Performance monitoring and generation of predictions regarding orthopedic prostheses

Abstract

A computer-implemented method comprising: obtaining feature data that is based at least in part on signals generated by one or more sensors included in an orthopedic prosthesis implanted in a joint of a patient; applying, by one or more processors, a machine learning (ML) model to the feature data to generate one or more joint reaction force (JRF) vectors; and generating, by the one or more processors, a prediction regarding the orthopedic prosthesis based at least in part on the one or more JRF vectors.

Description

1262-250W001PERFORMANCE MONITORING AND GENERATION OF PREDICTIONS REGARDING ORTHOPEDIC PROSTHESES

[0001] This application claims priority to U.S. Provisional Patent Application 63 / 788,553, filed April 14, 2025, the entire content of which is incorporated herein by reference.BACKGROUND

[0002] Orthopedic surgeries often involve implanting one or more orthopedic prostheses into a patient. For example, a total shoulder arthroplasty (TSA) involves implanting a humeral prosthesis on a proximal end of a humerus of a patient s shoulder joint and a corresponding glenoid prosthesis on the glenoid of a scapula of the patient’s shoulder joint. In an ankle replacement surgery, a surgeon may attach orthopedic prostheses to a tibia and a talus of a patient. Incorrect selection or placement of one or more prostheses, such as a humeral prosthesis and / or glenoid prosthesis, may lead to patient discomfort, impingement, subluxation, dislocation, premature failures of the prostheses and / or other events. Even if the one or more prostheses, such as a humeral prosthesis and / or glenoid prosthesis, are placed appropriately, a patient’s anatomy and / or activity may lead to patient discomfort, impingement, subluxation, dislocation, premature failures of the prostheses and / or other events.SUMMARY

[0003] During preoperative planning and / or after implantation of orthopedic prostheses (such as a humeral prosthesis and glenoid prosthesis), a user may want to predict a likelihood of whether particular events involving the orthopedic prostheses may occur for a patient. This disclosure describes devices, systems, and / or methods for using sensors included in orthopedic prostheses, such as humeral prosthesis and / or glenoid prosthesis, to determine one or more joint reaction force (JRF) vectors that indicate a magnitude and direction of a force that is being applied by one prosthesis (such as a humeral or glenoid prosthesis) on another prosthesis (such as a corresponding glenoid or humeral prosthesis), and generating, by one or more processors, a prediction regarding an event involving the orthopedic prostheses based at least in part on the one or more JRF vectors.

[0004] A medical system may include sensors, such as Hall effect sensors, that may be included in orthopedic prostheses (such as a humeral prosthesis and glenoid prosthesis)1262-250W001and may generate signal data that represent the distance of one prosthesis from another prosthesis and / or the angle of one prosthesis relative to a corresponding prosthesis. The medical system may generate feature data based, at least in part, on the signal data. In some examples, feature data may directly include the signal data. In some examples, the feature data may include data characterizing a patient's anatomy and / or shapes and positions of an implanted orthopedic prosthesis. In some examples, a medical system may additionally or alternatively include one or more processors configured to obtain the feature data based on signals generated by the one or more sensors included in an orthopedic prosthesis and apply a machine learning (ML) model to the feature data to generate one or more JRF vectors. In some examples, the medical system may use a combination of physics-based numerical simulations and one or more ML models to generate the one or more JRF vectors. Furthermore, in some examples, the one or more processors may be configured to map sequences of the feature data to sequences of data indicating the one or more JRF vectors.

[0005] A JRF vector changes over time, such as when a patient moves a limb (e.g., arm) and / or performs various activities of daily living (ADL). For example, for a humeral prosthesis and glenoid prosthesis, a JRF vector may be different when a patient has an arm extended over their head than when a patient’s arm is resting at their side, and a JRF vector may be different when a patient is carrying a heavy load versus when a patient is not carrying anything. The one or more processors may collect JRF vectors over a period of time and may generate a prediction regarding the orthopedic prostheses based at least in part on the JRF vectors. In some examples, the prediction relates to one or more of prosthesis performance, excessive wear on articulating surfaces of one or more prostheses, bone remodeling, loosening of a prosthesis, insufficient force on a prosthesis, impingement, subluxation, and / or dislocation.

[0006] For example, excessive force by one prosthesis on another prosthesis may erode polyethylene articulating surfaces prematurely and / or may lead to bone remodeling or loosening of the prostheses, which may lead to a revision surgery to replace the prostheses. In accordance with the techniques described herein, one or more processors may generate a prediction of excessive wear on articulating surfaces of one or more prostheses, bone remodeling, and / or loosening of a prosthesis based on JRF vector(s) having a magnitude that satisfy an excessive force threshold.

[0007] In some examples, one or more processors may generate a prediction regarding orthopedic prostheses using a physics-based model, such as a musculoskeletal model1262-250W001(MSM) and a finite element analysis (FEA) model. An FEA model may represent a patient’s bones, prostheses, and soft tissue structures (e.g., tendons, ligaments, muscles). For example, soft tissue structures may be modeled as linear elements that connect discrete points on the bone elements. In some examples, elements representing the soft tissue structures may have attributes representative of strength and elasticity of the represented soft tissue structures. In some examples, one or more processors may generate elements of the FEA model based on imaging scans, such as a computed tomography (CT) scan or a magnetic resonance imaging (MRI) scan, of a patient’s anatomy. In some examples, the imaging scans may include preoperative scans and / or postoperative scans. In some examples, the one or more processors may adjust values of parameters of the soft tissue structural elements of the FEA model based on the JRF vectors. In some examples, the one or more processors generate recommendations for the user or the patient based on the generated prediction, such as a recommendation on how to reduce the likelihood of an event from occurring,

[0008] In one example, this disclosure describes a computer-implemented method comprising: obtaining, by one or more processors, feature data that is based at least in part on signals generated by one or more sensors included in an orthopedic prosthesis implanted in a joint of a patient; applying, by the one or more processors, a machine learning (ML) model to the feature data to generate one or more joint reaction force (JRF) vectors; and generating, by the one or more processors, a prediction regarding the orthopedic prosthesis based at least in part on the one or more JRF vectors.

[0009] In another example, this disclosure describes a system comprising: an orthopedic prosthesis configured to be implanted in a joint of a patient, the orthopedic prosthesis including at least one sensor configured to generate signal data; and one or more processors configured to: obtain feature data based on the signal data, apply a machine learning (ML) model to the feature data to generate one or more joint reaction force (JRF) vectors, and generate a prediction regarding the orthopedic prosthesis based at least in part on the one or more JRF vectors.

[0010] In another example, this disclosure describes one or more non-transitory computer-readable storage media having processor-executable instructions stored thereon that, when executed by one or more processors, cause the one or more processors to obtain feature data that is based at least in part on signals generated by one or more sensors included in an orthopedic prosthesis implanted in a joint of a patient; apply a machine learning (ML) model to the feature data to generate one or more joint reaction force (JRF)1262-250W001vectors; and generate a prediction regarding the orthopedic prosthesis based at least in part on the one or more JRF vectors.

[0011] The details of various examples of the disclosure are set forth in the accompanying drawings and the description below. Various features, objects, and advantages will be apparent from the description, drawings, and claims.BRIEF DESCRIPTION OF DRAWINGS

[0012] FIG. 1 is a conceptual diagram illustrating an example computing system in which one or more techniques of this disclosure may be performed,

[0013] FIG. 2 is a block diagram illustrating example components of a prosthesis, in accordance with one or more techniques of this disclosure.

[0014] FIG. 3 is a schematic diagram of a glenoid prosthesis and a humeral prosthesis, in accordance with one or more techniques of this disclosure.

[0015] FIG. 4 is a conceptual diagram illustrating an example user interface showing an animation of a simulated movement of a joint, in accordance with one or more techniques of this disclosure.

[0016] FIG. 5 is a conceptual diagram illustrating an example user interface for showing a track of a JRF vector as a patient moves, in accordance with one or more techniques of this disclosure.

[0017] FIG. 6 is a conceptual diagram illustrating an example user interface showing a heat map indicating locations of predicted contact pressure, in accordance with one or more techniques of this disclosure of one prosthesis on another prosthesis.

[0018] FIG. 7 is a flowchart illustrating an example operation in accordance with one or more techniques of this disclosure,

[0019] FIG. 8 is a conceptual diagram illustrating an example architecture of a neural network for determining JRF vectors of a prosthesis, in accordance with one or more techniques of this disclosure.

[0020] FIG. 9 is a conceptual diagram illustrating an example system for training a neural network, in accordance with one or more techniques of this disclosure.

[0021] FIG. 10 is a conceptual diagram illustrating an example process for generating a musculoskeletal model, in accordance with one or more techniques of this disclosure.1262-250W001DETAILED DESCRIPTION

[0022] Orthopedic surgeries often involve implanting one or more orthopedic prostheses into a patient. For example, in a total shoulder arthroplasty, a surgeon replaces an articulating surface of a glenoid fossa of a scapula of a patient’s shoulder with a glenoid prosthesis and an articulating surface of a humerus of the patient’s shoulder with a humeral prosthesis. A patient may undergo a total shoulder arthroplasty for a variety of reasons. For example, the patient may be experiencing pain or joint instability due to bone erosion or trauma. Bone erosion may be caused by arthritis or other processes.

[0023] During preoperative planning and / or after implantation of orthopedic prostheses (such as a humeral prosthesis and glenoid prosthesis), a user may want to predict likely effects of implanting the orthopedic prostheses. Such effects may include events such as excessive wear on articulating surfaces of one or more prostheses, bone remodeling, loosening of a prosthesis, insufficient force on a prosthesis, impingement, subluxation, and / or dislocation. In some examples, the effects may include an anticipated range of motion. Furthermore, the techniques of the disclosure provide the advantage of real-time intraoperative verification. By utilizing a trial prosthesis equipped with one or more sensors, tire system can obtain and analyze joint reaction force vectors during the surgical procedure. This allows a user to assess the stability and force balance of a candidate prosthesis system before final implantation, facilitating immediate adjustments to the placement or selection of the permanent prosthesis to optimize patient outcomes and reduce the likelihood of postoperative complications such as subluxation or dislocation.

[0024] This disclosure describes devices, systems, and / or computer-implemented methods or techniques for using feature data based on signals generated by one or more sensors included in orthopedic prostheses, such as humeral and / or glenoid prostheses, to determine one or more JRF vectors that indicate a magnitude and direction of a force that is being applied by one prosthesis on another prosthesis. A prediction regarding the orthopedic prostheses, such as excessive wear on articulating surfaces of one or more prostheses, bone remodeling, loosening of a prosthesis, insufficient force on a prosthesis, impingement, subluxation, and / or dislocation, may be generated based at least in part on the one or more JRF vectors. Because the sensors are included in a first prosthesis and / or a second prosthesis, the one or more JRF vectors may be significantly more accurate than if the one or more JRF vectors were obtained from external sources, such as cameras or wearable devices. In addition, because the sensors are included in a first prosthesis and / or1262-250W001a second prosthesis, the JRF vectors may be continuously or periodically sensed, which may provide significantly more accurate predictions than if the JRF vectors were obtained from visits to a doctor’s office. Furthermore, obtaining kinematic data and joint reaction force vectors continuously via internal sensors 118 provides the technical advantage of reducing the patient's cumulative radiation exposure. Because prediction system 116 can monitor joint kinematics, identify subluxation risks, and detect potential prosthesis loosening directly from the generated signal data, the necessity for repeated postoperative X-rays, fluoroscopy, or CT scans to visually confirm the mechanical status of orthopedic prostheses 104 is significantly reduced. Thus, the JRF vectors are useful in generating a prediction regarding the orthopedic prosthesis, such as a prediction regarding a performance or function of the orthopedic prosthesis during activities of daily living. The techniques may use one or more machine learning (ML) models, one or more physics-based models, and / or one or more other types of techniques to generate the prediction regarding the orthopedic prosthesis. In some examples, the computer-implemented techniques may additionally include generating recommendations based on the generated prediction regarding the orthopedic prostheses, such as a recommendation on how to reduce the likelihood of an event from occurring.

[0025] FIG. 1 is a conceptual diagram illustrating an example system 100 in which one or more techniques of this disclosure may be performed. In the example of FIG. 1, system 100 includes a computing system 102 and at least one of a first orthopedic prosthesis 104A or a second orthopedic prosthesis 104B. This disclosure may refer to orthopedic prosthesis 104A and orthopedic prosthesis 104B collectively as "‘orthopedic prostheses 104.” In some examples, orthopedic prosthesis 104A is a humeral prosthesis configured to be implanted on a patient’s humerus. In some examples, orthopedic prosthesis 104B is a glenoid prosthesis configured to be implanted on a patient’s scapula. In some examples, orthopedic prosthesis 104A is a tibial prosthesis and orthopedic prosthesis 104B is a talar prosthesis. In some examples, orthopedic prosthesis 104A is an acetabular prosthesis and orthopedic prosthesis 104B is a femoral prosthesis. In some examples, orthopedic prosthesis 104A is a femoral prosthesis and orthopedic prosthesis 104B is a tibial prosthesis.

[0026] Computing system 102 is configured to generate a prediction regarding one or more of orthopedic prostheses 104. Examples of such predictions may include a prediction regarding whether excessive wear will occur on articulating surfaces of one or more orthopedic prostheses 104, a prediction regarding whether bone remodeling will1262-250W001occur in bones onto which orthopedic prostheses 104 are implanted, a prediction regarding whether loosening of one or more of orthopedic prostheses 104 will occur, a prediction regarding whether one of orthopedic prostheses 104 will exert insufficient force on the other one of orthopedic prosthesis 104, a prediction regarding whether an impingement will occur in a joint involving orthopedic prostheses 104, a prediction regarding whether a subluxation will occur in a joint involving orthopedic prostheses 104, a prediction regarding whether a dislocation will occur in a joint involving orthopedic prostheses 104, a prediction regarding an end of a lifespan of an orthopedic prosthesis system that includes orthopedic prostheses 104, a prediction regarding a range of motion achievable with orthopedic prosthesis 104, and so on. In some examples, to assist with surgical planning (e.g., choosing and positioning orthopedic prostheses), computing system 102 generates the prediction before a patient undergoes a surgery to implant the one or more orthopedic prostheses 104.

[0027] Computing system 102 may include one or more computing devices. In one example, computing system 102 includes a personal computer used by a user, such as a surgeon, clinician, or other type of user. In this example, the personal computer may generate a prediction regarding orthopedic prostheses 104 without interaction with other computing devices. In other examples, computing system 102 includes a server device and a client device (e.g., a personal computer). In such examples, the server device may generate a prediction regarding orthopedic prostheses 104 based on input initially received via the client device. In any case, one or more computing devices of computing system 102 may output user interfaces for display to a user and may receive, directly or indirectly, indications of user input.

[0028] Computing system 102 includes one or more processors 108, a storage system 110, a communication interface 112, and a display 114, In other examples, computing system 102 may include more, fewer, or different components. Tire components of computing system 102 may be in one or more computing devices. For example, processors 108 may be in a single computing device or distributed among multiple computing devices of computing system 102, storage system 110 may be in a single computing device or distributed among multiple computing devices of computing system 102, and so on. In some examples, computing system 102 is a personal computer, a system of computing devices, one or more server devices (e.g., local or cloud-based server devices), or a system comprising one or more other types of computing devices.1262-250W001Processors 108, storage system 110, communication interface 112, and display 114 are communicatively coupled.

[0029] Processors 108 may be implemented in circuitry and include one or more microprocessors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), hardware, or any combinations thereof In general, processors 108 may be implemented as fixed-function circuits, programmable circuits, or a combination thereof, Fixed-function circuits refer to circuits that provide particular functionality and are preset on the operations that can be performed. Programmable circuits refer to circuits that can be programmed to perform various tasks and provide flexible functionality in the operations that can be performed. For instance, programmable circuits may execute software or firmware that cause the programmable circuits to operate in the manner defined by instructions of the software or firmware. Fixed-function circuits may execute software instructions (e.g., to receive parameters or output parameters), but the types of operations that the fixed-function circuits perform are generally immutable. In some examples, one or more of tire units may be distinct circuit blocks (fixed-function or programmable), and in some examples, the one or more units may be integrated circuits.

[0030] Processors 108 may include arithmetic logic units (ALUs), elementary’ function units (EFUs), digital circuits, analog circuits, and / or programmable cores, formed from programmable circuits. In examples where the operations of processors 108 are performed using software executed by the programmable circuits, storage system 110 may store the object code of the software that processors 108 receives and executes, or another memory within processors 108 (not shown) may store such instructions. Examples of the software include software designed for surgical planning. Processors 108 may perform the actions ascribed in this disclosure to processors 108.

[0031] Processors 108 may output data (e.g., a user interface, models, etc.) for display. Outputting data for display may include one or more of processors 108 generating and sending signals to a display device (e.g., display 114) that the display device can directly use to display the data. Outputting data for display may include one or more of processors 108 outputting data for transmission to another computing device (e.g., another computing device of computing system 102) that processes the data to generate signals that a display device (e.g., display 114) may directly use to display the data.

[0032] Processors 108 may receive indications of user input from one or more users. Processors 108 may receive an indication of user input directly from a user input device1262-250W001(e.g., keyboard, mouse, touchscreen, etc.). For instance, in an example where computing system 102 is implemented on a single computing device, processors 108 may receive the indications of user input from one or more user input devices of the computing device. In other examples, processors 108 may receive the indications of user input by way of one or more computing devices. For instance, in an example where computing system 102 is implemented using a server device and a client device and processors 108 are located in the server device, processors 108 may receive indications of the user input from the client device. For example, processors 108 may receive an indication from the client device that the user has selected a displayed element, typed specific text, and so on.

[0033] Storage system 110 may store various types of data used by processors 108. Storage system 110 may include any of a variety of memory devices, such as dynamic random access memory (DRAM), including synchronous DRAM (SDRAM), magnetoresistive RAM (MRAM), resistive RAM (RRAM), or other types of memory devices. Examples of display 114 include a liquid crystal display (LCD), a plasma display, an organic light emitting diode (OLED) display, or another type of display device. Display 114 may be positioned in a user’s office, an operating room, or another location.

[0034] Communication interface 112 allows computing system 102 to output data and instructions to and receive data and instructions from a medical imaging system, orthopedic prostheses 104, or other devices via one or more communication links or networks. Communication interface 112 may include hardware circuitry that enables computing system 102 to communicate (e.g., wirelessly or using wires) with other computing systems and devices. Example networks may include various types of communication networks including one or more wide-area networks, such as the Internet, local area networks, and so on. In some examples, the network may include wired and / or wireless communication links.

[0035] In the example of FIG. 1, storage system 110 stores a prediction system 116 and a training system 132. In other examples, storage system 110 may store more, fewer, or different types of data or units. Moreover, the data and units illustrated in the example of FIG. 1 are provided for purposes of explanation and may not represent how data is actually stored or how software is actually implemented. Prediction system 116 may comprise instructions that are executable by processors 108. For ease of explanation, this disclosure may describe prediction system 116 and training system 132 as performing various actions when processors 108 execute instructions of prediction system 116 and1262-250W001training system 132. In some examples, the actions performed by prediction system 116 and training system 132 are performed by separate computing systems. In some examples, storage system 110 does not include training system 132.

[0036] Prediction system 116 may be an instance of a computer-assisted orthopedic surgery (CAOS) system or a computer-assisted surgical system (CASS). In some examples, prediction system 116 may be used for preoperative planning. In some examples, prediction system 116 may assist users after a surgery. For example, prediction system 116 may provide information to assist with recovery, physical therapy, or other postoperative activities. In some examples, prediction system 116 may also assist users during a surgery. For instance, prediction system 116 may output navigation information to one or more devices (e.g., monitors, tablet computers, head-mounted displays, etc.) to a user during a surgery to help the user execute a surgical plan.

[0037] During preoperative planning and / or after implantation of orthopedic prostheses (such as a humeral prosthesis and glenoid prosthesis), a user, such as a physician, may want to predict whether particular events involving the orthopedic prostheses are likely to occur, such as excessive wear on articulating surfaces of one or more prostheses, an end of a lifespan of one or more prostheses, bone remodeling, loosening of a prosthesis, insufficient force on a prosthesis, impingement, subluxation, and / or dislocation, may occur for a patient.

[0038] Orthopedic prosthesis 104A and / or orthopedic prosthesis 104B may include sensors 118, such as Hall effect sensors. In some examples, only one of orthopedic prosthesis 104A or orthopedic prosthesis 104B includes sensors 118, Sensors 118 may generate signals usable to determine a position of orthopedic prosthesis 104A relative to orthopedic prosthesis 104B. For example, sensors 118 may include a Hall sensor that generates an electrical signal having a voltage correlated with a motion of a corresponding magnetic element (i.e., a Hall element). For instance, orthopedic prosthesis 104B may include a Hall sensor and orthopedic prosthesis 104A may include a magnetic element. In some examples, sensors 118 include inertial measurement units (IMUs). In some examples, one or more of orthopedic prostheses 104 is a trial prosthesis that is temporarily attached to the patient during surgery. By using trial prostheses, a surgeon may be able to test and confirm whether a corresponding permanent orthopedic prosthesis is appropriate for a patient,

[0039] Prediction system 116 may obtain feature data 128. Feature data 128 is based at least in part on signals generated by sensors 118 included in one or more of orthopedic1262-250W001prostheses 104. Feature data 128 may represent the distance of orthopedic prosthesis 104A from orthopedic prosthesis 104B and / or the angle of orthopedic prosthesis 104A relative to orthopedic prosthesis 104B. In some examples, prediction system 116 may, as at least part of obtaining feature data 128, generate at least part of feature data 128. In some examples, prediction system 116 may, as at least part of obtaining feature data 128, receive at least part of feature data 128 from one or more of orthopedic prostheses 104. Prediction system 116 may obtain feature data 128 during or after a surgery involving orthopedic prostheses 104.

[0040] In some examples, feature data 128 directly includes signal data generated by one or more of sensors 118. The signal data may indicate a relative orientation of orthopedic prostheses 104, a distance between orthopedic prostheses 104, an overall motion of orthopedic prostheses 104, and / or other information. The overall motion of orthopedic prostheses 104 may be measured by one or more IMUs. The one or more IMUs may be integrated into orthopedic prostheses 104, In some examples, the overall motion of orthopedic prostheses 104 provides information about scapular motion, while Hall sensor signals provide information about movement of the humerus relative to the scapula. In some examples, feature data 128 includes data characterizing a patient’s anatomy, shapes / positions of one or more orthopedic prostheses 104, demographic information about the patient (e.g., gender, age, weight, etc.), and so on.

[0041] In the example of FIG. 1, prediction system 116 includes a JRF generation unit 129 that uses feature data 128 to generate one or more JRF vectors 130. For ease of explanation, this disclosure may refer to the one or more JRF vectors 130 simply as JRF vectors 130. In some examples, processors within one or more of orthopedic prostheses 104 implement JRF generation unit 129 to generate JRF vectors 130.

[0042] Each of the generated JRF vectors 130 indicate a magnitude and direction of a force that is being applied by one of orthopedic prostheses 104 on the other one of orthopedic prostheses 104. The JRF vector of a joint changes over time, such as when a patient moves a limb (e.g., arm) and / or performs various ADLs. In some examples, JRF vectors 130 may be a force vector applied to orthopedic prosthesis 104A by orthopedic prosthesis 104B or a force vector applied to orthopedic prosthesis 104B by orthopedic prosthesis 104A. For example, one or more of JRF vectors 130 may be a force vector applied to a glenoid prosthesis by a humeral prosthesis. For example, for a humeral prosthesis and glenoid prosthesis, a JRF vector may be different when a patient has an arm extended over their head than when a patient’s arm is resting at their side and a JRF1262-250W001vector may be different when a patient is carrying a heavy load versus when a patient is not carrying anything.

[0043] JRF generation unit 129 may generate time-series data indicating a series of JRF vectors 130 over a period of time. In other words, prediction system 116 may obtain timeseries feature data 128 corresponding to different points in time or time spans. JRF generation unit 129 may use the time-series feature data 128 to generate the time-series data indicating the series of JRF vectors 130. In this way, JRF vectors 130 may describe the joint reaction force as the anatomical structures of the patient’s joint move.

[0044] In some examples, JRF generation unit 129 applies a JRF machine learning (ML) model 126 to feature data 128 to generate the one or more JRF vectors 130. For example, JRF ML model 126 may comprise a neural network model and JRF generation unit 129 may provide feature data 128 as input to the neural network. An example architecture of JRF ML model 126 is described below with respect to FIG. 8.

[0045] In some examples, prediction system 116 includes a musculoskeletal model (MSM) 125. MSM 125 is a physics-based model that represents bones and soft tissue structures (e.g., bones, tendons, ligaments, cartilage, etc.) of the patient. MSM 125 may include attributes that provide information about the bones and soft tissue structures. Example types of atributes may include bone density, estimated muscle strength, estimated elasticity, fatty infiltration ratios, and so on. The sizes and shapes of the bones, bone density, and values of other attributes may be estimated based on medical imaging, such as x-rays, computed tomography (CT) slices, magnetic resonance imaging, and so on. Other attributes maybe determined based on physical examination of the patient, such as by measuring the patient’s height and weight, measuring the physical abilities of the patient to perform specific motions or exercises, and other preoperative planning data. Prediction system 116 may refine MSM 125 based on feature data 128. For example, prediction system 116 may adjust values of attributes of MSM 125 such that simulated movements within MSM 125 are more consistent with feature data 128. In some examples, JRF generation unit 129 uses MSM 125 to generate the one or more JRF vectors 130. For instance, JRF generation unit 129 may use an inverse dynamic approach with respect to MSM 125 to predict the JRF vectors for various motions. In other words, JRF generation unit 129 may obtain kinematic data, such as feature data 128, that may indicate positions, velocities, and accelerations of orthopedic prostheses 104, JRF generation unit 129 may then use the kinematic data and MSM 125 to calculate the JRF1262-250W001vectors by using equations of motion in reverse, starting from the observed motion and working backward to determine forces that cause the observed motion.

[0046] Prediction system 116 may generate a prediction regarding orthopedic prostheses 104 based at least in part on JRF vectors 130, For example, prediction system 116 may generate a prediction regarding orthopedic prostheses 104 based on at least the one or more JRF vectors 130 having a magnitude and / or direction that satisfy a corresponding particular orthopedic prosthesis event threshold and based on at least one of the JRF vectors 130. In other words, prediction system 116 may determine whether one or more of a magnitude or a direction of the one or more JRF vectors 130 satisfy a particular orthopedic prosthesis event threshold. In response to a determination that one or more of a magnitude or a direction of the one or more JRF vectors satisfy’ the particular orthopedic prosthesis event threshold, prediction system 116 may generate a prediction regarding the orthopedic prosthesis implanted in the joint of the patient having a particular orthopedic prosthesis event. In other words, prediction system 116 may generate the prediction in response to a determination that one or more of the magnitude or the direction of the one or more JRF vectors 130 satisfy the particular orthopedic prosthesis event threshold, wherein the prediction is that the orthopedic prosthesis implanted in the joint of the patient will have a particular orthopedic prosthesis event. For instance, prediction system 116 generate a prediction of excessive wear will occur on articulating surfaces of one or more of orthopedic prostheses 104, a prediction that bone remodeling will occur, and / or a prediction that loosening of one or more of orthopedic prostheses 104 will occur based on JRF vectors 130 having a magnitude that satisfy an excessive force threshold. The ability to predict adverse events such as subluxation or loosening based on objective force data provides the technical benefit of significantly reducing surgical revision rates. By flagging these risks postoperatively during activities of daily living (ADL), the system allows for early clinical intervention, such as activity modification or targeted physical therapy, before the mechanical condition of the joint degrades to a point requiring additional surgery.

[0047] In some examples, prediction system 116 generates a prediction of an end of a lifespan of one or more of orthopedic prostheses 104. For instance, to generate the prediction of the end of the lifespans of orthopedic prostheses 104, prediction system 116 may use an empirically determined lookup table that maps the magnitudes of JRF vectors to rates of erosion of articulating surfaces of orthopedic prostheses 104. Prediction system 116 may use these rates of erosion and data indicating dimensions of articulating surface1262-250W001elements or orthopedic prostheses 104 to determine a time at which a thickness of the articulating surface element is less than a minimum thickness threshold.

[0048] In some examples, prediction system 116 uses a prediction machine learning (ML) model 131, such as a regression model or a neural network model, that outputs an end-of-lifespan prediction based at least in part on JRF vectors 130. In some examples, prediction system 116 may provide JRF vectors 130 together with joint kinematic predictions from MSM 125 as input into FEA model 124 to predict implant wear or loosening outputs at an end-of-lifespan state based on the sensed movements and JRF through the sensors 118. For example, the JRF vectors 130 may be used as a load in FEA model 124 and the joint kinematic predictions from MSM 125 may be used as boundary conditions for FEA model 124. FEA model 124 may provide predictions of the micromotion or implant wear for these given loading and kinematic conditions. Information about the patient, such as a CT scan would be used to set up the bone material properties and shape and the implant material properties are also fed into FEA model 124, In some examples, FEA model 124 can be part of an ML model that predicts implant wear using the so-called foundation models in machine learning.

[0049] In some examples, training system 132 may use predictions generated using FEA model 124 to train prediction ML model 131, For example, training system 132 may use end-of-lifespan predictions based on FEA model 124 as training data to train prediction ML model 131 to generate end-of-lifespan predictions. In this example, training system 132 may obtain training data examples. Each of the training data examples may specify a set of JRF vectors and an end-of-lifespan prediction based on FEA model 124. In this example, training system 132 may use the training data examples to train prediction ML model 131 to generate the end-of-lifespans predictions based on sets of JRF vectors.

[0050] In an example where prediction ML model 131 is a neural network model, prediction ML model 131 may include a multi-layer perceptron (MLP) having one or more hidden layers, input neurons of the MLP may receive data include JRF vectors, data characterizing one or more aspects of MSM 125 (e.g., tissue elasticity, tissue strength, etc.), data characterizing orthopedic prostheses 104, and so on. Output neurons of the MLP may include different output neurons for different predictions (e.g., different end- of-lifespan predictions).

[0051] In some examples, prediction system 116 generates a prediction regarding an amount of loosening of one or more of orthopedic prostheses 104 based on one or more JRF vectors 130. Loosening of an orthopedic prosthesis refers to movement of the1262-250W001orthopedic prosthesis relative to a bone on which the orthopedic prosthesis is implanted. For example, the bone density of a bone on which one of orthopedic prostheses 104 is implanted is below a first threshold and a magnitude of one or more of JRF vectors 130 is above a second threshold, prediction system 116 may generate a prediction that loosening of the orthopedic prosthesis will occur. In some examples, prediction system 116 uses FEA model 124 to predict amounts of loosening. FEA model 124 may be configured based on JRF vectors 130, joint kinematics, and bone data. For example, FEA model 124 may be configured based on the bone data to include elements corresponding to bones; attributes of elements corresponding to bones may be set based on bone data (e.g., bone density); attributes of elements corresponding to soft tissue structures may be set using an inverse dynamics approach so that FEA model 124 produces results consistent with JRF vectors 130 and the joint kinematics. Furthermore, in some examples, prediction system 116 may train an ML model (e.g., prediction ML model 131) to predict amounts of loosening based on training examples generated using predictions of amounts of loosening generated using FEA model 124.

[0052] In some examples, insufficient force exerted by one prosthesis on another prosthesis may lead to dislocation. In some examples, prediction system 116 generates a prediction that one of orthopedic prostheses 104 will exert insufficient force on the other one of orthopedic prosthesis 104 based on JRF vectors 130 having a magnitude that satisfy an insufficient force threshold. In some examples, prediction system 116 generates a prediction that a dislocation of a joint may occur based on the prediction that one of orthopedic prostheses 104 will exert insufficient force on the other one of orthopedic prosthesis 104.

[0053] In some examples, prediction system 116 generates a prediction that relates to a dislocation of a joint based on JRF vectors 130. For example, prediction system 116 may generate a prediction that a dislocation of the joint is likely to occur if an angle of one or more of JRF vectors 130 exceeds a predetermined angular threshold. Similarly, in some examples, prediction system 116 generates a prediction that relates to a subluxation of a joint based on JRF vectors 130. For example, prediction system 116 may generate a prediction that a subluxation of the joint is likely to occur if an angle of one or more of JRF vectors 130 exceeds a predetermined angular threshold. In some examples, prediction system 116 may use MSM 125 to predict muscle forces and allow the user to analyze contributions of different muscle groups in joint stability. For instance, prediction system 116 may use MSM 125 to predict whether there are imbalances between muscle force1262-250W001contributions. That is, MSM 125 may include elements correspond to individual muscles and that indicate forces applied by various muscles of a joint. By analyzing the muscle group contributions in conjunction with the real-time JRF vectors 130, the techniques of the disclosure allow for the identification of specific muscle groups prone to premature fatigue or chronic imbalance. Identifying these imbalances enables the prediction system 116 to generate a specialized rehabilitation protocol that targets antagonistic muscle pairs. This approach provides a technical advantage by preventing eccentric loading on the articulating surfaces of the orthopedic prostheses 104, thereby extending the mechanical lifespan of the prosthesis system. Prediction system 116 may compare the forces applied by the muscles on either side of the joint to determine whether there are imbalances in the muscle force contributions, and hence whether there is joint instability.

[0054] In some examples, prediction system 116 uses a physics-based model, such as an FEA model 124, to generate a prediction. FEA model 124 may represent a patient's bones, prostheses 104, and soft tissue structures (e.g., tendons, ligaments, muscles). Thus, in some examples, a first orthopedic prosthesis and a second orthopedic prosthesis are implanted in the joint such that the first orthopedic prosthesis articulates with respect to the second orthopedic prosthesis and prediction system 116 uses FEA model 124 of the joint to perform a finite element analysis of the joint. In this example, FEA model 124 of the joint is based at least in part on a model of the first orthopedic prosthesis, a model of the second orthopedic prosthesis, and the one or more JRF vectors. In some examples, elements representing bones and prostheses 104 may be modeled as rigid elements.

[0055] Soft tissue structures may be modeled as linear elastic elements that connect discrete points on the bone elements. In some examples, elements representing the soft tissue structures (i.e., soft tissue structure elements) may have a set of attributes. Values of these attributes may be representative of strength and elasticity of the represented soft tissue structures. In some examples, prediction system 116 may generate elements of FEA model 124 based on imaging scans, such as a CT scan or an MRI scan, of a patient’s anatomy. In some examples, the imaging scans may include preoperative scans and / or postoperative scans. In some examples, values of attributes of soft tissue structure elements are determined based on factors such as the patient’s age, gender, weight, measured strengths of the soft tissue structured, measured sizes of the soft tissue structures, and so on. In some examples where an MRI scan of the patient’s anatomy is available, prediction system 116 may determine values of one or more attributes of one1262-250W001or more soft tissue structures based on sizes of muscle fibers of the soft tissue structures measured in the MRI scan.

[0056] In some examples, prediction system 116 adjusts values of one or more attributes of the soft tissue structure elements of FEA model 124 based on JRF vectors 130. For example, prediction system 116 may apply FEA model 124 with various combinations of values of the one or more attributes of the soft tissue structure elements to determine sets of trial JRF vectors. The trial JRF vectors are estimated JRF vectors as determined using FEA model 124. Prediction system 116 may compare the sets of trial JRF vectors to JRF vectors 130 generated by JRF generation unit 129. The values of the attributes that result in the set of trial JRF vectors that most closely match JRF vectors 130 are likely to accurately represent the values of attributes of the soft tissue structures in the patient. Thus, prediction system 116 may determine values of one or more attributes of one or more soft tissue structure elements in FEA model 124 based on a comparison of the trial JRF vectors and JRF vectors 130. In other words, prediction system 116 may determine the values of the one or more atributes of the one or more soft tissue structure elements as the values of the one or more attributes of the one or more soft tissue structure elements used to determine the determined set of trial JRFs vectors. By determining values of attributes of the soft tissue structural elements in this way, prediction system 116 may increase the accuracy of FEA model 124. Increasing the accuracy of FEA model 124 may result in improved predictions regarding orthopedic prostheses 104.

[0057] Furthermore, in some examples, prediction system 116 performs a multi-variable optimization process to determine the values of the attributes of the soft tissue structure elements. In some such examples, prediction system 116 may start from initial values of the attributes. Prediction system 116 may determine the initial values of the attributes based on the patient’s age, gender, measured strength, and so on. In some examples, a user of prediction system 116 manually sets one or more of the initial values. Prediction system 116 may use the initial values of tire attributes in FEA model 124 and determine a set of trial JRF vectors. Different trial JRF vectors in the set of trial JRF vectors may correspond to orthopedic prostheses 104 being at different positions relative to one another. Prediction system 116 may apply a difference function that determines a difference value based on the trial JRF vectors and corresponding JRF vectors 130. The difference function may be a sum of squared differences of angles and magnitudes of corresponding trial JRF vectors and JRF vectors 130. Prediction system 116 may determine a partial differential equation of the loss function with respect to each of the1262-250W001atributes, and thereby determine a slope of the partial differential function with respect to the attribute at the value of the attribute. For each of the attributes, prediction system 116 may modify the value of attribute in a direction determined based on the slope of the partial differential equation with respect to the attribute. Prediction system 116 may repeat this process multiple times. In other examples, prediction system 116 may use other optimization processes to determine the values of the attributes.

[0058] A patient may perform various ADLs. The ADLs are activities that the patient ■would perform during their daily lives. In some examples, a patient wears a virtual reality (VR) headset that shows the patient how to perform the ADLs. Integrating the virtual reality (VR) headset with prediction system 116 provides the technical benefit of enhanced patient compliance and real-time biomechanical correction during postoperative rehabilitation. By visually guiding the patient through activities of daily living (ADL) while simultaneously tracking the resulting joint reaction force vectors 130 via sensors 118, the system can provide immediate, closed-loop feedback to the patient. This ensures the patient performs the exercises within safe kinematic boundaries, reducing the risk of inadvertently causing impingement or edge loading on orthopedic prostheses 104 while at home. JRF generation unit 129 may generate JRF vectors 130 based on feature data 128. Feature data 128 may be generated based on signals generated by sensors 118 while the patient is performing ADLs. In some examples, electromyography (EMG) sensors may additionally be positioned on the patient to sense EMG signals while the patient is performing the ADLs. In some examples, prediction system 116 may refine the values of attributes of the soft tissue structures, such as ligaments and / or tendons activations sensed through the EMG signals, in FEA model 124 based on the sensed EMG signals while patient is performing ADLs. The sensed EMG signals together with muscle elongation predictions through the MSM 125 would allow predictions of the muscle forces. In some examples, a patient holds a tracking sensor, a tracking system, such as a VR headset, is configured to track as the patient performs ADLs which may help system 100 to correlate feature data 128 that is based on signals generated by sensors 118 with specific positions of a patient’s limb (e.g., arm).

[0059] Prediction system 116 may use FEA model 124 to generate one or more predictions regarding orthopedic prostheses 104. For example, prediction system 116 may generate a recommendation on how to reduce the likelihood of an event from occurring. For instance, prediction system 116 may determine, based on FEA model 124, that a particular soft tissue structure, such as a particular tendon (and therefore associated1262-250W001muscle) is relatively weak. For instance, prediction system 116 may determine, based on the values of attributes of soft tissue structure elements, that a strength of the soft tissue structure is below a specific threshold or that the elasticity of the soft tissue structure is above another threshold. Relative weakness of the particular soft tissue structure may increase the likelihood of occurrence of an event involving one or more of orthopedic prostheses 104, such as an impingement or dislocation. Accordingly, prediction system 116 may generate a recommendation of exercises a patient should perform to strengthen a muscle associated with the particular soft tissue structure. For instance, prediction system 116 may generate a recommendation of exercises the patient should perform to strengthen the muscle based on the muscle being relatively weak. In some examples, prediction system 116 may generate a recommendation that the patient avoid activities that rely on a muscle associated with the particular soft tissue structure.

[0060] Using FEA model 124, orthopedic prostheses 104 and supporting soft tissue structures may be represented as a system of finite elements. Prediction system 116 may use FEA model 124 to estimate a displacement of orthopedic prosthesis 104A relative to orthopedic prosthesis 104B when various forces (loads) are applied to a particular bone (e.g., a humerus, tibia, etc.) of a limb when the limb is at various positions. Prediction system 116 may generate a prediction that a subluxation is likely to occur if the displacement is greater than a predefined subluxation threshold. Prediction system 116 may generate a prediction that a dislocation is likely to occur if the displacement is greater than a predefined dislocation threshold. In some examples, the predefined dislocation threshold may correspond to a rim of the joint, such as a rim of a glenoid or acetabulum, In some examples, the predefined dislocation threshold corresponds to other limits.

[0061] In some examples, prediction system 116 may use FEA model 124 to generate a prediction that impingement is likely to occur. For instance, prediction system 116 may use the FEA model 124 to determine maximum ranges of motion of the joint in various directions. The maximum ranges of motion of the joint may be limited by the values of the attributes of the soft tissue structures as well as the shapes of the bone elements and the prosthesis elements. Prediction system 116 may determine distances between bone elements and / or distances between prosthesis elements and bone elements when the joint is at its maximum range of motion. If one or more of such distances is less than an applicable predetermined impingement threshold, prediction system 116 may generate a prediction that impingement is likely to occur.1262-250W001

[0062] In some examples, prediction system 116 uses FEA model 124 to estimate safe loads for orthopedic prostheses 104. For instance, prediction system 116 may use FEA model 124 as described above to predict whether subluxation, dislocation, or impingement is likely to occur but while simulating a load being applied to a bone of the joint. In some examples, if such loads result in subluxation, dislocation, or impingement, prediction system 116 may run the FEA again with different values of attributes of muscles of the joint (e.g., a particular muscle being stronger) and determine again whether the subluxation, dislocation, or impingement occurs. Through repeatedly adjusting values of attributes of the muscles, prediction system 116 may determine which of the muscles should be strengthened through exercise. In some examples, prediction system 116 use FEA model 124 to generate recommendations on particular exercises to perform preoperatively to strengthen muscles before implantation of particular prostheses at a particular location based on simulated loads of the particular prostheses at the particular location.

[0063] In some examples, prediction system 116 uses FEA model 124 to run simulations of simulated loads being applied to a patient’s limb (e.g., arm, leg, etc.) at a particular position to determine what particular applied loads (e.g., magnitude and / or direction) on the patient’s limb may cause pressures or excursions on predefined boundaries that help cause a prediction of an event involving one or more of orthopedic prostheses 104. In some examples, in response to a determination by FEA model 124 that excessive pressure or excursion from predefined boundaries occurs when particular simulated loads are applied to a patient’s limb (e.g., arm) when the patient’s limb is at a particular position, prediction system 116 may generate a recommendation that the patient not perform particular activities that may apply loads to the patient's limb that correspond to tire simulated loads (e.g., carrying heavy items, holding children, swinging a tennis racket overhead, etc.). In some examples, the loads may be static (e.g., holding a heavy item) or dynamic (swinging an item).

[0064] In some examples, prediction system 116 may use FEA model 124 to generate a prediction regarding an end of lifespan of one or more of orthopedic prostheses 104. For instance, prediction system 116 may use FEA model 124 while simulating application of various forces associated with ADLs on the bone elements and / or prosthesis elements. From such simulations, prediction system 116 may determine an accumulated force applied to components of prostheses, such as metallic components of the prostheses. Levels of fatigue of the components of prostheses may have predefined relationships with1262-250W001the accumulated forces applied to the components. Accordingly, prediction system 116 may determine, based on such predefined relationships, an amount of time that is likely to elapse before a level of fatigue exceeds a threshold. In a similar way, prediction system 116 may use FEA model 124 to generate predictions regarding loosening or wear.

[0065] In some examples, prediction system 116 may use FEA model 124 to generate simulations of the patient’s anatomy with various particular sets and / or positions of prostheses and generate predictions of events for the particular sets and / or positions of prostheses. For example, prediction system 116 may use FEA model 124 to generate a prediction that, for a first set of prostheses, a patient has a likelihood to experience an impingement above an impingement threshold but, for a second set of prostheses, the patient has a likelihood to experience an impingement that is below an impingement threshold, and then generate a recommendation to use the second set of prostheses. In another example, prediction system 116 may execute FEA model 124 to generate a prediction that, for a first position of prostheses, a patient has a likelihood to experience an impingement above an impingement threshold but, for a second position of prostheses, the patient has a likelihood to experience an impingement below an impingement threshold, and generate a recommendation to use the second position of prostheses.

[0066] In some examples, prediction system 116 may use FEA model 124 with multiple combinations of prostheses (and / or positions of prostheses) during preoperative planning. For each of the combinations, prediction system 116 may use information generated in the FEA of the combination to generate predictions regarding the orthopedic prostheses of the combination. Prediction system 116 may generate the predictions regarding the orthopedic prostheses of the combination in the same manner as described elsewhere in this disclosure with respect to generating predictions. Prediction system 116 may use the predictions to generate a recommendation of a set and position of prostheses that includes the lowest chances of events (e.g., impingement, subluxation, dislocation, etc.) for the joint and / or prostheses.

[0067] In some examples, prediction system 116 may use a recommendation ML model 127 to generate a recommendation of a set and position of prostheses that includes the lowest chances of events (e.g., impingement, subluxation, dislocation, etc.) for the joint and / or prostheses. For example, during a preoperative planning process for implanting a candidate orthopedic prosthesis of a candidate orthopedic prosthesis system, prediction system 116 may apply recommendation ML model 127 to input for a patient to generate a prediction regarding the candidate orthopedic prosthesis system and output an indication1262-250W001of a recommended orthopedic prosthesis system based on the prediction regarding the candidate orthopedic prosthesis system. In some examples, a candidate orthopedic prosthesis system is a combination of different prosthesis types and / or positions that are under consideration to be used on a different patient, such as a second patient. In some examples, the input to be input to recommendation ML model 127 may include different prosthesis types, different prosthesis positioning, and parameters of the second patient, such as second patient’s age, gender, weight, measured strengths of the soft tissue structured, measured sizes of the soft tissue structures, and so on. In some examples, in response to applying recommendation ML model 127 to the input, recommendation ML model 127 may determine a score for each prostheses combination based on resistance to dislocation, impingement, length of like, resistance to loosening, etc., and select a candidate orthopedic prosthesis system having the best score.

[0068] Storage system 110 may also store a training system 132. In some examples, training system 132 may train ML models, such as JRF ML model 126 and / or recommendation ML model 127 used by prediction system 116. In other examples, training system 132 is implemented in a different computing system from computing system 102.

[0069] In some examples, prediction system 116 may obtain a plurality of training examples where each of the training examples includes a set of training input data and ground-truth data. In some examples, the training input data may include parameters of prosthesis systems implanted in patients and anatomic characteristics of the patients. The parameters of prosthesis systems may include positions and types of prostheses. In some examples, the ground-truth data may include events that involve the prosthesis systems implanted in the patients. In some examples, the training input data includes parameters of orthopedic prostheses 104 implanted in a first patient and the ground-truth data of the training example may include the generated prediction regarding an event involving one or more of orthopedic prostheses 104. In some examples, training system 132 trains recommendation ML model 127 based on the training examples. In some examples, prediction system 116 applies recommendation ML model 127 to input data for a second patient to generate a prediction regarding a candidate orthopedic prosthesis for the second patient.

[0070] In some examples, training examples are generated in which the different prostheses types, different prostheses positioning, and patient parameter data, such as second patient’s age, gender, weight, measured strengths of the soft tissue structured,1262-250W001measured sizes of the soft tissue structures, is training input data and the generated score or predictions are training ground-truth data. During training, the training input data of a training example is provided as input to recommendation ML model 127 to generate an output score or prediction of the candidate orthopedic prosthesis. An error function may then be applied to the output score or prediction and the training ground-truth data of the training example. The resulting error value may be used as a basis for backpropagation for updating parameter values of recommendation ML model 127. This may be repeated for multiple rounds using the training examples.

[0071] In some examples, the training examples may include training input data that includes parameters of a prostheses in the first patient and anatomic characteristics of the first patient, and ground-truth data of the training example is the prediction regarding the orthopedic prostheses implanted in the first patient.

[0072] In some examples, prediction system 116 generates patient specific recommendations for the second patient based on the generated prediction regarding an event involving a candidate orthopedic prosthesis for the second patient. In some examples, the recommendations include particular exercises to perform during the preoperative period to strengthen particular muscles as strengthening the particular muscles may prevent or reduce the chances the event involving a candidate orthopedic prosthesis for the second patient occurs after surgery.

[0073] In some examples, orthopedic prosthesis 104A is a humeral prosthesis that includes an articulation element and a base element, and orthopedic prosthesis 104B is a glenoid prosthesis that includes a baseplate and an articulation element. The articulation element of the humeral prosthesis provides an articulation surface that slides over a corresponding articulation surface of the glenoid prosthesis. The articulation surface may be convex or concave. For example, the articulation surface may be convex if an anatomic TSA surgery' is performed or concave if a reverse TSA surgery' is performed. The base element of humeral prosthesis may secure humeral prosthesis to the patient’s humerus.

[0074] In some examples, a glenoid prosthesis also includes an augment element. The baseplate may comprise a disk-shaped metal element. In other examples, the baseplate may have other shapes. The augment element may be attached to a bone-facing side of the baseplate. The augment element may occupy a space between the baseplate and the bone. In some examples, the augment element has a patient-specific shape. In some examples, the augment element has wedge-shaped bone-facing surface. In some examples, the augment element has a flat bone-facing surface and a wedge-shaped bone-1262-250W001facing surface. The baseplate may define one or more screw holes. After the placement of the baseplate and augment element onto the patient's scapula, the surgeon may fix the baseplate and augment element to the patient's scapula by passing fixation screws into the patient’s scapula through the screw holes of the baseplate. In some instances, one or more of the fixation screws also pass through corresponding screw holes defined in the augment element.

[0075] During a total shoulder arthroplasty', a surgeon prepares the patient’s humerus and scapula to receive a humeral prosthesis and a glenoid prosthesis. For example, the surgeon may resect a head of the patient’s humerus along a preplanned cutting plane. In some examples, the surgeon may sound, punch, and / or compact cancellous bone tissue within an intramedullary canal of the humerus to make room for a stem of the humeral prosthesis. In some examples, to prepare the scapula to receive the glenoid prosthesis, the surgeon may ream one or more areas of the glenoid fossa, form a hole in the scapula for a central fixation element (e.g,, a peg or keel) of the glenoid prosthesis or otherwise modify the scapula to receive the glenoid prosthesis. In some examples, preparing the humerus or scapula may involve removing osteophytes from the humerus or scapula.

[0076] In some examples, prediction system 116 outputs indications of JRF vectors 130 and / or generated prediction(s) to a user interface for display to a user. The user interface may also display a virtual first orthopedic prosthesis (e.g., humeral) model relative to a virtual first bone (e.g., humerus) model. The virtual first bone model is a representation of at least a portion of a particular bone of a patient, such as a humerus. The virtual first orthopedic prosthesis model is a representation of orthopedic prosthesis 104A (e.g., humeral prosthesis) to be implanted on a first bone (e.g., humerus). The user interface may display a virtual second orthopedic prosthesis (e.g., glenoid) model relative to a virtual second bone (e.g., scapula) model. The virtual second bone model is a representation of at least a portion of a particular bone of a patient, such as a scapula. The virtual second orthopedic prosthesis model is a representation of orthopedic prosthesis 104B (e.g., glenoid prosthesis) to be implanted on the scapula.

[0077] The virtual first orthopedic prosthesis model and / or first orthopedic prosthesis model and the respective virtual first bone and / or second bone model may be 3- dimensional (3D) models, in such examples, a 3D virtual bone model may be generated based on a computed tomography (CT) scan and / or other ty pe of medical imaging of the particular bone. In some examples, the first orthopedic prosthesis model and / or first orthopedic prosthesis model and the respective virtual first bone and / or second bone1262-250W001model may additionally or alternatively be 2-dimensional (2D) models. In such examples, the respective virtual bone model be displayed as a 2D slice of the particular bone, and the particular 2D virtual orthopedic prosthesis model may be displayed as an outline of the corresponding orthopedic prosthesis.

[0078] Prediction system 116 may receive indications of user input to adjust values of various parameters of one or more of orthopedic prostheses 104. Example parameters may include types of prostheses, positions of prostheses, orientations of prostheses, and so on. Different types of prostheses are associated with different parameters. For example, in some examples where orthopedic prosthesis 104B is a glenoid prosthesis, parameters of the glenoid prosthesis in a total shoulder arthroplasty may include one or more of a lateralization / medialization of glenoid prosthesis, an inclination of glenoid prosthesis, a version / anteversion of glenoid prosthesis, a size of an articulating surface (e.g., radius of a glenosphere, diameter of a concave articulating surface, etc.), eccentricity of the articulating surface, and so on. For example, in some examples where first orthopedic prosthesis 104 is a humeral prosthesis, some example parameters of humeral prosthesis may include one or more of a size of an articulation element of the humeral prosthesis, an angle of a base of the humeral prosthesis, a fixation type of the humeral prosthesis (e.g., stemmed or stemless), and so on.

[0079] In some examples, prediction system 116 may automatically recommend values of one or more parameters of one or more of orthopedic prostheses 104. An example process for automatically recommending parameters of humeral prosthesis and glenoid prosthesis is described in PCT publication WO 2022 / 169678, filed January 28, 2022, the entire content of which is incorporated by reference. In some examples, prediction system 116 may check whether a selected humeral prosthesis is compatible with a selected glenoid prosthesis.

[0080] FIG. 2 is a block diagram illustrating example components of a prosthesis 200, in accordance with one or more techniques of this disclosure. Prosthesis 200 may be one of orthopedic prostheses 104 (FIG. 1). In some examples, prosthesis 200 may be a glenoid prosthesis, a humeral prosthesis, or another type of orthopedic prosthesis. For ease of explanation, mechanical components of prosthesis 200 (e.g., a baseplate, articulation element, augment element, stem, fixation elements, etc.) are omitted from prosthesis 200 in FIG. 2.

[0081] In the example of FIG. 2, prosthesis 200 includes a power source 202, a Hall sensor array 204, an inertial measurement sensor (IMU) 206, a memory 208, one or more1262-250W001processors 210, and a communication unit 212. In other examples, prosthesis 200 may include one or more additional components. For instance, in some examples, prosthesis 200 may include a temperature sensor configured to measure a body temperature of a patient. Furthermore, in some examples, one or more components shown in FIG, 2 are omitted. For instance, in some examples, prosthesis 200 does not include IMU 206 or processors 210. Hall sensor array 204, IMU 206, memory 208, processors 210, and / or communication unit 212 may be communicatively coupled directly or indirectly such that information may be communicated between such components.

[0082] Power source 202 includes one or more batteries or other devices configured to store energy. Power source 202 is configured to provide electrical energy to Hall sensor array 204, IMU 206, memoiy 208, processors 210, communication unit 212, and / or other components of prosthesis 200 via power distribution channels (not shown).

[0083] Hall sensor array 204 is a sensor that incorporates one or more Hall elements, each of which produces a voltage proportional to one axial component of a magnetic field vector using the Hall effect. Tire voltages may represent a magnetic field intensity observed in the x, y, and z axes, e.g., in milli-Tesla (mT). Hall sensor array 204 is configured to generate Hall signals based on the voltages caused by magnetic flux. Each respective Hall sensor of Hall sensor array 204 may generate a respective Hall signal. The Hall signals generated by Hall sensor array 204 may represent movement of prosthesis 200 relative to another prosthesis (e.g., a movement of orthopedic prosthesis 104A relative to orthopedic prosthesis 104B). Hall sensor array 204 may output data via a data interface, such as a Serial Peripheral Interface (SPI) interface, an I2C interface, or another type of interface. IMU 206 is configured to measure a specific force and angular rate of prosthesis 200. IMU 206 may include one or more accelerometers and gyroscopes. IMU 206 is configured to generate one or more IMU signals that represent bias from motion, such as humeral-glenoid bias from scapular motion. Memory 208 is configured to store data, such as data representing the Hall signals and the IMU signals.

[0084] Processors 210 are configured to process data, such as data representing the Hall signals and the IMU signals. Processors 210 may be implemented in any of the ways described above with respect to processors 108 (FIG. 1). Processors 210 are configured to generate position information based on the data representing the Hall signal and / or the IMU signals. In other examples, processors 108 (FIG. 1) may generate or obtain feature data 128 based on the data representing the Hall signal and / or the IMU signals. In other examples, processors 210 may generate feature data 128. In some examples,1262-250W001communication unit 212 of prosthesis 200 may communicate the feature data 128 to computing system 102.

[0085] In some examples, processors 210 may implement a ML model 214, such as a neural network, that is configured to generate JRF vector(s) 130 based on feature data 128. In some examples, ML model 214, as shown in FIG. 2, may correspond to ML model 126, as shown in FIG 1. In examples where ML model 214 is implemented as a neural network, the neural network may include an input layer, an output layer, and one or more hidden layers between the input layer and the output layer. For each layer of the input layer and hidden layers, output of the layer is input to the next layer of the neural network. Individual neurons generate output based on application of an activation function to a weighted sum of input values. The activation function may be a Rectified Linear Unit (ReLU) activation function, a sigmoid activation function, or another type of activation function. Output of the output layer may indicate JRF vector(s) that indicate a direction and / or magnitude of force / pressure that is being applied from one prosthesis (e.g., orthopedic prosthesis 104A) on another prosthesis (e.g., orthopedic prosthesis 104B). FIG. 8, which is described in greater detail below, provides an example architecture of a neural network that may implement ML model 214. The neural network may take the feature data representing the Hall signals as input. The neural network may output JRF vector(s) 130.

[0086] The neural network may be trained based on simulation data generated using FEA. For example, feature data may be generated based on signals from sensors (e.g., Hall sensor arrays) of orthopedic prostheses implanted in multiple patients, FEA models (e.g., FEA model 124) may be constructed for the patients. JRF vectors may be estimated for the patients using the FEA models constructed for the patients as elements representing orthopedic prostheses are moved relative to one another. Training examples are generated in which the feature data is training input data and the estimated JRF vectors are training ground-truth data. During training, the training input data of a training example is provided as input to ML model 214 to generate an output JRF vector. An error function may then be applied to the output JRF vector and the training ground-truth data of the training example. The resulting error value may be used as a basis for backpropagation for updating parameter values of ML model 214. This may be repeated for multiple rounds using the training examples.

[0087] Tire ML model may utilize specific hyperparameters and architectures optimized for correlating sensor-based feature data with the magnitude and direction of joint1262-250W001reaction force vectors. In some examples, the training system may implement a learning rate between 0.0001 and 0.01 and a batch size ranging from 16 to 256 training examples. The hidden layers of the neural network may include between 32 and 1024 neurons per layer, with the number of hidden layers selected to balance computational efficiency and prediction accuracy.

[0088] To improve the robustness of the model, the training system may apply data augmentation techniques to the training input data, such as injecting Gaussian noise into the signal data or applying random rotations to the ground-truth kinematics data. The loss function used during training may include a mean squared error (MSE) loss for generating the joint reaction force vectors and a cross-entropy loss for classification-based predictions, such as the likelihood of impingement or dislocation. Furthermore, preprocessing of the feature data may include Z-score normalization or min-max scaling to ensure consistent weighting across different sensor modalities, such as Hall signals and IMU signals. The ML model may also implement dropout layers with a dropout rate between 0.1 and 0.5 to prevent overfitting to the synthetic training examples generated by the FEA models.

[0089] Thus, in some examples, training system 132 may use a FEA model to generate training data and may use the training data to train a second ML model to generate predictions regarding the orthopedic prosthesis. In some examples, ML model 214 is initially pretrained for a generic patient and then further trained for an individual patient based on training examples from the FEA models. This may accelerate training ML model 214 to be specialized for an individual patient.

[0090] JRF vector(s) 130 provide information about a direction and / or magnitude of force / pressure that is being applied from one prosthesis (e.g., orthopedic prosthesis 104A) on another prosthesis (e.g., orthopedic prosthesis 104B). JRF vector(s) 130 may change overtime, such as when a patient moves a limb (e.g., ami) and / or performs various ADLs. For example, for a humeral prosthesis and glenoid prosthesis, a JRF vector 130 may be different when a patient has an arm extended over their head than when a patient’s arm is resting at their side and a JRF vector 130 may be different when a patient is carrying a heavy load versus when a patient is not carrying anything. In some examples, processors 210 collect a plurality of JRF vector(s) 130 over a period of time and determine various features, such as a stability ratio, based on the JRF vector(s) 130. In some examples, processors 210 may collect additional data, such as joint angle abduction, and determine features, such as a stability ratio, based on the JRF vector(s) and joint angle abduction1262-250W001over a period of time as the joint moves through various angles of motion. In some examples, processors 210 may generate a prediction of excessive wear on articulating surfaces of one or more of orthopedic prostheses 104, bone remodeling, and / or loosening of one or more of orthopedic prostheses 104 based on JRF vector(s) 130 having a magnitude and / or direction that satisfy an excessive force threshold. In some examples, processors 210 may generate a prediction of impingement, subluxation, and / or dislocation based on JRF vector(s) 130, over a period of time, having a change in direction that satisfy an impingement threshold, subluxation threshold, and / or dislocation threshold.

[0091] In some examples, feature data 128 provides information about a position of prosthesis 200. For instance, in an example where prosthesis 200 is a glenoid prosthesis, feature data 128 may include information about a position of a first orthopedic prosthesis relative to a second orthopedic prosthesis, such as a position of a humeral prosthesis to a glenoid prosthesis. In some examples, feature data 128 may provide information about the position of a particular bone of the patient’s, such as a scapula. Thus, in such examples, feature data 128 may include information indicative of a position of the particular bone (e.g., scapula) relative to the patient’s torso. In some examples, feature data 128 may include information indicating one or more of degrees of forward extension of the joint (e.g., shoulder joint), degrees of abduction / adduction of the joint, or degrees of internal / external rotation of the joint. In some examples, feature data 128 may indicate the position and motion of orthopedic prostheses 104 in 6 degrees of freedom (e.g., 3 spatial translation dimensions, anterior / posterior rotation, superior / inferior rotation, and axial rotation). In some examples, feature data 128 may include information indicating a distance between prosthesis 200 and a corresponding prosthesis (e.g., between a glenoid prosthesis and a humeral prosthesis).

[0092] In some examples, processors 210 of prosthesis do not implement ML model 214. Rather, an ML model that fulfills the role of ML model 214 may be implemented by processors 108 of computing system 102. Communication unit 212 of prosthesis 200 may transmit data to computing system 102 usable in the ML model.

[0093] Communication unit 212 is configured to transmit feature data, JRF vector(s) 130, signal data, or other data to one or more computing devices, such as a computing device of computing system 102 (FIG. 1). Communication unit 212 may be configured to communicate JRF vector(s) 130 and / or predictions regarding the orthopedic prosthesis using one or more types of wireless communication. For example, communication unit 212 may communicate JRF vector(s) 130 and / or predictions regarding the orthopedic1262-250W001prosthesis using a BLUETOOTH LOW ENERGY ™ wireless communication protocol. In some examples, communication unit 212 may use near field magnetic induction to communicate JRF vector(s) 130 and / or predictions regarding the orthopedic prosthesis. In some examples, communication unit 212 is configured to transmit other information, such as the status of power source 202, maximum and minimum positions, information about the time of motion, body temperature, and so on.

[0094] FIG. 3 is a schematic diagram of a glenoid prosthesis 300 and a humeral prosthesis 302, in accordance with one or more techniques of this disclosure. In some examples, orthopedic prosthesis 104A may be humeral prosthesis 302 and orthopedic prosthesis 104B may be glenoid prosthesis 300. In the example of FIG. 3, glenoid prosthesis 300 includes an articulation element 304, a baseplate 306, and a Hall element 308, Humeral prosthesis 302 includes an articulation element 310, a stem element 312, and electronic components 314. Articulation element 304 is coupled to baseplate 306. Articulation element 310 is coupled to stem element 312. Articulation element 304 and articulation element 310 may include polyethylene articulation surfaces configured to contact and slide over one another. Baseplate 306 defines one or more holes through which fixation elements (e.g., screws) may be passed in order to fix glenoid prosthesis 300 to a patient’s scapula. Stem element 312 is configured to attach humeral prosthesis 302 to the patient’s humerus. In some examples, stem element 312 extends into an intermedullary canal of the patient’s humerus. In other examples, humeral prosthesis 302 is stemless and humeral prosthesis 302 includes a baseplate for attaching humeral prosthesis 302 to the patient's humerus.

[0095] Electronic components 314 in humeral prosthesis 302 may include the components shown in the example of FIG. 2, such as a Hall sensor, IMU, processors, memory, communication unit, and so on. The Hall sensor may generate a voltage based on motion of Hall element 308 relative to the Hall sensor.

[0096] FIG. 4 is a conceptual diagram illustrating an example user interface 400 showing an animation, such as an animated diagram, of simulated movement of a joint with orthopedic prostheses 404 implanted in the joint. In some examples, prediction system 116 may output predictions regarding orthopedic prostheses 404 for display in user interface 400. For example, prediction system 116 may output a warning of a subluxation risk, dislocation risk, or an impingement risk,

[0097] FIG. 5 is a conceptual diagram illustrating an example user interface 500 for showing a track 502 of a JRF vector as a patient moves, in accordance with one or more1262-250W001techniques of this disclosure. Track 502 indicates a displacement of a current direction of the JRF vector relative to the other implant component (e.g., glenoid component). The default direction of the JRF vector may be a direction of the JRF vector while a joint is at rest. For instance, if the joint is the shoulder joint, the default direction of the JRF vector may be directed toward the patient’s glenoid or glenoid prosthesis. Prediction system 116 may animate the current direction of the JRF vector in user interface 500 as the patient's joint moves. Thus, prediction system 116 may output, for display on a display device, an animated diagram showing the one or more JRF vectors as the joint is positioned throughout a range of motion. In some examples, computing system 102 may output user interface 400 and user interface 500 for display concurrently so that a user may be able to see positions of the bones and prostheses with the corresponding current direction of the JRF vector as the joint moves through various positions. Since prediction system 116 may output warnings of subluxation, dislocation, or impingement risks, the user may be able to see in user interface 500 the JRF vectors that lead to such warnings.

[0098] In other examples, prediction system 116 may generate a user interface similar to user interface 500 except that track 502 indicates a displacement of a current direction of a muscle group force. For example, the track may indicate a muscle group force of deltoid muscles or rotator cuff muscles. Prediction system 116 or a user may determine, based on the track, muscle group contributions so that postoperative rehabilitations can be optimized. Prediction system 116 may use MSM 125 to provide predictions for all the muscle forces. In its post processing, MSM 125 may also provide groupings of the muscle forces, e.g., a grouping the forces of the posterior, middle, and anterior deltoids to deltoid and provide an overall deltoid force.

[0099] In some examples, prediction system 116 may generate user interface 500 and track 502 in real time based on feature data from orthopedic prostheses 104. In examples where one or more of orthopedic prostheses 104 is atrial, generating user interface 500 and track 502 in real time may help a surgeon during surgery to confirm whether a permanent prosthesis corresponding to the trial is appropriate for the patient. Generating user interface 500 and track 502 in real time may also be useful in a postoperative setting, e.g., to help a user develop a physical therapy plan while during a patient visit. Tire use of real-time joint reaction force vectors 130 provides a technical advantage by allowing a clinician to visualize the impact of specific movements on joint loading immediately. By observing track 502 on user interface 500, a clinician can identify if a particular range of motion causes edge loading or excessive predicted contact pressure. This objective1262-250W001feedback enables the clinician to refine a physical therapy plan during a patient visit, ensuring that prescribed exercises strengthen the joint while avoiding detrimental loads that could compromise the longevity of orthopedic prostheses 104.

[0100] FIG. 6 is a conceptual diagram illustrating an example user interface 600 showing a heat map indicating locations of predicted contact pressure 602 of one prosthesis on another prosthesis. In the example of FIG. 6, areas 604 of greater contact pressure are indicated with a different color or pattern than areas of lesser contact pressure. Additionally, user interface 600 may directly show edge loading.

[0101] In some examples, prediction system 116 may display an animation on user interface 400 (or an interface similar to user interface 400) to a user prior to surgery (i.e., preoperatively). In such examples, the source of information used by prediction system 116 to generate an animation on user interface 400 may include one or more devices other than implanted orthopedic prostheses 104. For example, the source of the information may include one or more cameras, wearable devices, or other types of devices. Based on the information provided by user interface 400, the user may recommend a presurgical plan for the patient. The presurgical plan may involve exercises to strengthen one or more muscles (e.g., rotator cuff muscles). In some cases, strengthening specific muscles may lead to better post-surgical outcomes, e.g., less risk of impingement, subluxation, or dislocation.

[0102] In some examples, prediction system 116 displays an animation on user interface 400 (or an interface similar to user interface 400) to a user after completion of the surgery (i.e., postoperatively). For example, prediction system 116 may display an animation on user interface 400 to a user (e.g., a practitioner or the patient) during the patient’s recovery or during physical therapy. ’Hie information provided by user interface 400 may help guide the course of the patient’s recovery or physical therapy. For instance, a practitioner may recommend the patient perform specific exercises to increase range of motion, decrease risk of subluxation or dislocation, or prevent other adverse outcomes. In some examples, prediction system 116 may generate a prediction regarding an adverse event involving one or more of orthopedic prostheses 104 based on the information or data received from the sensors after completion of the surgery. In some examples, prediction system 116 may then provide a recommendation on how to intervene to address the predicted adverse event. For instance, the user may use the information provided by an animation displayed on the user interface 400 that shows how one or more orthopedic prostheses 104 are functioning while implanted in patient to determine whether one or1262-250W001more of orthopedic prostheses 104 has failed and a revision surgery is needed based on how one or more orthopedic prostheses 104 are functioning while implanted in patient. In some examples, prediction system 116 may generate user interface 600 in real time based on feature data from orthopedic prostheses 104.

[0103] FIG. 7 is a flowchart illustrating an example operation 700 in accordance with one or more techniques of this disclosure. In the example of FIG. 7, one or more processors (e.g., processors 108, processors 210, etc.) obtain feature data 128 based on signals generated by one or more sensors 118 included in one or more of orthopedic prostheses 104 (702). In some examples, feature data 128 includes data characterizing a patient’s anatomy and / or shapes / positions of one or more of orthopedic prostheses 104. Sensors 118 may include one or more Hall sensors.

[0104] One or more processors (e.g., processors 108, processors 210, etc.) may apply a ML model (e.g., ML model 126, ML model 214, etc.) to feature data 128 to generate JRF vector( s) 130 (704). In some examples, the generated JRF vector(s) 130 may indicate a magnitude and / or direction of a force that is being applied by orthopedic prosthesis 104A and / or orthopedic prosthesis 104B. JRF vector(s) 130 may change over time, such as when a patient moves a limb (e.g., arm) and / or performs various ADLs. For example, for a humeral prosthesis and glenoid prosthesis, a JRF vector may be different when a patient has an arm extended over their head than when a patient’s arm is resting at their side and a JRF vector may be different when a patient is carrying a heavy load versus when a patient is not carrying anything.

[0105] One or more processors (e.g., processors 108, processors 210, etc.) may generate a prediction regarding one or more of orthopedic prostheses 104 based, at least in part, on JRF vector(s) 130 (706). The same or different processors may generate JRF vectors 130 and the prediction. In some examples, the one or more processors may generate a prediction that there will be excessive wear on articulating surfaces of first orthopedic prosthesis 104A and / or second orthopedic prosthesis 104B, bone remodeling, and / or loosening of one or more of orthopedic prostheses 104 based on JRF vector(s) 130 having a magnitude that satisfy an excessive force threshold. In other words, if the magnitude of the JRF vectors 130 exceeds the excessive force threshold, excessive wear, bone remodeling, or loosening may occur. Bone remodeling or prosthesis loosening may lead to a revision surgery' to replace one or more of orthopedic prostheses 104.

[0106] In some examples, generating the prediction regarding the orthopedic prosthesis is based directly on the generated joint reaction force vectors. A prediction system may1262-250W001compare the magnitude and direction of the joint reaction force vectors against empirical thresholds or lookup tables to predict events such as bone remodeling, prosthesis loosening, or excessive wear without executing an intermediate finite element analysis model. A training system may train a machine learning model using historical patient data to directly map sequences of joint reaction force vectors to specific predicted outcomes. Applying direct mapping or threshold comparisons may reduce computational requirements while providing predictions regarding the orthopedic prosthesis based on the joint reaction force vectors.

[0107] In some examples, the one or more processors may generate a prediction of an end of a lifespan of one or more of orthopedic prostheses 104, or a prediction of an amount of loosening of one or more of orthopedic prostheses 104 based on one or more JRF vectors 130 having a magnitude and / or direction that satisfy a particular corresponding threshold. In some examples, the one or more processors 108 may generate a prediction regarding one or more of orthopedic prostheses 104 using a physics-based model, such as an FEA model 124, that includes elements representing soft tissue structures and values of attributes of the soft tissue structures are determined based on JRF vectors 130.

[0108] FIG. 8 is a conceptual diagram illustrating an example architecture of a neural network 800 for generating JRF vector(s) 130 and associated processing steps, in accordance with one or more techniques of this disclosure. Neural network 800 may be an example of ML model 214, which is configured to generate JRF vector(s) 130.

[0109] In the example of FIG. 8, one or more processors (e.g., processors 108 (FIG. 1), processors 210 (FIG. 2), etc.) receive feature data 802 based on signals generated by sensors 118 included in one or more orthopedic prostheses 104. Signals generated by sensors 118 may include data representing Hall signals. As discussed above, the Hall signals may represent a magnetic field intensity- observed in the x, y, and z axes. Processors 210 may then process feature data 802 prior to providing input to neural network 800. In the example of FIG. 8, processors 210 may cut off bits (e.g., less significant bits) of feature data 802 (804). This reduces the number of bits that represent feature data 802. Additionally, processors 210 may decompose the feature data into magnitude data and direction data (806). In other words, processors 210 may separate the feature data by which of the generated portions of the feature data (which correspond to different directions) as well as the magnitudes (e.g., indications of acceleration and duration) of the magnetic field intensity in the different directions.1262-250W001

[0110] The magnitude data and the direction data may be used as input to a series of layers of neural network 800. In the example of FIG. 8. neural network 800 includes a series of alternating batch normalization layers and dense leaky ReLU layers and ends with a dense linear layer. Output of the dense linear layer of neural network 800 includes scaled translation data 808 and scaled modified Rodrigues parameters 810. The scaled translation data 808 indicates a translation of the prosthesis in a 3-dimensional space. The scaled modified Rodrigues parameters 810 are a minimal parameterization for representing rotation in the 3-dimensional space. Processors 210 may un-scale the scaled translation data (812). Additionally, processors 210 may un-scale the scaled modified Rodrigues parameters and convert the un-scaled modified Rodrigues parameters to an anatomical representation of the rotation (814). Un-scaling data, such as the translation data and modified Rodrigues parameters may involve multiplying the data by a scaling factor to convert the data to a different scale. The anatomical representation may indicate JRF vector(s) relative to anatomical planes of the patient’s body. Processors 210 may use the un-scaled translation data and the anatomical representation of the rotation to define the JRF vector(s) 816, which may be used to determine a pose of the prosthesis and the motion of tire prosthesis.

[0111] FIG. 9 is a conceptual diagram illustrating an example system 900 for training neural network 800, in accordance with one or more techniques of this disclosure. In the example of FIG. 9, training system 132 may train a kinematics estimation model 902 to generate JRF vector(s) 904, such as JRF vector(s) 816 (FIG. 8). ML model 126, ML model 214 and neural network 800 may be examples of kinematics estimation model 902.

[0112] As part of training kinematics estimation model 902, training system 132 may obtain sets of training examples. Each training example includes ground-truth kinematics data 906, The ground-truth kinematics data may indicate a position of a prosthesis of a prosthesis system, a velocity of the prosthesis, and an acceleration of tire prosthesis. Training system 132 may generate the ground-truth kinematics data based on data from sensors 118 of orthopedic prostheses 104. For each training example, training system 132 may use a finite element model of a prosthesis system (908) to generate a set of estimated feature data 910 associated with the training example based on the ground-truth kinematics data of the training example.

[0113] Training system 132 may then use the sets of estimated feature data 910 associated with the training examples to train kinematics estimation model 902 (912). Training system 132 may perform a plurality of training iterations to train kinematics estimation1262-250W001model 902. During each training iteration, training system 132 may preprocess a set of feature data 802 associated with a training example (e.g., as described with respect to FIG. 8), apply kinematics estimation model 902 to generate output (e.g., scaled translation data and scaled modified Rodrigues parameters), and then convert the output to JRF vector(s) (e.g., as described above with respect to FIG. 8) associated with the training example. Training system 132 may apply a loss function to the resulting patient kinematics data associated with the training example and the ground-truth kinematics data associated with the training example to generate a loss value. Training system 132 may perform a backpropagation process that updates parameters (e.g., weights) of neurons of kinematics estimation model 902 based on a gradient of the loss function. In this way, kinematics estimation model 902 may be trained to generate accurate patient JRF vector(s) 904.

[0114] Different kinematics estimation models may be trained for different prosthesis systems. Each different prosthesis system may have a different combination of prosthesis parameters (e.g., size, shape, eccentricity, lateralization / medialization, etc. Training of kinematics estimation model 902 may be performed by a computing system separate from a prosthesis in which kinematics estimation model 902 is used to generate patient kinematics data for a patient. For instance, computing system 102 (FIG. 1) or another computing system may perform the training process.

[0115] In some examples, kinematics estimation model 902 may provide outputs other than, or in addition to, JRFs. For example, kinematics estimation model 902 may generate predictions regarding implant wear, dislocation risk, and so on.

[0116] FIG. 10 is a conceptual diagram illustrating an example process for generating a musculoskeletal model, in accordance with one or more techniques of this disclosure. In the example of FIG. 10, CT slices 1000 are obtained for an arm and shoulder of a patient. Computing system 102 may use CT slices 1000 to generate a 3D model 1002 of the bones of the patient. Additionally, computing system 102 may identify landmarks in 3D model 1002. In the example of FIG. 10, the landmarks are labeled ulna styloid (US), radius styloid (RS), epicondylus lateralis (EL), epicondylus medialis (EM), glenohumeral (GH), acromion angulus (AA), acromion (AC), trigonum spinae (TS), angulus inferior (Al), scapula (SC), 7th cervical vertebra (C7), sternoclavicular (SC), incisura jugularis (IJ), 8th thoracic vertebra (T8), and processus xiphoideus (PX). Furthermore, computing system 102 may soft tissue elements representing soft tissue structures to 3D model 1002. The soft tissue elements may represent muscles, tendons, ligaments, cartilage, and so on.1262-250W001Computing system 102 may associate the soft tissue elements with attributes, such as estimated strength, estimated elasticity, and so on.

[0117] The following is a non-limiting list of clauses in accordance with one or more techniques of this disclosure.

[0118] Clause 1. A computer-implemented method comprising: obtaining, by one or more processors, feature data that is based at least in part on signals generated by one or more sensors included in an orthopedic prosthesis implanted in a joint of a patient; applying, by the one or more processors, a machine learning (ML) model to the feature data to generate one or more joint reaction force (JRF) vectors; and generating, by the one or more processors, a prediction regarding tire orthopedic prosthesis based at least in part on the one or more JRF vectors.

[0119] Clause 2. The computer-implemented method of clause 1, wherein: the orthopedic prosthesis is a first orthopedic prosthesis implanted in the joint, a second orthopedic prosthesis is also implanted in the joint such that the first orthopedic prosthesis articulates with respect to the second orthopedic prosthesis, and generating the prediction comprises using, by the one or more processors, a finite element analysis (FEA) model of the joint to perform a finite element analysis of the joint, wherein the FEA model of the joint is based at least in part on a model of the first orthopedic prosthesis, a model of the second orthopedic prosthesis, and the one or more JRF vectors.

[0120] Clause 3. The computer-implemented method of clause 2, further comprising: adjusting, by the one or more processors, values of one or more attributes of one or more soft tissue structural elements of the FEA model based on the JRF vectors.

[0121] Clause 4. The computer-implemented method of clause 3, wherein adjusting the values of the one or more attributes of the one or more soft tissue structure elements of the FEA model compri ses: applying, by the one or more processors, the FEA model with a plurality of combinations of values of the one or more attributes of the one or more soft tissue structural elements to determine sets of trial JRF vectors; comparing, by the one or more processors, the sets of trial JRF vectors to the one or more JRF vectors to determine a set of trial JRF vectors that most closely match the one or more JRF vectors; and determining, by the one or more processors, the values of the one or more attributes of the one or more soft tissue structure elements as the values of the one or more attributes of the one or more soft tissue structure elements used to determine the determined set of trial JRFs vectors.1262-250W001

[0122] Clause 5. The computer-implemented method of any of clauses 2-4, wherein generating the prediction comprises: determining whether one or more of a magnitude or a direction of the one or more JRF vectors satisfy a particular orthopedic prosthesis event threshold; and generating the prediction in response to a determination that one or more of the magnitude or the direction of tire one or more JRF vectors satisfy the particular orthopedic prosthesis event threshold, wherein the prediction is that the orthopedic prosthesis implanted in the joint of the patient will have a particular orthopedic prosthesis event.

[0123] Clause 6. The computer-implemented method of any of clauses 2-5, further comprising: using, by the one or more processors, information generated in the FEA model to generate predictions regarding orthopedic prostheses of a plurality of combinations of prostheses; and using, by the one or more processors, the predictions to generate a recommendation of a set of prostheses that includes a lowest chance of an event,

[0124] Clause 7. The computer-implemented method of any of clauses 2-7, wherein the ML model is a first ML model, the method further comprising: using, by the one or more processors, the FEA model to generate training data; and using, by the one or more processors, the training data to train a second ML model to generate predictions regarding the orthopedic prosthesis.

[0125] Clause 8. The computer-implemented method of any of clauses 1-7, wherein the joint is a shoulder joint and at least one of tire one or more JRF' vectors is a force vector applied to a glenoid prosthesis by a humeral prosthesis.

[0126] Clause 9. The computer-implemented method of any of clauses 1-8, further comprising: generating, by the one or more processors, a recommendation of exercises the patient should perform to strengthen a muscle based on the generated prediction regarding the orthopedic prosthesis.

[0127] Clause 10. The computer-implemented method of any of clauses 1-9, further comprising: generating, by the one or more processors, a recommendation of activities the patient is to avoid based on the generated prediction regarding the orthopedic prosthesis.

[0128] Clause 11. The computer-implemented method of any of clauses 1-10, further comprising: outputting, by the one or more processors, for display on a display device, an animated diagram showing the one or more JRF vectors as the joint is positioned throughout a range of motion.1262-250W001

[0129] Clause 12. The computer-implemented method of any of clauses 1-11, wherein the prediction relates to a subluxation, dislocation, or impingement of the joint.

[0130] Clause 13. The computer-implemented method of any of clauses 1-11, wherein the prediction relates to one or more of: an end of a lifespan of an orthopedic prosthesis system that includes tire orthopedic prosthesis, a loosening of one or more orthopedic prostheses of the orthopedic prosthesis system, implant wear, or muscle force.

[0131] Clause 14. The computer-implemented method of any of clauses 1-13, wherein: the patient is a first patient; the ML model is a first ML model; the method further comprises: obtaining a plurality of training examples, each of the training examples including a set of training input data and ground-truth data, wherein the training input data includes parameters of prosthesis systems implanted in patients and anatomic characteristics of the patients, wherein the ground-truth data regarding the prosthesis systems implanted in the patients; training a second ML model based on the training examples to generate predictions; and during a preoperative planning process for implanting a candidate orthopedic prosthesis into a second patient: applying the second ML model to input data for the second patient to generate a prediction regarding a candidate orthopedic prosthesis; and outputting an indication of a recommended orthopedic prosthesis based on the prediction regarding the candidate orthopedic prosthesis.

[0132] Clause 15. The computer-implemented method of clause 14, wherein the training examples include a training example having training input data that includes parameters of the orthopedic prosthesis implanted in the first patient and anatomic characteristics of the first patient, and ground-truth data of the training example is the prediction regarding the orthopedic prosthesis.

[0133] Clause 16, The computer-implemented method of any of clauses 1-15, wherein the one or more sensors include a Hall sensor.

[0134] Clause 17. A system comprising: an orthopedic prosthesis configured to be implanted in a joint of a patient, the orthopedic prosthesis including at least one sensor configured to generate signal data; and one or more processors configured to: obtain feature data based on the signal data, apply a machine learning (ML) model to the feature data to generate one or more joint reaction force (JRF) vectors, and generate a prediction regarding the orthopedic prosthesis based at least in part on the one or more JRF vectors.

[0135] Clause 18. The system of clause 17, wherein the orthopedic prosthesis is a first orthopedic prosthesis, the system further comprises: a second orthopedic prosthesis1262-250W001configured to be implanted in the joint such that the first orthopedic prosthesis articulates with respect to the second orthopedic prosthesis, and tlie one or more processors are configured to generate the prediction by using a finite element analysis (FEA) model of the joint to perform a finite element analysis of the joint, wherein the FEA of the joint is based at least in part on a model of the first orthopedic prosthesis, a model of the second orthopedic prosthesis, and the one or more JRF vectors.

[0136] Clause 19. The system of clause 18, wherein the one or more processors are configured to: adjust values of one or more attributes of one or more soft tissue structural elements of the FEA model based on the JRF vectors.

[0137] Clause 20. The system of clause 19, wherein to adjust the values of the one or more attributes of the one or more soft tissue structure elements of the FEA model, the one or more processors: apply the FEA model with a plurality of combinations of values of the one or more attributes of the one or more soft tissue structural elements to determine sets of trial JRF vectors; compare the sets of trial JRF vectors to the one or more JRF vectors to determine a set of trial JRF vectors that most closely match the one or more JRF vectors; and determine the values of the one or more attributes of the one or more soft tissue structure elements as the values of the one or more attributes of the one or more soft tissue structure elements used to determine the determined set of trial JRFs vectors.

[0138] Clause 21. The system of any of clauses 18-20, wherein, to generate the prediction, the one or more processors are configured to: determine whether one or more of a magnitude or a direction of the one or more JRF vectors satisfy a particular orthopedic prosthesis event threshold; and in response to a determination that one or more of the magnitude or the direction of the one or more JRF vectors satisfy the particular orthopedic prosthesis event threshold, generate the prediction regarding the orthopedic prosthesis implanted in the joint of the patient having a particular orthopedic prosthesis event.

[0139] Clause 22. The system of any of clauses 18-21, wherein the one or more processors are further configured to: use information generated in the FEA of the combination to generate predictions regarding orthopedic prostheses of a plurality of combinations of prostheses; and use the predictions to generate a recommendation of a set of prostheses that includes a lowest chance of an event.

[0140] Clause 23. The system of any of clauses 17-22, wherein the joint is a shoulder joint and at least one of the one or more JRF vectors is a force vector applied to a glenoid prosthesis by a humeral prosthesis.1262-250W001

[0141] Clause 24. The system of any of clauses 17-23, wherein the one or more processors are further configured to: output for display on a display device, an animated diagram showing the one or more JRF vectors as the j oint is positioned throughout a range of motion.

[0142] Clause 25. The system of any of clauses 17-24, wherein the prediction relates to a subluxation, dislocation or impingement of the joint.

[0143] Clause 26. The system of any of clauses 17-24, wherein the prediction related to one or more of: an end of a lifespan of an orthopedic prosthesis system that includes the orthopedic prosthesis, or a loosening of one or more orthopedic prostheses of the orthopedic prosthesis system.

[0144] Clause 27, The system of any of clauses 17-26, wherein the one or more processors are further configured to: generate a recommendation of exercises a patient should perform to strengthen a muscle based on the generated prediction regarding the orthopedic prosthesis.

[0145] Clause 28. The system of any of clauses 17-27, wherein the one or more processors are further configured to: generate a recommendation of activities the patient is to avoid based on the generated prediction regarding the orthopedic prosthesis.

[0146] Clause 29, The system of any of clauses 17-28, wherein the patient is a first patient, the ML model is a first ML model, and the one or more processors are further configured to: obtain a plurality of training examples, each of the training examples including a set of training input data and ground-troth data, wherein the training input data includes parameters of prosthesis systems implanted in patients and anatomic characteristics of the patients, wherein the ground-truth data regarding the prosthesis systems implanted in the patients; train a second ML model based on the training examples to generate predictions; during a preoperative planning process for implanting a candidate orthopedic prosthesis into a second patient: apply the second ML model to input data for the second patient to generate a prediction regarding a candidate orthopedic prosthesis; and output an indication of a recommended orthopedic prosthesis based on the prediction regarding the candidate orthopedic prosthesis,

[0147] Clause 30. The system of clause 29, wherein the training examples include a training example having training input data that includes parameters of the orthopedic prosthesis implanted in the first patient and anatomic characteristics of the first patient, and the ground-truth data of the training example is the prediction regarding the orthopedic prosthesis.1262-250W001

[0148] Clause 31. The system of any of clauses 17-30, wherein the one or more sensors include a Hall sensor.

[0149] Clause 32. A computing system comprising: a storage system; and one or more processors implemented in circuitry and communicatively coupled to the storage system, the one or more processors configured to perform the methods of any of clauses 1-16.

[0150] Clause 33. A computing system comprising means for performing the methods of any of clauses 1-16.

[0151] Clause 34. One or more computer-readable storage media having instructions stored thereon that, when executed by one or more processors, cause the one or more processors to perform the methods of any of clauses 1-16.

[0152] While the techniques been disclosed with respect to a limited number of examples, those skilled in the art, having the benefit of this disclosure, will appreciate numerous modifications and variations there from. For instance, it is contemplated that any reasonable combination of the described examples may be performed. It is intended that the appended claims cover such modifications and variations as fall within the true spirit and scope of the disclosure.

[0153] It is to be recognized that depending on the example, certain acts or events of any of the techniques described herein can be performed in a different sequence, may be added, merged, or left out altogether (e.g., not all described acts or events are necessary for the practice of the techniques). Moreover, in certain examples, acts or events may be performed concurrently, e.g., through multi-threaded processing, interrupt processing, or multiple processors, rather than sequentially.

[0154] In one or more examples, the functions described may be implemented in hardware, software, firmware, or any combination thereof. If implemented in software, the functions may be stored on or transmitted over as one or more instructions or code on a computer-readable medium and executed by a hardware-based processing unit. Computer-readable media may include computer-readable storage media, which corresponds to a tangible medium such as data storage media, or communication media including any medium that facilitates transfer of a computer program from one place to another, e.g., according to a communication protocol. In this manner, computer-readable media generally may correspond to (1) tangible computer-readable storage media which is non-transitory or (2) a communication medium such as a signal or carrier wave. Data storage media may be any available media that can be accessed by one or more computers1262-250W001or one or more processors to retrieve instructions, code and / or data structures for implementation of the techniques described in this disclosure. A computer program product may include a computer-readable medium.

[0155] By way of example, and not limitation, such computer-readable storage media can comprise random access memory (RAM), read only memory (ROM), electrically erasable programmable read only memory (EEPROM), compact disc read only memory (CD-ROM) or other optical disk storage, magnetic disk storage, or other magnetic storage devices, flash memory, or any other medium that can be used to store desired program code in the form of instructions or data structures and that can be accessed by a computer. Also, any connection is properly termed a computer-readable medium. For example, if instructions are transmitted from a website, server, or other remote source using a coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of medium. It should be understood, however, that computer-readable storage media and data storage media do not include connections, carrier waves, signals, or other transitory media, but are instead directed to non-transitory, tangible storage media. Disk and disc, as used herein, includes compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk and Blu-ray disc, where disks usually reproduce data magnetically, while discs reproduce data optically with lasers. Combinations of the above should also be included within the scope of computer-readable media.

[0156] Operations described in this disclosure may be performed by one or more processors, which may be implemented as fixed-function processing circuits, programmable circuits, or combinations thereof, such as one or more digital signal processors (DSPs), general purpose microprocessors, application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), or other equivalent integrated or discrete logic circuitry. Fixed-function circuits refer to circuits that provide particular functionality and are preset on the operations that can be performed. Programmable circuits refer to circuits that can programmed to perform various tasks and provide flexible functionality in the operations that can be performed. For instance, programmable circuits may execute instructions specified by software or firmware that cause the programmable circuits to operate in the manner defined by instructions of the software or firmware. Fixed-function circuits may execute software instructions (e.g., to receive1262-250W001parameters or output parameters), but the types of operations that the fixed-function circuits perform are generally immutable. Accordingly, the terms “processor” and “processing circuitry,” as used herein may refer to any of the foregoing structures or any other structure suitable for implementation of the techniques described herein.

Claims

1262-250W001What is claimed is:

1. A computer-implemented method comprising:obtaining, by one or more processors, feature data that is based at least in part on signals generated by one or more sensors included in an orthopedic prosthesis implanted in a joint of a patient;applying, by the one or more processors, a machine learning (ML) model to the feature data to generate one or more joint reaction force (JRF) vectors; and generating, by the one or more processors, a prediction regarding the orthopedic prosthesis based at least in part on the one or more JRF vectors.

2. The computer-implemented method of claim 1, wherein:the orthopedic prosthesis is a first orthopedic prosthesis implanted in the joint, a second orthopedic prosthesis is also implanted in the joint such that the first orthopedic prosthesis articulates with respect to the second orthopedic prosthesis, and generating the prediction comprises using, by the one or more processors, a finite element analysis (FEA) model of the joint to perform a finite element analysis of the joint, wherein the FEA model of the joint is based at least in part on a model of the first orthopedic prosthesis, a model of the second orthopedic prosthesis, and the one or more JRF vectors.

3. The computer-implemented method of claim 2, further comprising:adjusting, by the one or more processors, values of one or more attributes of one or more soft tissue structural elements of the FEA model based on the JRF vectors.

4. The computer-implemented method of claim 3, wherein adjusting the values of the one or more attributes of the one or more soft tissue structure elements of the FEA model comprises:applying, by the one or more processors, the FEA model with a plurality of combinations of values of the one or more attributes of the one or more soft tissue structural elements to determine sets of trial JRF vectors;comparing, by the one or more processors, the sets of trial JRF vectors to the one or more JRF vectors to determine a set of trial J RF vectors that most closely match the one or more JRF vectors; and1262-250W001determining, by the one or more processors, the values of the one or more attributes of the one or more soft tissue structure elements as the values of the one or more attributes of the one or more soft tissue structure elements used to determine the determined set of trial JRFs vectors.

5. The computer-implemented method of claim 2, wherein generating the prediction comprises:determining whether one or more of a magnitude or a direction of the one or more JRF vectors satisfy a particular orthopedic prosthesis event threshold; and generating the prediction in response to a determination that one or more of the magnitude or the direction of the one or more JRF vectors satisfy the particular orthopedic prosthesis event threshold, wherein the prediction is that the orthopedic prosthesis implanted in the joint of the patient will have a particular orthopedic prosthesis event.

6. The computer-implemented method of claim 2, further comprising:using, by the one or more processors, information generated in the FEA model to generate predictions regarding orthopedic prostheses of a plurality of combinations of prostheses; andusing, by the one or more processors, the predictions to generate a recommendation of a set of prostheses that includes a lowest chance of an event.

7. The computer-implemented method of claim 2, wherein the ML model is a first ML model, the method further comprising:using, by the one or more processors, the FEA model to generate training data; andusing, by the one or more processors, the training data to train a second ML model to generate predictions regarding the orthopedic prosthesis.

8. The computer-implemented method of claim 1, wherein the joint is a shoulder joint and at least one of the one or more JRF vectors is a force vector applied to a glenoid prosthesis by a humeral prosthesis.1262-250W0019. The computer-implemented method of claim 1, further comprising:generating, by the one or more processors, a recommendation of exercises the patient should perform to strengthen a muscle based on the generated prediction regarding the orthopedic prosthesis.

10. The computer-implemented method of claim 1, further comprising:generating, by the one or more processors, a recommendation of activities the patient is to avoid based on the generated prediction regarding the orthopedic prosthesis.

11. The computer-implemented method of claim 1, further comprising:outputting, by the one or more processors, for display on a display device, an animated diagram showing the one or more JRF vectors as the joint is positioned throughout a range of motion.

12. The computer-implemented method of claim 1, wherein the prediction relates to a subluxation, dislocation, or impingement of the joint.

13. The computer-implemented method of claim 1, wherein the prediction relates to one or more of:an end of a lifespan of an orthopedic prosthesis system that includes the orthopedic prosthesis,a loosening of one or more orthopedic prostheses of the orthopedic prosthesis system,implant wear, ormuscle force.

14. The computer-implemented method of claim 1, wherein:the patient is a first patient;the ML model is a first ML model;the method further comprises:obtaining a plurality of training examples, each of the training examples including a set of training input data and ground-truth data, wherein the training input data includes parameters of prosthesis systems implanted in patients and1262-250W001anatomic characteristics of the patients, wherein the ground-truth data regarding the prosthesis systems implanted in the patients;training a second ML model based on the training examples to generate predictions; andduring a preoperative planning process for implanting a candidate orthopedic prosthesis into a second patient:applying the second ML model to input data for the second patient to generate a prediction regarding a candidate orthopedic prosthesis; andoutputting an indication of a recommended orthopedic prosthesis based on the prediction regarding the candidate orthopedic prosthesis.

15. The computer-implemented method of claim 14, wherein the training examples include a training example having training input data that includes parameters of the orthopedic prosthesis implanted in the first patient and anatomic characteristics of the first patient, and ground-truth data of the training example is the prediction regarding the orthopedic prosthesis.

16. The computer-implemented method of claim 1, wherein the one or more sensors include a Hall sensor.

17. A system comprising:an orthopedic prosthesis configured to be implanted in a joint of a patient, the orthopedic prosthesis including at least one sensor configured to generate signal data; andone or more processors configured to:obtain feature data based on the signal data,apply a machine learning (ML) model to the feature data to generate one or more joint reaction force (JRF) vectors, andgenerate a prediction regarding the orthopedic prosthesis based at least in part on the one or more JRF vectors.

18. The system of claim 17, wherein the orthopedic prosthesis is a first orthopedic prosthesis, the system further comprises:1262-250W001a second orthopedic prosthesis configured to be implanted in the joint such that the first orthopedic prosthesis articulates with respect to the second orthopedic prosthesis, andthe one or more processors are configured to generate the prediction by using a finite element analysis (FEA) model of the joint to perform a finite element analysis of the joint, wherein the FEA of the joint is based at least in part on a model of the first orthopedic prosthesis, a model of the second orthopedic prosthesis, and the one or more JRF vectors.

19. The system of claim 18, wherein the one or more processors are configured to:adjust values of one or more attributes of one or more soft tissue structural elements of the FEA model based on the JRF vectors.

20. The system of claim 19, wherein to adjust the values of the one or more attributes of the one or more soft tissue structure elements of the FEA model, the one or more processors:apply the FEA model with a plurality of combinations of values of the one or more attributes of the one or more soft tissue structural elements to determine sets of trial JRF vectors;compare the sets of trial JRF vectors to the one or more JRF vectors to determine a set of trial JRF vectors that most closely match the one or more JRF vectors; anddetermine the values of the one or more attributes of the one or more soft tissue structure elements as the values of the one or more attributes of the one or more soft tissue structure elements used to determine the determined set of trial JRFs vectors.

21. The system of claim 18, wherein, to generate the prediction, the one or more processors are configured to:determine whether one or more of a magnitude or a direction of the one or more JRF vectors satisfy a particular orthopedic prosthesis event threshold; andin response to a determination that one or more of the magnitude or the direction of the one or more JRF vectors satisfy the particular orthopedic prosthesis event threshold, generate the prediction regarding tire orthopedic prosthesis implanted in the joint of the patient having a particular orthopedic prosthesis event.1262-250W00122. The system of claim 18, wherein the one or more processors are further configured to:use information generated in the FEA to generate predictions regarding orthopedic prostheses of a plurality of combinations of prostheses; anduse the predictions to generate a recommendation of a set of prostheses that includes a lowest chance of an event.

23. The system of claim 17, wherein the joint is a shoulder joint and at least one of the one or more JRF vectors is a force vector applied to a glenoid prosthesis by a humeral prosthesis.

24. The system of claim 17, wherein the one or more processors are further configured to:output for display on a display device, an animated diagram showing the one or more JRF vectors as the joint is positioned throughout a range of motion.

25. The system of claim 17, wherein the prediction relates to a subluxation, dislocation or impingement of the joint.

26. The system of claim 17, wherein the prediction related to one or more of:an end of a lifespan of an orthopedic prosthesis system that includes the orthopedic prosthesis, ora loosening of one or more orthopedic prostheses of the orthopedic prosthesis system,27. The system of claim 17, wherein the one or more processors are further configured to:generate a recommendation of exercises a patient should perform to strengthen a muscle based on the generated prediction regarding the orthopedic prosthesis.

28. The system of claim 17, wherein the one or more processors are further configured to:1262-250W001generate a recommendation of activities the patient is to avoid based on the generated prediction regarding the orthopedic prosthesis.

29. The system of claim 17, wherein the patient is a first patient, the ML model is a first ML model, and the one or more processors are further configured to:obtain a plurality of training examples, each of the training examples including a set of training input data and ground-truth data, wherein the training input data includes parameters of prosthesis systems implanted in patients and anatomic characteristics of the patients, wherein the ground-truth data regarding the prosthesis systems implanted in the patients;train a second ML model based on the training examples to generate predictions; during a preoperative planning process for implanting a candidate orthopedic prosthesis into a second patient:apply the second ML model to input data for the second patient to generate a prediction regarding a candidate orthopedic prosthesis; and output an indication of a recommended orthopedic prosthesis based on the prediction regarding the candidate orthopedic prosthesis.

30. Tire system of claim 29, wherein the training examples include a training example having training input data that includes parameters of the orthopedic prosthesis implanted in the first patient and anatomic characteristics of the first patient, and the ground-truth data of the training example is the prediction regarding the orthopedic prosthesis.

31. The system of claim 17, wherein the one or more sensors include a Hall sensor.

32. Tlie system of claim 17, wherein the orthopedic prosthesis is a trial prosthesis.

33. The system of claim 17, further comprising a wearable virtual reality headset; andthe one or more processors are configured to instruct the wearable virtual reality headset to display a simulated movement for a patient to perform an activity of daily living (ADL), wherein obtaining the feature data occurs while the patient performs the ADL.1262-250W00134. A computing system comprising:a storage system; andone or more processors implemented in circuitry and communicatively coupled to the storage system, the one or more processors configured to perform the methods of any of claims 1-16.

35. A computing system comprising means for performing the methods of any of claims 1-16.

36. One or more computer-readable storage media having instructions stored thereon that, when executed by one or more processors, cause the one or more processors to perform the methods of any of claims 1-16.

Images