A sperm flagellum movement parameter calculation method and a computer readable storage medium

Abstract

The application relates to the technical field of computer-aided semen analysis, and discloses a sperm flagellum movement parameter calculation method and a computer readable storage medium, which comprises the following steps: based on a preset sampling frequency, a plurality of sample images are extracted from a flagellum movement video of sperm to be analyzed; for each sample image, the sample image is input into an improved U-Net network to obtain a binary flagellum segmentation mask, after a morphological thinning algorithm is used to extract a flagellum skeleton, a parameterized flagellum skeleton curve is obtained; a convolution kernel in the improved U-Net network is improved into a convolution kernel with a constrained deformable offset; based on the parameterized flagellum skeleton curve, the flagellum length, the flagellum tangent angle, the flagellum propagation speed, the flagellum curvature, the flagellum amplitude, the flagellum frequency, the flagellum wavelength, the flagellum bending amplitude and the flagellum bending index are calculated. Through the convolution kernel with the constrained deformable offset, the receptive field is increased, the segmentation continuity is ensured, and the movement parameters of the flagellum can be quantitatively calculated.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer-aided semen analysis, in particular to a sperm flagellum motion parameter calculation method and a computer-readable storage medium. BACKGROUND

[0002] In medical and biological research, the motion characteristics of sperm are key indicators for assessing their functional status and fertility. According to the World Health Organization (WHO), sperm flagellum motion characteristics are important factors for diagnosing potential male infertility, and quantifying flagellum motion parameters (such as amplitude, frequency, and waveform) not only helps to fundamentally understand the generation mechanism of sperm motion, but also plays a crucial role in diagnosing male infertility and improving artificial reproductive technology.

[0003] Sperm flagellum motion is a highly dynamic process that requires high-throughput data processing. In order to capture the dynamic process of rapid sperm flagellum motion, the imaging frame rate is generally not less than 30 frames per second. In addition, taking the clinical diagnosis of infertility as an example, the World Health Organization requires analyzing at least 200 sperm per patient, i.e. 6000 sperm images per second for each patient to be processed to quantify flagellum motion parameters. Traditionally, sperm motion analysis requires the observation of experienced embryologists under a microscope, which is extremely time-consuming (2-3 days of image annotation per patient) and prone to inconsistency and subjectivity. Therefore, in order to fully exploit the rich information contained in sperm flagellum motion behavior, it is urgent to develop automated, high-throughput quantitative calculation and analysis tools.

[0004] The technical requirement of high-throughput fully automatic sperm flagellum motion parameter quantitative calculation and analysis is to completely and continuously segment the entire sperm flagellum. Incomplete or broken flagellum segmentation will lead to errors in flagellum parameter quantification. However, this task faces the following challenges: (1) the diameter of sperm flagellum is about 1 μm, which corresponds to 5-10 pixel sizes in the image (less than 0.01% of the entire image), with low imaging contrast and difficult to distinguish from the background; (2) sperm swings the flagellum in three-dimensional space, and part of the flagellum may be located outside the depth of field of the microscope, causing the local disappearance of the sperm flagellum to be invisible, further increasing the difficulty of segmenting the entire flagellum.

[0005] The existing analysis method is mainly divided into two categories: one is to use basic image processing and combine prior knowledge (such as geometric topology), and to enhance the perception performance by modeling the topological characteristics of the curve structure, but this method is not only sensitive to the threshold height, but also depends on the background and the anisotropy of the sperm, which leads to the discontinuity of the sperm flagellum segmentation result or even cannot be segmented; the second is to use standard convolutional neural networks such as U-Net, FC-DenseNet, Deeplabv3+ and the like, but experiments show that they are difficult to accurately and continuously segment the sperm flagellum; the reason is that the standard convolutional shape is fixed and cannot learn the irregular and time-varying sperm flagellum structure. In order to more accurately learn the complex and variable curve characteristics, deformable convolution is introduced, such as DUNet, DCU-Net, but due to the unconstrained offset, it may pay too much attention to the background area and ignore the narrow area of the sperm flagellum, leading to discontinuous segmentation results and fragmentation.

[0006] In summary, the existing method of using convolutional network to segment the sperm flagellum for quantitative calculation and analysis of sperm flagellum motion parameters cannot accurately distinguish the background area and the flagellum area due to the fixed kernel shape of the standard convolution which cannot adapt to the irregular shape of the flagellum, and the arbitrary offset of the deformable convolution which may lead to segmentation result fragmentation, resulting in discontinuous segmentation results and low accuracy. SUMMARY

[0007] Therefore, the technical problem to be solved by the present application is to overcome the difficulty of the convolution kernel in the prior art to adapt to the irregular shape of the flagellum, so as to accurately distinguish the flagellum and the background, and further to accurately identify the flagellum motion parameters.

[0008] To solve the above technical problems, the present application provides a sperm flagellum motion parameter calculation method, comprising:

[0009] Based on a preset sampling frequency, a plurality of sample images are extracted from a sperm flagellum motion video to be analyzed;

[0010] For each sample image, it is input into an improved U-Net network to obtain a corresponding binary flagellum segmentation mask; the convolution kernel in the improved U-Net network is a convolution kernel with a constrained deformable offset;

[0011] A morphological thinning algorithm is used to extract a flagellum skeleton from the binary flagellum segmentation mask;

[0012] The flagellum skeleton is parameterized to obtain a parameterized flagellum skeleton curve;

[0013] Based on the parameterized flagellum skeleton curve, the flagellum length, flagellum tangent angle, flagellum propagation speed and flagellum curvature are calculated;

[0014] Based on the flagellum curvature, the flagellum amplitude, the flagellum frequency, the flagellum wavelength, the flagellum bending amplitude and the flagellum bending index are calculated.

[0015] Preferably, the obtaining of the improved U-Net network comprises:

[0016] The normal convolution kernel in the standard U-Net network is linearly expanded along the x-axis and y-axis directions respectively to obtain an initial state of a continuous convolution kernel;

[0017] A variable deformable offset with a constraint is introduced to the initial state of the continuous convolution kernel to obtain the continuous convolution kernel;

[0018] The continuous convolution kernel is integrated into the standard U-Net network to obtain the improved U-Net network.

[0019] Preferably, the convolution process of the continuous convolution kernel is represented as:

[0020] ;

[0021] wherein, represents an axis feature of an output feature map at a pixel position ; represents an axis feature of an output feature map at a pixel position ; represents a feature of an output feature map at a pixel position ; represents a feature of an output feature map at a pixel position ; represents an i-th grid in a convolution kernel, represents a preset weight corresponding to an i-th grid in a convolution kernel, represents a preset modulation coefficient corresponding to an i-th grid in a convolution kernel; , , , , represents a total number of grids in the convolution kernel;

[0022] represents a deformable offset of an i-th grid in the convolution kernel in the x-axis direction, and the expression is ; ;

[0023] represents a deformable offset of an i-th grid in the convolution kernel in the y-axis direction, and the expression is ; ;

[0024] represents a preset constraint,​​ Represents the convolution kernel Axis direction The offset value of the grid, Represents the convolution kernel Axis direction The offset value of the grid.

[0025] Preferably, improving the training of the U-Net network includes:

[0026] Live sperm were selected from semen samples using the water bath-swim-up method and used as sperm samples. A 40x objective lens was used to capture a video of the sperm flagellar movement.

[0027] The flagellar motion video is converted into frame sequence images according to the preset sampling frequency, and the outline of the sperm flagella on each image is marked using the polygon tool to obtain the sperm flagella annotation dataset;

[0028] Each frame of the sperm flagellum annotation dataset is converted into a grayscale image and assigned to three color channels: R, G, and B, to obtain the corresponding three-channel pseudo-color image;

[0029] The sperm flagella annotation dataset is divided into a training set and a test set, and the training set data is expanded by image rotation, translation and flipping;

[0030] Based on the expanded data set, the cross entropy loss function and Adam optimizer are used to update the parameters of the improved U-Net network until the preset number of training times is reached to obtain the trained improved U-Net network.

[0031] Preferably, the flagellar skeleton is parameterized to obtain a parameterized flagellar skeleton curve, which is expressed as:

[0032] ;

[0033] in, Represents the starting point to point of the parameterized flagellar skeleton curve The arc length.

[0034] Preferably, the flagellar length, flagellar tangential angle, flagellar propagation velocity and flagellar curvature are calculated based on the parameterized flagellar skeleton curve, including:

[0035] Discretize the parameterized flagellar skeleton curve, obtain multiple sampling points, calculate the Euclidean distance between adjacent sampling points, and sum them to obtain the flagellar length , expressed as:

[0036] ;

[0037] , denotes the total number of sampling points, denotes the x-axis coordinate and y-axis coordinate of the i-th sampling point, respectively; denotes the x-axis coordinate and y-axis coordinate of the i-th sampling point, respectively; denotes the x-axis coordinate and y-axis coordinate of the i-th sampling point, respectively;

[0038] denotes the first derivative of the x-axis coordinate and y-axis coordinate at the arc length s, denotes the first derivative of the x-axis coordinate and y-axis coordinate at the arc length s, denotes the first derivative of the x-axis coordinate and y-axis coordinate at the arc length s, denotes the first derivative of the x-axis coordinate and y-axis coordinate at the arc length s, denotes the first derivative of the x-axis coordinate and y-axis coordinate at the arc length s, denotes the first derivative of the x-axis coordinate and y-axis coordinate at the arc length s, ;

[0039] denotes the first derivative of the x-axis coordinate and y-axis coordinate at the arc length s, denotes the first derivative of the x-axis coordinate and y-axis coordinate at the arc length s, ; denotes the first derivative of the x-axis coordinate and y-axis coordinate at the arc length s, denotes the first derivative of the x-axis coordinate and y-axis coordinate at the arc length s,

[0040] denotes the first derivative of the x-axis coordinate and y-axis coordinate at the arc length s, denotes the first derivative of the x-axis coordinate and y-axis coordinate at the arc length s, denotes the first derivative of the x-axis coordinate and y-axis coordinate at the arc length s, ; denotes the first derivative of the x-axis coordinate and y-axis coordinate at the arc length s, denotes the first derivative of the x-axis coordinate and y-axis coordinate at the arc length s, denotes the first derivative of the x-axis coordinate and y-axis coordinate at the arc length s,

[0041] denotes the first derivative of the x-axis coordinate and y-axis coordinate at the arc length s, denotes the first derivative of the x-axis coordinate and y-axis coordinate at the arc length s,

[0042] denotes the first derivative of the x-axis coordinate and y-axis coordinate at the arc length s, denotes the first derivative of the x-axis coordinate and y-axis coordinate at the arc length s, denotes the first derivative of the x-axis coordinate and y-axis coordinate at the arc length s, denotes the first derivative of the x-axis coordinate and y-axis coordinate at the arc length s, ; denotes the first derivative of the x-axis coordinate and y-axis coordinate at the arc length s, denotes the first derivative of the x-axis coordinate and y-axis coordinate at the arc length s,

[0043] denotes the first derivative of the x-axis coordinate and y-axis coordinate at the arc length s, denotes the first derivative of the x-axis coordinate and y-axis coordinate at the arc length s, denotes the first derivative of the x-axis coordinate and y-axis coordinate at the arc length s, denotes the first derivative of the x-axis coordinate and y-axis coordinate at the arc length s, ;

[0044] denotes the first derivative of the x-axis coordinate and y-axis coordinate at the arc length s, denotes the first derivative of the x-axis coordinate and y-axis coordinate at the arc length s, , expressed as: ;

[0045] The flagellar curvature of the parameterized flagellar skeleton curve is integrated along the arc length to obtain the flagellar bending amplitude , expressed as: ;

[0046] Based on flagellar curvature The mean and standard deviation , get the flagellar bending index , expressed as: .

[0047] Preferably, it also includes a flagellar amplitude-based Flagellar wavelength Calculation of flagellar amplitude-wavelength ratio , expressed as: .

[0048] Preferably, it also includes calculating the flagellar tangential angle Arc length The derivative of , as the flagellar wavelength index , expressed as: .

[0049] This embodiment further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the method for calculating sperm flagellar motion parameters as described above are implemented.

[0050] The above technical solution of the present invention has the following beneficial effects compared with the prior art:

[0051] The sperm flagella motion parameter calculation method described in the present invention introduces a constrained deformable offset into the convolution kernel of the standard U-Net network to obtain an improved U-Net network, and extracts a binary flagella segmentation mask from the sample image, effectively overcoming the limitations of standard convolution and deformable convolution in segmenting sperm flagella. By linearizing the convolution kernel, limiting the offset range, and adopting an offset accumulation strategy, the continuity of the segmentation result is ensured while increasing the receptive field, thereby achieving accuracy and continuity in flagella segmentation.

[0052] Meanwhile, the flagella skeleton is extracted from the binary flagella segmentation mask, and parameterization is performed on the flagella skeleton, the curvature of the parameterized flagella skeleton curve is calculated, and various parameters including flagella length, flagella tangent angle, flagella propagation speed, flagella amplitude, flagella frequency, flagella wavelength, flagella bending amplitude and flagella bending index are quantitatively calculated by taking the curvature as a basic feature, so that the calculation process is simplified, the use range of the calculation method is expanded, the sperm flagella motion parameter analysis throughput and efficiency are greatly improved, and the motion parameter quantitative extraction and statistical analysis from single cell single time to group level time sequence are realized. BRIEF DESCRIPTION OF DRAWINGS

[0053] In order to make the content of the present application more easily understood, the present application will be further described in detail below according to specific embodiments of the present application and in conjunction with the drawings, in which:

[0054] Figure 1 is a step flow chart of the sperm flagella motion parameter calculation method provided by the present application;

[0055] Figure 2 is a flow chart of the sperm flagella motion parameter quantitative analysis method;

[0056] Figure 3 is a structure diagram of the semantic segmentation backbone network U-Net model based on an encoder-decoder;

[0057] Figure 4 is a working principle diagram of a continuous convolution kernel;

[0058] Figure 5 is a schematic diagram of the difficulty of accurate continuous segmentation of sperm flagella;

[0059] Figure 6 is a schematic diagram of the limitations of standard convolution and deformable convolution;

[0060] Figure 7 is a comparative schematic diagram of sperm flagella segmentation results;

[0061] Figure 8 is a sperm flagella segmentation result rendering diagram;

[0062] Figure 9 is a comparative diagram of sperm flagella motion parameter statistical analysis data. DETAILED DESCRIPTION

[0063] The present application will be further described below in conjunction with the drawings and specific embodiments, so that those skilled in the art can better understand the present application and implement it, but the embodiments are not limiting to the present application.

[0064] Reference Figure 1 As shown in FIG, the flow chart of the method for calculating sperm flagellar motion parameters provided by the present invention includes the following steps:

[0065] S101: extracting multiple frames of sample images from the flagellar movement video of the sperm to be analyzed based on a preset sampling frequency;

[0066] S102: For each frame of sample image, input it into an improved U-Net network to obtain a corresponding binary flagellum segmentation mask; the convolution kernel in the improved U-Net network is a convolution kernel that introduces a constrained deformable offset;

[0067] S103: Extract the flagellar skeleton from the binary flagellar segmentation mask using a morphological thinning algorithm;

[0068] S104: Parameterize the flagellar skeleton to obtain a parameterized flagellar skeleton curve, which is expressed as:

[0069] ;

[0070] in, Represents the starting point to point of the parameterized flagellar skeleton curve The arc length.

[0071] S105: Calculate flagellar length, flagellar tangential angle, flagellar propagation velocity and flagellar curvature based on the parameterized flagellar skeleton curve;

[0072] S106: Based on the flagellar curvature, the flagellar amplitude, flagellar frequency, flagellar wavelength, flagellar bending amplitude and flagellar bending index are calculated.

[0073] Specifically, in this embodiment, the acquisition of the U-Net network is improved, including:

[0074] The ordinary convolution kernel in the standard U-Net network is linearly expanded along the x-axis and y-axis respectively to obtain the initial state of the continuous convolution kernel;

[0075] Introduce a constrained deformable offset to the initial continuous convolution kernel to obtain a continuous convolution kernel;

[0076] The continuous convolution kernel is integrated into the standard U-Net network to obtain an improved U-Net network.

[0077] Among them, the convolution process of the continuous convolution kernel is expressed as:

[0078] ;

[0079] in, Indicates the output feature map at pixel location at Axis features, Indicates the output feature map at pixel location at Axis features; Indicates the output feature map at pixel location The characteristics of the place, Represents the convolution kernel grids, Represents the convolution kernel The preset weights corresponding to the grids, Represents the convolution kernel The preset modulation coefficient corresponding to each grid; , Represents the total number of grids in the convolution kernel;

[0080] Represents the convolution kernel Axis direction The deformable offset of the grid is expressed as ;

[0081] Represents the convolution kernel Axis direction The deformable offset of the grid is expressed as ;

[0082] Represents a preset constraint, Represents the convolution kernel Axis direction The offset value of the grid, Represents the convolution kernel Axis direction The offset value of the grid.

[0083] This embodiment uses the trained improved U-Net network to obtain the binary flagella segmentation mask corresponding to the sample image, and improves the training process of the U-Net network, including:

[0084] Live sperm were selected from semen samples using the water bath-swim-up method and used as sperm samples. A 40x objective lens was used to capture a video of the sperm flagellar movement.

[0085] The flagellar motion video is converted into frame sequence images according to the preset sampling frequency, and the outline of the sperm flagella on each image is marked using the polygon tool to obtain the sperm flagella annotation dataset;

[0086] Each frame of the sperm flagellum annotation dataset is converted into a grayscale image and assigned to three color channels: R, G, and B, to obtain the corresponding three-channel pseudo-color image;

[0087] The sperm flagellum annotation dataset is divided into a training set and a test set, and the training set data is expanded through image rotation, translation and flipping;

[0088] Based on the expanded dataset, the parameters of the improved U-Net network are updated using the cross-entropy loss function and the Adam optimizer until the preset training number is reached, and the trained improved U-Net network is obtained.

[0089] The sperm flagellum motion parameter calculation method introduces a constrained deformable offset into the convolution kernel of the standard U-Net network, obtains an improved U-Net network, extracts a binary flagellum segmentation mask from a sample image, effectively overcomes the limitations of standard convolution and deformable convolution in sperm flagellum segmentation, and ensures the continuity of the segmentation result while increasing the receptive field by using a linearized convolution kernel, limiting the offset range and using an offset accumulation strategy, thereby realizing the accuracy and continuity of flagellum segmentation.

[0090] Specifically, in step S105, based on the parameterized flagellum skeleton curve, the flagellum length, flagellum tangent angle, flagellum propagation speed and flagellum curvature are calculated, including:

[0091] S105-1: Discretize the parameterized flagellum skeleton curve to obtain a plurality of sampling points, calculate the Euclidean distance between adjacent sampling points, and sum up to obtain the flagellum length , which is expressed as:

[0092] ;

[0093] , , wherein n represents the total number of sampling points, and respectively represent the x-axis coordinate and the y-axis coordinate corresponding to the i-th sampling point;

[0094] S105-2: Calculate the first derivative of the x-axis coordinate and the y-axis coordinate at the arc length , and calculate the inverse tangent function of the ratio of the first derivative of the y-axis coordinate to the first derivative of the x-axis coordinate as the flagellum tangent angle , which is expressed as: ;

[0095] S105-3: Based on the change rate of the cumulative length of the flagellum curve at adjacent time points, the flagellum propagation speed is calculated, which is expressed as: ; , wherein L(t) represents the cumulative arc length of the flagellum curve at t.​​ denotes a preset time interval;

[0096] S105-4: first and second derivatives of the x-axis coordinate and the y-axis coordinate at the arc length s are calculated to obtain the flagellum curvature , which is denoted as: ; and denote the second derivative of the x-axis coordinate and the y-axis coordinate at the arc length s, respectively.

[0097] Specifically, in step S106, based on the flagellum curvature, the flagellum amplitude, the flagellum frequency, the flagellum wavelength, the flagellum bending amplitude and the flagellum bending index are calculated, including:

[0098] S106-1: a plurality of pairs of local maxima and local minima of the flagellum curvature are obtained as the wave peaks and the wave troughs , respectively; the sum of the vertical distances between all adjacent wave peaks and wave troughs is calculated and averaged to obtain the flagellum amplitude , which is denoted as: ; denotes the vertical distance between adjacent wave peaks and wave troughs, denotes the total number of pairs of wave peaks and wave troughs;

[0099] S106-2: based on the ratio between the number of times of sign change of the flagellum curvature at the arc length s per unit time and the unit time , the flagellum frequency is calculated, which is denoted as: ;

[0100] S106-3: based on the ratio between the propagation speed and the flagellum frequency, the flagellum wavelength is calculated, which is denoted as: ;

[0101] S106-4: the flagellum bending amplitude is obtained by integrating the flagellum curvature along the arc length of the parameterized flagellum skeleton curve, which is denoted as: ;

[0102] S106-5: based on the mean value and the standard deviation of the flagellum curvature , the flagellum bending index is obtained, which is denoted as: .

[0103] Based on the above embodiments, in this embodiment, based on the flagellum amplitude and the flagellum wavelength​​ The ratio of the amplitude wavelength of the flagellum is calculated , and is expressed as: .

[0104] Based on the above embodiment, in this embodiment, the tangential angle of the flagellum is calculated The derivative of the arc length is taken as the flagellum wavelength index , and is expressed as: .

[0105] Meanwhile, the present application extracts the flagellum skeleton from the binary flagellum segmentation mask and parameterizes it, calculates the curvature of the parameterized flagellum skeleton curve, and uses the curvature as a basic feature to quantitatively calculate a plurality of parameters including flagellum length, flagellum tangential angle, flagellum propagation speed, flagellum amplitude, flagellum frequency, flagellum wavelength, flagellum bending amplitude, and flagellum bending index, simplifying the calculation process, expanding the use range of the calculation method, enabling the analysis of non-periodic complex flagellum movement patterns, providing a general tool for characterizing diversified flagellum behavior, and realizing quantitative extraction and statistical analysis of movement parameters from single-cell single-time to population-level time series, greatly improving the throughput and efficiency of sperm flagellum movement parameter analysis.

[0106] Based on the above embodiment, in the present application, the improved U-Net network of the present application is used as a sperm flagellum segmentation network model, and the sperm flagellum movement parameter calculation method provided by the present application is used to calculate the flagellum movement parameters of the sperm sample, as shown in Figure 2 , which is a flowchart of the quantitative analysis method of sperm flagellum movement parameters, and specifically includes:

[0107] S201: Obtain sperm movement video and establish sperm flagellum data set;

[0108] According to the guidance of the World Health Organization's Semen Examination and Processing Laboratory Manual, the sperm sample is screened for live sperm by the "water bath-upstream method", and sperm movement video is collected under a microscope with a 40x objective lens;

[0109] The sperm movement video is converted into a frame sequence image according to a preset sampling frequency, and the sperm flagellum contour is marked using a polygon tool to form a sperm flagellum annotation data set;

[0110] The frame sequence image is used as the input of the network model training process;

[0111] S202: Preprocess and expand the sperm flagellum annotation data set;

[0112] For each sperm flagellum image in the sperm flagellum annotation dataset, it is converted into a gray image, and then assigned to three color channels of red (R), green (G) and blue (B) to form a three-channel pseudo-color image to eliminate color errors caused by different microscope settings, lighting conditions and white balance during microscopic imaging;

[0113] The sperm flagellum annotation dataset is divided into a training set and a test set; image rotation, translation, flipping and other data augmentation operations are performed on the training set to increase data diversity, simulate different observation conditions, and normalize all images;

[0114] S203: Construct a semantic segmentation network based on an encoder-decoder structure;

[0115] Referring to Figure 3 , it is an encoder-decoder-based semantic segmentation backbone network U-Net model structure diagram;

[0116] The encoder extracts image features through multi-layer downsampling, and the decoder restores the original resolution through multi-layer upsampling to segment the sperm flagellum contour and finally output the sperm flagellum segmentation result;

[0117] The encoder and the decoder adopt a symmetrical multi-layer structure;

[0118] S204: Construct a continuous convolution kernel and integrate it into the above semantic segmentation network;

[0119] The ordinary convolution kernel is linearly expanded along the x-axis and y-axis directions to form the initial state of the continuous convolution kernel, which expands the receptive field and adapts to different flagellum orientations through linearization operation;

[0120] Introduce a deformable offset and , respectively, represent the offset of the kth convolution kernel grid in the x-axis and y-axis directions, , is the total number of convolution kernel grids in the x-axis and y-axis directions;

[0121] In order to maintain the connectivity of the segmentation result, the deformable offset is constrained, and the offset value is limited to the range of [-1, 1], which ensures that the offset difference between adjacent grids does not exceed 1 pixel, thereby maintaining the connectivity of the segmentation result in space;

[0122] Through the offset accumulation method, the offset of each grid is recursively calculated, and the offset of the grid far from the center is recursively obtained from the offset of the previous grid, so that the continuous convolution kernel can adapt to local curvature and orientation changes, which is represented as:

[0123] ;

[0124] ;

[0125] The continuous convolution kernel is integrated into the encoder-decoder network to obtain the sperm flagellum segmentation model;

[0126] Reference Figure 4 As shown in the figure, it is the working principle diagram of the continuous convolution kernel; the calculation process of the continuous convolution is expressed as:

[0127] ;

[0128] in, and The output and input feature maps are at positions The characteristics of the place, For the convolution kernel grid coordinates, is the corresponding weight, is the deformable offset in the y-axis direction, is the corresponding modulation coefficient;

[0129] Reference Figure 5 The figure shows the technical difficulties in accurately and continuously segmenting sperm flagella; Figure 6 Figure 2 shows the limitations of standard and deformable convolution. This paper proposes a continuous convolution module that effectively overcomes the limitations of standard and deformable convolution in sperm flagellum segmentation by introducing a constrained deformable offset. The fixed kernel shape of standard convolution is difficult to adapt to the irregular shape of flagella, while the arbitrary offset of deformable convolution may cause segmentation results to be broken. The continuous convolution kernel linearizes the convolution kernel, limits the offset range, and adopts an offset accumulation strategy. This ensures the continuity of the segmentation results while increasing the receptive field, achieving accurate and continuous flagellum segmentation.

[0130] S205: Using the training set to train the sperm flagellum segmentation model, using the cross entropy loss function and Adam optimizer, to obtain a trained sperm flagellum segmentation network model; using the test set to evaluate the model performance and calculate various accuracy indicators;

[0131] S206: using the trained sperm flagellum segmentation network model, processing the sperm image sequence to be segmented frame by frame to obtain a binary flagellum segmentation mask sequence;

[0132] S207: extracting the skeleton of each flagellum segmentation mask, and using a morphological thinning algorithm to obtain a flagellum skeleton with a single pixel width;

[0133] The skeleton curve is parameterized as , is the normalized arc length along the skeleton curve;

[0134] S208: Calculate the curvature of sperm flagellum by parameterizing the skeleton curve coordinates The calculation formula is:

[0135] ;

[0136] wherein, , and , respectively represent , the first derivative and the second derivative of .

[0137] S209: Calculate sperm flagellum movement parameters based on the curvature of sperm flagellum, including flagellum length, flagellum amplitude, flagellum frequency, flagellum wavelength, flagellum tangent angle, flagellum propagation speed, flagellum bending arc, flagellum amplitude wavelength ratio, flagellum wave form index, flagellum bending index, respectively represented as:

[0138] ① Flagellum length : Discretize the skeleton curve, calculate the Euclidean distance between adjacent sampling points and sum them up, is the number of sampling points on the skeleton curve.

[0139] ② Flagellum amplitude : Find the local maximum (peak) and minimum (valley) points of the curvature , calculate the vertical distance between adjacent peaks and valleys and average them, is the vertical distance between adjacent peaks and valleys (local extreme points of curvature ), represents the total number of peaks and valleys.

[0140] ③ Flagellum frequency : Select the arc length position, count the number of curvature sign changes (turns) at a specific arc length position in unit time .

[0141] ④ Flagellum wavelength : Where the propagation wave speed is estimated by tracking the movement of the curvature extreme point over time.

[0142] ⑤ Flagellum tangent angle :

[0143] ⑥ Flagellum propagation speed :​ ,in is the cumulative arc length of the flagellar curve at time t, is the time interval.

[0144] ⑦Flagellar curvature : , for curvature Integrate along the arc length.

[0145] ⑧ Flagellar amplitude-wavelength ratio :

[0146] ⑨Flagellar waveform index : , the derivative of the tangential angle with respect to the arc length.

[0147] ⑩ Flagellar bending index : ,in, , The curvature The mean and standard deviation of .

[0148] S210: Repeat steps S206-S209 to calculate the motion parameters of each sperm flagellum in each frame of the frame sequence image of the sperm motion video, and then analyze the change pattern of the parameters over time.

[0149] Reference Figure 7 The following is a comparative diagram of the results of sperm flagella segmentation; Figure 8 The following is a rendering of the sperm flagellum segmentation result; refer to Figure 9 As shown, it is a comparison chart of statistical analysis data of some motion parameters of sperm flagella. After obtaining flagellar motion parameters using the method of the present invention, the flagellar motion parameters of different groups (for example, normal or infertile) can be statistically tested and data mined in combination with clinical pathological data to reveal the potential association between abnormal flagellar motion and disease. At the same time, the present invention organically combines sperm flagella segmentation, skeleton extraction, parameter calculation and other links to form a set of fully automated analysis processes, and has been successfully applied to the analysis of large-scale sperm samples, realizing the quantitative extraction and statistical analysis of motion parameters from single cell single moment to population-level time series, greatly improving the throughput and efficiency of sperm flagellar motion parameter analysis, and has important scientific significance and application value in basic research on reproductive biology, diagnosis of male infertility, assisted reproduction, etc., providing a quantitative analysis tool for further studying the mechanism of sperm movement and assisting in the clinical diagnosis of male infertility, and providing technical support for the diagnosis and treatment of sperm motility dysfunction.

[0150] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the method for calculating sperm flagellar motion parameters as described above are implemented.

[0151] The sperm flagella motion parameter calculation method described in the present invention introduces a constrained deformable offset into the convolution kernel of the standard U-Net network to obtain an improved U-Net network, and extracts a binary flagella segmentation mask from the sample image, effectively overcoming the limitations of standard convolution and deformable convolution in segmenting sperm flagella. By linearizing the convolution kernel, limiting the offset range, and adopting an offset accumulation strategy, the continuity of the segmentation result is ensured while increasing the receptive field, thereby achieving accuracy and continuity in flagella segmentation. At the same time, the present invention extracts the flagellar skeleton from the binary flagellar segmentation mask and parameterizes it, calculates the curvature of the parameterized flagellar skeleton curve, and uses the curvature as the basic feature to quantitatively calculate multiple parameters including flagellar length, flagellar tangential angle, flagellar propagation speed, flagellar amplitude, flagellar frequency, flagellar wavelength, flagellar bending amplitude and flagellar bending index, which simplifies the calculation process and expands the scope of application of the calculation method, enabling it to analyze non-periodic complex flagellar motion patterns, providing a general tool for characterizing diverse flagellar behaviors, and realizing the quantitative extraction and statistical analysis of motion parameters from single-cell single moment to population-level time series, greatly improving the throughput and efficiency of sperm flagellar motion parameter analysis.

[0152] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0153] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0154] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the flow Figure 1 one or more flows and / or blocks Figure 1 one or more blocks or steps of the functions specified in the flow

[0155] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flow Figure 1 one or more flows and / or blocks Figure 1 one or more blocks or steps of the functions specified in the flow

[0156] Obviously, the above-mentioned embodiments are only examples for clearly illustrating the present application, and are not intended to limit the embodiments. Based on the above description, other different forms of changes or variations can be made by those skilled in the art. Here, it is not necessary and also impossible to enumerate all the embodiments. The obvious changes or variations derived therefrom are still within the protection scope of the present application.

Claims

1. A method for calculating sperm flagellar motion parameters, characterized in that: include: Extracting multiple frames of sample images from the flagellar movement video of the sperm to be analyzed based on a preset sampling frequency; For each frame of sample image, input it into the improved U-Net network to obtain the corresponding binary flagella segmentation mask; The convolution kernel in the improved U-Net network is a convolution kernel that introduces a constrained deformable offset. Specifically, the method includes: linearly expanding the ordinary convolution kernel in the standard U-Net network along the x-axis and y-axis directions to obtain the initial state of the continuous convolution kernel; introducing a constrained deformable offset to the initial state continuous convolution kernel to obtain a continuous convolution kernel; integrating the continuous convolution kernel into the standard U-Net network to obtain the improved U-Net network; wherein the convolution process of the continuous convolution kernel is expressed as: ； in, Indicates the output feature map at pixel location at Axis features, Indicates the output feature map at pixel location at Axis features; Indicates the output feature map at pixel location The characteristics of the place, Represents the convolution kernel A grid, Represents the convolution kernel The preset weights corresponding to the grids, Represents the convolution kernel The preset modulation coefficient corresponding to each grid; , Represents the total number of grids in the convolution kernel; Represents the convolution kernel Axis direction The deformable offset of the grid is expressed as ; Represents the convolution kernel Axis direction The deformable offset of the grid is expressed as ; Represents a preset constraint, Represents the convolution kernel Axis direction The offset value of the grid, Represents the convolution kernel Axis direction The offset value of the grid; The flagellar skeleton is extracted from the binary flagellar segmentation mask using a morphological thinning algorithm; Parameterize the flagellar skeleton to obtain a parameterized flagellar skeleton curve; Based on the parameterized flagellar skeleton curve, the flagellar length, flagellar tangential angle, flagellar propagation speed and flagellar curvature are calculated; Based on the flagellar curvature, the flagellar amplitude, flagellar frequency, flagellar wavelength, flagellar bending amplitude and flagellar bending index are calculated.

2. sperm flagella motion parameter calculation method according to claim 1, is characterized in that, Improvements to U-Net network training, including: Live sperm were selected from semen samples using the water bath-swim-up method and used as sperm samples. A 40x objective lens was used to capture a video of the sperm flagellar movement. The flagellar motion video is converted into frame sequence images according to the preset sampling frequency, and the outline of the sperm flagella on each image is marked using the polygon tool to obtain the sperm flagella annotation dataset; Each frame of the sperm flagellum annotation dataset is converted into a grayscale image and assigned to three color channels: R, G, and B, to obtain the corresponding three-channel pseudo-color image; The sperm flagella annotation dataset is divided into a training set and a test set, and the training set data is expanded by image rotation, translation and flipping; Based on the expanded data set, the cross entropy loss function and Adam optimizer are used to update the parameters of the improved U-Net network until the preset number of training times is reached to obtain the trained improved U-Net network.

3. sperm flagella motion parameter calculation method according to claim 1, is characterized in that, Parameterize the flagellar skeleton and obtain the parameterized flagellar skeleton curve, which is expressed as: ； in, Represents the starting point to point of the parameterized flagellar skeleton curve The arc length.

4. sperm flagella motion parameter calculation method according to claim 3, is characterized in that, Based on the parameterized flagellar skeleton curve, the flagellar length, flagellar tangential angle, flagellar propagation speed and flagellar curvature are calculated, including: Discretize the parameterized flagellar skeleton curve, obtain multiple sampling points, calculate the Euclidean distance between adjacent sampling points, and sum them to obtain the flagellar length , expressed as: ； , Indicates the total number of sampling points, and Respectively represent The x-axis coordinate and y-axis coordinate corresponding to each sampling point; Arc length Calculate the first-order derivative of the x-axis coordinate and the y-axis coordinate at The first derivative of the x-axis coordinate The inverse tangent function of the ratio , as the flagellar tangential angle , expressed as: ; The flagellar propagation velocity is calculated based on the rate of change of the cumulative length of the flagellar curve at adjacent moments , expressed as: ; represents the cumulative arc length of the flagellar curve at time t, Indicates a preset time interval; Arc length Calculate the first and second derivatives of the x-axis and y-axis coordinates at the flagellar curvature , expressed as: ; and Represents arc length The second derivative of the x-axis coordinate and the second derivative of the y-axis coordinate at .

5. sperm flagella motion parameter calculation method according to claim 4, is characterized in that, Based on the flagellar curvature, the flagellar amplitude, flagellar frequency, flagellar wavelength, flagellar bending amplitude and flagellar bending index are calculated, including: Get multiple pairs of local maxima and local minima of the flagellar curvature as peaks and troughs ; Calculate the sum of the vertical distances between all adjacent peaks and troughs and average them to obtain the flagellar amplitude , expressed as: ; Indicates the vertical distance between adjacent peaks and troughs, represents the total number of pairs of peaks and troughs; Based on unit time Number of flagellar curvature sign changes at inner arc length s The ratio of the unit time to the flagellar frequency , expressed as: ; Based on the speed of propagation The ratio of the flagellar frequency to the flagellar wavelength , expressed as: ; The flagellar curvature of the parameterized flagellar skeleton curve is integrated along the arc length to obtain the flagellar bending amplitude , expressed as: ; Based on flagellar curvature The mean and standard deviation , get the flagellar bending index , expressed as: .

6. The method for calculating sperm flagellar motion parameters according to claim 5, wherein: It also includes flagellar amplitude-based Flagellar wavelength Calculation of flagellar amplitude-wavelength ratio , expressed as: .

7. The method for calculating sperm flagellar motion parameters according to claim 5, wherein: Also includes calculation of flagellar tangential angle Arc length The derivative of , as the flagellar wavelength index , expressed as: .

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for calculating sperm flagellar motion parameters according to any one of claims 1 to 7 are implemented.

Images