Data processing method, medium and equipment in knee arthroplasty

Abstract

The invention provides a data processing method, medium and equipment in knee arthroplasty, and belongs to the technical field of medical electronics. The method comprises the following steps: acquiring pressure distribution data, measured by an array sensor, of an articular compartment in a complete period of joint movement; based on the pressure distribution data, pressure centers under all buckling degrees in the whole period of joint movement are calculated, and corresponding pressure center tracks are formed; calculating shear stress distribution data under the complete period of joint movement based on the pressure center track data; based on the pressure distribution data, calculating a dynamic pressure envelope line under the complete period of the joint movement; and generating an intraoperative correction suggestion for knee arthroplasty based on the pressure center trajectory, the shear stress distribution data and the dynamic pressure envelope. The correction suggestions given by the method are more accurate for abnormal conditions occurring in knee replacement.

Description

Technical Field

[0001] The present application relates to the field of medical electronic technology, and in particular to a data processing method, medium and device for knee replacement surgery. Background Art

[0002] Knee replacement surgery, including total knee arthroplasty (TKA) and unicompartmental knee arthroplasty, is a core treatment for end-stage knee disease. Its core goal is to restore joint mechanical balance, alleviate pain, and restore function by implanting an artificial prosthesis. Accurate intraoperative assessment of the joint's biomechanical state is crucial to surgical success, directly impacting prosthesis stability, postoperative function, and longevity. However, existing intraoperative assessment techniques for knee replacement surgery still have numerous flaws, making it difficult to meet increasingly stringent clinical needs.

[0003] Taking unicompartmental knee replacement as an example, the traditional intraoperative data detection and evaluation process of unicompartmental knee replacement usually relies on mechanical measurement relying on gap measurement blocks of different thicknesses, or meniscus pads of different thicknesses, in conjunction with the doctor's passive knee movement to evaluate compartment pressure and joint stability. The pressure detection device used is usually a rigid sensor that can only provide static measurement, cannot conform to the joint surface, and interferes with the natural motion trajectory, resulting in the measured pressure data being inaccurate. It is also difficult to accurately measure or derive and calculate relevant mechanical indicators other than pressure to assist in intraoperative evaluation.

[0004] Regarding TKA intraoperative evaluation technology, it is currently mainly divided into two categories: traditional mechanical measurement evaluation and evaluation based on electronic pressure sensing technology. As for traditional mechanical measurement evaluation technology, it is similar to unicompartmental knee replacement, and it is also very dependent on tools such as gap measuring instruments and tensioners. It physically measures the gap parameters of the knee joint in static positions such as flexion and extension, and combines the doctor's touch to evaluate joint stability and soft tissue balance. However, it has the following significant limitations: 1) It is highly subjective. The evaluation of soft tissue balance relies on the doctor's tactile perception and lacks quantitative pressure data support. The judgment results of different doctors vary greatly, which can easily lead to evaluation deviations; 2) Static measurement limitations. It can only obtain gap values ​​in specific static positions (such as 0° and 90° flexion). It cannot capture the dynamic pressure changes of the knee joint from 0° to 120° flexion throughout the entire motion cycle, and it is difficult to reflect the mechanical characteristics under real motion conditions; 3) The detection range is limited. The contact area of ​​the measuring tool is usually less than 20 , cannot cover the edge area of ​​the prosthesis, and it is difficult to identify an area smaller than 10 Micro-scale stress concentration areas are formed, and such stress concentration is an important cause of early wear and loosening of the prosthesis.

[0005] Electronic pressure sensing technology can achieve quantitative measurement of the contact pressure between the femoral and tibial prostheses, and obtain pressure distribution data at various points between the femoral and tibial prostheses. However, in the existing intraoperative guidance process based on pressure distribution data, the pressure distribution center is usually only calculated based on the pressure distribution data, and correction suggestions such as osteotomy angle adjustment are evaluated and guided based on the pressure distribution center. However, this adjustment and correction suggestion has the defect of a single data reference dimension, and it is impossible to conduct a more comprehensive intraoperative evaluation, resulting in inaccurate intraoperative biomechanical evaluation, which can easily lead to postoperative complications (such as prosthesis loosening, accelerated wear, joint instability, etc.), reduce the patient's postoperative quality of life, and increase the risk of revision surgery.

[0006] Therefore, there is an urgent need for a multi-dimensional data processing system for knee replacement surgery to overcome the shortcomings of existing technologies and improve the accuracy and reliability of intraoperative data detection and evaluation of knee replacement surgery. Summary of the Invention

[0007] The object of the present invention is to provide a data processing method, medium and device in knee replacement surgery to solve at least one of the above technical problems.

[0008] In a first aspect, the present application provides a data processing method for knee replacement surgery, the method comprising: Obtaining pressure distribution data of the joint compartment during the full cycle of joint motion measured by an array sensor; Calculating the pressure center at each flexion degree in the full cycle of joint movement based on the pressure distribution data to form a corresponding pressure center trajectory; Calculating shear stress distribution data during the entire joint motion cycle based on the pressure center trajectory data; Calculating the dynamic pressure envelope of the joint during the full cycle of motion based on the pressure distribution data; Intraoperative correction recommendations for knee replacement surgery are generated based on the center of pressure trajectory, shear stress distribution data, and the dynamic pressure envelope.

[0009] Optionally, calculating the dynamic pressure envelope of the joint during the full cycle of motion based on the pressure distribution data includes: Calling a preset convolutional neural network model to extract spatial and temporal features from the pressure distribution data to obtain gradient distribution data of the medial-lateral pressure ratio and phase relationship data of the pressure change rate and the joint angular velocity; The dynamic pressure envelope is generated based on the gradient distribution data and the phase relationship data.

[0010] Optionally, the calling of a preset convolutional neural network model to extract spatial features and temporal features from the pressure distribution data to obtain gradient distribution data of the medial-lateral pressure ratio and phase relationship data of the pressure change rate and the joint angular velocity includes: Calculating the pressure ratio of the inner and outer sides of the knee joint at each time step based on the pressure distribution data to obtain inner and outer side pressure ratio data; Inputting the inside-outside pressure ratio data into the convolutional neural network model, capturing the spatial relationship in the inside-outside pressure ratio data through the convolution layer and the pooling layer in the convolutional neural network model, and obtaining the gradient distribution data; Calculating pressure change rate data and joint angular velocity data at each time step based on the pressure distribution data; The pressure change rate data and the joint angular velocity data are used as inputs of a network structure including a recurrent layer in the convolutional neural network model. The phase relationship between the pressure change rate data and the joint angular velocity data is captured through the recurrent layer, and the phase relationship data is output.

[0011] Optionally, the method further includes: Acquire a pressure distribution training data set, and preprocess the pressure distribution training data in the pressure distribution training data set to form a preprocessed data set that conforms to a standard normal distribution; Iteratively training the pre-trained model using the pre-processed data set, calculating a loss value based on a feature vector output from each iterative training, and adjusting parameters in the pre-trained model based on the calculated loss value until the number of iterations reaches a preset number threshold or the latest loss value is less than a preset loss threshold, ending the iterative training and outputting a trained convolutional neural network model; The process of each iterative training includes sequentially performing tensor processing, convolution processing, normalization processing, activation processing and pooling processing on the preprocessed data set.

[0012] Optionally, the pressure distribution data includes pressure values ​​measured by each sensing unit in the array sensor at each moment; and calculating the pressure center at each flexion degree in the full cycle of joint motion based on the pressure distribution data to form a corresponding pressure center trajectory includes: The pressure center at the corresponding moment is calculated based on the pressure value of each sensing unit at the same moment and the position coordinates of the corresponding sensing unit; and the pressure center trajectory is formed based on the pressure center at each moment.

[0013] Optionally, calculating the shear stress distribution data during the full cycle of joint motion based on the pressure center trajectory data includes: Determining the height of the point of action of the friction force according to the elevation of the surface friction layer of the array sensor; Calculating the moving distance in the tangential direction based on the pressure center trajectory; The shear stress at the corresponding moment is calculated based on the height of the action point, the movement distance and the pressure center at the corresponding moment.

[0014] Optionally, generating intraoperative correction suggestions for knee replacement surgery based on the pressure center trajectory, shear stress distribution data, and the dynamic pressure envelope includes: Establishing a transfer function between the osteotomy angle adjustment amount and the pressure distribution change amount, wherein the pressure distribution change amount is calculated based on the offset of the pressure center trajectory and the gradient change of the shear stress distribution data; An optimization suggestion vector for osteotomy angle adjustment is generated according to the transfer function and the normal range of the dynamic pressure envelope.

[0015] Optionally, establishing a transfer function between the osteotomy angle adjustment amount and the pressure distribution change amount includes: training the intelligent agent through reinforcement learning, using the osteotomy angle adjustment amount as the action input, using the pressure distribution change amount and the feedback signal obtained after executing the action as the reward feedback, iteratively optimizing to obtain a mapping relationship, and converting the mapping relationship into the transfer function.

[0016] In a second aspect of the present application, a computer-readable storage medium is provided, on which executable instructions are stored. When the executable instructions are executed by a processor, the processor executes the method as described in any embodiment of the present application.

[0017] In a third aspect of the present application, an electronic device is provided, comprising: one or more processors; and a memory for storing one or more programs, which, when executed by the one or more processors, enables the one or more processors to execute the method described in any one of the embodiments of the present application.

[0018] The data processing method, medium and equipment in knee replacement surgery in this application cover the edge of the prosthesis and the entire motion cycle through array sensors, so that the information of the obtained pressure distribution data is richer, and by obtaining the pressure center trajectory and shear stress based on the pressure distribution data, the doctor's subjective feel is converted into digital data (such as pressure center offset, shear stress value); and the dynamic pressure envelope is further calculated based on the pressure distribution data, providing an objective judgment standard for intraoperative evaluation, avoiding misjudgment caused by reliance on "empirical thresholds"; through the fusion analysis of multi-dimensional data such as the pressure center trajectory, shear stress, and dynamic pressure envelope, the correction suggestions given for abnormal conditions that occur during knee replacement surgeries such as total knee replacement and unicompartmental knee replacement are more accurate and comprehensive. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope of the present application.

[0020] Figure 1 is a flow chart of a data processing method in knee replacement surgery according to one embodiment; Figure 2 is a schematic structural diagram of an array sensor in one embodiment; Figure 3 A schematic structural diagram of an array sensor in another embodiment; Figure 4 A pressure distribution cloud diagram in one embodiment; Figure 5 A schematic diagram of a process for calling a preset convolutional neural network model to extract spatial and temporal features from pressure distribution data in one embodiment to obtain gradient distribution data of the medial-lateral pressure ratio and phase relationship data of the pressure change rate and the joint angular velocity; Figure 6 FIG. 1 is a schematic structural diagram of an electronic device in an embodiment. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solutions and advantages of this application more clearly understood, the present application is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0022] All terms (including technical and scientific terms) used in this application have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0023] For example, the terms "first" and "second" used in this application are only used to distinguish similar objects and to differentiate the first object from another object, rather than to describe a specific order or sequence, and cannot be understood as indicating or implying relative importance.

[0024] This application proposes a data processing method for knee replacement surgery, such as Figure 1 As shown, the method includes: Step 110 , obtaining pressure distribution data of the joint compartment during the full cycle of joint motion measured by the array sensor.

[0025] In this embodiment, the knee replacement surgery may include total knee replacement surgery and unicompartmental knee replacement surgery. The array sensor can be placed between the tibia after osteotomy during surgery for pressure measurement. The shape of the array sensor matches the shape of the knee joint and can cover the load-bearing surface of the tibial prosthesis. Specifically, the array sensor can cover the load-bearing surface of the tibial prosthesis and its edge extension area, such as covering the 5mm extension area of ​​its edge, can fit the surface of the tibial prosthesis, and adaptively deform with the prosthesis shape when the knee joint flexes and extends to ensure measurement stability. The array sensor can be a flexible composite sensor array, such as an array capacitive sensor, which realizes distributed measurement of inter-articular pressure through pressure-electrical signal conversion.

[0026] The joint compartment refers to the space between the femur and tibia, and between the femur and patella in the knee joint, including the medial compartment (between the medial femoral condyle and tibial plateau of the knee joint) and the lateral compartment (between the lateral femoral condyle and tibial plateau of the knee joint). The joint compartment is the core area of ​​pressure transmission, and its pressure distribution directly reflects the mechanical equilibrium state of the knee joint. Pressure distribution data refers to the set of pressure values ​​measured by each sensing unit in the array sensor at different times. The pressure distribution data can be presented in the form of a two-dimensional matrix or a one-dimensional vector, the elements of which reflect the pressure size and distribution characteristics of the corresponding position of the joint compartment (spatial resolution <10 micro-scale stress concentration area).

[0027] The full cycle of joint motion refers to the continuous movement process of the knee joint from full extension (such as 0°) to maximum flexion (such as 120°), covering the flexion angles commonly used in daily activities.

[0028] During the intraoperative evaluation phase of a knee replacement, the sensor array is attached to the load-bearing surface of the tibial prosthesis, ensuring complete coverage of the joint contact area. As the knee joint undergoes continuous flexion from 0° to 120°, driven by a mechanical assist device, the sensor array collects pressure values ​​from each sensor unit in real time at a preset acquisition frequency (e.g., 100 Hz or higher), generating pressure distribution data at each moment.

[0029] like Figure 2 As shown, it shows a form of an array sensor, which matches the surface of the tibial prosthesis in total knee replacement surgery. The array sensor can be specifically an array capacitive sensor, covering the tibial prosthesis load-bearing surface and a 5mm extension area at the edge, on which multiple sensing units 201 ( Figure 2 (Only the left half of the sensing unit is shown in the figure). The sensing unit can be a pressure sensing unit that can measure the pressure at the corresponding position. The pressure at each position can be used to calculate the corresponding pressure center 202, and combined with the time change, the corresponding pressure center trajectory 203 is formed.

[0030] like Figure 3 As shown, another form of the array sensor is shown. The form of the array sensor matches the surface of the tibial prosthesis in unicompartmental knee replacement. The array sensor is also an array capacitive sensor, covering the tibial prosthesis load-bearing surface and the 5mm extension area at the edge, on which multiple sensing units 301 ( Figure 3 (Only some of the sensing units are shown in the figure). The sensing units can be pressure sensing units that can measure the pressure at corresponding locations. The pressure at each location can be used to calculate the corresponding pressure center 302, and combined with the changes over time, a corresponding pressure center trajectory 303 is formed.

[0031] Pressure distribution data can be reflected through pressure cloud diagrams, such as Figure 4 As shown, it shows the distribution of pressure of the array sensor at different positions. The closer the color is to red, the greater the pressure; the closer the color is to blue, the smaller the pressure.

[0032] Step 120 , calculating the pressure center at each flexion degree in the entire joint motion cycle based on the pressure distribution data, and forming a corresponding pressure center trajectory.

[0033] The pressure center refers to the "point of net force application" of pressure distribution within the joint compartment. It is the point of moment equilibrium within the pressure field, analogous to the center of mass of an object. The pressure center trajectory is the continuous path of the pressure center as it changes with flexion angle throughout the joint's full motion cycle (0° to 120° of flexion), reflecting the shifting trend of the center of gravity in dynamic pressure distribution. The pressure center can be calculated based on the pressure values ​​of each sensor unit and its location.

[0034] Specifically, the pressure distribution data includes the pressure values ​​measured by each sensing unit in the array sensor at each moment. Step 120 includes: calculating the pressure center at the corresponding moment based on the pressure values ​​of each sensing unit at the same moment and the position coordinates of the corresponding sensing unit; and forming a pressure center trajectory based on the pressure center at each moment.

[0035] The electronic device can establish a corresponding spatial coordinate system based on the pressure sensor or joint, map each sensing unit in the pressure sensor to the spatial coordinate system, and obtain the coordinate position of the center point of each sensing unit in the spatial coordinate system. For example, the sensing units in the pressure sensor are arranged in an m×n rectangular shape, and P is denoted as ij represents the pressure value measured by the sensor unit in the i-th row and j-th column; (x ij ,y ij ) represents the coordinate position of the center point of the sensing unit in the coordinate system; the corresponding pressure center coordinate on the two-dimensional pressure field is marked as (X CoP , Y CoP ).in, , .

[0036] The pressure center coordinates corresponding to all flexion angles in the full cycle are connected in angular order to form the pressure center trajectory.

[0037] Through real-time pressure center tracking during surgery, the optimal prosthesis rotation angle of the tibial plateau is determined, and the prosthesis position is dynamically adjusted to make the postoperative pressure center trajectory close to the natural knee joint characteristics. At the same time, it can also avoid the errors of traditional reliance on anatomical landmarks (such as avoiding abnormal patellar trajectory caused by excessive internal rotation of the tibial prosthesis). By digitally displaying the balance effect, the subjectivity of traditional "tactile assessment" is reduced.

[0038] Step 130 : Calculate the shear stress distribution data during the entire joint motion cycle based on the pressure center trajectory data.

[0039] Shear stress distribution data refers to the tangential stress distribution data parallel to the joint contact surface. Shear stress reflects the frictional mechanical characteristics of the knee joint during shear, sliding, or torsional motion, and its unit is MPa. In actual contact, in addition to normal pressure, there is also tangential friction (shear stress), which is a manifestation of external shear, sliding, or torsional forces. Because sensors cannot generally directly sense tangential forces, we indirectly infer the presence and direction of shear stress through the dynamic changes in the center of pressure, thereby generating shear stress distribution data.

[0040] Based on the pressure center trajectory, the offset of the pressure center at different flexion angles (Δx, Δy, i.e., the difference between the current coordinates and the baseline coordinates) is analyzed, and the shear stress is calculated in combination with the moment balance principle and / or Hooke's law to form the shear stress distribution data.

[0041] In one embodiment, step 130 includes: determining the height of the point of action of the friction force based on the elevation of the surface friction layer of the array sensor; calculating the movement distance in the tangential direction based on the pressure center trajectory; and calculating the shear stress at the corresponding moment based on the height of the point of action, the movement distance, and the pressure center at the corresponding moment.

[0042] The surface friction layer elevation refers to the thickness of the friction layer in the sensor array that directly contacts the articular cartilage or prosthetic surface. This friction layer is typically made of medical-grade silicone or polyurethane. Its elevation determines the location of the friction force's point of application. The surface friction layer elevation parameter is typically provided by the sensor manufacturer and stored in the device database, where it can be retrieved. The friction force's point of application height refers to the vertical height between the joint contact surface and the sensor's friction layer, where the friction force actually applies.

[0043] Tangential movement distance It refers to the displacement of the pressure center in the tangential direction (i.e., the direction parallel to the joint contact surface) between two adjacent moments (or flexion angles), reflecting the dynamic changes of the joint during sliding or torsion. The coordinates of the pressure center at two adjacent moments (or flexion angles) are recorded as (X CoP1 , Y CoP1 ) and (X CoP2 , Y CoP2 ), then the tangential movement distance of the pressure center at two adjacent moments (or flexion angles) is It can be expressed as the following formula 1: Formula 1 shear stress It can be calculated according to the following formula 2.

[0044] Formula 2 This formula expresses that: when the contact state is constant and the friction model is linear, the shear stress is proportional to the distance the pressure center moves. The greater the shear stress, the easier it is to generate a large tangential friction force, but the shear stress corresponding to the unit distance the pressure center moves also increases. Where h represents the height of the equivalent friction point (approximately the thickness of the material contact layer or the height of the friction layer on the sensor surface). Indicates the displacement of the pressure center in the tangential direction, which is approximately equal to the moving distance , is the total normal pressure.

[0045] Step 140 : Calculate the dynamic pressure envelope during the entire joint motion cycle based on the pressure distribution data.

[0046] The dynamic pressure envelope (DPE) is the boundary of the normal pressure range, generated by training based on pressure distribution data from a large number of normal knee replacement surgeries. Specifically, the DPE is expressed as the mean (μ) ± 2 standard deviations (2σ), covering the full range of pressure distribution from 0° to 120° of flexion.

[0047] The envelope is used to determine whether intraoperative pressure distribution is normal. The upper limit (μ + 2σ) and lower limit (μ - 2σ) correspond to the maximum and minimum normal pressure values, respectively. For example, the dynamic pressure envelope of the medial compartment at 0° of knee flexion is 3.0-5.0 MPa. If the intraoperative value is 5.8 MPa, it exceeds the upper limit and indicates abnormality.

[0048] Specifically, spatial and temporal features can be extracted from the acquired pressure distribution data, and based on these features, a dynamic pressure envelope can be generated for the entire joint transport cycle. The spatial feature can be the gradient distribution of the medial-lateral pressure ratio, and the temporal feature can be the phase relationship between the pressure change rate and the joint angular velocity.

[0049] Step 150 : Generate intraoperative correction suggestions for knee replacement surgery based on the center of pressure trajectory, shear stress distribution data, and dynamic pressure envelope.

[0050] In this embodiment, after obtaining the pressure center trajectory, shear stress distribution data and dynamic pressure envelope, it is possible to determine whether the current knee replacement surgery is normal based on these data. When an abnormality occurs, corresponding correction suggestions are generated.

[0051] Specifically, the system can detect whether the pressure center trajectory deviation exceeds the upper or lower limit of the dynamic pressure inclusion line. If so, it indicates an abnormality and provides corresponding correction suggestions based on the specific situation. It also detects whether the shear stress exceeds the corresponding shear stress threshold. If so, it also provides corresponding correction suggestions to reduce the shear stress.

[0052] For example, if the pressure center trajectory deviates by more than 1.5mm (such as a medial deviation of 2.0mm) and the medial pressure exceeds the upper limit of the envelope (6.2MPa>5.0MPa), it may indicate that the medial collateral ligament is too tight, and it is recommended to loosen the medial ligament by 0.5mm; if the shear stress is 7MPa at 90° of knee flexion (the corresponding shear stress threshold is 5MPa), and the pressure center deviates posteriorly, it indicates that the posterior joint capsule is tight, and it is recommended to adjust the posterior tilt angle of the tibial prosthesis by 1°.

[0053] In one embodiment, when there is a shear stress area greater than a shear stress threshold in the shear stress distribution data, a correction suggestion is generated in combination with the coordinate offset corresponding to the pressure center trajectory in the area. The correction suggestion can specifically be a suggestion for adjusting the prosthesis rotation angle, where the shear stress threshold is determined based on the shear stress safety range of the dynamic pressure envelope.

[0054] The system monitors shear stress distribution data throughout the entire joint motion cycle in real time, determining whether there are areas with shear stress greater than a threshold (e.g., 5 MPa) through point-by-point comparison. It locates the corresponding coordinates of the area exceeding the shear stress threshold in the pressure center trajectory, and calculates the difference (offset) between these coordinates and the normal pressure center coordinates in the dynamic pressure envelope. Based on the location and offset of the area exceeding the shear stress threshold, combined with a preset "shear stress-rotation angle" mapping relationship (e.g., a transfer function), it generates corresponding correction suggestions.

[0055] For example, this mapping relationship indicates that for every 1 MPa increase in the medial shear stress threshold and a 1 mm medial shift in the center of pressure, the femoral prosthesis needs to be externally rotated 1°. If the current situation is that the shear stress exceeds the shear stress threshold of 0.8 MPa and the offset is 2.0 mm, the generated correction suggestion is: it is recommended to externally rotate the femoral prosthesis 3°.

[0056] Correction suggestions can be presented in the form of one or more combinations of three-dimensional cloud map highlights, voice prompts, correction suggestion pop-ups, etc., to achieve early warning prompts during surgery. For example, when it is detected that the local pressure of the relevant part exceeds the preset pressure threshold (for example, 2.5Mpa) and lasts for more than a preset time (for example, 2 seconds), a three-dimensional cloud map is generated. In this cloud map, the part where the local pressure exceeds 2.5Mpa and lasts for more than 2 seconds is locally highlighted; in addition, when it is detected that the pressure difference between the inside and outside exceeds the preset percentage threshold (for example, 35%), and the trajectory offset exceeds the preset offset threshold (for example, 1.5mm), the feedback form can be voice prompts + correction suggestion pop-ups.

[0057] In one embodiment, the pressure distribution data is compared with the upper and lower limits of the dynamic pressure envelope in real time. When the pressure distribution data exceeds the upper and lower limits, the offset direction and offset of the pressure center trajectory at the corresponding timestamp are extracted, and combined with the directional characteristics of the shear stress distribution data, corresponding correction suggestions are generated. The correction suggestions can specifically be suggestions for soft tissue release or gasket thickness adjustment.

[0058] The upper and lower limits (upper and lower limits) of the dynamic pressure envelope represent the boundaries of the normal pressure range of the dynamic pressure envelope and constitute the "normal range" of the pressure distribution. The offset direction refers to the deviation of the pressure center from its normal position (the center of the envelope) (e.g., inward, outward, forward, backward); the offset refers to the distance of the deviation. The directional characteristics of the shear stress distribution data refer to the primary direction of shear stress (e.g., forward, backward, inward, outward), which is determined by the direction of movement of the pressure center trajectory.

[0059] Soft tissue release recommendations refer to intraoperative release plans for soft tissue tension (such as ligament contracture), including the release site (such as medial collateral ligament, posterior joint capsule) and the degree of release (such as release length).

[0060] The recommendation for gasket thickness adjustment refers to a solution to adjust the size of the joint gap by replacing polyethylene gaskets of different thicknesses to balance the pressure distribution.

[0061] Specifically, the pressure distribution data is compared with the upper and lower limits of the dynamic pressure envelope, marking areas outside the range. The pressure center coordinates corresponding to these areas are located, and their deviation from the normal coordinates (the center of the envelope) is calculated. The shear stress direction corresponding to these areas is analyzed, and the sliding trend is determined based on the pressure center offset direction. Targeted correction recommendations are generated based on the offset, shear stress direction, and the magnitude of the pressure exceeding the upper limit.

[0062] For example, if the pressure exceeds the upper limit by ≤1.0MPa and the offset is ≤2mm: it is recommended to loosen the tense soft tissue (such as the lateral collateral ligament). For example, when the offset is 2.5mm, it is recommended to loosen it by 0.5mm. If the pressure exceeds the upper limit by >1.0MPa and the offset is >2mm: it is recommended to adjust the gasket thickness at the same time, for example, replace the original 2mm gasket with 2.5mm, increase the lateral gap, and reduce the pressure.

[0063] The data processing method for knee replacement surgery in this application covers the edge of the prosthesis and the entire motion cycle through an array sensor, so that the information of the obtained pressure distribution data is richer, and by obtaining the pressure center trajectory and shear stress based on the pressure distribution data, the doctor's subjective feel is converted into digital data (such as pressure center offset, shear stress value); and the dynamic pressure envelope is further calculated based on the pressure distribution data, providing an objective judgment standard for intraoperative evaluation, avoiding misjudgment caused by relying on "empirical thresholds"; through the fusion analysis of multi-dimensional data such as the pressure center trajectory, shear stress, and dynamic pressure envelope, the accuracy of identifying abnormal conditions that occur during total knee replacement surgery such as unicompartmental knee replacement and the rationality of the correction suggestions given for abnormal conditions can be improved.

[0064] In one embodiment, step 140 includes: calling a preset convolutional neural network model to extract spatial features and temporal features from the pressure distribution data, obtaining gradient distribution data of the inner and outer pressure ratio and phase relationship data of the pressure change rate and the joint angular velocity; generating a dynamic pressure envelope based on the gradient distribution data and the phase relationship data.

[0065] In this embodiment, the convolutional neural network model includes multi-layer structures such as convolution layers and pooling layers, through which spatial and temporal features of the pressure distribution data are extracted. The spatial features represent the characteristics of the differences and changing trends in the spatial dimension of the pressure distribution, focusing on the uniformity of pressure distribution in the anatomical area of ​​the knee joint; the temporal features reflect the dynamic characteristics of the pressure distribution over time (or joint movement angle), focusing on the synchronization of pressure changes and joint movements. By extracting spatial features, the gradient distribution data of the medial and lateral pressure ratio can be obtained; by extracting temporal features, the phase relationship data of the pressure change rate and the joint angular velocity can be obtained.

[0066] The gradient distribution of the medial-lateral pressure ratio refers to the spatial rate of change of the ratio of medial to lateral knee pressure (ML Ratio), reflecting the transitional trend of pressure from the medial to the lateral side. The gradient of the medial-lateral pressure ratio (ML Ratio = medial mean pressure / lateral mean pressure) represents the change in ML Ratio per unit distance. A larger absolute value of the gradient indicates a more significant difference in medial-lateral pressure. The phase relationship between the pressure change rate and joint angular velocity refers to the time difference (or phase difference) between the pressure change rate (dP / dt) and the joint flexion angular velocity (dθ / dt), reflecting their synchronization. By calculating the angular difference (or phase difference) between the peak pressure and the peak angular velocity, we can determine whether the pressure changes match the movement rhythm. A normal phase difference is usually within ±15°; an excess of this phase difference indicates dynamic imbalance.

[0067] Specifically, the pressure distribution data acquired during the entire joint motion cycle (0°-120° flexion) can first be preprocessed. This preprocessing process includes filling in missing values, removing outliers, converting data formats, and normalizing data. After the data format conversion, the pressure distribution data can be converted into a tensor format, including dimensions such as the number of samples, time steps, and the number of sensors (or the dimensions of the combination of pressure and motion information). The time step corresponds to the flexion angle (e.g., 1200 steps, 0.1° per step); the feature dimensions include the pressure value of each sensor unit (100 units) and the joint angular velocity at the corresponding moment (extracted from the motion trajectory data). For example, for the first sample, its time step is 1s, the length of the sensor value recorded in this 1s is n, and the coordinates of the pressure center are (x, y). Then the dimensions of the corresponding preprocessed tensor format are [1, 1, n+2].

[0068] During the data standardization process, the pressure value and angular velocity are standardized and converted into standard normal distribution data with a mean of 0 and a standard deviation of 1. The following formula 3 can be used for processing: Formula 3 Where μ and σ are the mean and standard deviation of the data, respectively. x is the original data, and x' is the normalized data. For example, the normalized pressure value is in the range [-1, 1], and the normalized angular velocity value is in the range [-0.5, 0.5].

[0069] The convolutional neural network model can specifically be a 16-layer network structure, wherein the first layer is the input layer, the second to 12 layers are a convolutional layer group, which includes 4 convolutional layers, 3 activation layers, and 4 batch normalization layers; the 13th and 14th layers are a pooling layer group, which includes two pooling layers; the 15th layer is a fully connected layer, and the 16th layer is the output layer. It is understandable that the convolutional neural network model can also be any other suitable network structure.

[0070] During spatial feature extraction, the mean medial pressure (denoted as Pinside) and mean lateral pressure (denoted as Poutside) at each time step are calculated based on the distribution of the sensing units in the array sensor on the inside and outside of the knee joint (for example, the medial compartment corresponds to sensing units 1-50, and the lateral compartment corresponds to sensing units 51-100). Based on these mean medial and lateral pressures, the corresponding medial-to-lateral pressure ratio (ML Ratio = Pinside / Poutside) can be calculated. The convolutional layer of the convolutional neural network model is then used to extract the spatial rate of change of the medial-to-lateral pressure ratio (i.e., the change in ML Ratio per millimeter from the inside to the outside). This gradient distribution data for the corresponding medial-to-lateral pressure ratio can be obtained.

[0071] During temporal feature extraction, the pressure distribution data is time-differentiated to obtain the pressure change rate (MPa / s) at each time step. Based on the joint motion trajectory, the flexion angle change rate at each time step is calculated. Using a convolutional neural network model, the peak time difference between the pressure change rate and the joint angular velocity is calculated to determine the phase relationship between the two.

[0072] After extracting spatial and temporal features, the gradient distribution data and phase relationship data are integrated by time step (flexion angle) to form a comprehensive feature representation. For example, the two feature vectors can be concatenated or combined in other appropriate ways. Based on the integrated features, the dynamic pressure envelope is calculated using statistical methods. The upper and lower limits of the pressure distribution at each angle can be determined based on the mean μ and standard deviation σ under normal conditions. For example, the upper limit is μ + 2σ and the lower limit is μ - 2σ.

[0073] The generated dynamic pressure envelope is presented in a visual form (such as the colored area on the intraoperative display screen, with green indicating the normal range and red indicating abnormality), and the pressure distribution data of the current surgery is compared in real time to see if it is within the range.

[0074] In one embodiment, Figure 5 As shown, the preset convolutional neural network model is called to extract spatial and temporal features from the pressure distribution data, and the gradient distribution data of the medial and lateral pressure ratio and the phase relationship data of the pressure change rate and the joint angular velocity are obtained, including: Step 510 , calculating the pressure ratio between the inner and outer sides of the knee joint at each time step based on the pressure distribution data, and obtaining the inner and outer side pressure ratio data.

[0075] For example, if a full joint motion cycle (0°-120° flexion) is divided into 1200 time steps at a 100Hz sampling frequency, each step corresponds to 0.01 seconds and 0.1° of flexion angle. Each step corresponds to a pressure distribution data set, such as the pressure values ​​of 10×10 sensor units. Medial and lateral region division: Based on the anatomical positioning of the sensor array, for example, the medial compartment can be pre-assigned to sensor units 1-50 (covering the medial tibial plateau), and the lateral compartment can be pre-assigned to sensor units 51-100 (covering the lateral tibial plateau).

[0076] For each time step, the mean inner pressure (Pin) and the mean outer pressure (Pout) are calculated, and the corresponding inner-outer pressure ratio can be calculated based on the mean inner pressure and the mean outer pressure.

[0077] In step 520, the inside-outside pressure ratio data is input into the convolutional neural network model, and the spatial relationship in the inside-outside pressure ratio data is captured by the convolution layer and pooling layer in the convolutional neural network model to obtain gradient distribution data.

[0078] First, the inner and outer pressure ratio data are converted into the input format of the convolutional neural network model, and the gradient distribution data is obtained through the processes of capturing local spatial features through the convolution layer, extracting key features through the pooling layer, and generating gradient distribution data.

[0079] Specifically, the inside-outside pressure ratio data can be converted into a 1200×1 time series matrix, where the rows correspond to the time steps and the columns correspond to the pressure ratios, and expanded into a two-dimensional feature map (1200×1×1) to adapt to the convolution operation.

[0080] To capture local spatial features, the convolutional layers use 16 3×1 convolution kernels (covering three consecutive time steps). Sliding calculations are performed on the pressure ratio data to capture local variation patterns (e.g., the increasing trend of the ratio from 1.18 to 1.19 to 1.20 for steps t1-t3). Activation and normalization: Nonlinearity is introduced through a Relu activation layer, and a batch normalization layer stabilizes training. The output is 16 feature maps (each 1198×1 due to edge truncation). The second convolutional layer uses 32 5×1 convolution kernels (covering five time steps) to further capture more complex spatial relationships (e.g., the gradient of the ratio over five consecutive steps).

[0081] For the pooling layer to extract key features, the first pooling layer uses 2×1 max pooling with a stride of 2. This downsamples the feature maps output by the convolutional layer, retaining the maximum value in each 2-step window (for example, retaining the peak of the ratio change), and outputs 32 feature maps (599×1). The second pooling layer uses 2×1 max pooling with a stride of 2 to further simplify the data and output 32 feature maps (300×1).

[0082] For gradient distribution data generation, the outputs of the convolution layer and the pooling layer are integrated through the fully connected layer to generate the gradient distribution data of the inside-outside pressure ratio, that is, the rate of change of the pressure ratio from the inside to the outside at each time step.

[0083] Step 530 : Calculate the pressure change rate data and joint angular velocity data at each time step based on the pressure distribution data.

[0084] Step 540: Use the pressure change rate data and the joint angular velocity data as inputs to a network structure including a recurrent layer in a convolutional neural network model, capture the phase relationship between the pressure change rate data and the joint angular velocity data through the recurrent layer, and output the phase relationship data.

[0085] This recurrent layer can use an LSTM (Long Short-Term Memory) structure with 64 memory cells to capture long-term dependencies in time series data (e.g., the relationship between pressure change rate and angular velocity). The pressure change rate data (e.g., a 1200×1 time series matrix) and the joint angular velocity data (e.g., a 1200×1 time series matrix) are aligned by time step to form a two-dimensional input matrix (e.g., a 1200×2 time series matrix) and input into the LSTM layer.

[0086] The LSTM uses memory cells to store historical information (such as the angular velocity at step t-10) and compares the synchronization of the current pressure change rate with the historical angular velocity. It identifies the time step corresponding to the peak pressure change rate (t_p) and the time step corresponding to the peak angular velocity (t_θ), and calculates the phase difference φ = (t_p - t_θ) × 0.1° (the angle corresponding to the step).

[0087] The circulation layer outputs a full-cycle phase difference array, i.e., phase relationship data, which reflects the pressure-motion synchronization of each key motion stage (such as accelerated buckling and decelerated buckling).

[0088] The execution order of the above steps 510 to 520 and 530 to 540 is not limited. For example, steps 510 to 520 and steps 530 to 540 can be executed in parallel.

[0089] In one embodiment, the above method also includes a process for model training of a convolutional neural network, which process includes: obtaining a pressure distribution training data set, preprocessing the pressure distribution training data in the pressure distribution training data set to form a preprocessed data set that conforms to the standard normal distribution; iteratively training the pretrained model with the preprocessed data set, calculating the loss value based on the feature vector output of each iterative training, and adjusting the parameters in the pretrained model based on the calculated loss value until the number of iterations reaches a preset number threshold or the latest loss value is less than the preset loss threshold, ending the iterative training and outputting the trained convolutional neural network model; wherein, each iterative training process includes sequentially performing tensor processing, convolution processing, normalization processing, activation processing and pooling processing on the preprocessed data set.

[0090] In this embodiment, the pressure distribution training data set includes multiple pressure distribution training data, and the pressure distribution training data may be pressure distribution collected by using a knee joint-shaped array sensor during historical knee replacement surgery.

[0091] The preprocessing process primarily involves data cleaning and standardization. Data cleaning involves checking for missing values ​​and outliers. Missing values ​​can be interpolated using pressure values ​​at adjacent time points or at similar motion angles. Outliers can be identified and removed using the statistical 3σ principle. The 3σ range encompasses nearly all (99.73%) of the data. Therefore, data outside the [μ-3σ, μ+3σ] interval in the pressure distribution training dataset are considered extreme values ​​or outliers and are removed.

[0092] The pressure data and motion trajectory data are standardized and converted into standard normal distribution data with a mean of 0 and a standard deviation of 1. This helps improve the efficiency and stability of subsequent model training.

[0093] The first layer in the convolutional neural network model is the input layer. In the input layer, the training data containing pressure values ​​and motion trajectory information is organized into tensor form based on the characteristics of the preprocessed training data.

[0094] Layers 2 through 12 of the convolutional neural network model form a convolutional layer group, specifically consisting of four convolutional layers, three activation layers, and four batch normalization layers. Within this convolutional layer group, the convolution kernel slides over the input data, performing element-wise multiplication and summation operations to automatically extract local features from the data, such as specific patterns in pressure distribution and local trends in motion trajectories. This reduces the workload of manual feature extraction while better capturing the inherent structure of the data.

[0095] Convolutional layers are distributed across the 2nd, 5th, 8th, and 11th layers. In the 2nd layer, a 3×3 convolution kernel with 16 kernels is used to convolve the input data to extract local features. A batch normalization layer is then added to normalize the convolution output. In the 5th layer, the number of kernels is adjusted to 32, while the kernel size remains at 3×3. Convolution continues to extract more complex features, and a batch normalization layer is also added. In the 8th layer, the number of kernels is increased to 64, with a kernel size of 5×5, to further extract deeper features. A batch normalization layer follows. In the 11th layer, the number of kernels is increased to 128, with a 5×5 kernel size, to deepen feature extraction. Finally, a batch normalization layer is added to accelerate training convergence and mitigate gradient issues. Finally, the feature tensor after convolution and batch normalization is output.

[0096] The activation layer group is immediately after each batch normalization layer. The activation function used by the activation layer group can be one or more of ReLU, Sigmoid, Tanh, etc.

[0097] The 13th and 14th layers of the convolutional neural network model are pooling layers. In the 13th layer, a 2×2 max pooling window with a stride of 2 is used to downsample the activation output data of the 13th layer. In the 14th layer, a 2×2 max pooling window with a stride of 2 is also used to downsample the pooled data from the 13th layer, reducing the data dimension and computational complexity while retaining the key features.

[0098] The 15th layer of the convolutional neural network model is a fully connected layer. The number of neurons is set according to the task requirements and model complexity. The features extracted by the previous convolution and pooling (or recurrent layer) are integrated to form a feature vector.

[0099] The 15th layer of the convolutional neural network model is the output layer. The output layer structure is designed according to the feature extraction target, and the feature vector results related to the corresponding final task are output.

[0100] After building the convolutional neural network model's architecture, initialize the network parameters. This initialization process can use Xavier initialization or Kaiming initialization to assign appropriate initial values ​​to parameters such as the weights of the convolutional layers and the fully connected layers. This prevents problems like vanishing or exploding gradients during training, ensuring proper training and convergence.

[0101] During data training, the pressure distribution training data set is divided into a training set, a validation set, and a test set. This is generally done in a certain ratio, such as a 7:1:2 ratio. 70% of the pressure distribution training data is used to train the model, 10% is used to verify the model's performance and adjust hyperparameters during training, and 20% is used to ultimately evaluate the model's generalization ability and accuracy.

[0102] For the loss function, the Mean Squared Error (MSE) loss function can be used to calculate the error between the eigenvalues ​​predicted by the model and the actual eigenvalues, and optimize the model parameters by minimizing this error. The calculation formula of the loss function MES is as follows: Formula 4 in, is the sample size, is the true eigenvalue, The feature values ​​predicted by the model.

[0103] The relevant parameters in the convolutional neural network model can be updated using optimizers such as Stochastic Gradient Descent (SGD) and the Adam optimizer. These optimizers automatically adjust the learning rate during training, accelerating model convergence. For this model, using the Adam optimizer as an example, the initial learning rate is set to 0.001. As training progresses, the learning rate can be adjusted based on performance changes on the validation set.

[0104] During model training, the divided training set data is fed into the constructed convolutional neural network model and trained in batches. Within each training batch, the model's output is computed using a forward propagation algorithm. The loss is calculated using the defined loss function. Backpropagation is then used to calculate the gradients, and the network parameters are updated using an optimizer. During training, the model's performance is regularly evaluated using the validation set data, and the validation set loss and related metrics (such as feature extraction accuracy) are recorded. Based on performance changes on the validation set, network hyperparameters such as the learning rate, number of convolution kernels, and number of network layers are adjusted to prevent overfitting or underfitting. The model continues training until performance on the validation set reaches a stable state or a preset stopping criterion is met. This stopping criterion can be one or more of the following: reaching a certain number of training epochs, or the validation set loss no longer significantly decreasing.

[0105] In one embodiment, step 150 includes: establishing a transfer function between the osteotomy angle adjustment amount and the pressure distribution change amount, where the pressure distribution change amount is calculated based on the offset of the pressure center trajectory and the gradient change of the shear stress distribution data; generating an optimization recommendation vector for osteotomy angle adjustment based on the normal range of the transfer function and the dynamic pressure envelope.

[0106] In this embodiment, the osteotomy angle adjustment refers to the intraoperative correction value (unit: degrees) to the angle of the femoral or tibial osteotomy surface, used to adjust the prosthesis installation alignment (e.g., distal femoral external rotation angle, tibial plateau posterior tilt angle). The pressure distribution change refers to the change in joint compartment pressure distribution (unit: MPa or percentage) after the osteotomy angle adjustment, reflecting the degree of improvement in pressure uniformity.

[0107] The transfer function is a mathematical model that describes the quantitative relationship between osteotomy angle adjustment (denoted as Δθ) and change in pressure distribution (denoted as ΔP), typically expressed as ΔP = f(Δθ). An optimization recommendation vector is a multidimensional set of recommended data generated based on the transfer function and the normal range of the dynamic pressure envelope, including specific osteotomy angle adjustment values ​​and expected pressure improvement effects. Parameters in the optimization recommendation vector can include the adjustment amount, expected effect, and target range. For example, the optimization recommendation vector is [Δθ = +1.7°, expected ΔP = -1.2 MPa, target pressure = 5.0 MPa].

[0108] The pressure distribution change, ΔP, can be calculated based on the offset of the pressure center trajectory (denoted as Δx) and the gradient change of the shear stress distribution data (denoted as ▽τ). For example, the pressure distribution change can be obtained by taking a weighted sum of the two. For example, ΔP = k1 × |Δx| + k2 × ▽τ, where k1 and k2 are weight coefficients that can be set based on experience.

[0109] The transfer function can be obtained by combining one or more methods such as experimental measurement, finite element analysis, and reinforcement learning.

[0110] For experimental measurement, osteotomies can be performed at different angles on a skeletal model or a living subject, and the corresponding pressure distribution changes can be measured to fit a transfer function. For finite element analysis, a three-dimensional model of the knee joint can be constructed using computer simulation software (such as ANSYS or Abaqus). The mechanical response at different osteotomy angles can be simulated, and the relationship between Δθ and ΔP can be obtained to derive the transfer function. For reinforcement learning, the goal is to learn the optimal strategy by maximizing the cumulative reward.

[0111] For example, by conducting an external rotation angle adjustment experiment on 100 specimens, recording Δθ (-5° to +5°) and the corresponding ΔP, a transfer function was derived through linear regression fitting: ΔP = -3.33 × Δθ. The negative sign indicates that as the external rotation angle increases, the medial knee pressure decreases. For example, for every 1° increase in the osteotomy angle of external rotation, the medial knee pressure decreases by 3.33%. If the osteotomy angle is adjusted to Δθ = +3°, the transfer function can be calculated to result in a change in pressure distribution of ΔP = -10%, indicating a 10% decrease in medial knee pressure.

[0112] In one embodiment, a transfer function between the osteotomy angle adjustment amount and the pressure distribution change amount is established, including: training an intelligent agent through reinforcement learning, using the osteotomy angle adjustment amount as action input, using the pressure distribution change amount and the feedback signal obtained after executing the action as reward feedback, iteratively optimizing to obtain a mapping relationship, and converting the mapping relationship into a transfer function.

[0113] In this embodiment, reinforcement learning is a machine learning method that uses a trial-and-error approach to learn optimal strategies through continuous interaction between an agent and its environment. The goal is to find the optimal mapping from state to action by maximizing cumulative rewards. In reinforcement learning, the agent refers to the algorithmic entity that executes actions, receives environmental feedback, and optimizes its strategy. It is responsible for determining the osteotomy angle adjustment and updating its behavior based on this feedback. Action input refers to the actions that the agent can perform in a specific state. In this scenario, it is the osteotomy angle adjustment (Δθ), including the adjustment direction (internal rotation / external rotation, anteversion / posterior tilt) and the specific angle value. Reward feedback refers to the environment's evaluation signal (numerical value) of the agent's actions. It is determined by factors such as the change in pressure distribution and surgical trauma. Positive rewards encourage effective actions, while negative penalties inhibit ineffective or harmful actions. The mapping relationship refers to the state-action correspondence learned by the agent after reinforcement learning training. This is the functional relationship that outputs the optimal osteotomy angle adjustment given the current pressure distribution (e.g., center of pressure offset, shear stress).

[0114] The core elements of reinforcement learning include state, action, reward, and policy.

[0115] The state represents the state of the agent's environment, such as the current knee joint pressure distribution, bone geometry, and patient weight. The state includes key biomechanical parameters of the current knee joint, the offset of the center of pressure trajectory (e.g., Δx medially / laterally, Δy anteriorly / posteriorly), the maximum gradient of the shear stress distribution data, and the deviation of the current pressure distribution from the dynamic pressure envelope. For example, a state of S = (Δx = 2mm, Δy = 0.5mm, ▽τ_max = 0.6MPa / mm, ΔP_total = 1.8MPa) indicates a 2mm medial offset, a high shear stress gradient, and an overall pressure exceeding the normal range by 1.8MPa.

[0116] Actions represent operations that the agent can perform in the current state, such as different osteotomy angle adjustment schemes (such as external rotation +3° and internal rotation -2°); rewards represent feedback signals obtained after executing actions, such as the degree of improvement in pressure distribution (such as a reward of +10 points for a 10% reduction in medial pressure) and the size of surgical trauma (such as a penalty of -5 points for a 1% increase in trauma); strategies represent the mapping relationship from state to action, and select the best osteotomy angle adjustment scheme based on the current state.

[0117] By constructing a digital simulation environment for knee replacement surgery, inputting the patient's knee joint three-dimensional model and pressure distribution training data, simulating the pressure changes at different osteotomy angles (such as a 1.5MPa decrease in medial pressure when externally rotated 3°), and obtaining the corresponding transfer function.

[0118] The iterative training process of the intelligent agent is as follows: the intelligent agent randomly selects an action (such as Δθ = +2°), the environment feedback pressure changes (such as ΔP = -1.0MPa) and rewards (R = -22), the intelligent agent records "state S-action A-reward R", and adjusts the strategy (to reduce the probability of selecting the action).

[0119] When the average reward of multiple consecutive iterations stabilizes within the preset range, the strategy is judged to have converged and training stops.

[0120] After the iteration is completed, the mapping relationship can be extracted and converted into a transfer function. The mapping relationship of "state-optimal action" can be extracted from the trained agent strategy, for example: When the state S=(Δx=2mm,▽τ_max=0.6MPa / mm), the optimal action A=+3°; When the state S=(Δx=1mm,▽τ_max=0.3MPa / mm), the optimal action A=+1.5°.

[0121] Transfer function conversion: The transfer function is obtained by curve fitting the Δθ in the mapping relationship and the corresponding ΔP. For example, the transfer function can be fitted as: ΔP = -3.33 × Δθ.

[0122] In this embodiment, the transfer function is generated through reinforcement learning, which can improve the accuracy of osteotomy angle adjustment suggestions and provide key technical support for the intelligent optimization of knee replacement surgery.

[0123] In one embodiment, a computer-readable storage medium is provided, on which executable instructions are stored. When the instructions are executed by a processor, the processor executes the steps in the above-mentioned method embodiments.

[0124] In one embodiment, an electronic device is also provided, comprising one or more processors; a memory, wherein one or more programs are stored in the memory, wherein when the one or more programs are executed by one or more processors, the one or more processors execute the steps in the above-mentioned method embodiments.

[0125] In one embodiment, Figure 6 6 shows a schematic diagram of the structure of an electronic device for implementing an embodiment of the present application. The electronic device includes a central processing unit (CPU) 601, which can perform various appropriate actions and processes according to programs stored in a read-only memory (ROM) 602 or programs loaded from a storage unit 608 into a random access memory (RAM) 603. The RAM 603 also stores various programs and data required for the operation of the electronic device 600. The CPU 601, ROM 602, and RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0126] The following components are connected to the I / O interface 605: an input section 606 including a keyboard, mouse, and the like; an output section 607 including devices such as a cathode ray tube (CRT), a liquid crystal display (LCD), and speakers; a storage section 608 including devices such as a hard disk; and a communication section 609 including a network interface card such as a LAN card or a modem. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the I / O interface 605 as needed. Removable media 611, such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, is installed in the drive 610 as needed, so that computer programs read from the removable media can be installed in the storage section 608 as needed.

[0127] In particular, according to embodiments of the present application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present application include a computer program product comprising a computer-readable medium carrying instructions. In such embodiments, the instructions can be downloaded and installed from a network via communication portion 609 and / or installed from removable media 611. When the instructions are executed by central processing unit (CPU) 601, the various method steps described in this application are performed.

[0128] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some or all of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present application.

[0129] Furthermore, those skilled in the art will appreciate that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are intended to be within the scope of this application and to form different embodiments. For example, all of the above embodiments may be used in any combination. The information disclosed in this background section is intended solely to enhance understanding of the overall background of this application and should not be construed as an admission or any form of implication that such information constitutes prior art known to those skilled in the art.

Claims

1. A data processing method for knee replacement surgery, characterized in that: The method comprises: Obtaining pressure distribution data of the joint compartment during the full cycle of joint motion measured by an array sensor; Calculating the pressure center at each flexion degree in the full cycle of joint movement based on the pressure distribution data to form a corresponding pressure center trajectory; Calculating shear stress distribution data during the entire joint motion cycle based on the pressure center trajectory data; Calculating the dynamic pressure envelope of the joint during the full cycle of motion based on the pressure distribution data; Intraoperative correction recommendations for knee replacement surgery are generated based on the center of pressure trajectory, shear stress distribution data, and the dynamic pressure envelope.

2. The method according to claim 1, characterized in that Calculating the dynamic pressure envelope of the joint during the full cycle of motion based on the pressure distribution data includes: Calling a preset convolutional neural network model to extract spatial and temporal features from the pressure distribution data to obtain gradient distribution data of the medial-lateral pressure ratio and phase relationship data of the pressure change rate and the joint angular velocity; The dynamic pressure envelope is generated based on the gradient distribution data and the phase relationship data.

3. The method according to claim 2, characterized in that The calling of a preset convolutional neural network model to extract spatial features and temporal features from the pressure distribution data to obtain gradient distribution data of the medial-lateral pressure ratio and phase relationship data of the pressure change rate and the joint angular velocity includes: Calculating the pressure ratio of the inner and outer sides of the knee joint at each time step based on the pressure distribution data to obtain inner and outer side pressure ratio data; Inputting the inside-outside pressure ratio data into the convolutional neural network model, capturing the spatial relationship in the inside-outside pressure ratio data through the convolution layer and the pooling layer in the convolutional neural network model, and obtaining the gradient distribution data; Calculating pressure change rate data and joint angular velocity data at each time step based on the pressure distribution data; The pressure change rate data and the joint angular velocity data are used as inputs of a network structure including a recurrent layer in the convolutional neural network model. The phase relationship between the pressure change rate data and the joint angular velocity data is captured through the recurrent layer, and the phase relationship data is output.

4. The method according to claim 2, characterized in that The method further comprises: Acquire a pressure distribution training data set, and preprocess the pressure distribution training data in the pressure distribution training data set to form a preprocessed data set that conforms to a standard normal distribution; Iteratively training the pre-trained model using the pre-processed data set, calculating a loss value based on a feature vector output from each iterative training, and adjusting parameters in the pre-trained model based on the calculated loss value until the number of iterations reaches a preset number threshold or the latest loss value is less than a preset loss threshold, ending the iterative training and outputting a trained convolutional neural network model; The process of each iterative training includes sequentially performing tensor processing, convolution processing, normalization processing, activation processing and pooling processing on the preprocessed data set.

5. The method according to claim 1, wherein The pressure distribution data includes the pressure values ​​measured by each sensing unit in the array sensor at each moment; the pressure center at each flexion degree in the full cycle of joint movement is calculated based on the pressure distribution data to form a corresponding pressure center trajectory, including: Calculate the pressure center at the corresponding moment based on the pressure value of each sensing unit at the same moment and the position coordinates of the corresponding sensing unit; The pressure center trajectory is formed based on the pressure center at each moment.

6. The method according to claim 1, characterized in that The calculating of the shear stress distribution data during the full cycle of the joint motion based on the pressure center trajectory data includes: Determining the height of the point of action of the friction force according to the elevation of the surface friction layer of the array sensor; Calculating the moving distance in the tangential direction based on the pressure center trajectory; The shear stress at the corresponding moment is calculated based on the height of the action point, the movement distance and the pressure center at the corresponding moment.

7. The method according to any one of claims 1 to 6, characterized in that Generating intraoperative correction suggestions for knee replacement surgery based on the pressure center trajectory, shear stress distribution data, and the dynamic pressure envelope includes: Establishing a transfer function between the osteotomy angle adjustment amount and the pressure distribution change amount, wherein the pressure distribution change amount is calculated based on the offset of the pressure center trajectory and the gradient change of the shear stress distribution data; An optimization suggestion vector for osteotomy angle adjustment is generated according to the transfer function and the normal range of the dynamic pressure envelope.

8. The method according to claim 7, characterized in that The step of establishing a transfer function between the osteotomy angle adjustment amount and the pressure distribution change amount includes: The intelligent agent is trained through reinforcement learning, with the osteotomy angle adjustment amount as the action input, the pressure distribution change and the feedback signal obtained after executing the action as the reward feedback, and the mapping relationship is obtained by iterative optimization, which is then converted into the transfer function.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores executable instructions, which, when executed by a processor, enable the processor to perform the method according to any one of claims 1 to 8.

10. An electronic device, characterized in that: include: one or more processors; A memory for storing one or more programs, which, when executed by the one or more processors, causes the one or more processors to perform the method according to any one of claims 1 to 8.

Images