A high-precision sperm morphology screening system and screening method

Abstract

The application discloses a high-precision sperm morphology screening system and a screening method, and relates to the technical field of sperm cell recognition. Raw semen is obtained and filtered to obtain initial semen; the initial semen is spread into a preset single-layer sperm liquid strip to obtain initial single-layer motile sperm; a first magnification operation is performed on the initial single-layer motile sperm to obtain a first initial single-layer motile sperm image; a second magnification operation is performed on the first initial single-layer motile sperm image to obtain a second initial single-layer motile sperm image; and the second initial single-layer motile sperm image is substituted into a high-precision sperm morphology selection model to obtain target high-quality sperm. Through the preliminary filtration of the raw semen and the secondary screening of the preset single-layer sperm strip spreading method, and by adopting the progressive magnification strategy of the first magnification operation and the second magnification operation, the high-precision sperm morphology screening can be performed when high-quality sperm is selected by using the existing IMSI technology, and the screening efficiency of the high-precision sperm is improved.

Description

Technical Field

[0001] This invention belongs to the field of sperm cell recognition technology, specifically relating to a high-precision sperm morphology screening system and screening method. Background Technology

[0002] Intracytoplasmic sperm injection (ICSI) is a commonly used assisted reproductive technology for treating male infertility, suitable for patients with abnormal sperm count, low sperm motility, or abnormal sperm morphology. However, clinical data shows that even when embryos are obtained through conventional ICSI, patients with severe teratospermia have a significantly higher rate of embryonic developmental abnormalities compared to infertile men with normal sperm morphology. This often results in recurrent pregnancy failures, placing a heavy psychological and emotional burden on patients and their families. One important reason for this is the poor quality of sperm used for microinjection. Currently, the magnification of microscopes used in ICSI internationally and domestically is approximately 200-400x, which can only screen sperm with no obvious morphological abnormalities for injection, but makes it difficult to identify subtle structural abnormalities in the sperm head, such as vacuolation and mitochondrial abnormalities.

[0003] Intracytic morphological screening sperm injection (IMSI) technology and sperm selection have been publicized in medical research and education. Selecting morphologically superior sperm for assisted reproductive technology is an effective way to improve clinical outcomes. Normal sperm selected by low-magnification optical microscopy (x200) may contain some minor structural defects, which can sometimes lead to fertilization failure or embryonic developmental arrest when used for intracytoplasmic sperm injection (ICSI). Intracytic morphological screening sperm injection technology is a perfect combination of ICSI and morphological examination of motile sperm organelles. IMSI is equipped with differential interferometry phase-contrast microscopy and digital imaging technology, which allows for morphological quality evaluation of motile live sperm at high magnification (x6600). Sperm with no or few vacuoles in the head are selected for microinjection, thereby improving fertilization rate, implantation rate, pregnancy rate, and reducing miscarriage rate. This article mainly reviews the development of IMSI technology and its application in sperm selection, and analyzes the clinical outcomes of IMSI, providing theoretical guidance for its application expansion and further functional development.

[0004] Patent CN109064469A discloses a sperm quality analyzer and a sperm quality testing system. The sperm quality analyzer includes: an image magnification device, an image acquisition device, and an image processing device. The image magnification device is used to magnify the semen sample to be tested; the image acquisition device is used to acquire images of the magnified semen sample, generate image information, and send it to the image processing device; the image processing device is used to analyze and calculate the image information, generate calculation results, and the calculation results include sperm concentration and / or sperm motility.

[0005] However, existing IMSI technology simply makes a basic judgment on sperm with few or no vacuoles in the head when selecting high-quality sperm, resulting in low efficiency of high-precision sperm morphology screening. Summary of the Invention

[0006] The purpose of this invention is to solve the problem that existing IMSI technology only makes a simple judgment on sperm with no or few vacuoles in the head when selecting high-quality sperm, resulting in low efficiency of high-precision sperm morphology screening. Therefore, this invention proposes a high-precision sperm morphology screening system and screening method.

[0007] In a first aspect of this invention, a high-precision sperm morphology screening method is first proposed, the method comprising:

[0008] Obtain raw semen, and filter the raw semen to obtain initial semen;

[0009] The initial semen was spread into a pre-set single-layer sperm strip to obtain an initial single-layer motile sperm;

[0010] A first magnification operation is performed on the initial monolayer of motile sperm to obtain a first image of the initial monolayer of motile sperm;

[0011] The first initial monolayer motile sperm image is magnified to the target resolution to obtain the magnified initial monolayer motile sperm image;

[0012] A second initial single-layer motile sperm image is obtained by performing a second magnification operation on the first initial single-layer motile sperm image based on the magnified initial single-layer motile sperm image.

[0013] For the second initial monolayer motile sperm image, the structural information image is obtained using the Sobel operator;

[0014] The second initial monolayer motile sperm image and the structural information image are substituted into a high-precision sperm morphology selection model to obtain the target high-quality sperm.

[0015] Optionally, the initial semen is spread into a preset monolayer of sperm to obtain an initial monolayer of motile sperm, including:

[0016] Step 1, Glassware Preparation: Prepare N 20µL Gamete liquid strips and one 2µL polygonal tentacles-like PVP droplet in an IMSI glassware.

[0017] Step 2, Semen Strip Preparation: For each Gamete strip, 10uL of liquid is removed from the strip, and 10uL of initial semen is injected. The strip is then centrifuged and precipitated over a preset time period to obtain centrifuged semen strips.

[0018] Step 3, Preliminary sperm screening: Under an inverted microscope, all the initial normal sperm in the centrifuged semen strip are selected using an ICSI needle according to preset rules and added to the PVP droplets to obtain an initial monolayer of motile sperm.

[0019] Optionally, performing a second magnification operation on the first initial monolayer motile sperm image based on the magnified initial monolayer motile sperm image to obtain a second initial monolayer motile sperm image includes:

[0020] The first initial single-layer motile sperm image and the magnified initial single-layer motile sperm image are respectively substituted into the pre-trained VAE encoder to obtain low-resolution latent vectors and high-resolution latent vectors;

[0021] Noise is added to the low-resolution latent vector and the high-resolution latent vector respectively to obtain a low-resolution noisy latent vector and a high-resolution noisy latent vector;

[0022] The updated high-resolution latent vector is obtained by updating the high-resolution latent vector based on the low-resolution latent vector.

[0023] Generate mean squared error loss based on the low-resolution latent vector and the updated high-resolution latent vector;

[0024] The second initial monolayer motile sperm image is obtained by gradient updating the magnified initial monolayer motile sperm image based on the mean square error loss.

[0025] Optionally, updating the high-resolution noise latent vector based on the low-resolution noise latent vector to obtain the updated high-resolution latent vector includes:

[0026] The first initial single-layer motile sperm image is segmented by a preset first region size to obtain a set of labeled sub-initial single-layer motile sperm images, and a corresponding text prompt is generated for each sub-initial single-layer motile sperm image using a large language model.

[0027] The low-resolution latent noise vector and the text prompt corresponding to each initial single-layer motile sperm image are substituted into a preset Unet network to obtain the low-resolution attention score corresponding to each initial single-layer motile sperm image.

[0028] The high-resolution latent vectors are segmented by a preset second region size to obtain a labeled sub-high-resolution latent vector set; the labels in the sub-high-resolution latent vector set correspond one-to-one with the labels in the sub-initial single-layer active sperm image set;

[0029] Substitute the text prompt, low-resolution attention score and sub-high-resolution latent vector corresponding to the same label into the attention synthesizer for denoising to obtain the sub-updated high-resolution latent vector corresponding to the label.

[0030] The updated high-resolution latent vector is obtained by merging the sub-updated high-resolution latent vectors corresponding to all labels.

[0031] Optionally, substituting the second initial single-layer motile sperm image and the structural information image into a high-precision sperm morphology selection model to obtain target high-quality sperm includes:

[0032] The second initial single-layer active sperm image and the structural information image are respectively substituted into a single residual block to obtain the basic features and structural basic features;

[0033] Substituting the basic features and the structural basic features into the multi-scale sampling model, we obtain low-scale morphological features, medium-scale morphological features and high-scale morphological features;

[0034] Substituting the low-scale morphological features, the mid-scale morphological features, and the high-scale morphological features into the dual-path feature optimization model yields the low-scale final fusion feature, the mid-scale final fusion feature, and the high-scale final fusion feature.

[0035] Based on the low-scale final fusion features, the medium-scale final fusion features, and the high-scale final fusion features, target high-quality sperm are obtained by screening in combination with morphological quality standards.

[0036] Optionally, the basic features and structural basic features are obtained by substituting the second initial single-layer active sperm image and the structural information image into a single residual block, respectively, including:

[0037] The first feature and the second feature are obtained by substituting the second initial monolayer active sperm image and the structural information image into a single residual block, respectively.

[0038] The first and second features are compressed using average pooling and max pooling, and then merged to obtain a fused feature.

[0039] The fused feature, the first feature, and the second feature are concatenated to obtain a joint feature;

[0040] The joint features are processed through a 5×5 convolutional layer with Sigmoid activation to generate a first weight and a second weight; the first weight is multiplied by the first feature to obtain the basic features; the second weight is multiplied by the second feature to obtain the structural basic features.

[0041] Optionally, substituting the basic features and the structural basic features into a multi-scale sampling model to obtain low-scale morphological features, mesoscale morphological features, and high-scale morphological features includes:

[0042] The basic features and the structural basic features are fused to obtain the fused basic features;

[0043] The fused basic features are split into two branches using a 1×1 convolution to obtain the first sub-fused basic features and the second sub-fused basic features;

[0044] The first sub-convolutional feature is obtained by performing a convolution operation on the first sub-fusion basic feature using a first convolutional kernel; the second sub-convolutional feature is obtained by performing a convolution operation on the second sub-fusion basic feature using a second convolutional kernel.

[0045] The second sub-convolutional feature is subjected to a 3×3 convolution plus a 1×1 convolution plus a Sigmoid activation to obtain the second sub-convolutional attention.

[0046] The second sub-attention feature is obtained by multiplying the second sub-convolutional attention by the second sub-convolutional feature.

[0047] The enhanced multi-stage features are obtained by concatenating the second sub-attention features and the first sub-convolution features and then performing 3×3 convolution and residual connection. The multi-stage features include low-scale morphological features, medium-scale morphological features and high-scale morphological features.

[0048] Optionally, substituting the low-scale morphological features, the mesoscale morphological features, and the high-scale morphological features into the dual-path feature optimization model to obtain the low-scale final fusion feature, the mesoscale final fusion feature, and the high-scale final fusion feature includes:

[0049] First, a 1×1 convolution is performed on the first-scale feature and the second-scale feature. Then, the second-scale feature is upsampled and fused with the first-scale feature. Finally, bilinear interpolation is performed to generate the first semantic feature. The first-scale feature is any one of the medium-scale morphological feature and the high-scale morphological feature. The second-scale feature is any one of the medium-scale morphological feature and the low-scale morphological feature. The scale of the second-scale feature is smaller than that of the first-scale feature.

[0050] After performing a 1×1 convolution on the first semantic feature, a sigmoid activation function is passed in to obtain the foreground attention map and the background attention map;

[0051] The foreground path is obtained by multiplying the foreground attention map by the second-scale feature, and the background path is obtained by multiplying the background attention map by the second-scale feature. When the first-scale feature is a high-scale morphological feature and the second-scale feature is a mid-scale morphological feature, the first foreground path and the first background path are obtained. When the first-scale feature is a high-scale morphological feature and the second-scale feature is a low-scale morphological feature, the second foreground path and the second background path are obtained. When the first-scale feature is a mid-scale morphological feature and the second-scale feature is a low-scale morphological feature, the third foreground path and the third background path are obtained.

[0052] The first foreground path, the second foreground path, and the third foreground path are fused to obtain the fused foreground path;

[0053] The first background path, the second background path, and the third background path are merged to obtain the merged background path;

[0054] The foreground and background paths are optimized using residual blocks, and then concatenated and generated by 1×1 convolution to produce the final fused features.

[0055] The low-scale final fusion feature is obtained by averaging the low-scale morphological features and the final fusion feature; the mid-scale final fusion feature is obtained by averaging the mid-scale morphological features and the final fusion feature; and the high-scale final fusion feature is obtained by averaging the high-scale morphological features and the final fusion feature.

[0056] Optionally, the first magnification operation is performed using a microscope; the microscope uses a 60x objective lens and a 10x eyepiece.

[0057] In a second aspect of this invention, a high-precision sperm morphology screening system is provided, comprising:

[0058] A filtration module is used to obtain raw semen and filter the raw semen to obtain initial semen.

[0059] The initial monolayer motile sperm screening module is used to spread the initial semen into a preset monolayer sperm strip to obtain an initial monolayer motile sperm.

[0060] The first magnification module is used to perform a first magnification operation on the initial monolayer motile sperm to obtain a first initial monolayer motile sperm image;

[0061] The module for generating an enlarged initial monolayer motile sperm image is used to enlarge the first initial monolayer motile sperm image to a target resolution to obtain an enlarged initial monolayer motile sperm image.

[0062] The second magnification module is used to perform a second magnification operation on the first initial single-layer motile sperm image based on the magnified initial single-layer motile sperm image to obtain a second initial single-layer motile sperm image.

[0063] The structural information image generation module is used to obtain a structural information image from the second initial monolayer motile sperm image using the Sobel operator.

[0064] The high-precision sperm morphology selection module is used to input the second initial single-layer motile sperm image and the structural information image into the high-precision sperm morphology selection model to obtain target high-quality sperm.

[0065] The beneficial effects of this invention are:

[0066] This invention proposes a high-precision sperm morphology screening method. First, the raw semen is filtered to obtain initial semen, effectively removing impurities, dead sperm, and other useless components. Then, the initial semen is spread into a pre-set monolayer sperm strip to ensure uniform sperm distribution, forming an initial monolayer of motile sperm and preventing sperm stacking from affecting observation and screening. Subsequently, two magnification operations are performed: first, a first initial monolayer motile sperm image is obtained; then, it is magnified to the target resolution, and a second initial monolayer motile sperm image is optimized based on this, gradually improving image clarity and detail, making sperm morphology features easier to capture; finally, using So… The bel operator extracts structural information images, which can further highlight key structural features such as the outline and texture of sperm, providing richer judgment criteria for accurate identification. Finally, the second initial single-layer motile sperm image and structural information image are substituted into the high-precision sperm morphology selection model to accurately screen out target high-quality sperm with excellent morphology and strong motility, which greatly improves the screening efficiency and accuracy of high-quality sperm. It is not just a simple judgment of identifying sperm with no or few vacuoles in the head, but provides a reliable source of high-quality sperm for assisted reproduction and other related fields, ensuring the success rate of subsequent culture and other stages, and improving the efficiency of high-precision sperm screening. Attached Figure Description

[0067] The invention will now be further described with reference to the accompanying drawings.

[0068] Figure 1 A flowchart illustrating a high-precision sperm morphology screening method provided in an embodiment of the present invention;

[0069] Figure 2 A schematic flowchart illustrating a high-precision sperm morphology screening method provided in an embodiment of the present invention;

[0070] Figure 3 This invention provides sperm images at different magnifications as an embodiment of the invention;

[0071] Figure 4 A comparison image of abnormal sperm provided in an embodiment of the present invention;

[0072] Figure 5 This is a framework diagram of a high-precision sperm morphology screening system provided in an embodiment of the present invention. Detailed Implementation

[0073] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0074] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0075] This invention provides a high-precision method for sperm morphology screening. See also... Figure 1 , Figure 1 A flowchart illustrating a high-precision sperm morphology screening method provided in this embodiment of the invention. The method includes the following steps:

[0076] S101, Obtain raw semen and filter the raw semen to obtain initial semen;

[0077] S102, the initial semen is spread into a preset single-layer sperm strip to obtain an initial single-layer motile sperm;

[0078] S103, Perform a first magnification operation on the initial monolayer of motile sperm to obtain a first image of the initial monolayer of motile sperm;

[0079] S104, the first initial monolayer motile sperm image is magnified to the target resolution to obtain the magnified initial monolayer motile sperm image;

[0080] S105, A second initial monolayer motile sperm image is obtained by performing a second magnification operation on the first initial monolayer motile sperm image based on the magnified initial monolayer motile sperm image;

[0081] S106, For the second initial monolayer motile sperm image, obtain the structural information image using the Sobel operator;

[0082] S107, Substitute the second initial single-layer motile sperm image and structural information image into the high-precision sperm morphology selection model to obtain the target high-quality sperm.

[0083] Based on the high-precision sperm morphology screening method provided by this invention, the initial semen is first filtered to remove impurities, dead sperm, and other useless components. Then, it is spread into a pre-set monolayer sperm strip to ensure uniform sperm distribution, forming an initial monolayer of motile sperm and preventing stacking that could affect observation and screening. Subsequently, two magnification operations are performed: first, a first image of the initial monolayer of motile sperm is obtained; then, it is magnified to the target resolution and optimized to obtain a second image, gradually improving clarity and detail to facilitate the capture of sperm morphological features. Next, the Sobel operator is used to extract structural information images, highlighting key features such as sperm outline and texture, providing a basis for accurate identification. Finally, the two images are substituted into a high-precision sperm morphology selection model to accurately screen out high-quality sperm with excellent morphology and strong motility. This method not only simply determines the absence or scarcity of vacuoles in the sperm head but also significantly improves the efficiency and accuracy of high-quality sperm screening, providing a reliable source of high-quality sperm for assisted reproduction and other fields, ensuring a high success rate for subsequent culture.

[0084] In one implementation, the original semen is from patients with severe teratospermia or occult azoospermia; the patient's sperm is liquefied at 37°C for 30-60 minutes, 20 microliters of liquefied semen are collected for microscopic examination, and the sperm motility and quantity are recorded (the quantity is observed and counted using existing microscopes).

[0085] In one implementation, if the original semen sample comes from a patient with severe teratospermia, the sperm density is determined to be below 1×10⁻⁶. 6 (Unit: quantity / ml) If the sperm density is less than 1×10 6 If the sperm density is equal to or higher than 1×10⁻⁶, then the initial semen is obtained by direct washing; 6 Then, the semen is subjected to density gradient centrifugation (300g / 20min) and the precipitate is retained to obtain the initial semen.

[0086] In one implementation, see [link to implementation details]. Figure 2 , Figure 2 A flowchart illustrating a high-precision sperm morphology screening method is provided. Referring to Part I (implementation step S101), semen is first collected. Gradient solution 1 (90% colloidal silica particles) and gradient solution 2 (45% colloidal silica particles) are added to the semen, followed by centrifugation (centrifugation time is 20 min). The supernatant is removed from the centrifuged mixture to obtain sperm precipitate. Culture medium (G-IVF plus) is added to the sperm precipitate, and the mixture is mixed and washed to obtain initial semen. After obtaining the initial semen, referring to Part II (implementation step S102), the initial semen is placed in a glass dish and filtered using a monolayer sperm strip method to obtain an initial monolayer of motile sperm. After obtaining the initial monolayer of motile sperm, referring to Part III (implementation steps S103-S105).

[0087] In one implementation, the original semen is the semen of a patient with occult azoospermia. After liquefying at 37°C for 30-60 minutes, it is directly centrifuged (300g / 20min), and the supernatant is discarded, leaving 100 microliters of precipitate to obtain the initial semen.

[0088] In one implementation, the first magnification operation specifically involves magnifying the initial monolayer of motile sperm obtained by adding it to the PVP droplet using a dedicated objective (60x) and eyepiece (10x) of an IMSI system (e.g., referring to the Cell Screening Manual published by OCTAX Microscience GmbH in 2008) and acquiring a digital image to obtain a first initial monolayer of motile sperm image. The first magnification operation is performed using a microscope with a 60x objective and a 10x eyepiece. At this point, the first initial monolayer of motile sperm image can be magnified 600x using the combination of the objective and eyepiece. Then, a second magnification operation is performed on the first initial monolayer of motile sperm image (600x) to obtain a second initial monolayer of motile sperm image. The second magnification operation is implemented by software, which can magnify the 600x first initial monolayer of motile sperm image to a second initial monolayer of motile sperm image (2000-6000x).

[0089] In one implementation, see [link to implementation details]. Figure 3 , Figure 3 This invention provides sperm images at different magnifications. At 200x and 300x magnification using ICSI, only the shape of the sperm can be seen, and internal information cannot be observed. However, at 1200x magnification using SLS-IMSI (single-layer sperm strip laying combined with ultra-high resolution microinjection), the morphological structure of the sperm can be roughly seen, such as the subtle features of the sperm head and tail. This is crucial for assessing sperm quality and screening out morphologically normal sperm.

[0090] In one implementation, see [link to implementation details]. Figure 4 , Figure 4 A comparison chart of abnormal sperm is provided, with abnormal sperm shown in the chart and normal sperm on the right.

[0091] In one implementation, the target high-quality sperm is defined as sperm with symmetrical head morphology, an oval head structure, homogeneous nuclear chromatin with no more than one vacuole, or a vacuole area of ​​less than 4% of the nuclear area, and a normal nuclear appearance. The average length and width of the sperm are defined as: length: 4.75±0.28μm, width: 3.28±0.20μm.

[0092] In one implementation, the raw semen is filtered: basic contaminants such as seminal plasma impurities, dead sperm, and abnormal fragments are removed from the semen to obtain initial semen, which reduces interference for subsequent screening; a pre-set monolayer sperm strip filtration method is used: the initial monolayer of motile sperm is further separated through a standardized strip method. This step can selectively retain sperm with strong motility and remove immotile and asthenosperm, completing the first round of quality screening from the perspective of motility, laying a high-quality foundation for subsequent morphological analysis.

[0093] In one implementation, two magnification operations are performed. The first magnification is achieved through the objective lens and eyepiece, while the second magnification is achieved through a software AI model. This process gradually focuses on individual sperm from macroscopic to microscopic perspectives, ensuring that subsequent morphological analysis can clearly capture the details of the sperm head, neck, and tail, thus avoiding misjudgments caused by insufficient image resolution.

[0094] In one implementation, compared to manual morphological assessment, a high-precision sperm morphology selection model can establish standardized judgment criteria based on massive labeled data, accurately identify high-quality sperm with regular heads, no deformities, and complete tails. This reduces human error and covers subtle morphological features that are difficult for humans to detect, ultimately ensuring that the selected high-quality sperm have better morphological integrity.

[0095] In one embodiment, obtaining an initial monolayer of motile sperm by filtering the initial semen using a preset monolayer sperm strip method includes:

[0096] Step 1, Glassware Preparation: Prepare N 20µL Gamete liquid strips and one 2µL polygonal tentacles-like PVP droplet in an IMSI glassware.

[0097] Step 2, Semen Strip Preparation: For each Gamete strip, 10uL of liquid is removed from the strip, and 10uL of initial semen is injected. The strip is then centrifuged and precipitated over a preset time period to obtain centrifuged semen strips.

[0098] Step 3, Preliminary sperm screening: Under an inverted microscope, all the initial normal sperm in the centrifuged semen strips are selected using an ICSI needle according to preset rules and added to the PVP droplets to obtain an initial monolayer of motile sperm.

[0099] In one implementation, the glass dish preparation for the improved monolayer sperm strip placement method uses dedicated IMSI (version number OctaxEyeware) glass dishes (OOPW-IC03, OosafeICSI, IMSIDish, etc.). The IMSI system requires cell culture dishes made of glass with better light transmittance, and the glass injection dishes should be prepared at least 2 hours in advance before the actual IMSI procedure. Using a micropipette, three Gamete strips with a volume of 20 μL, a length of 4 cm, and a width of 0.5 cm are prepared in the dedicated dish by connecting the ends of the strips. Two 10 μL Gamete strips are placed on the bottom of the dish to hold the oocytes, and a 2 μL polygonal tentacles-like PVP strip is placed in the center of the dish to hold the sperm. Finally, the dish is covered with paraffin oil and placed in an incubator for equilibration.

[0100] In one implementation, a very small amount of sperm is collected using a single-layer sperm strip method. First, 10 microliters of Gamete are aspirated from the strip using a micropipette, and then 10 microliters of semen are centrifuged and precipitated. The precipitate is slowly added to the strip from beginning to end. Each strip is filled with the centrifuged semen precipitate in the same way. After all the semen precipitate strips are completely spread, the strips are placed in an incubator to equilibrate for 15 minutes. Then, the glass dish is removed to observe whether any motile sperm have swum to the edges of the strips.

[0101] In one implementation, sperm are initially screened using a single IMSI operation. After the culture dish is fully plated and cultured for 15-30 minutes, all sperm with relatively normal morphology are collected into PVP droplets using an inverted microscope using an ICSI injection needle. If the number of sperm swimming out is particularly small, the center of the strip can be searched for any slightly trembling motile sperm, which can be used as the initial monolayer of motile sperm. The preset time period is 15-30 minutes. The preset rule is that the sperm head is symmetrical and has an oval structure.

[0102] In one embodiment, obtaining a second initial monolayer motile sperm image by performing a second magnification operation on the first initial monolayer motile sperm image based on the magnified initial monolayer motile sperm image includes:

[0103] Substituting the first initial single-layer motile sperm image and the magnified initial single-layer motile sperm image into the pre-trained VAE encoder, respectively, yields low-resolution latent vectors and high-resolution latent vectors;

[0104] Adding noise to the low-resolution latent vector and the high-resolution latent vector respectively yields a low-resolution noisy latent vector and a high-resolution noisy latent vector.

[0105] The updated high-resolution latent vector is obtained by updating the high-resolution noise latent vector based on the low-resolution noise latent vector.

[0106] Generate mean squared error loss based on the low-resolution latent vector and update the high-resolution latent vector;

[0107] The second initial monolayer motile sperm image is obtained by gradient updating the magnified initial monolayer motile sperm image based on the mean square error loss.

[0108] In one implementation, the target resolution is the resolution that the technician needs to magnify, typically by magnifying the first initial single-layer active sperm image by 10 times; the pre-trained VAE encoder is a variational autoencoder; noise is added to both the low-resolution latent vector and the high-resolution latent vector, both of which are standard normally distributed noise.

[0109] In one implementation, generating the mean squared error loss based on the low-resolution latent vector and the updated high-resolution latent vector involves performing wavelet transforms on the low-resolution latent vector and the updated high-resolution latent vector respectively, extracting the corresponding low-frequency components to obtain low-frequency low-resolution latent features and low-frequency high-resolution latent features, and then calculating the square of the difference between the low-frequency low-resolution latent features and the low-frequency high-resolution latent features to obtain the mean squared error loss.

[0110] In one implementation, a second initial monolayer motile sperm image is obtained by gradient updating the magnified initial monolayer motile sperm image based on the mean square error loss, specifically through the formula... Where A is the high-resolution latent vector corresponding to the second initial monolayer motile sperm image, and A0 is the high-resolution latent vector of the magnified initial monolayer motile sperm image. For cosine decay scheduling, The gradient of the high-resolution latent vector of the magnified initial single-layer motile sperm image is calculated. The mean squared error loss is used to calculate the high-resolution latent vector corresponding to the second initial monolayer motile sperm image. This vector is then input into the decoder of the variational autoencoder to obtain the second initial monolayer motile sperm image.

[0111] In one implementation, the mean squared error loss is calculated by comparing the low-resolution latent vector with the updated high-resolution latent vector. Essentially, this utilizes the global structure of the low-resolution image to constrain the generation of the high-resolution image. This constraint effectively avoids head morphological anomalies common in traditional magnification methods, ensuring that the magnified sperm image is consistent with the original low-resolution image in terms of global morphology, meeting the core requirement of structural realism in biomedical images. In the low-resolution image, the sperm head is elliptical and smoothly connected to the tail. Through loss constraint, the high-resolution image retains this global feature, preventing the overall structure from being destroyed by the generation of details.

[0112] In one implementation, the operation of adding noise to low- and high-resolution latent vectors borrows from the noise and denoising logic of the diffusion model: noise provides flexibility for generating details in high-resolution images, and the high-resolution noise latent vector is subsequently updated by the low-resolution noise latent vector; for sperm images, this approach can generate the blurred fine structures in low-resolution images without generating details that do not conform to biological laws out of thin air, thus balancing the enhancement of clarity and biological rationality.

[0113] In one implementation, a closed loop of mean squared error loss and gradient update is used to achieve iterative optimization of high-resolution images. Each update makes the high-resolution image closer to the global structure of the low-resolution image while retaining reasonable details. The final output image achieves a better balance between structural consistency and detail richness. This optimization allows the magnified image to clearly show the tail swing amplitude without distorting the head shape, thus improving the accuracy of subsequent analysis.

[0114] In one embodiment, updating the high-resolution noise latent vector based on the low-resolution noise latent vector to obtain the updated high-resolution latent vector includes:

[0115] The first initial single-layer motile sperm image is segmented by a preset first region size to obtain a set of labeled sub-initial single-layer motile sperm images, and a corresponding text prompt is generated for each sub-initial single-layer motile sperm image using a large language model.

[0116] Substitute the low-resolution latent noise vector and the text prompt corresponding to each initial single-layer motile sperm image into the preset Unet network to obtain the low-resolution attention score corresponding to each initial single-layer motile sperm image.

[0117] By segmenting the high-resolution latent vectors by a preset second region size, a labeled sub-high-resolution latent vector set is obtained; the labels in the sub-high-resolution latent vector set correspond one-to-one with the labels in the sub-initial single-layer active sperm image set.

[0118] Substitute the text prompt, low-resolution attention score and sub-high-resolution latent vector corresponding to the same label into the attention synthesizer for denoising to obtain the sub-updated high-resolution latent vector corresponding to the label.

[0119] The updated high-resolution latent vector is obtained by merging the sub-updated high-resolution latent vectors corresponding to all labels.

[0120] In one implementation, a first initial single-layer motile sperm image is segmented by a preset first region size to obtain a set of labeled sub-initial single-layer motile sperm images. A set of labeled sub-high-resolution latent vectors is obtained by segmenting high-resolution latent vectors by a preset second region size. The labeled sub-initial single-layer motile sperm image set and the corresponding positions in the labeled sub-high-resolution latent vector set are in one-to-one correspondence. First, the low-resolution first initial single-layer motile sperm image is segmented according to the preset first region size. The label of each sub-image is bound to its spatial position in the low-resolution image (e.g., row and column number) (e.g., the first sub-image in the upper left corner is labeled 1, the adjacent block on the right is labeled 2, and so on). When the high-resolution latent vector is segmented according to the preset second region size, the corresponding position of the low-resolution sub-image in the high-resolution latent space is determined by proportional scaling from low to high resolution and VAE compression mapping from high-resolution pixel space to latent space. Then, the sub-high-resolution latent vector at that position is labeled with the same number, ultimately achieving sub-regions with the same label.

[0121] In one implementation, the Unet network is pre-defined as a Unet with an attention synthesizer. Different regions of the sperm image (such as the head, tail, and connective tissue) have distinct biological characteristics (the head needs to retain an elliptical acrosome, and the tail needs to exhibit a slender, unbroken texture). By segmenting the image into labeled sub-regions based on a pre-defined region size, and then generating exclusive text prompts for each sub-region (e.g., the head acrosome region has an arc shape in the first 1 / 3, and the middle section of the tail has clear fiber texture), the attention mechanism can accurately focus on the region-specific features. After the low-resolution attention score is aligned with the high-resolution sub-region through labeling, it can prevent the tail region from incorrectly learning the acrosome features of the head, ensuring that the denoising process of each region is only guided by its own exclusive features, and preventing cross-regional detail confusion.

[0122] In one implementation, the attention synthesizer simultaneously receives textual cues (semantic constraints), low-resolution attention scores (low-resolution feature guidance), and sub-high-resolution latent vectors (high-resolution features to be optimized), achieving multi-source information fusion and denoising: the textual cues provide what should be generated in the region (e.g., no tail breakage), avoiding biological errors; the low-resolution attention scores provide which features should be emphasized in the region (e.g., high weighting of elliptical features in the head region), ensuring consistency with the low-resolution image; the sub-high-resolution latent vectors provide the basic features to be optimized, gradually generating high-resolution details through denoising; this fusion mechanism ensures that the local details of sperm (e.g., acrosome edge clarity, tail fiber texture) conform to biological laws while remaining faithful to the feature distribution of the original low-resolution image.

[0123] In one embodiment, substituting the second initial monolayer motile sperm image and the structural information image into a high-precision sperm morphology selection model to obtain target high-quality sperm includes:

[0124] The basic features and structural basic features are obtained by substituting the second initial single-layer active sperm image and structural information image into a single residual block, respectively.

[0125] Substituting the basic features and structural basic features into the multi-scale sampling model yields low-scale morphological features, medium-scale morphological features, and high-scale morphological features.

[0126] Substituting low-scale, mid-scale, and high-scale morphological features into the dual-path feature optimization model yields the final fused features at low scale, mid-scale, and high scale.

[0127] Based on low-scale final fusion characteristics, medium-scale final fusion characteristics, and high-scale final fusion characteristics, combined with morphological quality standards, target high-quality sperm were selected.

[0128] In one implementation, the structural information image extracted by the Sobel operator complements the original sperm image: the original image is good at preserving biological features such as the staining depth of the sperm head, while the structural information image enhances the morphological edges; after processing by a single residual block, the basic features and structural basic features respectively preserve the integrity of the two types of information, laying the foundation for feature-completeness in subsequent multi-scale analysis.

[0129] In one implementation, the first feature and the second feature are obtained by substituting the second initial single-layer active sperm image and the structural information image into a single residual block, respectively; the first feature and the second feature are compressed by average pooling and max pooling, and then merged to obtain a fused feature; the fused feature, the first feature and the second feature are concatenated to obtain a joint feature; the joint feature is then passed through a 5×5 convolutional layer with Sigmoid activation to generate a first weight and a second weight; the first weight is multiplied by the first feature to obtain the basic feature; and the second weight is multiplied by the second feature to obtain the structural basic feature.

[0130] In one implementation, basic visual features (such as sperm morphology and contour) of the second initial single-layer motile sperm image and detailed features (such as internal texture) of the structural information image are extracted separately. These two types of information are then completely fused through concatenation, avoiding the loss of crucial details from single-type features. The first and second weights generated by 5×5 convolution with a sigmoid function can automatically determine the importance of the two types of features in the current task (such as sperm motility assessment). For example, when structural details are more critical to the judgment, the second weight will increase, allowing basic structural features to contribute more and reducing interference from useless features.

[0131] In one implementation, the basic features and structural basic features are substituted into a multi-scale sampling model to obtain low-scale morphological features, mid-scale morphological features, and high-scale morphological features. Specifically, the basic features and structural basic features are fused to obtain fused basic features; the fused basic features are split into two branches through a 1×1 convolution to obtain a first sub-fused basic feature and a second sub-fused basic feature; the first sub-fused basic feature is convolved with a first convolution kernel (with 3 kernels) to obtain a first sub-convolutional feature; the second sub-fused basic feature is convolved with a second convolution kernel (with 5 kernels) to obtain a second sub-convolutional feature; the second sub-convolutional feature is subjected to a 3×3 convolution plus a 1×1 convolution plus a Sigmoid activation to obtain a second sub-convolutional attention; the second sub-convolutional attention is multiplied by the second sub-convolutional feature to obtain a second sub-attention feature; the second sub-attention feature and the first sub-convolutional feature are concatenated and then subjected to a 3×3 convolution plus a residual connection (added to the fused basic features) to generate enhanced multi-stage features (low-scale morphological features, mid-scale morphological features, and high-scale morphological features).

[0132] In one implementation, the division of low-, medium-, and high-scale morphological features accurately matches the scale differences of different parts of the sperm. Low-scale features capture the delicate structure of the tail and the presence of multiple malformations (such as short tail, curled tail, folded tail, no tail, and irregular tail). Medium-scale features capture neck information and identify any abnormalities in the arrangement of mitochondria in the neck. High-scale features focus on the size of the head, determine whether the head has an acrosome, whether the acrosome morphology is regular, and whether the head has vacuoles. The combination of these three features achieves complete coverage of the global morphological characteristics of the sperm, solving the problems of traditional single-scale analysis.

[0133] In one implementation, after splicing features from different branches, 3×3 convolution is used to further fuse them and combine them with residual connections (added to the original fused basic features). This not only strengthens the correlation between the head, neck, and tail of sperm in multi-scale features, but also avoids the feature degradation problem in deep networks.

[0134] In one implementation, low-scale, mid-scale, and high-scale morphological features are substituted into a dual-path feature optimization model to obtain low-scale final fused features, mid-scale final fused features, and high-scale final fused features. Specifically, the first-scale feature and the second-scale feature are first convolved with a 1×1 convolution. Then, the second-scale feature is upsampled and fused with the first-scale feature, followed by bilinear interpolation to generate the first semantic feature. The first semantic feature is then convolved with a 1×1 convolution and fed into a Sigmoid activation function to obtain a foreground attention map A. Therefore, 1-A yields a background attention map. The foreground attention map is multiplied by the second-scale feature to obtain the foreground path, and the background attention map is multiplied by the second-scale feature to obtain the background path. When the first-scale feature is a high-scale morphological feature and the second-scale feature is a mid-scale morphological feature, the first foreground path and the first background path are obtained. When the first scale feature is a high-scale morphological feature and the second scale feature is a low-scale morphological feature, a second foreground path and a second background path are obtained; when the first scale feature is a mid-scale morphological feature and the second scale feature is a low-scale morphological feature, a third foreground path and a third background path are obtained; the first foreground path, the second foreground path, and the third foreground path are fused to obtain a fused foreground path; the first background path, the second background path, and the third background path are fused to obtain a fused background path; the fused foreground path and the fused background path are optimized by residual blocks respectively, and after concatenation, they are convolved and then subjected to 1×1 convolution to generate the final fused feature; the low-scale final fused feature is obtained by averaging the low-scale morphological feature and the final fused feature; the mid-scale final fused feature is obtained by averaging the mid-scale morphological feature and the final fused feature; the high-scale final fused feature is obtained by averaging the high-scale morphological feature and the final fused feature.

[0135] In one implementation, features at different scales are deeply fused at the semantic level through pairwise interactions at high-to-medium, high-to-low, and medium-to-low scales (first convolution to align channels, then upsampling to match dimensions). The global structural information at the high scale can guide the optimization of local features at the medium and low scales, avoiding the one-sidedness of features at a single scale.

[0136] One implementation introduces a foreground-background attention mechanism (generating a foreground attention map A and a background attention map A using a sigmoid function). Figure 1-A) can automatically distinguish sperm targets (foreground) from background noise; the foreground path enhances sperm-related features (such as tail swing area and head shape) and suppresses irrelevant background interference; the background path can assist model learning and indirectly improve the purity of foreground features; this separation mechanism is particularly important for scenarios where sperm are small targets and easily affected by background interference. In one implementation, target high-quality sperm are selected based on low-scale final fusion features, mid-scale final fusion features, and high-scale final fusion features, combined with morphological quality standards. Specifically, an "anchor-based detection head" is deployed on the low-scale, mid-scale, and high-scale final fusion features, and the detection results are output through classification (whether it is a high-quality sperm) and bounding box regression (locating the sperm position). High-quality sperm are selected based on morphological quality standards: the sperm head is symmetrical, the head is an elliptical structure, the nuclear chromatin has homogeneity with no more than one vacuole, or the vacuole area is less than 4% of the nuclear area, and the average length and width of sperm with normal nuclear appearance are limited to length: 4.75±0.28μm and width: 3.28±0.20μm. Sperm that meet the above conditions are marked as target high-quality sperm.

[0137] In one implementation, the dual-path feature optimization model uses parallel processing of foreground enhancement and background suppression: the foreground path focuses on the core region of sperm, amplifying the weight of effective morphological features; the background path filters out interference information such as cell debris and bubbles, avoiding noise being misjudged as sperm features. The final output fused features retain key morphological details while reducing redundant information interference, making subsequent screening more accurate.

[0138] Based on the same inventive concept, this invention also provides a high-precision sperm morphology screening system. See [link to related document]. Figure 5 , Figure 5 A framework diagram of a high-precision sperm morphology screening system provided in an embodiment of the present invention includes:

[0139] A filtration module is used to obtain raw semen and filter the raw semen to obtain initial semen.

[0140] The initial monolayer motile sperm screening module is used to spread the initial semen into a preset monolayer sperm strip to obtain an initial monolayer motile sperm.

[0141] The first magnification module is used to perform a first magnification operation on the initial monolayer motile sperm to obtain a first initial monolayer motile sperm image;

[0142] The module for generating an enlarged initial monolayer motile sperm image is used to enlarge the first initial monolayer motile sperm image to a target resolution to obtain an enlarged initial monolayer motile sperm image.

[0143] The second magnification module is used to perform a second magnification operation on the first initial single-layer motile sperm image based on the magnified initial single-layer motile sperm image to obtain a second initial single-layer motile sperm image.

[0144] The structural information image generation module is used to obtain a structural information image from the second initial single-layer motile sperm image using the Sobel operator.

[0145] The high-precision sperm morphology selection module is used to input the second initial single-layer motile sperm image and the structural information image into the high-precision sperm morphology selection model to obtain target high-quality sperm.

[0146] Based on the high-precision sperm morphology screening system provided in this invention, the initial semen is first filtered to remove impurities, dead sperm, and other useless components. Then, it is spread into a pre-set monolayer sperm strip to ensure uniform sperm distribution, forming an initial monolayer of motile sperm and preventing stacking that could affect observation and screening. Subsequently, two magnification operations are performed: first, a first image of the initial monolayer of motile sperm is obtained; then, it is magnified to the target resolution and optimized to obtain a second image, gradually improving clarity and detail to facilitate the capture of sperm morphological features. Next, the Sobel operator is used to extract structural information images, highlighting key features such as sperm outline and texture, providing a basis for accurate identification. Finally, the two images are substituted into a high-precision sperm morphology selection model to accurately screen out high-quality sperm with excellent morphology and strong motility. This not only simply determines the absence or scarcity of vacuoles in the sperm head but also significantly improves the efficiency and accuracy of high-quality sperm screening, providing a reliable source of high-quality sperm for assisted reproduction and other fields, ensuring a high success rate for subsequent culture.

[0147] In one embodiment, the initial single-layer motile sperm screening module includes:

[0148] The glass preparation module is used to prepare N 20uL Gamete liquid strips and one 2uL polygonal tentacles-like PVP droplet in an IMSI glass dish;

[0149] The semen strip module is used to aspirate 10uL of liquid from each Gamete liquid strip and inject 10uL of initial semen. After centrifugation and sedimentation for a preset time period, centrifuged semen strips are obtained.

[0150] The sperm preliminary screening module is used to select all the initial normal sperm in the centrifuged semen strips under an inverted microscope according to preset rules using an ICSI needle, and add them to the PVP droplets to obtain an initial monolayer of motile sperm.

[0151] In one embodiment, the second amplification module includes:

[0152] The module for determining the magnified initial monolayer motile sperm image is used to magnify the first initial monolayer motile sperm image to the target resolution to obtain the magnified initial monolayer motile sperm image;

[0153] The image encoding module is used to input the first initial single-layer motile sperm image and the magnified initial single-layer motile sperm image into the pre-trained VAE encoder to obtain low-resolution latent vectors and high-resolution latent vectors, respectively.

[0154] The noise addition module is used to add noise to the low-resolution latent vector and the high-resolution latent vector respectively to obtain the low-resolution noise latent vector and the high-resolution noise latent vector.

[0155] The high-resolution noise latent vector update module is used to update the high-resolution noise latent vector based on the low-resolution noise latent vector to obtain the updated high-resolution latent vector.

[0156] The mean squared error loss calculation module is used to generate mean squared error loss based on the low-resolution latent vector and the updated high-resolution latent vector.

[0157] The gradient update module is used to perform gradient update on the magnified initial monolayer motile sperm image based on the mean square error loss to obtain a second initial monolayer motile sperm image.

[0158] In one embodiment, the high-resolution noisy latent vector update module includes:

[0159] The text prompt generation module is used to segment the first initial monolayer motile sperm image by a preset first region size, obtain a set of labeled sub-initial monolayer motile sperm images, and generate corresponding text prompts for each sub-initial monolayer motile sperm image through a large language model.

[0160] The attention score generation module is used to input the low-resolution latent noise vector and the text prompt corresponding to each initial single-layer motile sperm image into the preset Unet network to obtain the low-resolution attention score corresponding to each initial single-layer motile sperm image.

[0161] The high-resolution latent vector segmentation module is used to segment high-resolution latent vectors by a preset second region size to obtain labeled sub-high-resolution latent vector sets; the labels in the sub-high-resolution latent vector sets correspond one-to-one with the labels in the sub-initial single-layer active sperm image set.

[0162] The sub-updated high-resolution latent vector denoising module is used to substitute the text prompts, low-resolution attention scores and sub-high-resolution latent vectors corresponding to the same label into the attention synthesizer for denoising, so as to obtain the sub-updated high-resolution latent vectors corresponding to the label.

[0163] The sub-updated high-resolution latent vector merging module is used to merge the sub-updated high-resolution latent vectors corresponding to all labels to obtain the updated high-resolution latent vector.

[0164] In one embodiment, the high-precision sperm morphology selection module includes:

[0165] The basic feature extraction module is used to substitute the second initial single-layer active sperm image and structural information image into a single residual block to obtain basic features and structural basic features, respectively.

[0166] The multi-scale feature extraction module is used to substitute basic features and structural basic features into the multi-scale sampling model to obtain low-scale morphological features, medium-scale morphological features and high-scale morphological features.

[0167] The multi-scale feature fusion module is used to substitute low-scale morphological features, medium-scale morphological features and high-scale morphological features into the dual-path feature optimization model to obtain low-scale final fused features, medium-scale final fused features and high-scale final fused features;

[0168] The target high-quality sperm screening module is used to screen for target high-quality sperm based on low-scale final fusion features, medium-scale final fusion features, and high-scale final fusion features, combined with morphological quality standards.

[0169] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A high precision method for sperm morphology screening, characterized by, The method comprises: obtaining raw semen, filtering the raw semen to obtain initial semen; spreading the initial semen into a preset single-layer sperm liquid strip to obtain initial single-layer active sperm; performing a first magnification operation on the initial single-layer active sperm to obtain a first initial single-layer active sperm image; magnifying the first initial single-layer active sperm image to a target resolution to obtain a magnified initial single-layer active sperm image; performing a second magnification operation on the first initial single-layer active sperm image according to the magnified initial single-layer active sperm image to obtain a second initial single-layer active sperm image; obtaining a structure information image through a Sobel operator for the second initial single-layer active sperm image; substituting the second initial single-layer active sperm image and the structure information image into a high-precision sperm morphology selection model to obtain target high-quality sperm; substituting the second initial single-layer active sperm image and the structure information image into a high-precision sperm morphology selection model to obtain target high-quality sperm comprises: respectively substituting the second initial single-layer active sperm image and the structure information image into a single residual block to obtain basic features and structure basic features; substituting the basic features and the structure basic features into a multi-scale sampling model to obtain low-scale morphological features, medium-scale morphological features, and high-scale morphological features; substituting the low-scale morphological features, the medium-scale morphological features, and the high-scale morphological features into a double-path feature optimization model to obtain low-scale final fusion features, medium-scale final fusion features, and high-scale final fusion features; screening to obtain target high-quality sperm according to the low-scale final fusion features, the medium-scale final fusion features, and the high-scale final fusion features in combination with morphological quality standards; substituting the low-scale morphological features, the medium-scale morphological features, and the high-scale morphological features into a double-path feature optimization model to obtain low-scale final fusion features, medium-scale final fusion features, and high-scale final fusion features comprises: performing 1x1 convolution on first scale features and second scale features, then performing upsampling on the second scale features, and performing feature fusion on the first scale features to generate first semantic features through bilinear interpolation; the first scale features are any one of the medium-scale morphological features and the high-scale morphological features; the second scale features are any one of the medium-scale morphological features and the low-scale morphological features; the scale of the second scale features is smaller than that of the first scale features; performing 1x1 convolution on the first semantic features and substituting them into a Sigmoid activation function to obtain foreground attention maps and background attention maps; The foreground path is obtained by multiplying the foreground attention map by the second scale feature, and the background path is obtained by multiplying the background attention map by the second scale feature; the first foreground path and the first background path are obtained when the first scale feature is a high scale morphological feature and the second scale feature is a medium scale morphological feature; the second foreground path and the second background path are obtained when the first scale feature is a high scale morphological feature and the second scale feature is a low scale morphological feature; the third foreground path and the third background path are obtained when the first scale feature is a medium scale morphological feature and the second scale feature is a low scale morphological feature; The first foreground path, the second foreground path, and the third foreground path are fused to obtain a fused foreground path; The first background path, the second background path, and the third background path are fused to obtain a fused background path; The fused foreground path and the fused background path are respectively optimized through a residual block, and after splicing, a 1x1 convolution is performed to generate a final fused feature; The low scale final fused feature is obtained by averaging the low scale morphological feature and the final fused feature; the medium scale final fused feature is obtained by averaging the medium scale morphological feature and the final fused feature; and the high scale final fused feature is obtained by averaging the high scale morphological feature and the final fused feature.

2. The method of claim 1, wherein the high-precision sperm morphology screening method is characterized by, The initial sperm is spread into a preset single-layer sperm liquid strip to obtain an initial single-layer motile sperm, including: Step 1, glass dish preparation: preparing N 20uL Gamete liquid strips and a drop of 2uL multi-edge antenna-like PVP liquid drop in IMSI glass dishes; Step 2, sperm smearing: for each Gamete liquid strip, 10uL of liquid in the Gamete liquid strip is sucked away, and 10uL of initial sperm is injected, and centrifugal sperm smearing is performed through a preset time period to obtain a centrifugal sperm smearing; Step 3, preliminary screening of sperm: under an inverted microscope, all initial normal sperm in the centrifugal sperm smearing are selected by a preset rule using an ICSI needle, and are added to the PVP liquid drop to obtain an initial single-layer motile sperm.

3. The method of claim 1, wherein the method is a high precision method of sperm morphology screening. The first initial single-layer motile sperm image is subjected to a second magnification operation to obtain a second initial single-layer motile sperm image according to the magnified initial single-layer motile sperm map, including: The first initial single-layer motile sperm image and the magnified initial single-layer motile sperm image are respectively substituted into a pre-trained VAE encoder to obtain a low-resolution latent vector and a high-resolution latent vector; Noise is added to the low-resolution latent vector and the high-resolution latent vector respectively to obtain a low-resolution noise latent vector and a high-resolution noise latent vector; The high-resolution noise latent vector is updated according to the low-resolution noise latent vector to obtain an updated high-resolution latent vector; A mean square error loss is generated according to the low-resolution latent vector and the updated high-resolution latent vector; The magnified initial single-layer motile sperm image is subjected to gradient update according to the mean square error loss to obtain a second initial single-layer motile sperm image.

4. The high precision sperm morphology screening method according to claim 3, wherein, The high-resolution noise latent vector is updated according to the low-resolution noise latent vector to obtain an updated high-resolution latent vector, including: The first initial single-layer active sperm image is segmented by a preset first region size to obtain a labeled sub-initial single-layer active sperm image set, and a large language model is used to generate a corresponding text prompt for each sub-initial single-layer active sperm image; The low-resolution noise latent vector and the text prompt corresponding to each sub-initial single-layer active sperm image are substituted into a preset Unet network to obtain a low-resolution attention score corresponding to each sub-initial single-layer active sperm image; The high-resolution latent vector is segmented by a preset second region size to obtain a labeled sub-high-resolution latent vector set; the labels in the sub-high-resolution latent vector set correspond one-to-one to the labels in the sub-initial single-layer active sperm image set; The text prompt, the low-resolution attention score, and the sub-high-resolution latent vector corresponding to the same label are substituted into an attention synthesizer for denoising to obtain a sub-updated high-resolution latent vector corresponding to the label; The sub-updated high-resolution latent vectors corresponding to all labels are combined to obtain an updated high-resolution latent vector.

5. The method of claim 1, wherein the method is a high precision method of sperm morphology screening. The second initial single-layer active sperm image and the structure information image are substituted into a single residual block to obtain a basic feature and a structure basic feature, including: The second initial single-layer active sperm image and the structure information image are substituted into a single residual block to obtain a first feature and a second feature; The first feature and the second feature are compressed by average pooling and maximum pooling, and are combined to obtain a fusion feature; The fusion feature, the first feature, and the second feature are spliced to obtain a joint feature; The joint feature is substituted into a 5x5 convolution layer plus Sigmoid activation to generate a first weight and a second weight; the first weight is multiplied by the first feature to obtain a basic feature; and the second weight is multiplied by the second feature to obtain a structure basic feature.

6. The method of claim 1, wherein the method is a high precision method of sperm morphology screening. The basic feature and the structure basic feature are substituted into a multi-scale sampling model to obtain a low-scale morphological feature, a medium-scale morphological feature, and a high-scale morphological feature, including: The basic feature and the structure basic feature are fused to obtain a fused basic feature; The fused basic feature is split into two branches by a 1x1 convolution to obtain a first sub-fused basic feature and a second sub-fused basic feature; The first sub-fused basic feature is convolved by a first convolution kernel to obtain a first sub-convolution feature; and the second sub-fused basic feature is convolved by a second convolution kernel to obtain a second sub-convolution feature; The second sub-convolution feature is subjected to 3x3 convolution plus 1x1 convolution plus Sigmoid activation to obtain a second sub-convolution attention; The second sub-convolution attention is multiplied by the second sub-convolution feature to obtain a second sub-attention feature; The second sub-attention feature and the first sub-convolution feature are spliced and then subjected to 3x3 convolution and residual connection to obtain an enhanced multi-stage feature; the multi-stage feature includes a low-scale morphological feature, a medium-scale morphological feature, and a high-scale morphological feature.

7. The method of claim 1, wherein the method is a high precision method of sperm morphology screening. The first magnification operation is implemented by a microscope; the microscope uses a 60x objective lens and a 10x eyepiece.

8. A high-precision sperm morphology screening system for implementing a high-precision sperm morphology screening method according to any one of claims 1 to 7, characterized in that, The system comprises: A filtering module configured to obtain raw semen, and filter the raw semen to obtain initial semen; An initial single-layer active sperm screening module is configured to spread initial semen into a preset single-layer sperm liquid strip to obtain initial single-layer active sperm; A first amplification module is configured to perform a first amplification operation on the initial single-layer active sperm to obtain a first initial single-layer active sperm image; An amplified initial single-layer active sperm image generation module is configured to amplify the first initial single-layer active sperm image to a target resolution to obtain an amplified initial single-layer active sperm image; A second amplification module is configured to perform a second amplification operation on the first initial single-layer active sperm image according to the amplified initial single-layer active sperm image to obtain a second initial single-layer active sperm image; A structural information image generation module is configured to obtain a structural information image by using a Sobel operator for the second initial single-layer active sperm image; A high-precision sperm morphology selection module is configured to input the second initial single-layer active sperm image and the structural information image into a high-precision sperm morphology selection model to obtain target high-quality sperm.

Images