Simulator for converting supine three-dimensional geometry to standing position measurements

Abstract

Systems and methods for converting supine three-dimensional geometry into standing measurements. One example system may include an electronic processor. The electronic processor is configured to extract a scan of a spinal region in a supine position and process the scan using a machine learning model trained to generate standing spinal measurements from the supine spinal scan. The electronic processor is configured to receive multiple weight-bearing measurements of the spinal region from the machine learning model and generate a weight-bearing two-dimensional scan of the spinal region based on the multiple weight-bearing measurements. The electronic processor is configured to provide the supine spinal region scan and the two-dimensional scan of the spinal region as training data to a second machine learning model.

Description

Technical Field

[0001] This application pertains to the field of medical imaging systems. Background Technology

[0002] Predictive and recommendation models are used to plan and execute robot-assisted spinal surgeries. These models are trained using the patient's electronic medical records, X-ray images, computed tomography scans, and other planning information relevant to the surgical procedure to be performed. Attached Figure Description

[0003] The same reference numerals in the accompanying drawings denote the same or similarly functional elements in the various views, and are incorporated in and form part of the specification together with the following detailed description, and are also used to illustrate various embodiments, examples, aspects and features including the claimed subject matter, and to explain the various principles and advantages of the aspects of those embodiments, examples, aspects and features.

[0004] Figure 1 This is a block diagram illustrating various examples of medical computing systems.

[0005] Figure 2 Examples are given based on various examples. Figure 1 The flowchart of the system execution method.

[0006] Figure 3 Flowcharts illustrating various methods for training machine learning models to estimate spinal geometry are provided.

[0007] Those skilled in the art will understand that the elements in the accompanying drawings are illustrated for simplicity and clarity and are not necessarily drawn to scale. For example, the dimensions of some elements in the drawings may be enlarged relative to other elements to aid in understanding the illustrated examples, aspects, and features.

[0008] In some instances, device and method components have been indicated by conventional symbols in the accompanying drawings where appropriate, so as to show only those specific details relevant to understanding the various embodiments, examples, aspects and features, so as not to confuse the contents of this disclosure with details that are readily apparent to those skilled in the art and have the benefit of the description herein. Detailed Implementation

[0009] When planning and performing robot-assisted spinal surgery, surgeons use predictive and recommendation models. These models are trained using the patient's electronic medical records, X-ray images, computed tomography (CT) scans, and other planning information relevant to the surgery to be performed. These models are used to predict targets defined by the geometry of the spine in a standing position. Targets include adjacent segmental disease (ALD), proximal junctional kyphosis (PJK), and measures related to alignment with spinal geometry (e.g., disc height, disc angle, spondylolisthesis, and vertebral endplate orientation). It is also desirable that the planning and execution phases of spinal surgery (whether robot-assisted or freehand) be performed on a properly standing CT scan.

[0010] Two-dimensional X-rays (2D XR) are typically used to acquire geometric measurements of the spine in a standing position. The advantage of 2D XR is that it is usually performed in a standing position, which allows for a good representation of the patient's global alignment in a weight-bearing and balanced posture. However, 2D XR can be susceptible to the influence of unknown relative pose between the camera and the patient, which can reduce measurement accuracy. Another disadvantage of 2D XR is its limited accuracy when measuring precise spinal geometric parameters such as disc height, spondylolisthesis grade, vertebral endplate orientation, etc. Three-dimensional imaging (e.g., CT and MRI) scans can improve accuracy, but these scans are acquired in a supine position and lack relevant global alignment information (compared to the standing position).

[0011] To address these issues, the proposed implementation schemes and aspects provide machine learning models for converting supine three-dimensional spinal measurements into standing two-dimensional measurements. Among other capabilities, such models can be used to estimate accurate weight-bearing spinal geometry from supine CT scans. These models combine the advantages of both two-dimensional and three-dimensional modalities and define 3D-based criteria for predicting geometric complications.

[0012] The examples and aspects provided in this paper can be used to improve the performance of predictive spine analysis models. Predictive spine analysis models are trained on datasets consisting of preoperative and postoperative patient data. Such models are used to predict postoperative outcomes using preoperative and planning data as input. The examples provided in this paper can provide enhanced predictive measurement quality by converting measurements from a supine position to a weight-bearing position. Additionally, the predictive power of such models is limited by factors such as the quality of the input data. The examples provided in this paper can improve the input data by automatically converting between CT and X-ray measurements to fill in missing data (e.g., where the patient has both preoperative X-ray images and postoperative CT scans).

[0013] By using the examples and aspects presented in this paper, spinal surgical planning can be improved. For example, surgical planning can be based on simulated weight-bearing 3D scans generated from supine 3D scans to achieve desired alignment corrections. The execution of spinal surgeries can also be improved. By utilizing the examples presented in this paper, surgeons can verify weight-bearing alignment during surgery, enabling corrections to be made during the procedure if necessary.

[0014] By estimating accurate weight-bearing spinal geometry based on supine CT scan data, the model presented in this paper can also be used by medical imaging systems to generate simulated standing (weight-bearing) CT scans from supine CT scans, including but not limited to. As mentioned above, such simulated CT scans can be used for surgical planning and improvement of spinal analysis models for diagnostic purposes, etc.

[0015] In some aspects, the technology described herein relates to a medical system. This system includes an electronic processor. The electronic processor is configured to extract a scan of a spinal region in a supine position. The electronic processor is configured to process the scan using a machine learning model trained to generate standing spinal measurements from the supine spinal scan. The electronic processor is configured to receive multiple weight-bearing measurements of the spinal region from the machine learning model. The electronic processor is configured to generate a two-dimensional weight-bearing scan of the spinal region based on the multiple weight-bearing measurements of the spinal region. The electronic processor is configured to generate a surgical plan for the spinal region based on the two-dimensional weight-bearing scan of the spinal region.

[0016] In some aspects, the techniques described herein relate to a method for training a machine learning model to estimate the geometry of the spine in a weight-bearing position based on supine spinal X-ray image data. The method includes generating multiple three-dimensional standing images from multiple standing spinal X-ray images, wherein each of the multiple three-dimensional standing images originates from one of the multiple spinal X-ray images. The method includes associating each of the multiple supine spinal scans with one of the multiple three-dimensional standing images. The method includes feeding the multiple supine spinal scans as input data to the machine learning model. The method includes providing the machine learning model with the multiple three-dimensional standing images as ground truth data.

[0017] In some aspects, the technology described herein relates to a medical imaging system. The system includes a display and an electronic processor coupled to the display. The electronic processor is configured to extract a scan image of a spinal region in a supine position. The electronic processor is configured to process the scan using a machine learning model trained to generate standing spinal measurements from the supine spinal scan image. The electronic processor is configured to receive multiple weight-bearing measurements of the spinal region from the machine learning model. The electronic processor is configured to generate a two-dimensional weight-bearing scan image of the spinal region based on the multiple weight-bearing measurements of the spinal region. The electronic processor is configured to control the display to present the two-dimensional weight-bearing scan image of the spinal region.

[0018] In some aspects, the technology described herein relates to a medical system. This system includes an electronic processor. The electronic processor is configured to extract a scan image of a spinal region in a supine position. The electronic processor is configured to process the scan using a machine learning model trained to generate standing spinal measurements from the supine spinal scan image. The electronic processor is configured to receive multiple weight-bearing measurements of the spinal region from the machine learning model. The electronic processor is configured to generate a two-dimensional weight-bearing scan image of the spinal region based on the multiple weight-bearing measurements of the spinal region. The electronic processor is configured to provide the supine spinal scan image and the two-dimensional scan image of the spinal region as training data to a second machine learning model.

[0019] Specific embodiments of the present disclosure are now described with reference to the accompanying drawings, wherein the same reference numerals denote the same or functionally similar elements.

[0020] Before explaining any examples in detail, it should be understood that the examples presented herein are not limited in their application to the details of the construction and arrangement of the components illustrated in the following description or illustrated in the following figures. These examples are capable of having other embodiments and can be practiced or implemented in various ways. For ease of description, the example systems presented herein can be illustrated by a single example of each of their components. Some examples may not describe or illustrate all the components of the system. Other exemplary embodiments may include more or fewer of each of the illustrated components, may combine some components, or may include additional or alternative components.

[0021] Figure 1This is a block diagram of a medical system 100. The medical system 100 can be used to perform medical imaging, robot-assisted surgery, and other functions, including one or more aspects of one or more of the methods disclosed herein. The medical system 100 includes a computing device 102 communicatively coupled (e.g., via one or more wired or wireless connections) to one or more imaging devices 112, a robot 114, a navigation system 118, one or more sensors 126, a database 130, and / or a cloud (or another network) 134. Systems according to other embodiments of this disclosure may include more or fewer components than the medical system 100. The computing device 102 includes an electronic processor 104, a memory 106, a communication interface 108, and a user interface 110. In some aspects, the computing device 102 may include more or fewer components than illustrated in the examples.

[0022] The computing device 102 includes an electronic processor 104 (e.g., a microprocessor, an application-specific integrated circuit, etc.), a memory 106, a communication interface 108, and a user interface 110. The electronic processor 104, the memory 106, the communication interface 108, the user interface 110, and various other modules (not illustrated) are directly coupled via one or more control or data buses (e.g., bus 140) or combinations thereof.

[0023] Memory 106 may be constituted by one or more non-transitory computer-readable media and includes at least a program storage area and a data storage area. The program storage area and data storage area may include combinations of several types of memory, such as read-only memory (“ROM”), random access memory (“RAM”) (e.g., dynamic RAM (“DRAM”, synchronous DRAM (“SDRAM”), etc.), electrically erasable programmable read-only memory (“EEPROM”), flash memory, or any other suitable tangible non-transitory memory for storing computer-readable data and / or instructions. Memory 106 may store information or data that can be used to perform any aspect of the methods described herein or any other methods. Memory 106 may store instructions and / or machine learning models, for example, that support one or more functions of system 100. For example, memory 106 may store content (e.g., instructions and / or machine learning models) that enables image processing 120, sensor processing 122, and / or spinal geometry estimator 124 when executed by electronic processor 104. In some embodiments, such content may be organized into one or more applications, modules, packages, layers, or engines.

[0024] Image processing 120 enables electronic processor 104 to process image data (received from, for example, the camera of imaging device 112, navigation system 118, or any imaging device) of images in order to, for example, identify information about anatomical elements in the images.

[0025] In some aspects, image processing 120 enables electronic processor 104 to generate, analyze, manipulate, and otherwise process medical images, such as those generated by imaging device 112. For example, in some aspects, image processing 120 may enable system 100 to generate and / or process two-dimensional X-ray images, computed tomography (CT) scans, magnetic resonance imaging (MRI) scans, etc.

[0026] Sensor processing 122 enables processor 104 to process sensor output data (received from, for example, one or more sensors 126) for purposes such as determining the position of robotic arm 116 and / or surgical tool 128. Sensor outputs can be received as signals and can be processed by electronic processor 104 using sensor processing 122 to output data such as, for example, force data, acceleration data, pose data, time data, position data, etc.

[0027] As described herein, the spinal geometry estimator 124 enables the processor 104 to estimate accurate weight-bearing spinal geometry from supine imaging scans. As described more specifically herein, the spinal geometry estimator 124 implements a machine learning model. Machine learning generally refers to the ability of a computer program to learn without being explicitly programmed. In some implementations, the computer program (sometimes referred to as a learning engine) is configured to build a model (e.g., one or more algorithms) based on example inputs. Supervised learning involves providing the computer program with example inputs and their expected (actual) outputs. The computer program is configured to learn general rules (models) that map inputs in training data to outputs. Various types of methods and mechanisms can be used to perform machine learning. Example methods and mechanisms include decision tree learning (e.g., random forests), association rule learning, artificial neural networks, deep neural networks, inductive logic programming, support vector machines, clustering, Bayesian networks, reinforcement learning, representation learning, similarity and metric learning, sparse dictionary learning, and genetic algorithms. By using some or all of these methods, the computer program can ingest, parse, and understand data, and progressively refine models for data analysis, including image analysis. After training, the computer system can be referred to as an intelligent system, an artificial intelligence (AI) system, a cognitive system, etc. Alternatively or additionally, memory 106 may store other types of content or data (e.g., other types of machine learning models) that can be processed by electronic processor 104 to implement the various methods and features described herein.

[0028] Although the various contents of memory 106 can be described as instructions, it should be understood that the functionality described herein can be achieved using instructions, algorithms, and / or machine learning models. Data, algorithms, and / or instructions enable electronic processor 104 to manipulate data stored in memory 106 and / or received from or via imaging device 112, robot 114, database 130, sensor 126, and / or cloud 134.

[0029] Electronic processor 104 sends and receives information (e.g., from memory 106, communication interface 108, and / or user interface 110) and processes that information by executing one or more software instructions or modules that can be stored in memory 106 or another non-transitory computer-readable medium. The software may include firmware, one or more applications, program data, filters, rules, one or more program modules, and other executable instructions. Electronic processor 104 is configured to retrieve from memory 106 and execute software, etc., for performing the techniques and methods described herein, including image processing 120, sensor processing 122, and a spinal geometry estimator 124.

[0030] Communication interface 108 sends and receives information from devices outside computing device 102 (e.g., components of medical system 100). Communication interface 108 receives input (e.g., from user interface 110), provides system output, or a combination of both. Communication interface 108 can be used to receive image data or other information from external sources (such as imaging device 112, robot 114, navigation system 118, sensor 126, database 130, cloud 134, and / or any other system or component not part of medical system 100), and / or to transmit instructions, images, or other information to external systems or devices (e.g., another computing device, imaging device, robot, navigation system, sensor, database, cloud, and / or any other system or component not part of medical system 100).

[0031] Communication interface 108 may include one or more wired interfaces (e.g., USB port, Ethernet port, etc.) and / or one or more wireless transceivers or interfaces (configured to send and / or receive information, for example, via one or more wireless communication protocols such as 802.11 a / b / g / n, Bluetooth, NFC, ZigBee, etc.). In some embodiments, communication interface 108 may be used to enable computing device 102 to communicate with one or more other electronic processors or computing devices, whether to reduce the time required to complete computationally intensive tasks or for any other reason.

[0032] The computing device 102 may also include one or more user interfaces 110. User interface 110 may be or include a keyboard, mouse, trackball, monitor, television, screen, touchscreen, and / or any other device for receiving information from and / or providing information to a user. User interface 110 may be used, for example, to receive user selections or other user input regarding any step of any method described herein. Nevertheless, any required input for any step of any method described herein may be automatically generated by medical system 100 (e.g., by electronic processor 104 or another component of medical system 100) or received by medical system 100 from a source external to medical system 100. In some embodiments, user interface 110 may be used to allow a surgeon or other user to modify instructions to be executed by electronic processor 104, and / or modify or adjust settings displayed on or corresponding to other information on user interface 110, according to one or more embodiments of this disclosure.

[0033] Although user interface 110 is shown as part of computing device 102, in some embodiments, computing device 102 may utilize user interface 110 which is housed separately from one or more other components of computing device 102. In some embodiments, user interface 110 may be located close to one or more other components of computing device 102, while in other embodiments, user interface 110 may be located away from one or more other components of computing device 102.

[0034] It should be understood that, despite Figure 1 Only a single electronic processor 104, memory 106, communication interface 108, and user interface 110 are illustrated, but alternative embodiments of computing device 102 may include multiple electronic processors, memory modules, communication interfaces, and / or user interfaces. It should also be noted that medical system 100 may include other computing devices, each comprising components similar to and configured similarly to computing device 102. In some embodiments, portions of computing device 102 are implemented partially or entirely on semiconductor chips (e.g., application-specific integrated circuits (ASICs), field-programmable gate arrays (“FPGAs”), etc.). Similarly, the various modules and controllers described herein may be implemented as separate controllers, as illustrated, or as components of a single controller. In some aspects, combinations of methods may be used. Furthermore, any one of image processing 120, sensor processing 122, and / or spinal geometry estimator 124 may be implemented on a separate computing platform, but accessible and / or controlled by computing device 102.

[0035] Continuing with other aspects of the medical system 100, the imaging device 112 may be operable to image anatomical features (e.g., bones, veins, tissues, etc.) and / or other aspects of the patient's anatomy to generate image data (e.g., image data depicting or corresponding to bones, veins, tissues, etc.). As used herein, "image data" refers to data generated or captured by the imaging device 112, including data in machine-readable form, graphical / visual form, and any other form. In various examples, image data may include data corresponding to a portion or part of a patient's anatomical features. In some aspects, image data is stored in a Medical Digital Imaging and Communication (DICOM) format (e.g., stored in database 130, cloud 134, or both). Image data may be or include preoperative images, intraoperative images, postoperative images, or images taken independently of any surgical procedure. The imaging device 112 may be able to capture 2D images (e.g., X-ray images) or 3D images (e.g., CT scans or MRI scans) to generate image data. The imaging device 112 may be or include, for example, an ultrasound scanner (which may include, for example, physically separate transducers and receivers, or a single ultrasound transceiver), an O-arm, a C-arm, a G-arm, or any other device that utilizes X-ray-based imaging (e.g., a fluorescence microscope, a CT scanner, or other X-ray machine), a magnetic resonance imaging (MRI) scanner, an optical coherence tomography (OCT) scanner, an endoscope, a microscope, an optical camera, a thermal imaging camera (e.g., an infrared camera), a radar system (which may include, for example, a transmitter, a receiver, a processor, and one or more antennas), or any other imaging device 112 adapted to obtain images of the patient's anatomical features.

[0036] Sensor 126 is configured to provide sensor output. Sensor 126 may include a position sensor, proximity sensor, magnetometer, or accelerometer. In some embodiments, sensor 126 may include a linear encoder, rotary encoder, or incremental encoder (e.g., positioned to sense movement or position of robot arm 116). Sensor output or output data from sensor 126 may be provided to the electronic processor of robot 114, the electronic processor 104 of computing device 102, and / or navigation system 118. Output data from sensor 126 may also be used to determine position information of robot 114. It should be understood that in some embodiments, sensor 126 may be a component separate from robot arm 116. In other embodiments, sensor 136—which may be the same as or similar to sensor 126—may be integrated with robot 114. Sensor 126 enables electronic processor 104 (or the electronic processor of robot 114) to determine the precise pose of robot arm 116 (and any object or element held or secured to robot arm) in space. In other words, the sensor output, or the output data from sensor 126, can be used to calculate the position of the robotic arm 116 (and therefore the surgical tool 128) in space relative to one or more coordinate systems.

[0037] Robot 114 can be any surgical robot or part of any robot-assisted surgical system, either of which is capable of operating as described herein. Robot 114 can be, or includes, for example, Mazor X. TM Stealth robot guidance system, or any derivative thereof. Robot 114 may be configured to position surgical tool 128 at one or more precise poses (e.g., position and orientation). Surgical tool 128 may be any tool capable of cutting, drilling, milling, and / or segmenting anatomical elements. In one example, surgical tool 128 may be a drill. In some embodiments, robot 114 may be configured to rotate surgical tool 128 using, for example, one or more motors, to rotate surgical tool 128 and / or to travel the surgical tool.

[0038] Robot 114 may be additionally or alternatively configured to manipulate any component (whether or not based on guidance from navigation system 118) to perform or assist surgical tasks. In some embodiments, robot 114 may be configured to hold and / or manipulate anatomical elements during or in conjunction with surgical procedures. Robot 114 may include one or more robotic arms 116. In some embodiments, robotic arms 116 may include a first robotic arm and a second robotic arm, but robot 114 may include more than two robotic arms. In some embodiments, one or more robotic arms 116 may be used to hold and / or manipulate surgical instruments 128. Each robotic arm 116 may be able to be positioned independently of the other robotic arms. Robotic arms 116 may be controlled in a single shared coordinate space or in separate coordinate spaces.

[0039] The robot 114, together with the robotic arm 116, may have, for example, one, two, three, four, five, six, seven, or more degrees of freedom. Furthermore, the robotic arm 116 can be or is capable of being positioned in any pose, plane, and / or focal position. Pose includes position and orientation. Therefore, a surgical tool 128 or another object held by the robot 114 (or more specifically, by the robotic arm 116) can be precisely positioned in one or more desired and specific locations and orientations.

[0040] In some implementations, reference markers (i.e., navigation markers) may be placed on robot 114 (including, for example, on robotic arm 116), surgical instruments 128, or any other object in the surgical space. The reference markers may be tracked by navigation system 118, and the results of the tracking may be used by the operator of robot 114 and / or by the operator of medical system 100 or any component thereof.

[0041] During operation, navigation system 118 provides navigation for the surgeon and / or surgical robot. Navigation system 118 can be any navigation system currently known or developed in the future, including, for example, Medtronic StealthStation. TMThe S8 surgical navigation system or any subsequent system thereof. Navigation system 118 may include one or more cameras or other sensors for tracking one or more reference markers, navigation trackers, or other objects within the operating room or other room where part or all of the medical system 100 is located. The one or more cameras may be optical cameras, infrared cameras, or other cameras. In some embodiments, navigation system 118 may include one or more electromagnetic sensors. In various embodiments, navigation system 118 may be used to track the position and orientation (e.g., pose) of imaging device 112, robot 114 and / or robotic arm 116, surgical tools 128 and / or one or more other tools (or more specifically, for tracking the pose of navigation trackers directly or indirectly attached in a fixed relationship to one or more of the foregoing). Navigation system 118 may include a display for displaying one or more images from an external source (e.g., computing device 102, imaging device 112, or other sources), or for displaying images and / or video streams from one or more cameras or other sensors of navigation system 118. In some embodiments, medical system 100 may operate without using navigation system 118. The navigation system 118 may be configured to provide guidance to the surgeon or other users of the medical system 100 or its components, to the robot 114, or to any other element of the medical system 100 regarding, for example, the pose of one or more anatomical elements, whether the tool is in the appropriate trajectory, and / or how to move the tool into the appropriate trajectory to perform the surgical task in accordance with preoperative or other surgical planning.

[0042] In some embodiments, database 130 stores patient data associated with one or more patients. Patient data may include, for example, medical images of one or more patients, created by imaging device 112 or loaded into database 130 from another source or system. Patient data may also include patient-associated age, patient-associated body mass index (“BMI”), patient-associated bone mineral density, patient-associated sex, patient-associated implant history (e.g., date and placement location associated with implants such as screws, rods, retainers, prostheses, etc.), other aspects of the patient's medical history (e.g., smoker / non-smoker, chronic and acute illnesses, etc.), or combinations thereof. In some embodiments, patient data is associated with patients undergoing surgical procedures, where the systems and methods described herein are utilized. In some embodiments, patient data may be stored in database 130 prior to surgical procedures. In some embodiments, electronic processor 104 is configured to send a query to database 130 requesting patient data associated with a unique patient identifier and to receive patient data associated with a unique patient identifier from database 130.

[0043] Database 130 may store information relating one coordinate system to another (e.g., relating one or more robot coordinate systems to a patient coordinate system and / or a navigation coordinate system). Database 130 may additionally or alternatively store, for example, one or more surgical plans (including, for example, pose information about the target and / or image information about the anatomical structures of the patient at and / or proximal to the surgical site, for use by the user of robot 114, navigation system 118 and / or computing device 102 or medical system 100); one or more images of surgical procedures performed by or with the assistance of one or more other components of medical system 100; and / or any other useful information.

[0044] Database 130 may be configured to provide any such information to computing device 102 or any other device of medical system 100 or any other device outside medical system 100, whether directly or via cloud 134. In some embodiments, database 130 may be or include part of a hospital image storage system, such as a Picture Archiving and Communication System (PACS), a Health Information System (HIS), and / or another system for collecting, storing, managing, and / or transmitting electronic medical records including image data.

[0045] Cloud 134 can be or represents the Internet or any other wide area network. Computing device 102 can connect to cloud 134 via communication interface 108 using a wired connection, a wireless connection, or both. In some embodiments, computing device 102 can communicate with database 130 and / or external devices (e.g., computing devices) via cloud 134.

[0046] The medical system 100 or similar system may be used, for example, to implement one or more aspects of the methods described herein. The medical system 100 or similar system may also be used for other purposes.

[0047] Figure 2 Examples of operations are shown. Figure 1 The system described herein is an example method 200 for performing robotic surgery using estimated weight-bearing spinal geometry. Although method 200 is described in conjunction with medical system 100 as described herein, method 200 can be used with other systems and devices. Furthermore, method 200 can be modified or performed differently from the provided example. In particular, method 200 is also applicable to surgery using freehand surgical tools.

[0048] As an example, method 200 is described as being executed by computing device 102, and in particular by electronic processor 104. However, it should be understood that in some examples, portions of method 200 may be executed by other components of medical system 100 or by components external to system 100.

[0049] At box 202, electronic processor 104 extracts a scan of the spinal region (e.g., the lumbar region) in a supine position. For example, electronic processor 104 may extract a computed tomography scan from database 130 (e.g., acquired during preoperative evaluation). In some aspects, electronic processor 104 may control imaging device 112, sensor 126, or another medical scanning device coupled to system 100 to perform an intraoperative scan of the spinal region as part of a surgical procedure. As another example, electronic processor 104 may extract or generate a magnetic resonance imaging scan.

[0050] At box 204, the electronic processor 104 processes the scan using a machine learning model trained to generate standing spine measurements from a supine spine scan. For example, the electronic processor 104 may execute a spine geometry estimator 124, which implements a machine learning model trained using the method 300 described below. In some examples, scan image data is provided to the machine learning model. In some examples, the electronic processor 104 first segments the image of the spine region into multiple spine segments (e.g., identifying and segmenting each vertebra constituting the spine region in the scan image). The electronic processor 104 generates a three-dimensional surface model (e.g., depicting the vertebra) for each of the segmented spine segments. By using these surface models, the electronic processor 104 determines at least one three-dimensional geometric feature for each three-dimensional surface model (e.g., intervertebral disc volume, vertebral slippage vector / surface, three-dimensional intervertebral disc orientation, etc.). In some aspects, the electronic processor provides the three-dimensional geometric features of each of the multiple spine segments as input to the machine learning model.

[0051] At box 206, the electronic processor 104 receives multiple weight-bearing measurements of the spinal region from a machine learning model. Weight-bearing measurements are measurements describing the geometry of the spinal region when the patient is in a standing position, where gravity and the patient's body exert a compressive effect on the spine. Weight-bearing measurements include anterior disc height, posterior disc height, disc angle, spondylolisthesis grade, and spinal endplate orientation. In some respects, these measurements are two-dimensional, with some derived from corresponding three-dimensional geometric features (determined at box 204). For example, the machine learning model can predict anterior and posterior disc heights based on disc volume, predict spondylolisthesis grade based on spondylolisthesis vector / surface, and predict spinal endplate orientation based on three-dimensional disc orientation. In some respects, two-dimensional measurements may originate from combinations of their corresponding three-dimensional geometric features with other three-dimensional geometric features of the spinal region, from combinations of two or more three-dimensional geometric features that do not directly correspond to the two-dimensional measurements, or from an analysis of all three-dimensional geometric features of the spinal region.

[0052] At box 208, the electronic processor 104 generates a two-dimensional weight-bearing scan of the spinal region based on multiple weight-bearing measurements of the spinal segments. For example, image processing techniques can be used to modify the scan extracted at box 202 by applying the weight-bearing measurements to the scan to produce a simulated two-dimensional X-ray scan with dimensions associated with the weight-bearing measurements. In some examples, the electronic processor 104 controls a display (e.g., part of user interface 110) to present the two-dimensional weight-bearing scan (e.g., for analysis by a radiologist or surgeon). Optionally, the electronic processor 104 may generate a three-dimensional simulated scan based on lateral and anteroposterior measurements. In some aspects, axial rotation is missing, resulting in the simulated three-dimensional scan lacking axial rotation information.

[0053] At box 208, the electronic processor 104 generates a surgical plan for the spinal region based on a weight-bearing two-dimensional scan of the spinal region. For example, the electronic processor 104 can use the weight-bearing two-dimensional scan as input (e.g., input by a surgeon into the plan) to execute an existing planned path.

[0054] In some respects, the electronic processor 104 controls the robot 114 to perform operations on the spinal region based on a surgical plan. For example, the electronic processor 104 may control the robotic arm 116 according to the surgical plan and / or user input to remove a volume of bone from the spinal region using surgical tools 128. The surgeon may also directly manipulate the surgical tools 128 according to the surgical plan to perform a robotic arm-assisted bone removal operation.

[0055] As part of the surgical protocol, the electronic processor 104 can determine the outcome of the procedure. For example, while surgical instruments are being manipulated, the electronic processor 104 tracks the movement of surgical instruments 128. By tracking the movement of surgical instruments 128, the electronic processor 104 can use knowledge of the movement and state of surgical instruments 128 (e.g., the type of tool head and its operation) to generate data representing the volume of bone removed. Based on this result, the electronic processor 104 can utilize a machine learning model to generate at least one updated weight-bearing position measurement. For example, if a certain volume of bone has been removed, the machine learning model can generate an updated intervertebral disc height value. In some examples, as the surgical protocol progresses, the electronic processor 104 can generate an updated surgical plan based on the updated weight-bearing position measurements. In some examples, the electronic processor 104 can also take subsequent supine images of spinal segments during the protocol and use these images to generate updated weight-bearing two-dimensional scans, which can be presented to the surgeon, used to generate updated surgical plans, or both.

[0056] In some respects, the electronic processor 104 may alternatively or additionally use a weight-bearing two-dimensional scan as input to a second machine learning model. For example, as described below, a supine spinal scan and its associated two-dimensional weight-bearing scan can be used to supplement the data provided for training other machine learning models.

[0057] Figure 3 An example method 300 for training a machine learning model to estimate the geometry of the weight-bearing spine based on supine image data is illustrated. Although method 300 is described in conjunction with medical system 100 as described herein, method 300 can be used with other systems and devices. Furthermore, method 300 can be modified or performed differently from the provided example.

[0058] As an example, method 300 is described as being executed by computing device 102, and in particular by electronic processor 104. However, it should be understood that in some examples, portions of method 300 may be executed by other components of medical system 100 or by components external to system 100.

[0059] In the example below, the machine learning model is a random forest regressor. In other examples, the machine learning model may be a convolutional neural network, a transformer network, or a support vector regression network. Other variations of the machine learning model are also possible without departing from the disclosure presented herein.

[0060] At box 302, method 300 includes generating multiple three-dimensional standing images from multiple standing spinal X-ray images. The spinal X-ray images are posteroanterior and lateral views obtained from a patient sample, where both standing X-ray images and supine CT images (or other three-dimensional supine images) are available. However, because the target is novel (three-dimensional weight-bearing images), no ground truth is available for training the machine learning model. Therefore, method 300 includes generating ground truth for training. Each of the multiple three-dimensional standing images originates from one of the multiple spinal X-ray images.

[0061] In some aspects, for each spinal X-ray image in the spinal X-ray images, the electronic processor 104 segments the spine depicted in the X-ray image into multiple vertebrae. The electronic processor 104 performs three-dimensional pose reconstruction to generate vertebral poses of the vertebrae (e.g., using six degrees of freedom). In some examples, the electronic processor 104 uses computer vision (CV) algorithms for stereo vision to generate vertebral poses with the lowest vertebra as a reference. To reconstruct the pose of each vertebra, at least three corresponding points are required, preferably four or more. To accommodate this, one of the X-ray images should be tilted so that all four corners of one side are visible in the tilted image. Alternatively, computer vision techniques can be used to detect additional corresponding feature points on the image.

[0062] Alternatively or otherwise, the pose can be estimated using the Mazor registration algorithm, which optimizes both lateral and anterior / posterior registration to provide a pose estimate for each vertebra with 6 degrees of freedom.

[0063] To construct a 3D standing image, the lowest vertebra among multiple vertebrae is used as an anchor point, and the vertebrae are stitched together according to the predicted pose of each vertebra.

[0064] In some examples, the resulting three-dimensional standing images are simulated three-dimensional computed tomography scans and / or simulated magnetic resonance imaging scans.

[0065] At box 304, method 300 associates each of a plurality of supine spinal scan images with one of a plurality of three-dimensional standing images. For example, database 130 may include a patient record that includes both X-ray images and supine spinal scan images. Using this information, electronic processor 104 is able to assign each of the generated three-dimensional standing images to its corresponding supine spinal scan image. The supine spinal scan image may be a three-dimensional computed tomography scan, a magnetic resonance imaging scan, or both.

[0066] At box 306, method 300 provides multiple supine spinal scans as input data to a machine learning model. The supine spinal scans are the input type, and the trained machine learning model will generate weight-bearing geometric measurements for them.

[0067] At box 308, method 300 provides a machine learning model with multiple three-dimensional standing position images as ground truth data. In some aspects, more fine-grained ground truth is provided. For example, electronic processor 104 may determine multiple weight-bearing position measurements (as described above) for each of the multiple three-dimensional standing position images and include these multiple weight-bearing position measurements in the ground truth data.

[0068] In some examples, cross-validation is used to train machine learning models. For instance, the input and ground truth data can be split into multiple subsets of training data and used to perform k-fold cross-validation.

[0069] In some examples, machine learning models normalize the dimensions of the intervertebral discs and vertebral bodies (width, depth, and height, divided by the length of the spine) and the aspect ratio of the vertebral bodies and intervertebral discs when generating their output.

[0070] In some respects, the machine learning models described in this paper can be used to improve other analytical machine learning models. For example, for CT-based standards, applying a transformation (D = data feature input, t = target value, D' and t' are the transformed corresponding values) to the training dataset [D, t] used for complication prediction or planning prediction in three dimensions yields [D', t']. In other respects, the machine learning models described in this paper can be used for data enrichment. For example, existing two-dimensional and three-dimensional features can be enriched with the transformed features (DU D', t) for training with the original target t.

[0071] Similarly, the machine learning model described in this paper can be used for modality integration for database expansion. For example, to define a dynamic objective, images from two different time points are compared. Those images must have the same modality for comparison. However, in practice, available scan images are often mixed (e.g., X-ray images at time t1, but CT scan images at time t2). Furthermore, many medical databases are missing data, resulting in the absence of crucial "history" records for model training. The proposed transformation model can supplement the database. In the described example, CT instances are transformed into equivalent representations of X-ray instances, thus significantly increasing the training data.

[0072] Regarding the processes, systems, methods, and inspirations described herein, it should be understood that although the steps of such processes are described as occurring according to an ordered sequence, such processes can be practiced using the described steps performed in an order other than that described herein. It should also be understood that some steps may be performed simultaneously, other steps may be added, or certain steps described herein may be omitted. In other words, the description of processes herein is provided for the purpose of illustrating certain specific embodiments and should in no way be construed as limiting the claims.

[0073] Therefore, it should be understood that the above description is intended to be illustrative rather than restrictive. Many embodiments and applications beyond the examples provided will become apparent upon reading the above description. The scope should not be determined by reference to the above description, but rather by reference to the appended claims and the full scope of their equivalents. Future developments are anticipated and intended in the art discussed herein, and the disclosed systems and methods will be incorporated into such future embodiments. In conclusion, it should be understood that modifications and variations are possible in this application.

[0074] Unless expressly indicated otherwise herein, all terms used in the claims are intended to be given their broadest reasonable structure and their common meaning as understood by one skilled in the art described herein. In particular, the use of singular articles such as “a,” “the,” “the,” etc., should be understood to enumerate one or more of the elements shown, unless expressly limited to the contrary is set forth in the claims.

[0075] Unless otherwise explicitly stated, every value and range should be interpreted as an approximation, as if the words “about” or “approximate” preceded the value or range.

[0076] In this document, the reference to “an embodiment” or “implementation” means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this disclosure. The appearance of the phrase “in an embodiment” in various places in the specification does not necessarily refer to the same embodiment, nor is it necessarily a single or alternative embodiment mutually exclusive with other embodiments. The same applies to the term “specific implementation”.

[0077] Unless otherwise specified herein, the use of ordinal adjectives such as “first,” “second,” “third,” etc., to refer to one of a plurality of similar objects merely indicates that different instances of such similar objects are being referred to, and does not imply that the similar objects referred to in this way must have a corresponding order or sequence in time, space, ranking, or any other way.

[0078] Unless otherwise specified herein, the conjunction “if” may also be interpreted, in addition to its explicit meaning, as “when”, “in response to determination”, or “in response to detection,” depending on the specific context. For example, the phrase “if determination” or “if detection of [the stated condition]” may be interpreted as “in response to determination”, “in response to detection of [the stated condition or event]”, or “in response to detection of [the stated condition or event].”

[0079] Furthermore, for the purposes of this specification, the terms “coupled” and “connected” refer to any manner known in the art or subsequently developed, in which energy is allowed to be transferred between two or more elements, and the insertion of one or more additional elements is considered, but not required. Conversely, the terms “direct coupling,” “direct connection,” etc., imply the absence of such additional elements. The same type of distinction applies to the use of the terms “attachment” and “direct attachment” when applied to the description of physical structures. For example, such a “direct attachment” of two corresponding parts in such a physical structure may be implemented using a relatively thin layer of adhesive or other suitable bonding agent.

[0080] The described embodiments should be considered illustrative rather than restrictive in all respects. In particular, the scope of this disclosure is indicated by the appended claims, and not by the description and drawings herein. All modifications falling within the meaning and scope of the equivalents of the claims are to be covered by the claims.

[0081] The functionality of the various elements shown in the accompanying figures, including any functional blocks labeled “processor” and / or “controller,” can be provided using dedicated hardware and hardware capable of executing software in association with appropriate software. When provided by a processor, the functionality can be provided by a single dedicated processor, a single shared processor, or multiple separate processors, some of which may be shared. Furthermore, the explicit use of the terms “processor” or “controller” should not be construed as referring only to hardware capable of executing software, and may implicitly include, but is not limited to, digital signal processor (DSP) hardware, network processors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), read-only memory (ROM) for storing software, random access memory (RAM), and non-volatile memory. Other conventional and / or custom hardware may also be included. Similarly, any switches shown in the accompanying figures are conceptual only. Their functionality can be implemented through the operation of program logic, through dedicated logic, through the interaction of program control and dedicated logic, or even manually, with the specific technology selectable by the implementer, as understood more specifically from the context.

[0082] As used in this application, the term "circuit" may refer to one or more or all of the following: (a) a circuit implementation that is only hardware (such as an implementation that is only in analog and / or digital circuits); (b) a combination of hardware circuits and software, such as (if applicable): (i) a combination of analog and / or digital hardware circuits with software / firmware, and (ii) any part of a hardware processor with software (including digital signal processors), software, and memory that work together to enable a device (such as a mobile phone or server) to perform various functions; and (c) hardware circuits and / or processors, such as microprocessors or parts thereof, that require software (e.g., firmware) to operate, but may be absent when the software is not required to operate. "The definition of 'circuit' applies to all uses of the term in this application, including in any claim. As a further example, as used in this application, the term 'circuit' also covers specific implementations of hardware circuitry or processors (or processors) or portions thereof and their accompanying software and / or firmware. The term 'circuit' also covers (e.g., and if applicable to a particular claim element) baseband integrated circuits or processor integrated circuits for mobile devices or similar integrated circuits in servers, cellular network devices, or other computing or network devices."

[0083] Those skilled in the art will understand that any block diagram herein represents a conceptual view of an exemplary circuit embodying the principles of this disclosure. Similarly, it should be understood that any flowchart, flow diagram, state transition diagram, pseudocode, etc., represents various processes that can be represented substantially in a computer-readable medium and executed by a computer or processor, whether or not such a computer or processor is explicitly shown.

[0084] It should be understood that although some of the figures presented herein illustrate hardware and software located within a particular device, these depictions are for illustrative purposes only. In some embodiments, the illustrated components may be combined or divided into separate software, firmware, and / or hardware. For example, instead of being located within and executed by a single electronic processor, logic and processing may be distributed across multiple electronic processors. Regardless of how they are combined or divided, hardware and software components may reside on the same computing device or may be distributed among different computing devices connected by one or more networks or other suitable communication links.

[0085] The following paragraphs provide various embodiments that illustrate the examples and alternatives disclosed herein.

[0086] Example 1. A medical system comprising: an electronic processor configured to: extract a scan image of a spinal region in a supine position; process the scan image using a machine learning model trained to generate standing spinal measurements from the supine spinal scan image; receive multiple weight-bearing measurements of the spinal region from the machine learning model; generate a two-dimensional weight-bearing scan image of the spinal region based on the multiple weight-bearing measurements of the spinal region; and generate a surgical plan for the spinal region based on the two-dimensional weight-bearing scan image of the spinal region.

[0087] Example 2. The medical system according to Example 1, the medical system further includes: a surgical robot communicatively coupled to the electronic processor; wherein the electronic processor is further configured to: perform an operation on the spinal region based on the surgical plan; determine the result of the operation; generate at least one updated weight-bearing position measurement based on the result using the machine learning model; and generate an updated surgical plan based on the at least one updated weight-bearing position measurement.

[0088] Example 3. The medical system according to any one of Examples 1 and 2, the medical system further comprising: a display, the display being communicatively coupled to the electronic processor; wherein the electronic processor is further configured to: control the display to present the weighted two-dimensional scan image.

[0089] Example 4. A medical system according to any one of Examples 1 to 3, wherein the electronic processor is further configured to process the scan image using a machine learning model in the following manner: segmenting the spinal region into a plurality of spinal segments; generating a three-dimensional surface model for each of the plurality of spinal segments; determining at least one three-dimensional geometric feature for each three-dimensional surface model; and providing the at least one three-dimensional geometric feature of each of the plurality of spinal segments as input to the machine learning model.

[0090] Example 5. A medical system according to any one of Examples 1 to 4, wherein the electronic processor is configured to train the machine learning model by: generating a plurality of three-dimensional standing images from a plurality of standing spinal X-ray images, wherein each of the plurality of three-dimensional standing images originates from one of the plurality of spinal X-ray images; associating each of a plurality of supine spinal CT scan images with one of the plurality of three-dimensional standing images; providing the plurality of supine spinal CT scan images as input data to the machine learning model; and providing the plurality of three-dimensional standing images as ground truth data to the machine learning model.

[0091] Example 6. The medical system according to Example 5, wherein the electronic processor is further configured to: generate the plurality of three-dimensional standing images by: for each of the plurality of spinal X-ray images, segmenting the spine in the image into a plurality of vertebrae; for each of the plurality of vertebrae, performing three-dimensional pose reconstruction to generate a plurality of vertebrae poses; and using the lowest vertebra among the plurality of vertebrae as an anchor point, stitching together each of the plurality of vertebrae according to the predicted pose of each vertebra.

[0092] Example 7. A medical system according to any one of Examples 5 and 6, wherein the electronic processor is configured to train the machine learning model by: dividing the input data and the ground truth data into multiple training data subsets; and performing k-fold cross-validation based on the multiple training data subsets.

[0093] Example 8. The medical system according to any one of Examples 5 to 7, wherein the plurality of spinal X-ray images include a plurality of posteroanterior views and lateral views.

[0094] Example 9. The medical system according to any one of Examples 1 to 8, wherein the plurality of weight-bearing measurements include anterior intervertebral disc height, posterior intervertebral disc height, intervertebral disc angle, spondylolisthesis grade, and spinal endplate orientation.

[0095] Example 10. The medical system according to any one of Examples 1 to 9, wherein the scan image of the spinal region is one of a three-dimensional computed tomography scan image and a magnetic resonance imaging scan image.

[0096] Example 11. The medical system according to any one of Examples 1 to 10, wherein the weight-bearing two-dimensional scan of the spinal region is a simulated two-dimensional computed tomography scan.

[0097] Example 12. A method for training a machine learning model to estimate the geometry of a weight-bearing spine based on supine spinal scans: generating multiple three-dimensional standing images from multiple standing spinal X-ray images, wherein each of the multiple three-dimensional standing images originates from one of the multiple spinal X-ray images; associating each of the multiple supine spinal scans with one of the multiple three-dimensional standing images; providing the multiple supine spinal scans as input data to the machine learning model; and providing the multiple three-dimensional standing images as ground truth data to the machine learning model.

[0098] Example 13. The method according to Example 12, the method further comprising: determining a plurality of weight-bearing measurements for each of the plurality of three-dimensional standing images; and including the plurality of weight-bearing measurements in the true data; wherein the plurality of weight-bearing measurements for each of the plurality of three-dimensional standing images includes at least one selected from the group consisting of: anterior intervertebral disc height, posterior intervertebral disc height, intervertebral disc angle, spondylolisthesis grade, and spinal endplate orientation.

[0099] Example 14. The method according to any one of Examples 12 and 13, wherein generating the plurality of three-dimensional standing images comprises: for each of the plurality of spinal X-ray images, segmenting the spine in the image into a plurality of vertebrae; for each of the plurality of vertebrae, performing three-dimensional pose reconstruction to generate a plurality of vertebrae poses; and using the lowest vertebra among the plurality of vertebrae as an anchor point, stitching together each of the plurality of vertebrae according to the predicted pose of each vertebra.

[0100] Example 15. The method according to any one of Examples 12 to 14, the method further comprising: dividing the input data and the ground truth data into a plurality of training data subsets; and performing k-fold cross-validation based on the plurality of training data subsets.

[0101] Example 16. The method according to any one of Examples 12 to 15, wherein the plurality of spinal X-ray images include a plurality of posteroanterior views and lateral views.

[0102] Example 17. The method according to any one of Examples 12 to 16, wherein the plurality of supine spinal scans include at least one selected from the group consisting of: three-dimensional computed tomography scans and magnetic resonance imaging scans.

[0103] Example 18. The method according to any one of Examples 12 to 17, wherein the plurality of three-dimensional standing images include at least one selected from the group consisting of: simulated three-dimensional computed tomography scans and simulated magnetic resonance imaging scans.

[0104] Example 19. The method according to any one of Examples 12 to 18, wherein the machine learning model is selected from the group consisting of: random forest regressors, convolutional neural networks, transformer networks, and support vector regression networks.

[0105] Example 20. A medical imaging system, the system comprising: a display; an electronic processor coupled to the display and configured to: extract a scan image of a spinal region in a supine position; process the scan image using a machine learning model trained to generate standing spinal measurements from the supine spinal scan image; receive multiple weight-bearing measurements of the spinal region from the machine learning model; generate a weight-bearing two-dimensional scan image of the spinal region based on the multiple weight-bearing measurements of the spinal region; and control the display to present the weight-bearing two-dimensional scan image of the spinal region.

[0106] Example 21. A medical system comprising: an electronic processor configured to: extract a scan image of a spinal region in a supine position; process the scan image using a machine learning model trained to generate standing spinal measurements from the supine spinal scan image; receive multiple weight-bearing measurements of the spinal region from the machine learning model; generate a two-dimensional weight-bearing scan image of the spinal region based on the multiple weight-bearing measurements of the spinal region; and provide the scan image of the supine spinal region and the two-dimensional scan image of the spinal region as training data to a second machine learning model.

[0107] The various features and advantages of the embodiments and implementations presented herein are set forth in the following claims.

Claims

1. A medical system (100), the system comprising: Electronic processor (104), the electronic processor being configured to: Extract the scan image of the spinal region in the supine position; The scan image is processed using a machine learning model trained to generate standing spine measurements from a supine spine scan image; Receive multiple weight-bearing position measurements of the spinal region from the machine learning model; A two-dimensional scan of the spinal region is generated based on the multiple weight-bearing position measurements of the spinal region. as well as A surgical plan for the spinal region is generated based on the two-dimensional weight-bearing scan of the spinal region.

2. The medical system (100) according to claim 1, further comprising: A surgical robot (114) is communicatively coupled to the electronic processor (104). The electronic processor (104) is further configured to: The procedure is performed on the spinal region based on the surgical plan; Determine the result of the operation; The machine learning model is used to generate at least one updated negative weight measurement based on the results; as well as An updated surgical plan is generated based on the at least one updated weight-bearing position measurement.

3. The medical system (100) according to any one of claims 1 and 2, further comprising: A display, which is communicatively coupled to the electronic processor (104). The electronic processor (104) is further configured to: Control the display to present the two-dimensional scan of the weighted position.

4. The medical system (100) according to any one of claims 1 to 3, wherein the electronic processor (104) is further configured to process the scan image using a machine learning model in the following manner: The spinal region is divided into multiple spinal segments; Generate a three-dimensional surface model for each of the plurality of spinal segments; Determine at least one three-dimensional geometric feature for each three-dimensional surface model; and The machine learning model is provided with at least one three-dimensional geometric feature of each of the plurality of spinal segments as input.

5. The medical system (100) according to any one of claims 1 to 4, wherein the electronic processor (104) is configured to train the machine learning model in such a way as: Multiple three-dimensional standing images are generated from multiple standing spinal X-ray images, wherein each of the multiple three-dimensional standing images is derived from one of the multiple spinal X-ray images; Each supine spinal CT scan image in multiple supine positions is associated with one of the multiple three-dimensional standing images; The multiple supine spinal CT scans are provided as input data to the machine learning model; and The machine learning model is provided with the multiple 3D standing images as ground truth data.

6. The medical system (100) according to claim 5, wherein the electronic processor (104) is further configured to: The multiple three-dimensional standing images are generated in the following manner: For each of the multiple spinal X-ray images The spine in the image is segmented into multiple vertebrae; For each of the plurality of vertebrae, perform three-dimensional pose reconstruction to generate multiple vertebral poses; and Using the lowest vertebra among the plurality of vertebrae as an anchor point, each of the plurality of vertebrae is spliced ​​together according to the predicted pose of each vertebra.

7. The medical system (100) according to any one of claims 5 and 6, wherein the electronic processor (104) is configured to train the machine learning model in such a way as: The input data and the ground truth data are divided into multiple training data subsets; and Perform k-fold cross-validation based on the multiple subsets of training data.

8. The medical system (100) according to any one of claims 5 to 7, wherein the plurality of spinal X-ray images include a plurality of posteroanterior views and lateral views.

9. The medical system (100) according to any one of claims 1 to 8, wherein the plurality of weight-bearing measurements include anterior intervertebral disc height, posterior intervertebral disc height, intervertebral disc angle, spondylolisthesis grade, and spinal endplate orientation.

10. The medical system (100) according to any one of claims 1 to 9, wherein the scan image of the spinal region is one of a three-dimensional computed tomography scan image and a magnetic resonance imaging scan image.

11. The medical system (100) according to any one of claims 1 to 10, wherein the weight-bearing two-dimensional scan of the spinal region is a simulated two-dimensional computed tomography scan.

12. A method for training a machine learning model to estimate the geometry of the weight-bearing spine from supine scan data in the following manner (300) (200): Multiple three-dimensional standing images are generated from multiple standing spinal X-ray images, wherein each of the multiple three-dimensional standing images is derived from one of the multiple spinal X-ray images; Each supine spinal scan image in the plurality of supine spinal scan images is associated with one of the plurality of three-dimensional standing images; The multiple supine spinal scan images are provided as input data to the machine learning model; and The machine learning model is provided with the multiple 3D standing images as ground truth data.

13. The method (300) (200) according to claim 12, further comprising: For each of the plurality of three-dimensional standing position images, determine multiple load position measurements; as well as The multiple negative weight measurement values ​​are included in the true value data; The weight-bearing measurements of each of the plurality of three-dimensional standing images include at least one selected from the group consisting of: anterior intervertebral disc height, posterior intervertebral disc height, intervertebral disc angle, spondylolisthesis grade, and spinal endplate orientation.

14. The method (300) (200) according to any one of claims 12 and 13, wherein generating the plurality of three-dimensional standing images comprises: For each of the multiple spinal X-ray images The spine in the image is segmented into multiple vertebrae; For each of the plurality of vertebrae, three-dimensional pose reconstruction is performed to generate multiple vertebrae poses; as well as Using the lowest vertebra among the plurality of vertebrae as an anchor point, each of the plurality of vertebrae is spliced ​​together according to the predicted pose of each vertebra.

15. The method (300) (200) according to any one of claims 12 to 14, the method further comprising: The input data and the true data are divided into multiple training data subsets; as well as Perform k-fold cross-validation based on the multiple subsets of training data.

Images