An algorithm for automatically identifying and quantifying endometrial motility in magnetic resonance images

Abstract

The present invention discloses an algorithm for automatically identifying and quantifying endometrial peristalsis in magnetic resonance images, comprising the following steps: step one, constructing a 3D-UNET network model; step two, identifying endometrial exfoliation movement; step three, identifying peristaltic wave conduction in the uterine juncture; step four, quantifying the amplitude of endometrial peristalsis; and step five, determining the direction of endometrial peristalsis. The present invention realizes automatic identification of endometrial exfoliation movement and peristaltic wave conduction in the uterine juncture based on the 3D-UNET network model, can automatically quantify the amplitude of endometrial peristalsis, and determine the direction of endometrial peristalsis. Compared with the existing method relying on naked eye observation by radiologists, the identification result of the present invention has better objectivity and repeatability, can obtain the standard range of female uterine movement-related parameters by expanding the amount of data, provide a reference standard for clinical application, and expand the application of endometrial peristalsis as a uterine function evaluation standard in clinical practice.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to an algorithm for automatically identifying and quantifying endometrial motility in magnetic resonance images. Background Art

[0002] Like other organs such as the heart and intestines, the uterus has its own inherent movement patterns. Endometrial peristalsis is caused by contractions of the subuterine myometrium and smooth muscle layers, creating a pressure gradient within the uterine cavity and causing the endometrium to move in a pattern similar to intestinal peristalsis. Endometrial peristalsis exhibits distinct patterns throughout a woman's menstrual cycle and plays an important physiological role. During ovulation, the endometrium peristsals at a high frequency from the cervix to the fundus, helping sperm reach the fundus for fertilization. The luteal phase is the period with the lowest endometrial peristalsis frequency, creating a quiet uterine environment and promoting embryo implantation. During menstruation, the uterus peristsals at a high amplitude from the fundus to the cervix, expelling shed endometrial structures and preventing endometriosis. Therefore, regular endometrial peristalsis plays a crucial role throughout a woman's menstrual cycle.

[0003] However, clinical studies have found that gynecological diseases such as uterine fibroids, adenomyosis, and endometriosis can disrupt the peristaltic function of the patient's endometrium to varying degrees, leading to a series of clinical symptoms such as dysmenorrhea, infertility, embryo implantation failure, spontaneous abortion, and premature birth. Therefore, observing and evaluating the peristaltic function of the female endometrium is of great clinical significance.

[0004] Traditional monitoring of endometrial motility uses intrauterine pressure sensors and ultrasound. Intrauterine sensors can detect changes in intrauterine pressure and measure the amplitude and frequency of endometrial motility. However, since this is an invasive procedure, the stimulation of foreign bodies can induce uterine stress and contraction, so the measured results will have human errors and are almost never used. Transabdominal ultrasound can dynamically and continuously observe endometrial motility and is non-invasive. However, due to the pressure of the abdominal probe, it can also induce uterine contractions or inhibit existing endometrial motility, making it impossible to accurately detect the inherent patterns of endometrial motility.

[0005] In recent years, thanks to technological innovations in magnetic resonance imaging (MRI), MRI has gradually become the primary method for observing endometrial motility. MRI can rapidly and dynamically acquire images of endometrial motility, generating endometrial motility movies that allow for intuitive observation of endometrial motility. Due to MRI's high soft tissue contrast, it can clearly visualize the uterine synaptic zone. Unlike abdominal ultrasound, which requires the use of a pressure probe that increases intrauterine pressure and disrupts the inherent regularity of endometrial motility, MRI dynamic movie sequences have gradually become the primary method for observing endometrial motility. MRI dynamic sequences intermittently acquire and generate endometrial motility movies. Visual observation by experienced radiologists can only assess the direction and frequency of endometrial motility. However, this assessment method is subjective and cannot yield objective, quantitative indicators related to endometrial motility. Therefore, there is an urgent need for an algorithm that can automatically and objectively assess and quantify endometrial motility. Software based on this algorithm can be developed to objectively assess women's uterine motility and monitor the recovery of uterine motility after treatment. Summary of the Invention

[0006] The object of the present invention is to provide an algorithm for automatically identifying and quantifying endometrial motility in magnetic resonance images, so as to solve the problems raised in the above background technology.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an algorithm for automatically identifying and quantifying endometrial peristalsis in magnetic resonance images, comprising the following steps: step 1, constructing a 3D-UNET network model; step 2, identifying endometrial exfoliation movement; step 3, identifying peristaltic wave conduction in the uterine synaptic zone; step 4, quantifying the amplitude of endometrial peristalsis; step 5, determining the direction of endometrial peristalsis;

[0008] In the above step 1, an initial 3D-UNET network model is established, a data set of endometrial contour magnetic resonance images is obtained, and the data set is used to train the 3D-UNET network model to obtain a final model;

[0009] In the above step 2, the final model obtained in step 1 is used to identify the endometrial exfoliation movement, and an endometrial contour is fixed as a reference contour to monitor the contour grayscale intensity at different times; when the peristaltic wave reaches a certain point on the reference contour, the endometrium becomes thinner, and the signal intensity of the endometrial pixel at this point decreases; when the peristaltic wave leaves this point, the signal intensity recovers. Using these characteristics, it is possible to determine whether the pixel intensity on the endometrial contour has a periodic pattern based on the data, that is, whether the endometrium has periodic peristalsis, and calculate parameters such as the number of peristalsis;

[0010] In the above step 3, the final model obtained in step 1 is used to identify the peristaltic wave conduction of the uterine juncture. For a certain point on the uterine juncture, the intensity is minimum when the peristaltic wave arrives and recovers when the peristaltic wave leaves. This characteristic can be used to determine whether the uterine juncture has a periodic pattern and calculate the number of peristaltic waves. Specifically, the following steps are included:

[0011] 3.1 Endometrial segmentation: Use the final model obtained in step 1 to segment the endometrium at each moment and extract the contour;

[0012] 3.2 Statistical analysis of the intensity of the junction zone at different times: Based on the corresponding endometrial contour, expand outward by 3 pixels and calculate the average intensity of these three points, which is regarded as the intensity of the junction zone;

[0013] 3.3 Calculate the peristaltic cycle using FFT transformation: Use the method mentioned in step 2.4 to obtain the final peristaltic number result. At the same time, fix the monitoring range length and align the slices according to the uterine fundus to reduce the influence of the morphological changes of other organs on the horizontal position of the uterus.

[0014] In the above step 4, the endometrial peristalsis amplitude is calculated, that is, the difference in endometrial thickness is calculated, and the range of motion of the upper and lower sides of the endometrial contour is calculated;

[0015] In the above step five, the direction of peristalsis on the image and the orientation of the uterine fundus are determined to obtain the direction of uterine peristalsis.

[0016] Preferably, in the step one, the specific method for training the 3D-UNET network model is: dividing the data set into a training set, a validation set and a test set, using the training set to train the initial 3D-UNET network model, and after the training is completed, substituting the validation set into the trained 3D-UNET network model, adjusting the model parameters and selecting the model with the highest accuracy, and substituting the test set into the model with the highest accuracy to test the model recognition accuracy. If the recognition accuracy is greater than or equal to the set threshold, the model is used as the final model. If the recognition accuracy is less than the set threshold, the model training is performed again until the recognition accuracy is greater than or equal to the set threshold.

[0017] Preferably, the step 2 specifically includes the following steps:

[0018] 2.1 Endometrial segmentation: Import the scanned female pelvic magnetic resonance image into the final model obtained in step 1, segment the endometrium, obtain the uterine ROI area, and obtain the endometrial contour;

[0019] 2.2 Determine the reference contour: Rotate the ROI area of the uterus so that the line connecting the uterine fundus and the cervix is approximately parallel to the horizontal axis of the plane. Obtain the minimum bounding rectangle of the obtained contour to get the rotation angle. After obtaining the angle, rotate the three-dimensional image around the coronal axis by the corresponding angle. Subsequently, take each point on the reference contour as a sampling point;

[0020] 2.3 Statistically analyze the pixel intensities of each sampling point at different times: Monitor the gray-scale intensities of the contour at different times. The gray-scale values of each sampling point at different times t, where 0 < t < 60. To improve accuracy, the actual evaluation is the average of the gray-scale values of three pixels in the vertical direction of each sampling point;

[0021] 2.4 Calculate the peristaltic period using the fft transform: Each sampling point corresponds to 60 gray-scale values, that is, a gray-scale sequence with a length of 60. Perform fft analysis on the gray-scale sequence corresponding to each sampling point. Perform fft transform on this sequence to obtain the frequency and the corresponding amplitude. Assume there are N points on the sample reference contour, corresponding to N sequences. Calculate each sequence in this way to obtain N corresponding frequencies, and then take the mode among them as the final peristaltic count result.

[0022] Preferably, in step 4, it specifically includes the following steps:

[0023] 4.1 Difference in endometrial thickness: During the sampling time, the thicknesses corresponding to each point on the line connecting the cervix and the uterine fundus are changing. Record the maximum and minimum values of the thickness at each position. The difference between the maximum and minimum values is regarded as the peristaltic amplitude at that position. At the same time, calculate the maximum amplitude, minimum amplitude, and average amplitude;

[0024] 4.2 Movement ranges of the upper and lower sides of the endometrial contour respectively: During the sampling time, extract the endometrial contour respectively. Record the longitudinal coordinate positions of each point on the upper and lower sides of the contour. Extract the maximum and minimum values. The difference between the maximum and minimum values is regarded as the peristaltic amplitude at that position. At the same time, calculate the maximum amplitude, minimum amplitude, and average amplitude.

[0025] Preferably, in step 5, it specifically includes the following steps:

[0026] 5.1 Determine the direction of peristalsis on the sample image: Process the baseband intensity map obtained when detecting the wave conduction of the junction zone. Calculate the slope of each line segment in the map. If the slope is less than 0, it means that the peristaltic direction on the image is from left to right; if the slope is greater than 0, it means that the peristaltic direction on the image is from right to left; when there is no obvious peristalsis, the slope is close to 0;

[0027] 5.2 Determine the direction of the uterine fundus: Draw a horizontal line through the center D of the circumscribed rectangle of the endometrium, intersecting each side of the body at points A and B. Take the midpoint C of AB. If the distance CD is greater than 5 pixels, if point D is to the right of point C, the fundus is on the right; if point D is to the left of point C, the fundus is on the left. If the distance CD is less than or equal to 5 pixels, calculate the average thickness of the left and right sides of the endometrium. The side with the greater average thickness is the fundus.

[0028] 5.3 Obtaining the direction of uterine peristalsis: If the peristalsis direction on the image is from left to right and the fundus is on the right, the uterine peristalsis direction is from the cervix to the fundus; if the peristalsis direction on the image is from left to right and the fundus is on the left, the uterine peristalsis direction is from the fundus to the cervix.

[0029] Compared with the existing technology, the beneficial effects of the present invention are: the present invention realizes automatic recognition of endometrial exfoliation movement and uterine junctional peristaltic wave conduction based on the 3D-UNET network model, can automatically quantify the amplitude of endometrial peristalsis, and judge the direction of endometrial peristalsis. Compared with the existing method relying on naked eye observation by radiologists, the recognition results of the present invention have better objectivity and repeatability, and can obtain the standard range of female uterine movement-related parameters by expanding the amount of data, providing a reference standard for clinical application, and at the same time expanding the application of endometrial peristalsis in clinical practice as a standard for uterine function evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 Schematic diagram of the endometrial outline;

[0031] Figure 2 Schematic diagram of the minimum contiguous rectangle of the endometrial outline;

[0032] Figure 3 This is a schematic diagram of the endometrial contour after rotation;

[0033] Figure 4 This is a schematic diagram of the sampling points of the endometrial contour;

[0034] Figure 5 The gray intensity statistical graph of the endometrial contour at different moments;

[0035] Figure 6 It is a line graph of the pixel intensity change at the sampling point;

[0036] Figure 7 It is the frequency line graph after FFT transformation;

[0037] Figure 8 It is a schematic diagram of peristaltic wave conduction in the direction of cervix-uterine fundus or uterine fundus-uterine cervix;

[0038] Figure 9 This is a schematic diagram of the endometrial outline after expanding outward by 3 pixels;

[0039] Figure 10 Combined signal strength graph at different moments;

[0040] Figure 11 It is a line graph of the changes in endometrial thickness;

[0041] Figure 12 It is a line graph of the longitudinal peristalsis position of the endometrium;

[0042] Figure 13 It is a line graph of endometrial motility amplitude;

[0043] Figure 14 is a flowchart for processing the baseband intensity map obtained when combining the wave conduction of the band;

[0044] Figure 15 It is an auxiliary judgment diagram for the position of the uterine fundus;

[0045] Figure 16 It is the algorithm flow chart of the present invention. DETAILED DESCRIPTION

[0046] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0047] See also Figure 1-16 The present invention provides an embodiment of an algorithm for automatically identifying and quantifying endometrial peristalsis in magnetic resonance images, comprising the following steps: Step 1, constructing a 3D-UNET network model; Step 2, identifying endometrial exfoliation movement; Step 3, identifying peristaltic wave conduction in the uterine juncture; Step 4, quantifying the amplitude of endometrial peristalsis; Step 5, determining the direction of endometrial peristalsis;

[0048] In the above step one, an initial 3D-UNET network model is established, an endometrial contour magnetic resonance image dataset is obtained, and the 3D-UNET network model is trained using the dataset to obtain a final model. The specific method is: the dataset is divided into a training set, a validation set, and a test set, the initial 3D-UNET network model is trained using the training set, and after the training is completed, the validation set is substituted into the trained 3D-UNET network model, the model parameters are adjusted and the model with the highest accuracy is selected, and the test set is substituted into the model with the highest accuracy to test the model recognition accuracy. If the recognition accuracy is greater than or equal to the set threshold, the model is used as the final model. If the recognition accuracy is less than the set threshold, the model training is performed again until the recognition accuracy is greater than or equal to the set threshold.

[0049] In the above step two, the final model obtained in step one is used to identify the endometrial stripping movement. A fixed endometrial contour is used as the reference contour to monitor the contour gray intensity at different times. When the peristaltic wave reaches a certain point on the reference contour, the endometrium thins and the signal intensity of the endometrial pixels here decreases. When the peristaltic wave leaves this point, the signal intensity recovers. Using these characteristics, it is possible to determine whether there is a periodic pattern in the pixel intensity on the endometrial contour, that is, whether there is periodic peristalsis of the endometrium, and calculate parameters such as the number of peristalsis. The specific steps are as follows:

[0050] 2.1 Endometrial segmentation: Import the female pelvic magnetic resonance image collected by scanning into the final model obtained in step one, segment the endometrium, obtain the uterine ROI region, and get the endometrial contour (as Figure 1 shown);

[0051] 2.2 Determine the reference contour: Rotate the uterine ROI region so that the line connecting the fundus and the cervix is approximately parallel to the horizontal axis of the plane, and obtain the minimum bounding rectangle of the obtained contour (as Figure 2 shown) to obtain the rotation angle. After obtaining the angle, rotate the three-dimensional image around the coronal axis by the corresponding angle (as Figure 3 shown), and then take each point on the reference contour as a sampling point (as Figure 4 shown);

[0052] 2.3 Statistically analyze the pixel intensity of each sampling point at different times: Monitor the contour gray intensity at different times (as Figure 5 shown). The gray value of each sampling point at different times t, where 0 < t < 60. To improve accuracy, the actual evaluation is the average value of the gray values of three pixels in the vertical direction of each sampling point;

[0053] 2.4 Calculate the peristaltic period using fft transformation: Each sampling point corresponds to 60 gray values, that is, a gray sequence with a length of 60. Perform fft analysis on the gray sequence corresponding to each sampling point (as Figure 6 shown), perform fft transformation on this sequence to obtain the frequency and the corresponding amplitude (as Figure 7 shown). Assume that there are N points on the sample reference contour, corresponding to N sequences. Calculate each sequence in this way to obtain N corresponding frequencies, and then take the mode among them as the final peristaltic count result;

[0054] In the above step three, the final model obtained in step one is used to identify the conduction of the peristaltic wave in the uterine junction zone (as Figure 8 shown). For a certain point on the uterine junction zone, when the peristaltic wave arrives, the intensity is the smallest, and when the peristaltic wave leaves, the intensity recovers. Using this characteristic, it is possible to determine whether there is a periodic pattern in the uterine junction zone and calculate its peristaltic count. The specific steps are as follows:

[0055] 3.1 Endometrial segmentation: Use the final model obtained in step 1 to segment the endometrium at each moment and extract the contour;

[0056] 3.2 Statistical analysis of the intensity of the combined band at different times: Based on the corresponding endometrial contour, expand outward by 3 pixels (such as Figure 9 ), and the average intensity of these three points was calculated and regarded as the intensity of the binding band (as shown in Figure 10 shown);

[0057] 3.3 Calculate the peristaltic cycle using FFT transformation: Use the method mentioned in step 2.4 to obtain the final peristaltic number result. At the same time, fix the monitoring range length and align the slices according to the uterine fundus to reduce the influence of the morphological changes of other organs on the horizontal position of the uterus.

[0058] In the above step 4, the endometrial peristalsis amplitude is calculated, that is, the difference in endometrial thickness is calculated, and the range of motion of the upper and lower sides of the endometrial contour is calculated. Specifically, the following steps are included:

[0059] 4.1 Endometrial thickness variation: The thickness of each point on the line connecting the cervix to the uterine fundus varies during the sampling period. The maximum and minimum values ​​of the thickness at each location are recorded. The difference between the maximum and minimum values ​​is considered the peristaltic amplitude at that location (e.g. Figure 11 Calculate the maximum amplitude, minimum amplitude and mean amplitude at the same time;

[0060] 4.2 The range of motion of the upper and lower sides of the endometrial contour: The endometrial contour was extracted during the sampling time, and the longitudinal coordinate positions of each point on the upper and lower sides of the contour were recorded. The maximum and minimum values ​​were extracted, and the difference between the maximum and minimum values ​​was regarded as the peristaltic amplitude at that position. The maximum amplitude, minimum amplitude, and mean amplitude were calculated at the same time (e.g. Figure 12 、 13 As shown, Figure 12 The horizontal axis represents the position from the uterine fundus to the cervix, and the vertical axis represents the longitudinal position. Figure 13 The horizontal axis represents the position from the fundus to the cervix, and the vertical axis represents the difference between the upper and lower limits, i.e., the amplitude);

[0061] In the above step 5, the direction of the peristalsis on the image and the orientation of the uterine fundus are determined to obtain the direction of uterine peristalsis. Specifically, the following steps are included:

[0062] 5.1 Determine the direction of peristalsis on the sample image: Process the baseband intensity image obtained when detecting the wave conduction of the binding band, and calculate the slope of each line segment in the image. If the slope is less than 0, it means that the peristalsis direction is from left to right on the image; if the slope is greater than 0, it means that the peristalsis direction is from left to right on the image; when there is no obvious peristalsis, the slope is close to 0; (e.g. Figure 14As shown, the baseband intensity obtained when detecting the wave conduction of the combined band Figure 14 a) as an example, clustering with k=3 is performed to obtain Figure 14 b), yes Figure 14 b) Perform contour detection and find the minimum adjacent rectangle of the contour, impose some height and width restrictions, and filter out Figure 14 The line segments in b) are filtered as follows Figure 14 c);

[0063] 5.2 Determine the direction of the uterine fundus: Draw a horizontal line through the center D of the circumscribed rectangle of the endometrium, intersecting the two sides of the body at points A and B. Take the midpoint C of AB. When the distance CD is greater than 5 pixels, if point D is to the right of point C, the uterine fundus is on the right; if point D is to the left of point C, the uterine fundus is on the left. When the distance CD is less than or equal to 5 pixels, calculate the average thickness of the left and right sides of the endometrium. The side with the greater average thickness is the uterine fundus. (e.g. Figure 15 shown)

[0064] 5.3 Obtaining the direction of uterine peristalsis: If the peristalsis direction on the image is from left to right and the fundus is on the right, the uterine peristalsis direction is from the cervix to the fundus; if the peristalsis direction on the image is from left to right and the fundus is on the left, the uterine peristalsis direction is from the fundus to the cervix.

[0065] Based on the above, the advantages of the present invention are that when the invention is used, the present invention can automatically identify the endometrial peristalsis waves appearing on the magnetic resonance image, can automatically determine the direction of endometrial peristalsis, and can automatically quantify the number of endometrial peristalsis, which will replace the existing method relying on naked eye observation by radiologists, and the evaluation results are more objective and repeatable; the present invention is based on an algorithm platform that can process large-scale data, which helps to obtain the standard range of female uterine movement-related parameters by expanding the amount of data, and provide a reference standard for clinical application; at the same time, the present invention can also quantify uterine movement parameters that cannot be observed by the naked eye of the physician, that is, the amplitude of endometrial peristalsis. Under some pathological conditions, such as adenomyosis, uterine fibroids, etc., the amplitude of endometrial peristalsis will be affected, thereby affecting the relevant physiological functions of women. By realizing the automatic quantification of endometrial peristalsis amplitude parameters, the application of endometrial peristalsis in clinical practice as a uterine function evaluation standard can be further expanded.

[0066] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be included therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

Claims

1. An algorithm for automatically identifying and quantifying endometrial motility in magnetic resonance images, comprising the following steps: Step 1: construct a 3D-UNET network model; Step 2: identify endometrial exfoliation movement; Step 3: identify peristaltic wave conduction in the uterine junction zone; Step 4: quantify the amplitude of endometrial peristalsis; Step 5: determine the direction of endometrial peristalsis; characterized by: In the above step 1, an initial 3D-UNET network model is established, a data set of endometrial contour magnetic resonance images is obtained, and the data set is used to train the 3D-UNET network model to obtain a final model; In the above step 2, the final model obtained in step 1 is used to identify the endometrial exfoliation movement, and an endometrial contour is fixed as a reference contour to monitor the contour grayscale intensity at different times; when the peristaltic wave reaches a certain point on the reference contour, the endometrium becomes thinner, and the signal intensity of the endometrial pixel at this point decreases; when the peristaltic wave leaves this point, the signal intensity recovers. Using these characteristics, it is possible to determine whether the pixel intensity on the endometrial contour has a periodic pattern based on the data, that is, whether the endometrium has periodic peristalsis, and calculate parameters such as the number of peristalsis; In the above step 3, the final model obtained in step 1 is used to identify the peristaltic wave conduction of the uterine juncture. For a certain point on the uterine juncture, the intensity is minimum when the peristaltic wave arrives and recovers when the peristaltic wave leaves. This characteristic can be used to determine whether the uterine juncture has a periodic pattern and calculate the number of peristaltic waves. Specifically, the following steps are included: 3.1 Endometrial segmentation: Use the final model obtained in step 1 to segment the endometrium at each moment and extract the contour; 3.2 Statistical analysis of the intensity of the junction zone at different times: Based on the corresponding endometrial contour, expand outward by 3 pixels and calculate the average intensity of these three points, which is regarded as the intensity of the junction zone; 3.3 Calculate the peristaltic cycle using FFT transformation: Use the method mentioned in step 2.4 to obtain the final peristaltic number result. At the same time, fix the monitoring range length and align the slices according to the uterine fundus to reduce the influence of the morphological changes of other organs on the horizontal position of the uterus. In the above step 4, the endometrial peristalsis amplitude is calculated, that is, the difference in endometrial thickness is calculated, and the range of motion of the upper and lower sides of the endometrial contour is calculated; In the above step five, the direction of peristalsis on the image and the orientation of the uterine fundus are determined to obtain the direction of uterine peristalsis.

2. The algorithm for automatically identifying and quantifying endometrial motility in magnetic resonance images according to claim 1, characterized in that: In the step one, the specific method for training the 3D-UNET network model is: dividing the data set into a training set, a validation set and a test set, using the training set to train the initial 3D-UNET network model, and after the training is completed, substituting the validation set into the trained 3D-UNET network model, adjusting the model parameters and selecting the model with the highest accuracy, substituting the test set into the model with the highest accuracy to test the model recognition accuracy; if the recognition accuracy is greater than or equal to the set threshold, the model is used as the final model; if the recognition accuracy is less than the set threshold, the model training is performed again until the recognition accuracy is greater than or equal to the set threshold.

3. The algorithm for automatically identifying and quantifying endometrial motility in magnetic resonance images according to claim 1, characterized in that: The step 2 specifically includes the following steps: 2.1 Endometrial segmentation: Import the magnetic resonance images of the female pelvic cavity collected by scanning into the final model obtained in Step 1, segment the endometrium, obtain the uterine ROI region, and get the endometrial contour; 2.2 Determine the reference contour: Rotate the uterine ROI region so that the line connecting the fundus and the cervix is approximately parallel to the horizontal axis of the plane, and find the minimum bounding rectangle of the obtained contour to obtain the rotation angle; After obtaining the angle, rotate the three-dimensional image around the coronal axis by the corresponding angle, and then take each point on the reference contour as a sampling point; 2.3 Statistically analyze the pixel intensities of each sampling point at different times: Monitor the gray-scale intensities of the contour at different times, and the gray-scale values of each sampling point at different times t, where 0 < t < 60. To improve accuracy, the actual evaluation is the average value of the gray-scale values of three pixels in the vertical direction of each sampling point; 2.4 Calculate the peristaltic period using the fft transform: Each sampling point corresponds to 60 gray-scale values, that is, a gray-scale sequence with a length of 60. Perform fft analysis on the gray-scale sequence corresponding to each sampling point, perform fft transform on this sequence to obtain the frequency and the corresponding amplitude. Assume that there are N points on the sample reference contour, corresponding to N sequences. Calculate each sequence in this way to obtain N corresponding frequencies, and then take the mode among them as the final peristaltic frequency result.

4. The algorithm for automatically identifying and quantifying endometrial motility in magnetic resonance images according to claim 1, characterized in that: In Step 4, it specifically includes the following steps: 4.1 Endometrial thickness difference: During the sampling time, the thicknesses corresponding to each point on the line connecting the cervix to the fundus are changing. Record the maximum and minimum values of the thickness at each position, and the difference between the maximum and minimum values is regarded as the peristaltic amplitude at that position; At the same time, calculate the maximum amplitude, minimum amplitude, and average amplitude; 4.2 Movement ranges of the upper and lower sides of the endometrial contour respectively: During the sampling time, extract the endometrial contour respectively, record the longitudinal coordinate positions of each point on the upper and lower sides of the contour, extract the maximum and minimum values, and the difference between the maximum and minimum values is regarded as the peristaltic amplitude at that position; At the same time, calculate the maximum amplitude, minimum amplitude, and average amplitude.

5. The algorithm for automatically identifying and quantifying endometrial motility in magnetic resonance images according to claim 1, characterized in that: In Step 5, it specifically includes the following steps: 5.1 Determine the direction of peristalsis on the sample image: Process the baseband intensity map obtained when detecting the wave conduction of the junction zone, calculate the slope of each line segment in the map. If the slope is less than 0, it means that the peristaltic direction on the image is from left to right; If the slope is greater than 0, it means that the peristaltic direction on the image is from left to right; When there is no obvious peristalsis, the slope is close to 0; 5.2 Determine the direction of the uterine fundus: Draw a horizontal line through the center D of the circumscribed rectangle of the endometrium, which intersects the two sides of the body at points A and B respectively. Take the midpoint C of AB. When the distance of CD > 5 pixels, if point D is on the right side of point C, the fundus is on the right side; if point D is on the left side of point C, the fundus is on the left side; When the distance of CD <= 5 pixels, calculate the average thickness of the left and right sides of the endometrium, and the side with the greater average thickness is the fundus; 5.3 Obtain the uterine peristaltic direction: If the peristaltic direction on the image is from left to right and the fundus is on the right side, the uterine peristaltic direction is from the cervix to the fundus; If the peristaltic direction on the image is from left to right and the fundus is on the left side, the uterine peristaltic direction is from the fundus to the cervix.

Images