Multi-source clinical information fusion-based mammal ovulation prediction method and system

Abstract

The embodiment of the invention discloses a mammal ovulation prediction method and system based on multi-source clinical information fusion, and the method comprises the steps: obtaining the medical image data of a target animal generated in a conventional clinical examination process, and carrying out the preprocessing of the obtained medical image data; performing analysis according to the follicle image to obtain reference ovulation remaining time and a reference ovulation state in a preset time window; analyzing according to the cervical image to obtain an ovulation stage evaluation result; analyzing according to the uterus image to obtain the corresponding ovulation condition for reflecting the state change of the uterus tissue; and generating a comprehensive ovulation prediction result of the target animal according to the ovulation remaining time, a reference ovulation state in a preset time window, an ovulation stage evaluation result, a result obtained by uterus image analysis and an artificial experience evaluation result. Therefore, an ovulation prediction result which is available in small samples, easy to deploy, high in interpretability, accurate in prediction result and stable can be provided based on conventional clinical data.

Description

Technical Field

[0001] The embodiments of the present invention relate to the fields of animal reproduction and medical image analysis technology, and in particular to a method and system for predicting ovulation in mammals based on the fusion of multi-source clinical information. Background Technology

[0002] In the routine reproductive management of large mammals (such as horses, cattle, and donkeys), clinical veterinarians typically rely on single or periodic ultrasound examinations, combined with observation of cervical condition and experience in manual palpation, to empirically determine the ovulation time of the target animal. Accurate determination of ovulation time in large mammals is a crucial aspect of reproductive management, directly impacting mating success rates and overall reproductive efficiency.

[0003] In clinical practice, rectal ultrasound is often used to observe changes in the size and morphology of follicles, supplemented by manual palpation and a comprehensive assessment of the cervix to estimate ovulation time. However, this type of assessment largely depends on the operator's experience level and lacks a unified quantitative standard.

[0004] Current clinical methods for determining and predicting ovulation in large mammals mainly fall into the following categories:

[0005] 1) Methods that rely solely on a single indicator such as follicle diameter or follicle area to determine ovulation based on experience;

[0006] 2) A method for feature extraction and ovulation time prediction based solely on ultrasound images;

[0007] 3) A method that mainly relies on manual palpation experience to determine ovulation.

[0008] The existing technology has at least the following shortcomings:

[0009] 1) In practical applications, ovulation determination relies heavily on the clinical experience of clinicians and is highly subjective. Different operators may have significantly different judgments on the same animal, affecting the consistency and stability of the prediction results.

[0010] 2) Some existing technologies are mainly based on information from a single modality for judgment. When the quality of ultrasound images is unstable or there are large individual differences, the reliability and stability of the prediction results are easily affected.

[0011] 3) Most existing methods only provide qualitative judgments of ovulation status and are difficult to output continuous ovulation time prediction results, thus limiting their guiding role in clinical reproductive management and mating timing selection;

[0012] 4) Some prediction methods based on complex models have high requirements for sample quantity and data quality. It is difficult to obtain large-scale, high-quality samples under clinical conditions, making them unsuitable for small-sample application scenarios.

[0013] 5) The endometrial texture features and echo changes reflected in uterine ultrasound images have not been fully quantified and utilized in the current technology, resulting in the failure to effectively tap the potential useful information in the ovulation determination process. Summary of the Invention

[0014] This invention provides a method and system for predicting ovulation in mammals based on the fusion of multi-source clinical information. It can provide ovulation prediction results that are available for small samples, easy to deploy, highly interpretable, accurate and stable based on routine clinical data.

[0015] In a first aspect, embodiments of the present invention provide an ovulation prediction system based on the fusion of multi-source clinical information, comprising:

[0016] The image acquisition module is used to acquire medical image data of the target animal generated during routine clinical examinations, wherein the medical image data includes at least two of the following: follicle images, cervical images, and uterine images;

[0017] An image preprocessing module is used to preprocess the acquired medical image data. In one embodiment, the preprocessing includes, but is not limited to, size normalization, grayscale conversion, contrast enhancement, and noise suppression, in order to improve the stability of subsequent feature extraction.

[0018] The follicle source prediction module is used to analyze the preprocessed follicle images to obtain the baseline ovulation time remaining and the baseline ovulation status within a preset time window.

[0019] The cervical source prediction module is used to obtain ovulation stage assessment results based on the analysis of preprocessed cervical images.

[0020] The uterine-origin prediction module is used to obtain uterine-origin ovulation prediction results based on the analysis of preprocessed uterine images. The uterine-origin ovulation prediction results are used to reflect the ovulation situation corresponding to changes in uterine tissue status.

[0021] The multi-source information fusion module is used to generate a comprehensive ovulation prediction result for the target animal based on the remaining ovulation time, the baseline ovulation status within a preset time window, the ovulation stage assessment result, the uterine-derived ovulation prediction result, and the artificial experience assessment result.

[0022] As one embodiment, the follicle source prediction module includes:

[0023] The follicle feature extraction submodule is used to segment the follicle region based on the preprocessed follicle image to obtain the follicle segmentation result, and extract the morphological features of the follicle based on the segmentation result. The morphological features include: follicle area, equivalent diameter, roundness, principal axis length, and secondary axis length.

[0024] The ovulation prediction submodule is used to input the morphological characteristics of the follicles into a preset follicle ovulation regression model and output the baseline ovulation time remaining for the target animal. It also inputs the morphological characteristics of the follicles into a preset follicle ovulation classification model and outputs the baseline ovulation status of the target animal within a preset time window.

[0025] As one embodiment, the cervical origin prediction module includes:

[0026] The cervical feature extraction submodule is used to extract the cervical features of the target animal based on the preprocessed cervical image. The cervical features include: cervical color features and cervical texture features.

[0027] The cervical prediction submodule is used to input the cervical features into a preset cervical analysis model to obtain ovulation stage assessment results based on the cervix, and to map the ovulation stage assessment results into a continuous form of ovulation correlation score that reflects the relationship between cervical status and the proximity of ovulation.

[0028] As one embodiment, the uterine origin prediction module includes:

[0029] The uterine feature analysis submodule is used to extract uterine features that reflect the structure and texture of the endometrial tissue from the preprocessed uterine image. The uterine features include: gray-level distribution features, gradient intensity distribution features, local texture pattern features, and statistical features that reflect the overall gray-level statistical characteristics.

[0030] The uterine prediction submodule is used to input the uterine features into a preset uterine ovulation classification model and output the uterine-derived ovulation prediction result.

[0031] As an example, the uterine-derived ovulation prediction results include one of the following: category, distribution, or score.

[0032] As one embodiment, the comprehensive ovulation prediction result includes: the corrected remaining ovulation time; the multi-source information fusion module includes:

[0033] The normalization submodule is used to convert the baseline ovulation remaining time based on follicular ultrasound prediction into a follicular ovulation probability score O, normalize the ovulation stage assessment results to obtain a cervical ovulation probability score C, normalize the uterine ultrasound prediction results to obtain a uterine ovulation probability score U, and normalize the artificial experience assessment results to obtain an artificial ovulation probability score H.

[0034] The fusion submodule is used to perform a weighted summation based on the follicle-derived ovulation probability score O, the cervical-derived ovulation probability score C, the uterine-derived ovulation probability score U, and the artificial-derived ovulation probability score H to obtain a comprehensive ovulation confidence score.

[0035] As an example, the comprehensive ovulation prediction result further includes a dynamic ovulation time correction step, which adjusts the remaining time of ovulation based on the baseline ovulation time predicted by follicular ultrasound. If the remaining time for ovulation exceeds a preset time threshold, and the sum of the follicle-derived ovulation probability score O, cervical-derived ovulation probability score C, uterine-derived ovulation probability score U, and artificially induced ovulation probability score H exceeds a preset synergistic threshold A, the baseline remaining ovulation time is corrected according to the following formula:

[0036]

[0037] in, The corrected remaining ovulation time is represented by k, which is the correction sensitivity coefficient. When the sum of the follicular ovulation probability score O, cervical ovulation probability score C, uterine ovulation probability score U, and artificial ovulation probability score H exceeds a preset synergistic threshold A, the exponential term is negative, thus achieving a decreasing correction of the baseline remaining ovulation time. An exponential decay approach is used to perform non-linear compression correction of the baseline remaining ovulation time when the multi-source ovulation probability score significantly increases, to better reflect the actual characteristics of accelerated physiological processes in the approaching ovulation stage.

[0038] The fusion and dynamic correction steps are performed by computer equipment to reduce the impact of single-modal errors on ovulation time prediction results in the case of inconsistencies between multi-source medical images and clinical data, thereby improving prediction stability and clinical usability.

[0039] Secondly, embodiments of the present invention provide an ovulation prediction method based on the fusion of multi-source clinical information, including:

[0040] Acquire medical imaging data of the target animal generated during routine clinical examinations, wherein the medical imaging data includes at least two of the following: follicle images, cervical images, and uterine images;

[0041] The acquired medical image data is preprocessed;

[0042] The baseline ovulation time and the baseline ovulation status within the preset time window are obtained by analyzing the pre-processed follicle images.

[0043] The ovulation phase assessment results were obtained based on the analysis of preprocessed cervical images;

[0044] The uterine ovulation prediction results are obtained by analyzing the preprocessed uterine images. These uterine ovulation prediction results are used to reflect the ovulation situation corresponding to changes in uterine tissue status.

[0045] The comprehensive ovulation prediction result for the target animal is generated based on the remaining ovulation time, the baseline ovulation status within the preset time window, the ovulation stage assessment result, the uterine ovulation prediction result, and the artificial experience assessment result.

[0046] Thirdly, embodiments of the present invention provide an ovulation prediction system based on the fusion of multi-source clinical information, including a memory and a processor;

[0047] The memory is used to store computer programs; the processor is used to read the computer programs in the memory and, when executing the programs, implement the ovulation prediction method based on multi-source clinical information fusion as described above.

[0048] Fourthly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the ovulation prediction method based on multi-source clinical information fusion as described in the first aspect.

[0049] The technical solution provided by the embodiments of the present invention has at least the following positive effects compared with the prior art:

[0050] In the technical solution of this invention embodiment, medical image data of the target animal generated during routine clinical examination is acquired through an image acquisition module. The medical image data includes at least two of the following: follicle images, cervical images, and uterine images. The acquired medical image data is preprocessed by an image preprocessing module. A follicle source prediction module analyzes the preprocessed follicle images to obtain the baseline remaining ovulation time and the baseline ovulation status within a preset time window. A cervical source prediction module analyzes the preprocessed cervical images to obtain an ovulation stage assessment result. A uterine source prediction module analyzes the preprocessed uterine images to obtain a uterine source ovulation prediction result. The uterine source ovulation prediction result is used to reflect the ovulation status. The ovulation status corresponding to changes in uterine tissue state is analyzed by a multi-source information fusion module. This module generates a comprehensive ovulation prediction result for the target animal based on the remaining time before ovulation, the baseline ovulation status within a preset time window, the ovulation stage assessment results, the uterine-derived ovulation prediction results, and the results of human experience assessment. By unifying and fusion analysis of heterogeneous information from different sources, such as follicle images, cervical images, uterine images, and results of human experience assessment (e.g., manual palpation information), the overall consistency of ovulation judgment can be improved, and the impact of different examiners and examination conditions on the prediction results can be reduced. Furthermore, this invention can predict based on small samples, is easy to deploy and expand, and can adapt to the needs of different clinical application scenarios. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is a schematic diagram of the ovulation prediction system based on multi-source clinical information fusion provided in Embodiments 1 and 2 of the present invention;

[0053] Figure 2 The flowcharts for ovulation prediction based on follicular ultrasound images provided in Embodiments 1 and 2 of the present invention are as follows;

[0054] Figure 3 The flowcharts for predicting ovulation status from uterine ultrasound images provided in Embodiments 1 and 2 of the present invention are as follows:

[0055] Figure 4 The flowcharts for the ovulation prediction methods based on multi-source clinical information fusion provided in Embodiments 1 and 2 of the present invention are shown below.

[0056] Figure 5 This is a schematic diagram of the hardware architecture of the ovulation prediction method based on multi-source clinical information fusion provided in Embodiments 1 and 2 of the present invention. Detailed Implementation

[0057] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0058] 1. System Overall Implementation

[0059] Figure 1 This is a schematic diagram of the structure of an ovulation prediction system based on multi-source clinical information fusion provided in an embodiment of the present invention. In a preferred embodiment, the present invention provides an ovulation prediction system for large mammals based on multi-source clinical information fusion, including an image acquisition module, an image preprocessing module, a follicle ultrasound analysis module, a cervical image analysis module, a uterine ultrasound prediction module, a multi-source information fusion module, and a prediction result output module.

[0060] The system uses a server or local computing device as its carrier, and each module is integrated into a unified software service through a program. It supports receiving input data and outputting prediction results through a graphical user interface or application programming interface.

[0061] 2. Image Acquisition and Preprocessing Implementation Methods

[0062] Image acquisition module 102 is used to acquire medical image data of the target animal generated during routine clinical examinations. The medical image data includes at least two of the following: follicle images, cervical images, and uterine images.

[0063] Follicular images, cervical images, and uterine images can be ultrasound images of follicles, cervix, and uterus obtained during routine clinical examinations. Follicular ultrasound images obtained during routine clinical examinations contain rich ovulation-related information. Extractable follicular features include the follicle's equivalent diameter, roundness, principal axis length, and secondary axis length. Based on these features, follicles can be classified into different stages, such as development, maturity, ovulation, and post-ovulation. Furthermore, the remaining time before ovulation and the ovulation status can be predicted. Cervical color and texture features can be extracted from cervical images, which can aid in identifying ovulation. Grayscale distribution features, gradient intensity distribution features, local texture pattern features, and statistical features reflecting overall grayscale statistical characteristics can be extracted from uterine ultrasound images. Ovulation correlation information of follicles, cervix, and uterus can be obtained through manual palpation. It is understood that the medical imaging data in this embodiment can be B-mode ultrasound data or color Doppler ultrasound data, etc., without excessive limitations.

[0064] The image preprocessing module 104 is used to preprocess the acquired medical image data.

[0065] The image preprocessing module 104 can perform noise reduction, enhancement, and standardization processing on follicular ultrasound images, cervical images, and uterine ultrasound images. It can smooth noise, preserve edges, improve contrast, highlight dark areas, correct brightness, unify the baseline, and enhance edge gradients to obtain high-quality, easily analyzable ultrasound image data.

[0066] Anisotropic diffusion filtering, based on partial differential equations, uses a diffusion coefficient to smoothly diffuse noise regions while stopping diffusion at edges with large gradients. This method effectively removes speckle noise while maintaining edge sharpness, making it one of the gold standards for medical ultrasound noise reduction. Alternatively, nonlocal mean denoising can be used. Leveraging the self-similarity of images and the repetitive texture structures within the image, it considers not only the neighborhood of each pixel but also searches for similar blocks across the entire image for weighted averaging, effectively preserving texture details (such as the folds of the cervix). Another approach is wavelet / shear wave thresholding. This transforms the image to the wavelet or shear wave domain, thresholds high-frequency coefficients (removing small noise coefficients), and then performs an inverse transform. This method is suitable for handling multi-scale noise and preserves edges and linear structures (such as the contour of the uterine horn) well.

[0067] Image enhancement can solve the problems of low contrast and blurred details in ultrasound images. Common image enhancement techniques include:

[0068] Histogram equalization methods: such as global histogram equalization, which redistributes image gray values ​​to make the gray distribution more uniform and improves the overall contrast, or contrast-limited adaptive histogram equalization: dividing the image into blocks and performing histogram equalization on each block can improve the contrast between the inside of the follicle and the cyst wall.

[0069] Homomorphic filtering treats an image as a product of illuminance and reflectance. By taking the logarithm, the multiplicative relationship is transformed into an additive one, and then filtered in the frequency domain. This method is suitable for correcting uneven gray levels in an image (such as excessive brightness in the near field and excessive darkness in the far field) while enhancing details.

[0070] Enhancement methods based on Retinex theory: Simulating the human visual system, removing the influence of lighting, and preserving the reflective properties of the object itself, can enhance the local details and texture of tissues, making the tiny structures of the cervix and uterus clearer.

[0071] Edge sharpening: For example, using algorithms such as the Laplacian operator and Unsharp Masking, the edges of areas with abrupt changes in grayscale are enhanced, making the boundaries of the follicle wall and endometrium clearer.

[0072] Image standardization can eliminate differences caused by different devices or different gain settings, preparing for subsequent feature recognition. Specifically, it can linearly map pixel values ​​to the interval [0, 1] or [0, 255], accelerating the convergence of the prediction model and improving the model's generalization ability.

[0073] Example of a unified preprocessing workflow:

[0074] In one embodiment, the image acquisition module is used to acquire at least one of the following image data:

[0075] • Ultrasound images of follicles;

[0076] • Cervical images;

[0077] • Uterine ultrasound images.

[0078] Images can be obtained using transrectal ultrasound equipment, portable image acquisition equipment, or digital imaging equipment commonly used in veterinary clinics.

[0079] In a preferred embodiment, the image preprocessing module performs at least one of the following processing operations on the acquired image:

[0080] • Image size normalization processing;

[0081] • Grayscale conversion or color space conversion;

[0082] • Contrast enhancement processing;

[0083] • Noise suppression processing.

[0084] The above preprocessing steps improve the stability and consistency of subsequent feature extraction and model prediction.

[0085] 3. Implementation method of follicular ultrasound analysis

[0086] The follicle source prediction module 106 is used to analyze the preprocessed follicle images to obtain the baseline ovulation time remaining and the baseline ovulation status within a preset time window.

[0087] 1) Follicular region segmentation

[0088] In one embodiment, the follicular ultrasound analysis module performs follicular region segmentation on the preprocessed follicular ultrasound image to generate corresponding follicular segmentation results.

[0089] The segmentation result can be a binary mask image, which represents the spatial distribution relationship between follicular and non-follicular regions. Follicular region segmentation can be achieved using thresholding, edge detection, region growing, morphological processing, or model-based image segmentation methods; this invention does not limit this approach.

[0090] 2) Extraction of follicular morphological features

[0091] In a preferred embodiment, based on the follicle segmentation results, morphological features reflecting the follicle development status are extracted, including but not limited to:

[0092] • Follicle area;

[0093] • Equivalent diameter;

[0094] • Follicle roundness;

[0095] • Spindle length;

[0096] • Secondary axis length.

[0097] The above features are all obtained based on follicle segmentation mask calculations and are used to characterize the size, morphology and development trend of follicles.

[0098] 3) Ovulation time prediction and status assessment

[0099] In one implementation, the ovulation time prediction module predicts the remaining ovulation time of the mother based on the follicle morphological characteristics using a regression model.

[0100] Meanwhile, the ovulation status judgment module determines whether the mother is in the ovulation state within a preset time window based on the same or partial follicle morphological characteristics and a classification model.

[0101] The regression and classification models can be implemented using linear regression, support vector machines, random forests, or other machine learning models suitable for processing numerical features; this invention does not limit the specific implementation of these models.

[0102] 4. Implementation method of cervical image analysis

[0103] The cervical source prediction module 108 is used to obtain ovulation stage assessment results based on the analysis of preprocessed cervical images. The cervical source prediction module 108 may include: a cervical feature extraction submodule and a cervical prediction submodule.

[0104] In one embodiment, the cervical image analysis module extracts features from cervical images to obtain image features that reflect the physiological state of the cervix.

[0105] In a preferred embodiment, the image features include cervical color features and texture features, such as color histogram features obtained through color space conversion.

[0106] Based on the above image features, a classification model is used to classify the cervical condition into stages, including but not limited to the anovulatory sign stage, the mild change stage, the moderate change stage, and the abnormal inflammation stage.

[0107] Furthermore, the stage classification results are mapped to a continuous form of ovulation correlation score for fusion with other modal information.

[0108] 5. Uterine ultrasound prediction implementation method

[0109] The uterine-origin prediction module 110 is used to obtain uterine-origin ovulation prediction results based on the analysis of preprocessed uterine images. The uterine-origin ovulation prediction results are used to reflect the ovulation situation corresponding to changes in uterine tissue status.

[0110] The uterine origin prediction module may include: a uterine feature analysis submodule and a uterine prediction submodule.

[0111] The uterine feature analysis submodule is used to extract uterine features that reflect the structure and texture of the endometrial tissue from the preprocessed uterine image. Uterine features include, but are not limited to: grayscale distribution features, gradient intensity distribution features, local texture pattern features, and statistical features that reflect the overall grayscale statistical characteristics.

[0112] In livestock such as horses, cattle, and donkeys, the gray distribution characteristics mainly reflect the edema subsidence of the endometrium and the state of glandular secretion, and are important indicators for judging whether ovulation has been completed and whether the corpus luteum function is sound.

[0113] Gray-scale analysis of the uterine horn and endometrium can help determine the stage of the estrous cycle, for example:

[0114] 1) Estrus (pre-ovulation period): The endometrial area shows a lower gray value. Under the influence of estrogen, the endometrium exhibits significant edema, and the interstitial spaces are filled with fluid. At this time, the endometrium is soft, and on ultrasound images, it appears as numerous dark areas with weak echoes.

[0115] 2) Ovulation period (transition phase): The grayscale value begins to change, transitioning from low to high, indicating that the follicle has ruptured and released the egg, and the corpus luteum begins to form. As progesterone levels rise, endometrial edema begins to subside, and the previously dark areas on the image gradually brighten.

[0116] 3) Luteal phase (post-mating / pregnancy preparation period): The endometrium shows a medium to high gray value (bright and homogeneous), indicating that the corpus luteum function is good, the endometrial edema has completely subsided, the stromal cells have become dense, and the glands have developed and secreted nutrients.

[0117] Gradient intensity distribution characteristics are the statistical distribution of gradient intensity (amplitude), which can include the average gradient and the standard deviation of the gradient. These characteristics can be used to analyze the clarity of tissue boundaries and texture roughness. By calculating the gradient amplitude of an image using the Sobel / Prewitt operator and then statistically analyzing the average gradient and standard deviation of the gradient in the endometrial region, the gradient intensity distribution characteristics describing the boundaries and texture can be obtained.

[0118] Local texture pattern features describe the local gray-level spatial arrangement of an image, reflecting texture complexity. Based on the relative gray-level relationship between pixels and their neighbors, they are robust to illumination. Local texture pattern features can include the average LBP value (characterizing texture complexity) and the LBP standard deviation (characterizing texture heterogeneity). The LBP (Local Binary Patterns) operator can be used to generate feature maps, compare and encode the center and neighboring pixels, and statistically analyze the average and standard deviation of LBP values ​​in the endometrial region as texture features.

[0119] Statistical features that reflect the overall grayscale characteristics may include the average brightness of all pixels within the ROI (Region of Interest, such as the endometrial region), or variance and standard deviation, which reflect the degree to which pixel grayscale values ​​within the ROI deviate from the mean (dispersion).

[0120] The uterine prediction submodule is used to input uterine features into a preset uterine ovulation classification model and output uterine-derived ovulation prediction results. Uterine-derived ovulation prediction results include, but are not limited to, one of the following: category, distribution, or score. The process of obtaining uterine status prediction results based on uterine ultrasound images is as follows: Figure 3 As shown, in one implementation, the result is a classification result; in another implementation, the classification result is mapped to a continuous score; both are equivalent representations of uterine-derived ovulation prediction results.

[0121] In one embodiment, the system further includes a uterine ultrasound prediction module for analyzing uterine ultrasound images to assist in ovulation prediction.

[0122] In a preferred embodiment, the uterine ultrasound prediction module extracts at least one of the following image features from the preprocessed uterine ultrasound image:

[0123] • Gray-scale distribution characteristics;

[0124] • Gradient or edge strength features;

[0125] • Local texture features;

[0126] • Global statistical characteristics.

[0127] Based on the aforementioned uterine image features, the uterine state is predicted using a classification model or a regression model, outputting uterine-related ovulation auxiliary assessment results. These assessment results can be in discrete category form or mapped to a continuous ovulation correlation score.

[0128] The uterine prediction results serve as supplementary predictive information independent of follicle and cervical analysis, enhancing the stability and accuracy of multi-source information fusion.

[0129] 6. Implementation methods for multi-source information fusion

[0130] The multi-source information fusion module 112 is used to generate a comprehensive ovulation prediction result for the target animal based on the follicle-source prediction module, the cervix-source prediction module, the uterus-source prediction module, and the results of artificial experience evaluation.

[0131] In one implementation, the multi-source information fusion module performs fusion processing on the following information:

[0132] • Ovulation time or ovulation status information predicted based on follicular ultrasound images;

[0133] • Ovulation correlation score based on cervical image analysis;

[0134] • Auxiliary assessment results obtained based on the uterine ultrasound prediction module;

[0135] • Manual palpation score.

[0136] In a preferred embodiment, the fusion module maps various types of information into a continuous numerical form and performs normalization processing.

[0137] Furthermore, a weighted fusion method is used to generate comprehensive ovulation prediction results, where the weights of different information sources can be adjusted based on clinical experience or system configuration.

[0138] In this embodiment, the fusion and dynamic correction process is automatically executed by the computer system. It is used to correct the baseline ovulation time when there are deviations between the follicular ultrasound prediction results and the results of cervical images, uterine ultrasound, and manual experience assessment, so as to reduce the impact of single modality error on the prediction results.

[0139] 7. Implementation Method for Prediction Result Output

[0140] The comprehensive ovulation prediction results may include: the corrected remaining time to ovulation. The multi-source information fusion module includes: a normalization submodule and a fusion submodule.

[0141] The normalization submodule is used to convert the baseline ovulation remaining time based on follicular ultrasound prediction into a follicular ovulation probability score O, normalize the ovulation stage assessment results to obtain a cervical ovulation probability score C, normalize the uterine ultrasound prediction results to obtain a uterine ovulation probability score U, and normalize the artificial experience assessment results to obtain an artificial ovulation probability score H.

[0142] The fusion submodule is used to perform a weighted summation based on the follicle-derived ovulation probability score O, the cervical-derived ovulation probability score C, the uterine-derived ovulation probability score U, and the artificial-derived ovulation probability score H to obtain a comprehensive ovulation confidence score.

[0143] As an example, the comprehensive ovulation prediction result further includes a dynamic ovulation time correction step, which adjusts the remaining time of ovulation based on the baseline ovulation time predicted by follicular ultrasound. If the remaining time for ovulation exceeds a preset time threshold, and the sum of the follicle-derived ovulation probability score O, cervical-derived ovulation probability score C, uterine-derived ovulation probability score U, and artificially induced ovulation probability score H exceeds a preset synergistic threshold A, the baseline remaining ovulation time is corrected according to the following formula:

[0144]

[0145] in, The corrected remaining ovulation time is represented by k, which is the correction sensitivity coefficient. When the sum of the follicle-derived ovulation probability score O, the cervical-derived ovulation probability score C, the uterine-derived ovulation probability score U, and the artificially induced ovulation probability score H is greater than a preset synergistic threshold A, the exponential term is negative, thereby achieving a decreasing correction of the baseline remaining ovulation time.

[0146] Among them, the ovulation probability score O from follicle source, ovulation probability score C from cervix source, ovulation probability score U from uterus source, and ovulation probability score H from artificial source all range from 0 to 1; the preset synergistic threshold A is used to reflect the typical sum of multiple source scores in the state of near ovulation, and can be set to 2.5 for example; the preset time threshold can be set to 36 hours for example; the correction sensitivity coefficient k ranges from 0.3 to 0.8 for example, and can be adjusted according to the historical reproductive data of the target species or individual.

[0147] In one implementation, the prediction result output module is used to output at least one of the following information:

[0148] • Predicted time remaining until ovulation;

[0149] • Results of ovulation status assessment;

[0150] •Comprehensive ovulation prediction score after multi-source fusion.

[0151] The prediction results can be displayed through a graphical interface or generated into structured report files for clinical decision support and medical record archiving.

[0152] The ovulation prediction method based on multi-source clinical information fusion in this invention has the following significant advantages compared with the prior art:

[0153] 1) By performing multi-source information fusion analysis on follicular ultrasound images, cervical images, uterine ultrasound images and manual palpation information, the stability and accuracy of ovulation prediction results under different individuals and different examination conditions were improved;

[0154] 2) The combination of regression prediction and classification based on follicular morphology characteristics enables continuous prediction of ovulation time in target animals, thereby enhancing the guiding significance of prediction results for clinical mating and reproductive management.

[0155] 3) Modeling is based on explicit features of image morphology, color and texture features, which makes the prediction process and results highly interpretable and easy for clinicians to understand and verify.

[0156] 4) The prediction method does not rely on large-scale training data or continuous monitoring data, and can still achieve effective prediction under small sample conditions, which reduces the cost of data collection and annotation and is conducive to clinical application.

[0157] 5) The overall system structure adopts a modular design, with each functional module being relatively independent, which facilitates deployment, maintenance, and functional expansion in practical applications;

[0158] 6) By introducing uterine ultrasound image analysis, the texture features and tissue state information of the endometrium are further utilized to supplement the judgment of ovulation prediction, thereby improving the comprehensiveness and stability of the overall prediction results.

[0159] 7) Without relying on large-scale sample data, it can stably predict the ovulation status of large mammals under routine clinical imaging examinations. In addition to being applicable to routine livestock breeding management, it is also particularly applicable to assisted breeding and conservation breeding scenarios of endangered large mammals, which has positive significance for wildlife protection and population continuation.

[0160] 8) Since the prediction model is based on explicit features rather than an end-to-end deep network, it can still maintain stable prediction performance even with a small number of samples.

[0161] 9) Compared with ovulation prediction methods based on a single ultrasound modality, the present invention can still output stable prediction results even when there are deviations or inconsistencies in multimodal information;

[0162] 10) Through the multi-source information fusion and dynamic time correction mechanism executed by computer equipment, stable ovulation prediction results can still be output even when there are inconsistencies in multi-source medical images and clinical data, thereby reducing the impact of single-modal error on prediction accuracy and improving the reliability of the system in clinical applications.

[0163] The beneficial effects described in this invention are based on the rationality of the technical solution in terms of structure and process. Its technical effects do not require specific experimental data or statistical results as a necessary prerequisite. The purpose of this invention is to provide auxiliary decision-making means for clinical reproductive management, rather than to replace human judgment or guarantee specific reproductive results.

[0164] The extraction methods of follicular morphological features, the types of cervical image features, the types of uterine ultrasound image features, the specific form of the prediction model, the multi-source information fusion method, and the weight allocation of each information source can all be adjusted according to the actual application scenario and data conditions.

[0165] The prediction model is not limited to a deep learning model, nor does it depend on a specific network structure or time series modeling method. Any model that can process the above-mentioned image features or fused features and output ovulation prediction results can be applied to this invention.

[0166] Without departing from the technical concept of this invention, any technical solution that can achieve the fusion analysis of at least two information sources among follicular ultrasound images, cervical images, uterine ultrasound images and manual palpation information, and thereby complete the prediction of ovulation in large mammals, is an alternative to this invention and should fall within the protection scope of this invention.

[0167] This invention provides an ovulation prediction method based on the fusion of multi-source clinical information, such as... Figure 4 As shown, the method includes the following steps:

[0168] Step 401: Acquire medical imaging data of the target animal generated during routine clinical examination, wherein the medical imaging data includes at least two of the following: follicle images, cervical images, and uterine images;

[0169] Step 402: Preprocess the acquired medical image data;

[0170] Step 403: Based on the analysis of the preprocessed follicle images, obtain the baseline ovulation time remaining and the baseline ovulation status within the preset time window.

[0171] Step 404: Obtain the ovulation stage assessment results based on the analysis of the preprocessed cervical images;

[0172] Step 405: Based on the analysis of the preprocessed uterine images, the uterine-derived ovulation prediction results are obtained. The uterine-derived ovulation prediction results are used to reflect the ovulation situation corresponding to changes in the state of uterine tissue.

[0173] Step 406: Generate a comprehensive ovulation prediction result for the target animal based on the remaining ovulation time, the baseline ovulation status within the preset time window, the ovulation stage assessment result, the uterine-derived ovulation prediction result, and the artificial experience assessment result;

[0174] The prediction result output module is used to output at least one of the following information:

[0175] • Predicted time remaining until ovulation;

[0176] • Results of ovulation status assessment;

[0177] •Comprehensive ovulation prediction score after multi-source fusion.

[0178] The prediction results can be displayed through a graphical interface or generated into structured report files for clinical decision support and medical record archiving.

[0179] Optionally, in step 403, based on the preprocessed follicle image, the follicle region is segmented to obtain the follicle segmentation result. Based on the segmentation result, the morphological features of the follicle are extracted, including: follicle area, equivalent diameter, roundness, principal axis length, and secondary axis length. The morphological features of the follicle are input into a preset follicle ovulation regression model to output the baseline ovulation time remaining of the target animal. The morphological features of the follicle are input into a preset follicle ovulation classification model to output the baseline ovulation status of the target animal within a preset time window.

[0180] Optionally, in step 404, cervical features of the target animal are extracted based on the preprocessed cervical image. The cervical features include cervical color features and cervical texture features. The cervical features are input into a preset cervical analysis model to obtain ovulation stage assessment results based on the cervix. The ovulation stage assessment results are then mapped into a continuous form of ovulation correlation score to reflect the correlation between cervical status and the proximity of ovulation.

[0181] In step 405, uterine features reflecting the structure and texture of the endometrial tissue are extracted from the preprocessed uterine image. These uterine features include: grayscale distribution features, gradient intensity distribution features, local texture pattern features, and statistical features reflecting the overall grayscale statistical characteristics. The uterine features are then input into a preset uterine ovulation classification model, and the uterine-derived ovulation prediction result is output. The uterine-derived ovulation prediction result includes one of the following: category, distribution, or score.

[0182] The comprehensive ovulation prediction results include: the corrected remaining ovulation time. In step 406, the baseline remaining ovulation time is converted into a follicle ovulation probability score O. The ovulation stage assessment results are normalized to obtain a cervical ovulation probability score C. The uterine ovulation prediction results are normalized to obtain a uterine ovulation probability score U. The artificial experience assessment results are normalized to obtain an artificial ovulation probability score H. The comprehensive ovulation confidence score is obtained by weighting and summing the follicle ovulation probability score O, the cervical ovulation probability score C, the uterine ovulation probability score U, and the artificial ovulation probability score H.

[0183] Step 406 may also include: when If the time exceeds a preset threshold, and the sum of the follicle-derived ovulation probability score O, cervical-derived ovulation probability score C, uterine-derived ovulation probability score U, and artificially induced ovulation probability score H exceeds a preset synergistic threshold A, the corrected remaining ovulation time is calculated using the following formula:

[0184]

[0185] in, The corrected remaining ovulation time is given by k, where k is the correction sensitivity coefficient. When the sum of the follicle-derived ovulation probability score O, the cervical-derived ovulation probability score C, the uterine-derived ovulation probability score U, and the artificially induced ovulation probability score H is greater than a preset synergistic threshold A, the exponential term is negative, thereby achieving a decreasing correction of the baseline remaining ovulation time.

[0186] In this embodiment, the fusion and dynamic correction process is automatically executed by the computer system. It is used to correct the baseline ovulation time when there are deviations between the follicular ultrasound prediction results and the results of cervical images, uterine ultrasound, and manual experience assessment, so as to reduce the impact of single modality error on the prediction results.

[0187] The present invention will be further described below with reference to specific embodiments.

[0188] Example 1: A General Example of Ovulation Prediction in Large Mammalians

[0189] In one specific embodiment, the present invention uses mare as the target large mammal to illustrate the ovulation prediction method and system:

[0190] The system first acquires ultrasound images of the mares' follicles and cervix, and records the corresponding manual palpation scores. The acquired ultrasound images are preprocessed and segmented to obtain follicle segmentation results, and morphological features of the follicles are extracted based on these results. Based on these morphological features, a regression model predicts the mares' baseline remaining ovulation time, and a classification model determines whether the mares are in the ovulation state within a preset time window.

[0191] Simultaneously, the collected cervical images are preprocessed and image features are extracted. The extracted image features are used to classify the cervical status into stages, obtain the stage assessment results related to ovulation, and further map them into a continuous form of ovulation correlation score.

[0192] Based on this, the system further acquires ultrasound images of the mare's uterus. After preprocessing the ultrasound images, image features reflecting the structure and texture of the endometrial tissue are extracted. Based on these ultrasound image features, a classification model is used to predict the uterine condition, resulting in a uterine ultrasound prediction.

[0193] Finally, the system weighted and fused the baseline remaining ovulation time or baseline ovulation status obtained from follicular ultrasound images, the ovulation correlation score obtained from cervical images, the uterine ultrasound prediction results obtained from uterine ultrasound images, and the manual palpation score to generate a comprehensive ovulation prediction result for the mare. This result reflects the mare's current ovulation status or the proximity of ovulation, thus providing auxiliary support for clinical reproductive management and mating decisions. The manual palpation score or equivalent clinical experience assessment information is qualitative or semi-quantitative information reflecting the reproductive status of the target animal. The specific acquisition method can be adjusted according to the physiological characteristics of the target animal, but the technical concept of its participation in multi-source information fusion as an auxiliary information source remains unchanged.

[0194] Example 2: Application of Assisted Reproduction in Endangered Large Mammals

[0195] In this embodiment, the target large mammal is an endangered large mammal individual under artificial or semi-artificial rearing conditions, whose reproductive physiological characteristics are the same as or similar to those of conventional large mammals, and allows for routine medical imaging examinations and manual assessments without causing additional harm to the animal.

[0196] The system first acquires ultrasound images of follicles, uterus, and cervix of endangered large mammals without relying on continuous physiological monitoring devices, and simultaneously records manual palpation scores or equivalent clinical experience assessments given by veterinarians.

[0197] Subsequently, the acquired follicular ultrasound images were preprocessed and segmented to extract morphological features reflecting the follicular growth status. Based on these morphological features, a regression model was used to predict the remaining ovulation time (i.e., baseline remaining ovulation time) of the target animal. At the same time, a classification model was used to determine whether the animal had entered the ovulation state within the preset time window.

[0198] Simultaneously, feature extraction and stage classification analysis were performed on the acquired cervical images to obtain stage assessment results related to the proximity of ovulation, and these stage assessment results were mapped into a continuous form of ovulation correlation score; feature extraction and state prediction were performed on the acquired uterine ultrasound images to obtain uterine ultrasound prediction results reflecting changes in uterine tissue state.

[0199] Based on this, the system weighted and fused the remaining ovulation time or ovulation status obtained from follicular ultrasound image analysis, the ovulation correlation score obtained from cervical image analysis, the uterine ultrasound prediction results, and the manual palpation score to generate a comprehensive ovulation prediction result for the target endangered large mammal. When it is determined that there is follicular growth retardation but the physiological process has started based on the baseline remaining ovulation time and multi-source scoring results, the final remaining ovulation time can be obtained by correcting it according to the aforementioned exponential decay function (i.e., formula (1)).

[0200] The comprehensive ovulation prediction results are used to provide decision support for artificial insemination, mating timing selection, or other assisted reproductive operations in endangered large mammals, thereby improving the scientific nature and stability of reproductive management under conditions of limited samples and high individual value.

[0201] This invention does not rely on invasive sensors or long-term monitoring devices and is suitable for species with strict ethical restrictions.

[0202] Figure 5 This is a schematic diagram of the architecture of the ovulation prediction system based on multi-source clinical information fusion provided in Embodiments 1 and 2 of the present invention. The prediction system 50 includes a memory 51 and a processor 52;

[0203] The memory 51 is used to store computer programs; the processor 52 is used to read the computer programs in the memory 51 and, when executing the programs, implement the ovulation prediction method based on multi-source clinical information fusion as described in the foregoing embodiments.

[0204] This invention provides a computer-readable storage medium storing a computer program thereon, which, when executed by a computer processor, is used to perform the technical solution of any method embodiment.

[0205] In one embodiment, the system of the present invention is implemented using a modular software architecture, with each analysis module deployed in a backend service and interacting with the frontend interface through an interface.

[0206] Model parameters and configurations can be stored locally or on a server, supporting model loading, updating, and inference operations. This invention is not limited to a specific software framework or hardware platform.

[0207] Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or grid device, etc.) to execute the methods described in the various embodiments of the present invention.

[0208] It is worth noting that in the embodiments of the above-mentioned device, the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of the present invention.

[0209] It should be noted that the endangered large mammals described in this embodiment are merely examples of application scenarios. The methods and systems of this invention are not limited to specific species. Without departing from the technical concept of this invention, any application scenario suitable for the reproductive management of large mammals can adopt the technical solutions of this invention. Furthermore, the methods and systems of this invention are applicable to large mammals, whose reproductive anatomy and imaging examination methods are common across different species. Those skilled in the art can apply this invention to different large mammal individuals without creative effort, simply by adjusting parameters or retraining the model. The parameter adjustments or model retraining are routine engineering adjustments based on differences in target animal image size, follicle scale range, or image acquisition conditions, and do not involve substantial changes to the technical solutions of this invention.

Claims

1. A mammalian ovulation prediction method based on multi-source clinical information fusion, characterized in that, Includes the following steps: Steps for follicular ultrasound analysis: Acquire ultrasound images of follicles of the target individual, preprocess the ultrasound images and segment the follicle regions, and extract morphological features of the follicles based on the segmentation results. The morphological features include at least follicle area, equivalent diameter, roundness, principal axis length and secondary axis length. Steps for predicting ovulation: Based on the follicular morphological characteristics, the remaining ovulation time of the target individual is predicted by a regression model, and the target individual is judged by a classification model to determine whether the target individual is in the ovulation state within a preset time window. Uterine ultrasound analysis steps: Uterine ultrasound images of the target individual are acquired, the uterine ultrasound images are preprocessed, and image features reflecting the texture and tissue state of the endometrium are extracted. Based on the image features, a uterine state prediction score is generated to characterize the correlation between the uterine environment and the ovulation stage. Cervical image analysis steps: Cervical images of the target individual are acquired, the cervical images are preprocessed, and image features reflecting the color distribution and texture characteristics of the cervix are extracted. Based on the image features, the cervical status is classified into stages, and the stage classification results are mapped into a continuous form of ovulation correlation score. Steps for obtaining information through manual palpation: Obtain manual palpation scores provided by clinicians to characterize the current reproductive physiological status of the target individual; Multi-source information fusion steps: The follicle ovulation prediction results, uterine status prediction scores, cervical ovulation correlation scores, and manual palpation scores are normalized and then weighted and fused according to preset or adaptively adjusted weights to generate a comprehensive ovulation prediction result, which is used to characterize the ovulation proximity of the target individual.

2. The method according to claim 1, characterized in that, The image features extracted in the uterine ultrasound analysis step include at least one of grayscale statistical features, gradient features, and local texture features.

3. The method according to claim 1, characterized in that, The uterine condition prediction score is used to reflect the homogeneity, degree of edema, or structural regularity of the endometrium.

4. The method according to claim 1, characterized in that, The cervical stage classification includes the estrus period, mild pre-ovulatory opening, moderate pre-ovulatory opening, and abnormal inflammatory state.

5. The method according to claim 1, characterized in that, In the multi-source information fusion step, the weights corresponding to each information source are adjustable parameters and satisfy the constraint conditions of non-negative and normalized weights.

6. The method according to claim 1, characterized in that, The prediction is based on explicitly extracted image features and does not rely on an end-to-end pixel-level deep learning model.

7. A mammalian ovulation prediction system based on multi-source clinical information fusion, characterized in that, include: The image acquisition module is used to acquire ultrasound images of follicles, ultrasound images of the uterus, and images of the cervix. The follicular ultrasound analysis module is used to segment the follicular ultrasound images and extract follicular morphological features, and to perform ovulation time prediction and ovulation status judgment based on the morphological features. The uterine ultrasound analysis module is used to extract features from uterine ultrasound images and generate a uterine prediction score that reflects the texture and condition of the endometrium. The cervical image analysis module is used to extract color and texture features from cervical images, classify cervical status into stages, and generate ovulation-related scores. The manual palpation input module is used to receive manual palpation scores; The multi-source information fusion module is used to normalize and weight the prediction results output by the above modules to generate a comprehensive ovulation prediction result. The results output module is used to output the comprehensive ovulation prediction results and supports visualization or report generation.

8. The system according to claim 7, characterized in that, The prediction is based on explicitly extracted image features and does not rely on an end-to-end pixel-level deep learning model.

9. A computer-readable storage medium having a computer program stored thereon, the computer program, when executed by a processor, implementing the method of any one of claims 1 to 6.

Images