State recognition method for adaptive gene editing laboratory mouse

Abstract

The invention discloses a state recognition method of an adaptive gene editing laboratory mouse, and aims to solve the problem of insufficient state recognition specificity of the gene editing laboratory mouse in the prior art. The method comprises the following steps: acquiring genotypes, disease model types and experimental environment parameters of laboratory mice; optimizing detection and optical flow analysis parameters, generating a time sequence behavior track map and segmenting a key area; extracting general and pathological specific composite features and standardizing the general and pathological specific composite features; calling a genotype pathological feature association database to calibrate a weight and a threshold value, and inputting a random forest classifier recognition state of transfer learning optimization; scores are calculated through a weighted scoring model, the state is judged, and credibility is calculated through a multi-dimensional confidence evaluation model. The method is suitable for various gene editing laboratory mice and experimental environments, the accuracy and suitability of state recognition are improved, and powerful support is provided for gene therapy research and development and disease mechanism research.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and more specifically, to a method for state recognition of genetically modified experimental mice. Background Technology

[0002] In the fields of life science research and drug development, accurate identification of the state of laboratory mice is a crucial step in assessing animal physiological characteristics, disease progression, and the effectiveness of interventions. Existing technologies include methods that use machine learning to analyze the behavioral pathmaps of laboratory mice to identify a binary state of anxiety and calmness. The core of these methods is to extract mouse location information using optical flow and target detection algorithms. Based on common behavioral characteristics such as the proportion of peripheral activity and the frequency of drinking, a random forest classifier and dynamic threshold calculation are used to determine the state, effectively reducing the errors and labor costs associated with traditional manual observation.

[0003] With the rapid development of gene editing technology, CRISPR / Cas9, base editing, and other technologies have been widely used in the construction of animal models of diseases. These gene-edited disease models have clear genotypic characteristics and specific pathological phenotypes. However, existing methods for identifying the state of laboratory mice have significant limitations: First, the behavioral features of existing technologies are only designed for ordinary mice and do not cover the pathologically related behaviors of gene-edited disease models (such as asymmetrical limb movements and pain-induced curling behavior in arthritis mice, and metabolic abnormalities related to fluctuations in drinking rhythms in PKU mice), resulting in insufficient feature specificity; Second, existing dynamic thresholds and feature weight calibration mechanisms do not consider genotypic differences. The behavioral baselines (such as the proportion of rest time and the range of activity areas) of different gene-edited models are fundamentally different from those of ordinary mice, and directly applying existing thresholds will cause a significant decrease in recognition accuracy; Third, existing training sample libraries rely on manually recorded data from ordinary mice and lack standardized phenotypic data from gene-edited mice, limiting the model's generalization ability; Fourth, existing methods are not adapted to the SPF-grade breeding environment and standardized cage specifications commonly used in gene-editing experiments, and environmental factors can easily interfere with location extraction and feature analysis.

[0004] The aforementioned shortcomings mean that existing technologies cannot meet the state recognition requirements of gene-edited disease models, making it difficult to accurately reflect the pathological state of model mice and limiting their application in gene-editing related research. There is an urgent need for a mouse state recognition technology that can be adapted to the specificity of gene-edited disease models. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a method for state recognition in gene-edited experimental mice, comprising: acquiring the genotype information, disease model type, and experimental environment parameters of the experimental mice; optimizing an improved target detection algorithm and dense optical flow field analysis parameters based on the experimental environment parameters, capturing the spatial coordinates and motion vectors of the experimental mice in real time, and generating a temporal behavioral trajectory map; performing region segmentation on the temporal behavioral trajectory map to identify the coordinate boundaries of drinking areas, nesting areas, and pathologically sensitive areas; extracting a composite feature set from the temporal behavioral trajectory map, and standardizing the feature data, wherein the composite feature set includes general behavioral features and pathologically specific features; calling a preset genotype-pathological feature association database, calibrating feature weights and dynamic judgment thresholds through an adaptive learning algorithm, and inputting the standardized composite features into a random forest classifier optimized by transfer learning for state recognition; calculating a comprehensive state score of the experimental mice using an improved weighted scoring model, wherein when the comprehensive score is higher than the calibrated dynamic judgment threshold, the mice are judged to be in a pathologically related anxiety state, otherwise they are judged to be in a physiologically calm state; and calculating the confidence level of the state judgment using a multi-dimensional confidence assessment model.

[0006] Furthermore, the genotype information includes mutated genes and mutation sites; the disease model type includes single-gene genetic disease models and spontaneous inflammation models; the experimental environment parameters include feeding level, cage size, light intensity, and background noise-related information; the general behavioral characteristics include activity density, regional dwell time percentage, variance of movement acceleration, drinking interval period, and frequency of resting events; and the pathological specific characteristics include the percentage of activity in the affected limb, the time difference between feeding and drinking, the frequency of inflammation-related curling, and the coefficient of fluctuation in drinking due to metabolic abnormalities.

[0007] Furthermore, the pathological specific characteristics are defined as follows: the proportion of activity in the affected limb is the ratio of the duration of activity in the affected limb to the total activity duration in the gene-edited inflammatory model mouse; the time difference between eating and drinking is the mean and standard deviation of the time interval between the end of a single feeding and the start of the next drinking; the frequency of inflammation-related curling is the number of times the experimental mouse's body curling amplitude exceeds a preset threshold per unit time; and the metabolic abnormality drinking fluctuation coefficient is the coefficient of variation of the number of drinking times within a continuous period of time, which is adapted to metabolic disease models.

[0008] Furthermore, the technical process for optimizing and improving the target detection algorithm and dense optical flow field analysis parameters to generate a time-series behavioral trajectory map includes: collecting body size data and motion characteristic data of experimental mice, wherein the body size data includes body length, body width, and weight-related body size dimensional parameters, and the motion characteristic data includes motion speed, motion direction change frequency, and motion trajectory continuity parameters; performing cluster analysis on the body size data to obtain multiple anchor frame size cluster centers, and generating initial anchor frames based on the cluster centers; calculating the intersection-union ratio (IU) between the initial anchor frames and the target region of the experimental mice, and determining whether the average IU meets a preset standard; if it does not meet the preset standard, adjusting the length and width parameters of the anchor frames according to the body size data corresponding to the anchor frames with low IU, and recalculating the IU between the adjusted anchor frames and the target region; repeating the above adjustment and calculation steps until the average IU meets the preset standard, thus completing the anchor frame size optimization; and statistically analyzing the motion speed distribution and trajectory change of the experimental mice based on the motion characteristic data. The detection threshold is optimized by setting an initial detection threshold and adjusting the amplitude of the change in the movement trajectory. When the mouse's movement speed exceeds a preset speed threshold, the detection threshold is lowered; when the amplitude of the movement trajectory change is below a preset amplitude threshold, the detection threshold is increased. Illumination intensity data and background noise-related data are extracted from the experimental environment parameters, including environmental noise intensity and noise frequency distribution. A mapping relationship between illumination intensity, background noise, and optical flow sensitivity is established. When the illumination intensity exceeds a preset strong light threshold, the brightness sensitivity parameter for optical flow field calculation is lowered; when the illumination intensity is below a preset weak light threshold, the contrast sensitivity parameter for optical flow field calculation is increased; when the background noise intensity exceeds a preset noise threshold, the time window parameter of the optical flow field is adjusted according to the noise frequency distribution. An optimized improved target detection algorithm and adjusted dense optical flow field analysis parameters are used to capture the spatial coordinates and movement vectors of the mouse in real time. Based on the spatial coordinates and movement vectors, a time-series behavioral trajectory map is generated.

[0009] Furthermore, the calibration formula for the dynamic determination threshold is: ; in, This represents the baseline threshold for the pathological state of the corresponding genotype. Baseline weighting coefficients, Let Median be the feature sequence within the sliding window, and Median be the median function. It is an adaptive learning factor that is dynamically updated based on the feature distribution of previous time nodes.

[0010] Furthermore, the feature weights are obtained through a calculation model based on mutual information entropy, as shown in the formula: ; in, Let f be the mutual information entropy between feature f and state category S. It is a composite feature set. Let f be the set of features other than feature f. Let f be the average correlation coefficient between feature f and other features. It is a redundancy suppression factor.

[0011] Furthermore, the calculation formula for the improved weighted scoring model is as follows: ; in, For feature weights, These are the standardized eigenvalues. The pathological association enhancement coefficient, Features With disease models The degree of correlation.

[0012] Furthermore, the formula for the multi-dimensional confidence assessment model is as follows: ; in, The overall score is based on the current state. The set of scores for all possible states. It is the minimum value in the score set.

[0013] Furthermore, the training process of the random forest classifier optimized by transfer learning includes: constructing a behavioral sample library of gene-edited experimental mice, which contains temporal behavioral trajectory maps of experimental mice with different genotypes and different growth stages, with each sample labeled with genotype, disease status, and pathological index data; using a pre-trained model to initially train the behavioral data of ordinary mice to obtain a basic classification model; extracting features from gene-edited experimental mouse samples in the sample library, and updating the decision tree node splitting threshold and feature weights of the basic classification model through a fine-tuning strategy; introducing a cross-validation mechanism to evaluate model performance, and completing model training when the model performance reaches a preset standard; storing the trained model to support rapid loading and adaptive adjustment when experimental environment parameters change.

[0014] Furthermore, the adaptation process for cage size in the experimental environment parameters includes: pre-setting coordinate transformation rules corresponding to multiple standard cage sizes and storing them in the rule base; obtaining cage size information in the experimental environment parameters and determining whether the cage size is a standard cage size; if it is a standard cage size, directly calling the corresponding coordinate transformation rule in the rule base; if it is a non-standard cage size, generating an adapted transformation rule based on the standard coordinate transformation rule through a coordinate scaling algorithm; applying the generated adapted transformation rule to the time-series behavior trajectory map, correcting its regional proportion parameters, and ensuring the consistency of feature extraction under different cage specifications.

[0015] The positive and progressive effects of this invention are as follows: This invention addresses the shortcomings of existing technologies in adapting to the specific pathological conditions of gene-edited mice. By optimizing and improving target detection and dense optical flow field analysis algorithms, adding pathologically specific features, introducing a genotype calibration mechanism, and employing a transfer learning-optimized classifier, it achieves highly specific identification of gene-edited mouse states. Adaptable to different experimental environments and cage specifications, it effectively improves the accuracy and robustness of state identification. Furthermore, it covers various gene-editing models, including single-gene genetic diseases and spontaneous inflammation, providing a precise behavioral analysis tool for evaluating gene therapy efficacy and studying disease mechanisms, significantly reducing human observation errors and experimental costs. Attached Figure Description

[0016] Figure 1 This is a flowchart of the steps of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0018] A method for state recognition of genetically edited experimental mice includes: acquiring the genotype information, disease model type, and experimental environment parameters of the experimental mice; optimizing an improved target detection algorithm and dense optical flow field analysis parameters based on the experimental environment parameters to capture the spatial coordinates and motion vectors of the experimental mice in real time and generate a temporal behavioral trajectory map; performing region segmentation on the temporal behavioral trajectory map to identify the coordinate boundaries of drinking areas, nesting areas, and pathologically sensitive areas; extracting a composite feature set from the temporal behavioral trajectory map and standardizing the feature data, wherein the composite feature set includes general behavioral features and pathologically specific features; calling a preset genotype-pathological feature association database, calibrating feature weights and dynamic judgment thresholds through an adaptive learning algorithm, and inputting the standardized composite features into a random forest classifier optimized by transfer learning for state recognition; calculating a comprehensive state score of the experimental mice using an improved weighted scoring model, and determining a pathology-related anxiety state when the comprehensive score is higher than the calibrated dynamic judgment threshold, otherwise determining a physiological calm state; and calculating the confidence level of the state judgment using a multi-dimensional confidence assessment model.

[0019] Furthermore, the genotype information includes the mutated gene and mutation site; the disease model type includes a single-gene genetic disease model and a spontaneous inflammation model; the experimental environment parameters include housing level, cage size, light intensity, and background noise information; the general behavioral characteristics include activity density, percentage of time spent in a given area, variance of movement acceleration, drinking interval period, and frequency of resting events; the pathological specific characteristics include the percentage of activity in the affected limb, the time difference between eating and drinking, the frequency of inflammation-related curling, and the coefficient of fluctuation in metabolically abnormal drinking. In one example, for gene-edited mice in a spontaneous inflammation model, the genotype information can specifically be a specific mutated gene and its corresponding mutation site. The housing level in the experimental environment parameters is SPF, the cage size is selected according to the experimental design, the light intensity is controlled within a suitable range for mouse activity, and background noise information is collected in real time through environmental monitoring equipment to ensure the targeted nature of subsequent algorithm optimization.

[0020] Furthermore, the pathological specific characteristics are defined as follows: the proportion of activity in the affected limb is the ratio of the activity duration of the diseased limb to the total activity duration in the gene-edited inflammatory model mouse; the time difference between eating and drinking is the mean and standard deviation of the time interval from the end of a single feeding to the start of the next drinking session; the frequency of inflammation-related curling is the number of times the mouse's body curling exceeds a preset threshold per unit time; the metabolic abnormality drinking fluctuation coefficient is the coefficient of variation of the number of drinking sessions over a continuous period, adapted to metabolic disease models. In one example, for gene-edited experimental mice with spontaneous arthritis, the proportion of activity in the affected limb is determined by distinguishing the activity trajectories of the diseased limb and the normal limb using image segmentation technology, and the ratio is calculated after statistically analyzing their respective durations; the preset threshold for the frequency of inflammation-related curling is set based on the physiological characteristics of this type of model mouse to ensure accurate capture of curling behavior induced by inflammation; in the calculation of the metabolic abnormality drinking fluctuation coefficient of metabolic disease model mice, the continuous time can be set according to the experimental observation period to fully reflect the fluctuation pattern of drinking behavior.

[0021] Furthermore, the technical process for optimizing and improving the target detection algorithm and dense optical flow field analysis parameters to generate a time-series behavioral trajectory map includes: collecting body size data and motion characteristic data of experimental mice, wherein the body size data includes body length, body width, and weight-related body size dimensional parameters, and the motion characteristic data includes motion speed, motion direction change frequency, and motion trajectory continuity parameters; performing cluster analysis on the body size data to obtain multiple anchor frame size cluster centers, and generating initial anchor frames based on the cluster centers; calculating the intersection-union ratio (IU) between the initial anchor frames and the target region of the experimental mice, and determining whether the average IU meets a preset standard; if it does not meet the preset standard, adjusting the length and width parameters of the anchor frames according to the body size data corresponding to the anchor frames with low IU, and recalculating the IU between the adjusted anchor frames and the target region; repeating the above adjustment and calculation steps until the average IU meets the preset standard, thus completing the anchor frame size optimization; and statistically analyzing the motion speed distribution and trajectory change of the experimental mice based on the motion characteristic data. The detection threshold is optimized by setting an initial detection threshold and adjusting the amplitude of the change in the movement trajectory. When the mouse's movement speed exceeds a preset speed threshold, the detection threshold is lowered; when the amplitude of the movement trajectory change is below a preset amplitude threshold, the detection threshold is increased. Illumination intensity data and background noise-related data are extracted from the experimental environment parameters, including environmental noise intensity and noise frequency distribution. A mapping relationship between illumination intensity, background noise, and optical flow sensitivity is established. When the illumination intensity exceeds a preset strong light threshold, the brightness sensitivity parameter for optical flow field calculation is lowered; when the illumination intensity is below a preset weak light threshold, the contrast sensitivity parameter for optical flow field calculation is increased; when the background noise intensity exceeds a preset noise threshold, the time window parameter of the optical flow field is adjusted according to the noise frequency distribution. An optimized improved target detection algorithm and adjusted dense optical flow field analysis parameters are used to capture the spatial coordinates and movement vectors of the mouse in real time. Based on the spatial coordinates and movement vectors, a time-series behavioral trajectory map is generated.In one example, when collecting body size data, multi-angle images of the experimental mice were captured using an image acquisition device. Image measurement technology was used to obtain parameters such as body length and width, and a body size dimension parameter library was established by combining this with weight data. Cluster analysis employed the K-means algorithm, determining the number of clusters based on the body size differences within the experimental mouse population, generating initial anchor frames covering different body size ranges. The intersection-union ratio (IUU) was preset to meet the accuracy requirements of target detection, ensuring that the anchor frames effectively encompassed the target area of ​​the experimental mice. During the detection threshold adjustment process, the preset speed threshold and trajectory change amplitude threshold were determined based on statistical data of the movement characteristics of the gene-edited experimental mice. For example, for more active model mice, the preset speed threshold was appropriately increased. The mapping relationship between light intensity and optical flow sensitivity was established through fitting with a large amount of experimental data. In strong light environments, reducing brightness sensitivity can avoid interference from reflections in target capture, while increasing contrast sensitivity in weak light environments can enhance the recognition effect of motion vectors. In background noise processing, adjusting the time window parameters according to the noise frequency distribution can filter out false optical flow information caused by high-frequency environmental noise, ensuring the accuracy of spatial coordinate and motion vector capture.

[0022] Furthermore, the calibration formula for the dynamic determination threshold is: ; in, This represents the baseline threshold for the pathological state of the corresponding genotype. Baseline weighting coefficients, Let Median be the feature sequence within the sliding window, and Median be the median function. It is an adaptive learning factor that is dynamically updated based on the feature distribution of previous time nodes.

[0023] In one example, for a single-gene genetic disease model mouse, the baseline threshold for pathological state is determined based on the correlation analysis between the historical behavioral data and pathological indicators of the mouse of that genotype; the value of the baseline weight coefficient is adjusted according to the disease type and growth stage of the mouse to ensure a reasonable weight allocation between the baseline threshold and the real-time feature sequence; the adaptive learning factor β(t) dynamically adjusts the sensitivity of the threshold calibration by continuously monitoring the feature distribution changes of the preceding time nodes, so that the threshold can accurately adapt to the state changes of the mouse.

[0024] Furthermore, the feature weights are obtained through a calculation model based on mutual information entropy, as shown in the formula: ; in, Let f be the mutual information entropy between feature f and state category S. It is a composite feature set. Let f be the set of features other than feature f. Let f be the average correlation coefficient between feature f and other features. It is a redundancy suppression factor.

[0025] Furthermore, the calculation formula for the improved weighted scoring model is as follows: ; in, For feature weights, These are the standardized eigenvalues. The pathological association enhancement coefficient, Features With disease models The degree of correlation.

[0026] Furthermore, the formula for the multi-dimensional confidence assessment model is as follows: ; in, The overall score is based on the current state. The set of scores for all possible states. It is the minimum value in the score set.

[0027] In one example, when calculating the mutual information entropy between features and state categories, the contribution of different features to state recognition is quantified based on labeled sample data from gene-edited mice. The value of the redundancy suppression factor λ is determined based on the correlation analysis results of features in the composite feature set. For feature combinations with high correlation, the value of λ is appropriately increased to reduce the weight ratio of redundant features and ensure the rationality of feature weight allocation.

[0028] In one example, the pathological association enhancement coefficient γ is set according to the type of disease model. For models with high correlation between pathological specific features and disease, the value of γ is appropriately increased to enhance the contribution of such features to the score. The correlation degree between features and disease models, Relev (f,D), is determined based on historical data in the genotype pathological feature association database. It is obtained by statistically analyzing the frequency of different features in the corresponding disease models and their correlation with state determination.

[0029] In one example, the score set for all possible states was obtained by simulating different states of gene-edited mice (such as pathological anxiety, physiological calm, and different degrees of intermediate states) to ensure that the confidence assessment covers all potential states and accurately reflects the reliability of the current state determination.

[0030] Furthermore, the training process of the random forest classifier optimized by transfer learning includes: constructing a behavioral sample library of gene-edited experimental mice, which contains temporal behavioral trajectory maps of experimental mice with different genotypes and different growth stages, with each sample labeled with genotype, disease status, and pathological index data; using a pre-trained model to initially train the behavioral data of ordinary mice to obtain a basic classification model; extracting features from gene-edited experimental mouse samples in the sample library, and updating the decision tree node splitting threshold and feature weights of the basic classification model through a fine-tuning strategy; introducing a cross-validation mechanism to evaluate model performance, and completing model training when the model performance reaches a preset standard; storing the trained model to support rapid loading and adaptive adjustment when experimental environment parameters change. In one example, during the sample library construction process, behavioral data of experimental mice at different growth stages from various gene-edited disease models (including single-gene genetic disease models, spontaneous inflammation models, etc.) were collected. A standardized collection process ensured the consistency of the time-series behavioral trajectory maps. The genotype, disease state, and pathological indicators of each sample were determined through a combination of professional detection equipment and manual annotation. The pre-trained model used a random forest model trained on a large amount of ordinary mouse behavioral data. During fine-tuning, the splitting threshold and feature weight allocation ratio of the decision tree nodes were adjusted to address the specific characteristics of gene-edited experimental mice. The cross-validation mechanism employed 5-fold cross-validation, with preset performance standards including state recognition accuracy and pathological state recall. When both accuracy and recall reached preset thresholds, the model training was considered complete. The model storage used an efficient file format, supporting rapid model loading and adaptive adjustments when experimental environment parameters changed, eliminating the need for retraining.

[0031] Furthermore, the adaptation process for cage size in the experimental environment parameters includes: pre-setting coordinate transformation rules corresponding to multiple standard cage sizes and storing them in the rule base; obtaining cage size information in the experimental environment parameters and determining whether the cage size is a standard cage size; if it is a standard cage size, directly calling the corresponding coordinate transformation rule in the rule base; if it is a non-standard cage size, generating an adapted transformation rule based on the standard coordinate transformation rule through a coordinate scaling algorithm; applying the generated adapted transformation rule to the time-series behavior trajectory map, correcting its regional proportion parameters, and ensuring the consistency of feature extraction under different cage specifications. In one example, the preset standard cage size covers commonly used experimental specifications, and the corresponding coordinate transformation rules are determined by combining geometric transformation and experimental verification. For non-standard cage sizes, the coordinate scaling algorithm adjusts the coordinate parameters of the temporal behavior trajectory map proportionally according to the difference between the length and width ratio of the cage and the standard size. For example, when the length of the non-standard cage is 1.2 times that of the standard cage, the length direction coordinate of the map is scaled by 1.2 times to ensure that the relative positions and ratios of the drinking area, nest area and pathologically sensitive area remain unchanged, thus ensuring the consistency of feature extraction.

[0032] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for identifying the state of a genetically modified experimental mouse, characterized in that, include: The experiment involves acquiring the genotype information, disease model type, and experimental environment parameters of the experimental mice; optimizing and improving the target detection algorithm and dense optical flow field analysis parameters based on the experimental environment parameters to capture the spatial coordinates and motion vectors of the experimental mice in real time, generating a temporal behavioral trajectory map; segmenting the temporal behavioral trajectory map to identify the coordinate boundaries of the drinking area, nest area, and pathologically sensitive area; extracting a composite feature set from the temporal behavioral trajectory map, and standardizing the feature data, wherein the composite feature set includes general behavioral features and pathologically specific features; A pre-defined genotype pathological feature association database is invoked. Feature weights and dynamic judgment thresholds are calibrated using an adaptive learning algorithm. The standardized composite features are then input into a random forest classifier optimized by transfer learning for state identification. An improved weighted scoring model is used to calculate the comprehensive state score of the experimental mice. When the comprehensive score is higher than the calibrated dynamic judgment threshold, the mice are judged to be in a pathology-related anxiety state; otherwise, they are judged to be in a physiological calm state. A multi-dimensional confidence assessment model is used to calculate the credibility of the state judgment.

2. The method for state recognition of a genetically modified experimental mouse according to claim 1, characterized in that, The genotype information includes mutated genes and mutation sites; the disease model type includes single-gene hereditary disease models and spontaneous inflammation models; the experimental environment parameters include feeding level, cage size, light intensity, and background noise; the general behavioral characteristics include activity density, percentage of time spent in a given area, variance of movement acceleration, drinking interval period, and frequency of resting events; the pathological specific characteristics include the percentage of activity in the affected limb, the time difference between feeding and drinking, the frequency of inflammation-related curling, and the coefficient of fluctuation in drinking due to metabolic abnormalities.

3. The method for state recognition of a genetically modified experimental mouse according to claim 2, characterized in that, The pathological specific features are defined as follows: the proportion of activity in the affected limb is the ratio of the duration of activity in the affected limb to the total activity duration in the gene-edited inflammatory model mouse; the time difference between eating and drinking is the mean and standard deviation of the time interval between the end of a single feeding and the start of the next drinking. Inflammation-related curling frequency is the number of times the experimental mouse's body curls up beyond a preset threshold per unit time; metabolic abnormality water consumption fluctuation coefficient is the coefficient of variation of water consumption frequency over a continuous period of time, which is adapted to metabolic disease models.

4. The method for state recognition of a genetically modified experimental mouse according to claim 1, characterized in that, The technical process for optimizing and improving the target detection algorithm and dense optical flow field analysis parameters to generate a time-series behavioral trajectory map includes: collecting body shape data and motion characteristic data of experimental mice, wherein the body shape data includes body length, body width, and weight-related body shape dimension parameters, and the motion characteristic data includes motion speed, motion direction change frequency, and motion trajectory continuity parameters; performing cluster analysis on the body shape data to obtain multiple anchor frame size cluster centers, and generating initial anchor frames based on the cluster centers; calculating the intersection-union ratio (IU) between the initial anchor frames and the target area of ​​the experimental mice, and determining whether the average IU meets a preset standard; if it does not meet the preset standard, adjusting the length and width parameters of the anchor frames according to the body shape data corresponding to the anchor frames with low IU, and recalculating the IU between the adjusted anchor frames and the target area; repeating the above adjustment and calculation steps until the average IU meets the preset standard, thus completing the anchor frame size optimization; and statistically analyzing the motion speed distribution and trajectory change amplitude of the experimental mice based on the motion characteristic data. An initial detection threshold is set; when the mouse's movement speed exceeds a preset speed threshold, the detection threshold is lowered; when the amplitude of the movement trajectory change is lower than a preset amplitude threshold, the detection threshold is increased, thus optimizing the detection threshold. Illumination intensity data and background noise-related data are extracted from the experimental environment parameters. The background noise-related data includes environmental noise intensity and noise frequency distribution. A mapping relationship between illumination intensity, background noise, and optical flow sensitivity is established. When the illumination intensity exceeds a preset strong light threshold, the brightness sensitivity parameter for optical flow field calculation is lowered; when the illumination intensity is lower than a preset weak light threshold, the contrast sensitivity parameter for optical flow field calculation is increased; when the background noise intensity exceeds a preset noise threshold, the time window parameter of the optical flow field is adjusted according to the noise frequency distribution. An optimized improved target detection algorithm and adjusted dense optical flow field analysis parameters are used to capture the spatial coordinates and movement vectors of the mouse in real time. Based on the spatial coordinates and movement vectors, a time-series behavioral trajectory map is generated.

5. The method for state recognition of a genetically modified experimental mouse according to claim 1, characterized in that, The calibration formula for the dynamic determination threshold is: ; in, This represents the baseline threshold for the pathological state of the corresponding genotype. Baseline weighting coefficients, Let Median be the feature sequence within the sliding window, and Median be the median function. It is an adaptive learning factor that is dynamically updated based on the feature distribution of previous time nodes.

6. The method for state recognition of a genetically modified experimental mouse according to claim 1, characterized in that, The feature weights are obtained through a calculation model based on mutual information entropy, and the formula is as follows: ; in, Let f be the mutual information entropy between feature f and state category S. It is a composite feature set. The set of features other than feature f. Let f be the average correlation coefficient between feature f and other features. It is a redundancy suppression factor.

7. The method for state recognition of a genetically modified experimental mouse according to claim 1, characterized in that, The calculation formula for the improved weighted scoring model is as follows: ; in, For feature weights, These are the standardized eigenvalues. The pathological association enhancement coefficient, Features With disease models The degree of correlation.

8. The method for state recognition of a genetically modified experimental mouse according to claim 1, characterized in that, The formula for the multi-dimensional confidence assessment model is: ; in, The overall score is based on the current state. The set of scores for all possible states. It is the minimum value in the score set.

9. The method for state recognition of a genetically modified experimental mouse according to claim 1, characterized in that, The training process of the random forest classifier optimized by transfer learning includes: constructing a behavioral sample library of gene-edited experimental mice, which contains temporal behavioral trajectory maps of experimental mice with different genotypes and different growth stages, with each sample labeled with genotype, disease status, and pathological index data; using a pre-trained model to initially train the behavioral data of ordinary mice to obtain a basic classification model; extracting features from gene-edited experimental mouse samples in the sample library, and updating the decision tree node splitting threshold and feature weights of the basic classification model through a fine-tuning strategy; introducing a cross-validation mechanism to evaluate model performance, and completing model training when the model performance reaches a preset standard; and storing the trained model to support rapid loading and adaptive adjustment when experimental environment parameters change.

10. The method for state recognition of a genetically modified experimental mouse according to claim 1, characterized in that, The adaptation process for cage size in experimental environment parameters includes: pre-setting coordinate transformation rules corresponding to multiple standard cage sizes and storing them in the rule base; obtaining cage size information in the experimental environment parameters and determining whether the cage size is a standard cage size; if it is a standard cage size, directly calling the corresponding coordinate transformation rule in the rule base; if it is a non-standard cage size, generating an adapted transformation rule based on the standard coordinate transformation rule through a coordinate scaling algorithm; applying the generated adapted transformation rule to the time-series behavior trajectory map, correcting its regional ratio parameters, and ensuring the consistency of feature extraction under different cage specifications.

Images