search for the most similar algorithm ROI compared to the user ROI (by a metric such as intersection over union). As an example, assuming a generous 100ms runtime for optimizing the ROI, a 100 x 100 parameter space search could be performed for a set of 100 images in 28 hours, and this would only need to be run by the manufacturer once prior to the release of the system.

[0121] Reference is now made to FIG. 8. If the phase of the procedure is known, the algorithm may accordingly modify feature weights in order to generate outputs that are appropriate for each procedure phase. For instance, when a diagnostic digitally subtracted angiography (DSA) of brain vasculature is acquired, the automated collimation algorithm should generate an output that positions collimation shutters and wedges such that all air around the patient head is collimated out (FIG. 8, left side). The algorithm should weight head anatomy features such that this output is computed, and any devices present in the anatomy may be ignored. However, in the next phase where navigation to the treatment target is performed, both devices and anatomy near the device must be given higher weight, while anatomy away from the device is given lower weight and collimated out (FIG. 8, right side). Procedure phase may be specified by the user via a userinterface, or it could be automatically inferred via sensors in the operating room that can capture and evaluate context from activities in the operating room.

[0122] While the invention has been illustrated and described in detail in the drawings and foregoing description, such illustration and description are to be considered illustrative or exemplary and not restrictive; the invention is not limited to the disclosed embodiments.

[0123] For example, in the examples given above, the automatic collimation algorithm is implemented by artificial intelligence (Al). However, the inventive concepts are not limited in this manner. Rather, the inventive concepts encompass any processor implemented automatic collimation algorithm, such as software-driven algorithms.

[0124] Further, in the examples given above, the collimator includes a shutter and wedge to achieve collimation. Again, however, the inventive concepts are not limited in this manner. For example, other collimations mechanisms may instead be adopted such as multi-leaf collimators or more futuristic collimators like liquid or gas collimators.

[0125] Other variations to the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed invention, from a study of the drawings, the disclosure, and the appended claims. While representative embodiments are disclosed herein, one of ordinary skill in the art will appreciate that many variations that are in accordance with the present teachings are possible and remain within the scope of the appended claim set. The invention therefore is not to be restricted except within the scope of the appended claims.

Claims

WHAT IS CLAIMED IS:

1. A system for controlling collimation, the system comprising: an automated collimation controller that includes at least one processor configured to: evaluate one or more images from an x-ray imaging system to identify features of interest, wherein the x-ray imaging system comprises a collimator controlled by the automated collimation controller, obtain at least one feature weight parameter or dose weight parameter of the automated collimation controller, determine a weighted combination of the identified features of interest based on the at least one feature weight parameter or dose weight parameter, and tune collimation settings based on the weighted combinations of the identified features of interest.

2. The system of claim 1, wherein the automatic collimation controller includes artificial intelligence (Al) models to tune the collimation settings.

3. The system of claim 1, further comprising a user interface for adjusting parameters of the automated collimation controller, wherein the user interface comprises a set of interactable sliders, each slider corresponding to a parameter of the automated collimation controller.

4. The system of claim 1, further comprising: the collimator that includes a shutter and a wedge, wherein the at least one feature weight parameter or dose weight parameter includes at least one of a shutter feature weight, a shutter dose weight, a wedge feature weight, or a wedge dose weight.

5. The system of claim 3, wherein the automatic collimation controller is further configured to adapt the collimation settings based on at least one of user feedback provided via interaction with the user interface or eye tracking data indicating a gaze of the user on regions of the image.

6. The system of claim 1, wherein the automatic collimation controller is further configured to run in the background and provide collimation recommendations when manual settings of the user are within a predetermined range.

7. The system of claim 1, wherein the at least one processor is further configured to adjust the at least one feature weight parameter or dose weight parameter based on a known phase of a procedure being performed.

8. The system of claim 1, wherein the automatic collimation controller is further configured to tune the collimation settings based on the presence or absence of interventional devices in the image.

9. The system of claim 1, wherein the at least one processor is further configured to: extract a plurality of features from an image of the one or more images, the plurality of features corresponding to relevance of displaying each part of the image to a user; generate a pixel-level map indicating relevance of revealing each pixel of the image to the user based on the extracted features; and adjust the collimation settings based on the pixel-level map and user preferences.

10. The system of claim 9, wherein the at least one processor is further configured to use a convolutional neural network to extract the plurality of features from the image and generate a heatmap indicating the relevance of revealing each pixel to the user, wherein the convolutional neural network is trained using a dataset of interventional images annotated to highlight relevance of different anatomical features.

11. The system of claim 9, wherein to adjust the collimation settings, the at least one processor is further configured to optimize an objective function that balances the relevance of revealing each pixel to the user and the reduction of radiation dose.

12. The system of claim 1, wherein the at least one processor is further configured to calibrate the x-ray imaging system by: presenting a user with a set of representative interventional images, receiving a user input indicating preferred collimation settings for each representative interventional image, determining parameters of the automated collimation controller that would result in the preferred collimation settings; and storing the determined parameters for use in subsequent interventional procedures.

13. A method for controlling collimation, the method comprising: evaluating one or more images from an x-ray imaging system to identify features of interest, wherein the x-ray imaging system comprises a collimator controlled by the automated collimation controller; obtaining at least one feature weight parameter or dose weight parameter of the automated collimation controller; determining a weighted combination of the identified features of interest based on the at least one feature weight parameter or dose weight parameter; and tuning collimation settings based on the weighted combinations of the identified features of interest.

14. The method of claim 13, wherein the automatic collimation controller includes artificial intelligence (Al) models to tune the collimation settings.

15. The method of claim 13, wherein parameters of the automated collimation controller are adjustable by a user interface, wherein the user interface comprises a set of interactable sliders, each slider corresponding to a parameter of the automated collimation controller.

16. The method of claim 13, wherein the collimator that includes a shutter and a wedge, andwherein the at least one feature weight parameter or dose weight parameter includes at least one of a shutter feature weight, a shutter dose weight, a wedge feature weight, or a wedge dose weight.

17. A non-transitory computer-readable storage medium having stored a computer program comprising instructions, which, when executed by a processor of an automated collimation controller, cause the processor to: evaluate one or more images from an x-ray imaging system to identify features of interest, wherein the x-ray imaging system comprises a collimator controlled by the automated collimation controller; obtain at least one feature weight parameter or dose weight parameter of the automated collimation controller; determine a weighted combination of the identified features of interest based on the at least one feature weight parameter or dose weight parameter; and tune collimation settings based on the weighted combinations of the identified features of interest.

18. The non-transitory computer-readable storage medium of claim 17, wherein the automatic collimation controller includes artificial intelligence (Al) models to tune the collimation settings.

19. The non-transitory computer-readable storage medium of claim 17, wherein parameters of the automated collimation controller are adjustable by a user interface, wherein the user interface comprises a set of interactable sliders, each slider corresponding to a parameter of the automated collimation controller.

20. The non-transitory computer-readable storage medium of claim 19, wherein the collimator that includes a shutter and a wedge, and wherein the at least one feature weight parameter or dose weight parameter includes at least one of a shutter feature weight, a shutter dose weight, a wedge feature weight, or a wedge dose weight.

Images