A sperm image recognition system for spermatogenic disorders based on deep learning

Abstract

The present invention discloses a sperm image recognition system for spermatogenesis disorders based on deep learning, which relates to the field of sperm image recognition technology, including an image recognition management center, wherein the image recognition management center is communicatively connected with an image acquisition module, a sperm positioning module, a feature extraction module, a motion feature analysis module, a sperm morphology classification module, a sperm state comprehensive recognition module, and a result display feedback module, wherein electrical signals are connected between each module. The present invention learns and extracts morphological and motion features from a large amount of sperm image data through deep learning technology, which not only improves the accuracy of sperm classification, but also significantly reduces the time and error of manual operation, and can accurately identify subtle differences in sperm, greatly improving diagnostic efficiency, enabling doctors to obtain more accurate diagnostic results in a shorter time, and provide patients with more timely treatment recommendations.

Description

Technical Field

[0001] The present invention relates to the technical field of sperm image recognition, and in particular to a sperm image recognition system for spermatogenesis disorders based on deep learning. Background Art

[0002] Sperm morphology assessment plays an important role in the diagnosis and treatment of male infertility. Abnormal sperm morphology is one of the important causes of male infertility. Therefore, accurate and efficient assessment of sperm morphology is of great significance for guiding clinical treatment. Deep learning, as an important branch of artificial intelligence, has achieved remarkable results in image recognition, classification and other fields in recent years. Deep learning algorithms can automatically extract features from images and perform classification and recognition, greatly improving the intelligent level of image processing. With the continuous deepening of research on spermatogenesis disorders, people have gradually realized the close relationship between abnormal sperm morphology and spermatogenesis disorders.

[0003] In the existing technology, due to the high individual differences and diversity of sperm movement patterns, sperm morphological classification is affected by abnormal samples such as deformed sperm and dead sperm, which affects the accuracy of sperm morphological identification. Therefore, how to refine sperm morphological classification and integrate the morphological and movement characteristics of sperm to improve the accuracy of sperm image recognition of sperm with spermatogenesis is the problem we need to solve. To this end, we propose a sperm image recognition system for spermatogenesis with spermatogenesis based on deep learning. Summary of the Invention

[0004] The purpose of the present invention is to provide a sperm image recognition system for spermatogenesis disorders based on deep learning to solve the problems raised in the above background technology.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0006] A sperm image recognition system for spermatogenesis disorders based on deep learning includes an image recognition management center, wherein the image recognition management center is communicatively connected to an image acquisition module, a sperm positioning module, a feature extraction module, a motion feature analysis module, a sperm morphology classification module, a sperm status comprehensive recognition module, and a result display and feedback module, wherein the modules are electrically connected;

[0007] The image acquisition module is used to acquire sperm image data to ensure that the image is clear and complete for subsequent processing, and to pre-process the acquired sperm image data to improve image quality, reduce noise and interference in the image, and enhance image contrast and clarity;

[0008] The sperm localization module is used to locate the sperm head, identify the position of the sperm in the image, and segment the sperm using a semantic segmentation network to extract and accurately segment the various parts of the sperm, providing a clear image basis for subsequent morphological and motion feature analysis;

[0009] The feature extraction module uses a deep learning algorithm to extract features in the sperm image from the preprocessed sperm image, including morphological features and motion features, wherein the morphological features include head shape, tail shape and acrosome size, and the motion features include motion speed, motion trajectory and motion direction;

[0010] The motion characteristics analysis module evaluates the motion quality of sperm based on the motion characteristics of sperm, and further refines the motility status of sperm;

[0011] The sperm morphology classification module uses a deep learning model combined with existing semen analysis standards to perform sperm morphology classification based on the extracted sperm morphological characteristics and sperm motility quality assessment results, identifying normal sperm and abnormal sperm, where abnormal sperm include deformed sperm and dead sperm, thereby improving the accuracy and robustness of sperm morphology classification;

[0012] The sperm status comprehensive identification module integrates sperm morphology and motility characteristics, uses a weighted algorithm to assign a comprehensive score to each sperm, and performs graded identification of spermatogenic disorders, comprehensively clarifying sperm quality and providing a more accurate spermatogenic disorder assessment. Based on the classification and motility analysis results, the patient's spermatogenic status and potential fertility level are clarified;

[0013] The result display and feedback module displays the classification results to the user in a graphical and report-like form and provides a feedback mechanism for the user to make corrections and confirmations, thereby improving the user-friendliness and interactivity of the system and enabling the user to intuitively understand the classification results and make necessary adjustments.

[0014] A further improvement of the technical solution of the present invention is that the image acquisition module specifically includes:

[0015] Mix the semen sample with a diluent in appropriate proportions to reduce sperm density for ease of subsequent image acquisition and analysis; stain the sperm with a contrast dye to enhance the contrast and clarity of the sperm and facilitate identification of sperm morphological characteristics; and place the processed sample on a glass slide to ensure that the sample is evenly distributed and avoid overlap or accumulation so that the camera can clearly capture the sperm image;

[0016] Using a high-resolution camera to collect sperm images based on multiple angles and time points, thereby capturing the required sperm image data from the sperm sample;

[0017] Preprocessing the captured sperm image data includes preprocessing steps of grayscale conversion, noise removal, image enhancement, and image sharpening, wherein grayscale conversion is performed on color images to reduce the amount of data to be processed and improve processing speed, and frequency domain or spatial domain methods are used to enhance the image. The frequency domain method converts the image into the frequency domain through two-dimensional Fourier transform, and then uses low-pass filtering or high-pass filtering to enhance the signal. The spatial domain method directly operates on image pixels and uses histogram equalization, contrast stretching, and other methods to enhance the contrast and clarity of the image. A filter method is used to remove noise and interference in the image, effectively smooth the image, and reduce the impact of noise on subsequent processing. The image is sharpened to enhance the edge and detail features of the image to ensure improved image quality, reduce noise and interference, and enhance the contrast and clarity of the image.

[0018] The pre-processed sperm image data are integrated to form a complete image data set, and the integrated image data are temporarily stored in the data warehouse for further analysis and processing.

[0019] A further improvement of the technical solution of the present invention is that the sperm positioning module specifically includes:

[0020] Preprocessed sperm image data is retrieved from the data warehouse. The Faster R-CNN algorithm is used to locate and mark sperm heads. The preprocessed sperm images are fed into the Faster R-CNN model. Faster R-CNN generates candidate regions through its Region Proposal Network (RPN). The model then combines classification and regression branches to locate and mark the sperm heads with bounding boxes, thereby obtaining a sperm image dataset with labeled sperm heads. Each sperm head is accurately marked.

[0021] After the sperm head is successfully located, a U-Net-based semantic segmentation network is selected to segment the sperm image. The U-Net network is trained using a well-labeled sperm image dataset (containing the sperm head bounding box and corresponding segmentation labels) to map the located sperm head region to the semantic segmentation network. The U-Net, through its encoder-decoder architecture and skip connections, learns the multi-scale features of the sperm image and maps the sperm head region to the segmentation network.

[0022] The located sperm head region is input into the trained U-Net network, which outputs a segmentation map of the same size as the input image. Based on the output segmentation map of the U-Net network, each pixel is classified as a part of the sperm, including the head, midsection, and tail. The segmentation result is then fine-tuned to extract clear sperm contours and images of each part to ensure that all parts of the sperm are accurately and completely segmented.

[0023] A further improvement of the technical solution of the present invention is that the feature extraction module specifically includes:

[0024] Receive the segmented sperm image from the sperm localization module, ensure that the sperm head, mid-piece, and tail are clearly visible in the image, and identify the features in the sperm image, namely morphological features and motion features, and then prepare a sperm image dataset containing diverse shape features and motion features;

[0025] According to the type of features to be extracted, the corresponding deep learning model is selected for training. For morphological features, the convolutional neural network model is selected. For motion features, the convolutional neural network model and the recurrent neural network model are combined. The convolutional neural network model is used to capture spatial features, and the recurrent neural network model is used to capture time series features.

[0026] A convolutional neural network model was trained using a sperm image dataset containing diverse shape features. The convolutional neural network model was then used to extract head shape features, tail morphology features, and acrosome size features. Head shape features included head outline, aspect ratio, and area; tail morphology features included length, curvature, and thickness variation; and acrosome size features included acrosome volume and its size ratio relative to other parts of the head.

[0027] Convolutional neural network models and recurrent neural network models are used to continuously capture and analyze the position changes of the same sperm at different time points, and then extract sperm motion characteristics including motion speed characteristics, motion trajectory characteristics and motion direction characteristics. Specifically, by calculating the distance moved by the sperm in a unit time, its motion speed including average speed, instantaneous speed and path speed is obtained. The complexity and directionality of the sperm motion path are analyzed, and different trajectory patterns including linear motion, curvilinear motion and spiral motion are identified. By analyzing the motion direction of the sperm at different time points, the directional characteristics are extracted.

[0028] The extracted morphological features and motion features are integrated into a feature vector, and for time series data, the feature vectors at different time points are combined into a feature sequence to represent the dynamic characteristics of sperm, thereby forming a complete sperm feature sequence.

[0029] A further improvement of the technical solution of the present invention is that the motion feature analysis module specifically includes:

[0030] receiving a sperm image dataset containing sperm morphology and movement characteristics, and obtaining processed sperm movement data, including sperm position, velocity, and trajectory information;

[0031] For the motion speed feature, the position of sperm in consecutive image frames is tracked, the distance of position change between adjacent frames is calculated, and then the distance moved by sperm in unit time is calculated to evaluate its motion speed. The motion speed feature includes average speed, instantaneous speed and path speed. The average speed is the average speed of sperm along the straight line, the instantaneous speed is the instantaneous speed of sperm along the actual motion trajectory, and the path speed is the speed of sperm along the average path.

[0032] For motion trajectory characteristics, the complexity and directionality of the sperm movement path are analyzed. Among them, the motion trajectory characteristics include linearity and oscillation. Linearity is the ratio of the calculated average speed to the instantaneous speed, which reflects the linearity of sperm movement. Oscillation is the ratio of the calculated path speed to the instantaneous speed, which reflects the oscillation of sperm movement. Sperm that move in a straight line indicates that its motion trajectory is stable, while sperm that move in a curve or oscillate in place have motion disorders.

[0033] For the movement direction characteristics, we analyze whether the movement direction of sperm is consistent at different time points. The movement direction characteristics include direction change rate and direction consistency. The direction change rate is to calculate the change rate of the sperm movement direction at different time points. The direction consistency is to evaluate whether the sperm movement direction is consistent. Sperm with stable direction are more likely to successfully reach the fertilization site.

[0034] Based on the extracted sperm movement speed characteristics, movement trajectory characteristics and movement direction characteristics, the sperm motility value is calculated to evaluate the movement quality of the sperm. The higher the sperm motility value, the better the sperm movement quality.

[0035] Based on the evaluation results of sperm motility quality, the classification standard of sperm motility grade is preset, and the sperm motility state is divided into different motility grades according to the sperm motility value, namely A, B, C and D. Among them, the sperm motility of grade A motility grade is the strongest, and the sperm motility of grade D motility grade is the worst. Among them, the average speed ranges from 10-30μm / s, the path speed ranges from 25-50μm / s, the instantaneous speed ranges from 20-100μm / s, the linearity ranges from 0-100%, and the higher the value, the more linear the movement. The swing ranges from 0-100%, and the higher the value, the more complex the movement trajectory. The direction change rate ranges from 0-1, and the lower the value, the smaller the direction change. The direction consistency ranges from 0-1, and the higher the value, the more consistent the direction.

[0036] A further improvement of the technical solution of the present invention is that the classification criteria for the sperm motility levels are:

[0037] Class A vitality level is fast forward motion: average speed ≥ 25 μm / s, linearity ≥ 80%, and high directional consistency;

[0038] Level B vitality level is medium-speed forward movement: 10 μm / s ≤ average speed < 25 μm / s, 60% ≤ linearity < 80%, and medium directional consistency;

[0039] Grade C vitality level is slow or non-forward movement: average speed <10 μm / s, linearity <60%, and frequent changes in direction;

[0040] D-level activity level is inactivity: average speed ≈ 0 μm / s, no obvious movement trajectory.

[0041] A further improvement of the technical solution of the present invention is that the sperm morphology classification module specifically includes:

[0042] The existing sperm morphology database HSID (Human Semen Image Database) was searched to collect sperm images of normal morphology from multiple healthy donors, as well as images of deformed sperm (abnormal head and tail morphology) and dead sperm (lack of motility or obvious morphological abnormalities). Sperm images were annotated using image annotation software (LabelImg). The annotation information included sperm type (normal, deformed, dead) and existing morphological parameters (head shape, tail length, degree of curvature) to clearly distinguish normal sperm from abnormal sperm.

[0043] The labeled data was divided into training and test sets to ensure data balance, that is, the number of samples in each category was similar. The training set data was combined with a convolutional neural network model to train a sperm morphology classification model, enabling it to extract features from sperm images and classify them. The deep learning model was used to extract parameters associated with morphological characteristics and potential motility characteristics from sperm images, including sperm head shape, tail length, curvature, and motility.

[0044] The sperm image to be classified is input into a trained sperm morphology classification model, which automatically extracts parameters associated with morphological features and potential motion features in the sperm image, identifies normal sperm and abnormal sperm, and then classifies the sperm. The classification results include normal sperm, deformed sperm and dead sperm.

[0045] A further improvement of the technical solution of the present invention is that the process of classifying sperm is:

[0046] A convolutional neural network is used to extract a feature vector F from the sperm image. The features include head shape, tail length, curvature, and motility.

[0047] Based on the large amount of annotated sperm image data collected, sperm types are classified into three categories: normal, deformed, and dead sperm. The annotated data is then combined to train a sperm morphology classification model to learn the characteristic distribution of each category.

[0048] The probability that the input feature vector F belongs to the pth class of sperm is calculated based on the Gaussian distribution to analyze the degree of match between the input feature and the feature distribution of each class. The closer the feature value is to the mean of the sperm class, the higher the probability. The proportional relationship between the feature value and the mean is calculated through the radical formula to enhance the sensitivity to abnormal features and further distinguish different classes through the proportional relationship.

[0049] Using arg max p The category p with the highest comprehensive score is selected as the classification result C(F), and the input image is further classified as normal sperm, deformed sperm or dead sperm.

[0050] A further improvement of the technical solution of the present invention is that the sperm status comprehensive identification module specifically includes:

[0051] Sperm morphology classification results are synthesized, and sperm morphology characteristics including head shape, tail length, and curvature, and motion characteristics including average speed, linearity, and directional consistency are received from the sperm morphology classification module and the motion characteristic analysis module;

[0052] According to the importance of morphological and movement characteristics, a corresponding weight is assigned to each feature, among which the weight of head shape is 0.3, the weight of tail length is 0.2, the weight of curvature is 0.1, the weight of average speed is 0.2, the weight of linearity is 0.1, and the weight of directional consistency is 0.1. A weighted algorithm is used to assign a comprehensive score to each sperm;

[0053] Based on clinical experience and research data, different levels of spermatogenic disorder thresholds are set, and combined with the comprehensive score of sperm status, they are classified into different spermatogenic status levels, including A-level spermatogenic status level, B-level spermatogenic status level, C-level spermatogenic status level and D-level spermatogenic status level. Among them, the comprehensive score decreases step by step from A to D.

[0054] According to the spermatogenesis status grade identification results, the patient's fertility grade is determined, which are high fertility grade, medium fertility grade and low fertility grade. Among them, the proportion of sperm with grade A spermatogenesis status grade in the high fertility grade is high, indicating that the patient has strong fertility. The proportion of sperm with grade B and grade C spermatogenesis status grade in the medium fertility grade is high, indicating medium fertility. The proportion of sperm with grade D spermatogenesis status grade in the low fertility grade is high, indicating that the patient has poor fertility. A comprehensive evaluation report including sperm quality, comprehensive status score, spermatogenesis status grade and fertility grade is then generated to provide a basis for clinical diagnosis and treatment.

[0055] A further improvement of the technical solution of the present invention is that the result display feedback module specifically includes:

[0056] Extract classification results from the system, including sperm morphology classification, motility characteristics analysis, comprehensive score, spermatogenesis grade and fertility grade information, and present the sperm classification results to the user in an intuitive graphical form;

[0057] The sperm quality, comprehensive score, spermatogenesis grade, and fertility grade are integrated into a detailed report with embedded graphical presentations. The sperm quality assessment lists the sperm morphology and motility characteristics, along with the corresponding comprehensive score. The spermatogenesis grade (A, B, C, D) is clearly defined, with an explanation of each grade. The patient's fertility is then assessed based on the spermatogenesis grade (high, medium, or low fertility).

[0058] Feedback buttons are set up in the graphical interface and reports to collect user feedback, and then the collected feedback is analyzed to find out the deficiencies in the system. Based on the analysis results, the system is optimized and improved to improve the accuracy and user-friendliness of the classification results.

[0059] Due to the adoption of the above technical solution, the present invention has the following technical advancements compared to the prior art:

[0060] 1. The present invention provides a sperm image recognition system for spermatogenesis disorders based on deep learning. By using deep learning technology to learn and extract morphological and motion features from a large amount of sperm image data, it not only improves the accuracy of sperm classification, but also significantly reduces the time and errors of manual operation. At the same time, it can accurately identify subtle differences in sperm, greatly improving diagnostic efficiency, enabling doctors to obtain more accurate diagnostic results in a shorter time and provide patients with more timely treatment recommendations.

[0061] 2. The present invention provides a sperm image recognition system for spermatogenesis disorders based on deep learning. Through deep learning technology, the system not only provides detailed classification results of sperm morphology and movement characteristics, but also combines clinical experience and research data to assign a comprehensive score to each sperm, and classify it into different spermatogenesis status levels, thereby generating a comprehensive evaluation report including sperm quality, comprehensive score, spermatogenesis status level and fertility level, which provides an important basis for doctors to formulate personalized treatment plans, thereby more effectively improving patients' fertility. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0063] Figure 1Schematic diagram of the system function modules of the present invention;

[0064] Figure 2 Schematic diagram of the working process of the motion feature analysis module of the present invention;

[0065] Figure 3 Schematic diagram of the workflow of the sperm morphology classification module of the present invention;

[0066] Figure 4 Schematic diagram of the workflow of the sperm status comprehensive identification module of the present invention. DETAILED DESCRIPTION

[0067] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0068] Example 1, as Figure 1 As shown, the present invention provides a sperm image recognition system for spermatogenesis disorders based on deep learning, including an image recognition management center, which is communicatively connected to an image acquisition module, a sperm positioning module, a feature extraction module, a motion feature analysis module, a sperm morphology classification module, a sperm status comprehensive recognition module, and a result display and feedback module, wherein the modules are electrically connected;

[0069] The image acquisition module is used to acquire sperm image data, ensure that the image is clear and complete, and facilitate subsequent processing. It also pre-processes the acquired sperm image data to improve image quality, reduce noise and interference in the image, and enhance the contrast and clarity of the image. It mixes the semen sample with the diluent in proportion to reduce the sperm density to facilitate subsequent image acquisition and analysis. It uses a contrast dye to stain the sperm to enhance the contrast and clarity of the sperm and facilitate the identification of the morphological characteristics of the sperm. The processed sample is placed on a slide to ensure that the sample is evenly distributed and avoids overlapping or accumulation so that the camera can In order to clearly capture sperm images, a high-resolution camera is used to collect sperm images based on multiple angles and time points, and then the required sperm image data is captured from the sperm sample to ensure that the subtle features and dynamic changes of the sperm sample can be captured. When acquiring images, the camera parameters need to be adjusted, including shutter speed, focus setting, ISO sensitivity and aperture setting, to determine the best image quality. The appropriate shutter speed is set to capture the dynamic changes of sperm and avoid motion blur. The focus setting is performed to ensure that the camera focuses accurately so that the sperm image is clear. The ISO value is adjusted to suit In response to different lighting conditions, ensure that the image brightness is moderate, adjust the aperture size to control the depth of field, ensure that all parts of the sperm can be clearly imaged, and preprocess the captured sperm image data, including grayscale, noise removal, image enhancement and image sharpening preprocessing steps. Among them, grayscale processing is performed on color images to reduce the amount of data required for processing and increase processing speed. Frequency domain method or spatial domain method is used to enhance the image. The frequency domain method converts the image into the frequency domain through two-dimensional Fourier transform, and then uses low-pass filtering or high-pass filtering to enhance the signal. The spatial domain method directly Perform operations on image pixels, using methods such as histogram equalization and contrast stretching to enhance the contrast and clarity of the image, using filter methods to remove noise and interference in the image, effectively smoothing the image and reducing the impact of noise on subsequent processing, sharpening the image to enhance the edge and detail features of the image to ensure improved image quality, reduce noise and interference, and enhance the contrast and clarity of the image, integrate the pre-processed sperm image data to form a complete image data set, and temporarily store the integrated image data in the data warehouse for further analysis and processing;

[0070] The sperm positioning module is used to locate the sperm head, identify the position of the sperm in the image, and use the semantic segmentation network to segment the sperm, extract the various parts of the sperm, and accurately segment the various parts of the sperm, providing a clear image basis for subsequent morphological and motion feature analysis. The preprocessed sperm image data is called from the data warehouse, and the sperm head is located and marked using the Faster R-CNN algorithm. The preprocessed sperm image is input into the Faster R-CNN model. Faster R-CNN generates candidate regions through its region proposal network (RPN), and combines classification and regression branches to locate and mark the sperm head with a bounding box, thereby obtaining a sperm image dataset with annotated sperm heads. Each sperm head is accurately marked. After the sperm head is successfully located, the semantic segmentation network based on U-Net is selected to segment the sperm image. The labeled sperm image dataset (including the bounding box of the sperm head and the corresponding segmentation label) is used to train the U-Net network so that the located sperm head area is mapped to the semantic segmentation network. Among them, U-Net Through its encoder-decoder architecture and skip connections, it learns multi-scale features of sperm images and maps the sperm head region to a segmentation network. The located sperm head region is then fed into a trained U-Net network, which outputs a segmentation map of the same size as the input image. Based on the U-Net output, each pixel is classified as a sperm part, including the head, midsection, and tail. The segmentation results are then fine-tuned to extract clear sperm outlines and images of each part, ensuring that all sperm parts are accurately and completely segmented.

[0071] Furthermore, the process of classifying each pixel as a part of the sperm based on the output segmentation map of the U-Net network is:

[0072] In the sperm image segmentation task, the input image is determined to be I(x, y), where x and y represent the horizontal and vertical coordinates of the image, respectively. For each pixel (x, y), its distance d(x, y) from the center of the sperm is calculated. The probability that the pixel (x, y) belongs to each sperm part is calculated using Gaussian distribution and trigonometric functions. The probabilities of all sperm parts are summed to obtain a comprehensive probability. The sperm part with the largest probability is selected as the classification result of the pixel (x, y), which is S(x, y). Each pixel (x, y) in the input image is then accurately classified as the head, middle or tail of the sperm. S(x, y) represents the classification result of the pixel with coordinates (x, y) in the segmentation map. S(x, y) indicates which part of the sperm the pixel belongs to (head, middle or tail). The classification result is obtained by calculating the probability of each sperm part and selecting the category with the largest probability, where argmax kIndicates the selection of the k value that maximizes the expression in the brackets, that is, the sperm part with the highest probability is selected as the classification of the pixel. Therefore, the value of S(x, y) is the k value that maximizes the probability, which represents the classification result of the pixel. If the value range of k is {1, 2, 3}, corresponding to the head, middle and tail of the sperm respectively, then S(x, y) = 1 means that the pixel is classified as the head of the sperm, S(x, y) = 2 means that the pixel is classified as the middle of the sperm, and so on. S(x, y) = 3 means that the pixel is classified as the tail of the sperm, thus achieving accurate segmentation of the sperm image;

[0073] The calculation formula of S(x, y) is as follows:

[0074]

[0075] Where S(x, y) is the classification result of pixel (x, y), k is the different parts of sperm (head, midsection, tail), M is the number of features used for classification, u is the index of the feature used for classification, μ is the k and σ represent the mean and standard deviation of the k-th sperm fraction, respectively, and λ is the wavelength of the periodic characteristic. is the phase offset of the k-th sperm part, d(x, y) is the distance from the pixel (x, y) to the sperm center, (x c ,y c ) are the coordinates of the sperm center;

[0076] The feature extraction module uses a deep learning algorithm to extract features from the preprocessed sperm image, including morphological features and motion features, wherein the morphological features include head shape, tail shape and acrosome size, and the motion features include motion speed, motion trajectory and motion direction. The segmented sperm image is received from the sperm positioning module to ensure that the head, midsection and tail of the sperm in the image are clearly visible, and the features in the sperm image are determined, which are morphological features and motion features, respectively. Then, a sperm image dataset containing various shape features and motion features is prepared. According to the type of features to be extracted, the corresponding deep learning model is selected for training. For morphological features, a convolutional neural network model is selected. For motion features, a convolutional neural network model and a recurrent neural network model are combined. Among them, the convolutional neural network model is used to capture spatial features, and the recurrent neural network model is used to capture time series features. The sperm image dataset containing various shape features is used to train the convolutional neural network model, and then the convolutional neural network model is used to extract head shape features and tail morphological features. and acrosome size characteristics, wherein the head shape characteristics include the head outline, aspect ratio and area, the tail morphological characteristics include length, curvature and thickness changes, and the acrosome size characteristics include the volume of the acrosome and the size ratio relative to other parts of the head. The convolutional neural network model and the recurrent neural network model are used to continuously capture and analyze the position changes of the same sperm at different time points, and then extract the sperm motion characteristics including motion speed characteristics, motion trajectory characteristics and motion direction characteristics. Among them, by calculating the distance moved by the sperm in unit time, its motion speed including average speed, instantaneous speed and path speed is obtained, the complexity and directionality of the sperm motion path are analyzed, and different trajectory patterns including linear motion, curved motion and spiral motion are identified. By analyzing the motion direction of the sperm at different time points, the directional characteristics are extracted, and the extracted morphological characteristics and motion characteristics are integrated into a feature vector. In addition, for time series data, the feature vectors at different time points are combined into a feature sequence to represent the dynamic characteristics of the sperm, thereby forming a complete sperm feature sequence;

[0077] The motion characteristics analysis module combines the motion characteristics of sperm to evaluate the movement quality of sperm and further refine the sperm motility status;

[0078] The sperm morphology classification module uses a deep learning model combined with existing semen analysis standards to classify sperm morphology based on extracted sperm morphological characteristics and sperm motility quality assessment results. It can identify normal sperm and abnormal sperm, including deformed sperm and dead sperm, to improve the accuracy and robustness of sperm morphology classification.

[0079] The sperm status comprehensive identification module integrates sperm morphology and motility characteristics, uses a weighted algorithm to assign a comprehensive score to each sperm, and conducts graded identification of spermatogenic disorders. This module comprehensively determines sperm quality and provides a more accurate assessment of spermatogenic disorders. Based on the classification and motility analysis results, it also clarifies the patient's spermatogenic status and potential fertility level.

[0080] The result display and feedback module displays the classification results to users in graphical and report forms, and provides a feedback mechanism for users to make corrections and confirmations, thereby improving the user-friendliness and interactivity of the system and enabling users to intuitively understand the classification results and make necessary adjustments.

[0081] Example 2, as Figure 2 As shown, based on Example 1, the present invention provides a technical solution: preferably, the motion feature analysis module specifically includes:

[0082] Receive a sperm image dataset containing sperm morphology and motion characteristics, and obtain processed sperm motion data, including sperm position, speed and trajectory information. For the motion speed feature, by tracking the position of sperm in consecutive image frames, calculate the distance of position change between adjacent frames, and then calculate the distance moved by sperm in unit time to evaluate its motion speed. Among them, the motion speed feature includes average speed, instantaneous speed and path speed. The average speed is the average speed of sperm along the straight line direction, the instantaneous speed is the instantaneous speed of sperm along the actual motion trajectory, and the path speed is the speed of sperm along the average path. For the motion trajectory characteristics, the complexity and directionality of the sperm motion path are analyzed. Among them, the motion trajectory characteristics include linearity and oscillation. Linearity is the ratio of the calculated average speed to the instantaneous speed, which reflects the linearity of the sperm motion. Oscillation is the ratio of the calculated path speed to the instantaneous speed, which reflects the oscillation degree of the sperm motion. Sperm that moves in a straight line indicates that its motion trajectory is stable, while sperm that moves in a curve or oscillates in place has motion disorders. For the motion direction characteristics, whether the sperm motion direction is consistent at different time points is analyzed. Among them, the motion direction characteristics include the direction change rate and direction consistency. The direction change rate is used to calculate the change rate of the sperm's movement direction at different time points. The direction consistency is used to evaluate whether the sperm's movement direction is consistent. Sperm with stable directionality are more likely to successfully reach the fertilization site. According to the extracted sperm movement speed characteristics, movement trajectory characteristics and movement direction characteristics, the sperm motility value is calculated to evaluate the movement quality of the sperm. The higher the sperm motility value, the better the sperm movement quality. Based on the evaluation results of the sperm movement quality, the sperm motility level classification standard is preset, and the sperm motility state is divided into different motility levels in combination with the sperm motility value, namely A The motility of sperm is graded as A, B, C, and D, with grade A being the most motile and grade D being the least motile. The average velocity ranges from 10-30 μm / s, the path velocity ranges from 25-50 μm / s, the instantaneous velocity ranges from 20-100 μm / s, the linearity ranges from 0-100%, with higher values ​​indicating more linear motion, the oscillation ranges from 0-100%, with higher values ​​indicating more complex motion trajectories, the direction change rate ranges from 0-1, with lower values ​​indicating smaller direction changes, and the direction consistency ranges from 0-1, with higher values ​​indicating more consistent directions.

[0083] The calculation formula of sperm motility is as follows:

[0084]

[0085] Where, SMI is the sperm motility value, V i is the value of the i-th velocity feature, including average velocity, instantaneous velocity and path velocity, V max , iis the maximum normal value of the i-th velocity feature, LIN is linearity, which reflects the linearity of sperm movement, WOB is wiggling, which reflects the wiggling degree of sperm movement, T is the threshold of linearity and wiggling, which is used to adjust the nonlinear part in the formula, with a value of 50%, DC is directional consistency, which evaluates the consistency of sperm movement direction, with a value range of 0 to 1, and DV is the direction change rate, which calculates the change rate of sperm movement direction at different time points, with a value range of 0 to 1. When all movement characteristics perform well, the SMI value will be close to or exceed 1. When When the motility characteristics are poor, the SMI value will be close to 0. The higher the SMI value, the better the sperm motility quality and the higher the fertilization potential. The lower the SMI value, the sperm may have motility disorders. The higher the speed characteristics, the higher the SMI value. The higher the linearity, the higher the SMI value, indicating that the sperm motion trajectory is more stable. The lower the oscillation, the higher the SMI value, indicating that the sperm motion trajectory is closer to a straight line. The higher the directional consistency, the higher the SMI value, indicating that the sperm motion direction is more stable. The lower the directional change rate, the higher the SMI value, indicating that the sperm motion direction changes less.

[0086] In addition, the classification criteria for sperm motility are:

[0087] Class A vitality level is fast forward motion: average speed ≥ 25 μm / s, linearity ≥ 80%, and high directional consistency;

[0088] Level B vitality level is medium-speed forward movement: 10 μm / s ≤ average speed < 25 μm / s, 60% ≤ linearity < 80%, and medium directional consistency;

[0089] Grade C vitality level is slow or non-forward movement: average speed <10 μm / s, linearity <60%, and frequent changes in direction;

[0090] D-level activity level is inactivity: average speed ≈ 0 μm / s, no obvious movement trajectory.

[0091] Example 3, as Figure 3 、 Figure 4 As shown, based on Examples 1-2, the present invention provides a technical solution: preferably, the sperm morphology classification module specifically includes:

[0092] Search the existing sperm morphology database HSID (Human Semen Image Database), collect sperm images of normal morphology from multiple healthy donors, and collect images of deformed sperm and dead sperm. Use the image annotation software LabelImg to annotate the sperm images. The annotation information includes sperm type and existing morphological parameters, clearly distinguishing normal sperm from abnormal sperm. The annotated data is divided into a training set and a test set to ensure data balance, that is, the number of samples in each category is similar. Use the training set data in combination with the convolutional neural network model to train the sperm morphology classification model, so that it can extract features from sperm images and classify them. Use the deep learning model to extract parameters associated with morphological features and potential motion features from sperm images, including The model is trained to accurately identify normal and abnormal sperm, including the head shape, tail length, curvature and motility of sperm. The cross-entropy loss function is used for classification tasks. The accuracy, recall rate, F1 score and other indicators are used to evaluate the model performance on the validation set. The model structure, hyperparameters or data enhancement strategy are adjusted according to the evaluation results. Then, through continuous iteration and optimization of model parameters, the model is able to accurately identify normal and abnormal sperm. The sperm images to be classified are input into the trained sperm morphology classification model, and the parameters associated with the morphological features and potential motion features in the sperm images are automatically extracted to identify normal and abnormal sperm, and then the sperm are classified. The classification results include normal sperm, deformed sperm and dead sperm.

[0093] The process of sorting sperm is:

[0094] A convolutional neural network is used to extract feature vectors F from sperm images. The features include head shape, tail length, curvature, and motility. Based on a large amount of annotated sperm image data collected, sperm types are divided into three categories: normal, deformed, and dead sperm. The annotated data is combined to train a sperm morphology classification model and learn the feature distribution of each category (mean μ p，j and variance ), calculate input features based on Gaussian distribution

[0095] The probability that F belongs to the pth class of sperm is used to analyze the matching degree between the input features and the distribution of features of each class. The closer the eigenvalue is to the mean of the sperm of this class, the higher the probability. The proportional relationship between the eigenvalue and the mean is calculated by the radical formula to enhance the sensitivity to abnormal features. Different classes are further distinguished by the proportional relationship. The arg max is used p Select the category p with the highest comprehensive score as the classification result C(F), and then classify the input image into normal sperm, deformed sperm or dead sperm;

[0096] The calculation formula of the classification result C(F) is as follows:

[0097]

[0098] Where C(F) is the classification result, which indicates the sperm type that the input feature vector F is classified into, p is the sperm type index, p∈{1,2,3}, corresponding to normal sperm, deformed sperm and dead sperm respectively, and f j is the jth eigenvalue, m is the number of features, μ p，j is the mean value of the jth characteristic of the pth type sperm, is the variance of the p-th type sperm characteristics, α p is the weight coefficient, which is used to adjust the contribution of the radical part. Different types of sperm have different weights. If C(F) = 1, it means that the input sperm image is classified as normal sperm. If C(F) = 2, it means that the input sperm image is classified as deformed sperm. C(F) = 3, it means that the input sperm image is classified as dead sperm.

[0099] The sperm status comprehensive identification module specifically includes:

[0100] The results of sperm morphology classification are integrated, and the morphological characteristics of sperm including head shape, tail length, and curvature and the motion characteristics including average speed, linearity, and directional consistency are received from the sperm morphology classification module and the motion characteristic analysis module. According to the importance of morphological characteristics and motion characteristics, corresponding weights are assigned to each feature, among which the weight of head shape is 0.3, the weight of tail length is 0.2, the weight of curvature is 0.1, the weight of average speed is 0.2, the weight of linearity is 0.1, and the weight of directional consistency is 0.1. A weighted algorithm is used to assign a comprehensive score to each sperm. According to clinical experience and research data, different levels of spermatogenesis disorder thresholds are set, and combined with the comprehensive score of sperm status, they are classified into different spermatogenesis status levels, including A-level spermatogenesis status level, B-level spermatogenesis status level, C-level spermatogenesis status level and D-level spermatogenesis status level. Among them, the comprehensive score decreases step by step from A to D. The A-level spermatogenesis status level has the highest comprehensive score, and the sperm morphology is normal and the motility is strong. The spermatogenic function is good. The comprehensive score of grade B spermatogenic status is high, the sperm morphology and motility are moderate, and the spermatogenic function is good. The comprehensive score of grade C spermatogenic status is moderate, there are mild abnormalities in sperm morphology or motility, and the spermatogenic function is limited. The comprehensive score of grade D spermatogenic status is low, the sperm morphology and motility are significantly abnormal, and the spermatogenic function is poor. According to the spermatogenic status grade identification results, the patient's fertility grade is determined as high fertility grade, medium fertility grade, and low fertility grade. Among them, in the high fertility grade, the proportion of sperm in grade A spermatogenic status grade is high, indicating that the patient has strong fertility. In the medium fertility grade, the proportion of sperm in grade B and grade C spermatogenic status grade is high, and the fertility is moderate. In the low fertility grade, the proportion of sperm in grade D spermatogenic status grade is high, indicating that the patient has poor fertility. A comprehensive assessment report including sperm quality, comprehensive score, spermatogenic status grade, and fertility grade is generated to provide a basis for clinical diagnosis and treatment.

[0101] The formula for calculating the comprehensive condition score is as follows:

[0102]

[0103] Where P is the comprehensive score of the situation, ω g is the weight of the g-th feature, f g is the corresponding eigenvalue, N is the number of features;

[0104] The result display and feedback module specifically includes:

[0105] Extract classification results from the system, including sperm morphology classification, motion feature analysis, comprehensive score, spermatogenesis status grade and fertility grade information, and present the sperm classification results to users in an intuitive graphical form, using bar charts or pie charts to show the proportion of normal sperm, deformed sperm and dead sperm, draw the movement trajectory of sperm, mark parameters such as movement speed, linearity and direction consistency, and display indicators such as sperm classification accuracy, recall rate and F1 score to help users understand the performance of the model. Support users to view detailed information of individual sperm through the interactive interface, including morphological characteristics and motion characteristics, and present sperm quality, comprehensive score, spermatogenesis status grade and fertility grade. The sperm quality assessment lists the morphological and motility characteristics of sperm, as well as the corresponding comprehensive scores, clarifies the spermatogenic status levels (A, B, C, D), and explains the meaning of each level. According to the spermatogenic status level, the patient's fertility is assessed (high, medium, and low fertility levels). Feedback buttons are set in the graphical interface and report to collect user feedback, and then the collected feedback is analyzed to find out the deficiencies in the system. Based on the analysis results, the system is optimized and improved to improve the accuracy and user-friendliness of the classification results.

[0106] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A sperm image recognition system for spermatogenesis disorders based on deep learning, including an image recognition management center, characterized by: The image recognition management center is communicatively connected to an image acquisition module, a sperm positioning module, a feature extraction module, a motion feature analysis module, a sperm morphology classification module, a sperm status comprehensive identification module, and a result display and feedback module, wherein electrical signals are connected between the modules; The image acquisition module is used to acquire sperm image data and pre-process the acquired sperm image data; The sperm localization module is used to locate the sperm head, identify the position of the sperm in the image, and segment the sperm using a semantic segmentation network to extract the various parts of the sperm; The feature extraction module uses a deep learning algorithm to extract features in the sperm image from the preprocessed sperm image, including morphological features and movement features; The motion characteristics analysis module evaluates the motion quality of sperm based on the motion characteristics of sperm, and further refines the motility status of sperm; The sperm morphology classification module uses a deep learning model combined with existing semen analysis standards to classify sperm morphology based on the extracted sperm morphological characteristics and sperm motility quality assessment results, and identifies normal sperm and abnormal sperm, wherein abnormal sperm includes deformed sperm and dead sperm; The sperm status comprehensive identification module integrates the morphological and motion characteristics of sperm, uses a weighted algorithm to assign a comprehensive score to each sperm, performs graded identification of spermatogenic disorders, and clarifies the patient's spermatogenic status and potential fertility level based on the classification and motion analysis results. The sperm status comprehensive identification module specifically includes: integrating the results of sperm morphological classification, and receiving sperm morphological characteristics including head shape, tail length, and curvature, and motion characteristics including average speed, linearity, and directional consistency from the sperm morphological classification module and the motion characteristic analysis module; According to the importance of morphological and movement characteristics, a corresponding weight is assigned to each feature, among which the weight of head shape is 0.3, the weight of tail length is 0.2, the weight of curvature is 0.1, the weight of average speed is 0.2, the weight of linearity is 0.1, and the weight of directional consistency is 0.

1. A weighted algorithm is used to assign a comprehensive score to each sperm; Based on clinical experience and research data, different levels of spermatogenic disorder thresholds are set, and combined with the comprehensive score of sperm status, they are classified into different spermatogenic status levels, including A-level spermatogenic status level, B-level spermatogenic status level, C-level spermatogenic status level and D-level spermatogenic status level. Among them, the comprehensive score decreases step by step from A to D. Based on the spermatogenesis status identification results, the patient's fertility level is determined, which are high fertility level, medium fertility level, and low fertility level. Among them, the high fertility level has a high proportion of sperm with grade A spermatogenesis status, indicating that the patient has strong fertility. The medium fertility level has a high proportion of sperm with grade B and grade C spermatogenesis status, indicating medium fertility. The low fertility level has a high proportion of sperm with grade D spermatogenesis status, indicating that the patient has poor fertility. A comprehensive assessment report including sperm quality, comprehensive status score, spermatogenesis status level, and fertility level is then generated; The result display and feedback module displays the classification results to the user in a graphical and report form.

2. The deep learning-based sperm image recognition system for spermatogenesis disorders according to claim 1, characterized in that: The image acquisition module specifically includes: Mix the semen sample with the diluent in appropriate proportions, stain the sperm with a contrasting stain, and place the processed sample on a slide; Using a high-resolution camera to collect sperm images based on multiple angles and time points, thereby capturing the required sperm image data from the sperm sample; Preprocessing the captured sperm image data, including grayscale conversion, noise removal, image enhancement, and image sharpening; The pre-processed sperm image data are integrated to form a complete image data set, and the integrated image data are temporarily stored in the data warehouse.

3. The deep learning-based sperm image recognition system for spermatogenesis disorders according to claim 2, characterized in that: The sperm positioning module specifically includes: Retrieving preprocessed sperm image data from the data warehouse, using the Faster R-CNN algorithm to locate and mark sperm heads, then feeding the preprocessed sperm images into the Faster R-CNN model. Faster R-CNN generates candidate regions through its region proposal network, and combines classification and regression branches to locate and mark the sperm heads with bounding boxes, thereby obtaining a sperm image dataset with labeled sperm heads. After the sperm head is successfully located, a U-Net-based semantic segmentation network is selected to segment the sperm image. The U-Net network is trained using a well-labeled sperm image dataset so that the located sperm head region is mapped to the semantic segmentation network. The located sperm head region is input into the trained U-Net network, which outputs a segmentation map of the same size as the input image. Based on the output segmentation map of the U-Net network, each pixel is classified as a part of the sperm, including the head, midsection, and tail. The segmentation result is then fine-tuned to extract a clear sperm outline and images of each part.

4. The deep learning-based sperm image recognition system for spermatogenesis disorders according to claim 3, characterized in that: The feature extraction module specifically includes: Receive the segmented sperm image from the sperm localization module and determine the features in the sperm image, namely, morphological features and motion features, and then prepare a sperm image dataset containing diverse shape features and motion features; According to the type of features to be extracted, the corresponding deep learning model is selected for training. For morphological features, the convolutional neural network model is selected, and for motion features, the convolutional neural network model and the recurrent neural network model are combined; A convolutional neural network model was trained using a sperm image dataset containing diverse shape features. The convolutional neural network model was then used to extract head shape features, tail morphology features, and acrosome size features. Head shape features included head outline, aspect ratio, and area; tail morphology features included length, curvature, and thickness variation; and acrosome size features included acrosome volume and its size ratio relative to other parts of the head. Convolutional neural network models and recurrent neural network models are used to continuously capture and analyze the position changes of the same sperm at different time points, and then extract sperm motion characteristics including motion speed characteristics, motion trajectory characteristics and motion direction characteristics. Specifically, by calculating the distance moved by the sperm in a unit time, its motion speed including average speed, instantaneous speed and path speed is obtained. The complexity and directionality of the sperm motion path are analyzed, and different trajectory patterns including linear motion, curvilinear motion and spiral motion are identified. By analyzing the motion direction of the sperm at different time points, the directional characteristics are extracted. The extracted morphological features and motion features are integrated into a feature vector, and for time series data, the feature vectors at different time points are combined into a feature sequence to represent the dynamic characteristics of sperm, thereby forming a complete sperm feature sequence.

5. The deep learning-based sperm image recognition system for spermatogenesis disorders according to claim 4, characterized in that: The motion feature analysis module specifically includes: receiving a sperm image dataset containing sperm morphology and movement characteristics, and obtaining processed sperm movement data, including sperm position, velocity, and trajectory information; For the motion speed feature, the position of sperm in consecutive image frames is tracked, the distance of position change between adjacent frames is calculated, and then the distance moved by sperm in unit time is calculated to evaluate its motion speed. The motion speed feature includes average speed, instantaneous speed and path speed. The complexity and directionality of the sperm's motion trajectory were analyzed. These trajectory characteristics included linearity and oscillation. Linearity was calculated as the ratio of average velocity to instantaneous velocity, while oscillation was calculated as the ratio of path velocity to instantaneous velocity. Sperm that moved in a straight line indicated a stable trajectory, while sperm that moved in a curve or oscillated in place had motion disorders. For the movement direction characteristics, the consistency of sperm movement direction at different time points is analyzed. The movement direction characteristics include direction change rate and direction consistency. The direction change rate is used to calculate the change rate of sperm movement direction at different time points, and the direction consistency is used to evaluate whether the sperm movement direction is consistent. Based on the extracted sperm movement speed characteristics, movement trajectory characteristics and movement direction characteristics, the sperm motility value is calculated to evaluate the movement quality of the sperm; Based on the evaluation results of sperm motility quality, the sperm motility level classification standard is preset, and the sperm motility status is divided into different motility levels according to the sperm motility value, namely A, B, C and D. Among them, the sperm motility of grade A is the strongest, and the sperm motility of grade D is the worst.

6. The deep learning-based sperm image recognition system for spermatogenesis disorders according to claim 5, characterized in that: The classification criteria for the sperm motility grades are: Class A vitality level is fast forward motion: average speed ≥ 25 μm / s, linearity ≥ 80%, and high directional consistency; Level B vitality level is medium-speed forward movement: 10 μm / s ≤ average speed < 25 μm / s, 60% ≤ linearity < 80%, and medium directional consistency; Grade C vitality level is slow or non-forward movement: average speed <10 μm / s, linearity <60%, and frequent changes in direction; D-level activity level is inactivity: average speed ≈ 0 μm / s, no obvious movement trajectory.

7. The deep learning-based sperm image recognition system for spermatogenesis disorders according to claim 6, characterized in that: The sperm morphology classification module specifically includes: Search the existing sperm morphology database HSID and collect sperm images of normal morphology from multiple healthy donors, as well as images of deformed and dead sperm. Use image annotation software to annotate the sperm images with information including sperm type and existing morphological parameters to clearly distinguish normal sperm from abnormal sperm. The labeled data was divided into training and test sets. The training set data was combined with a convolutional neural network model to train a sperm morphology classification model. The deep learning model was used to extract parameters associated with morphological features and potential motility characteristics from sperm images, including sperm head shape, tail length, curvature, and motility. The sperm image to be classified is input into a trained sperm morphology classification model, which automatically extracts parameters associated with morphological features and potential motion features in the sperm image, identifies normal sperm and abnormal sperm, and then classifies the sperm. The classification results include normal sperm, deformed sperm and dead sperm.

8. The deep learning-based sperm image recognition system for spermatogenesis disorders according to claim 7, characterized in that: The process of classifying sperm is as follows: A convolutional neural network is used to extract a feature vector F from the sperm image. The features include head shape, tail length, curvature, and motility. Based on the large amount of annotated sperm image data collected, sperm types are classified into three categories: normal, deformed, and dead sperm. The annotated data is then combined to train a sperm morphology classification model to learn the characteristic distribution of each category. The probability that the input feature vector F belongs to the pth class of sperm is calculated based on the Gaussian distribution to analyze the degree of match between the input feature and the feature distribution of each class. The closer the feature value is to the mean of the sperm class, the higher the probability. The proportional relationship between the feature value and the mean is calculated through the radical formula to enhance the sensitivity to abnormal features and further distinguish different classes through the proportional relationship. Using arg max p The category p with the highest comprehensive score is selected as the classification result C(F), and the input image is further classified as normal sperm, deformed sperm or dead sperm.

9. The sperm image recognition system for spermatogenesis disorders based on deep learning according to claim 1, characterized in that: The result display feedback module specifically includes: Extract classification results from the system, including sperm morphology classification, motility characteristics analysis, comprehensive score, spermatogenesis grade and fertility grade information, and present the sperm classification results to the user in an intuitive graphical form; Integrate sperm quality, comprehensive score, spermatogenesis grade, and fertility grade into a detailed report with embedded graphical presentation. The sperm quality assessment lists sperm morphology and motility characteristics, along with the corresponding comprehensive score, identifies the spermatogenesis grade, explains the meaning of each grade, and assesses the patient's fertility based on the spermatogenesis grade. Set up feedback buttons in the graphical interface and reports to collect user feedback, then analyze the collected feedback to find out the shortcomings in the system, and optimize and improve the system based on the analysis results.

Images