Method and system for modeling predictive outcomes of arthroplasty procedures

Abstract

The device includes a processor and non-transitory memory. The processor is configured to receive pre-operative patient-specific data. The pre-operative patient-specific data is input to a first machine learning model to determine a first post-operative joint performance data output including a first predicted post-operative outcome metric. A reconstruction plan for the patient's joint is generated based on medical images of the joint and at least one arthroplasty surgery parameter obtained from a user. The at least one arthroplasty surgery parameter is input to a second machine learning model to determine a second predicted post-operative joint performance data output including a second predicted post-operative outcome metric. The second predicted post-operative joint performance data output is updated to include an arthroplasty surgery recommendation in response to a user changing the at least one arthroplasty surgery parameter before the arthroplasty surgery, during the arthroplasty surgery, or both.

Description

Technical Field

[0001] [Cross - Reference to Related Applications] This application is an international (PCT) patent application related to and claiming the benefit of co - owned and co - pending U.S. Provisional Patent Application No. 63 / 011,871, titled "Machine Learning Techniques for Predicting Clinical Outcomes After Shoulder Arthroplasty", having a filing date of April 17, 2020, the content of which is hereby incorporated by reference in its entirety.

[0002] The present disclosure relates to machine learning modeling for medical use, and more specifically, to methods and systems for modeling the predicted outcomes of arthroplasty surgical procedures.

Background Art

[0003] Supervised machine learning is a class of artificial intelligence in which a computer learns the complex structures and relationships in large datasets and creates a predictive model using labeled features. The machine learning model learns iteratively using the feature data and minimizes the prediction error. There are numerous commercial applications for a variety of machine learning techniques.

Summary of the Invention

Means for Solving the Problems

[0004] In some embodiments, the Disclosure provides an exemplary technically improved computer-based device, which includes at least the components of a processor and non-temporary memory storing instructions to be executed by the processor when executed, the instructions to the processor: receive preoperative patient-specific data about arthroplasty performed on a patient's joint, the preoperative patient-specific data may include the patient's medical history, measured range of motion for at least one type of joint movement of the joint, and at least one pain metric associated with the joint; input the preoperative patient-specific data into at least one first machine learning model to output a first predicted postoperative joint performance data output, the first predicted postoperative joint performance data output may include at least one first predicted postoperative outcome metric for the joint; display the first predicted postoperative joint performance output to the user on a display; receive at least one medical image of the joint obtained from at least one medical imaging procedure performed on the patient; and plan the reconstruction of the patient's joint, at least The system includes generating a reconstruction plan based on a medical image and at least one arthroplasty parameter obtained from the user in response to a displayed first predicted postoperative joint performance data output, wherein the reconstruction plan may include at least one arthroplasty parameter selected from at least one implant, at least one implant size, at least one arthroplasty procedure, at least one location for implanting at least one implant in the joint, or any combination thereof; inputting the at least one arthroplasty parameter into at least one second machine learning model to generate a second predicted postoperative joint performance data output including at least one second predicted postoperative outcome metric for the joint; displaying the second predicted postoperative joint performance data output to the user on a display; and updating the displayed second predicted postoperative joint performance data output to include at least one arthroplasty recommendation in response to the user changing some of the at least one arthroplasty parameter before, during, or both of the arthroplasty.

[0005] In some embodiments, the Disclosure provides an exemplary technically advanced computer-based method comprising at least the following steps: receiving preoperative patient-specific data about arthroplasty performed on a patient's joint by a processor. The preoperative patient-specific data may include the patient's medical history, measured range of motion for at least one type of joint movement of the joint, and at least one pain associated with the joint. The preoperative patient-specific data may be input by the processor into at least one machine learning model to determine a first predicted postoperative joint performance data output. The first predicted postoperative joint performance data output may include at least one first predicted postoperative outcome metric for the joint. The first predicted postoperative joint performance data output may be displayed to the user on a display by the processor. At least one medical image obtained from at least one medical imaging procedure performed on the patient may be received by the processor. A reconstruction plan for the patient's joint may be generated by the processor based on at least one medical image of the joint and at least one arthroplasty parameter obtained from the user in response to the displayed first predicted postoperative joint performance data output. The reconstruction plan may include at least one arthroplasty parameter selected from at least one implant, at least one implant size, at least one arthroplasty procedure, at least one location for implanting at least one implant in the joint, or any combination thereof. The at least one arthroplasty parameter may be input by the processor into at least one second machine learning model to determine a second predicted postoperative joint performance data output including at least one second predicted postoperative outcome metric for the joint. The second predicted postoperative joint performance data output may be displayed to the user on a display by the processor. The displayed second predicted postoperative joint performance data output may be updated by the processor to include at least one arthroplasty recommendation in response to the user changing some of the at least one arthroplasty parameter before, during, or both of the arthroplasty. [Brief explanation of the drawing]

[0006] [Figure 1] Figure 1 is a block diagram of a system for modeling the predicted outcomes of arthroplasty procedures according to one or more embodiments of the present disclosure. [Figure 2] Figure 2 is a graph showing preoperative range of motion (ROM) scores and preoperative outcome scores according to one or more embodiments of the present disclosure, comparing preoperative outcomes of anatomical whole shoulder arthroplasty (aTSA) patients in a clinical outcome database who describe themselves as "much better" or "worse" after surgery. [Figure 3] Figure 3 is a graph showing preoperative range of motion (ROM) scores versus preoperative outcome scores according to one or more embodiments of the present disclosure, comparing preoperative outcomes of reverse total shoulder arthroplasty (aTSA) patients in a clinical journal database in which patients describe themselves as "much better" or "worse" after surgery. [Figure 4] Figure 4 is a graph showing the age distribution at the time of surgery for anatomical total shoulder arthroplasty (aTSA) and reverse total shoulder arthroplasty (rTSA) according to one or more embodiments of the present disclosure. [Figure 5] Figure 5 is a table showing the minimum clinical importance difference (MCID) threshold and the substantial clinical benefit (SCB) threshold for each outcome metric of the cohort-wide, aTSA, and rTSA according to one or more embodiments of this disclosure. [Figure 6] Figure 6 is a table showing a comparison of mean absolute errors (MEA) associated with predictive models for the American Shoulder and Elbow Surgery Shoulder Score (ASES) according to one or more embodiments of the present disclosure. [Figure 7] Figure 7 is a table showing a comparison of mean absolute errors (MAE) associated with the University of California, Los Angeles (UCLA) prediction model according to one or more embodiments of the present disclosure. [Figure 8] Figure 8 is a table showing a comparison of mean absolute errors (MAE) associated with constant prediction models according to one or more embodiments of the present disclosure. [Figure 9]Figure 9 is a table showing a comparison of mean absolute errors (MAE) associated with comprehensive shoulder function score prediction models according to one or more embodiments of the present disclosure. [Figure 10] Figure 10 is a table showing a comparison of mean absolute errors (MAE) associated with predictive models for visual analog scale (VAS) pain scores according to one or more embodiments of the present disclosure. [Figure 11] Figure 11 is a table showing a comparison of mean absolute errors (MAE) associated with an active abduction prediction model according to one or more embodiments of the present disclosure. [Figure 12] Figure 12 is a table showing a comparison of mean absolute errors (MAE) associated with an active forward lift prediction model according to one or more embodiments of the present disclosure. [Figure 13] Figure 13 is a table showing a comparison of mean absolute errors (MAE) associated with prediction models for active external rotation according to one or more embodiments of the present disclosure. [Figure 14] Figure 14 is a table showing a comparison of the top five most predictive features identified by the XGBoost machine learning algorithm for predicting patient-reported outcome measures (PROMs) ranked by F-scores, according to one or more embodiments of the present disclosure. [Figure 15] Figure 15 is a table showing a comparison of the top five most predictive features identified by the XGBoost machine learning algorithm for predicting pain, function, and ROM, ranked by F-score, according to one or more embodiments of the present disclosure. [Figure 16] Figure 16 is a table showing a comparison of the accuracy of XGBoost for predicting aTSA and rTSA patients who will experience clinical improvement exceeding the MCID threshold, for each of the ASES score, UCLA score, and constant score, according to one or more embodiments of this disclosure. [Figure 17]Figure 17 is a table showing a comparison of the accuracy of the XGBoost algorithm for predicting aTSA and rTSA patients who experienced clinical improvement exceeding the MCID threshold, for each of the comprehensive shoulder function scores and VAS pain scores for measurements of active abduction, anterior elevation, and external rotation ROM, according to one or more embodiments of the present disclosure. [Figure 18] Figure 18 is a table showing a comparison of the accuracy of the XGBoost algorithm for predicting aTSA and rTSA patients who experienced clinical improvement exceeding the SCB threshold, for each of the ASES score, UCLA score, and constant score, according to one or more embodiments of the present disclosure. [Figure 19] Figure 19 is a table showing a comparison of the accuracy of the XGBoost algorithm for predicting aTSA and rTSA patients who experienced clinical improvement exceeding the SCB threshold, for each of the comprehensive shoulder function score and VAS pain score, and for measurements of active external rotation, anterior elevation, and external rotation ROM, according to one or more embodiments of the present disclosure. [Figure 20] Figure 20 is a table showing a list of predictive model inputs to a machine learning model that calculates a comprehensive shoulder function score and a VAS pain score, as well as active abduction, active forward elevation, and active external rotation, according to one or more embodiments of the present disclosure. [Figure 21] Figure 21 is a table showing a list of further predictive model inputs (beyond the inputs shown in Figure 20) to a machine learning model for calculating ASES scores, according to one or more embodiments of the present disclosure. [Figure 22] Figure 22 is a table showing a list of further predictive model inputs (beyond the inputs shown in Figure 20) to a machine learning model for calculating a constant score, according to one or more embodiments of the present disclosure. [Figure 23] Figure 23 is an exemplary flowchart for modeling the predicted outcomes of arthroplasty procedures according to one or more embodiments of the present disclosure. [Figure 24]Figure 24 is a table showing a comparison of mean absolute error (MAE) related to the prediction of ASES pre using a complete XGBoost machine learning model and a simplified XGBoost machine learning model according to one or more embodiments of the present disclosure. [Figure 25] Figure 25 is a table showing a comparison of mean absolute error (MAE) related to the prediction of constants using a complete XGBoost machine learning model and a simplified XGBoost machine learning model according to one or more embodiments of the present disclosure. [Figure 26] Figure 26 is a table showing a comparison of mean absolute error (MAE) related to the prediction of comprehensive shoulder function scores using a complete XGBoost machine learning model and a simplified XGBoost machine learning model according to one or more embodiments of the present disclosure. [Figure 27] Figure 27 is a table showing a comparison of mean absolute error (MAE) related to the prediction of VAS pain scores using a complete XGBoost machine learning model and a simplified XGBoost machine learning model according to one or more embodiments of the present disclosure. [Figure 28] Figure 28 is a table showing a comparison of mean absolute error (MAE) related to the prediction of active abduction using a complete XGBoost machine learning model and a simplified XGBoost machine learning model according to one or more embodiments of the present disclosure. [Figure 29] Figure 29 is a table showing a comparison of mean absolute error (MAE) related to the prediction of active forward elevation using a complete XGBoost machine learning model and a simplified XGBoost machine learning model according to one or more embodiments of the present disclosure. [Figure 30] Figure 30 is a table showing a comparison of mean absolute error (MAE) related to the prediction of active external rotation using a complete XGBoost machine learning model and a simplified XGBoost machine learning model according to one or more embodiments of the present disclosure. [Figure 31] Figure 31 is a table showing a comparison of predictions of a complete XGBoost model for aTSA patients and rTSA patients who experienced clinical improvement exceeding the MCID threshold for measurements of multiple different outcomes according to one or more embodiments of the present disclosure. [Figure 32]Figure 32 is a table showing a comparison of predictions of a simple XGBoost model for aTSA patients and rTSA patients who experienced clinical improvement exceeding the MCID threshold for measurements of multiple different outcomes according to one or more embodiments of the present disclosure. [Figure 33] Figure 33 is a table showing a comparison of predictions of a complete XGBoost model for aTSA patients and rTSA patients who experienced clinical improvement exceeding the SCB threshold for measurements of multiple different outcomes according to one or more embodiments of the present disclosure. [Figure 34] Figure 34 is a table showing a comparison of predictions of a simple XGBoost model for aTSA patients and rTSA patients who experienced clinical improvement exceeding the SCB threshold for measurements of multiple different outcomes according to one or more embodiments of the present disclosure. [Figure 35] Figure 35 is a table showing a comparison of a simple XGBoost model with an input from predictive CT planning data for aTSA patients and rTSA patients who experienced clinical improvement exceeding the MCID threshold for measurements of multiple different outcomes according to one or more embodiments of the present disclosure. [Figure 36] Figure 36 is a table showing a comparison of a simple XGBoost model with an input from predictive CT planning data for aTSA patients and rTSA patients who experienced clinical improvement exceeding the SCB threshold for measurements of multiple different outcomes according to one or more embodiments of the present disclosure. [Figure 37] Figure 37 is a flowchart of an exemplary method for modeling predicted outcomes of arthroplasty surgical procedures. Perspective view showing an energy receiving section of a vehicle according to a first embodiment of the present invention.

Embodiments for Carrying Out the Invention

[0007] Some embodiments of the present disclosure are described herein by method only, as examples, with reference to the accompanying drawings. With particular detail reference to the drawings, it is emphasized that the embodiments shown are intended to illustrate the embodiments of the present disclosure by method only. In this regard, the description with reference to the drawings will make it clear to those skilled in the art how embodiments of the present disclosure may be carried out.

[0008] Among the benefits and improvements disclosed, other purposes and advantages of this disclosure will become apparent from the following description, as interpreted in conjunction with the accompanying figures. While detailed embodiments of this disclosure are disclosed herein, it should be understood that the disclosed embodiments are merely illustrative examples of the various forms in which this disclosure may be embodied. Furthermore, each example given with respect to the various embodiments of this disclosure intended to be illustrative is not limiting.

[0009] Throughout this specification and the claims, the following terms have the meanings expressly associated herein unless indicated to be clearly different in context. The phrases “in one embodiment,” “in one embodiment,” and “in several embodiments” as used herein do not necessarily refer to the same, but possible, embodiments. Furthermore, the phrases “in another embodiment” and “in several other embodiments” as used herein do not necessarily mean different embodiments, although they may. All embodiments of this disclosure are intended to be combinable without departing from the scope or spirit of this disclosure.

[0010] As used herein, the term "based on" is not exclusive and may be based on additional factors not listed unless explicitly indicated otherwise by the context. Furthermore, in this specification, the meanings of "a," "an," and "the" include multiple references. The meaning of "in" includes "in" and "above."

[0011] As used herein, terms such as “equipment,” “includes,” and “possess” are not intended to limit the scope of a particular claim to the materials or steps enumerated by the claims.

[0012] All prior patents, publications, and test methods referenced herein are incorporated in their entirety by reference. EXAMPLES

[0013] Any variations, modifications, and alterations to the embodiments of the present disclosure described above will be obvious to those skilled in the art. All such variations, modifications, and alterations are intended to fall within the spirit and scope of the present disclosure, which is limited only by the appended claims.

[0014] While several embodiments of this disclosure have been described, it should be understood that these embodiments are illustrative and not restrictive, and that many modifications may become apparent to those skilled in the art. For example, all dimensions described herein are provided as examples only and are intended to be illustrative and not restrictive.

[0015] Any feature or element explicitly identified in this description may be explicitly excluded as a feature or element of the embodiment of the invention defined in the claims.

[0016] Machine learning techniques for healthcare applications offer the potential to transform complex healthcare data into actionable knowledge, enabling surgeons to better understand patients and their complex conditions. By leveraging large amounts of high-quality clinical outcome data, machine learning analysis can identify previously unknown correlations and relationships within datasets, creating predictive models that better communicate treatment strategies for individual patients.

[0017] In orthopedics, predictive models derived from high-quality outcomes and patient data represent patient-specific implementations of evidence-based decision-making tools, potentially transforming complex medical data into actionable knowledge to support more informed treatment decisions. While commercial use of machine learning may be new in orthopedics, its use in research has increased in recent years. Although machine learning applications have historically been largely image-based analysis, there is growing interest in using machine learning techniques to predict clinical outcomes. Predictive outcome models can help orthopedic surgeons better identify patients who would benefit from elective surgeries such as arthroplasty, and by leveraging demographic, diagnostic, comorbid, clinical history, and the experiences of past patients with similar treatments, they can help better align patient and surgeon expectations for clinical improvement. As the factors predicting patient-specific improvements are better understood and the consistency between predicted and actual outcomes increases, using such evidence-based predictive tools could improve patient satisfaction.

[0018] Embodiments of this disclosure disclose methods and systems for modeling predictive outcomes of arthroplasty procedures. Arthroplasty is used to repair or replace any joint of the body, including but not limited to the hip, knee, shoulder, elbow, and ankle. However, to further illustrate these methods and systems, shoulder arthroplasty is used herein as an exemplary embodiment.

[0019] Figure 1 is a block diagram of a system 10 for modeling predictive outcomes of arthroplasty procedures according to one or more embodiments of the present disclosure. The system 10 may include a server 15, a medical imaging system 35, a plurality of N electronic medical resources indicated as electronic resource 1 40A...electronic resource N 40B (where N is an integer), and a computing device 77 of a user 20, all of which communicate 32 via a communication network 30. The computing device 77 of the user 20 may also be coupled 37 to the server 15 so as to be able to communicate directly.

[0020] In some embodiments, the user 20, who can interact with a graphical user interface (GUI) 75 on the computing device 77, may be a physician discussing an arthroplasty procedure to be performed on a patient 25. In other embodiments, the computing device 77 may be located in any suitable location, such as an operating room, where the arthroplasty procedure may be performed.

[0021] The server 15 may include a processor 45, non-temporary memory 60, communication circuits 70 for performing communication 32 via a communication network 30, and / or I / O devices 65 such as a display for displaying a GUI 75 to a user 20, a keyboard 65A, and a mouse 65B.

[0022] In some embodiments, the server 15 may be configured to run different software modules to perform the functions in the system 10 described herein. These different software modules may include, but are not limited to, a patient-specific data acquisition module 46, a computed tomography (CT) image-based guided personalized surgery (GPS) joint reconstruction planning module 48, an initial preoperative predictive machine learning model (MLM) module 50, an image-based predictive MLM module 52, a machine learning model training module 54, and a GUI management module 56 that controls a GUI 75 on the user's computing device 77.

[0023] In some embodiments, the non-temporary memory 60 may be configured to store a clinical outcome database 62 having multiple clinical outcomes of different types of arthroplasty procedures performed on multiple patients.

[0024] In some embodiments, the patient-specific data acquisition module 46 can query one of several electronic medical resources 40A, 40B via a communication network 30 to obtain clinical data from the patient 25. The multiple electronic medical resources 40A, 40B may be managed, for example, by the patient's health management organization (HMO), the hospital where the patient 25 received treatment, or the physician who treated the patient 25.

[0025] In some embodiments, the CT image-based GPS joint reconstruction planning module 48 may analyze data from medical images received from the medical imaging system 35. The medical imaging system 35 may generate, for example, X-ray images, computed tomography (CT) images, magnetic resonance images, and / or three-dimensional (3D) medical images. The 3D medical image may be generated from multiple X-ray images. The medical image may include frames from a video of the joint.

[0026] In some embodiments, the machine learning model (MLM) training module 54 may generate a training dataset for training a machine learning model used in the system 10. For example, the MLM training module 54 may obtain patient outcome data from a clinical outcome database 62 and generate a dataset that partially maps preoperative patient-specific data and data vectors of arthroplasty parameters used in different types of arthroplasty procedures to known postoperative outcome metrics for joint replacement. The trained machine learning model may then generate predicted postoperative outcome metrics for joint replacement given input data vectors for new patients before arthroplasty.

[0027] In some embodiments, with respect to shoulder arthroplasty, machine learning techniques may be used to preoperatively predict clinical outcomes at various postoperative time points in patients undergoing total shoulder arthroplasty. These predictions can be used, for example, to inform shoulder surgeons of what a particular patient is expected to experience after anatomical total shoulder arthroplasty (aTSA) or reverse total shoulder arthroplasty (rTSA). The list of model inputs may be automatically obtained from the patient's electronic medical records through software integration by querying healthcare professionals and / or any of the electronic medical resources mentioned above. While this disclosure focuses on aTSA and rTSA outcome prediction, these models may also be applicable to other shoulder arthroplasty applications, such as hemiarthroplasty, fracture reconstruction, internal prostheses, surface replacement, and primary arthroplasty versus rearthroplasty outcome prediction.

[0028] In some embodiments, with respect to the input to these predictive models, the predicted outcomes for a given total shoulder arthroplasty patient may be further refined to provide recommendations for optimal clinical outcomes by taking into account patient-specific diagnoses and morphological considerations of bone / soft tissue, for example, different implant sizes such as different sized humeral heads, humeral stems, glenospheres, glenoid or humeral reinforcements, implant types such as aTSA, rTSA, hemiarthroplasty, surface replacement, short stem, stemless, fracture arthroplasty, internal prostheses, and corrective devices, and / or surgical techniques such as trigonothorax, superolateral, and subscapular preservation.

[0029] In some embodiments, using machine learning predictive outcome algorithms to predict a patient's postoperative clinical outcome preoperatively may have numerous additional practical applications that are valuable to both patients and surgeons. Firstly, being able to preoperatively identify which patients may achieve clinical improvement after aTSA and rTSA, compared to minimum clinical importance difference (MCID) and effective clinical benefit (SCB) thresholds based on patient satisfaction anchors, for multiple different patient-reported outcome metrics (PROMs) and active range of motion (ROM) measurements, may be beneficial for orthopedic surgeons in objectively identifying candidate patients suitable for these alternative procedures. It may also help orthopedic surgeons determine the type of implant for a particular patient. Since non-surgical treatment may be optimal for some patients, this prediction may represent a more efficient allocation of resources for patients, surgeons, hospitals, and / or payers.

[0030] In this disclosure, the terms “outcome metric” and “outcome measurement” may be used interchangeably herein. The terms “machine learning model,” “machine learning module,” “machine learning predictive outcome algorithm,” “predictive outcome algorithm,” and / or “predictive outcome model” may be used interchangeably herein.

[0031] In some embodiments, patient-specific predictions of clinical improvement at multiple postoperative time points can help align patient and surgeon expectations about what can be achieved after this selective procedure. Given the correlation between preoperative expectations and postoperative satisfaction, improved surgeon-patient alignment in both the magnitude and speed of clinical improvement may lead to improved patient satisfaction. Furthermore, a deeper understanding of the amount of clinical improvement that can be expected at various postoperative time points for a particular patient may help surgeons establish rehabilitation protocols. This may also help both surgeons and patients weigh these benefits against the procedure-specific risks associated with aTSA and rTSA, such as instability, aseptic loosening, and infection.

[0032] In some embodiments, the machine learning techniques disclosed herein may be extended to predict outcomes and improvements based on specific diagnoses, and to predict and / or identify patients with risk factors for various complications. Furthermore, predictive models may help appropriately risk-stratify patients and generate recommendations regarding medical workflows, such as identifying patients who are likely to undergo surgery safely in an outpatient surgery center and those who require both inpatient and outpatient surgery in a hospital. Predictive models may also recommend the length of hospital stay after arthroplasty for specific patients.

[0033] In some embodiments, predictive models can provide a better understanding of factors influencing outcomes, potentially assisting orthopedic surgeons in individualizing each patient's care in relation to their specific requirements for pain relief, function, and mobility, and enabling patients to better understand how well arthroplasty procedures can meet their needs based on the patient-specific characteristics input into and described in the predictive model output.

[0034] Figure 2 is a graph showing preoperative range of motion (ROM) scores versus preoperative outcome scores, comparing preoperative outcomes of patients undergoing anatomical total shoulder arthroplasty (aTSA) who, according to one or more embodiments of the present disclosure, would be described as "much better" or "worse" postoperatively in a clinical outcome database.

[0035] Figure 3 is a graph showing preoperative range of motion (ROM) scores versus preoperative outcome scores, comparing preoperative outcomes of reverse total shoulder arthroplasty (aTSA) patients who, according to one or more embodiments of this disclosure, would be described as "much better" or "worse" postoperatively in a clinical outcome database.

[0036] In both Figures 2 and 3, it should be noted that the preoperative outcomes of aTSA and rTSA patients in the clinical outcome database 62 may be based on aTSA and rTSA patients who rated their postoperative outcomes as "much better" and those who rated them as "worse" during the most recent follow-up. In both Figures 2 and 3, it should be noted that because the distribution of patients between the two cohorts is relatively equal, it may be difficult for orthopedic surgeons to identify these patients and distinguish preoperatively whether a particular patient will have a "much better" or "worse" outcome based solely on these parameters if they undergo a given procedure.

[0037] In some embodiments, because it is typically difficult for arthroplasty surgeons to preoperatively identify which patients will achieve poor outcomes and which will be dissatisfied with the procedure, based on currently available knowledge and clinical guidelines, as well as known risk factors, evidence-based preoperative predictive outcome tools greatly assist surgeons in objectively establishing patient-specific gains to be achieved after arthroplasty.

[0038] In patients who underwent aTSA and / or rTSA, approximately 90% reported being satisfied with the procedure (e.g., patients reporting being “better” or “much better” compared to the non-surgical state) compared to dissatisfied patients (e.g., patients reporting “no change” or “worse” compared to the non-surgical state), suggesting that positive outcomes may be common in patients after total shoulder arthroplasty. However, the predictability of patients achieving these poor outcomes may be less certain for both aTSA (Figure 2) and rTSA (Figure 3), as shown by the presentation of preoperative outcomes compared to those who “get much better” compared to those who “get worse” postoperatively.

[0039] Considering the functional improvements and the amount of range of motion patients achieve in specific planes at certain postoperative follow-ups, the predictability of outcomes after total shoulder arthroplasty may not be entirely certain. For example, most shoulder surgeons believe that the improvement and amount of active rotation after rTSA are unpredictable, and therefore may not be able to accurately advise patients on whether their ability to actively rotate their arm will improve.

[0040] The recovery time it takes for a patient to regain full range of motion after total shoulder arthroplasty, and the complete outcome as measured by various patient-reported outcome indicators (PROM: e.g., ASES, Constant, UCLA, Shoulder Function, Simple Shoulder Test) (SST), Shoulder Pain and Disability Index (SPADI), VAS Pain, Shoulder Arthroplasty Smart Score, etc.), the majority of the improvement a patient may experience is usually achieved within the first six months after the arthroplasty procedure. However, some patients may take up to two years after surgery to achieve full range of motion or the maximum PROM score.

[0041] Furthermore, total range of motion and / or maximum PROM scores may vary among patients due to many different factors, including but not limited to, patient demographics, comorbidities, diagnosis, severity / degenerative state of diagnosis, bone / soft tissue quality, bone morphology, implant selection type, implant size, implant placement, and / or surgical technical information. Therefore, surgeon and patient expectations may not be accurate and may not match due to all of the above factors, potentially leading to increased dissatisfaction with the surgery. Thus, to better assist patients and surgeons in achieving more accurate expectations, improved predictability, and improved satisfaction, it is necessary to consider all possible variables to more appropriately and accurately predict outcomes defined by PROM and ROM after total shoulder arthroplasty.

[0042] Figure 4 is a graph showing the age distribution at surgery for patients undergoing anatomical total shoulder arthroplasty (aTSA) and reverse total shoulder arthroplasty (rTSA) according to one or more embodiments of the present disclosure. Figure 4 shows that older patients are more likely to undergo rTSA than aTSA, and younger patients are more likely to undergo aTSA. The crossover age at which patients are more likely to undergo rTSA is 64 years of age at surgery. For patients aged 75 years or older at surgery, the ratio is 4:1 for rTSA compared to aTSA.

[0043] Furthermore, recent merging of signs between aTSA and rTSA, and a shift in recent trends toward increased use of rTSA by shoulder surgeons in older patients to reduce the incidence of rotator cuff-related complications that primarily occur with aTSA, necessitates that surgeons be able to more accurately predict which arthroplasty will yield a better outcome, rather than rTSA, as shown in Figure 4.

[0044] Embodiments herein describe a method, workflow, and computer software system, shown in System 10 of Figure 1, that predicts joint outcomes and range of motion after arthroplasty using multivariate machine learning analysis of outcome data from a clinical outcome database 62 (which may, for example, be used to train a machine learning predictive model implemented herein). Thus, the trained machine learning predictive model can extrapolate its statistical trends and relationships to those of patient-specific data of a particular patient who will undergo arthroplasty, in order to more accurately predict the postoperative outcome metrics that that particular patient may achieve preoperatively.

[0045] In some embodiments, surgeons may use the information derived from this predictive model to identify outcomes measured by multiple different outcome metrics at various postoperative time points for various implant types and sizes, and to compare their outcome results with those of other similar patients from a clinical outcome database 62, in order to extrapolate outcomes based on the experiences of other similar patients.

[0046] In some embodiments, predictive models may be used to compare the range of outcomes achieved with different implant types (e.g., aTSA vs. rTSA in shoulder arthroplasty), different implant sizes, and different implant locations, compared to other patients in the clinical outcome database62 for various defined diagnoses, comorbidities, bone deformities, and / or soft tissue conditions within the joint under consideration. All these considerations may be used to establish and communicate more accurate expectations for actual outcomes and to improve collaboration between surgeons and patients.

[0047] In some embodiments, the predictive models may utilize data from a clinical outcome database 62 to identify complex interactions in this data, classify the data, and / or identify the most important contributing factors and relevances to postoperative outcomes. These predictive algorithms can further model and predict postoperative outcomes for similar new cases of various different PROM and range of motion measurements. Each predictive model may be analyzed independently and / or coupled into a series where the results of one predictive model become input to another new predictive model.

[0048] In some embodiments, the predictive models for total shoulder arthroplasty developed for the exemplary embodiments shown herein were trained using data from a clinical outcome database62 from more than 8,000 patients and 20,000 postoperative patient visits. For each patient underlying the analysis, there were approximately 300 preoperative data inputs. This predictive analysis may perform regression analysis, deep learning-based analysis, at least one ensemble-based decision tree learning method, or any combination thereof, to combine results from multiple different decision trees to identify and rank the preoperative parameters most significantly associated with outcomes for total shoulder arthroplasty.

[0049] In some embodiments, predictive models may assist surgeons in providing the best possible outcomes for specific patients by leveraging a large database of clinical history, for example, by identifying and ranking these parameters and the most relevant risk factors from data related to patient demographics, comorbidities, diagnosis, severity of diagnosis / degenerative state, bone / soft tissue quality, bone morphology, implant selection type, implant size, implant location, and / or surgical procedure information. The predictive models can provide surgeons with actionable recommendations in identifying and communicating the complex interactions between these parameters.

[0050] In some embodiments, the system 10, which can be accessed by the surgeon 20 on a computing device 77, may be preoperative planning software that provides recommendations regarding the types and sizes of implants the surgeon can choose from, and recommendations on where these implants should be placed.

[0051] In some embodiments, a system 10 in which a predictive model can be accessed by a surgeon 20 on a computing device 77 may provide a GUI 75 for an intraoperative computer navigation or robotic system that enables on-the-fly modifications of the preoperative plan based on intraoperative findings (e.g., during the surgical procedure) by the surgeon and / or hospital staff. Each of the aforementioned actionable guidances may be communicated intraoperatively by the predictive model (e.g., implant type, implant size, and / or implant location). Conversely, the predictive model may be accessed via a standalone software application available on multiple different software platforms accessible to the patient, surgeon, or other healthcare professionals.

[0052] In some embodiments, three supervised machine learning techniques, including linear regression-based, tree-based, and / or deep learning-based machine learning, may be used to analyze data from a clinical outcome database62 of shoulder arthroplasty patients who underwent single-platform total shoulder arthroplasty (e.g., Equinoxe, Exactech Inc., Gainesville, FL) between November 2004 and December 2018. All total shoulder arthroplasty patients consented to data sharing, and all data were collected using standardized forms in accordance with institutional review board (IRB) approved protocols.

[0053] In some embodiments, patients with cases of revision surgery, a diagnosis of humeral fracture, or hemiarthroplasty were excluded to ensure a homogeneous dataset. Patients with a follow-up period of less than three months were also excluded. Based on these criteria, preoperative, intraoperative, and postoperative data from 5,774 patients with 17,427 postoperative follow-up visits, as well as postoperative follow-up data from 17,427 patients, were made available. An algorithm was trained and generated to predict postoperative scores at 3-6 months, 6-9 months, 1 year [9-18 months], 2 years [18-36 months], 3-5 years [36-60 months], and 5 years or more [60 months or more]. Active range of motion was measured using a goniometer during each patient's clinical visit.

[0054] In some embodiments, the predictive algorithm was trained and generated using demographic data, diagnosis, comorbidities, implant type, preoperative ROM, preoperative X-ray findings, and preoperative PROM scores (such as ASES, SPADI, SST, UCLA, and Constant Metric), including individual questions used to derive each score, with a total of 291 labeled features utilized. Using clinical data from 2,153 major aTSA patients (7,305 visits; mean follow-up period = 26.7 months) and 3,621 major rTSA patients (10,122 visits; mean follow-up period = 22.8 months), predictive models were trained and generated at various postoperative time points: 3–6 months (aTSA = 1282 visits and rTSA = 222 visits), 6–9 months (aTSA = 658 visits and rTSA = 1177 visits), 1 year (aTSA = 1451 visits and rTSA = 2445 visits), 2–3 years (aTSA = 1347 visits and rTSA = 1882 visits), 3–5 years (aTSA = 1321 visits and rTSA = 1482 visits), and 5 years or more (aTSA = 1246 visits and rTSA = 907 visits). 66.7% of this data was randomly selected to define the training cohort, and the remaining 33.3% was defined as the validation test cohort and used to evaluate the prediction error of each algorithm.

[0055] In some embodiments, the predictive model may include three trained supervised machine learning techniques: 1) linear regression, 2) XGBoost, and 3) width and depth.

[0056] As a general technical background for these predictive models, linear regression models model a linear relationship between preoperative data (input variables) and outcome data (output variables) from a complete training dataset. XGBoost models are ensemble techniques of multiple regression trees. These regression trees can be built by repeatedly splitting the entire training dataset into multiple smaller batches using a method called boosting. XGBoost can handle missing values ​​and data sparseness relatively well. Breadth and depth models are a hybrid of linear regression and deep learning models and are particularly useful for classification problems with sparse inputs. Since the features of the clinical outcome database62 can be categorized, width and depth models may be suitable for this technique.

[0057] In some embodiments, the deep learning component can utilize a hierarchical function that calculates model coefficients based on inputs from previous layers and ultimately propagates those coefficients to the top layer of the outcome prediction model. Width (or linear component) can be used for high-density / numerical functions, while depth (or feedforward neural network component) can be used for sparse / categorical functions. Baseline mean analysis can be used as a consideration control to evaluate the relative accuracy of each prediction model.

[0058] Figure 5 is a table showing the minimum clinical importance difference (MCID) and substantial clinical benefit (SCB) thresholds for each outcome metric (scale) of the cohort-wide, aTSA, and rTSA according to one or more embodiments of this disclosure. The primary targets of each model can be used to predict postoperative outcome measures at each postoperative time point. Secondary goals can be identified if patients experience a clinical improvement greater than the MCID and SCB patient satisfaction anchor-based thresholds previously established by Simovitch et al. See Figure 5. MCID can represent a lower threshold for improvement and can define the minimum improvement that patients perceive as a significant change from a given treatment. SCB may differ from MCID in that it may represent a target level of improvement to achieve a substantial benefit perceived by patients.

[0059] In some embodiments, the predictive performance of each model for its primary target can be quantified by the mean absolute error (MAE) between the actual and predicted values ​​of each outcome metric for aTSA and rTSA patients in a 33.3% validation trial cohort. To facilitate model interpretation, the F-score of the XGBoost model can be used to identify the most predictive features. The F-score can quantify how often a particular feature is likely to be used as a candidate for splitting in the decision tree algorithm. The accuracy of each model for determining the performance of secondary targets, or whether patients will achieve the threshold for MCID and SCB improvement for each outcome metric in a 2-3 year follow-up, can be quantified using an accuracy classification metric to quantify the model's ability to not identify negatives as positives, recall to quantify the model's ability to identify positives as positives, an F1 score to quantify the harmonic mean between the accuracy score and the recall score, precision to quantify the ratio of correcting predictions to the total number of predictions, and / or the area under the receiver operating curve (AUROC), all of which may determine the overall accuracy of the model. The results of these predictive models are shown in the table below.

[0060] Figure 6 is a table showing a comparison of the mean absolute error (MAE) associated with predictive models for the U.S. Shoulder and Elbow Surgery Score (ASES) according to one or more embodiments of the present disclosure.

[0061] Figure 7 is a table showing a comparison of the mean absolute error (MAE) associated with the University of California, Los Angeles (UCLA) prediction model in one or more embodiments of the present disclosure.

[0062] Figure 8 is a table showing a comparison of the mean absolute error (MAE) associated with the constant prediction model in one or more embodiments of the present disclosure.

[0063] Figure 9 is a table showing a comparison of the mean absolute error (MAE) associated with a comprehensive shoulder function score prediction model according to one or more embodiments of the present disclosure.

[0064] Figure 10 is a table showing a comparison of the mean absolute error (MAE) associated with a predictive model of a visual analog scale (VAS) pain score according to one or more embodiments of the present disclosure.

[0065] Figure 11 is a table showing a comparison of the mean absolute error (MAE) associated with the prediction model of active abduction according to one or more embodiments of the present disclosure.

[0066] Figure 12 is a table showing a comparison of the mean absolute error (MAE) associated with an active forward lift prediction model according to one or more embodiments of the present disclosure.

[0067] Figure 13 is a table showing a comparison of the mean absolute error (MAE) associated with the prediction models for active external rotation according to one or more embodiments of the present disclosure.

[0068] The primary target predictions for ASES (Figure 6), UCLA (Figure 7), Constant (Figure 8) PROM, Comprehensive Shoulder Function Score (Figure 9), VAS Pain Score (Figure 10), Active Abduction (Figure 11), Anterior Elevation (Figure 12), and External Rotation (Figure 13) at 1 year, 2-3 years, 3-5 years, and 5 years or more after aTSA and rTSA are shown in the tables in Figures 6-13. The width and depth models had the lowest MAE for all indicators at each time point, followed by the XGBoost model and the linear regression model. Despite differences in accuracy, all three predictive outcome algorithms had lower MAE than the mean model at baseline.

[0069] Based on weighted average MAE, each machine learning technique most accurately predicted the constant score (±7.56% MAE), followed by the UCLA score (±8.16% MAE), and finally the ASES score (±10.45% MAE). Across all postoperative time points analyzed, the mean MAE of the width and depth prediction models was ±1.2 for the comprehensive shoulder function score, ±1.9 for the VAS pain score, ±19.5° for active abduction, ±15.9° for anterior elevation, and ±11.4° for external rotation. The patient-to-patient differences between aTSA and rTSA were similar across each score, each motion plane analyzed, and postoperative time points, with only slight differences. Furthermore, this data and techniques can be used to generate other prediction models such as the internal rotation score, visual analog scale pain, and / or the shoulder arthroplasty smart score.

[0070] Figure 14 is a table showing a comparison of the top five most predictive features identified by the XGBoost machine learning algorithm to predict F-score ranked patient-reported outcome measures (PROM) in one or more embodiments of the present disclosure. Figure 15 is a table showing a comparison of the top five most predictive features identified by the XGBoost machine learning algorithm to predict F-score ranked pain, function, and ROM in one or more embodiments of the present disclosure.

[0071] In some embodiments, the top five most predictive features used by the XGBoost predictive model for each PROM (Figure 14) and pain, function, and ROM indices (Figure 15) are shown in the tables in Figures 14-15. In the embodiments of this disclosure, for the 291 features used, the XGBoost predictive model showed some differences between the PROM model and the pain, function, and ROM models, but demonstrated excellent agreement for the top five features in F-score rank. "Follow-up period," representing the postoperative recovery period, was identified as the most predictive feature used in all models.

[0072] In some embodiments, with respect to PROM, two different preoperative PROM (SPADI and ASES) and four different preoperative measures of active ROM were observed to have high predictive power, along with the classification question "Will the surgery be performed on the dominant hand?". With respect to pain, function, and ROM measures, the classification question "Will the surgery be performed on the dominant hand?" was identified as the second most predictive feature in all models. The classification question "Is the patient female?" was identified as the third most predictive feature in all but one model. Other highly predictive features included the preoperative SPADI score, two different preoperative measures of active ROM, and the classification question "Has the patient had previous shoulder surgery?".

[0073] Figure 16 is a table showing a comparison of the accuracy of the XGBoost algorithm for predicting aTSA and rTSA patients who experienced clinical improvement exceeding the MCID threshold for ASES, UCLA, and constant scores, respectively, according to one or more embodiments of the present disclosure.

[0074] Figure 17 is a table showing a comparison of the accuracy of the XGBoost algorithm in predicting aTSA and rTSA patients who experienced clinical improvement above the MCID threshold for each of the comprehensive shoulder function scores and VAS pain scores for active abduction, active anterior elevation, and active external rotation ROM measurements, according to one or more embodiments of the present disclosure.

[0075] Figure 18 is a table showing a comparison of the accuracy of the XGBoost algorithm in predicting aTSA and rTSA patients who experienced clinical improvement above the SCB threshold for ASES, UCLA, and Constant Score, respectively, according to one or more embodiments of the present disclosure.

[0076] Figure 19 is a table showing a comparison of the accuracy of the XGBoost algorithm in predicting aTSA and rTSA patients who experienced clinical improvement above the SCB threshold for each of the comprehensive shoulder function score and VAS pain score, as well as for the indices of active abduction, anterior elevation, and external rotation ROM, according to one or more embodiments of the present disclosure.

[0077] In some embodiments, the MCID predictions for the PROM model (Figure 16) and the secondary targets of the pain, function, and ROM model (Figure 17) at follow-up 2-3 years are shown in the tables in Figures 16-17. The XGBoost PROM model showed 93-95% accuracy for MCID and 0.87-0.94 for AUROC in aTSA patients, and 93-99% accuracy for MCID and 0.85-0.97 for AUROC in rTSA patients. In other embodiments, the XGBoost pain / function / ROM model showed 85-94% accuracy for MCID and 0.79-0.91 for AUROC in aTSA patients, and 90-94% accuracy for MCID and 0.78-0.90 for AUROC in rTSA patients.

[0078] In some embodiments, the SCB predictions for the PROM model (Figure 18) and the ROM model (Figure 19) at follow-up of 2-3 years are shown in the tables in Figures 18-19. The XGBoost PROM model showed an accuracy of 82-90% for SCB and 0.80-0.90 for AUROC in aTSA patients, and an accuracy of 87-93% for SCB and 0.81-0.89 for AUROC in rTSA patients. In other embodiments, the XGBoost pain / function / ROM model showed an accuracy of 76-89% for SCB and 0.73-0.86 for AUROC in aTSA patients, and an accuracy of 88-90% for SCB and 0.77-0.88 for AUROC in rTSA patients.

[0079] In some embodiments, predictive outcome analysis may demonstrate the effectiveness of multiple machine learning techniques for generating models that accurately predict three PROM scores, pain and function scores, and three active ROM indices at multiple postoperative follow-up points for both aTSA and rTSA. Predictive accuracy for PROM, pain relief, and function was comparable for aTSA and rTSA patients at each time point analyzed. Breadth and depth techniques consistently demonstrated the best overall predictive performance. Most importantly, these models can risk stratify patients by accurately identifying those at highest risk of a poor outcome (e.g., those unable to achieve the MCID threshold) and those most likely to achieve a good outcome (e.g., those likely to achieve the SCB threshold).

[0080] However, the use of 291 exemplary variable inputs in these shoulder joint replacement examples may not be a practical tool for orthopedic surgeons to use clinically, given the significant data entry and time burden on both surgeons and patients. Examining the F-score results of this analysis and the application of extensive knowledge related to total shoulder joint replacement, a simplified model requiring only 10–20% of the inputs of the original model was generated. Therefore, the clinical deployment of such software-based predictive outcome tools may be more practical for orthopedic surgeons to use clinically without sacrificing the predictive accuracy of the model.

[0081] Figure 20 is a table showing a list of predictive model inputs to a machine learning model for calculating a comprehensive shoulder function score, a VAS pain score, and active abduction, active forward elevation, and active external rotation, according to one or more embodiments of the present disclosure.

[0082] Figure 21 is a table showing a list of additional predictive model inputs to a machine learning model for calculating ASES scores, according to one or more embodiments of the present disclosure. These are predictive model inputs in addition to those shown in Figure 20.

[0083] Figure 22 is a table showing a list of additional predictive model inputs to a machine learning model for calculating a constant score, according to one or more embodiments of the present disclosure. These are predictive model inputs in addition to those shown in Figure 20. The CT pre-planning predictive model and CT post-predictive model in Figures 20-22 may be equivalent to the initial pre-operative predictive MLM 50 and image-based predictive MLM 52 in system 10 of Figure 1, respectively.

[0084] In some embodiments, a triple predictive outcome model (1. user input of active ROM, pain score, and comprehensive shoulder function score = 19, 2. additional user input of AES = 10, 3. additional user input of Constant = 20) may be formulated, which may be divided into two steps: a step of generating a first predictive model, also referred to herein as the initial preoperative predictive model, using data input prior to the image-based (e.g., 3DCT-based) surgical planning step, and a step of generating a second predictive model, also referred to herein as the final preoperative predictive model, which includes additional data obtained from the image-based (e.g., 3DCT-based) surgical planning step. The data used in the first predictive model may utilize patient demographics, diagnosis, comorbidities, patient history, physician's measurements of active range of motion, patient-specific answers to several highly predictable questions, and patient-specific answers to questions with ASES and Constant scores. A complete list of these questions for these triple outcome models is shown in the tables of Figures 20, 21, and 22, respectively.

[0085] In some embodiments, the data used in the second predictive model may utilize output from a surgeon indicating the ideal size, type, and positioning of an implant that anatomically fits the patient's bone during an image-based (e.g., 3D CT) reconstructive surgery step. A proposed workflow illustrating the patient flow from clinic to surgery, and how these predictive models are used with pre-medical imaging (CT) and post-medical imaging (post-CT) plans to determine appropriate treatment at each stage, is illustrated in Figure 23.

[0086] Figure 23 is an exemplary flowchart 100 for modeling predictive outcomes of arthroplasty surgical procedures according to one or more embodiments of the present disclosure. The exemplary flowchart 100, with reference to Figure 1, may include a patient 25 entering the clinic (step 105) and consulting with a physician 20 about an arthroplasty surgical procedure to improve or replace a joint. The physician 20 may collect preoperative patient-specific data from the patient 25, which may be input into a patient-specific data collection module 46 executed by a processor 45 of a computing device 77. Alternatively, and / or optionally, the patient-specific data collection module 46 may query several N electronic resources (40A and 40B) for patient-specific preoperative data that can be received by a server 15 via a communication network 30. The received dataset may include preoperative patient-specific data for arthroplasty surgery performed on the patient's joint, if the preoperative patient-specific data further includes the patient's medical history, a measured range of at least one movement for at least one type of joint movement, and at least one pain associated with the joint, or any combination thereof.

[0087] In some embodiments, the received preoperative patient-specific data may be input into an initial preoperative predictive machine learning model (MLM) 115 (e.g., the initial preoperative predictive MLM 50 in Figure 1), also referred to herein as the first machine learning model.

[0088] In some embodiments, the initial preoperative prediction MLM 115 can determine a first predicted postoperative joint performance data output that includes at least one first predicted postoperative joint performance metric, which can then be displayed to a user, such as a physician 20, on the display of a computing device 77.

[0089] In some embodiments, the physician 20 and the patient 25 may have an initial patient consultation 120. The physician 20 and / or the patient 25 may decide to proceed with arthroplasty of the joint, delay the surgery, or pursue other treatments 125 for the diseased joint.

[0090] In some embodiments, physician 20 may request that patient 25 receive at least one medical image of the joint, such as a computed tomography (CT) scan 130 obtained from at least one medical imaging procedure performed on patient 25. The at least one medical image of the joint may include an X-ray image, a computed tomography (CT) image, a magnetic resonance image, a three-dimensional (3D) image, and / or a 3D medical image based on multiple X-ray images. The at least one medical image of the joint may also include an image of the bone and / or connective tissue attached to and / or forming the joint.

[0091] In some embodiments, in the 135 steps of Guided Personalized Surgery (GPS) preoperative planning, the CT image-based (GPS) joint reconstruction planning module 48 may be a software program executed by a processor 45 on a server 15, which may generate a joint reconstruction plan displayed on a GUI 75. The CT image-based (GPS) joint reconstruction planning module 48 may also be referred to herein as GPS planning software, as shown in Figure 20.

[0092] In some embodiments, the reconstruction plan can utilize at least one arthroplasty parameter selected by the physician, depending on the physician's consideration of a first predicted postoperative joint performance data output. The reconstruction plan may include at least one arthroplasty parameter, selected from, but not limited to, at least one implant, at least one implant size, at least one arthroplasty procedure, and / or at least one location for implanting at least one implant in the joint. The reconstruction plan may include different views of at least one medical image of the joint, such as a CT scan 130, which may be displayed in GUI 75 along with an image of at least one implant implanted in the joint. In other embodiments, for shoulder arthroplasty, at least one arthroplasty parameter may include any of the user inputs from GPS planning software, as shown in the table in Figure 20.

[0093] In some embodiments, at least one arthroplasty surgical parameter may be input to a final preoperative prediction model 140 (e.g., image-based prediction MLM 52 in Figure 1), also referred to herein as a second machine learning model. The at least one arthroplasty surgical parameter may include, for example, in the case of shoulder arthroplasty, any of the data inputs to a post-CT planning prediction model (e.g., final preoperative prediction model 140) as shown in the table in Figure 20. In other embodiments, the data input to the second machine learning model may include any appropriate parameters extracted from the reconstruction plan, similar to any of the inputs to the first machine learning model. In some embodiments, the first machine learning model (e.g., initial preoperative prediction MLM 115) and the second machine learning model (e.g., final preoperative prediction model 140) may be the same machine learning model.

[0094] In some embodiments, a software application for modeling the predicted outcomes of arthroplasty procedures performed by the processor 45 may include any or all of the following software modules: a patient-specific data acquisition module 46, a CT image-based guided personalized surgery (GPS) joint reconstruction planning module 48, an initial preoperative predictive machine learning model (MLM) module 50, an image-based predictive MLM module 52, a machine learning model training module 54, and / or a GUI management module 56. In other embodiments, the initial preoperative predictive machine learning model (MLM) module 50 and the image-based predictive MLM module 52 may be the same machine learning model.

[0095] In some embodiments, a software application for modeling the predicted outcomes of arthroplasty surgical procedures may be run by a processor 45, and a GUI management 56 may remotely control a GUI 75 running on a computing device 77 for providing input and / or output from a server 15.

[0096] In some embodiments, the first predicted postoperative joint performance data output and / or the second predicted postoperative joint performance data output may be displayed to the physician 20 on the GUI 75 in any appropriate format, such as outputting a list of predicted postoperative outcome metrics for the joint to a predictive outcome machine learning model based on data inputs such as preoperative patient-specific data, medical images of the joint, and arthroplasty parameters. A visual representation of the implant embedded in the joint based on medical images of the joint. The visual representation of the implant embedded in the joint may include raw images, enhanced images, and / or augmented images of the joint that can be displayed on the GUI 75.

[0097] In some embodiments, the second predicted postoperative joint performance data output may include displaying on GUI75 at least one arthroplasty recommendation for a combination of surgical procedure, implant type, implant size, and implant location, along with a predicted postoperative outcome metric from the model for each combination, so that the surgeon can optimize postoperative joint performance by changing the surgical parameters of the arthroplasty. This optimization may be performed preoperatively or intraoperatively.

[0098] In some embodiments, at least one recommendation for arthroplasty may include a recommendation to discontinue the arthroplasty procedure and / or to pursue an alternative treatment.

[0099] In some embodiments, the final preoperative prediction model 140 may determine a second predicted postoperative joint performance data output that includes at least one second predicted postoperative performance metric for the joint, which may then be displayed to a user, such as a physician 20, on the GUI 75 of the computing device 77.

[0100] In some embodiments, the physician 20 can review the second predicted postoperative joint performance data output and conduct a final patient consultation 145 with the patient 25. The physician 20 and / or the patient 25 can decide to schedule arthroplasty 155 of the joint, postpone the surgery, or pursue other treatments 150 for the diseased joint.

[0101] Figure 24 is a table showing a comparison of the mean absolute error (MAE) associated with ASES predictions using a full XGBoost machine learning model and a simplified XGBoost machine learning model in one or more embodiments of the present disclosure.

[0102] Figure 25 is a table showing a comparison of the mean absolute error (MAE) associated with a given prediction using a full XGBoost machine learning model and a simplified XGBoost machine learning model, according to one or more embodiments of the present disclosure.

[0103] Figure 26 is a table showing a comparison of the mean absolute error (MAE) associated with comprehensive shoulder function score prediction using a full XGBoost machine learning model and a simplified XGBoost machine learning model in one or more embodiments of the present disclosure.

[0104] Figure 27 is a table showing a comparison of the mean absolute error (MAE) associated with VAS pain score prediction using a full XGBoost machine learning model and a simplified XGBoost machine learning model according to one or more embodiments of the present disclosure.

[0105] Figure 28 is a table showing a comparison of the mean absolute error (MAE) associated with active abduction prediction using a full XGBoost machine learning model and a simplified XGBoost machine learning model in one or more embodiments of the present disclosure.

[0106] Figure 29 is a table showing a comparison of the mean absolute error (MAE) associated with active forward lift prediction using a full XGBoost machine learning model and a simplified XGBoost machine learning model in one or more embodiments of the present disclosure.

[0107] Figure 30 is a table showing a comparison of the mean absolute error (MAE) associated with active external rotation prediction using a full XGBoost machine learning model and a simplified XGBoost machine learning model in one or more embodiments of the present disclosure.

[0108] Figure 31 is a table showing a comparison of full XGBoost model predictions for aTSA and rTSA patients who experienced clinical improvement exceeding the MCID threshold for several different outcome measures, according to one or more embodiments of the present disclosure.

[0109] Figure 32 is a table showing a comparison of simplified XGBoost model predictions for aTSA and rTSA patients who experienced clinical improvement exceeding the MCID threshold for several different outcome measures, according to one or more embodiments of the present disclosure.

[0110] Figure 33 is a table showing a comparison of full XGBoost model predictions for aTSA and rTSA patients who experienced clinical improvement exceeding the SCB threshold for several different outcome measures, according to one or more embodiments of the present disclosure.

[0111] Figure 34 is a table showing a comparison of simplified XGBoost model predictions for aTSA and rTSA patients who experienced clinical improvement exceeding the SCB threshold for several different outcome measures, according to one or more embodiments of the present disclosure.

[0112] Figure 35 is a table showing a comparison of simplified XGBoost models with input from CT plan data for predicting clinical improvement exceeding the MCID threshold for multiple different outcome measures in aTSA and rTSA patients, according to one or more embodiments of the present disclosure.

[0113] Figure 36 is a table showing a comparison of simplified XGBoost models with input from CT plan data for predicting aTSA and rTSA patients who experienced clinical improvement exceeding the SCB threshold for several different outcome measures, according to one or more embodiments of the present disclosure.

[0114] In some embodiments, the model inputs (in the pre- and post-planning stages of the predictive model) may be the most highly predictable parameters, which can provide a very similar level of predictive accuracy as using all variables from the clinical outcome database 62. As shown in the tables in Figures 24–30, the results of the simplified model may yield nearly the same accuracy for each outcome metric as a predictive model inputting data from the entire clinical outcome database 62.

[0115] In some embodiments, the predictive accuracy between aTSA and rTSA was observed to be similar for both the full and simplified models. Furthermore, in both the full and simplified predictive models, MAE was found to be slightly higher at the earlier postoperative time point than at the later postoperative time point. At all postoperative time points analyzed, the mean difference in MAE between the full model prediction and the simplified model prediction was found to be ±0.3 MAE (±0.3 aTSA and ±0.4 rTSA) for the ASES score, ±0.9 (±0.7 aTSA and ±0.8 rTSA) for the Constant score, ±0.1 (±0.1 aTSA and ±0.1 rTSA) for the Comprehensive Shoulder Function score, ±0.1 (±0.0 aTSA and ±0.2 rTSA) for the VAS pain score, ±1.4° (±1.1 aTSA and 1.2 rTSA) for abduction, ±1.6° (±1.7 aTSA and ±1.4 rTSA) for anterior elevation, and ±0.4° (±0.1 aTSA and ±0.4 rTSA) for external rotation.

[0116] In some embodiments, as shown in the tables in Figures 31–34, the simplified models also yielded nearly identical MCID and SCB accuracy results, demonstrating the ability of these models to effectively risk stratify patients before surgery based on their ability to achieve varying degrees of improvement at 2–3 year follow-up according to several different outcome metrics.

[0117] In some embodiments, particularly with respect to MCID, the fully predictive model achieved an accuracy of 82-96% in MCID with an AUROC of 0.75-0.97 for aTSA patients, while the simplified predictive model achieved an accuracy of 82-96% in MCID with an AUROC of 0.70-0.95 for aTSA patients. The fully predictive model achieved an accuracy of 91-99% in MCID with an AUROC of 0.82-0.98 for rTSA patients, while the simplified predictive model achieved an accuracy of 91-99% in MCID with an AUROC of 0.84-0.94 for rTSA patients.

[0118] In some embodiments, similarly for SCB, the fully predictive model achieved 79-90% accuracy in SCB for aTSA patients with an AUROC between 0.74 and 0.90, while the simplified predictive model achieved 76-90% accuracy in SCB for aTSA patients with an AUROC between 0.70 and 0.89. Ultimately, the fully predictive model achieved 83-92% accuracy in SCB for rTSA patients with an AUROC between 0.78 and 0.88, while the simplified predictive model achieved 81-90% accuracy in SCB for rTSA patients with an AUROC between 0.70 and 0.87. Regarding the interpretation of the AUROC values ​​used for these MCID and SCB predictions, for a predictive model, 0.5 is considered a guess, >0.7 is considered acceptable, >0.8 is considered good, and >0.9 is considered to have excellent discriminatory power.

[0119] In some embodiments, for the simplified model algorithms, the average MCID AUROC values ​​were 0.82 for aTSA and 0.89 for rTSA, and the average SCB AUROC values ​​were 0.85 for aTSA and 0.82 for rTSA, suggesting that these algorithms, generated from a minimal set of features, exhibit, on average, discriminative ability between good and excellent, and at worst, acceptable. These simplified model predictions may be improved by adding them to implant data selected from guided personalized surgery (GPS) CT plans, as shown in the tables in Figures 24-30 and 35-36. It should be noted that this data and technique disclosed herein can be used to generate other predictive models, such as internal rotation scores, worst-case visual analog pain, and shoulder arthroplasty smart scores.

[0120] Figure 24 is a table showing a comparison of the mean absolute error (MAE) associated with ASES predictions using a full XGBoost machine learning model and a simplified XGBoost machine learning model in one or more embodiments of the present disclosure.

[0121] Figure 25 is a table showing a comparison of the mean absolute error (MAE) associated with constant predictions using a full XGBoost machine learning model and a simplified XGBoost machine learning model in one or more embodiments of the present disclosure.

[0122] Figure 26 is a table showing a comparison of the mean absolute error (MAE) associated with predicting comprehensive shoulder function scores using a full XGBoost machine learning model and a simplified XGBoost machine learning model in one or more embodiments of the present disclosure.

[0123] Figure 27 is a table showing a comparison of the mean absolute error (MAE) associated with predicting VAS pain scores using a full XGBoost machine learning model and a simplified XGBoost machine learning model in one or more embodiments of the present disclosure.

[0124] Figure 28 is a table showing a comparison of the mean absolute error (MAE) associated with predicting active abduction using a full XGBoost machine learning model and a simplified XGBoost machine learning model in one or more embodiments of the present disclosure.

[0125] Figure 29 is a table showing a comparison of the mean absolute error (MAE) associated with active forward elevation prediction using a full XGBoost machine learning model and a simplified XGBoost machine learning model in one or more embodiments of the present disclosure.

[0126] Figure 30 is a table showing a comparison of the mean absolute error (MAE) associated with predicting active external rotation using a full XGBoost machine learning model and a simplified XGBoost machine learning model in one or more embodiments of the present disclosure.

[0127] Therefore, the machine learning predictive models described herein can effectively provide the same predictive accuracy for clinical outcomes for aTSA and rTSA for a given patient before arthroplasty, based on using more than 75% less user input for the simplified predictive model than for the fully predictive model. This significant reduction in user input data makes it possible to use such tools in a surgeon's clinic, as they require a similar input burden to other patient-reported outcome metrics commonly used to quantify clinical outcomes after aTSA and rTSA.

[0128] In some embodiments, the machine learning model used in the software application may be a simplified machine learning model for improving the computational efficiency and / or computational speed of the server 15, as shown in the table in the previous figure.

[0129] In other words, the initial preoperative prediction MLM50 and the image-based prediction MLM52 may be simplified machine learning models, which in this specification may be referred to as the first simplified MLM and the second simplified MLM, respectively.

[0130] In some embodiments, in addition to outcome metrics and range of motion predictions, the predictive outcome model may identify factors that influence the prediction. Specifically, for patient-modifiable factors, the predictive outcome model can provide the patient with recommendations on what they can do to improve their outcome predictions, enabling them to participate more actively in their surgeon-patient consultations.

[0131] In some embodiments, the predictive outcome model may incorporate a lookup table of typical comorbidity rates that may be associated with aTSA and rTSA for a given patient's demographics, diagnosis, patient history, and / or comorbidities.

[0132] In some embodiments, predictive outcome models may provide surgeons with additional capabilities that can help achieve better predicted outcomes. For example, outcomes may improve by 2% if the case is navigated correctly. Also, as another exemplary example, if a patient has 10 degrees of glenoid retroversion, using augmented glenoid components for aTSA and / or rTSA, in contrast to standard components (with or without eccentric glenoid reaming surgical techniques), may predict better outcomes.

[0133] In some embodiments, trade-offs between implant technologies may be implemented to improve the decision-making of surgeon users. For example, to inform surgeons when to use aTSA versus rTSA for patients with different rotator cuff tear dimensions; when to use aTSA versus rTSA for different Gutalie rotator cuff fat infiltration grades; when to use bone graft versus augmented glenoid components for different glenoid deformity classification types (Warch, Favar, Antuna, etc.) or specific glenoid wear measurements (retroversion, tilt, or beta angle); when and to what extent to perform glenoid eccentric reaming versus off-axial reaming to correct glenoid wear; and / or when to use standard-length humeral stem versus short humeral stem versus stemless humeral implant, and which size implant to select based on bone quality.

[0134] In some embodiments, these arthroplasty parameters can be changed on the fly in a second predicted postoperative joint performance data output on the software platform, either preoperatively or intraoperatively, in response to the surgeon (e.g., user) changing any one of the at least one arthroplasty parameters in the reconstruction plan, either preoperatively or intraoperatively.

[0135] In some embodiments, data from early postoperative follow-up visits at 2 weeks, 6 weeks, 8 weeks, 12 weeks, 4 months, or earlier can be used to predict outcomes at different postoperative time points. The advantage of such postoperative predictions is that they may allow for a more accurate estimation of patient-specific improvement. This data may be useful in establishing more patient-specific rehabilitation protocols that target improvement in a given metric compared to other outcome metrics.

[0136] In some embodiments, these predictive models can be further refined using this data or additional data (and / or additional data directly ingested from the patient's electronic medical records or other databases, such as data stored in the cloud and / or data generated from wearable devices capable of measuring the patient's movement and / or activity level, which may accept patient responses related to pain levels) to create more accurate inputs using the additional data. This data may also be useful for recommendations regarding medical workflows, such as risk stratification of patients for shoulder arthroplasty and identification of patients who can safely undergo surgery in an outpatient surgical clinic. The predictive models can make recommendations on whether a particular patient should undergo inpatient or outpatient surgery in a hospital. Furthermore, the predictive models can also provide recommendations for the length of hospital stay after surgery for a particular patient.

[0137] Finally, as more clinical data is added to the clinical outcome database 62 over time, the model learning module 54 may be used to update the machine learning algorithm accordingly to reduce prediction errors. This allows the predictive outcome algorithm to continuously learn based on the new data input using the tool. Furthermore, new parameters may be added in the future and the rank of existing parameters may be changed to further improve the predictive model from data directly from CT and / or MRI images, e.g., bone density, bone structure, soft tissue tears, and / or other soft tissue injuries such as rotator cuff fat infiltration, thereby further assisting physicians in their clinical decisions regarding treatment and / or outcome predictions.

[0138] In some embodiments, the bone-to-bone relationships of the shoulder joint or other joints may be evaluated from these images, and patient-specific data can influence predictive models as new inputs that further support clinical decisions for treatment or outcome prediction. With new data, predictive models may be more transplantable and generalizable to other total shoulder joint replacement systems, and perhaps even to other arthroplasty systems for different joints and applications such as spine, hip, knee, ankle, and trauma. Improved predictive accuracy of predictive outcome models allows for better clinical decisions regarding implant type, size, and location, resulting in increased patient and surgeon satisfaction and more realistic outcomes.

[0139] Figure 37 is a flowchart of an exemplary method 200 for modeling predictive outcomes of arthroplasty surgical procedures, according to one or more embodiments of the present disclosure. This method may be performed by a processor 45 of server 15.

[0140] Method 200 may include receiving preoperative patient-specific data for arthroplasty to be performed on the patient's joints 210.

[0141] Method 200 may also include inputting preoperative patient-specific data into at least one first machine learning model 220 to determine a first predicted postoperative joint performance data output, the first predicted postoperative joint performance data output including at least one first predicted postoperative outcome metric for the joint.

[0142] Method 200 may also include displaying first predicted postoperative joint performance data output on a display to the user 230.

[0143] Method 200 may include receiving at least one medical image of a joint obtained from at least one medical imaging procedure performed on a patient 240.

[0144] Method 200 includes generating a joint reconstruction plan for a patient based on at least one medical image of the joint and at least one arthroplasty parameter obtained from the user in response to a displayed first predicted postoperative joint performance data output, the reconstruction plan may include at least one arthroplasty parameter selected from at least one implant, at least one implant size, at least one arthroplasty procedure, at least one location for implanting at least one implant in the joint, or any combination thereof.

[0145] Method 200 may include inputting at least one arthroplasty parameter into at least one second machine learning model 260 to determine a second predicted postoperative joint performance data output that includes at least one second predicted postoperative outcome metric for the joint.

[0146] Method 200 may also include displaying to the user second predicted postoperative joint performance data output on a display 270.

[0147] Method 200 may include updating a second predicted postoperative joint performance data output, which is displayed to include at least one arthroplasty recommendation, in response to the user changing any of at least one arthroplasty parameter before, during, or both arthroplasty. This allows the surgeon 20 to adjust any of the surgical parameters on the fly, either preoperatively and / or during the arthroplasty procedure, to optimize any of the predicted postoperative outcome indicators.

[0148] In some embodiments, the device may include a processor and non-temporary memory that stores instructions to be executed by the processor when executed by the processor, and the general instructions are performed by the processor. To receive preoperative patient-specific data regarding arthroplasty performed on the patient's joints, The first predicted postoperative joint performance data output is determined by inputting preoperative patient-specific data into at least one machine learning model. Here, the first predicted postoperative joint performance data output can be included in at least one first predicted postoperative outcome metric of the joint. The first predictive postoperative joint performance data output is displayed to the user on the display. To receive at least one medical image of a joint obtained from at least one medical imaging procedure performed on the patient, To generate a joint reconstruction plan for a patient based on at least one medical image of the joint and at least one arthroplasty parameter obtained from the user in response to a displayed first predicted postoperative joint performance data output. Inputting at least one arthroplasty parameter into at least one machine learning model to determine a second predicted postoperative joint performance data output that includes at least one second predicted postoperative outcome metric for the joint, and This includes displaying a second set of predicted postoperative joint performance data to the user on a display.

[0149] In some embodiments, the device may include a processor and non-temporary memory that stores instructions to be executed by the processor when executed by the processor, and the general instructions are performed by the processor. To receive preoperative patient-specific data regarding arthroplasty performed on the patient's joints, Here, the preoperative patient-specific data is: (i) Patient's medical history, (ii) the range of motion for at least one type of joint movement of the joint, (iii) At least one pain metric related to the joint (iv) may be included, Inputting preoperative patient-specific data into at least one first machine learning model to determine a first predicted postoperative joint performance data output. Here, the first predicted postoperative joint performance data output may include at least one first predicted postoperative outcome metric for the joint. The first predictive postoperative joint performance data output is displayed to the user on the display. To receive at least one medical image of a joint obtained from at least one medical imaging procedure performed on the patient, To generate a joint reconstruction plan for a patient based on at least one medical image of the joint and at least one arthroplasty parameter obtained from the user in response to a displayed first predicted postoperative joint performance data output. Here, the reconstruction plan may include at least one arthroplasty parameter selected from the following: (i) at least one implant, (ii) at least one implant size (iii) at least one arthroplasty procedure, (iv) at least one location for implanting at least one implant in the joint, or (v) Any combination of them Inputting at least one arthroplasty parameter into at least a second machine learning model to determine a second predicted postoperative joint performance data output that includes at least one second predicted postoperative outcome metric for the joint, The second predictive postoperative joint performance data output is displayed to the user on the display, and The system includes updating a displayed second predicted postoperative joint performance data output to include at least one arthroplasty recommendation in response to a user changing at least one arthroplasty parameter before, during, or both of the arthroplasty procedures.

[0150] In some embodiments, the processor may be configured to receive preoperative patient-specific data by receiving preoperative patient-specific data from at least one electronic medical resource via a communication network.

[0151] In some embodiments, at least one medical image may include at least one of the following: (a) an X-ray image, (b) a computed tomography image, (c) a magnetic resonance image, (d) a three-dimensional (3D) image, (e) a 3D medical image generated from multiple X-ray images, (f) a video frame, or any combination thereof.

[0152] In some embodiments, at least one first predictive postoperative outcome metric and at least one second predictive postoperative outcome metric and trick may be predicted for at least one of (a) days, (b) months, and (c) years.

[0153] In some embodiments, the processor may be configured to display a second predicted postoperative joint performance data output along with recommendations for at least one arthroplasty parameter.

[0154] In some embodiments, the joints may be selected from a group including the hip joint, knee joint, shoulder joint, elbow joint, and ankle joint.

[0155] In some embodiments, the joint may be a shoulder joint.

[0156] In some embodiments, preoperative patient-specific data may include (a) patient demographics, (b) patient diagnosis, (c) patient comorbidities, (d) patient medical history, (e) measurement of active range of motion of the shoulder, (f) patient self-reported measurements of pain, function, or both, (g) patient score based on the American Shoulder and Elbow Society Surgical Shoulder Score (ASES), (h) patient score based on the Constant Shoulder Score (CSS), or any combination thereof.

[0157] In some embodiments, at least one arthroplasty procedure may be selected from a group including anatomical total shoulder arthroplasty, reverse total shoulder arthroplasty, deltoid approach, and superior lateral approach.

[0158] In some embodiments, at least one first postoperative outcome metric and at least one second postoperative outcome metric may be selected from a group including the United States Shoulder-Elbow (ASES) score, University of California, Los Angeles (UCLA) score, constant score, comprehensive shoulder function score, visual analog scale (VAS) pain score, smart shoulder arthroplasty score, internal rotation (IR) score, abduction measurement, anterior elevation measurement, and external rotation measurement.

[0159] In some embodiments, the method is The processor receives preoperative patient-specific data about arthroplasty performed on the patient's joints. The processor inputs preoperative patient-specific data into at least one machine learning model to determine the first predicted postoperative joint performance data output. Here, the first predicted postoperative joint performance data output may include at least one first predicted postoperative outcome metric for the joint. The processor displays the first predicted postoperative joint performance data output to the user on the display. The processor receives at least one medical image of a joint obtained from at least one medical imaging procedure performed on the patient. The processor generates a patient's joint reconstruction plan based on at least one medical image of the joint and at least one arthroplasty parameter, obtained from the user in response to the displayed first predicted postoperative joint performance data output. The processor inputs the reconstruction plan into at least one machine learning model to determine a second predicted postoperative joint performance data output, which includes at least one second predicted postoperative outcome metric for the joint, and The processor may also include displaying a second predicted postoperative joint performance data output to the user on a display.

[0160] In some embodiments, the method is The processor receives preoperative patient-specific data about arthroplasty performed on the patient's joints. Here, the preoperative patient-specific data includes the following: (i) Patient's medical history, (ii) the measured range of motion for at least one type of joint movement of the joint, and (iii) at least one pain metric related to the joint, The processor inputs preoperative patient-specific data into at least one machine learning model to determine the first predicted postoperative joint performance data output. Here, the first predicted postoperative joint performance data output may include at least one first predicted postoperative outcome metric for the joint. The processor displays the first predicted postoperative joint performance data output to the user on the display. The processor receives at least one medical image of a joint obtained from at least one medical imaging procedure performed on the patient. The processor generates a joint reconstruction plan for the patient based on at least one medical image of the joint and at least one arthroplasty parameter obtained from the user in response to the displayed first predicted postoperative joint performance data output. Here, the reconstruction plan may include at least one arthroplasty parameter selected from the following: (i) at least one implant, (ii) at least one implant size, (iii) at least one arthroplasty procedure, (iv) at least one location for implanting at least one implant in the joint, or (v) Any combination of them, The processor inputs the reconstruction plan into at least one second machine learning model to determine a second predicted postoperative joint performance data output that includes at least one second predicted postoperative outcome metric for the joint. The processor displays the second predicted postoperative joint performance data output to the user on the display, and The processor may include updating a displayed second predicted postoperative joint performance data output in response to a user changing any of at least one arthroplasty parameter of the reconstruction plan before, during, or both arthroplasty, to include at least one arthroplasty recommendation.

[0161] In some embodiments, receiving preoperative patient-specific data may include receiving preoperative patient-specific data from at least one electronic medical resource via a communication network.

[0162] In some embodiments, at least one medical image may include at least one of the following: (a) an X-ray image, (b) a computed tomography image, (c) a magnetic resonance image, (d) a three-dimensional (3D) image, (e) a 3D medical image generated from multiple X-ray images, (f) a video frame, or any combination thereof.

[0163] In some embodiments, at least one first predictive postoperative outcome index and at least one second predictive postoperative outcome index may be predicted for at least one of (a) days, (b) months, and (c) years.

[0164] In some embodiments, displaying a second predicted postoperative joint performance data output may include displaying the second predicted postoperative joint performance data output along with recommendations for at least one arthroplasty parameter.

[0165] In some embodiments, the joints may be selected from a group including the hip joint, knee joint, shoulder joint, elbow joint, and ankle joint.

[0166] In some embodiments, the joint may be a shoulder joint.

[0167] In some embodiments, preoperative patient-specific data may include (a) patient demographics, (b) patient diagnosis, (c) patient comorbidities, (d) patient medical history, (e) active range of motion measurement of the shoulder, (f) patient self-reported measurements of pain, function, or both, (g) patient score based on the American Shoulder and Elbow Society Surgical Shoulder Score (ASES), (h) patient score based on the Constant Shoulder Score (CSS), the Smart Score for Shoulder Arthroplasty, or any combination thereof.

[0168] In some embodiments, at least one arthroplasty procedure may be selected from a group including anatomical total shoulder arthroplasty, reverse total shoulder arthroplasty, deltoid approach, and superior lateral approach.

[0169] In some embodiments, at least one first postoperative outcome metric and at least one second postoperative outcome metric may be selected from a group including the United States Shoulder-Elbow (ASES) score, University of California, Los Angeles (UCLA) score, constant score, comprehensive shoulder function score, visual analog scale (VAS) pain score, smart shoulder arthroplasty score, internal rotation (IR) score, abduction measurement, anterior elevation measurement, and external rotation measurement.

[0170] In some embodiments, an exemplary specially programmed computing system / platform of the present invention, equipped with associated devices, is configured to operate in a distributed network environment and communicate with one or more suitable data communication networks (e.g., the Internet, satellite, etc.) and uses, but is not limited to, one or more suitable data communication protocols / modes such as IPX / SPX, X.25, AX.25, AppleTalk™, TCP / IP (HTTP, etc.), Near Field Communication (NFC), RFID, Narrow Band Internet of Things (NBIOT), 3G, 4G, 5G, GSM, GPRS, WiFi, WiMAX, CDMA, satellite, ZigBee, and other suitable communication modes. In some embodiments, NFC can represent a near-field communication technology that communicates by “swiping,” “bumping,” “tapping,” or otherwise moving an NFC-enabled device in close proximity. In some embodiments, NFC can include a set of short-range wireless technologies that typically require a distance of 10 cm or less. In some embodiments, NFC can operate at 13.56 MHz over an ISO / IEC 18000-3 air interface at rates ranging from 106 kbit / s to 424 kbit / s. In some embodiments, NFC can include an initiator and a target, the initiator actively generating an RF field that can power a passive target. In some embodiments, this allows the NFC target to take on a very simple form factor, such as a tag, sticker, key fob, or battery-free card. In some embodiments, peer-to-peer NFC communication can occur when multiple NFC-enabled devices (e.g., smartphones) are in close proximity to each other.

[0171] The materials disclosed herein are implemented as software or firmware or a combination thereof, or as instructions stored on a machine-readable medium, which are read and executed by one or more processors. The machine-readable medium may include any medium and / or mechanism for storing or transmitting information in a format readable by a machine (e.g., a computing device). For example, the machine-readable medium may include read-only memory (ROM), random-access memory (RAM); magnetic disk storage media; optical storage media; flash memory devices; electrically, optical, acoustic or other forms of propagating signals (e.g., carrier waves, infrared signals, digital signals, etc.), and others.

[0172] Examples of hardware elements may include processors, microprocessors, circuits, circuit elements (e.g., transistors, resistors, capacitors, inductors, etc.), integrated circuits, application-specific integrated circuits (ASICs), programmable logic devices (PLDs), digital signal processors (DSPs), field-programmable gate arrays (FPGAs), logic gates, registers, semiconductor devices, chips, microchips, chipsets, and the like. In some embodiments, one or more processors may be implemented as a composite instruction set computer (CISC) processor or a reduced instruction set computer (RISC) processor, an x86 instruction set compatible processor, a multicore, or other microprocessor or central processing unit (CPU). In various implementations, one or more processors may be a dual-core processor, a dual-core mobile processor, and the like.

[0173] The computer-related systems, computer systems, and systems used herein include any combination of hardware and software. Examples of software may include software components, operating system software, middleware, firmware, software modules, routines, subroutines, functions, methods, procedures, software interfaces, application programming interfaces (APIs), instruction sets, computer code, computer code segments, words, values, symbols, or any combination thereof. The decision of whether a particular embodiment is implemented using hardware and / or software elements may vary depending on any number of factors, such as desired computing speed, power level, thermal tolerance, processing cycle budget, input data rate, output data rate, memory resources, data bus speed, and other design or performance constraints.

[0174] One or more aspects of at least one embodiment may be implemented by targeted instructions stored on a machine-readable medium representing various logic within a processor, which, when read by a machine, cause the machine to create logic that performs the techniques described herein. Such representatives are known as “IP cores” and may be stored on a tangible machine-readable medium and provided to various customers or manufacturing facilities to be loaded into a creation machine or processor that creates the logic. Naturally, the various embodiments described herein can of course be implemented using any suitable hardware and / or computing software language (e.g., C++, Objective-C, Swift, Java, JavaScript, Python, Perl, QT, etc.).

[0175] In some embodiments, one or more of the exemplary computer-based systems / platforms of the present invention, exemplary computer-based devices of the present invention, and / or exemplary computer-based components of the present invention, such as computing device 77, may be partially or whole-time included in at least one personal computer (PC), laptop computer, ultralaptop computer, tablet, touchpad, portable computer, handheld computer, palmtop computer, personal digital assistant (PDA), mobile phone, mobile phone / PDA combination, television, smart device (e.g., smartphone, smart tablet, or smart television), mobile internet device (MID), messaging device, data communication device, etc.

[0176] As used herein, the term “server” should be understood to refer to a service point that provides processing, database, and communication functions. For example, rather than being limited, the term “server” may refer to a single physical processor with associated communication and data storage and database functions, a complex of networked or clustered processors and associated network and storage devices, or to operating software and one or more database systems and application software that support the services provided by the server. A cloud server is an example of this.

[0177] In some embodiments, as detailed herein, one or more of the exemplary original computer-based systems / platforms, exemplary original computer-based devices, and / or exemplary original computer-based components of the Disclosure may acquire, manipulate, transfer, store, transform, generate, and / or output (e.g., from inside and / or outside a particular application) any digital object and / or data unit, which may be in any suitable form such as files, contacts, tasks, emails, social media posts, maps, and complete applications (e.g., spreadsheets). In some embodiments, as detailed herein, one or more exemplary original computer-based systems / platforms, exemplary original computer-based devices, and / or exemplary original computer-based components of this disclosure are, but are not limited to, (1) FreeBSD, NetBSD, OpenBSD, (2) Linux, (3) Microsoft Windows, (4) OS X (MacOS), (5) MacOS 11, (6) Solaris, (7) Android, (8) iOS, (9) Embedded Linux, (10) Tizen, (11) WebOS, (12) IBM i, (13) IBM AIX, (14) Binary Runtime Environment for Wireless (BREW), (15) Cocoa (API), (16) Cocoa Touch, (17) Java Platform, (18) JavaFX, (19) JavaFX Mobile, (20) Microsoft It may be implemented across one or more different computer platforms, such as DirectX, (21) the .NET framework, (22) Silverlight, (23) the Open Web Platform, (24) Oracle Database, (25) Qt, (26) the Eclipse Rich Client Platform, (27) SAP NetWeaver, (28) SmartFace and / or (29) the Windows Runtime.

[0178] In some embodiments, the exemplary original computer-based systems / platforms, exemplary original computer-based devices, and / or exemplary original computer-based components of the Disclosure may be configured to utilize hardware-implemented circuits that can be used instead of, or in combination with, software instructions for implementing features consistent with the principles of the Disclosure. Therefore, implementations consistent with the principles of the Disclosure are not limited to any particular combination of hardware circuits and software. For example, various embodiments may be embodied in many different ways, such as software components including standalone software packages, combinations of software packages, or software packages incorporated as “tools” into larger software products.

[0179] For example, exemplary software specifically programmed in accordance with one or more principles of this disclosure may be available for download from a network, such as a website, as a standalone product or as an add-in package for installation in existing software. For example, exemplary software specifically programmed in accordance with one or more principles of this disclosure may also be available as a client-server software application or a web-enabled software application. For example, exemplary software specifically programmed in accordance with one or more principles of this disclosure may be embodied as a software package installed on a hardware device.

[0180] In some embodiments, the exemplary original computer-based systems / platforms, exemplary original computer-based devices, and / or exemplary original computer-based components of this disclosure are, but are not limited to, at least 100 people (e.g., not limited to 100-999 people), at least 1,000 people (e.g., not limited to 1,000-9,999 people), at least 10,000 people (e.g., not limited to 10,000-99,999 people), and at least 100,000 people (e.g., It may be configured to handle a large number of concurrent users, such as 100,000 to 999,999, at least 1,000,000 (e.g., not limited to, but 1,000,000 to 9,999,999), at least 10,000,000 (e.g., not limited to, but 10,000,000 to 99,999,999), at least 100,000,000 (e.g., not limited to, but 100,000,000 to 999,999,999), and at least 1,000,000,000 (e.g., not limited to, but 1,000,000,000 to 999,999,999,999).

[0181] In some embodiments, exemplary original computer-based systems / platforms, exemplary original computer-based devices, and / or exemplary original computer-based components of the Disclosure may be configured to output to a separate, particularly programmed graphical user interface implementation of the Disclosure (e.g., a desktop, a web application, etc.). In various implementations of the Disclosure, the final output may be displayed on a display screen, which may be, but not limited to, a computer screen, a mobile device screen, etc. In various implementations, the display may be a holographic display. In various implementations, the display may be a transparent surface that can receive visual projections. Such projections may convey various forms of information, images, and / or objects. For example, such projections may be visual overlays for mobile augmented reality (MAR) applications.

[0182] As used herein, terms such as “mobile electronic device” may refer to any portable electronic device that may or may not be capable of location tracking (e.g., MAC address, Internet Protocol (IP) address, etc.). For example, a mobile electronic device may include, but is not limited to, a mobile phone, personal digital assistant (PDA), Blackberry®, Pager, smartphone, or any other reasonable mobile electronic device.

[0183] As used herein, the terms “cloud,” “Internet cloud,” “cloud computing,” “cloud architecture,” and similar terms correspond to at least one of the following: (1) a large number of computers connected via a real-time communication network (e.g., the Internet); (2) providing the ability to run programs or applications simultaneously on a large number of connected computers (e.g., physical machines, virtual machines (VMs)); and (3) providing network-based services that appear to be provided by actual server hardware but are actually provided by virtual hardware (e.g., virtual servers) that simulates one or more real machines (e.g., which can be moved and scaled in place without affecting end users).

[0184] In some embodiments, exemplary original computer-based systems / platforms, exemplary original computer-based devices, and / or exemplary original computer-based components of the present disclosure may be configured to securely store and / or transmit data by utilizing one or more cryptographic techniques (e.g., private-key and public-key pairs, Triple Data Encryption Standard (3DES), block cipher algorithms (e.g., IDEA, RC2, RC5, CAST, and Skipjack), cryptographic hash algorithms (e.g., MD5, RIPEMD-160, RTR0, SHA-1, SHA-2), Tiger (TTH), WHIRLPOOL, RNGs).

[0185] The examples given above are, of course, illustrative and not limiting.

[0186] As used herein, the term "user" means at least one user. In the context as used herein, a user may be a physician or surgeon, or a person acting on behalf of a physician or surgeon, laboratory technician, surgical staff, etc.

[0187] In some embodiments, the exemplary original computer-based systems / platforms, exemplary original computer-based devices, and / or exemplary original computer-based components of this disclosure may be configured to utilize one or more exemplary AI / machine learning techniques selected from, but not limited to, decision trees, boosting, support vector machines, neural networks, nearest neighbor algorithms, Naive Bayes, bagging, random forests, and the like. In some embodiments, and optionally in any combination of the embodiments described above or below, the exemplary neural network technique may be, but not limited to, a feedforward neural network, a radial basis function network, a recurrent neural network, a convolutional network (e.g., U-net), or one of other suitable networks. In some embodiments, optionally in combination of any embodiments described above or below, exemplary implementations of neural networks can be carried out as follows: i) Defining a neural network architecture / model, ii) Transferring input data to an exemplary neural network model, iii) Training an exemplary model step by step iv) Determine the accuracy for a specific number of time steps. v) Apply an exemplary trained model to process newly received input data. vi) Continuously training an exemplary pre-trained model at a predetermined periodicity, both at will and in parallel.

[0188] In some embodiments, optionally, by combining any embodiments above or below, the exemplary trained neural network model may be identified by at least the neural network topology, a set of activation functions, and connection weights. For example, the topology of the neural network may include the configuration of the nodes of the neural network and the connections between such nodes. In some embodiments, optionally, by combining any embodiments above or below, the exemplary trained neural network model may be identified to include other parameters, but not limited to bias values / functions and / or aggregate functions. For example, the activation function of a node may be a step function, a sine function, a continuous linear function or a piecewise linear function, a sigmoid function, a hyperbolic tangent function, or other type of mathematical function that represents a threshold at which the node is activated. In some embodiments, optionally, by combining any embodiments above or below, the exemplary aggregate function may be a mathematical function that combines (e.g., sum, product, etc.) the input signals to the node. In some embodiments, optionally, by combining any embodiments above or below, the output of the exemplary aggregate function may be used as an input to the exemplary activation function. In some embodiments, and optionally in combination of any embodiments described above or below, the bias may be a constant value or function that can be used by aggregate functions and / or activation functions to make a node more easily or less easily activated.

[0189] The disclosures described herein can be implemented without any elements, limitations, or restrictions not specifically disclosed herein. For example, in each example herein, the terms “having,” “essentially consisting of,” and “including” can be replaced with any of the other two terms without altering their respective meanings as defined herein. The terms and expressions used are for illustrative purposes only and are not limitations; the use of such terms and expressions is not intended to exclude equivalents of the functions or parts thereof shown or described, but it should be understood that various modifications are possible within the scope of this disclosure.

Claims

1. Processor and The display and A non-temporary memory that stores instructions to be executed by the aforementioned processor when it is executed by the aforementioned processor, Equipped with, The aforementioned instruction is performed by the processor, To receive preoperative patient-specific data regarding arthroplasty performed on the patient's joints, Here, the aforementioned preoperative patient-specific data (i) The patient’s medical history, (ii) The measured range of motion for at least one type of joint movement of the joint, and (iii) at least one pain metric associated with the joint It has, The patient-specific data from the preoperative period is input into at least one first machine learning model to determine the output of first predicted postoperative joint performance data at multiple first postoperative time points. Here, the first predicted postoperative joint performance data output is at least, (i) A first predicted range of motion for at least one type of joint movement of the joint, (ii) at least one first predictive pain metric associated with the joint It has, Here, the at least one first machine learning model is trained to output data including a plurality of first values ​​for the first predicted postoperative joint performance data output at a plurality of first postoperative time points after surgery, where each first value corresponds to each specific first time point at the plurality of first postoperative time points after surgery. Here, the input data for training the at least one first machine learning model includes at least the preoperative patient-specific data. The first predicted postoperative joint performance data output is displayed to the user on the display as the first predicted postoperative joint performance data output. To receive at least one medical image of the joint obtained from at least one medical imaging procedure performed on the patient, The user receives at least one arthroplasty parameter based on the displayed first predicted postoperative joint performance data output. Here, the at least one arthroplasty parameter is (i) at least one implant, (ii) at least one implant size, (iii) at least one arthroplasty procedure, (iv) at least one location in the joint for embedding the at least one implant, or (v) these combinations Selected from, To generate a reconstruction plan for the patient's joint based at least partially on the at least one medical image of the joint and at least one arthroplasty parameter, The patient-specific data from the preoperative period and the reconstruction plan data are input into at least one second machine learning model to determine the output of second predicted postoperative joint performance data at multiple second postoperative time points. Here, the second predicted postoperative joint performance data output is at least (i) A second predicted range of motion for at least one type of joint movement of the joint, (ii) at least one second predictive pain metric associated with the joint and It includes, Here, the at least one second machine learning model is trained to output data including a plurality of second values ​​for the second predicted postoperative joint performance data output at the plurality of second postoperative time points after surgery, where each second value corresponds to a specific second time point among the plurality of second postoperative time points after surgery. Here, the input data for training the at least one second machine learning model is at least, (i) The patient-specific data from before the surgery and (ii) The reconstruction plan data and It includes, Displaying the reconstruction plan data and the second predicted postoperative joint performance data output at the multiple second postoperative time points after surgery to the user on the display, In response to the user changing some of the parameters of the reconstruction plan input to the at least one second machine learning model before, during, or both of the arthroplasty, the second predicted postoperative joint performance data output determined from the at least one second machine learning model is updated. A device including a device.

2. The processor is configured to receive the preoperative patient-specific data by receiving the preoperative patient-specific data from at least one electronic medical resource via a communication network. The apparatus according to claim 1.

3. The at least one medical image comprises at least one of the following: (a) an X-ray image, (b) a computed tomography image, (c) a magnetic resonance image, (d) a three-dimensional (3D) image, (e) a 3D medical image generated from multiple X-ray images, (f) a video frame, or any combination thereof. The apparatus according to claim 1.

4. The at least one first predicted postoperative joint performance data at a plurality of first postoperative time points after surgery, and the at least one second predicted postoperative joint performance data at a plurality of second postoperative time points after surgery are predicted for at least one of (a) days, (b) months, and (c) years. The apparatus according to claim 1.

5. The processor is configured to display the second predicted postoperative joint performance data output along with recommendations for at least one arthroplasty parameter. The apparatus according to claim 1.

6. The aforementioned joint is selected from the group including the hip joint, knee joint, shoulder joint, elbow joint, and ankle joint. The apparatus according to claim 1.

7. The aforementioned joint is the shoulder joint. The apparatus according to claim 1.

8. The aforementioned preoperative patient-specific data includes (a) patient demographics, (b) patient diagnosis, (c) patient comorbidities, (d) patient medical history, (e) measurement of active range of motion of the shoulder, (f) patient self-reported measurements of pain, function, or both, (g) patient score based on the American Shoulder and Elbow Surgeons Shoulder Score (ASES), (h) patient score based on the Constant Shoulder Score (CSS), or any combination thereof. The apparatus according to claim 7.

9. The aforementioned at least one arthroplasty procedure is selected from a group including anatomical total shoulder arthroplasty, reverse total shoulder arthroplasty, deltoid approach, and superior lateral approach. The apparatus according to claim 7.

10. The at least one first predicted postoperative joint performance data at a plurality of first postoperative time points after surgery, and the at least one second predicted postoperative joint performance data at a plurality of second postoperative time points after surgery, are selected from a group including the United States Shoulder-Elbow (ASES) score, the University of California, Los Angeles (UCLA) Patient Reported Outcome Index score, constant score, comprehensive shoulder function score, visual analog scale (VAS) pain score, abduction score, anterior elevation score, and external rotation score. The apparatus according to claim 7.

11. The processor receives preoperative patient-specific data about arthroplasty performed on the patient's joints. Here, the preoperative patient-specific data is: (i) The patient’s medical history, (ii) The measured range of motion for at least one type of joint movement of the joint, and (iii) at least one pain metric related to the joint It has, The processor inputs the preoperative patient-specific data into at least one first machine learning model to determine the output of first predicted postoperative joint performance data at multiple postoperative time points. Here, the first predicted postoperative joint performance data output is at least (i) A first predicted range of motion for at least one type of joint movement of the joint, (ii) at least one first predictive pain metric related to the joint and It includes, Here, the at least one first machine learning model is trained to output data containing a plurality of first values ​​for the first predicted postoperative joint performance data output at a plurality of first postoperative time points after surgery, where each first value corresponds to each specific first time point at the plurality of first postoperative time points after surgery. The processor displays the first predicted postoperative joint performance data output to the user on a display as the first predicted postoperative joint performance data output. The processor receives at least one medical image of the joint obtained from at least one medical imaging procedure performed on the patient. The processor receives from the user at least one arthroplasty parameter based on the displayed first predicted postoperative joint performance data output. Here, the at least one arthroplasty parameter is (i) at least one implant, (ii) at least one implant size, (iii) at least one arthroplasty procedure, (iv) at least one location in the joint for embedding the at least one implant, or (v) Any combination of these Selected from, The processor generates a reconstruction plan for the patient's joint, at least partially based on the at least one medical image of the joint and the at least one arthroplasty parameter. The processor inputs the preoperative patient-specific data and reconstruction plan data into at least one second machine learning model to determine the output of second predicted postoperative joint performance data at multiple second postoperative time points. Here, the second predicted postoperative joint performance data output is at least, (i) A second predicted range of motion for at least one type of joint movement of the joint, (ii) at least one second predictive pain metric related to the joint and It includes, Here, the at least one second machine learning model is trained to output data containing a plurality of second values ​​for the second predicted postoperative joint performance data output at the plurality of second postoperative time points after surgery, where each second value corresponds to a specific second time point among the plurality of second postoperative time points after surgery. Here, the input data for training the at least one second machine learning model is at least (i) The patient-specific data from before the surgery, (ii) The reconstruction plan data and It includes, The processor displays the reconstruction plan data and the output of the second predicted postoperative joint performance data at the plurality of second postoperative time points to the user on the display, and The processor updates the second predicted postoperative joint performance data output determined from the at least one second machine learning model in response to the user changing some parameters of the reconstruction plan data input to the at least one second machine learning model before, during, or both of the arthroplasty. Methods that include...

12. Receiving the aforementioned preoperative patient-specific data includes receiving the aforementioned preoperative patient-specific data from at least one electronic medical resource via a communication network. The method according to claim 11.

13. The at least one medical image comprises at least one of the following: (a) an X-ray image, (b) a computed tomography image, (c) a magnetic resonance image, (d) a three-dimensional (3D) image, (e) a 3D medical image generated from multiple X-ray images, (f) a video frame, or any combination thereof. The method according to claim 11.

14. The at least one first predicted postoperative joint performance data at a plurality of first postoperative time points after surgery, and the at least one second predicted postoperative joint performance data at a plurality of second postoperative time points after surgery are predicted for at least one of (a) days, (b) months, and (c) years. The method according to claim 11.

15. Displaying the second predicted postoperative joint performance data output includes displaying the second predicted postoperative joint performance data output together with the recommendations for at least one arthroplasty parameter. The method according to claim 11.

16. The aforementioned joint is selected from the group having hip, knee, shoulder, elbow, and ankle joints. The method according to claim 11.

17. The aforementioned joint is the shoulder joint. The method according to claim 11.

18. The aforementioned preoperative patient-specific data includes (a) patient demographics, (b) patient diagnosis, (c) patient comorbidities, (d) patient medical history, (e) measurement of active range of motion of the shoulder, (f) patient self-reported measurements of pain, function, or both, (g) patient score based on the American Shoulder and Elbow Surgeons Shoulder Score (ASES), (h) patient score based on the Constant Shoulder Score (CSS), or any combination thereof. The method according to claim 17.

19. The aforementioned at least one arthroplasty procedure is selected from a group including anatomical total shoulder arthroplasty, reverse total shoulder arthroplasty, deltoid approach, and superior lateral approach. The method according to claim 17.

20. The at least one first predicted postoperative joint performance data at a plurality of first postoperative time points after surgery, and the at least one second predicted postoperative joint performance data at a plurality of second postoperative time points after surgery, are selected from a group including the United States Shoulder-Elbow (ASES) score, the University of California, Los Angeles (UCLA) Patient Reported Outcome Index score, constant score, comprehensive shoulder function score, visual analog scale (VAS) pain score, abduction score, anterior elevation score, and external rotation score. The method according to claim 17.

21. The processor is further configured to determine at least one joint-plastic surgery recommendation from the at least one second machine learning model and to display it to the user on the display. The apparatus according to claim 1.

22. The processor further comprises determining at least one joint-plastic surgery recommendation from the at least one second machine learning model and displaying it to the user on the display, The method according to claim 11.

Images