Devices at imaging points to integrate AI algorithm training into clinical workflows.

Abstract

The present invention relates to image processing. To enable a sustainable infrastructure for training AI algorithms, an imaging system using a mobile annotation device is proposed for receiving images collected by a medical imaging device (e.g., an X-ray, CT, or MRI scanner) in real time, i.e., during an imaging session. The images collected by the medical imaging device are then displayed, allowing a user to annotate the acquired images. The user annotations may include one or more of a recommended workflow for the patient, image quality instructions for the first image, instructions for medical findings, and priority information representing the urgency of the medical findings. The collected images and user annotations are then stored in a database, thereby creating a training database for training AI algorithms. Alternatively or additionally, a user interface receives user annotations regarding a set of pre-image settings to be used by the medical imaging device to collect images of a patient. In an example of X-ray chest imaging, the user annotations may be collimation settings, exposure time settings, tube voltage settings, focal spot size settings, selection of an X-ray sensitive area for the X-ray imaging system to apply the correct dose to the patient, etc. The set of pre-image settings and user annotations are then stored in a training database, thereby creating a training database for training AI algorithms. In this way, images and sets of pre-image settings can be selected directly from the clinical workflow, without the need to select images and / or sets of pre-image settings and transfer them elsewhere (e.g., from another facility) for development. Thus, the parameters of AI algorithms trained using training data from the training database can be adapted to fit the needs and specifications of a particular facility, thereby resulting in a sustainable architecture that can be used to train AI algorithms for different applications based on customer needs.

Description

Technical Field

[0001] The present invention generally relates to image processing, and more particularly to an imaging system, a method of image processing, a computer program element, and a computer-readable medium.

Background Art

[0002] Previously, most of the medical imaging equipment was operated by specialized operators such as radiographers (X-ray, CT or MRI), sonographers (ultrasound), or nuclear medicine technologists (NM imaging). However, there is a new trend emerging where less qualified staff are being tasked with performing the examinations. This practice, without safety measures, can lead to a decline in clinical quality.

[0003] (referred to herein as the "user") The operator is responsible for performing a set of work steps throughout the examination, for example, depending on the modality and equipment specifications, (i) arranging the patient, (ii) adapting the parameters of the imaging scan, (iii) performing the acquisition itself, (iv) reviewing and post-processing the obtained images on the console of the imaging device.

Summary of the Invention

Problems to be Solved by the Invention

[0004] When the imaging examination is completed, the subsequent steps in the modern radiology workflow generally involve the operator electronically sending the images to an image database (PACS) for storage and simultaneously electronically sending the images for interpretation to another trained specialist (a radiographer with medical qualifications) via a reading-worklist. Depending on several factors such as the medical urgency of the condition and the institution-specific scheduling of workload, this interpretation is often done asynchronously, meaning there is a significant time delay between image acquisition and image interpretation.

[0005] Artificial intelligence (AI) has the potential to improve clinical efficiency while compensating for the lack of qualified personnel. AI systems are computer-implemented systems. They are based on machine learning algorithms that have been pre-trained on training data to perform tasks such as assisting users during examinations. Training machine learning algorithms requires creating training data for the development of AI algorithms, which implies significant effort in the development process. Furthermore, AI algorithms trained in different facilities do not always fit the needs and standards of different facilities, making it extremely difficult to obtain a sustainable architecture that can be used to train AI algorithms for different application cases based on customer needs.

[0006] A system and method are needed to address at least some of the aforementioned deficiencies. [Means for solving the problem]

[0007] The object of the present invention is solved by the subject matter of the independent claims, and further embodiments are incorporated into the dependent claims. It should be noted that the embodiments of the present invention described below also apply to imaging systems, image processing methods, computer program elements, and computer-readable media.

[0008] According to a first aspect of the present invention, A medical imaging device for using a set of pre-image settings to collect patient images during an imaging session, (i) a user interface for receiving user annotations relating to the collected patient images and / or (ii) a set of pre-image settings used by a medical imaging device for collecting patient images, (i) collected patient images and received user annotations, and / or (ii) at least one training database for storing a set of pre-image settings and received user annotations, A training module for training at least one data-driven model using training data obtained from at least one training database, and An imaging system is provided that includes the following features.

[0009] AI has demonstrated great potential for improving clinical workflows. Therefore, there are many attempts to develop AI algorithms that can be used in clinical workflows. However, training these algorithms is extremely difficult because it requires well-annotated clinical data. Generally, this implies that considerable effort must be put into finding and annotating suitable data.

[0010] To enable a sustainable infrastructure for training AI algorithms, an imaging system is proposed that includes a medical imaging system (e.g., X-ray, CT, or MRI scanner) for collecting patient images. The images collected by the medical imaging device are then displayed, and the user has the possibility of annotating the images via a user interface.

[0011] In one example, the user interface receives user annotations on patient images acquired by a medical imaging device (e.g., an X-ray, CT, or MRI scanner). Examples of user annotations include, but are not limited to, indications of image quality and clinical findings. The user interface is configured to receive the acquired patient images in real time, for example, immediately after image acquisition in the imaging system. The acquired images and user annotations are then stored in a training database, thereby creating a training database for training an AI algorithm.

[0012] Alternatively or additionally, the user interface receives user annotations regarding a set of pre-image settings used by a medical imaging device to acquire patient images. In the case of X-ray chest imaging, user annotations may include collimation settings, exposure time settings, tube voltage settings, focal spot size settings, and selection of X-ray sensitive areas for the X-ray imaging system to apply the correct dose to the patient. The set of pre-image settings and user annotations are then stored in a training database, thereby creating a training database for training AI algorithms.

[0013] In this way, sets of images and pre-image settings can be selected directly from the clinical workflow, eliminating the need to select sets of images and / or pre-image settings and transfer them to another location (e.g., from another facility) for development. Therefore, the parameters of the AI ​​algorithm trained using training data from the training database can be adapted to the specific needs and standards of the facility, thereby making it possible to obtain a sustainable architecture that can be used to train the AI ​​algorithm for different application cases based on customer needs.

[0014] In one example, the training module is implemented in a processor circuit configured for parallel computing, such as a multi-core processor, GPU, or a part thereof. In another example, the training module is included in a system-on-a-chip (SoC) circuit.

[0015] For example, the user interface is part of a handheld device, which may include one or more of the following: a mobile phone, a laptop computing device, and a tablet computer.

[0016] In another example, the user interface is part of a medical imaging device.

[0017] For example, a medical imaging device, training module, and user interface may have wired connections (e.g., USB, coaxial, or optical cable) and / or wireless connections (e.g., Bluetooth, NFC, WLAN).

[0018] In one example, the network connects a medical imaging device, a training module, and a user interface in a communicative manner. The network could be the internet. Alternatively, the network could be any other type and number of networks. For example, the network could be implemented by several local area networks connected to a wide area network. For instance, the network could include any combination of wired networks, wireless networks, wide area networks, and local area networks.

[0019] According to one embodiment of the present invention, at least one data-driven model is A data-driven model for analyzing patient images collected to compute medical decision-making support information, A data-driven model for analyzing a patient's camera image to compute a set of pre-image settings, wherein the camera image is generated based on sensor data obtained from a sensor device, and the sensor device has a field of view that includes at least a portion of the area where the patient is positioned for imaging. Includes one or more of the following.

[0020] Medical decision support information includes, for example, recommended workflows for patients, image quality guidelines for collected images, guidelines for medical findings, and priority information indicating the urgency of medical findings.

[0021] The camera image can be in the form of a depth image or an RGB image. The camera image is collected, for example, when a patient is placed for an imaging examination by lying or standing within the field of view of the imaging system. The camera image does not necessarily have to include the entire body surface of the patient, and the camera image can relate only to a part of the patient's body surface relevant for the imaging examination. For example, if the anatomy in question is the patient's neck, only the measurement image of the upper body of the patient is captured by the sensor device.

[0022] According to one embodiment of the present invention, the medical imaging device comprises a user interface.

[0023] For example, the medical imaging device comprises a touch screen that enables a user to enter user annotations.

[0024] According to one embodiment of the present invention, the imaging system (i) an input channel for receiving a set of pre-image settings used by a medical imaging device for collecting an image of a patient and / or (ii) the collected image of the patient, and (i) a display for displaying a set of pre-image settings used by a medical imaging device for collecting an image of a patient and / or (ii) the collected image of the patient, and (i) a user interface for receiving user annotations regarding a set of pre-image settings used by a medical imaging device for collecting an image of a patient and / or (ii) the collected image of the patient, and an output channel for providing (i) the collected image of the patient and the received user annotations and / or (ii) the set of pre-image settings and the received user annotations to at least one training database and further comprises a mobile annotation device.

[0025] For example, the mobile annotation device is a handheld device including one or more of a mobile phone, a laptop computing device, and a tablet computer.

[0026] The mobile annotation device is positioned near the radiologist's reading monitor, as shown in Figure 4. Each time the radiologist opens an image, that image is further displayed on the mobile annotation device 14. On the mobile annotation device, the radiologist can annotate the image regarding, for example, image quality, clinical findings, etc.

[0027] According to one embodiment of the present invention, the training module repeatedly trains at least one data-driven model.

[0028] In one example, a data-driven model is trained on the fly with respect to new instances of images acquired by a medical imaging device. In other words, the training mode continues through repeated training phases during the deployment phase. For example, a training module for an image processing device is configured to train the data-driven model on the fly (i.e., during normal use of the data-driven model in the deployment phase) with respect to acquired patient images and user annotations provided by a mobile annotation device. For example, each newly acquired image and each user annotation provided by the mobile annotation device are sent directly to the training module to update the parameters of the data-driven model. In this way, the data-driven model is continuously trained to conform to the needs and standards of a particular facility.

[0029] In another example, a data-driven model is not trained on the fly, but rather after a certain number of images with annotations have been collected.

[0030] In a further example, if a user disagrees with the algorithm's feedback displayed on their mobile device, they may select a new annotation for that image. The database is then enriched not only by this image but also by all possible variations of that image (such as the same image but slightly cropped). This data enrichment step allows the algorithm to increase the weight of this image that was incorrectly annotated by the algorithm, as it is then retrained using the new image and all newly generated image variations.

[0031] According to one embodiment of the present invention, the training module randomly generates user annotations to begin training a data-driven model.

[0032] In other words, the initial model is a naive model, meaning the first image feedback is generated randomly. This means that the model can be trained from scratch.

[0033] According to one embodiment of the present invention, a data-driven model provides suggestions based on computed medical decision support information to enable the user to actively accept or reject the suggestions given.

[0034] In other words, the model only provides suggestions, in contrast to the possibility that the output user input could be several predefined tags (e.g., different diseases), and there may be further possibilities where the user needs to actively accept or reject those suggestions.

[0035] According to one embodiment of the present invention, the medical imaging system is A first group of medical imaging devices, and a second group of medical imaging devices that are different from the first group of medical imaging devices. Equipped with, The user interface is (i) images of patients collected by medical imaging devices in the first group, and / or (ii) a first user annotation relating to a set of pre-image settings used by medical imaging devices in the first group, (i) images of the patient collected by the medical imaging device in the second group, and / or (ii) a second user annotation relating to a set of pre-image settings used by the medical imaging device in the second group. signal, At least one training database (16) (i) patient images and received user annotations collected by medical imaging devices in the first group, and / or (ii) a set of pre-image settings and received user annotations used by medical imaging devices in the first group, (i) patient images and received user annotations collected by medical imaging devices in the second group, and / or (ii) a second training database for storing a set of pre-image settings and received user annotations used by medical imaging devices in the second group. Includes, The training module trains a first data-driven model using training data obtained from a first training database, and trains a second data-driven model using training data obtained from a second training database.

[0036] In this way, images and / or pre-image settings for training purposes are taken directly from the clinical workflow, and training can be performed according to the standards of each group (e.g., user group or facility). The proposed IT infrastructure provides a method for integrating AI algorithm training into the clinical workflow and making that training group-specific.

[0037] According to one embodiment of the present invention, the first group of medical imaging devices and the second group of medical imaging devices are from different facilities and / or different user groups.

[0038] According to one embodiment of the present invention, user annotations are Instructions for image quality, Clinical findings and, Instructions for setting the desired pre-image settings and Includes one or more of the following.

[0039] For example, instructions regarding image quality may include one or more of the following: a) patient placement, b) collimator settings, c) contrast, d) resolution, e) noise, and f) artifacts.

[0040] The term "clinical finding" is sometimes also called "medical finding."

[0041] For example, instructions for a set of desired pre-image settings for CT include collimation settings, exposure time settings, tube voltage settings, focal spot size settings, and / or selection of X-ray sensitive regions for an X-ray imaging system.

[0042] According to one embodiment of the present invention, medical decision support information is Recommended workflows for patients, Instructions regarding the image quality of the collected images, Instructions regarding medical findings, Priority information indicating the urgency of medical findings and This includes one or more of the following. For example, a recommended workflow is a variation of a previously defined workflow assumed for a patient.

[0043] According to one embodiment of the present invention, the mobile annotation device is a handheld device that includes one or more of a mobile phone, a laptop computing device, and a tablet computer.

[0044] According to a second aspect of the present invention, The medical imaging device uses a set of pre-image settings to collect images of the patient during the imaging session. The steps include receiving user annotations via a user interface regarding (i) collected patient images and / or (ii) a set of pre-image settings used by a medical imaging device for collecting patient images, The steps include storing in at least one training database (i) collected patient images and received user annotations, and / or (ii) a set of pre-image settings and received user annotations, The training module includes the step of training at least one data-driven model using training data obtained from at least one training database, and A method for image processing is provided, which has the following characteristics.

[0045] According to another aspect of the present invention, a computer program is provided which, when executed by at least one processing unit, causes an imaging system according to the first aspect and any related example to perform a method according to the second aspect and any related example.

[0046] According to a further aspect of the present invention, a computer-readable medium containing a computer program is provided.

[0047] To the advantage of the terms, the benefits provided by any of the above-mentioned aspects shall apply equally to all other aspects, and the benefits provided by all other aspects shall apply equally to any of the above-mentioned aspects.

[0048] As used herein, the term "user" refers to a medical person who is at least partially involved in an imaging procedure in a managerial or organizational manner.

[0049] As used herein, the term “patient” refers to the person being imaged, or, in a veterinary setting, an animal (especially a mammal).

[0050] As used herein, the term “machine learning” refers to the field of computer science that studies the design of computer programs that can extract patterns, regularities, or rules from past experience in order to produce appropriate responses to future data or to explain that data in some meaningful way.

[0051] As used herein, the term "learning" in the context of machine learning refers to the identification and training of a suitable algorithm for accomplishing the task in question. Learning, i.e., the performance of machine learning on a task measurable by performance metrics, generally improves with the use of training data.

[0052] As used herein, the term “data-driven model” in the context of machine learning refers to a suitable algorithm trained on appropriate training data. A neural network model is shown in the example, as will be explained later. However, other machine learning techniques, such as support vector machines and decision trees, may be used instead of neural networks. An exemplary data-driven model is shown in Figure 8.

[0053] As used herein, the term “module” means, is a part of, or includes, application-specific integrated circuits (ASICs), electronic circuits, processors (shared, dedicated, or group processors) and / or memories (shared, dedicated, or group memory) that run one or more software or firmware programs, combinational logic circuits, and / or other suitable components that provide the functions described herein.

[0054] It should be understood that all combinations of the above concepts and any additional concepts described in more detail below (provided that such concepts are not mutually contradictory) are intended to be part of the inventive subject matter disclosed herein. In particular, all combinations of claimed subject matter appearing at the end of this disclosure are intended to be part of the inventive subject matter disclosed herein.

[0055] These and other aspects of the present invention will become apparent from the embodiments described below, and will be clarified by referring to those embodiments.

[0056] In drawings, the same reference numerals across different drawings generally refer to the same part. Furthermore, drawings are not necessarily to a fixed scale; instead, the focus is generally on illustrating the principles of the invention. [Brief explanation of the drawing]

[0057] [Figure 1] This figure shows a schematic block diagram of an exemplary medical imaging system. [Figure 2] This figure shows a schematic block diagram of an exemplary imaging processing device. [Figure 3] This figure shows a schematic block diagram of an exemplary mobile annotation device. [Figure 4] This figure shows an example of implementing a mobile annotation device for radiologic technologists. [Figure 5] This figure shows a schematic block diagram of a further exemplary medical imaging system. [Figure 6] This figure shows a schematic block diagram of a further exemplary imaging processing device. [Figure 7] This diagram shows a flowchart illustrating an example of an image processing method. [Figure 8] This figure shows a schematic diagram of an exemplary data-driven model. [Modes for carrying out the invention]

[0058] The following describes in detail a method relating to a data-driven model for analyzing patient images collected to compute medical decision support information. While the following detailed description is illustrative for specific data-driven models, those skilled in the art should understand that the methods and imaging systems described above and below may be applicable to any other data-driven models, such as a data-driven model for analyzing patient camera images to compute a set of pre-image settings. Therefore, the examples described below are provided without any loss of generality of the claimed invention and without imposing any limitations on the claimed invention.

[0059] Figure 1 shows a schematic block diagram of an exemplary medical imaging system 100. The medical imaging system 100 comprises a medical imaging device 10, an image processing device 12, a mobile annotation device 14, a training database 16, image databases 18a and 18b, and a display configuration 20.

[0060] The medical imaging device 10 is configured to collect images 22 of the patient during an imaging session. The medical imaging device 10 may be of any modality, such as transmission imaging or radiography. Transmission imaging includes, for example, X-ray-based imaging performed using a CT scanner. Magnetic resonance imaging (MRI) and ultrasound imaging are also possible. Radiography includes PET / SPECT and other nuclear medicine modalities. During the imaging session, images 22 of the patient are collected. The images 20 are preferably in digital format to assist the physician in diagnosis.

[0061] The image processing device 12 includes an image analyzer 26 configured to apply a data-driven model to analyze patient images collected for calculating medical decision support information. In one example, the image processing device 14 is a mobile device, such as a mobile phone, laptop computer, or tablet computer, but is not limited to these. In another example, the image processing device 12 is a server providing computing services. In yet another example, the image processing device 12 is a workstation with a display configuration 18.

[0062] Figure 2 shows a schematic block diagram of an exemplary image processing device 12. In this example, the image processing device 12 comprises an input channel 24, an image analyzer 26, and an output channel 28.

[0063] Input channel 24 is configured to receive patient images 22 collected during an imaging session. Input channel 12 is implemented, in one example, as an Ethernet interface, a USB® interface, a wireless interface such as Wi-Fi® or Bluetooth®, or any equivalent data transfer interface that enables data transfer between input peripherals and the image analyzer 26. Furthermore, the input channels access data via a network, such as the Internet, or any combination of wired networks, wireless networks, wide area networks, local area networks, etc.

[0064] The image analyzer 26 of the image processing device 12 may be driven by artificial intelligence. In particular, the image analyzer 26 may be included as a pre-trained data-driven model. The image analyzer runs on the processing unit of the image processing device 12. The processing unit may include general-purpose circuitry and / or dedicated computing circuits such as a GPU, or it may be a dedicated core of a multi-core multiprocessor. Preferably, the processing unit is configured for parallel computing. This is particularly advantageous when the underlying machine learning model is a neural network such as a convolutional network. Such types of machine learning models can be efficiently performed by vector multiplication, matrix multiplication, or tensor multiplication. Such types of computation can be accelerated in a parallel computing infrastructure.

[0065] The decision-supporting information calculated by the image analyzer 26 may therefore include one or more of the following: image quality, medical findings, and / or related priority levels. Image quality analysis includes patient placement, collimator settings (if any), contrast, resolution, and assessment of image noise or artifacts. Some or all of these factors may be considered and represented as a single image quality score on a preferred metric, or each factor may be measured by separate scores on different metrics.

[0066] If the decision support information indicates that the image quality is acceptable, the image is sent to a different image database 18a.

[0067] If decision support information indicates that the image quality is insufficient, the image is sent to the image database 18b. Furthermore, when the image quality is considered optimal, suggestive graphical instructions are provided. For example, a suggestive "tick" symbol is displayed in an appropriate color scheme, such as green. If the image quality is found to be insufficient, this is also indicated on the onboard display with a suggestive symbol, such as a red cross.

[0068] If a medical condition is identified, this will be indicated by a suitable text symbol or other symbol on the mobile display device's onboard display. Furthermore, or alternatively, if a medical condition is identified, the relevant workflow will be suggested. This suggested workflow may differ from the currently assigned plan. Additionally, a retake may be suggested, optionally along with suggestions for updating imaging parameters. In this case, the user may accept the retake using the user interface, adjust the imaging parameters, and / or initiate the image retake; a suitably formatted message will be sent to the operating console.

[0069] Preferably, the calculation of decision support information is performed in a two-stage sequential processing flow. In the first stage, image quality is established. If the image quality is found to be sufficient, then the imagery is analyzed for medical findings and / or workflow suggestions. The workflow calculated based on the analyzed images may differ from the workflow initially associated with the patient at check-in, for example. This change in workflow may be necessary, for example, if an unexpected medical condition that was not previously anticipated by the original workflow is detected in the images. For example, if a patient should be treated for cancer in a certain organ, such as the liver, a certain workflow is assumed. However, if analysis of a copy of the images incidentally reveals that the patient actually has pneumonia, the workflow needs to be changed to treat the pneumonia first before proceeding with cancer treatment.

[0070] The image processing device 12 may further include a training module 30 for training a data-driven module. Training data is collected from a training database 16 to train the data-driven module.

[0071] The output channel 28 is configured to provide decision support information to, for example, a display configuration 20 (e.g., a built-in screen, a connected monitor, or a projector), or to file storage (e.g., a hard drive or a solid-state drive), such as image databases 18a, 18b.

[0072] Figure 3 shows a schematic block diagram of an exemplary mobile annotation device 14. The mobile annotation device 14 includes any type of wireless device, such as a home electronics device, smartphone, tablet personal computer, wearable computing device, personal digital assistant (PDA), laptop computer, and / or any other similar physical computing device.

[0073] In the example in Figure 3, a single mobile annotation device 14 is shown. However, it should be understood that there may be multiple mobile annotation devices 14 to receive multiple user annotations from different users and provide the collected patient images along with multiple user annotations to a training database for training a data-driven model. In other words, there may be the possibility that two or more users add their user annotations to the training database. This provides comprehensive annotation of the collected images and minimizes opinion bias. In this way, the quality of the training data is improved.

[0074] The mobile annotation device 14 includes an input channel 14a, a display 14b, a user interface 14c, and an output channel 14d.

[0075] Input channel 14a is configured to receive acquired patient images. Input channel 14a can be a universal interface, providing interoperability with a range of different medical imaging devices, even those of different modalities. In one example, input channel 14a is configured to receive acquired images based on direct imaging of the displayed image ("image of the image"). In another example, if the imaging device does not have NFC or Bluetooth, input channel 14a is configured as NFC or Bluetooth. Other examples include LAN, WLAN, etc.

[0076] The display 14b is configured to display the collected patient images. When the collected images are displayed on the mobile annotation device 14, the user has the option to annotate the images.

[0077] User interface 14c is configured to receive user annotations regarding collected patient images. The term “user interface” refers to an interface between a user and a device that enables communication between a human user or operator and one or more devices. Further examples of user interfaces that may be employed in various implementations of this disclosure include, but are not limited to, switches, potentiometers, buttons, dials, sliders, trackballs, display screens, various types of graphical user interfaces (GUIs), touchscreens, microphones, and other types of sensors that receive and respond to any form of human-generated stimuli.

[0078] In one example, a user may annotate an image using predefined parameters, such as predefined tags (for example, for different diseases). In another example, a user uses a mobile annotation device 14, such as a web-based thin client, to access personalized content for annotating images. The mobile annotation device 14 may be used with the thin client as an app for accessing content for annotating images. The content may be stored in a database. The content can be customized, for example, by an administrator. For example, content for annotating images may include a list of image quality levels to be selected, such as "optimal," "suboptimal," and "poor," and a list of clinical findings related to a specific disease in a patient with, for example, chronic complex regional pain syndrome.

[0079] The nature of "annotations" is explained using the example of image quality for knee imaging. Whenever the user considers an image to be good, the user selects the "optimal" option on the mobile annotation device 14. If the image quality / patient placement is not optimal, the user selects "suboptimal" on the mobile annotation device 14. The input is then added to the training database 16 for the data-driven model. Alternatively or additionally, user annotations may include clinical findings. For example, if a radiologist notices pneumothorax on an image, the user selects "pneumothorax" on the mobile annotation device 14, and the image is automatically added to the corresponding training database 16 for the data-driven model. This workflow has the advantage that the radiologist can choose which images the user wants to add, meaning the user can exclude images for which they are unsure about the findings.

[0080] In another example, a data-driven model only provides suggestions, in contrast to the possibility that the output user input could be several predefined tags (for example, for different diseases), and the user must actively accept or reject those suggestions.

[0081] For example, to implement this tool for a radiologic technologist, the mobile annotation device 14 is positioned near the radiologic technologist's reading monitor 18, as shown in Figure 4. Each time the radiologic technologist opens an image, that image is further displayed on the mobile annotation device 14. On the mobile annotation device 14, the radiologic technologist can annotate the image with respect to, for example, image quality, clinical findings, etc.

[0082] Returning to Figure 3, the output channel 14d is configured to provide the collected patient images and user annotations to the training database 16 for training a data-driven model.

[0083] Once well-annotated images are collected in this manner, data-driven models can be used, for example, to automatically evaluate the image quality and / or clinical findings of the newly collected images.

[0084] Since images for training purposes are taken directly from the clinical workflow, training can be performed according to institutional standards. Institutions may determine who has sufficient experience to annotate the data. The proposed IT infrastructure provides a method for integrating AI algorithm training into the clinical workflow and making that training user-specific.

[0085] As shown in Figure 2, the training module 30 is configured to train a data-driven model using the following steps: receiving training data and applying the data-driven model to the training data one or more times. As a result of this application, a pre-trained model is obtained, which can then be used in deployment. In deployment, this new data can be applied to the pre-trained model to obtain desired decision support information about new data, for example, new images that are not from the training set.

[0086] Data-driven models can operate in two modes: "training mode / phase" and "deployment mode / phase". In training mode, the initial model of the data-driven model is trained based on a set of training data to generate a trained data-driven model. In deployment mode, the pre-trained data-driven model is supplied with newly acquired image data that has not been used for training, so that it can operate during normal use.

[0087] In a further example, a data-driven model can operate in hybrid mode, where the system operates in a pre-trained manner, but the user can "overrule" the system if they see an image for which they do not agree with the feedback. For example, as shown in Figure 1, a mobile annotation device 14 receives and displays medical decision support information calculated by an image processing device 12. If the user does not agree with the calculated medical decision support information, the user provides feedback to overrule the system.

[0088] The initial model of the data-driven model is a naive model, i.e., a model in which the first image feedback is generated randomly. In other words, the training module is configured to generate user annotations randomly in order to begin training the data-driven model, and the data-driven model is trained from scratch.

[0089] Preferably, the training mode is continued during repeated training phases in the deployment phase.

[0090] In one example, the training module 30 of the image processing device 12 is configured to train the data-driven model on the fly (i.e., during normal use of the data-driven model) with respect to each newly acquired image of a patient and the user annotations provided by the mobile annotation device 14. For example, each newly acquired image and each user annotation provided by the mobile annotation device 14 are sent directly to the training module to update the parameters of the data-driven model. In this way, the data-driven model is continuously trained so that its parameters are continuously adapted to meet the needs and standards of a particular facility.

[0091] In another example, a data-driven model is not trained on the fly, but rather after a certain number of images with annotations have been collected.

[0092] Figure 5 shows a further example of the medical imaging system 100. In this example, the medical imaging system 100 comprises a first group 50a of medical imaging devices and a second group 50b of medical imaging devices that are different from the first group 50a of medical imaging devices.

[0093] In one example, this represents different facilities from the first and second groups of medical imaging devices.

[0094] In one example, this represents a different user group from the first and second groups of medical imaging devices.

[0095] Each group includes one or more medical imaging devices. In the example shown in Figure 5, the first group of medical imaging devices 50a includes three medical imaging devices, and the second group of medical imaging devices 50b includes two medical imaging devices.

[0096] The image analyzer 26 of the image processing device 12 is configured to apply a first data-driven model to analyze images collected by a first group of medical imaging devices and to apply a second data-driven model to analyze images collected by a second group of medical imaging devices.

[0097] Figure 6 shows a schematic block diagram of an exemplary imaging processing device 12. The image analyzer 26 is configured to apply a first data-driven model to analyze images acquired by a first group 50a of the medical imaging device. The image analyzer 26 is configured to apply a second data-driven model to analyze images acquired by a second group 50b of the medical imaging device.

[0098] For example, the first data-driven model and the second data-driven model use the same neural network, such as a CNN.

[0099] In another example, the first data-driven model and the second data-driven model use different neural networks, such as CNNs and recurrent NNs.

[0100] In a further example, the first data-driven model uses a neural network, while the second data-driven model uses other machine learning techniques such as support vector machines and decision trees.

[0101] As shown in Figure 5, the training data for the first group 50a of medical imaging devices and the training data for the second group 50b of medical imaging devices are collected differently. In particular, the mobile annotation device 14 is configured to provide images collected by the first group 50a of medical imaging devices, along with user annotations, to the first training database 16a for training the first data-driven model. The mobile image annotation device 14 is configured to provide images collected by the second group 50b of medical imaging devices, along with user annotations, to the second training database 16b for training the second data-driven model. The second training database 16b is different from the first training database 16a.

[0102] In one example, the metadata for each collected image includes a group identifier. The mobile annotation device 14 sends the collected images and user annotations to the corresponding training databases 16a and 6b according to the group identifier.

[0103] In another example, the metadata for each collected image includes an identifier for the medical imaging device. The mobile annotation device 14 checks a lookup table using the identifier for the medical imaging device to find the corresponding training databases 16a, 16b.

[0104] In this way, images for training purposes can be taken directly from each group's clinical workflow, and training can be performed according to the standards of each group (e.g., user group or facility). The proposed IT infrastructure provides a method for integrating AI algorithm training into clinical workflows and making that training group-specific.

[0105] The exemplary imaging systems in Figures 1 to 6 show the user interface within the mobile annotation device 14. Alternatively or additionally (not shown), the user interface may be part of the mobile imaging system. For example, a medical imaging device (e.g., X-ray, MRI, CT, etc.) may include a display with a touchscreen configured to receive user annotations.

[0106] Next, referring to Figure 7, which shows a flowchart of the image processing method for the system described above. However, the method described below is not necessarily related to the system described above, and it will be seen that the following method can therefore be understood as teaching in itself.

[0107] In step 210, the medical imaging device collects images of the patient during the imaging session.

[0108] In step 220, the user interface receives user annotations regarding the collected patient images. Alternatively or additionally, the user interface receives a set of pre-image settings used by the medical imaging device for collecting the patient images.

[0109] In step 230, the collected patient images and received user annotations are stored in at least one training database. In one example, the user annotations include image quality indications. In another example, the user annotations include clinical findings.

[0110] Alternatively or additionally, a set of pre-image settings and received user annotations are stored in at least one training database.

[0111] In step 240, the training module trains at least one data-driven model using training data obtained from at least one training database.

[0112] In one example, the initial model of a data-driven model is trained in training mode based on a set of training data to generate a trained data-driven model. In deployment mode, the pre-trained data-driven model is fed newly acquired, untrained image data so that it can operate during normal use.

[0113] In another example, a data-driven model can operate in hybrid mode, where the system operates in an already trained manner, but the user can “reject” the system if they see an image for which they do not agree with the feedback. For example, as shown in Figure 1, a mobile annotation device 14 receives and displays medical decision support information calculated by an image processing device 12. If the user does not agree with the calculated medical decision support information, the user provides feedback to reject the system.

[0114] Optionally, there is a first group of medical imaging devices and a second group of medical imaging devices distinct from the first group. Each group includes one or more medical imaging devices. The first group and the second group of medical imaging devices may come from different facilities and / or different user groups.

[0115] Method 200 may further include the step of providing images collected by a first group of medical imaging devices, along with user annotations, to a first training database for training a first data-driven model for analyzing images collected by a first group of medical imaging devices, using a mobile annotation device.

[0116] Method 200 may further include the step of providing images collected by a second group of medical imaging devices and user annotations to a second training database for training a first data-driven model for analyzing images collected by a first group of medical imaging devices, using a mobile annotation device. The second training database is different from the first training database.

[0117] In this way, images for training purposes can be taken directly from the clinical workflow, and training can be performed according to the standards of each group (e.g., user group or facility). The proposed IT infrastructure provides a method for integrating AI algorithm training into the clinical workflow and making that training group-specific.

[0118] Furthermore, unless otherwise expressly stated, in any method claimed herein that includes two or more steps or actions, the order of the steps or actions of that method is not necessarily limited to the order in which the steps or actions of that method are presented.

[0119] Referring to Figure 8, a neural network model is shown as an exemplary data-driven model. However, other machine learning techniques, such as support vector machines and decision trees, can be used instead of neural networks. Nevertheless, neural networks, especially convolutional networks, have been shown to have particular advantages, particularly with respect to image data.

[0120] More specifically, Figure 8 is a schematic diagram of a convolutional neural network (CNN). A well-configured NN obtained after training (described more fully below) can be thought of as an approximate representation of a latent mapping between two spaces: the image and one or more of the spaces of image quality metrics, medical findings, and treatment plans. These spaces can be represented as points in a potentially high-dimensional space, such as an image being an N×N matrix where N is the number of pixels. Image quality metrics, medical findings, and treatment plans can similarly be encoded as vectors, matrices, or tensors. For example, a workflow can be implemented as a matrix or vector structure where each entry represents a workflow step. The learning task can be one or more of classification and / or regression. The image input space may include a 4D matrix to represent a time series of matrices, and thus a video sequence.

[0121] A suitable trained machine learning model or component attempts to approximate this mapping. This approximation is achieved in a learning or training process where parameters, which themselves form a high-dimensional space, are adjusted using an optimization method based on the training data.

[0122] More specifically, the machine learning components can be implemented as neural networks ("NN"), particularly convolutional neural networks ("CNN"). Referring again to Figure 11, this illustrates in more detail the CNN architecture assumed herein in some embodiments.

[0123] CNNs can operate in two modes: "training mode / phase" and "deployment mode / phase". In training mode, the initial model of the CNN is trained on a set of training data to generate a trained CNN model. In deployment mode, the pre-trained CNN model is fed new, untrained data so that it can operate during normal use. Training mode is a one-off operation, or this is continued in repeated training phases to improve performance. Everything described so far regarding the two modes is applicable to any kind of machine learning algorithm and is not limited to CNNs, or even NNs, for that matter.

[0124] A CNN contains a set of interconnected nodes organized in layers. A CNN includes an output layer (OL) and an input layer (IL). The input layer (IL) may be a matrix with a size matching the size (rows and columns) of the training input images. The output layer (OL) may be a vector or matrix with a size matching the sizes selected for image quality metrics, medical findings, and treatment planning.

[0125] A CNN preferably has a deep learning architecture, i.e., there is at least one, preferably two or more, hidden layers between the OL and IL. The hidden layers include one or more convolutional layers CL1, Cl2 ("CL") and / or one or more pooling layers PL1, PL2 ("PL") and / or one or more fully connected layers FL1, FL2 ("FL"). CL is not fully connected, and / or the connection from CL to the next layer may vary, but is generally fixed in FL.

[0126] Each node is associated with a number called a "weight," which represents how the node responds to inputs from previous nodes in the preceding layer.

[0127] The set of all weights defines the structure of the CNN. During the training phase, the initial structure is adjusted based on the training data using a learning algorithm, such as forward-backward ("FB") propagation, other optimization methods, or other gradient descent methods. The gradient is taken over the parameters of the object function.

[0128] The training mode is preferably supervised, i.e., based on annotated training data. For each pair, one item is training input data, and the other item is target training data that is known a priori to be correctly associated with the training input data item of that pair. This association is defined and preferably given by a human expert. The training pair includes images as training input data, associated with each training image, and is a target label for one or more of the following: image quality indication, indication of the medical finding represented by the image, priority level indication, or indication of the workflow step required for a given image.

[0129] As described above, in this disclosure, annotated training data includes pairs or training data items taken directly from the clinical workflow. Therefore, training may be performed according to institutional standards. The institution may determine which individuals have sufficient experience to annotate the data. The IT infrastructure proposed above provides a method for integrating AI algorithm training into the clinical workflow and making that training user-specific. Optionally, training may be performed according to user group standards. The proposed IT infrastructure provides a method for integrating AI algorithm training into the clinical workflow and making that training group-specific.

[0130] In training mode, preferably, multiple such pairs are applied to the input layer and propagated through the CNN until the output appears in the OL. Initially, the output is generally different from the target. During optimization, the initial configuration is readjusted to achieve a good match between the input training data and their respective targets for all pairs. The match is measured by a similarity measure, which can be constructed with respect to an object function or a cost function. The goal is to tune the parameters to result in low cost, i.e., a good match.

[0131] More specifically, in a NN model, input training data items are applied to the input layer (IL), passed through a cascaded group of convolutional layers CL1, CL2, and possibly one or more pooling layers PL1, PL2, and finally through one or more fully connected layers. The convolutional modules are responsible for feature-based learning (e.g., identifying features in patient characteristics and contextual data), while the fully connected layers are responsible for more abstract learning, such as the impact of features on treatment. The output layer OL contains output data representing estimates for each target.

[0132] The strict grouping and ordering of layers shown in Figure 8 is merely an illustrative embodiment, and other groupings and orderings of layers are conceivable in different embodiments. Furthermore, the number of each type of layer (i.e., any one of CL, FL, or PL) may differ from the configuration shown in Figure 8. The depth of the CNN may also differ from the depth shown in Figure 8. All of the above applies equally to other neural networks conceivable herein, such as fully connected typical perceptron-type NNs and recurrent NNs, whether deep or not.

[0133] The annotated (labeled) training data assumed in this specification may need to be reformatted into a structured form. As previously mentioned, annotated training data may consist of vectors, matrices, or tensors (arrays of more than 2 dimensions). This reformatting is performed by a data preprocessor module (not shown), such as a scripting program or filter that thoroughly searches the patient records of the current facility's HIS to extract a set of patient characteristics.

[0134] The training dataset is applied to the initially configured CNN and then processed according to a learning algorithm, such as the FB propagation algorithm described above. At the end of the training phase, the thus pre-trained CNN is then used in the expansion phase to compute decision support information about new data, i.e., newly collected copy images that were not present in the training data.

[0135] Alternatively, the data-driven model is trained on the fly (i.e., during normal use of the data-driven model in the deployment phase) with respect to the collected patient images and user annotations provided by the mobile annotation device 14. For example, each newly collected image and each user annotation provided by the mobile annotation device 14 are sent directly to a training module to update the parameters of the data-driven model. In this way, the data-driven model is continuously trained to conform to the needs and standards of a particular facility.

[0136] Some or all of the steps described above can be implemented in hardware, software, or a combination thereof. Hardware implementations preferably include a programmed FPGA (Field-Programmable Gate Array) or hardwired IC chip. For good responsiveness and high throughput, and especially for neural networks, a multi-core processor such as a GPU or TPU may be used to perform the training and deployment of machine learning models as described above.

[0137] One or more features disclosed herein may be configured or implemented as / using circuits encoded in a computer-readable medium, and / or in combination thereof. Circuits include discrete circuits and / or integrated circuits, application-specific integrated circuits (ASICs), systems on a chip (SoCs), and combinations thereof, machines, computer systems, processors and memories, and computer programs.

[0138] Figures 1-8 illustrate a data-driven modeling method for analyzing patient images collected to compute medical decision-making support information. It can be seen that the method described above can be applied to any data-driven model.

[0139] For example, a data-driven model is a model for analyzing a patient's camera images to compute a set of pre-image settings. Such a data-driven model correlates one or more features in the camera image with a set of pre-image settings. In one example, a set of pre-image settings for an X-ray imaging system includes at least one of the following: tube voltage, tube current, grid, collimation window, and collimator geometry parameters. In a further example, a set of pre-image settings for a CT imaging system includes at least one of the following: power level, tube current, dose adjustment, scan planning parameters, and reconstruction parameters. Inputs to a data-driven model may include non-image patient data. Non-image patient data may include supplementary information to measurement images captured by sensor devices. For example, lung size is known to correlate with the patient's weight, age, and sex, and can be affected by several diseases such as COPD. By further adding non-patient image data obtained during the imaging procedure to the training set database, the data-driven model is trained to better model the relationship between the patient and the scanning configuration, such as collimation settings, which should be used in examination preparation and / or the imaging procedure. A data-driven model may include an artificial neural network and at least one of the following: a classification tree that uses at least one of Haar-like image features, scale-invariant feature transformation (SIFT) image features, and speed-up robust feature (SURF) image features.

[0140] In this exemplary data-driven model, the user interface 14c shown in Figure 1 is configured to receive user annotations regarding a set of pre-image settings used by a medical imaging device for acquiring patient images. The training database 16 is configured to store the set of pre-image settings and the received user annotations. The training module 30 is configured to train the exemplary data-driven model using training data obtained from at least one training database.

[0141] All definitions defined and used herein should be understood to govern dictionary definitions, definitions in documents incorporated by reference, and / or the ordinary meanings of the terms defined.

[0142] In use herein and in the claims, the singular element should be understood to mean “at least one” unless expressly otherwise.

[0143] As used herein and in the claims, the phrase “and / or” should be understood to mean “either or both” of the elements thus combined, that is, elements that may be present either conjunctly or disjunctly. Multiple elements listed using “and / or” should be interpreted in the same manner, that is, “one or more” of the elements thus combined. Other elements other than those specifically identified by the “and / or” clause may be present, whether related to or unrelated to those specifically identified elements.

[0144] As used herein and in the claims, “or” should be understood to have the same meaning as “and / or” as defined above. For example, when separating items in a list, “or” or “and / or” should be interpreted as inclusive, that is, including at least one of several elements or lists of elements, and optionally, any additional unlisted items. Only terms that explicitly state the opposite, such as “exactly one of” or “strictly one of” or “consisting of” as used in the claims, refer to including exactly one element of several elements or lists of elements. In general, as used herein, the term “or” should be interpreted only when preceded by a term of exclusivity, such as “either,” “one of,” “exactly one of,” or “strictly one of,” to indicate an exclusive choice (i.e., “one or the other, but not both”).

[0145] When used herein and in the claims, the phrase “at least one” with respect to a list of one or more elements should be understood to mean at least one element selected from any one or more of the elements in the list of elements, but not necessarily including at least one of each individual element specifically mentioned in the list of elements, and not excluding any combination of elements in the list of elements. This definition also allows for the optional presence of elements other than those specifically identified in the list of elements to which the phrase “at least one” refers, whether or not they are related to the specifically identified elements.

[0146] In the claims and the above-mentioned specification, all transitional phrases, such as “equipment,” “includes,” “carries,” “possesses,” “includes,” “accompanys,” “holds,” and “composed of,” should be understood to be open-ended, meaning “includes but not limited to.” Only the transitional phrases “consist of” and “essentially consist of” are closed or semi-closed transitional phrases, respectively.

[0147] Another exemplary embodiment of the present invention provides a computer program or computer program element characterized by being adapted to perform steps of a method according to one of the prior embodiments on a suitable system.

[0148] Computer program elements may, therefore, be stored on a computer unit, which may also be part of one embodiment of the present invention. This computing unit may be adapted to perform or induce the performance of the steps of the method described above. Furthermore, this computing unit may be adapted to operate the components of the apparatus described above. The computing unit may be adapted to operate automatically and / or to execute user instructions. The computer program may be loaded into the working memory of a data processor. The data processor may, therefore, be equipped to perform the method of the present invention.

[0149] This exemplary embodiment of the present invention covers both computer programs that use the present invention from the outset and computer programs that, through updates, make existing programs use the present invention.

[0150] Furthermore, a computer program element may provide all the necessary steps to carry out the procedure of an exemplary embodiment of the method described above.

[0151] According to a further exemplary embodiment of the present invention, a computer-readable medium such as a CD-ROM is presented, on which computer program elements are stored, the computer program elements being described in the previous section.

[0152] Computer programs may be stored and / or distributed on suitable media, such as optical or solid-state media supplied together with or as part of other hardware, but may also be distributed in other forms, such as via the Internet or other wired or wireless telecommunications systems.

[0153] However, computer programs can also be presented via networks such as the World Wide Web and downloaded from such networks into the working memory of a data processor. According to one further exemplary embodiment of the present invention, a medium is provided for making a computer program element available for download, and the computer program element is configured to perform a method according to one of the previously described embodiments of the present invention.

[0154] While several inventive embodiments have been described and shown herein, those skilled in the art will readily conceive of various other means and / or structures for performing the functions described herein and / or obtaining one or more of the results and / or advantages described herein, and each of such variations and / or modifications will be considered to fall within the scope of the inventive embodiments described herein. More generally, those skilled in the art will readily understand that all parameters, dimensions, materials, and configurations described herein are illustrative, and that actual parameters, dimensions, materials, and / or configurations depend on the specific one or more applications for which the inventive teachings are used. Those skilled in the art will be able to recognize many equivalents of the particular inventive embodiments described herein, or to verify those equivalents by simply using routine experimentation. Therefore, it should be understood that the embodiments described above are presented merely as examples, and that within the scope of the appended claims and their equivalents, the inventive embodiments may be carried out in ways other than those specifically described and claimed. The inventive embodiments of this disclosure cover each individual feature, system, article, material, kit, and / or method described herein. Furthermore, any combination of two or more such features, systems, articles, materials, kits, and / or methods is included within the inventive scope of this disclosure, provided that such features, systems, articles, materials, kits, and / or methods are not mutually inconsistent.

Claims

1. A medical imaging device for using a set of pre-image settings to collect patient images during an imaging session, (i) a user interface for receiving user annotations relating to the collected images of the patient and / or (ii) a set of pre-image settings used by the medical imaging device for collecting the images of the patient, and for creating annotated training data; (i) images of the patient collected and user annotations received, and / or (ii) a set of pre-image settings and user annotations received, for storing the created annotated training data, The system comprises a training module for training at least one data-driven model using training data obtained from the at least one training database, A first group of medical imaging devices, and a second group of medical imaging devices that are different from the first group of medical imaging devices. Furthermore, The aforementioned user interface (i) images of a patient collected by the medical imaging devices in the first group, and / or (ii) a first user annotation relating to a set of pre-image settings used by the medical imaging devices in the first group, (i) images of the patient collected by the medical imaging device in the second group, and / or (ii) a second user annotation relating to a set of pre-image settings used by the medical imaging device in the second group. signal, The aforementioned at least one training database, (i) images of the patient collected by the medical imaging devices in the first group and the user annotations received, and / or (ii) a set of pre-image settings and the user annotations used by the medical imaging devices in the first group; (i) images of the patient collected by the medical imaging devices in the second group and the user annotations received, and / or (ii) a set of pre-image settings used by the medical imaging devices in the second group and the user annotations received, in a second training database for storing these: Includes, An imaging system in which the training module trains a first data-driven model using training data obtained from the first training database, and trains a second data-driven model using training data obtained from the second training database.

2. The first group of medical imaging devices and the second group of medical imaging devices are from different facilities and / or different user groups. The imaging system according to claim 1.

3. The at least one data-driven model, A data-driven model for analyzing patient images collected to calculate medical decision-making support information, A data-driven model for analyzing the patient's camera image to calculate the set of pre-image settings, wherein the camera image is generated based on sensor data obtained from a sensor device, and the sensor device has a field of view that includes at least a portion of the area in which the patient is positioned for imaging. Including one or more of the following: The imaging system according to claim 1 or 2.

4. The medical imaging device is equipped with the user interface. The imaging system according to any one of claims 1 to 3.

5. (i) an input channel for receiving the collected images of the patient, and / or (ii) the set of pre-image settings used by the medical imaging device for collecting the images of the patient; (i) a display for displaying the collected images of the patient, and / or (ii) a set of pre-image settings used by the medical imaging device for collecting the images of the patient; (i) the collected images of the patient, and / or (ii) the user interface for receiving user annotations relating to the set of pre-image settings used by the medical imaging device for collecting the images of the patient, The at least one training database has an output channel for providing (i) the collected patient images and the received user annotations, and / or (ii) the set of pre-image settings and the received user annotations. Mobile annotation device The imaging system according to any one of claims 1 to 4, further comprising:

6. The aforementioned mobile annotation device is a handheld device that includes one or more of the following: a mobile phone, a laptop computing device, and a tablet computer. The imaging system according to claim 5.

7. The training module repeatedly trains the at least one data-driven model. The imaging system according to any one of claims 1 to 6.

8. The training module randomly generates the user annotations in order to begin training the at least one data-driven model. The imaging system according to any one of claims 1 to 7.

9. The at least one data-driven model provides suggestions based on the calculated medical decision support information, enabling the user to actively accept or reject the suggestions given. The imaging system according to claim 3.

10. The aforementioned user annotation, Instructions for image quality, Clinical findings and, Instructions for setting the desired pre-image settings and Including one or more of the following: The imaging system according to any one of claims 1 to 9.

11. The aforementioned medical decision-making support information, The recommended workflow for the aforementioned patient, Display of the image quality of the collected images, Instructions regarding medical findings, Priority information indicating the urgency of medical findings and Including one or more of the following: The imaging system according to claim 3.

12. The medical imaging device uses a set of pre-image settings to collect images of the patient during the imaging session. The steps include: receiving user annotations via a user interface regarding (i) images of the patient collected and / or (ii) a set of pre-image settings used by the medical imaging device for collecting images of the patient, and creating annotated training data; The steps include storing in at least one training database the created annotated training data, which includes (i) the collected patient images and the received user annotations, and / or (ii) the set of pre-image settings and the received user annotations; The training module includes the step of training at least one data-driven model using training data obtained from the at least one training database, Via the aforementioned user interface, (i) images of a patient collected by a medical imaging device in a first group of medical imaging devices, and / or (ii) a first user annotation relating to a set of pre-image settings used by the medical imaging devices in the first group, (i) images of a patient collected by a medical imaging device in a second group of medical imaging devices, which is different from the first group of medical imaging devices, and / or (ii) a second user annotation relating to a set of pre-image settings used by the medical imaging devices in the second group. The step of receiving, The aforementioned at least one training database, (i) images of the patient collected by the medical imaging devices in the first group and the user annotations received, and / or (ii) a set of pre-image settings and the user annotations used by the medical imaging devices in the first group; (i) images of the patient collected by the medical imaging devices in the second group and the user annotations received, and / or (ii) a set of pre-image settings used by the medical imaging devices in the second group and the user annotations received, in a second training database for storing these: Includes, An image processing method comprising the steps of the training module training a first data-driven model using training data obtained from the first training database, and training a second data-driven model using training data obtained from the second training database.

13. A computer program, when executed by at least one processing unit, that causes the imaging system according to any one of claims 1 to 11 to perform the method according to claim 12.

14. A computer-readable medium comprising the computer program described in claim 13.

Images