A female pelvic floor dysfunction disease risk model construction system

Abstract

The present application relates to the technical field of pelvic floor MRI risk modeling, and discloses a female pelvic floor dysfunction disease risk model construction system, comprising: first, dynamic magnetic resonance data containing resting, contraction, force and emptying stages are acquired under a supine position, and time information of each stage is recorded; the pubococcygeal line is determined in the mid-sagittal position, a fixed reference coordinate is established, and the pubococcygeal line functional area in the resting stage is mapped to each frame. Subsequently, the displacement field relative to the resting stage is calculated by registration in the reference coordinate, and the normalized displacement curve is extracted along the tail side direction of the pubococcygeal line; the first slope change maximum point of the displacement curve is detected as a critical time in the logarithmic time domain with the starting point of the force stage as a reference, and weighted fitting is performed in the window after the critical time, so that the second slope is obtained as a critical post-power law softening index. Finally, the female pelvic floor dysfunction risk prediction model is constructed with the index, individualized risk probability output is realized, and high sensitivity and quantitative advantages are achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of pelvic floor MRI risk modeling, more particularly, it relates to a female pelvic floor dysfunction disease risk model construction system. BACKGROUND

[0002] The formation process of female pelvic floor dysfunction is affected by many factors such as pregnancy, childbirth, age and gravity stress, and its pathological basis is the continuous deformation and mechanical degradation of the pelvic floor support system. Although the traditional pelvic floor magnetic resonance examination can provide high-resolution anatomical images, it usually only measures quantitatively at the resting or single force stage, and cannot reflect the dynamic evolution of the tissue from the initial stress to instability. With the development of dynamic imaging technology, cine magnetic resonance can continuously record the timing changes of organs and muscle groups during force and emptying processes, however, clinical analysis is mostly limited to geometric parameters such as displacement amplitude, angle or area, which can only reflect the result of structural deformation and cannot describe the evolution path of the mechanical process. Due to the lack of description of the post-yield mechanical behavior, the existing risk model can only make macroscopic classification or grade prediction, and cannot reveal the true damage potential of different individuals under dynamic loading.

[0003] Specifically, although magnetic resonance imaging can provide high temporal resolution dynamic images, there is still no method to extract the softening behavior of the tissue in the post-yield stage from the image sequence. Soft tissue exhibits typical viscoelastic and time-dependent softening characteristics under high stress, and this process often follows a nonlinear decay law, which is difficult to express with linear or static models. In addition, the existing image analysis process lacks a recognition mechanism for this dynamic stage, that is, it cannot determine when to enter the post-yield response region, and there is no corresponding parameter to quantify the softening rate. Therefore, the risk model cannot utilize the time sequence information recorded by magnetic resonance imaging during training, and can only rely on static indicators or empirical thresholds, resulting in limited model generalization ability and prediction accuracy. SUMMARY

[0004] The present application provides a female pelvic floor dysfunction disease risk model construction system, which solves the technical problem proposed in the background art: how to use the observable time course changes of the tissue in the magnetic resonance imaging sequence to quantitatively describe the softening law of the female pelvic floor in the post-yield stage, so as to realize dynamic prediction modeling of the disease risk.

[0005] The present application provides a female pelvic floor dysfunction disease risk model construction system, which includes:

[0006] A data acquisition module acquires dynamic magnetic resonance data containing resting, contraction, force and emptying stages in a supine position, and records the time information of each stage.

[0007] The reference coordinate establishing module determines a pubococcygeal line in the mid-sagittal plane, establishes a fixed reference coordinate based on the pubococcygeal line registration data, generates a functional area corresponding to the pubococcygeal line in the resting phase, and maps to each frame;

[0008] The displacement feature extraction module obtains a relative resting displacement field in the reference coordinate and each frame functional area through registration, extracts regional displacement features in the tail side direction of the pubococcygeal line, and normalizes the length of the pubococcygeal line to obtain a continuous-time displacement curve;

[0009] The critical moment detection module detects the maximum position of the first slope change of the displacement curve in the logarithmic time domain as the critical moment, taking the start time of the force phase as the starting point;

[0010] The softening index calculation module performs weighted fitting on the logarithmic relationship between the displacement increment and the time in the post-critical window to obtain the second slope as the post-critical power-law softening index; wherein the post-critical window is determined based on the critical moment;

[0011] The risk prediction module constructs a female pelvic floor dysfunction risk prediction model based on the post-critical power-law softening index, and outputs an individual risk probability, including:

[0012] The post-critical power-law softening index is inputted;

[0013] The model training data consists of multiple sample pairs, each sample pair containing the post-critical power-law softening index of the jth sample and the event indicator of the jth sample, the event indicator taking the value of zero or one, j being a positive integer of the sample number, and the total number of samples being a positive integer;

[0014] The linear prediction is defined as the product of the model intercept parameter and the model slope parameter and the post-critical power-law softening index of the jth sample;

[0015] The logistic function is defined as a function that maps any real number to between zero and one, and the calculation method of the logistic function is one divided by one plus the negative linear prediction raised to the power of the natural exponential function;

[0016] The individual risk probability of the jth sample is calculated by the logistic function;

[0017] The objective function is defined as the negative logarithmic likelihood based on the individual risk probability, specifically the negative value of the logarithm of the product of the event indicator and the risk probability of each sample, plus one minus the difference between the event indicator and the difference between the risk probability of the sample, to obtain the sample parameter corresponding to each sample; then summing up the sample parameters of all samples to obtain the objective function;

[0018] Adjust the values of the model intercept parameter and the model slope parameter so that the value of the objective function reaches a minimum value, and obtain the optimal model intercept parameter and the optimal model slope parameter;

[0019] For any individual to be evaluated, a linear predictor is calculated using the corresponding critical post-power law softening index and the optimal model intercept parameter, the optimal model slope parameter, and the risk probability of the individual is obtained as the model output by the logic function.

[0020] The beneficial effects of the present application include: by establishing a fixed reference coordinate system based on the pubococcygeal line and realizing multi-stage image registration, the present application can accurately extract the continuous displacement curve of the pelvic floor soft tissue in the whole process of rest, contraction, force and emptying, overcoming the limitation of traditional MRI which only relies on static indicators and is difficult to identify the dynamic characteristics of the tissue. Through the slope analysis of the displacement curve in the logarithmic time domain, the present application can automatically identify the critical moment when the tissue enters the yield, and extract the power law softening index as a quantitative parameter, which truly reflects the mechanical degradation degree of the tissue under the loading of abdominal pressure. The logic function established based on the index can output individualized risk probability, significantly improving the sensitivity and interpretability of disease risk assessment, and realizing the breakthrough from anatomical structure measurement to mechanical behavior modeling. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 is a flowchart of a female pelvic floor dysfunction disease risk model construction system of the present application. DETAILED DESCRIPTION

[0022] The subject matter described herein will now be discussed with reference to example implementations. It should be understood that the discussion of these implementations is merely meant to provide a better understanding of the subject matter described herein and can be changed in function and arrangement without departing from the scope of the present description. Various processes or components can be omitted, substituted, or added according to desired implementations. Additionally, features described with respect to some examples can be combined in other examples.

[0023] As shown in Figure 1 , a female pelvic floor dysfunction disease risk model construction system includes:

[0024] A data acquisition module acquires dynamic magnetic resonance data in a supine position including rest, contraction, force and emptying stages, and records time information of each stage;

[0025] A reference coordinate establishment module determines a pubococcygeal line in the mid-sagittal plane, establishes a fixed reference coordinate based on the pubococcygeal line registration data, generates a functional area corresponding to the pubococcygeal line in the rest stage and maps it to each frame;

[0026] A displacement feature extraction module, in the reference coordinate and each frame functional area, obtains a displacement field relative to rest through registration, extracts regional displacement features along the tail side direction of the pubococcygeal line and normalizes them by the length of the pubococcygeal line, and obtains a continuous time displacement curve;

[0027] a critical moment detection module, detecting a first slope change maximum position of the displacement curve in a logarithmic time domain as a critical moment, starting from a force stage start moment;

[0028] a softening index calculation module, fitting a logarithmic relationship between displacement increment and time in a post-critical window to obtain a second slope as a post-critical power law softening index, wherein the post-critical window is determined based on the critical moment;

[0029] a risk prediction module, constructing a female pelvic floor dysfunction risk prediction model based on the post-critical power law softening index, and outputting an individual risk probability, including:

[0030] the post-critical power law softening index as input;

[0031] The model training data consists of a plurality of sample pairs, each sample pair containing the post-critical power law softening index of the jth sample and the event indicator of the jth sample, the event indicator taking a value of zero or one, j being a positive integer of sample number, and the total number of samples being a positive integer;

[0032] defining a linear prediction as the product of the model intercept parameter and the model slope parameter and the post-critical power law softening index of the jth sample;

[0033] defining a logistic function as a function that maps any real number to between zero and one, the logistic function being calculated as one divided by one plus the negative linear prediction raised to the power of the natural exponential function;

[0034] calculating the individual risk probability of the jth sample through the logistic function;

[0035] defining a target function as a negative log likelihood based on the individual risk probability, specifically the negative value of the log of the risk probability of each sample multiplied by the event indicator of the sample, plus one minus the difference between the event indicator multiplied by the negative value of the log of the difference between the risk probability of the sample, to obtain a sample parameter corresponding to each sample; and summing up the sample parameters of all samples to obtain the target function;

[0036] adjusting the values of the model intercept parameter and the model slope parameter so that the value of the target function reaches a minimum value, to obtain the optimal model intercept parameter and the optimal model slope parameter;

[0037] For any individual to be evaluated, the linear prediction is calculated using the corresponding post-critical power law softening index and the optimal model intercept parameter and the optimal model slope parameter, and the risk probability of the individual is obtained through the logistic function as the model output.

[0038] In an embodiment of the present application, the dynamic magnetic resonance data of the supine position containing the rest, contraction, force and emptying stages is obtained, and the time information of each stage is recorded, including:

[0039] determining the start time and the end time of the resting phase, forming the resting phase time information;

[0040] collecting image frames corresponding to the resting phase, the contraction phase, the straining phase and the emptying phase at fixed time intervals, forming dynamic magnetic resonance data.

[0041] Resting phase: The patient takes a supine position, the knee joint is flexed, the whole body is relaxed, and natural breathing is maintained. The pelvic floor muscles are not actively contracted or strained. This phase is used to obtain baseline data of the basic anatomy.

[0042] Contraction phase: The patient performs a Kegel exercise under the guidance of the physician, that is, actively tightens the muscles around the anus and maintains it for a few seconds, until receiving the relaxation instruction. This phase is used to evaluate the contraction function of the pelvic floor muscles.

[0043] Straining phase: The patient performs an abdominal pressure increasing action, that is, after deep inhalation, holds breath, and forces the abdominal wall inward and downward (similar to the forced state before defecation), and maintains it until the action is stable.

[0044] Emptying phase: For rectal or bladder function evaluation, if evaluating the rectum, the patient tries to expel the simulated feces (such as a balloon containing ultrasonic coupling agent) in the rectum, if evaluating the bladder, the patient tries to expel urine in the bladder, until there is no obvious expelling action or the device display shows that the organ is close to the emptying state. This phase needs to be performed after the straining phase.

[0045] The four phases are collected in the order of resting → contraction → straining → emptying. After the end of the previous phase, the patient needs to adjust to the starting state of the next phase for no more than ten seconds, to ensure the continuity of the dynamic data.

[0046] End time of the resting phase: The moment when the patient completes the resting state preparation, the device displays stable image signals (such as signal noise value decreases to a pre-set low level), and the physician issues the start contraction instruction is used as the reference. This moment is also used as the start time of the contraction phase.

[0047] Start time of each phase: The start time of the contraction, straining and emptying phases is determined by the moment when the physician issues the corresponding action instruction. The system time when the instruction is issued is recorded by the device, which is used as the official start time point of the phase.

[0048] Time information composition: The time information of each phase needs to include the start time and the end time. The end time is determined by the moment when the patient completes the action of the phase (such as relaxation of the contraction action, release of the straining state, stop of the emptying action) and the physician issues the phase end instruction.

[0049] Fixed time interval needs to be set in combination with the fast scanning capability of the magnetic resonance device. The interval corresponding to the collection of two to three image frames per second is usually selected, that is, the time difference between the collection of adjacent two image frames is between one third of a second and half a second.

[0050] The fixed time interval needs to meet two conditions, specifically:

[0051] 1. Can completely capture the dynamic changes of the pelvic floor tissue in each stage (such as the displacement process of the tissue in the force stage);

[0052] 2. Avoiding too small interval leading to data redundancy and increasing the storage burden of the device. Specifically, it can be adjusted according to the device model, and both 1.5T and 3.0T magnetic resonance devices need to follow this range.

[0053] Preparation before operation: if the rectal emptying function is evaluated, 100 to 120 milliliters of diluted ultrasonic coupling agent (temperature close to body temperature) needs to be injected into the rectum before the examination to simulate the normal state of feces; if the bladder emptying function is evaluated, the bladder needs to be in a semi-filled state (hold urine until no distending pain) before the examination.

[0054] Action specification: immediately after the end of the force stage, the patient tries to empty according to the natural action of daily defecation or urination, maintains the supine position during the process, avoids body displacement, and the physician observes the morphological changes of the organ through the real-time image observer until the image display organ content is significantly reduced or the patient indicates that he / she cannot continue to empty, which is determined as the end of the emptying stage.

[0055] In an embodiment of the present application, the pubococcygeal line is determined in the mid-sagittal plane, the fixed reference coordinates are established based on the pubococcygeal line registration data, and the functional area corresponding to the pubococcygeal line in the resting stage is generated and mapped to each frame, including:

[0056] Determining the pubic inferior edge point and the coccyx end point in the image frame in the resting stage, and taking the straight line connecting the pubic inferior edge point and the coccyx end point as the pubococcygeal line;

[0057] Taking the image frame corresponding to the beginning of the resting stage as the reference image frame, for any image frame corresponding to a time, constructing a rigid body transformation containing rotation and translation, adjusting the rotation and translation parameters to make the pubic inferior edge point and the coccyx end point in the image frame align with the pubic inferior edge point and the coccyx end point in the reference image frame, respectively, to establish the fixed reference coordinates;

[0058] Rotating the direction vector from the pubic inferior edge point to the coccyx end point counterclockwise by ninety degrees and normalizing to obtain a unit normal vector pointing to the coccyx side half plane;

[0059] On the reference image frame, taking the pixel set located on the coccyx side of the pubococcygeal line and satisfying that the projection of the unit normal vector to the connecting line of each point in the region to the pubic inferior edge point is less than or equal to zero as the functional area;

[0060] Mapping the functional area on the reference image frame to the image frame corresponding to each time through the inverse transformation of the rigid body transformation to obtain the functional area in each image frame.

[0061] Identification of the pubic inferior margin point: On the sagittal image frame in the resting phase, find the lowest edge of the pubic symphysis (a high-density bony structure, shown as a bright white area in the image), and take the midpoint or the point of maximum curvature of this edge line as the pubic inferior margin point. If there is pubic symphysis hyperplasia, take the lower edge of the normal bone cortex under the hyperplasia area as the reference.

[0062] Identification of the coccyx end point: Along the coccyx (a beaded bony structure), trace from top to bottom to the last coccyx segment, and take the end point of the farthest end of this coccyx segment; if the coccyx has fusion or bending variations, take the distal edge point of the last coccyx joint that can be clearly identified as the reference.

[0063] Confirmation requirements for landmark points: Two physicians with more than 3 years of experience in pelvic floor MRI diagnosis are required to identify the points respectively. If the coordinate error of the two points exceeds 1 pixel, the final point position is determined through joint review to ensure the consistency of the landmark points.

[0064] Adjustment of rigid body transformation parameters, as follows:

[0065] Adjustment target: The sum of the straight-line distances between the two landmark points of the target image frame and the corresponding points of the reference frame is minimized as the optimization target, and the distance unit uses the image pixel spacing (e.g. 1.5 mm / pixel).

[0066] Operation mode: Use an automatic optimization algorithm (such as the gradient descent method) to adjust the parameters. First, fix the rotation parameters to adjust the translation, and then fine-tune the rotation and translation parameters simultaneously until the sum of the distances is less than 0.5 pixels.

[0067] Stopping condition: When the parameter iteration is performed for 3 consecutive times, the change in the sum of the distances is less than 0.1 pixel, or the maximum number of iterations (e.g. 50) is reached, the adjustment is stopped and the current parameters are saved.

[0068] Unit normal vector generation, as follows:

[0069] Definition of the reference coordinate system: Take the conventional display orientation of the MRI image as the reference, i.e. the top of the image corresponds to the patient's head side, the bottom corresponds to the patient's foot side, the left side corresponds to the patient's right side, and the right side corresponds to the patient's left side.

[0070] Rotation operation: First, determine the direction vector from the pubic inferior margin point to the coccyx end point (e.g. in the image, it appears to point from the top left to the bottom right), and then rotate the vector by 90 degrees clockwise around the starting point (pubic inferior margin point) according to the rule that clockwise rotation is to the right and counterclockwise rotation is to the left, forming the initial normal vector.

[0071] Normalization: Calculate the length of the initial normal vector (i.e. the straight-line distance between the two endpoints of the vector), and divide each directional component of the initial normal vector by the length to obtain a unit normal vector with a length of 1, ensuring that it only represents direction and does not contain length information.

[0072] Projection calculation and functional area definition, including:

[0073] Projection operation: for each pixel on the tail side of the pubococcygeal line in the reference image frame, first draw a line from the pubic inferior edge point to the pixel (i.e. the line from the point to the pubic inferior edge point), then determine the direction relationship between the line and the unit normal vector, if the line is inclined to the direction indicated by the unit normal vector, or perpendicular to the normal vector, the projection result is less than or equal to zero; if it is inclined in the opposite direction, the projection result is greater than zero.

[0074] Functional area screening rule: first retain all pixels located below the pubococcygeal line (tail side), then remove the pixels with projection results greater than zero from them, and the remaining pixel set is the functional area. It can be simply understood as: taking the pubococcygeal line as the boundary, taking the pixel area below it and towards the unit normal vector direction.

[0075] Area verification: the functional area needs to include the pixels corresponding to the pelvic floor support structures such as levator ani muscle and anterior rectal wall, if the area range deviates from the actual anatomy obviously, the rotation direction of the normal vector needs to be checked again.

[0076] Inverse transformation area mapping, specifically as follows:

[0077] Inverse transformation parameters: the rigid body transformation parameters (rotation angle, translation) of each image are calculated in reverse to obtain the inverse transformation parameters, the rotation angle is taken as the opposite number, and the translation is taken as the same distance in the opposite direction.

[0078] Pixel mapping rule: the coordinates of each pixel in the functional area of the reference frame are calculated by substituting the inverse transformation parameters of the corresponding frame, to obtain the corresponding position of the pixel in the target frame; if the calculation result is a non-integer coordinate, the nearest neighbor interpolation method is used to take the nearest integer coordinate pixel.

[0079] Edge processing: if there are scattered isolated pixels (with an area less than 3 pixels) on the edge of the functional area after mapping, they are removed; if there are area breaks, fill in the adjacent pixels of the same type (i.e. pixels on the tail side and with projection meeting the conditions) at the break to ensure the continuity and integrity of each frame of functional area.

[0080] In an embodiment of the present application, in the reference coordinate and each frame of functional area, a relatively static displacement field is obtained by registration, the regional displacement characteristics are extracted along the tail side direction of the pubococcygeal line and normalized by the length of the pubococcygeal line, to obtain a continuous-time displacement curve, including:

[0081] Taking the image gray value of the reference image frame as the reference image gray value;

[0082] For each corresponding image frame, based on the rigid body transformation, the image frame is converted into a rigid body registration image aligned to the reference coordinate;

[0083] By non-rigid registration, the gray difference between the rigid registration image and the reference image frame is minimized while adding a smoothing term of the deformation field, to determine the deformation field of the pixel in the reference coordinate to the corresponding position at each time;

[0084] The deformation field is subtracted from the pixel position in the reference coordinate to obtain the displacement field relative to the resting stage;

[0085] Determine the functional area of each frame and the caudal unit direction pointing to the caudal direction, and the distance between the subpubic edge point and the coccyx terminal point as the pubococcygeal line length;

[0086] In the functional area of each frame, calculate the average value of the projection of the displacement field in the caudal unit direction after rigid transformation, and normalize the average value by dividing the length of the pubococcygeal line to obtain the single-frame area displacement at each time;

[0087] Correlate each time with the corresponding single-frame area displacement to form a displacement curve in continuous time.

[0088] The reference image gray scale only takes the gray scale values of all pixels in the functional area of the resting reference frame, and does not include irrelevant areas (such as background, intestinal gas area) outside the functional area. Specific operation: first determine the functional area (caudal functional area of pubococcygeal line) on the resting reference frame, then extract the gray scale value of each pixel in the area to form a reference gray scale data set for subsequent non-rigid registration gray difference comparison.

[0089] Non-rigid registration smoothing term setting, specifically as follows:

[0090] Smoothing term type: Gaussian smoothing term is adopted, which can suppress local severe deformation by applying Gaussian filtering to the deformation field, and ensure that the deformation conforms to the physiological movement law of the pelvic soft tissue (such as the gradual deformation of the levator ani muscle contraction).

[0091] Smoothing term weight value: according to the field strength adjustment of MRI equipment, the smoothing term weight of 1.5T equipment is 0.1-0.3, and the weight of 3.0T equipment is 0.05-0.2 (the higher the field strength, the higher the image resolution, and the smoothing weight can be appropriately reduced). The principle of weight adjustment: the goal is to make the gray difference in the functional area after registration less than 5%, and the deformation field has no obvious local protrusion. If the gray difference is too large, the weight should be reduced, and if the deformation field appears abnormal protrusion, the weight should be increased.

[0092] Tail side unit direction is determined based on pubococcygeal line: taking pubococcygeal line (connecting the pubic inferior edge point and the coccyx end point) as the reference, the direction perpendicular to the pubococcygeal line and pointing to the lower part of the pelvic floor (the foot side of the patient) is taken as the tail side unit direction. Specific determination: taking the pubococcygeal line as the horizontal reference line, the tail side unit direction is the vertical downward direction (consistent with the tail side in human anatomy), which can be confirmed by the image coordinate system. If the top of the MRI image is the head side of the patient and the bottom is the foot side, the tail side unit direction is the vertical downward direction of the image.

[0093] Rigid body transformation of displacement field is to transform the displacement field of each image frame, so that the displacement field is aligned with the reference coordinate. The specific steps are as follows: first, obtain the rigid body transformation parameters (rotation angle, translation amount, rigid body transformation parameters) corresponding to the image frame at this time, and then adjust each pixel displacement vector in the displacement field according to the inverse transformation of the rigid body transformation (i.e. reverse rotation, reverse translation), eliminate the influence of body motion on the displacement field, and ensure that the adjusted displacement field only reflects the active deformation of the pelvic floor soft tissue, which is consistent with the reference coordinate.

[0094] Pubococcygeal line length is fixed to the straight line distance between the pubic inferior edge point and the coccyx end point in the reference image frame at rest, and the real-time length of each frame is not used. The reason is that the length of the pubococcygeal line in the rest frame represents the reference size of the pelvic bone structure, and the relative position of the bone landmark points may change slightly due to body motion or soft tissue deformation. If the real-time length is used, it will cause the normalization reference to fluctuate, affecting the comparability of the displacement characteristics. Measurement method: directly measure the straight line distance between the two points on the rest reference frame using image measurement tools (such as the distance measurement function provided by the MRI device), and the unit is unified as millimeter.

[0095] Displacement projection mean value calculation rule, including:

[0096] Exclusion of abnormal pixels: first, exclude non-soft tissue pixels (such as bone structure pixels, pixels with gray value higher than 200HU; noise pixels, pixels with gray value lower than 0HU) in the functional area, and only keep soft tissue pixels (gray value between 0-200HU).

[0097] Mean value calculation: for the remaining soft tissue pixels, calculate the projection value of the displacement field after rigid body transformation in the tail side unit direction, and then calculate the arithmetic mean value of all projection values to obtain the regional displacement mean value at this time. If the number of soft tissue pixels in the functional area at a certain time is less than 100 (due to image cropping or organ position change), the functional area mapping needs to be rechecked to ensure that the sample size meets the reliability of mean value calculation.

[0098] Each time of the continuous time displacement curve adopts the relative time of the beginning time of the relative rest phase, instead of the absolute time collected by the device. Specifically, the first frame of image collected at the beginning of the rest phase is taken as the 0 time, and the time of each subsequent frame of image is (the collection sequence of the frame-1) x the fixed time interval (for example, if the fixed interval is 0.5 seconds, the time of the third frame is 1 second). The time axis unit is unified as seconds, the horizontal coordinate of the curve is the relative time, and the vertical coordinate is the single-frame regional displacement amount (millimeters) at the corresponding time.

[0099] In an embodiment of the application, the first slope change maximum position of the displacement curve is detected in the logarithmic time domain as the critical time, starting from the beginning time of the force phase, including:

[0100] A positive small constant is set;

[0101] The difference between each time and the beginning time of the force is added to the small constant, and the logarithm is taken, to obtain the logarithmic time corresponding to each time;

[0102] The difference between the displacement value of the displacement curve at each time and the displacement value of the displacement curve at the beginning of the force is added to the small constant, and the logarithm is taken, to obtain the logarithmic displacement corresponding to each time;

[0103] The difference between the logarithmic displacements corresponding to two adjacent times is calculated, and divided by the difference between the logarithmic times corresponding to the two adjacent times, to obtain the logarithmic time domain slope between the two adjacent times;

[0104] The difference between the two adjacent logarithmic time domain slopes is calculated, to obtain the slope change;

[0105] The absolute value of the slope change is taken, to obtain the slope change absolute value;

[0106] All positions satisfying the first preset condition are screened out to form a local maximum position set of the slope change absolute value; wherein the first preset condition is that the previous slope change absolute value is less than the current slope change absolute value, and the current slope change absolute value is greater than or equal to the next slope change absolute value;

[0107] The position corresponding to the earliest time is selected from the local maximum position set as the first maximum position, and the time corresponding to the first maximum position is determined as the critical time.

[0108] The positive small constant needs to be determined in combination with the actual magnitude of the pelvic floor displacement data and the accuracy of the MRI device, and the specific value range is 0.1-0.5 millimeters (corresponding to the typical small change amount of pelvic floor tissue displacement):

[0109] If a 1.5T MRI device (displacement measurement accuracy is about 0.3mm) is used, the micro constant is 0.3mm; if a 3.0T MRI device (displacement measurement accuracy is about 0.1mm) is used, the micro constant is 0.1mm;

[0110] Value principle: Ensure that the result of the displacement difference value + the micro constant is greater than 0 (avoiding the meaninglessness of logarithm), and the constant value is much smaller than the normal pelvic floor displacement difference (usually 2-10mm), without interfering with the authenticity of the displacement trend.

[0111] Each time is only limited to the time corresponding to all image frames after the beginning time of the force stage and before the end time of the emptying stage, specifically excluding two types of time:

[0112] Time before the beginning time of the force stage (such as rest, contraction stage): no abdominal pressure is applied in this stage, the pelvic floor tissue has no active stress deformation, and the data is irrelevant to the critical time (mechanical turning point after abdominal pressure loading);

[0113] Time after the end of the emptying stage: the tissue has completed the emptying action, and the deformation tends to be stable, with no effective mechanical turning point information;

[0114] When operating, all frame times within the above limited range need to be selected from the time sequence of dynamic MRI first, and then subsequent logarithmic transformation is performed.

[0115] The adjacent two times specifically refer to the time corresponding to the image frames continuously collected by dynamic MRI, that is, the adjacent frame times arranged in turn according to the fixed acquisition interval of the device (such as 0.3 seconds / frame, 0.5 seconds / frame), and skipping intermediate frame selection interval times is not allowed. For example, if the device collects at an interval of 0.5 seconds / frame, the first frame time after the beginning time of the force stage is t1, the second frame is t2, and the third frame is t3. The adjacent times are only (t1, t2) and (t2, t3) such continuous frame combinations; specifically, the deformation of the pelvic floor tissue is a continuous time process, and continuous frame times can fully reflect the continuity of the slope change, avoiding information loss caused by interval frames.

[0116] The former and the latter in the first preset condition both specifically refer to one time point directly adjacent to the current time, rather than multiple times. Specifically, taking the current time t0 as an example, the former time is the frame time t-1 before t0 (continuously adjacent to t0), and the latter time is the frame time t+1 after t0 (continuously adjacent to t0); only when the absolute value of the slope change corresponding to t-1 is < the absolute value of the slope change corresponding to t0, and the absolute value of the slope change corresponding to t0 is ≥ the absolute value of the slope change corresponding to t+1, t0 is included in the local maximum position set; avoiding the complexity and misjudgment of the judgment standard caused by multi-neighborhood comparison (such as the former 2 and the latter 2).

[0117] To avoid misjudging the small amplitude slope change caused by noise as a local maximum, a minimum judgment threshold of the absolute value of the slope change needs to be set. Calculate the arithmetic mean of the absolute values of all slope changes in the limited range (from the beginning of the force stage to the end of the emptying stage), and record it as Savg; set the threshold value as 1.5-2 times of Savg (recorded as Sth), only when the absolute value of the slope change at a certain time is greater than or equal to Sth and the first preset condition is met, the time is included in the local maximum position set; for example, if Savg=0.2, Sth takes 0.3 (1.5 times), although the time when the absolute value of the slope change is 0.25 meets the condition of "small before large", it is not included in the local maximum set because it is lower than 0.3. It should be noted that 1.5T MRI equipment (image noise is slightly high) takes 2 times of Savg, and 3.0T equipment (low noise) takes 1.5 times of Savg, which balances noise filtering and effective signal retention.

[0118] The critical moment of judgment needs to meet the physiological mechanical properties of the pelvic floor tissue, and the verification standard is as follows:

[0119] The critical moment needs to be located within 0.5-5 seconds after the beginning of the force stage, that is, after the pelvic floor tissue is loaded by abdominal pressure, the elastic deformation usually lasts for about 0.5 seconds, and after more than 0.5 seconds, it enters the viscoelastic softening stage, and after 5 seconds, the deformation tends to be stable, and the critical moment beyond this range needs to be rechecked for data processing;

[0120] Before the critical moment, the logarithmic time domain slope of the displacement curve is usually stable at about 1.0 (approximate elastic deformation, displacement and time are linearly related) ; after the critical moment, the slope decreases to 0.3-0.8 (viscoelastic softening, displacement increases slowly), if the slope change does not meet this trend, the small constant or threshold needs to be adjusted.

[0121] In an embodiment of the present application, in the post-critical window, the logarithmic relationship between the displacement increment and time is weighted and fitted to obtain a second slope as a post-critical power-law softening index, which comprises:

[0122] Taking the critical moment as the starting point, a positive time interval and a positive continuous time window duration that avoid the critical transient state are set, and a time period from the critical moment plus the positive time interval that avoids the critical transient state to the critical moment plus the positive time interval that avoids the critical transient state plus the positive continuous time window duration is defined as the post-critical window.

[0123] The difference between each time in the post-critical window and the critical moment is taken as a logarithm to obtain the logarithmic time of the time relative to the critical moment.

[0124] The difference between the displacement value of the displacement curve at the time and the displacement value of the displacement curve at the critical moment is calculated, and the difference is taken as a logarithm after adding the small constant to obtain the logarithmic displacement increment of the time relative to the critical moment.

[0125] The ratio of the pixel area of each frame functional area to the pixel area of the functional area corresponding to the critical moment is calculated, and the ratio is taken as the weight corresponding to each frame;

[0126] A target function is constructed with the constant term and the to-be-solved slope as variables, which is the sum of the squares of the weight corresponding to each moment in the post-critical window, multiplied by the logarithmic displacement increment of the moment minus the constant term, and then minus the product of the to-be-solved slope and the logarithmic time of the moment;

[0127] The values of the constant term and the to-be-solved slope are adjusted so that the value of the target function reaches a minimum value, and a second slope is obtained, which is taken as the post-critical power-law softening index.

[0128] The time interval needs to match the transient fluctuation duration of the pelvic floor tissue after the critical moment, and the value range is 0.3-1 second, which is determined according to the following rules:

[0129] If the patient is a young group (≤45 years old, the pelvic floor tissue has good elasticity), the transient fluctuation is short, and 0.3-0.5 seconds are taken;

[0130] If the patient is an elderly group (>45 years old, the tissue elasticity degenerates), the transient fluctuation is longer, and 0.6-1 second is taken;

[0131] When operating, the stable starting point and the interval from the critical moment can be taken as the final value through pre-experiment observation: if the logarithmic displacement increment of the displacement curve has no obvious fluctuation (the difference between adjacent frames is <0.1) for 3 frames (according to the device acquisition interval, such as 0.5 seconds / frame, which is 1.5 seconds) after the critical moment, the stable starting point and the interval from the critical moment are taken as the final value.

[0132] The duration needs to meet the sufficient data amount and only contain the stable softening stage, and the value range is 2-4 seconds, with the following specific requirements:

[0133] The lower limit of the duration contains at least 5 frames of images (according to the minimum acquisition interval of 0.3 seconds / frame, 5 frames correspond to 1.5 seconds, and 2 seconds are taken to ensure the data amount), to avoid insufficient degrees of freedom during fitting;

[0134] The upper limit of the duration is not more than 1 / 2 of the interval from the critical moment to the end of the emptying stage, for example, if the interval from the critical moment to the end of the emptying stage is 6 seconds, the duration is at most 3 seconds, to prevent containing the stable displacement stage (non-softening stage) data after emptying;

[0135] Typical values include: 1.5T MRI device (acquisition interval 0.5 seconds / frame) takes 3 seconds (6 frames of data), and 3.0T MRI device (acquisition interval 0.3 seconds / frame) takes 2.4 seconds (8 frames of data).

[0136] The pixel area of the functional area is calculated according to the gray threshold and complete pixel determination, and the steps are as follows:

[0137] Take the average gray value of the functional area in the resting phase reference frame as the basis, and take 1 / 2 of the value as the threshold. Any pixel with a gray value ≥ the threshold in the functional area is considered a valid pixel.

[0138] If the center point of a pixel is located within the functional area boundary line (according to the functional area contour), it is included in the area regardless of whether the gray value meets the standard. If only part of the pixel is within the boundary (such as a blurred edge pixel), the proportion of the area within the boundary to the total area of the pixel needs to be calculated. If the proportion is ≥ 50%, it is included in the area, otherwise it is excluded.

[0139] Count the number of all valid pixels and multiply it by the actual area of a single pixel (such as a pixel spacing of 1.2 mm x 1.2 mm for an MRI device, then the area of a single pixel is 1.44 square millimeters) to get the actual pixel area of the functional area.

[0140] The gradient descent method is used to adjust the constant term and the to-be-solved slope, which specifically includes:

[0141] Initial value setting: the initial value of the constant term is taken as the average of the logarithmic displacement increments within the critical window, and the initial value of the to-be-solved slope is taken as 0.5 (a typical empirical value of the softening slope of the pelvic floor tissue).

[0142] Learning rate setting: the learning rate is taken as 0.01-0.05. If the target function decreases slowly, it is increased (such as 0.05), and if it oscillates, it is decreased (such as 0.01).

[0143] Iteration stopping condition: when the change in the target function is < 0.0001 after 100 consecutive iterations, or the number of iterations reaches 1000 (to prevent infinite iteration), the adjustment is stopped, and the current constant term and to-be-solved slope are taken.

[0144] Verification: if the correlation coefficient after fitting (linear correlation of logarithmic displacement increment and logarithmic time) ≥ 0.85, the fitting is considered valid; if < 0.85, the window length or the small constant needs to be adjusted again, and the fitting is performed again.

[0145] The second slope obtained by fitting (critical post-power law softening index) needs to meet the following physiological and data standards to be considered valid:

[0146] Numerical range: the softening index of normal pelvic floor tissue is usually 0.2-0.8. If the slope < 0.2, it means that the tissue softens very slowly (possibly due to data acquisition errors, such as insufficient abdominal pressure); if the slope > 0.8, it means that the tissue softens too quickly (possibly due to functional area division errors, such as including non-supporting tissue pixels), which needs to be checked again.

[0147] Residual error requirement: the minimum value of the objective function (i.e. the sum of squared fitting residuals) needs to be < 1 / 10 of the maximum value of the log displacement increment in the critical post-window, for example, if the maximum value of the log displacement increment is 2, the sum of squared residuals needs to be < 0.2, to ensure that the fitting error is small;

[0148] Clinical correlation: if the patient has a history of pelvic floor dysfunction (such as uterine prolapse), the softening index is usually > 0.5; if it is a healthy population, it is usually < 0.4, which can be used to assist in verifying the effectiveness of the slope.

[0149] Specific clinical definition of event indication, including:

[0150] Event time window: uniformly set to 24 months from the completion of MRI examination, only events occurring within this window are counted.

[0151] Event determination criteria (any one of the following can be recorded as event indication = 1):

[0152] Prolapse recurrence / progression: POP-Q staging is increased by ≥ 1 stage (e.g. from stage II to stage III) compared with MRI examination, confirmed by gynecological or pelvic floor surgeons through clinical examination;

[0153] Reoperation: secondary surgical treatment for pelvic floor dysfunction (such as uterine prolapse, anterior and posterior vaginal wall prolapse), based on surgical records;

[0154] Symptom aggravation: patients report a significant increase in pelvic discomfort symptoms (such as a feeling of heaviness, urinary incontinence), and the doctor confirms that it is related to the degradation of the pelvic support structure;

[0155] Event indication = 0: within 24 months of follow-up, none of the above events occurred, and there was no change in POP-Q staging or symptom aggravation.

[0156] Sample pair construction criteria, including:

[0157] Sample inclusion criteria:

[0158] Female patients aged 18-75 years;

[0159] Patients with diagnosed pelvic floor dysfunction (e.g. POP-Q staging stages I-IV) and completed supine dynamic MRI examination and calculation of critical post-power law softening index;

[0160] Can complete 24 months of follow-up (e.g. no serious underlying disease leading to loss of follow-up).

[0161] Sample exclusion criteria:

[0162] Patients with pelvic cavity malignant tumors and severe nervous system diseases (such as spinal cord injury);

[0163] Incomplete MRI data (e.g. missing stage, failed index calculation);

[0164] Patients who died or lost contact due to non-pelvic floor diseases during follow-up.

[0165] Sample size and balance requirements:

[0166] Total sample size of at least 100 cases (to ensure the model has enough data to learn);

[0167] The ratio of positive and negative samples is controlled between 1:1 and 1:3. If the number of event samples is small (e.g. only 20 cases), oversample positive samples (duplicate a small number of positive samples) or undersample negative samples (randomly delete part of the negative samples) to adjust the balance.

[0168] The initial value of the model parameters includes:

[0169] The initial value of the model intercept parameter is set to 0 (corresponding to no feature input, the risk probability is default 0.5, consistent with the initial neutral hypothesis);

[0170] The initial value of the model slope parameter is set to 0.1 (reference to the preliminary correlation trend between pelvic floor dysfunction risk and softening index, to avoid excessive initial value leading to optimization shock);

[0171] If the proportion of event indicator = 1 in the sample is high (e.g. > 40%), the intercept initial value can be adjusted to 0.2; if the proportion is low (< 20%), adjust to -0.2 to assist in optimizing fast convergence.

[0172] Natural exponential function definition: The base of the natural exponential function here is uniform as the mathematical constant e (approximately equal to 2.71828), which is the industry standard natural exponential base and does not require additional adjustment.

[0173] Specifically, the numerical precision control is that the linear prediction calculation retains 4 decimal places, the natural exponential calculation result retains 6 decimal places, and the final risk probability retains 3 decimal places (e.g. 0.325, i.e. 32.5%), which is convenient for clinicians to understand.

[0174] Sample weight processing of negative log-likelihood includes:

[0175] If the sample is unbalanced (e.g. positive sample proportion < 30%), different samples need to be assigned weights, and the adjusted target function calculation rule is:

[0176] Calculate sample weight: Set the number of positive samples as N1 and the number of negative samples as N0, positive sample weight = (N1+N0) / (2×N1), negative sample weight = (N1+N0) / (2×N0);

[0177] Weighted negative log-likelihood: Objective function =∑[sample weight × (event indicator × (-ln(risk probability)) + (1-event indicator) × (-ln(1-risk probability)))].

[0178] For example, if N1=30 and N0=70, then the positive sample weight = 100 / (2×30)≈1.667, and the negative sample weight = 100 / (2×70)≈0.714, balancing the contributions of the two types of samples to the objective function through the weights.

[0179] The specific algorithm and stopping conditions for parameter adjustment include:

[0180] Parameter adjustment algorithm: batch gradient descent method (suitable for small sample size, stable calculation), specific steps:

[0181] At each iteration, calculate the partial derivative of the objective function with respect to the intercept and slope using all training samples;

[0182] Update the parameters according to the new parameter value = old parameter value - learning rate × partial derivative;

[0183] Parameter settings:

[0184] Learning rate: fixed at 0.01 (balance optimization speed and stability, if the objective function fluctuates, reduce to 0.005, if the convergence is slow, increase to 0.02);

[0185] Stopping condition: after 50 consecutive iterations, the change in the objective function is <0.0001 (i.e., the difference between the objective functions of the last two iterations is less than one ten-thousandth), or the total number of iterations reaches 1000 (to prevent infinite iterations);

[0186] Optimization process monitoring: output the current objective function value and parameter value every 10 iterations to facilitate timely detection of abnormalities (such as an increase in the objective function, which requires checking the learning rate or initial value).

[0187] After the model training is completed, an independent test set (20%-30% of the total samples, not involved in parameter training) is needed to evaluate effectiveness, with the following core standards:

[0188] Main indicators: AUC value (area under the receiver operating characteristic curve) ≥0.75, the higher the AUC value, the stronger the model's ability to distinguish between event occurrence and non-occurrence;

[0189] Accuracy rate (number of correctly predicted samples / total test sample number) ≥0.7;

[0190] Recall rate (number of correctly predicted positive samples / actual positive sample number) ≥0.65 (to avoid missing high-risk patients);

[0191] If the evaluation is not up to standard (e.g. AUC < 0.75), the sample construction (e.g. supplementing samples) needs to be rechecked or the initial value of the parameter is adjusted, and then the training is performed again.

[0192] The above describes the embodiments of the present embodiment, but the present embodiment is not limited to the above-described specific embodiments, and the above-described specific embodiments are only illustrative but not restrictive, and a person of ordinary skill in the art can make many forms under the inspiration of the present embodiment, which all belong to the protection of the present embodiment.

Claims

1. A system for constructing a risk model for female pelvic floor dysfunction, characterized in that, include: The data acquisition module acquires dynamic magnetic resonance data of the supine position, including the resting, contraction, exertion, and emptying phases, and records the time information of each phase. The reference coordinate establishment module determines the pubococcygeal line in the midsagittal position, establishes fixed reference coordinates based on the pubococcygeal line registration data, generates the functional region corresponding to the pubococcygeal line in the resting phase, and maps it to each frame. The displacement feature extraction module obtains the relatively resting displacement field through registration within the reference coordinates and functional areas of each frame. It extracts the regional displacement features along the tail side of the pubococcygeal line and normalizes them by the length of the pubococcygeal line to obtain the displacement curve over continuous time. The critical moment detection module takes the start of the force application phase as the starting point and detects the position where the first slope of the displacement curve changes the most in the logarithmic time domain as the critical moment. The softening index calculation module performs a weighted fitting of the logarithmic relationship between displacement increment and time within the critical window, and uses the fitted second slope as the critical power-law softening index; wherein, the critical window is determined based on the critical moment. The risk prediction module constructs a risk prediction model for female pelvic floor dysfunction using the post-critical power-law softening index, and outputs individual risk probabilities, including: The input is the post-critical power-law softening exponent; The model training data consists of multiple sample pairs. Each sample pair contains the critical post-power-law softening exponent of the j-th sample and the event indicator of the j-th sample. The event indicator takes the value of zero or one, j is a positive integer representing the sample number, and the total number of samples is a positive integer. The linear predictor is defined as the product of the model intercept parameter, the model slope parameter, and the critical post-power-law softening exponent of the j-th sample. A logic function is defined as a function that maps any real number to the range between zero and one. The logic function is calculated by dividing one by one and raising the negative linear predictor of the natural exponential function to the power of one. The individual risk probability of the j-th sample is calculated using the aforementioned logical function; The objective function is defined as the negative log-likelihood based on the individual risk probability. Specifically, it is calculated by multiplying the event indication of each sample by the negative logarithm of the individual risk probability of that sample, and adding the negative logarithm of the difference between the event indication and the individual risk probability of that sample to obtain the sample parameter for each sample. The sample parameters of all samples are then summed to obtain the objective function. Adjust the values ​​of the model intercept parameter and the model slope parameter to minimize the value of the objective function, thereby obtaining the optimal model intercept parameter and the optimal model slope parameter; For any individual to be evaluated, the linear predictor is calculated using the corresponding critical post-power-law softening exponent, the optimal model intercept parameter, and the optimal model slope parameter. The risk probability of the individual is then obtained through the aforementioned logic function and used as the model output.

2. The system for constructing a risk model for female pelvic floor dysfunction according to claim 1, characterized in that, Acquire dynamic magnetic resonance imaging (MRI) data in the supine position, including the resting, contraction, exertion, and emptying phases, and record the time information for each phase, including: Determine the start and end times of the rest, contraction, exertion and emptying phases to form time information for each phase; Dynamic magnetic resonance data is generated by acquiring image frames corresponding to the resting, contraction, exertion, and emptying phases at fixed time intervals.

3. The system for constructing a risk model for female pelvic floor dysfunction according to claim 2, characterized in that, The pubococcygeal line is determined in the midsagittal plane. Based on the pubococcygeal line registration data, a fixed reference coordinate system is established. The functional regions corresponding to the pubococcygeal line during the resting phase are generated and mapped to each frame, including: In the image frame during the resting phase, the lower edge of the pubis and the end point of the coccyx are determined, and the straight line connecting the lower edge of the pubis and the end point of the coccyx is taken as the pubococcygeal line. Using the image frame corresponding to the start of the resting state as the reference image frame, for any image frame at any time, a rigid body transformation including rotation and translation is constructed. By adjusting the rotation and translation parameters, the lower edge of the pubis and the end point of the coccyx in the image frame are aligned with the lower edge of the pubis and the end point of the coccyx in the reference image frame, respectively, so as to establish a fixed reference coordinate. After rotating the direction vector from the lower edge of the pubis to the end of the coccyx counterclockwise by 90 degrees, normalize it to obtain the unit normal vector pointing to the caudal half-plane. On the reference image frame, the set of pixels located at the tail of the pubococcygeal line and whose unit normal vector and the projection of the line connecting each point in the region to the lower edge of the pubis are less than or equal to zero are defined as the functional region. By using the inverse transformation of rigid body transformation, the functional regions on the reference image frame are mapped to the corresponding image frames at each time point, thus obtaining the functional regions in each image frame.

4. The system for constructing a risk model for female pelvic floor dysfunction according to claim 3, characterized in that, Within the reference coordinates and functional regions of each frame, the relatively resting displacement field is obtained through registration. Regional displacement features are extracted along the tail side of the pubococcygeal line and normalized by the length of the pubococcygeal line to obtain continuous-time displacement curves, including: Use the grayscale of the reference image frame as the reference image grayscale; For each image frame at each time point, based on rigid body transformation, the image frame is converted into a rigid body registration image aligned to the reference coordinates; By using non-rigid registration, the grayscale difference between the rigid registration image and the reference image frame is minimized. At the same time, a smoothing term for the deformation field is added to determine the deformation field from the pixel in the reference coordinates to the corresponding position at each time step. The displacement field in the relative resting phase is obtained by subtracting the deformation field from the pixel position in the reference coordinates. The functional regions of each frame and the caudal unit direction pointing to the caudal direction are determined, and the distance between the lower edge of the pubis and the end point of the coccyx is defined as the length of the pubococcygeal line. Within each frame's functional region, calculate the average value of the projection of the displacement field onto the tail side unit direction after rigid body transformation, and normalize this average value by dividing it by the length of the tail line to obtain the single-frame region displacement corresponding to each moment. By associating each moment with the corresponding single-frame region displacement, a displacement curve over continuous time is formed.

5. The system for constructing a risk model for female pelvic floor dysfunction according to claim 4, characterized in that, Starting from the moment of the force application phase, the critical moment is defined as the location of the first maximum change in the slope of the displacement curve detected in the logarithmic time domain, including: Set a positive, small constant; Add the small constant to the difference between each moment and the moment when force is applied, and then take the logarithm to obtain the logarithmic time corresponding to each moment; The logarithmic displacement at each moment is obtained by adding the small constant to the difference between the displacement value of the displacement curve at each moment and the displacement value of the displacement curve at the moment of force application, and then taking the logarithm. Calculate the difference in logarithmic displacement between two adjacent time points, divide it by the difference in logarithmic time between these two adjacent time points, and obtain the logarithmic time domain slope between the two adjacent time points. Calculate the difference between the slopes of two adjacent logarithmic time domains to obtain the slope change; The absolute value of the slope change is obtained by taking the absolute value of the slope change. All positions that satisfy the first preset condition are selected to form a set of local maxima of the absolute value of the slope change; wherein, the first preset condition is: the absolute value of the previous slope change is less than the absolute value of the current slope change, and the absolute value of the current slope change is greater than or equal to the absolute value of the next slope change. The earliest position corresponding to the local maximum position is selected from the set of local maximum positions as the first maximum position, and the time corresponding to the first maximum position is determined as the critical time.

6. The system for constructing a risk model for female pelvic floor dysfunction according to claim 5, characterized in that, Within the critical window, a weighted fit is performed on the logarithmic relationship between displacement increment and time. The second slope obtained from the fit is used as the power-law softening exponent after the critical period, including: Starting from the critical moment, a positive time interval for avoiding the critical transient and a positive duration of the continuous time window are set. The time period from the critical moment plus the positive time interval for avoiding the critical transient to the critical moment plus the positive duration of the continuous time window is defined as the post-critical window. Take the logarithm of the difference between each time point within the critical window and the critical time point to obtain the logarithmic time of that time point relative to the critical time point; Calculate the difference between the displacement value of the displacement curve at this moment and the displacement value of the displacement curve at the critical moment, add the small constant to the difference and take the logarithm to obtain the logarithmic displacement increment at this moment relative to the critical moment. Calculate the ratio of the pixel area of ​​the functional region in each frame to the pixel area of ​​the functional region at the critical moment, and use this ratio as the weight for each frame. Construct an objective function with a constant term and the slope to be determined as variables. The objective function is the weight corresponding to each time point within the critical window, multiplied by the logarithmic displacement increment at that time point minus the constant term, and then subtracted by the sum of the squares of the product of the slope to be determined and the logarithmic time at that time point. Adjust the values ​​of the constant term and the slope to be determined so that the objective function reaches its minimum value, thus obtaining the second slope. Use the second slope as the critical post-power-law softening exponent.

Images