Experimental animal cognitive function evaluation system and detection method thereof

Abstract

The invention discloses an experimental animal cognitive function evaluation system and a detection method thereof, and the system comprises a behavior data collection module which is configured to receive unstructured original behavior data streams from a plurality of behavior testing devices; the physiological data acquisition module is configured to receive real-time physiological data from the plurality of physiological signal acquisition devices and extract characteristic indexes related to cognitive activities from the real-time physiological data; the task management and generation module is configured to store various cognitive function test normal form templates and generate an executable task instruction set according to the evaluation instruction or the self-adaptive optimization instruction; and the cognitive function analysis module is used for analyzing the cognitive function of the experimental animal based on the data. Through multi-modal data fusion, the system can capture neural activity evidence which cannot be reflected by a traditional single behavior index and is directly related to the cognitive process, interference of non-cognitive factors is effectively stripped, and the evaluation result can reflect the inherent cognitive state of the animal more truly.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of experimental animal science and biomedical technology, in particular to an experimental animal cognitive function evaluation system and a detection method thereof. BACKGROUND

[0002] In many scientific research fields such as neuroscience, pharmacology and psychology, accurate evaluation of the cognitive function of experimental animals is of great significance. It not only helps to understand the neural mechanisms of cognitive processes, but also provides key experimental evidence for new drug development, disease model research, etc.

[0003] Currently, the traditional experimental animal cognitive function evaluation method mainly relies on a single behavior index. For example, in the water maze experiment, the spatial memory ability of experimental animals is usually evaluated by measuring the time and path length of finding the hidden platform; in the object recognition experiment, the object recognition memory ability of animals is mainly judged according to the difference in exploration time of new and old objects. However, these single behavior index evaluation methods have obvious limitations.

[0004] On the one hand, single behavior index is easily disturbed by various non-cognitive factors. The difference in motor ability of experimental animals will directly affect their performance in behavior test, for example, animals with weaker motor ability may need more time to find the platform in the water maze, but this does not necessarily mean that their spatial memory ability is poorer; the appetite state will also affect the behavior of animals, hungry animals may be more inclined to find food, thereby changing their exploration behavior pattern in cognitive test, leading to biased evaluation results.

[0005] On the other hand, single behavior index is difficult to comprehensively and accurately reflect the complex cognitive process of experimental animals. Cognitive activity is a complex process involving multiple brain regions and neural pathways, and single external behavior often cannot cover all information in the cognitive process. For example, the neural activity and neurotransmitter level changes in the brain of animals when they complete a cognitive task are crucial for understanding the cognitive mechanism, but traditional methods cannot obtain these key information.

[0006] Therefore, how to overcome the limitations of traditional single behavior index evaluation method and obtain more comprehensive, accurate and non-cognitive factor disturbed experimental animal cognitive function evaluation results has become a specific problem to be solved in current scientific research field. In order to solve the above problems, we propose an experimental animal cognitive function evaluation system and a detection method thereof. SUMMARY

[0007] Therefore, the present application provides an experimental animal cognitive function evaluation system and a detection method thereof to solve the problem that the traditional experimental animal cognitive function evaluation method mainly relies on a single behavior index.

[0008] To achieve the above object, the present application provides the following technical solutions:

[0009] An experimental animal cognitive function evaluation system comprises:

[0010] A behavior data acquisition module configured to receive an unstructured raw behavior data stream from a plurality of behavior testing devices and preprocess the raw behavior data stream to obtain a standardized behavior event sequence;

[0011] A physiological data acquisition module configured to receive real-time physiological data from a plurality of physiological signal acquisition devices and extract feature indicators related to cognitive activities therefrom;

[0012] A task management and generation module configured to store a plurality of cognitive function test paradigm templates and generate executable task instruction sets according to evaluation instructions or adaptive optimization instructions;

[0013] A cognitive function analysis module in communication connection with the behavior data acquisition module, the physiological data acquisition module and the task management and generation module, the cognitive function analysis module being configured to receive the standardized behavior event sequence output by the behavior data acquisition module, the feature indicators extracted by the physiological data acquisition module and the task instruction sets generated by the task management and generation module and analyze cognitive functions of experimental animals based on the above data.

[0014] As a preferred solution, the cognitive function analysis module comprises:

[0015] A feature fusion unit configured to perform spatiotemporal alignment and associated fusion of the behavior event sequence and the feature indicators to generate a multi-modal behavior-physiological feature vector;

[0016] A cognitive evaluation model unit storing a cognitive function evaluation model configured to receive the multi-modal behavior-physiological feature vector and output a quantitative result comprising cognitive dimension scores and a comprehensive cognitive evaluation score;

[0017] A result output and visualization module configured to visually display the quantitative result, the multi-modal behavior-physiological feature vector and raw data indexes and generate a structured evaluation report;

[0018] A database configured to store the raw behavior data stream, the real-time physiological data, the task instruction sets, the multi-modal behavior-physiological feature vector and the quantitative result in association.

[0019] As a preferred solution, the cognitive function analysis module further comprises a cognitive function model construction unit configured to:

[0020] constructing a training sample set comprising multi-modal behavioral-physiological feature vectors corresponding to samples with known cognitive state labels based on data historically stored in the database;

[0021] As a preferred solution, the training sample set is trained by a machine learning algorithm to obtain or optimize the cognitive function evaluation model, wherein the cognitive function evaluation model is used to map the complex nonlinear relationship between the multi-modal behavioral-physiological feature vectors and the cognitive state.

[0022] As a preferred solution, an adaptive optimization module is further included, which is configured to:

[0023] receive the current quantification result from the cognitive function analysis module;

[0024] determine whether to trigger task parameter optimization based on the current quantification result;

[0025] If triggered, an updated task parameter set that can make the optimization objective function better is generated by searching in the parameter space through an optimization algorithm based on the predefined optimization objective function and the cognitive function evaluation model;

[0026] send the updated task parameter set to the task management and generation module to generate a new executable task instruction set.

[0027] As a preferred solution, the optimization objective function is to maximize the information entropy of the evaluation result.

[0028] As a preferred solution, the task management and generation module is further configured with a task verification unit, which is configured to:

[0029] After generating the executable task instruction set, logical consistency verification is performed on the task instruction set based on a pre-stored animal behavior rule library and task parameter constraint conditions;

[0030] If the verification fails, feedback adjustment suggestions or automatic adjustment parameters are provided until the executable task instruction set that passes the verification is generated.

[0031] As a preferred solution, the system further includes a longitudinal tracking analysis module configured to:

[0032] For a specific experimental animal individual, historical evaluation data of the individual at different time points, including historical multi-modal behavioral-physiological feature vectors and historical quantification results, are extracted from the database.

[0033] As a preferred solution, the longitudinal tracking analysis module performs time series analysis on the historical assessment data to generate a cognitive function change trajectory and trend prediction of the individual.

[0034] An experimental animal cognitive function detection method, comprising the following steps:

[0035] Receiving and preprocessing the original behavior data stream and real-time physiological data to obtain standardized behavior event sequences and feature indicators;

[0036] According to the evaluation requirements, an executable task instruction set is generated or called to guide the experimental animal to complete the cognitive test task;

[0037] The behavior event sequences generated during the test are fused with the feature indicators to generate multi-modal behavior-physiological feature vectors;

[0038] The multi-modal behavior-physiological feature vectors are input into a pre-trained cognitive function evaluation model to obtain a quantitative evaluation result containing at least one cognitive dimension score;

[0039] The quantitative evaluation result, related features and original data index are stored and visualized.

[0040] The present application has the following advantages:

[0041] Through multi-modal data fusion (behavior + physiology), the system can capture the neural activity evidence directly related to the cognitive process that cannot be reflected by traditional single behavior indicators, effectively eliminating the interference of non-cognitive factors such as motor ability and appetite, making the evaluation result more directly and truly reflect the internal cognitive state of the animal. The cognitive evaluation model trained based on machine learning realizes the automatic analysis and quantitative scoring of high-dimensional complex features, completely changing the traditional mode of relying on researchers to manually select indicators and subjective interpretation, greatly improving the objectivity, consistency and repeatability of the evaluation result.

[0042] The adaptive task generation and optimization mechanism of the system can dynamically adjust the task difficulty according to the real-time performance of the animal, so that the test is always near the critical point of its ability. This not only significantly improves the detection sensitivity of individual differences (especially slight cognitive impairment), but also enables the acquisition of the most informative data in a unit of time, greatly improving the evaluation efficiency.

[0043] The system builds a complete automatic closed loop from task generation, data acquisition, fusion analysis to report output, avoiding errors and batch differences introduced by manual operation. In particular, the task logic verification unit ensures that the automatically generated task meets the animal behavior logic, ensuring the scientificity and standardization of the experiment.

[0044] This highly automated and standardized process greatly liberates the productivity of researchers, reduces human error, ensures the comparability of data obtained at different laboratories and different time points, and lays a technical foundation for large-scale, multi-center preclinical research.

[0045] The system can record and analyze individual cognitive data for a long time and continuously, draw dynamic change curves of cognitive function, and make trend prediction, based on a complete database and a longitudinal tracking analysis module. This provides an unprecedented powerful tool for studying cognitive development, aging process, pathological evolution of neurodegenerative diseases, and long-term efficacy evaluation of drugs, and realizes the leap from static evaluation to dynamic process research. BRIEF DESCRIPTION OF DRAWINGS

[0046] Fig. 1 A block diagram of an experimental animal cognitive function evaluation system provided for an embodiment of the present application.

[0047] Fig. 2 A block diagram of a cognitive function analysis module in an experimental animal cognitive function evaluation system provided for an embodiment of the present application.

[0048] Fig. 3 A flowchart of an experimental animal cognitive function detection method provided for an embodiment of the present application. DETAILED DESCRIPTION

[0049] The embodiments of the present application will be described in detail by specific embodiments, and those skilled in the art can easily understand other advantages and effects of the present application from the content disclosed in the specification. Obviously, the described embodiments are part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0050] REFERENCE Figs. 1-3 As shown in the figure, an experimental animal cognitive function evaluation system comprises:

[0051] The behavior data acquisition module is configured to receive unstructured raw behavior data streams from a plurality of behavior test devices, and to preprocess the raw behavior data streams to obtain standardized behavior event sequences;

[0052] The physiological data acquisition module is configured to receive real-time physiological data from a plurality of physiological signal acquisition devices, and to extract feature indicators related to cognitive activities therefrom;

[0053] The task management and generation module is configured to store a plurality of cognitive function test paradigm templates, and to generate executable task instruction sets according to evaluation instructions or adaptive optimization instructions;

[0054] The cognitive function analysis module is in communication connection with the behavior data acquisition module, the physiological data acquisition module and the task management and generation module, and is configured to receive the standardized behavior event sequence output by the behavior data acquisition module, the feature index extracted by the physiological data acquisition module and the task instruction set generated by the task management and generation module, and analyze the cognitive function of the experimental animal based on the above data.

[0055] The cognitive function analysis module comprises:

[0056] The feature fusion unit is configured to perform spatio-temporal alignment and associated fusion of the behavior event sequence and the feature index, and generate a multi-modal behavior-physiological feature vector;

[0057] The cognitive assessment model unit stores a cognitive function assessment model, which is configured to receive the multi-modal behavior-physiological feature vector and output a quantitative result comprising cognitive dimension scores and a comprehensive cognitive assessment score;

[0058] The result output and visualization module is configured to visually display the quantitative result, the multi-modal behavior-physiological feature vector and the original data index, and generate a structured assessment report.

[0059] The database is configured to store the original behavior data stream, the real-time physiological data, the task instruction set, the multi-modal behavior-physiological feature vector and the quantitative result in association.

[0060] In the feature fusion unit, spatio-temporal alignment is the basis of fusion, aiming to solve the inconsistency of behavior and physiological data in time and space dimensions. First, the time base of all data sources needs to be unified. Precise time protocol (such as IEEE 1588 PTP) can be used for hardware clock synchronization to control the timestamp error of all sensors and devices within milliseconds. For data with different sampling frequencies (such as high-frequency physiological signals and low-frequency behavior video frames), dynamic time warping (DTW) algorithm or double-flow time convolution network can be used for flexible matching at the software level, so as to accurately align the key time points such as "start of specific behavior event" and "start of physiological index response".

[0061] Then, the mapping relationship between behavior and physiological changes in space or logic is established. For example, a heterogeneous multi-modal knowledge graph can be constructed. In this graph, nodes can be defined as different behavior types (such as "shuttle", "press lever"), physiological indicators (such as "local field potential power of specific brain area", "heart rate variability") and environmental parameters. The edges are used to represent the association strength between them (for example, the "complete a correct selection" behavior may have a strong association with "enhanced Theta rhythm energy in prefrontal cortex"). This structure provides semantic constraints for data fusion.

[0062] On the basis of alignment, advanced fusion strategies are used to mine deep cross-modal associations. For example, a hierarchical fusion mechanism can be used to capture dependencies from local to global through a granular interaction fusion strategy. For another example, fine-grained fusion focuses on instantaneous associations, such as using a cross-attention mechanism (Co-Attention) to calculate the mutual attention weights between behavior segments (such as “animal rigid posture sequence at the moment”) and physiological segments (such as “high-frequency activity of amygdala EEG at the same period”), to capture local dynamic dependencies.

[0063] The fused information needs to be encoded into a numerical form suitable for machine learning models. The high-dimensional features after alignment and association (which may include hundreds of behavior event features and physiological indicators) are extracted by an encoder (such as 1D-CNN, LSTM, or Transformer) and reduced in dimension using methods such as principal component analysis (PCA), generating a set of feature vectors with high information density and low redundancy.

[0064] The final generated multi-modal behavior-physiological feature vector is a numerical representation that can comprehensively reflect the overall state of the experimental animal under a specific cognitive task. For example, a feature vector may condense “the reaction time, movement trajectory, neural oscillation power in specific brain regions, and coordinated change pattern of heart rate of the animal during the decision-making phase”. This vector will be directly input into the cognitive function evaluation model.

[0065] The cognitive function analysis module also includes a cognitive function model construction unit, which is configured to:

[0066] Based on the data stored in the database, a training sample set is constructed, which includes multi-modal behavior-physiological feature vectors corresponding to samples with known cognitive state labels;

[0067] The training sample set is trained using a machine learning algorithm to obtain or optimize a cognitive function evaluation model, which is used to map the complex nonlinear relationship between multi-modal behavior-physiological feature vectors and cognitive states.

[0068] In the experimental animal cognitive function evaluation system, the machine learning algorithm is the core engine for converting multi-modal data into reliable cognitive evaluation conclusions. It is implemented through a gradient boosting machine, which is an ensemble learning algorithm. Its core idea is to train multiple weak learners (usually decision trees) in series, with each subsequent model focusing on learning the samples that the previous model predicts incorrectly, and by continuously reducing the prediction error, a powerful and accurate model is finally combined. It is particularly suitable for processing tabular data (such as your multi-modal feature vectors), and can effectively capture the complex nonlinear relationship between features and cognitive states.

[0069] Suppose our research goal is to assess whether mice have impaired working memory after a certain drug or genetic intervention. Working memory is an important dimension of cognitive function.

[0070] The specific cognitive assessment model building includes the following steps:

[0071] 1. Data preparation and feature engineering

[0072] First, we need to build a training sample set for model training.

[0073] Data source: Extract historical data from the database, which contains standardized behavioral event sequences and physiological feature indicators of a large number of mice in previous experiments.

[0074] Behavioral features: from the behavior data acquisition module. For example, in the "delayed alternation task" (a common working memory test paradigm), features that can be extracted include: correct selection rate, reaction time, duration of staying in the decision area, total length and speed change of the movement trajectory, etc.

[0075] Physiological features: from the physiological data acquisition module. For example, we can extract: eye movement parameters (such as pupil diameter change, saccade speed), local field potential (LFP) in specific brain regions (such as prefrontal cortex) in different frequency bands (such as Theta wave, 4-8 Hz) Energy, heart rate variability (HRV), etc.

[0076] Label generation: Each valid test trial of each mouse needs a "cognitive state label". This label is the goal of model learning. For example, we can label samples according to behavioral performance (such as correct selection rate) and expert evaluation as:

[0077] "Normal working memory" (label 0), "Mildly impaired working memory" (label 1), "Severely impaired working memory" (label 2), thus generating a training sample set (X, y) with clear labels, where X is the feature matrix (each row is a sample, each column is a feature), y is the corresponding label vector.

[0078] 2. Model training and optimization

[0079] Next, use the gradient boosting algorithm to train this dataset.

[0080] Model initialization: Initialize a basic GBDT model.

[0081] Serial training:

[0082] The first decision tree is trained according to the features of all samples and makes a preliminary prediction. It will make some mistakes, such as misjudging some mildly impaired mice as normal.

[0083] When training the second tree, the algorithm focuses on the samples that were misclassified by the first tree, adjusting their weights, and lets the second tree learn to correct these errors.

[0084] This process continues sequentially, adding new trees one after another. Each new tree aims to correct the residual errors made by the combination of all previous trees.

[0085] Combining the outputs: Ultimately, the predictions of hundreds or thousands of trees are combined through weighted averaging or voting to form a final, very robust prediction model. This model can finely capture complex patterns such as "a slight prolongation of reaction time accompanied by a decrease in frontal theta wave energy," which are difficult to determine from a single feature but become strong indicators of working memory impairment when combined.

[0086] Hyperparameter tuning: Techniques such as cross-validation are used to find the best parameters of the model (such as the maximum depth of the trees, learning rate, number of trees, etc.) to prevent overfitting and ensure the model's generalization ability.

[0087] 3. Model application and interpretation

[0088] When the trained model is used to evaluate a new mouse:

[0089] Feature extraction: The system collects the mouse's behavioral and physiological data in real-time and extracts the same features as during training, forming a feature vector.

[0090] Prediction: The feature vector is input into the trained GBDT model. The model outputs a probability value representing the probability that the mouse belongs to each of the three classes: "normal," "mild impairment," and "severe impairment."

[0091] Result generation: The class with the highest probability is chosen as the final cognitive state judgment. This result, along with scores from other dimensions, is ultimately incorporated into the "cognitive dimension score and comprehensive cognitive assessment score."

[0092] Interpretability: One of the key advantages of gradient boosting models is the ability to provide feature importance rankings. This helps researchers understand which behaviors or physiological indicators (for example, a delay in reaction time or an anomaly in pupil diameter) contribute most to the judgment of "working memory impairment," thereby not only knowing "what" but also speculating "why," enhancing the scientific depth of the research.

[0093] It also includes an adaptive optimization module configured to:

[0094] Receive the current quantification results from the cognitive function analysis module;

[0095] Based on the current quantization result, it is judged whether to trigger the task parameter optimization;

[0096] If triggered, based on the predefined optimization objective function and cognitive function evaluation model, an updated task parameter set that can make the optimization objective function better is generated through optimization algorithm in the parameter space;

[0097] The updated task parameter set is sent to the task management and generation module to generate a new executable task instruction set.

[0098] The optimization objective function is the maximum information entropy of the evaluation result. The maximum information entropy of the evaluation result as the objective function of the adaptive optimization module, its core purpose is to improve the information efficiency and quality of the cognitive function evaluation process, to ensure that each test can obtain the maximum amount of information, so as to more accurately and efficiently reveal the cognitive state of the experimental animal.

[0099] Where "evaluation result" is the probability distribution composed of the comprehensive cognitive evaluation scores of the animal's performance on different cognitive tasks. The maximum information entropy is to optimize the task parameters to maximize the uncertainty of this probability distribution.

[0100] The information entropy of a discrete random variable (representing the possible categories or values of the evaluation result here) is defined as:

[0101]

[0102] Where, represents the th possible result (for example, the cognitive level is divided into "excellent", "good", "general" and other categories, or a set of segmented comprehensive scores), is the probability of the result. The unit of information entropy is bit (bit).

[0103] The range of entropy value: when the result is completely determined (such as a result with a probability of 1 and the rest of 0), the entropy takes the minimum value 0. When all results are equally likely to occur (i.e. uniform distribution, ), the entropy takes the maximum value . Maximizing the information entropy essentially wants the probability distribution of the evaluation result to be as close to the uniform distribution as possible.

[0104] The specific algorithm implementation in the system is:

[0105] Define optimization variables: optimization variables are task parameters, such as delay time, stimulus difficulty, number of interference items, etc., denoted as vector .

[0106] Objective function: The objective function is the information entropy of the evaluation result, which is determined by the current task parameters and the cognitive state of the animal (mapped through the cognitive function evaluation model).

[0107]

[0108] where, represents the cognitive evaluation result of the animal (such as the category to which the comprehensive score belongs or the discrete level after discretization) after testing under the current parameters . The goal is to find the parameters that maximize the objective function :

[0109]

[0110] Probability estimation: The probability needs to be estimated through historical data or model prediction. For example, the system can predict the probability of the animal's performance falling into different result intervals under the given parameters according to the current cognitive function evaluation model.

[0111] Introduce constraints: The optimization process needs to be conducted within the feasible region, such as the physical limitations of parameters (minimum / maximum delay time) and the limitation of the total test duration. These constitute the constraints:

[0112]

[0113] Solve the optimization problem: The complete optimization problem can be expressed as:

[0114]

[0115]

[0116] Solving this constrained nonlinear optimization problem can use optimization algorithms such as gradient ascent, genetic algorithm, and harmony search algorithm. The adaptive optimization module generates an updated set of task parameters through optimization algorithms in the parameter space.

[0117] The objective function guides the system to automatically adjust the parameters (e.g. difficulty, stimulus type) of subsequent testing tasks so that the behavioral response results (i.e. "evaluation results") of the animals have the maximum uncertainty or information content. This means that the system will tend to select tasks that can best distinguish the cognitive abilities of the animals, avoiding tasks that are too simple (all animals can complete) or too difficult (all animals fail), so that each test can provide the most informative data. By pursuing the maximization of information entropy, the system no longer uses a fixed and unchanging test procedure, but dynamically adjusts the tasks according to the real-time performance of the animals. For example, when the system detects that the animals perform stably at the current difficulty, it will automatically increase the task difficulty to explore the boundaries of their abilities, thereby outlining a more accurate cognitive ability profile for the individual.

[0118] The task management and generation module is also configured with a task verification unit, which is configured to:

[0119] After generating the executable task instruction set, based on the pre-stored animal behavior rule library and task parameter constraint conditions, the task instruction set is logically consistent;

[0120] If the verification fails, feedback adjustment suggestions or automatically adjust the parameters until an executable task instruction set that passes the verification is generated.

[0121] The system also includes a longitudinal tracking analysis module configured to:

[0122] For a specific experimental animal individual, extract its historical evaluation data at different time points from the database, including historical multi-modal behavior-physiological feature vectors and historical quantitative results.

[0123] The longitudinal tracking analysis module performs time series analysis on the historical evaluation data to generate a cognitive function change trajectory and trend prediction for the individual.

[0124] An experimental animal cognitive function detection method, comprising the following steps:

[0125] Receive and preprocess the original behavior data stream and real-time physiological data to obtain standardized behavior event sequences and feature indicators;

[0126] According to the evaluation requirements, generate or call an executable task instruction set to guide the experimental animals to complete the cognitive testing tasks;

[0127] Fuse the behavior event sequences and feature indicators generated during the testing process to generate multi-modal behavior-physiological feature vectors;

[0128] Input the multi-modal behavior-physiological feature vectors into a pre-trained cognitive function evaluation model to obtain a quantitative evaluation result containing at least one cognitive dimension score;

[0129] Store and visualize quantification results, relevant features and raw data indices.

Claims

1. An experimental animal cognitive function evaluation system, comprising: a behavior data acquisition module configured to receive a stream of unstructured raw behavior data from a plurality of behavior testing devices and pre-process the stream of raw behavior data to obtain a sequence of standardized behavior events; a physiological data acquisition module configured to receive real-time physiological data from a plurality of physiological signal acquisition devices and extract feature indicators related to cognitive activities therefrom; a task management and generation module configured to store a plurality of templates of cognitive function testing paradigms and generate executable task instruction sets according to evaluation instructions or adaptive optimization instructions; a cognitive function analysis module in communication connection with the behavior data acquisition module, the physiological data acquisition module and the task management and generation module, the cognitive function analysis module being configured to receive the sequence of standardized behavior events output by the behavior data acquisition module, the feature indicators extracted by the physiological data acquisition module and the task instruction sets generated by the task management and generation module and analyze cognitive functions of experimental animals based on the above data.

2. The experimental animal cognitive function evaluation system according to claim 1, characterized by, The cognitive function analysis module comprises: a feature fusion unit configured to perform spatio-temporal alignment and correlation fusion of the sequence of behavior events and the feature indicators to generate a multi-modal behavior-physiological feature vector; a cognitive evaluation model unit storing a cognitive function evaluation model configured to receive the multi-modal behavior-physiological feature vector and output quantitative results including cognitive dimension scores and a comprehensive cognitive evaluation score; a result output and visualization module configured to visualize the quantitative results, the multi-modal behavior-physiological feature vector and raw data indexes and generate a structured evaluation report; a database configured to store the stream of raw behavior data, the real-time physiological data, the task instruction sets, the multi-modal behavior-physiological feature vector and the quantitative results in association.

3. The experimental animal cognitive function evaluation system according to claim 2, characterized by The cognitive function analysis module further comprises a cognitive function model construction unit configured to: construct a training sample set based on data stored in the database in the past, the training sample set comprising multi-modal behavior-physiological feature vectors corresponding to samples with known cognitive state labels.

4. The experimental animal cognitive function evaluation system according to claim 3, characterized by train the training sample set using a machine learning algorithm to obtain or optimize the cognitive function evaluation model, wherein the cognitive function evaluation model is configured to map a complex non-linear relationship between the multi-modal behavior-physiological feature vector and a cognitive state.

5. The experimental animal cognitive function evaluation system according to claim 4, characterized by further comprising an adaptive optimization module configured to: receive a current quantitative result from the cognitive function analysis module; determine whether to trigger optimization of task parameters based on the current quantitative result; if so, search in a parameter space based on a predefined optimization objective function and the cognitive function evaluation model using an optimization algorithm to generate an updated task parameter set that can make the optimization objective function better; send the updated task parameter set to the task management and generation module to generate a new executable task instruction set.

6. The experimental animal cognitive function evaluation system according to claim 5, wherein The optimization objective function is to maximize the information entropy of evaluation results.

7. The experimental animal cognitive function evaluation system according to claim 1, wherein The task management and generation module is further configured with a task verification unit configured to: After generating the executable task instruction set, perform logical consistency verification on the task instruction set based on a pre-stored animal behavior rule library and task parameter constraint conditions; If the verification fails, feedback adjustment suggestions or automatically adjust parameters until the executable task instruction set that passes the verification is generated.

8. The experimental animal cognitive function evaluation system according to claim 1, characterized by, The system further comprises a longitudinal tracking analysis module configured to: For a specific experimental animal individual, extract its historical evaluation data at different time points from the database, including historical multi-modal behavior-physiological feature vectors and historical quantitative results.

9. The experimental animal cognitive function evaluation system according to claim 8, characterized by, The longitudinal tracking analysis module performs time series analysis on the historical evaluation data to generate the cognitive function change trajectory and trend prediction of the individual.

10. A method for detecting cognitive function of an experimental animal based on the system according to any one of claims 1 to 9, characterized by, Comprising the following steps: Receive and pre-process raw behavior data streams and real-time physiological data to obtain standardized behavior event sequences and feature indicators; Generate or call executable task instruction sets according to evaluation requirements to guide experimental animals to complete cognitive test tasks; Fuse the behavior event sequences generated during the test with the feature indicators to generate multi-modal behavior-physiological feature vectors; Input the multi-modal behavior-physiological feature vectors into a pre-trained cognitive function evaluation model to obtain quantitative evaluation results containing at least one cognitive dimension score; Store and visualize the quantitative evaluation results, related features, and original data index.

Images