A pinn-based acoustic pathology cause-effect inference method for temporomandibular joint

Abstract

The application discloses a temporomandibular joint acoustic pathology cause-effect inference method based on PINN, relates to the technical field of medical signal processing, computational biomechanics and artificial intelligence, and comprises the following steps: S1000, signal acquisition; S2000, input variable calculation; S3000, model construction, a temporomandibular joint acoustic pathology cause-effect inference model is constructed based on a physical information neural network (PINN); S4000, model training; S5000, temporomandibular joint pathology inference. Through joint modeling of mechanical parameters, acoustic responses and their time evolution relationship, continuous quantitative evaluation of the degree of temporomandibular joint lesion is realized, the partial differential equation is used to constrain the network solution space, the displacement field, the sound pressure field and their space-time evolution output by the model have clear physical meanings, and therefore the credibility and traceability of the prediction result in clinical application are improved.

Description

Technical Field

[0001] This invention relates to the fields of medical signal processing, computational biomechanics and artificial intelligence, and in particular to a PINN-based method for causal inference of acoustic pathology of the temporomandibular joint. Background Technology

[0002] Temporomandibular joint disorders (TMD) are a common orofacial disease characterized by joint clicking, crepitus, pain, and limited function. Their pathogenesis is complex, involving multiple factors such as abnormal articular disc position, soft tissue degeneration, muscle dysfunction, and alterations in the articular surface biomechanics. In clinical diagnosis, abnormal acoustic signals generated during joint movement are considered important external indicators of the internal pathological state of the joint; therefore, acoustic signal-based pathological analysis of the temporomandibular joint has gradually gained attention.

[0003] In existing technologies, the main research methods for the acoustic pathology of the temporomandibular joint include the following categories:

[0004] One type of method relies on clinical experience or simple signal analysis techniques. It uses devices such as electronic stethoscopes and accelerometers to collect acoustic signals of joint movement and extracts time-domain or frequency-domain features, such as amplitude, spectral distribution, and energy indicators, to help doctors determine whether there are abnormal popping or friction sounds. This type of method is simple to implement, but it is highly dependent on human experience, has poor adaptability to different patients and different movement patterns, and is difficult to reveal the specific pathological mechanisms behind acoustic abnormalities.

[0005] Another type of approach incorporates traditional machine learning or deep learning techniques, using acoustic features as input to directly establish a mapping relationship between acoustic signals and disease labels. Examples include using support vector machines and convolutional neural networks to classify or grade temporomandibular joint disorders. While this type of approach improves automatic identification accuracy to some extent, it is essentially still a data-driven "black box model." The model output lacks physical and biomechanical explanations, making it difficult to answer the causes of acoustic anomalies and maintain stable performance when sample distribution changes.

[0006] In addition, some studies have attempted to model the biomechanical behavior of the temporomandibular joint using medical imaging or finite element methods to analyze disc displacement, stress distribution, and tissue deformation. While these methods can theoretically reflect the internal mechanical state of the joint, they typically rely on idealized geometric models and material parameters, making it difficult to incorporate acoustic information from the patient's actual movement. Furthermore, the modeling and computational costs are high, making them unsuitable for rapid, continuous clinical assessments.

[0007] In summary, existing technologies generally suffer from the following shortcomings: First, most methods use acoustic signals only as classification features, without establishing a causal relationship between acoustic abnormalities and joint movement and tissue mechanics; second, data-driven models lack physical constraints, resulting in insufficient interpretability of prediction results and difficulty in supporting mechanistic analysis of disease severity; third, purely mechanical or imaging modeling methods struggle to integrate real acoustic data from patients, limiting their ability to characterize individual differences; and fourth, existing methods often only provide a judgment of "whether it is abnormal," making it difficult to analyze the severity and progression of the disease.

[0008] Therefore, those skilled in the art are dedicated to developing a PINN-based method for causal inference of acoustic pathology of the temporomandibular joint. Summary of the Invention

[0009] In view of the above-mentioned deficiencies of the prior art, the technical problem to be solved by the present invention is how to analyze the acoustic abnormalities of the temporomandibular joint.

[0010] This invention uses a Physical Information Neural Network (PINN) to jointly model the mechanical state variables and acoustic response variables of the temporomandibular joint during functional movement, employing a continuous function approximation method. By uniformly mapping acoustic feature vectors, spatial coordinate parameters, temporal variables, and external loads and control parameters to the displacement and sound pressure fields of the temporomandibular joint, it enables inference of pathological behaviors of the temporomandibular joint.

[0011] In one embodiment of the present invention, a PINN-based method for causal inference of temporomandibular joint acoustic pathology is provided, comprising the following steps:

[0012] S1000, signal acquisition, acquires raw signals of the temporomandibular joint during functional movement;

[0013] S2000 Input variable calculation: Based on the original signal, construct multi-source pathological related variables and calculate the input variables;

[0014] S3000, Model Construction: Based on the Physical Information Neural Network (PINN), a causal inference model for the acoustic pathology of the temporomandibular joint is constructed. Through continuous function approximation, the mapping relationship between input variables and key state quantities of the temporomandibular joint is established.

[0015] S4000, model training, calculation of multi-physical constraint loss, training of the causal inference model of temporomandibular joint acoustic pathology until the multi-physical constraint loss is minimized;

[0016] S5000, Temporomandibular Joint Pathology Inference: The temporomandibular joint acoustic pathology causal inference model is used to perform pathological analysis of the temporomandibular joint pathology.

[0017] Optionally, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, the original signal includes the original acoustic signal, spatial coordinates of the temporomandibular joint region, functional movement cycle information, and individual movement strategy information.

[0018] Furthermore, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, step S2000 includes:

[0019] S2100, acoustic feature vector construction, performs denoising, segmentation, and time-frequency analysis on the original acoustic signal, and extracts acoustic feature vectors that reflect joint pathological characteristics;

[0020] S2200, Spatial coordinate variable construction, mapping the spatial coordinates of the temporomandibular joint region to a three-dimensional spatial coordinate system. The temporomandibular joint region includes the articular disc, articular head, and joint cavity of the temporomandibular joint. Spatial coordinate variables describing the spatial distribution of displacement field and sound pressure field are constructed.

[0021] S2300, Time Variable Construction: Based on functional motion cycle information, the motion process is mapped into continuous time variables to describe the location of pathological acoustic events on the time axis.

[0022] S2400, external load and control parameter construction, based on individual movement strategy information, including muscle driving force and masticatory muscle tension, parameterizes the force state of different patients under different movement conditions, constructs external load and control parameters, and enhances individualized analysis capabilities;

[0023] S2500, Input Vector Construction, Input Vector The definition is as follows:

[0024] ,

[0025] in, , This represents the acoustic feature vector of the temporomandibular joint that was actually collected and extracted from the patient. Let be the real number field, and its dimension be . ; For spatial coordinates, Here, represents the spatial coordinate vector of the temporomandibular joint region in three-dimensional space. Using three-dimensional coordinates, representing spatial components in three orthogonal directions, it is used to describe the spatial distribution of the displacement field and sound pressure field. For time variables, For external load and control parameters.

[0026] Furthermore, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, the original acoustic signals include snapping sounds, friction sounds, and continuous noise.

[0027] Optionally, in the PINN-based acoustic pathology causal inference method for temporomandibular joint in any of the above embodiments, the acoustic feature vector includes energy and energy mutation-related features, spectral distribution and high-frequency enhancement features, time-frequency statistical features, and nonlinear dynamic features.

[0028] Furthermore, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, the functional movement cycle information includes the complete opening and closing cycle.

[0029] Optionally, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in any of the above embodiments, the external load and control parameters include temporomandibular joint motion driving force, masticatory muscle and related muscle group tension, and individualized motion control strategy parameters.

[0030] Furthermore, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, the Physical Information Neural Network (PINN) adopts a multi-layer fully connected feedforward neural network structure. The input layer receives acoustic feature vectors, spatial coordinate variables, time variables, and external load and control parameters, and the output layer outputs the displacement field and sound pressure field of the temporomandibular joint, respectively.

[0031] Furthermore, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, the Physical Information Neural Network (PINN) calculates the temporal and spatial partial derivatives of the output results through an automatic differentiation mechanism, constructs a multi-physical constraint loss, and approximates the displacement field and sound pressure field that meet the multi-physical constraint conditions in a continuous function space, thereby realizing a physically consistent mapping relationship between the input variables and the key state quantities of the temporomandibular joint.

[0032] Furthermore, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, the key state quantities of the temporomandibular joint... The definition is as follows:

[0033] ,

[0034] Among them, displacement field It describes the deformation and displacement behavior of the articular disc, articular head and contact area during movement, and is used to judge abnormal displacement of the articular disc, reduction delay, sudden change of movement path and local strain concentration.

[0035] sound pressure field It describes the acoustic response caused by friction, impact and material nonlinearity, and transforms the subjectively perceptible joint sound anomalies into an analytical physical quantity.

[0036] Optionally, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in any of the above embodiments, step S4000 includes:

[0037] S4100. Construct multiple physical constraints. Based on the motion control characteristics, acoustic response characteristics, and elastic mechanical characteristics of the temporomandibular joint during functional movement, construct multiple physical constraints, including motion control physical constraint loss, patient real data driven physical constraint loss, and elastic mechanical physical constraint loss.

[0038] S4200: Calculate multi-physical constraint loss by combining multi-physical constraints with the patient's actual acoustic data. Furthermore, through a weight adjustment strategy, adaptive attention can be achieved for different pathological mechanisms;

[0039] S4300, Model Training: Minimize and optimize the loss of multiple physical constraints, train the causal inference model of temporomandibular joint acoustic pathology. Training is completed when the loss of multiple physical constraints converges to within the convergence threshold and satisfies the solution space of multiple physical constraints.

[0040] Furthermore, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, step S4100 includes:

[0041] S4110. Constructing the physical constraint loss for motion control: Based on the motion control characteristics of the temporomandibular joint during functional movement, predict the displacement and velocity fields of the temporomandibular joint using an acoustic pathological causal inference model, and construct the physical constraint loss for motion control. The formula is as follows:

[0042] ,

[0043] in, The displacement field is predicted by the acoustic pathology causal inference model of the temporomandibular joint. The velocity field predicted by the temporomandibular joint acoustic pathology causal inference model. The velocity is obtained by automatically differentiating the time derivative of the displacement field predicted by the temporomandibular joint acoustic pathology causal inference model. The acceleration is obtained by taking the time derivative of the velocity field predicted by the temporomandibular joint acoustic pathology causal inference model. In order to be in Below , and time The overall control force function determined by both parties For time variables, For external load and control parameters, This represents taking the square of the L2 norm of the physical residual term within the parentheses, used to measure the degree of deviation between the neural network prediction and the physical constraint equations;

[0044] S4120. Construct a patient-based real-data-driven physical constraint loss mechanism. Based on the individualized acoustic response characteristics of the temporomandibular joint during functional movements, construct a patient-based real-data-driven physical constraint loss mechanism. The physical consistency error of the sound pressure field prediction results over time is measured by the following formula:

[0045] ,

[0046] in, and These represent the sound pressure field and displacement field predicted by the temporomandibular joint acoustic pathology causal inference model, respectively. The time derivative of the predicted sound pressure field, To infer the causal relationship of the temporomandibular joint acoustic pathology model in terms of parameters The predicted acoustic-structure coupling term is used to constrain the physical consistency between the sound pressure field and structural motion during model training. The acoustic-structural coupling coefficient modulates the degree to which structural motion affects the acoustic pressure field response, reflecting the pathological mechanism corresponding to temporomandibular joint movement abnormalities. This represents the divergence of the velocity field predicted by the acoustic pathology causal inference model of the temporomandibular joint, measuring local volume changes, contact compression, or separation trends. This is a data-driven item describing the impact of patient temporomandibular joint roughness, soft tissue degeneration, and differences in movement rhythm on acoustic performance.

[0047] S4130. Constructing elastic mechanical physical constraint loss: Based on the elastic mechanical characteristics of the temporomandibular joint during functional movement, i.e., stress-strain physical constraints, constructing elastic mechanical physical constraint loss. The formula is as follows:

[0048] ,

[0049] in, and These represent the predicted stress tensor and the predicted displacement field, respectively, measuring the degree of deviation of the prediction results from the elastic equilibrium relationship in the spatial and temporal dimensions. This is a volumetric force term, representing gravity or equivalent distributed load. This represents the equivalent density.

[0050] Furthermore, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, step S4110 includes:

[0051] S4111. Construct basic kinematic state variables, obtain the spatial coordinates of the temporomandibular joint during mouth opening, mouth closing, protrusion and lateral movements, as well as the time variables used to describe the complete functional movement cycle, construct displacement field, velocity field and acceleration field, construct basic kinematic state variables, and describe the basic kinematics of the temporomandibular joint under the control of the neuromuscular system.

[0052] S4112. Establish hybrid dynamic differential equations, model the causal relationship between displacement, velocity and control force, and construct first- and second-order hybrid dynamic differential equations, including the acceleration field of the temporomandibular joint and the comprehensive control force function, so that the temporomandibular joint motion satisfies the continuous and smooth biomechanical characteristics in the time evolution process.

[0053] S4113. Calculate abnormal changes in motion control, introduce displacement, velocity, time and load parameters into the comprehensive control force function, and describe abnormal changes in control mechanisms closely related to pathological states.

[0054] S4114. Construct the physical constraint loss for motion control. Using the displacement and velocity fields predicted by the temporomandibular joint acoustic pathology causal inference model, and the hybrid dynamic differential equations, construct the physical constraint loss for motion control. The formula is as follows:

[0055] .

[0056] Furthermore, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, the hybrid dynamic differential equation formula is as follows:

[0057] ,

[0058] ,

[0059] It represents the acceleration field of the temporomandibular joint, reflecting the dynamic response caused by changes in muscle driving force and contact force; In order to be in Below , and time The combined control force function, jointly determined by neuromuscular drive, ligament constraint, and joint contact conditions, describes the equivalent control input. This represents the displacement field of the articular disc, articular head, and contact area. This is the first derivative of the displacement field with respect to time, i.e., the velocity field.

[0060] Furthermore, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, abnormal changes in control mechanisms include changes in acceleration response caused by abnormal muscle tension, abrupt changes in contact state caused by abnormal anterior displacement or repositioning of the articular disc, changes in control force in the time domain due to abnormal movement rhythm, and changes in load conditions caused by individual differences, through parameters. reflect.

[0061] Furthermore, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, step S4120 includes:

[0062] S4121. Construct individualized temporomandibular joint acoustic feature data for patients, and collect the original acoustic signals of different patients during mouth opening, mouth closing, protrusion or lateral movement, as well as the corresponding time stamp information;

[0063] S4122. Raw acoustic signal processing: Perform signal processing and feature extraction on the raw acoustic signal to obtain acoustic feature vectors representing abnormal acoustic events.

[0064] S4123. Establish coupling differential relationship: By constructing a time-first partial differential equation, based on the time evolution of the sound pressure field, the structural changes caused by joint movement, and the patient's real data, establish the acoustic-dynamic coupling differential relationship between the sound pressure field and joint movement.

[0065] S4124, Constructing Real Patient Data-Driven Physical Constraint Loss Based on the time derivative of the sound pressure field, the sound-structure coupling term, and the data-driven term, a physical constraint loss driven by real patient data is constructed to measure the physical consistency error of the sound pressure field prediction results over time. The formula is as follows:

[0066] .

[0067] Furthermore, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, the acoustic-dynamic coupling differential relationship formula is as follows:

[0068] ,

[0069] in, For sound pressure field The time derivative describes the rate of acoustic energy release, reflects the instantaneous changes in abnormal acoustic events, aligns acoustic anomalies with sudden events during joint movement in time, and describes the rate of change of sound pressure at the moment the abnormal acoustic event occurs. For velocity field; The acoustic-structural coupling term in the acoustic pathology causal inference model of the temporomandibular joint describes the intensity of the conversion of joint structural motion into acoustic radiation energy. This represents the acoustic feature vector of the temporomandibular joint that was actually collected and extracted from the patient.

[0070] Furthermore, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, abnormal acoustic events include snapping sounds, sudden friction sounds, and continuous noises.

[0071] Furthermore, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, joint movement abnormalities include: the impact effect generated during the repositioning of the articular disc; abnormal friction caused by irregular articular surface morphology; and sudden changes in local contact state in the joint contact area.

[0072] Furthermore, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, the acoustic-structural coupling coefficient... The range of values ​​is .

[0073] Furthermore, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, the acoustic-structural coupling coefficient... It is 0.95.

[0074] Furthermore, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, step S4130 includes:

[0075] S4131. Establish the geometric relationship between the strain tensor and the displacement field. Under the assumption of small deformation, establish the geometric relationship between the strain tensor and the displacement field of the joint disc and its surrounding soft tissue, as shown in the following formula:

[0076] ,

[0077] in, It represents the displacement field gradient, reflecting the tensile, compressive, and shear trends in different directions; This is the transpose of the displacement field gradient matrix, used to eliminate rigid body rotation components and retain only the true deformation information; The strain tensor of the articular disc and its surrounding soft tissues reflects the actual deformation state of the articular disc during functional movement.

[0078] S4132. Establish the constitutive relationship between the stress tensor and the strain tensor, as shown in the following formula:

[0079] ,

[0080] in, It represents the stress tensor of the articular disc and its surrounding soft tissues, describing the distribution of internal forces under stress. The equivalent elastic modulus tensor is a key implicit parameter that is highly correlated with the elastic pathological state. Its changes reflect the elastic pathological characteristics and comprehensively characterize the mechanical properties of the articular disc material. This represents the tensor inner product operation;

[0081] S4133. Establish force balance equations: Establish force balance equations for the articular disc and its surrounding soft tissues during functional movement, reflecting the changes in stress transmission path and inertial response characteristics when pathological changes occur in the material properties or structural morphology of the articular disc. The formulas are as follows:

[0082] ,

[0083] in, This represents the divergence of the stress field, corresponding to the distribution of internal forces; This is a volumetric force term, representing gravity or equivalent distributed load. Indicates equivalent density; This represents the second time derivative of the displacement field, corresponding to the inertial effect;

[0084] 4134. Construction of elasticity and physical constraint loss The following formula describes the consistency between the material parameters, stress-strain relationship, and elasticity model of the articulated disc and articulated surface:

[0085] ,

[0086] in, and These represent the predicted stress tensor of the joint disc and its surrounding soft tissue, and the predicted displacement field, respectively, measuring the degree of deviation of the prediction results from the elastic equilibrium relationship in the spatial and temporal dimensions.

[0087] Furthermore, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, the elastic pathological features include the increase or decrease in overall stiffness caused by articular disc degeneration; anisotropic changes caused by fiber structure destruction or rearrangement; and the degree of material performance degradation under long-term abnormal load.

[0088] Optionally, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in any of the above embodiments, different pathological mechanisms include anterior disc displacement, increased friction, and tissue sclerosis.

[0089] Furthermore, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, step S4200 includes:

[0090] S4210. Calculate multi-physical constraint loss by using multi-physical constraints and the patient's actual acoustic data. The formula is as follows:

[0091] ,

[0092] in, For multi-physical constraint loss, To assess the loss of physical constraints in motion control, this reflects whether the joint displacement-velocity-acceleration relationship conforms to the basic kinematic and dynamic laws under the regulation of the neuromuscular system. The weights for the physical constraint loss in motion control; Using real patient data to drive physical constraint loss, we characterize the degree of deviation between the predicted results and the patient's actual acoustic and imaging data. Weights for physical constraint loss are driven by real patient data. For loss due to physical constraints in elasticity, The weight of the loss due to the physical constraints of elasticity;

[0093] S4220, Weight Adjustment: Adjust according to the weight adjustment strategy. It adapts to different pathological focuses.

[0094] Optionally, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, The range of values ​​is .

[0095] Preferably, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, , , .

[0096] Furthermore, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, the weight adjustment strategy includes: when focusing on pathological types dominated by motor control mechanisms, increasing the weight of the loss of physical constraints on motor control. , , , When focusing on pathology types dominated by individualized load mechanisms, increase the weight of patient-based data-driven physical constraint loss. , , , When focusing on pathology types dominated by elastic mechanical mechanisms, the weight of elastic mechanical physical constraint loss should be increased. , , , .

[0097] Furthermore, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, the convergence threshold ranges as follows: .

[0098] Preferably, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, the convergence threshold is: .

[0099] Furthermore, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, the solution space of multiple physical constraints includes... The temporal continuity and dynamic consistency of the constraints are determined by Constrained acoustic-motion causal consistency, by The rationality of the elasticity of the constraint.

[0100] Optionally, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in any of the above embodiments, step S5000 includes:

[0101] S5100, Statistical Analysis of Loss Components: Statistical analysis is performed on the loss components, which include loss of physical constraints for motion control, loss of physical constraints driven by real patient data, and loss of physical constraints for elasticity. The mean of the loss components, the relative proportion of the loss components in the multiple physical constraints, and the slope of the trend of the loss components in the later stages of training are calculated.

[0102] S5200, Determining the dominant physical mechanism: Based on the rules for determining the dominant physical mechanism, the dominant physical mechanism of the abnormal acoustic event is determined.

[0103] S5300, Quantitative assessment of structural degeneration: When the dominant physical mechanism is elastic mechanical mechanism, quantitative assessment of the degree of temporomandibular joint structural degeneration is performed.

[0104] S5400, Temporomandibular Joint Pathology Inference: The temporomandibular joint acoustic pathology causal inference model is used to infer the pathology of the temporomandibular joint.

[0105] Furthermore, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, step S5100 includes:

[0106] S5110, Calculate the mean of the loss components, and finally train the temporomandibular joint acoustic pathology causal inference model. N The mean of the loss components in the next iteration The formula is as follows:

[0107] ,

[0108] in, To lose weight, Number the loss components. For the first Loss component This training session N The number of training rounds is an integer, ranging from 5% to 10%. ;

[0109] S5120. Calculate the relative proportion of the loss component in the multi-physics constraint loss. The formula is as follows:

[0110] ,

[0111] in, This represents the average of the multi-physics constraint losses.

[0112] S5130. Calculate the slope of the trend of the loss component in the later stage of training. The change in loss component per unit training epoch is given by the following formula:

[0113] ,

[0114] in, For the first Each loss component Indicates the training round, each This represents the state of the temporomandibular joint acoustic pathology causal inference model after one parameter update; if , indicating the first The loss component has converged. This is the convergence threshold for the rate of change of loss; if This indicates that the degree of violation of constraints still has an increasing trend.

[0115] Furthermore, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, the dominant physical mechanism judgment rule is: the relative proportion of the loss component in the multi-physical constraint loss. At that time, determine the loss component. The corresponding physical mechanism is the dominant physical mechanism. The threshold for the predefined dominant physical mechanism includes:

[0116] When the relative proportion of motion control physical constraint loss in multi-physics constraint loss At this time, the dominant physical mechanism is the motor control mechanism, including neuromuscular control instability, discontinuous joint movement, sudden acceleration or abnormal velocity fluctuation;

[0117] When real patient data drives the relative proportion of physical constraint loss in multi-physical constraint loss At that time, the dominant physical mechanism is the individualized load mechanism, including individual differences, special meshing loads and local frictional anomalies;

[0118] When the relative proportion of elastic mechanical physical constraint loss in multi-physical constraint loss At this time, the dominant physical mechanism is the elastic mechanical mechanism, including the degradation of the stiffness of the joint disc material, the destruction of the fiber structure, the decrease in the load-bearing capacity of the contact area, and the significant deviation of the elastic modulus or Poisson's ratio from the normal value.

[0119] Furthermore, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, a predefined threshold for the dominant physical mechanism is used. The range is 40%–70%.

[0120] Preferably, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, a predefined threshold for the dominant physical mechanism is used. It is 60%.

[0121] Furthermore, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, the loss rate of change convergence threshold is... The range is .

[0122] Preferably, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, the loss rate of change convergence threshold is... for .

[0123] Furthermore, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, step S5300 includes:

[0124] S5310, Time Trend Analysis of Elastic Mechanical Physical Constraint Loss, When Elastic Mechanical Physical Constraint Loss When the time trend exceeds the time trend threshold and the duration exceeds a specified percentage of the total observation time, it is determined that there is persistent structural degeneration of the temporomandibular joint.

[0125] S5320. Identification of concentrated areas of structural degradation; calculation of spatial gradient based on displacement and stress fields output by the temporomandibular joint acoustic pathology causal inference model. The formula is as follows:

[0126] ,

[0127] definition The continuous spatial region is a concentrated region of structural degradation, in which, This is the spatial gradient threshold.

[0128] Preferably, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, the time trend threshold is the mean value of elastic mechanical physical constraint loss in normal physiological samples. with standard deviation weighted combination value .

[0129] Preferably, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, the specified ratio is 30%.

[0130] Preferably, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, the spatial gradient threshold... ,in This represents the mean of the spatial gradient in a normal physiological sample. It represents the standard deviation of the spatial gradient in a normal physiological sample.

[0131] Optionally, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, the degeneration concentration area includes the posterior disc zone, the articular surface contact concentration area, and the displacement mutation area.

[0132] Furthermore, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, step S5400 includes:

[0133] S5410. Calculate the sound pressure change frequency, i.e., the number of sound pressure changes per unit time, using the following formula:

[0134] ,

[0135] in, This represents the number of sound pressure abrupt change events. The length of the observation time window, The frequency of sudden sound pressure change;

[0136] S5420, Define a sound pressure surge event, when at time... t Predicted sound pressure field The time derivative exceeds the time derivative threshold, that is:

[0137] ,

[0138] Or the predicted sound pressure amplitude abruptly exceeds the sound pressure amplitude threshold, i.e.:

[0139] ,

[0140] Defined as a sudden change in sound pressure level; where, Indicates time t The predicted sound pressure field, This represents the predicted sound pressure field at the previous moment. The time step is determined by the numerical calculation time interval. This indicates the magnitude of the change, regardless of whether it is positive or negative. The time derivative threshold, The sound pressure amplitude threshold;

[0141] S5430, Determine the frequency of sudden sound pressure changes The rate of change of , when the frequency of sudden changes in sound pressure increases per unit time, that is:

[0142] ,

[0143] This indicates a significant upward trend in the frequency of sudden sound pressure changes, suggesting increased discontinuity in temporomandibular joint contact, enhanced cartilage surface friction, increased frequency of articular disc displacement jumps, or a more severe degree of motor control instability.

[0144] S5440, Define the acoustic energy release intensity, as shown in the following formula:

[0145] ,

[0146] in, This refers to the acoustic region of the joint. The intensity of acoustic energy release;

[0147] When the acoustic energy release per unit time increases, that is:

[0148] ,

[0149] If the acoustic energy release per unit time shows an increasing trend, it is determined that the corresponding displacement field gradient increases, reflecting the change in the intensity level of the abnormal acoustic event.

[0150] S5450. Determine the trend of multi-physical coupling. When the frequency of sudden changes in sound pressure per unit time increases, the release of acoustic energy per unit time increases, and the loss component of the dominant physical mechanism continues to rise, it indicates that a positive feedback is formed between the acoustic abnormality event and the structural degeneration. The pathological state of the temporomandibular joint is evolving towards a more unstable state. Conversely, the pathological state of the temporomandibular joint is stable or improved, thus completing the pathological inference of the temporomandibular joint.

[0151] Preferably, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, the time derivative threshold is... The mean of the time derivative of the sound pressure field in a normal physiological sample. with standard deviation weighted combination value .

[0152] Preferably, in the PINN-based causal inference method for temporomandibular joint acoustic pathology in the above embodiments, the sound pressure amplitude threshold is... The mean value of the abrupt change in sound pressure amplitude in the sound pressure field of a normal physiological sample. with standard deviation weighted combination value .

[0153] This invention achieves continuous quantitative assessment of the severity of temporomandibular joint lesions by jointly modeling the mechanical parameters, acoustic response, and their temporal evolution. By constraining the network solution space through partial differential equations, the displacement field, sound pressure field, and their spatiotemporal evolution output by the model have clear physical meaning, thereby improving the reliability and traceability of the prediction results in clinical applications.

[0154] The following will further explain the concept, specific structure, and technical effects of the present invention in conjunction with the accompanying drawings, so as to fully understand the purpose, features, and effects of the present invention. Attached Figure Description

[0155] Figure 1 This is a flowchart of an exemplary embodiment of a PINN-based method for causal inference of acoustic pathology of the temporomandibular joint.

[0156] Figure 2 This is a flowchart illustrating the calculation of input variables in an exemplary embodiment;

[0157] Figure 3 This is a flowchart of model training in an exemplary embodiment;

[0158] Figure 4 This is a flowchart illustrating the pathological deduction of the temporomandibular joint, as exemplified in an exemplary embodiment. Detailed Implementation

[0159] The following description, with reference to the accompanying drawings, illustrates several preferred embodiments of the present invention to make its technical content clearer and easier to understand. The present invention can be embodied in many different forms, and the scope of protection of the present invention is not limited to the embodiments mentioned herein.

[0160] This invention provides a PINN-based method for causal inference of acoustic pathology in the temporomandibular joint, such as... Figure 1 As shown, it includes the following steps:

[0161] S1000 Signal Acquisition: Acquires raw signals of the temporomandibular joint during functional movement, including raw acoustic signals, spatial coordinates of the temporomandibular joint region, functional movement cycle information, and individual movement strategy information. Raw acoustic signals include snapping sounds, friction sounds, and continuous noise.

[0162] S2000 Input variable calculation: Based on the original signal, construct multi-source pathologically relevant variables and calculate the input variables; specifically including:

[0163] S2100, acoustic feature vector construction, performs denoising, segmentation, and time-frequency analysis on the original acoustic signal, and extracts acoustic feature vectors that reflect joint pathological characteristics, including energy and energy mutation-related features, spectral distribution and high-frequency enhancement features, time-frequency statistical features, and nonlinear dynamic features;

[0164] S2200, Spatial coordinate variable construction, mapping the spatial coordinates of the temporomandibular joint region to a three-dimensional spatial coordinate system. The temporomandibular joint region includes the articular disc, articular head, and joint cavity of the temporomandibular joint. Spatial coordinate variables describing the spatial distribution of displacement field and sound pressure field are constructed.

[0165] S2300, Time Variable Construction: Based on functional motion cycle information, the motion process is mapped to a continuous time variable to describe the location of pathological acoustic events on the time axis. Functional motion cycle information includes the complete opening and closing cycle.

[0166] S2400, external load and control parameter construction, based on individual movement strategy information, including muscle driving force and masticatory muscle tension, parameterizes the force state of different patients under different movement conditions, constructs external load and control parameters, including temporomandibular joint movement driving force, masticatory muscle and related muscle group tension, and individualized movement control strategy parameters, and enhances individualized analysis capability;

[0167] S2500, Input Vector Construction, Input Vector The definition is as follows:

[0168] ,

[0169] in, , This represents the acoustic feature vector of the temporomandibular joint that was actually collected and extracted from the patient. Let be the real number field, and its dimension be . ; For spatial coordinates, Here, represents the spatial coordinate vector of the temporomandibular joint region in three-dimensional space. Using three-dimensional coordinates, representing spatial components in three orthogonal directions, it is used to describe the spatial distribution of the displacement field and sound pressure field. For time variables, For external load and control parameters.

[0170] S3000 Model Construction: A causal inference model for the acoustic pathology of the temporomandibular joint (TMJ) is constructed based on a Physical Information Neural Network (PINN). A mapping relationship between input variables and key state variables of the TMJ is established through continuous function approximation. The PINN employs a multi-layer fully connected feedforward neural network structure. The input layer receives acoustic feature vectors, spatial coordinate variables, temporal variables, and external load and control parameters. The output layer outputs the displacement field and sound pressure field of the TMJ, respectively. An automatic differentiation mechanism calculates the temporal and spatial partial derivatives of the output results, constructing a multi-physical constraint loss. This approximates the displacement field and sound pressure field that meet the multi-physical constraint conditions within the continuous function space, achieving a physically consistent mapping relationship between input variables and key state variables of the TMJ. The definition is as follows:

[0171] ,

[0172] Among them, displacement field It describes the deformation and displacement behavior of the articular disc, articular head and contact area during movement, and is used to judge abnormal displacement of the articular disc, reduction delay, sudden change of movement path and local strain concentration.

[0173] sound pressure field It describes the acoustic response caused by friction, impact and material nonlinearity, and transforms the subjectively perceptible joint sound anomalies into an analytical physical quantity.

[0174] S4000, Model Training: Calculating multi-physical constraint loss, training the temporomandibular joint acoustic pathology causal inference model until the multi-physical constraint loss is minimized; specifically including:

[0175] S4100. Constructing multi-physical constraints: Based on the motion control characteristics, acoustic response characteristics, and elastic mechanical characteristics of the temporomandibular joint during functional movement, constructing multi-physical constraints, including motion control physical constraint loss, patient-data-driven physical constraint loss, and elastic mechanical physical constraint loss; specifically including:

[0176] S4110. Constructing the physical constraint loss for motion control: Based on the motion control characteristics of the temporomandibular joint during functional movement, predict the displacement and velocity fields of the temporomandibular joint using an acoustic pathological causal inference model, and construct the physical constraint loss for motion control. The formula is as follows:

[0177] ,

[0178] in, The displacement field is predicted by the acoustic pathology causal inference model of the temporomandibular joint. The velocity field predicted by the temporomandibular joint acoustic pathology causal inference model. The velocity is obtained by automatically differentiating the time derivative of the displacement field predicted by the temporomandibular joint acoustic pathology causal inference model. The acceleration is obtained by taking the time derivative of the velocity field predicted by the temporomandibular joint acoustic pathology causal inference model. In order to be in Below , and time The overall control force function determined by both parties For time variables, For external load and control parameters, This represents taking the square of the L2 norm of the physical residual term within the parentheses, used to measure the deviation between the neural network prediction and the physical constraint equations; specifically including:

[0179] S4111. Construct basic kinematic state variables, obtain the spatial coordinates of the temporomandibular joint during mouth opening, mouth closing, protrusion and lateral movements, as well as the time variables used to describe the complete functional movement cycle, construct displacement field, velocity field and acceleration field, construct basic kinematic state variables, and describe the basic kinematics of the temporomandibular joint under the control of the neuromuscular system.

[0180] S4112. Establish hybrid dynamic differential equations, model the causal relationship between displacement, velocity and control force, and construct first- and second-order hybrid dynamic differential equations, including the acceleration field of the temporomandibular joint and the comprehensive control force function, so that the temporomandibular joint motion satisfies the continuous and smooth biomechanical characteristics in the time evolution process.

[0181] S4113. Calculate abnormal changes in motor control. Introduce displacement, velocity, time, and load parameters into the comprehensive control force function to describe abnormal changes in control mechanisms closely related to the pathological state. This includes changes in acceleration response caused by abnormal muscle tension, abrupt changes in contact state caused by abnormal anterior disc displacement or repositioning, changes in control force in the time domain due to abnormal motor rhythm, and changes in load conditions caused by individual differences. This is achieved through parameter... reflect;

[0182] S4114. Construct the physical constraint loss for motion control. Using the displacement and velocity fields predicted by the temporomandibular joint acoustic pathology causal inference model, and the hybrid dynamic differential equations, construct the physical constraint loss for motion control. The formula is as follows:

[0183] ,

[0184] The differential equation for hybrid power is as follows:

[0185] ,

[0186] ,

[0187] It represents the acceleration field of the temporomandibular joint, reflecting the dynamic response caused by changes in muscle driving force and contact force; In order to be in Below , and time The combined control force function, jointly determined by neuromuscular drive, ligament constraint, and joint contact conditions, describes the equivalent control input. This represents the displacement field of the articular disc, articular head, and contact area. This is the first derivative of the displacement field with respect to time, i.e., the velocity field.

[0188] S4120. Construct a patient-based real-data-driven physical constraint loss mechanism. Based on the individualized acoustic response characteristics of the temporomandibular joint during functional movements, construct a patient-based real-data-driven physical constraint loss mechanism. The physical consistency error of the sound pressure field prediction results over time is measured by the following formula:

[0189] ,

[0190] in, and These represent the sound pressure field and displacement field predicted by the temporomandibular joint acoustic pathology causal inference model, respectively. The time derivative of the predicted sound pressure field, To infer the causal relationship of the temporomandibular joint acoustic pathology model in terms of parameters The predicted acoustic-structure coupling term is used to constrain the physical consistency between the sound pressure field and structural motion during model training. The acoustic-structural coupling coefficient is... The value is 0.95, which adjusts the degree of influence of structural movement on the sound pressure field response, reflecting the pathological mechanism corresponding to temporomandibular joint movement abnormalities. Joint movement abnormalities include: the impact effect generated during the repositioning of the articular disc; abnormal friction caused by irregular articular surface morphology; and sudden changes in local contact state in the joint contact area. This represents the divergence of the velocity field predicted by the acoustic pathology causal inference model of the temporomandibular joint, measuring local volume changes, contact compression, or separation trends. This is a data-driven item describing the impact of patient temporomandibular joint roughness, soft tissue degeneration, and differences in movement rhythm on acoustic performance; specifically including:

[0191] S4121. Construct individualized temporomandibular joint acoustic feature data for patients, and collect the original acoustic signals of different patients during mouth opening, mouth closing, protrusion or lateral movement, as well as the corresponding time stamp information;

[0192] S4122. Raw acoustic signal processing: Perform signal processing and feature extraction on the raw acoustic signal to obtain acoustic feature vectors that characterize abnormal acoustic events, including popping sounds, sudden friction sounds, and continuous noises.

[0193] S4123. Establish the coupling differential relationship. By constructing a first-order partial differential equation for time, and based on the time evolution of the sound pressure field, structural changes caused by joint movement, and real patient data, establish the acoustic-dynamic coupling differential relationship between the sound pressure field and joint movement. The formula is as follows:

[0194] ,

[0195] in, For sound pressure field The time derivative describes the rate of acoustic energy release, reflects the instantaneous changes in abnormal acoustic events, aligns acoustic anomalies with sudden events during joint movement in time, and describes the rate of change of sound pressure at the moment the abnormal acoustic event occurs. For velocity field; The acoustic-structural coupling term in the acoustic pathology causal inference model of the temporomandibular joint describes the intensity of the conversion of joint structural motion into acoustic radiation energy. This represents the acoustic feature vector of the temporomandibular joint that was actually collected and extracted from the patient.

[0196] S4124, Constructing Real Patient Data-Driven Physical Constraint Loss Based on the time derivative of the sound pressure field, the sound-structure coupling term, and the data-driven term, a physical constraint loss driven by real patient data is constructed to measure the physical consistency error of the sound pressure field prediction results over time. The formula is as follows:

[0197] .

[0198] S4130. Constructing elastic mechanical physical constraint loss: Based on the elastic mechanical characteristics of the temporomandibular joint during functional movement, i.e., stress-strain physical constraints, constructing elastic mechanical physical constraint loss. The formula is as follows:

[0199] ,

[0200] in, and These represent the predicted stress tensor and the predicted displacement field, respectively, measuring the degree of deviation of the prediction results from the elastic equilibrium relationship in the spatial and temporal dimensions. This is a volumetric force term, representing gravity or equivalent distributed load. Represents equivalent density; specifically including:

[0201] S4131. Establish the geometric relationship between the strain tensor and the displacement field. Under the assumption of small deformation, establish the geometric relationship between the strain tensor and the displacement field of the joint disc and its surrounding soft tissue, as shown in the following formula:

[0202] ,

[0203] in, It represents the displacement field gradient, reflecting the tensile, compressive, and shear trends in different directions; This is the transpose of the displacement field gradient matrix, used to eliminate rigid body rotation components and retain only the true deformation information; The strain tensor of the articular disc and its surrounding soft tissues reflects the actual deformation state of the articular disc during functional movement.

[0204] S4132. Establish the constitutive relationship between the stress tensor and the strain tensor, as shown in the following formula:

[0205] ,

[0206] in, It represents the stress tensor of the articular disc and its surrounding soft tissues, describing the distribution of internal forces under stress. The equivalent elastic modulus tensor is a key implicit parameter that is highly correlated with the elastic pathological state. Its changes reflect the characteristics of elastic pathology, including the increase or decrease in overall stiffness caused by joint disc degeneration, the anisotropic changes caused by fiber structure destruction or rearrangement, and the degree of material performance degradation under long-term abnormal loads. It comprehensively characterizes the mechanical properties of joint disc materials. This represents the tensor inner product operation;

[0207] S4133. Establish force balance equations: Establish force balance equations for the articular disc and its surrounding soft tissues during functional movement, reflecting the changes in stress transmission path and inertial response characteristics when pathological changes occur in the material properties or structural morphology of the articular disc. The formulas are as follows:

[0208] ,

[0209] in, This represents the divergence of the stress field, corresponding to the distribution of internal forces; This is a volumetric force term, representing gravity or equivalent distributed load. Indicates equivalent density; This represents the second time derivative of the displacement field, corresponding to the inertial effect;

[0210] 4134. Construction of elasticity and physical constraint loss The following formula describes the consistency between the material parameters, stress-strain relationship, and elasticity model of the articulated disc and articulated surface:

[0211] ,

[0212] in, and These represent the predicted stress tensor of the joint disc and its surrounding soft tissue, and the predicted displacement field, respectively, measuring the degree of deviation of the prediction results from the elastic equilibrium relationship in the spatial and temporal dimensions.

[0213] S4200: Calculate multi-physical constraint loss by combining multi-physical constraints with the patient's actual acoustic data. Furthermore, through a weighted adjustment strategy, adaptive attention is achieved for different pathological mechanisms, including anterior disc displacement, increased friction, and tissue sclerosis; specifically including:

[0214] S4210. Calculate multi-physical constraint loss by using multi-physical constraints and the patient's actual acoustic data. The formula is as follows:

[0215] ,

[0216] in, For multi-physical constraint loss, To assess the loss of physical constraints in motion control, this reflects whether the joint displacement-velocity-acceleration relationship conforms to the basic kinematic and dynamic laws under the regulation of the neuromuscular system. The weights for the physical constraint loss in motion control; Using real patient data to drive physical constraint loss, we characterize the degree of deviation between the predicted results and the patient's actual acoustic and imaging data. Weights for physical constraint loss are driven by real patient data. For loss due to physical constraints in elasticity, The weight of the loss due to the physical constraints of elasticity;

[0217] S4220, Weight Adjustment: Adjust according to the weight adjustment strategy. To adapt to different pathological focuses, the weighting adjustment strategy includes: when focusing on pathological types dominated by motor control mechanisms, increasing the weight of the loss of physical constraint of motor control. , , , When focusing on pathology types dominated by individualized load mechanisms, increase the weight of patient-based data-driven physical constraint loss. , , , When focusing on pathology types dominated by elastic mechanical mechanisms, the weight of elastic mechanical physical constraint loss should be increased. , , , .

[0218] S4300, Model Training: Minimize and optimize the multi-physical constraint loss to train a temporomandibular joint acoustic pathology causal inference model. The multi-physical constraint loss converges to a convergence threshold. Training is complete when the solution space satisfies the multi-physics constraints, which include the solution space consisting of... The temporal continuity and dynamic consistency of the constraints are determined by Constrained acoustic-motion causal consistency, by The rationality of the elasticity of the constraint.

[0219] S5000, Temporomandibular Joint Pathological Inference: This involves using a temporomandibular joint acoustic pathology causal inference model to analyze temporomandibular joint pathology, specifically including:

[0220] S5100, Statistical Analysis of Loss Components: This section performs statistical analysis on the loss components, which include loss due to motion control physical constraints, loss due to patient-driven physical constraints based on real-world data, and loss due to elasticity and mechanical constraints. It calculates the mean of each loss component, its relative proportion within the total loss due to multiple physical constraints, and the slope of its changing trend in the later stages of training. Specifically, this includes:

[0221] S5110, Calculate the mean of the loss components, and finally train the temporomandibular joint acoustic pathology causal inference model. N The mean of the loss components in the next iteration The formula is as follows:

[0222] ,

[0223] in, To lose weight, Number the loss components. For the first Loss component This training session N The number of training rounds is an integer, ranging from 5% to 10%. ;

[0224] S5120. Calculate the relative proportion of the loss component in the multi-physics constraint loss. The formula is as follows:

[0225] ,

[0226] in, This represents the average of the multi-physics constraint losses.

[0227] S5130. Calculate the slope of the trend of the loss component in the later stage of training. The change in loss component per unit training epoch is given by the following formula:

[0228] ,

[0229] in, For the first Each loss component Indicates the training round, each This represents the state of the temporomandibular joint acoustic pathology causal inference model after one parameter update; if , indicating the first The loss component has converged. The threshold for convergence of the rate of change of loss. for ;like This indicates that the degree of violation of constraints still has an increasing trend.

[0230] S5200. Dominant physical mechanism determination: The dominant physical mechanism of the anomalous acoustic event is determined according to the dominant physical mechanism judgment rule. The dominant physical mechanism judgment rule is: the relative proportion of the loss component in the multi-physical constraint loss. At that time, determine the loss component. The corresponding physical mechanism is the dominant physical mechanism. To preset the threshold of the dominant physical mechanism, It is 60%, specifically including:

[0231] When the relative proportion of motion control physical constraint loss in multi-physics constraint loss At this time, the dominant physical mechanism is the motor control mechanism, including neuromuscular control instability, discontinuous joint movement, sudden acceleration or abnormal velocity fluctuation;

[0232] When real patient data drives the relative proportion of physical constraint loss in multi-physical constraint loss At that time, the dominant physical mechanism is the individualized load mechanism, including individual differences, special meshing loads and local frictional anomalies;

[0233] When the relative proportion of elastic mechanical physical constraint loss in multi-physical constraint loss At this time, the dominant physical mechanism is the elastic mechanical mechanism, including the degradation of the stiffness of the joint disc material, the destruction of the fiber structure, the decrease in the load-bearing capacity of the contact area, and the significant deviation of the elastic modulus or Poisson's ratio from the normal value.

[0234] S5300, Quantitative Assessment of Structural Degeneration: When the dominant physical mechanism is elasticity, a quantitative assessment of the degree of temporomandibular joint structural degeneration is performed; specifically including:

[0235] S5310, Time Trend Analysis of Elastic Mechanical Physical Constraint Loss, When Elastic Mechanical Physical Constraint Loss The time trend exceeds the time trend threshold, which is the mean value of elastic mechanical physical constraint loss in normal physiological samples. with standard deviation weighted combination value If the duration exceeds 30% of the total observation time, it is determined that there is persistent structural degeneration of the temporomandibular joint.

[0236] S5320. Identification of concentrated areas of structural degradation; calculation of spatial gradient based on displacement and stress fields output by the temporomandibular joint acoustic pathology causal inference model. The formula is as follows:

[0237] ,

[0238] definition The continuous spatial region is a region of concentrated structural degeneration, including the posterior disc region, the joint surface contact concentration region, and the region of abrupt displacement. Spatial gradient threshold, ,in This represents the mean of the spatial gradient in a normal physiological sample. It represents the standard deviation of the spatial gradient in a normal physiological sample.

[0239] S5400, Temporomandibular Joint Pathology Inference: The temporomandibular joint acoustic pathology causal inference model infers the pathology of the temporomandibular joint, specifically including:

[0240] S5410. Calculate the sound pressure change frequency, i.e., the number of sound pressure changes per unit time, using the following formula:

[0241] ,

[0242] in, This represents the number of sound pressure abrupt change events. The length of the observation time window, The frequency of sudden sound pressure change;

[0243] S5420, Define a sound pressure surge event, when at time... t Predicted sound pressure field The time derivative exceeds the time derivative threshold. ,Right now:

[0244] ,

[0245] The mean of the time derivative of the sound pressure field in a normal physiological sample. with standard deviation weighted combination value ;

[0246] Or the predicted sound pressure amplitude abruptly exceeds the sound pressure amplitude threshold, i.e.:

[0247] ,

[0248] Defined as a sudden change in sound pressure level; where, Indicates time t The predicted sound pressure field, This represents the predicted sound pressure field at the previous moment. The time step is determined by the numerical calculation time interval. This indicates the magnitude of the change, regardless of whether it is positive or negative. The time derivative threshold, The sound pressure amplitude threshold. The mean value of the abrupt change in sound pressure amplitude in the sound pressure field of a normal physiological sample. with standard deviation weighted combination value ;

[0249] S5430, Determine the frequency of sudden sound pressure changes The rate of change of , when the frequency of sudden changes in sound pressure increases per unit time, that is:

[0250] ,

[0251] This indicates a significant upward trend in the frequency of sudden sound pressure changes, suggesting increased discontinuity in temporomandibular joint contact, enhanced cartilage surface friction, increased frequency of articular disc displacement jumps, or a more severe degree of motor control instability.

[0252] S5440, Define the acoustic energy release intensity, as shown in the following formula:

[0253] ,

[0254] in, This refers to the acoustic region of the joint. The intensity of acoustic energy release;

[0255] When the acoustic energy release per unit time increases, that is:

[0256] ,

[0257] If the acoustic energy release per unit time shows an increasing trend, it is determined that the corresponding displacement field gradient increases, reflecting the change in the intensity level of the abnormal acoustic event.

[0258] S5450. Determine the trend of multi-physical coupling. When the frequency of sudden changes in sound pressure per unit time increases, the release of acoustic energy per unit time increases, and the loss component of the dominant physical mechanism continues to rise, it indicates that a positive feedback is formed between the acoustic abnormality event and the structural degeneration. The pathological state of the temporomandibular joint is evolving towards a more unstable state. Conversely, the pathological state of the temporomandibular joint is stable or improved, thus completing the pathological inference of the temporomandibular joint.

[0259] To verify the technical effects of the above embodiments, the present invention constructed a verification sample including a normal control group and multiple pathological groups, and systematically verified the model convergence, physical consistency, dominant mechanism identification ability, structural degeneracy ability, and pathological trend prediction ability.

[0260] Sample Construction: Normal Physiological Samples 42 cases of patients with anterior displacement of the temporomandibular joint disc 38 cases of patients with increased temporomandibular joint friction 35 cases of temporomandibular joint structural degeneration patients 31 cases, with a total sample size of 146 cases. Complete opening and closing cycles were collected from each case. It synchronously records the original acoustic signals, spatial coordinate information, and motion control parameters to construct input variables.

[0261] First, during the model training phase, the convergence of the multi-physics constraint loss is verified by setting a convergence threshold for the multi-physics constraint loss. The model was trained under multi-physical constraints. Experimental results show that the PINN-based causal inference model for temporomandibular joint acoustic pathology achieves stable convergence after an average of approximately 18,500 iterations, with a convergence success rate of 96.6%. The final average multi-physical constraint loss stabilizes at 7.8 × 10⁻⁶. -5 The magnitude indicates that the above embodiments have good numerical stability and optimization feasibility under multi-physics coupling constraints.

[0262] Secondly, to verify the improvement in physical consistency, the above embodiments were compared with traditional neural network models that do not include physical constraints. Experimental results show that, regarding displacement and velocity consistency constraints, the average error of the above embodiments is 0.016, while that of the traditional neural network model is 0.064. Regarding the elastic equilibrium residuals, the residuals of the above embodiments are 0.018, while those of the traditional neural network model are 0.057. These results demonstrate that by introducing motion control physical constraints and elasticity physical constraints, the displacement and stress fields output by the above embodiments have stronger physical rationality and significantly reduce non-physical explanations.

[0263] Regarding the identification of dominant physical mechanisms, the results of human expert annotation were used as the gold standard to verify the dominant physical mechanisms obtained by the model in the above embodiments based on the statistical rules of loss components. Experimental results show that the above embodiments achieved identification accuracies of 92.1%, 89.4%, and 94.6% for the three pathological mechanisms of disc prolapse, increased friction, and elastic degeneration, respectively, with an overall accuracy of 91.8%. Compared to traditional models based solely on acoustic feature classification, this represents an accuracy of approximately 74.3%, an improvement of about 17%. These results demonstrate that by performing proportion and trend analysis on multiple loss components, it is possible to determine pathological mechanisms with causal explanatory significance, rather than merely relying on superficial acoustic feature classification.

[0264] Regarding the structural degradation mitigation capability, the equivalent elastic modulus predicted by the above embodiments was compared with the imaging evaluation results. Experimental results show that the average relative error of the predicted equivalent elastic modulus by the above embodiments is 6.8%, with a maximum error not exceeding 12.3%, and a correlation coefficient reaching 0.91, indicating that the material parameters output by the model of the above embodiments have high reliability. Simultaneously, the spatial overlap rate (Dice coefficient) between the degradation concentration region identified through spatial gradient analysis and the imaging diagnostic results reached 0.87, with an average spatial error of approximately 1.8 mm, verifying the accuracy of the above embodiments in structural localization.

[0265] Regarding the ability to predict pathological trends, a 6-month follow-up study was conducted on a subset of samples for validation. The experiment showed that when the model in the above embodiments determined that the frequency of sudden sound pressure spikes and the intensity of acoustic energy release were continuously increasing, over 84% of the samples experienced worsening structural degeneration during the follow-up period; while only about 12% of the samples predicted to be stable showed deterioration. The overall trend prediction accuracy reached over 88%. Furthermore, when the three conditions of increased frequency of sudden sound pressure spikes, enhanced acoustic energy release, and a continuous increase in the elastic mechanical loss component were simultaneously met, the proportion of structural degeneration occurring within 6 months exceeded 90%, with a risk ratio approaching 8 times, indicating that the above embodiments can identify the positive feedback evolution mechanism of structural degeneration.

[0266] In summary, the above-described embodiments significantly improve the physical consistency of the model through partial differential equation physical constraints, ensuring that the displacement field, sound pressure field, and their spatiotemporal evolution conform to the laws of dynamics and elasticity. These embodiments can transform subjectively perceptible acoustic anomalies into physically meaningful structural state variables, achieving a causal mapping between acoustic anomalies and structural pathology. Furthermore, these embodiments enable continuous quantitative assessment of temporomandibular joint material parameters, achieving a quantitative expression of the degree of structural degeneration, rather than merely making classification judgments. Finally, these embodiments can identify pathological evolution trends, providing early warnings of disease aggravation risks, possessing clinical predictive value, and outperforming existing purely data-driven methods.

[0267] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A PINN-based method for causal inference of acoustic pathology in the temporomandibular joint, characterized in that, Includes the following steps: S1000, signal acquisition, acquires raw signals of the temporomandibular joint during functional movement; S2000, Input variable calculation: Based on the original signal, construct multi-source pathological related variables and calculate the input variables; S3000, Model Construction: Based on the Physical Information Neural Network (PINN), a causal inference model for the acoustic pathology of the temporomandibular joint is constructed. Through continuous function approximation, the mapping relationship between the input variables and the key state quantities of the temporomandibular joint is established. S4000, Model Training: Calculate the multi-physical constraint loss and train the temporomandibular joint acoustic pathology causal inference model until the multi-physical constraint loss is minimized; specifically including: S4100. Constructing multi-physical constraints: Based on the motion control characteristics, acoustic response characteristics, and elastic mechanical characteristics of the temporomandibular joint during functional movement, constructing multi-physical constraints, including motion control physical constraint loss, patient-data-driven physical constraint loss, and elastic mechanical physical constraint loss; specifically including: S4110. Construct the motion control physical constraint loss. The formula is as follows: ， in, The displacement field is predicted by the acoustic pathology causal inference model of the temporomandibular joint. The velocity field predicted by the temporomandibular joint acoustic pathology causal inference model. To automatically differentiate the velocity from the time derivative of the displacement field predicted by the temporomandibular joint acoustic pathology causal inference model, The acceleration is obtained by taking the time derivative of the velocity field predicted by the aforementioned temporomandibular joint acoustic pathology causal inference model. In order to be in Below , and time The overall control force function determined by both parties For time variables, For external load and control parameters, This indicates taking the square of the L2 norm of the physical residual term within the parentheses; S4120. Construct a physical constraint loss mechanism driven by real patient data. The formula is as follows: ， in, and These represent the sound pressure field and displacement field predicted by the temporomandibular joint acoustic pathology causal inference model, respectively. The time derivative of the predicted sound pressure field, To infer the causal relationship of the temporomandibular joint acoustic pathology model in terms of parameters The predicted acoustic-structure coupling term is obtained below. The acoustic-structural coupling coefficient is... This represents the divergence of the velocity field predicted by the acoustic pathology causal inference model of the temporomandibular joint. For data-driven items; S4130. Constructing elastic mechanical physical constraint loss. The formula is as follows: ， in, and Let represent the predicted stress tensor and the predicted displacement field, respectively. This represents the body force term, indicating gravity or equivalent distributed load. Indicates equivalent density; S4200, Calculate the multi-physical constraint loss by using the multi-physical constraints and the patient's actual acoustic data. Furthermore, through a weight adjustment strategy, adaptive attention can be achieved for different pathological mechanisms; S4300, Model training: Minimize and optimize the multi-physical constraint loss to train the temporomandibular joint acoustic pathology causal inference model. Training is completed when the multi-physical constraint loss converges to within the convergence threshold and satisfies the solution space of the multi-physical constraints. S5000, Temporomandibular Joint Pathology Inference: The temporomandibular joint acoustic pathology causal inference model is used to perform pathological analysis of the temporomandibular joint pathology.

2. The PINN-based acoustic pathology causal inference method for temporomandibular joint as described in claim 1, characterized in that, The raw signals include raw acoustic signals, spatial coordinates of the temporomandibular joint region, functional movement cycle information, and individual movement strategy information.

3. The PINN-based acoustic pathology causal inference method for temporomandibular joint as described in claim 2, characterized in that, The S2000 includes: S2100, Acoustic feature vector construction: Denoise, segment, and perform time-frequency analysis on the original acoustic signal to extract acoustic feature vectors that reflect joint pathological characteristics; S2200, Spatial coordinate variable construction, maps the spatial coordinates of the temporomandibular joint region to a three-dimensional spatial coordinate system; S2300, Time Variable Construction: Based on functional motion cycle information, the motion process is mapped into continuous time variables to describe the location of pathological acoustic events on the time axis. S2400, external load and control parameter construction: Based on individual motion strategy information, the force state of different patients under different motion conditions is parameterized, external load and control parameters are constructed, and individualized analysis capability is enhanced; S2500, Input Vector Construction, Input Vector The definition is as follows: ， in, , This represents the acoustic feature vector of the temporomandibular joint that was actually collected and extracted from the patient. Let be the real number field, and its dimension be . ; For spatial coordinates, Here, represents the spatial coordinate vector of the temporomandibular joint region in three-dimensional space. In three-dimensional coordinates, the spatial components are represented in three orthogonal directions. For time variables, For external load and control parameters.

4. The PINN-based acoustic pathology causal inference method for temporomandibular joint as described in claim 1, characterized in that, Key state parameters of the temporomandibular joint The definition is as follows: ， Among them, displacement field It describes the deformation and displacement behavior of the articular disc, articular head and contact area during movement, and is used to judge abnormal displacement of the articular disc, reduction delay, sudden change of movement path and local strain concentration. sound pressure field It describes the acoustic response caused by friction, impact and material nonlinearity, and transforms the subjectively perceptible joint sound anomalies into an analytical physical quantity.

5. The PINN-based acoustic pathology causal inference method for temporomandibular joint as described in claim 1, characterized in that, The S4200 includes: S4210. Calculate the multi-physical constraint loss by using the multi-physical constraints and the patient's actual acoustic data. The formula is as follows: ， in, The weights for the physical constraint loss of the motion control; Weights for physical constraint loss are driven by real patient data. The weight of the loss due to the physical constraints of elasticity; S4220, Weight Adjustment: Adjust according to the weight adjustment strategy. It adapts to different pathological focuses.

6. The PINN-based acoustic pathology causal inference method for temporomandibular joint as described in claim 1, characterized in that, The S5000 includes: S5100, Statistical analysis of loss components: Statistical analysis is performed on the loss components, which include the loss of physical constraints for motion control, the loss of physical constraints driven by real patient data, and the loss of physical constraints for elasticity. The mean of the loss components, the relative proportion of the loss components in the multiple physical constraints, and the slope of the trend of the loss components in the later stage of training are calculated. S5200, Determining the dominant physical mechanism: Based on the rules for determining the dominant physical mechanism, the dominant physical mechanism of the abnormal acoustic event is determined. S5300, Quantitative assessment of structural degeneration: When the dominant physical mechanism is an elastic mechanical mechanism, a quantitative assessment of the degree of temporomandibular joint structural degeneration is performed. S5400, Temporomandibular Joint Pathology Inference: The temporomandibular joint acoustic pathology causal inference model infers the pathology of the temporomandibular joint.

7. The PINN-based acoustic pathology causal inference method for temporomandibular joint as described in claim 6, characterized in that, The S5100 includes: S5110. Calculate the mean of the loss components. The temporomandibular joint acoustic pathology causal inference model is finally trained. N The mean of the loss components in the next iteration The formula is as follows: ， in, To lose weight, Number the loss components. For the first Loss component This training session N The number of training rounds is an integer, ranging from 5% to 10%. ; S5120. Calculate the relative proportion of the loss component in the multi-physics constraint loss. The formula is as follows: ， in, The mean of the multi-physics constraint loss; S5130. Calculate the slope of the change trend of the loss component in the later stage of training. The change in loss component per unit training epoch is given by the following formula: ， in, For the first Each loss component Indicates the training round, each This represents the state of the temporomandibular joint acoustic pathology causal inference model after one parameter update; if , indicating the first The loss component has converged. This is the convergence threshold for the rate of change of loss; if This indicates that the degree of violation of constraints still has an increasing trend.

8. The PINN-based method for causal inference of temporomandibular joint acoustic pathology as described in claim 7, characterized in that, The S5400 includes: S5410. Calculate the frequency of sound pressure change, that is, the number of sound pressure changes per unit time. S5420, Define a sound pressure surge event, when at time... t Predicted sound pressure field The time derivative exceeds the time derivative threshold; S5430, Determine the frequency of sudden sound pressure changes The rate of change of sound pressure change frequency indicates a significant upward trend when the frequency of sound pressure change increases per unit time. S5440, Define the acoustic energy release intensity, as shown in the following formula: ， in, This refers to the acoustic region of the joint. The acoustic energy release intensity; when the acoustic energy release increases per unit time, it indicates that the acoustic energy release per unit time is increasing, and the corresponding displacement field gradient is determined to be increasing, reflecting the intensity level change of the abnormal acoustic event; S5450. Determine the multi-physical coupling trend. When the frequency of sudden sound pressure changes per unit time increases, the acoustic energy release per unit time increases, and the loss component of the dominant physical mechanism continues to rise, it indicates that a positive feedback is formed between the abnormal acoustic event and structural degeneration, and the pathological state of the temporomandibular joint is evolving towards a more unstable state. Conversely, the pathological state of the temporomandibular joint is stable or improved, thus completing the pathological inference of the temporomandibular joint.

Images