Method for constructing a mouse ischemic injury model for assessing the protective effect of GDF-15
By constructing an individual vulnerability index and using convolutional neural network dynamic simulation of a mouse ischemic injury model, the problems of poor model uniformity and high failure rate in existing technologies are solved, enabling real-time evaluation and optimization of model quality and improving the success rate and efficiency of the model.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- THE FIRST MEDICAL CENT CHINESE PLA GENERAL HOSPITAL
- Filing Date
- 2026-03-16
- Publication Date
- 2026-07-10
AI Technical Summary
Existing ischemic injury models do not consider individual differences during the construction process, resulting in poor model uniformity, high failure rate, lack of dynamic optimization mechanism, reliance on post-hoc biochemical index verification, and low efficiency.
By acquiring physiological parameter data from a mouse ischemic injury database, an individual vulnerability index was constructed. A convolutional neural network with an attention mechanism was used for dynamic simulation to evaluate the model quality in real time. If the model was deemed unqualified, similarity matching and model optimization were performed.
This approach enables the quantitative characterization of physiological differences among individual mice, improves the uniformity and success rate of the model, dynamically evaluates model quality, and reduces the modeling failure rate.
Smart Images
Figure CN122369972A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of ischemic injury model construction technology, and in particular to a method for constructing a mouse ischemic injury model for evaluating the protective effect of GDF-15. Background Technology
[0002] Ischemic injury models are important tools for studying the pathological mechanisms of ischemic diseases and assessing the protective effects of drugs. Growth differentiation factor-15 (GDF-15), as a stress response protein, has a potential protective role in ischemic injury. Therefore, constructing stable and reliable ischemic injury models is of great significance for assessing the protective effect of GDF-15.
[0003] Chinese Patent Publication No. CN118058234A discloses a method for constructing an animal model of cardiac and hepatic injury after limb ischemia-reperfusion. The relevant technical solution selects healthy Bama miniature pigs of similar age and weight, and implements ischemia for 6 hours by clamping the left external iliac artery and vein combined with a rotary tourniquet to restore blood flow. The success of the model construction is then verified by statistical analysis of postoperative serological and tissue homogenate indicators. The success criterion for this method is a significant difference in all indicators between the model group and the control group. However, the above method still has the following problems:
[0004] Existing technical solutions do not consider the physiological differences between individual experimental animals (such as baseline blood flow, body temperature, and heart rate variability), resulting in significant differences in the actual degree of damage caused by the same surgical procedure on different individuals, leading to poor model uniformity. The judgment of model quality relies on post-operative biochemical index detection and histological verification, making it impossible to conduct real-time evaluation and intervention during model construction, resulting in a high modeling failure rate. There is a lack of feedback optimization mechanisms, so when the model construction results are not ideal, it is impossible to adjust the modeling parameters based on existing data, and experiments can only be repeated, which is inefficient. Existing methods mainly rely on human experience to judge model quality, lacking quantitative and dynamically adjustable evaluation standards.
[0005] Therefore, there is an urgent need for a method to construct ischemic injury models that can take into account individual differences, achieve dynamic assessment, and execute optimization strategies based on assessment results, in order to solve the problems of poor model uniformity and high model failure rate. Summary of the Invention
[0006] To address this, the present invention provides a method for constructing a mouse ischemic injury model for evaluating the protective effect of GDF-15, thereby overcoming the problems in the prior art that fail to consider individual differences, rely on post-validation for model quality, and lack dynamic optimization mechanisms, resulting in poor model uniformity and high model failure rate.
[0007] To achieve the above objectives, the present invention provides a method for constructing a mouse ischemic injury model for evaluating the protective effect of GDF-15, comprising: Obtain the physiological parameter dataset from the mouse ischemic injury database, which includes blood flow data, body temperature data, and heart rate data; Physiological feature vectors are determined based on the aforementioned physiological parameter dataset, and individual vulnerability indices for several mice are determined based on the physiological feature vectors. The physiological feature vectors include mean blood flow, mean body temperature, and heart rate variability. The individual vulnerability indices are matched with a specific individual vulnerability index based on similarity, and the corresponding matched individual vulnerability index is determined based on the similarity matching results. A matching sample set is constructed based on the corresponding matched individual vulnerability index, intermediate monitoring dataset, and infarct area percentage dataset. The intermediate monitoring dataset includes time-series data of blood flow after ischemia, rate of blood flow decline, and duration of ischemia, while the infarct area percentage dataset includes data on the percentage of infarct area. An ischemic injury model is constructed using the matching sample set as training data through a convolutional neural network architecture that integrates attention mechanisms and multi-scale features. The specific individual vulnerability index, the specific intermediate monitoring dataset, and the specific infarct area percentage dataset are input into the ischemic injury model for dynamic simulation to output a protective effect simulation index. The specific intermediate monitoring dataset includes specific post-ischemic blood flow time-series data, specific blood flow decline rate, and specific ischemic duration. The specific infarct area percentage dataset includes specific infarct area percentage data. The suitability of the ischemic injury model is determined based on the comparison results between the simulated protective effect index and the preset simulated protective effect index. If the identification result is unqualified, the similarity matching and model are re-constructed based on the degree of deviation determined by several corresponding matching intermediate monitoring datasets and the specific intermediate monitoring dataset.
[0008] Furthermore, the process of determining the physiological feature vector based on the physiological parameter dataset includes: The mean values of the blood flow data and the body temperature data are calculated separately to determine the corresponding mean values of the blood flow and the body temperature. Frequency domain analysis was performed on the heart rate data to determine the corresponding heart rate variability.
[0009] Furthermore, the process of determining the individual vulnerability index of several mice based on physiological feature vectors includes: The corresponding individual vulnerability index is determined by weighting the mean blood flow, mean body temperature, and heart rate variability.
[0010] Furthermore, the process of performing similarity matching between the plurality of individual vulnerability indices and a specific individual vulnerability index includes: Calculate the Euclidean distance between several individual vulnerability indices and the specific individual vulnerability index to determine the corresponding deviation. Based on the comparison result between the deviation and the preset deviation, it is determined whether the current individual vulnerability index is the corresponding matching individual vulnerability index.
[0011] Furthermore, the process of determining whether the current individual vulnerability index is the corresponding matching individual vulnerability index includes: The individual vulnerability index that has a deviation less than a preset deviation is statistically analyzed, and the individual vulnerability index is recorded as the corresponding matching individual vulnerability index.
[0012] Furthermore, the process of constructing the matching sample set includes: Based on the corresponding matched individual vulnerability index, the corresponding matched physiological parameter dataset in the mouse ischemic injury database is determined, and based on the corresponding matched physiological parameter dataset, the corresponding matched intermediate monitoring dataset and the corresponding matched infarct area percentage dataset in the mouse ischemic injury database are determined. The corresponding matching sample set is determined by associating the individual vulnerability index, the intermediate monitoring dataset, and the infarct area percentage dataset with corresponding matching data using unique identifiers.
[0013] Furthermore, the process of constructing the ischemic injury model includes: The convolutional neural network is trained using the set of matched samples, wherein, The matching sample set is used as training samples. The input features of each training sample include the individual vulnerability index, the time series data of blood flow after ischemia, the rate of blood flow decline, and the duration of ischemia. The output is the evaluation result of the protective effect of GDF-15 determined based on the infarct area percentage dataset. Several convolutional layer branches are constructed to extract multi-scale features from the input features in order to generate a multi-scale feature neural network; The multi-scale feature neural network is based on channel attention and spatial attention to enhance the expression of key features, and outputs the pass probability after passing through a fully connected layer and a sigmoid activation function. The parameters of the multi-scale feature neural network are optimized by backpropagation to generate the ischemic injury model.
[0014] Furthermore, the process of determining whether the ischemic injury model is qualified includes: The specific individual vulnerability index, the specific intermediate monitoring dataset, and the specific infarct area percentage dataset are input into the ischemic injury model for dynamic simulation to output the protective effect simulation index used to evaluate whether the ischemic injury model is qualified. The ischemic injury model is deemed qualified based on the comparison results where the protective effect simulation index is greater than or equal to the preset protective effect simulation index. Based on the comparison results where the simulated protective effect index is less than the preset simulated protective effect index, the ischemic injury model is determined to be unqualified.
[0015] Furthermore, the process of re-performing similarity matching and rebuilding the model includes: The mean is calculated based on the intermediate monitoring dataset that is currently matched to obtain the intermediate monitoring mean set, wherein the intermediate monitoring mean set includes the average blood flow decline rate and the average ischemia duration. The difference between the intermediate monitoring mean set and a specific intermediate monitoring mean set is calculated to determine the corresponding degree of deviation, wherein the specific intermediate monitoring mean set includes the mean of a specific blood flow decrease rate and a specific ischemia duration. The preset deviation is adjusted based on the degree of deviation, and similarity matching and model construction are performed again based on the adjusted preset deviation.
[0016] Furthermore, the process of adjusting the preset deviation based on the degree of deviation includes: The increase in the preset deviation is determined based on the degree of deviation, wherein the increase in the preset deviation is positively correlated with the degree of deviation.
[0017] Compared with existing technologies, the mouse ischemic injury model construction method of the present invention for evaluating the protective effect of GDF-15 has the following advantages: It achieves quantitative characterization of individual physiological differences in mice by comprehensively considering mean blood flow, mean body temperature, and heart rate variability through individual vulnerability indices, providing a foundation for subsequent personalized model construction; by performing similarity matching of individual vulnerability indices, it can select historical sample groups with the closest physiological characteristics for each target sample, constructing a targeted matching sample set, thus solving the problem of poor model uniformity caused by neglecting individual differences when using a uniform surgical procedure in existing technologies; and by constructing an ischemic injury model using a convolutional neural network that integrates attention mechanisms and multi-scale features with the matching sample set as training data, it can effectively address the ischemic injury from the perspective of ischemia... Multi-scale features related to the assessment of GDF-15 protective effects are automatically extracted from post-blood flow time-series data, blood flow decline rate, and ischemia duration. The expression of key features is enhanced through an attention mechanism, enabling accurate prediction of model quality. By outputting protective effect simulation indicators and comparing them with preset protective effect simulation indicators, the applicability of the model to the assessment of GDF-15 protective effects can be dynamically determined during the model construction stage, overcoming the lag of existing technologies that rely on post-event biochemical indicator verification. When the identification result is unqualified, the similarity matching and model construction are re-performed based on the degree of deviation between the corresponding matched intermediate monitoring dataset and the specific intermediate monitoring dataset, enabling the model to have self-optimization capabilities and significantly improving the modeling success rate and model quality.
[0018] Furthermore, this invention also calculates the mean values of blood flow and body temperature separately to determine the corresponding mean values; and performs frequency domain analysis on heart rate data to determine the corresponding heart rate variability, thereby obtaining stable mean values of blood flow and body temperature, eliminating random errors that may be caused by a single measurement; by extracting heart rate variability through frequency domain analysis, the regulatory function of the mouse's autonomic nervous system can be quantified. Heart rate variability, as an important indicator reflecting an individual's stress response ability, together with the mean values of blood flow and body temperature, constitutes the basis of the individual vulnerability index, making the quantification of individual differences more comprehensive and accurate. The level of the individual vulnerability index directly reflects the mouse's intrinsic tolerance potential to ischemic injury. A high individual vulnerability index value indicates poorer tolerance and a greater likelihood of excessive ischemia, while a low individual vulnerability index value indicates better tolerance and a greater likelihood of insufficient ischemia. This provides a unified quantitative assessment standard, enabling mice with different physiological characteristics to be compared and matched on the same scale.
[0019] Furthermore, this invention determines whether the current individual vulnerability index is a corresponding matching individual vulnerability index by calculating the deviation degree and comparing it with a preset deviation degree. It also counts individual vulnerability indices with deviation degrees less than the preset deviation degree and records these as corresponding matching individual vulnerability indices. Then, based on the corresponding matching individual vulnerability indices, it determines the corresponding matching intermediate monitoring dataset and the corresponding matching infarct area percentage dataset in the mouse ischemic injury database. Finally, based on the corresponding matching individual vulnerability indices, the corresponding matching intermediate monitoring dataset, and the corresponding matching infarct area percentage dataset, it determines the corresponding matching sample set. This achieves a quantitative measurement of physiological similarity between samples, objectively filters historical samples with similar physiological characteristics to the target sample, provides a scientific basis for constructing the matching sample set, and improves the prediction accuracy of the model corresponding to the matching sample set in the local sample space.
[0020] Furthermore, this invention also performs multi-scale feature extraction by constructing several convolutional layer branches; calculates the importance weight of each feature channel through a channel attention mechanism to enhance the response of the feature channel most relevant to the qualification judgment; generates a spatial weight map through a spatial attention mechanism to enhance the response at key time locations; the feature vector enhanced by dual attention outputs the qualification probability through a fully connected layer and a sigmoid function; optimizes the network parameters through backpropagation to make the output probability approximate the true label; the finally generated ischemic injury model can output the evaluation result of the protective effect on any new sample input, and then, by using standard parameters (specific individual easy...) The model is validated by inputting the damage index, a specific intermediate monitoring dataset, and a specific infarct area percentage dataset into the currently generated ischemic injury model. The evaluation results output by the model reflect the response characteristics of the current model to the standard input, which is the protective effect simulation index. By comparing it with the preset protective effect simulation index, it is possible to objectively determine whether the currently constructed ischemic injury model is qualified. This realizes the dynamic evaluation of the quality of the model itself during the model construction stage, overcomes the lag of traditional methods that rely on post-event biochemical index validation, and can obtain quality feedback in a timely manner after the model is constructed.
[0021] Furthermore, this invention also quantifies the degree of deviation between the current sample and the similar group by calculating the mean of intermediate monitoring data of historical samples in the matching sample set and comparing it with specific intermediate monitoring data of the current sample when the model determines that it is unqualified. The greater the degree of deviation, the more the ischemic process characteristics of the current sample deviate from the typical pattern of its physiologically similar group, which may be the reason for the model's unqualified determination. By adjusting the preset deviation threshold based on the degree of deviation quantification, more historical samples with certain similarities to the current sample can be included in the matching set. Based on the adjusted preset deviation, similarity matching and model construction are carried out again, so that the model can learn the rules that can correctly evaluate the current sample in a wider sample group, forming a complete feedback optimization control strategy. Attached Figure Description
[0022] Figure 1 This is a flowchart of a method for constructing a mouse ischemic injury model to evaluate the protective effect of GDF-15 in an embodiment of the present invention; Figure 2 This is a logical decision diagram for constructing a matching sample set in an embodiment of the present invention; Figure 3 This is a logic diagram for determining whether an ischemic injury model is qualified in an embodiment of the present invention; Figure 4 This is a logical decision diagram for re-similarity matching and model construction in an embodiment of the present invention. Detailed Implementation
[0023] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.
[0024] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.
[0025] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.
[0026] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.
[0027] Please see Figure 1 The diagram shows a flowchart of a method for constructing a mouse ischemic injury model to evaluate the protective effect of GDF-15, as illustrated in this embodiment of the invention. The process includes at least the following steps:
[0028] S1: Obtain the physiological parameter dataset from the mouse ischemic injury database, which includes blood flow data, body temperature data, and heart rate data; S2: Determine physiological feature vectors based on physiological parameter datasets, and determine individual vulnerability indices for several mice based on physiological feature vectors. The physiological feature vectors include mean blood flow, mean body temperature, and heart rate variability. S3: Perform similarity matching between several individual vulnerability indices and a specific individual vulnerability index, and determine the corresponding matched individual vulnerability index based on the similarity matching results; S4: Construct a matching sample set based on the corresponding matched individual vulnerability index, intermediate monitoring dataset, and infarct area percentage dataset. The intermediate monitoring dataset includes time-series data of blood flow after ischemia, blood flow decline rate, and ischemia duration. The infarct area percentage dataset includes infarct area percentage data. S5: Construct an ischemic injury model using a convolutional neural network architecture that integrates attention mechanisms and multi-scale features, with a matching sample set as training data; S6: Input the specific individual vulnerability index, specific intermediate monitoring dataset, and specific infarct area percentage dataset into the ischemic injury model for dynamic simulation to output the protective effect simulation index. The specific intermediate monitoring dataset includes specific post-ischemic blood flow time series data, specific blood flow decline rate, and specific ischemic duration. The specific infarct area percentage dataset includes specific infarct area percentage data. S7: Determine whether the ischemic injury model is qualified based on the comparison results between the protective effect simulation index and the preset protective effect simulation index; S8: If the identification result is unqualified, re-match similarity and build the model based on the degree of deviation determined by several corresponding intermediate monitoring datasets and a specific intermediate monitoring dataset.
[0029] S9: If the identification result is qualified, it is determined that the currently constructed ischemic injury model can be used to evaluate the GDF-15 protective effect in the target mice.
[0030] In one specific embodiment, a large dataset of physiological parameters, including blood flow data, body temperature data, and heart rate data, of historical mice is obtained from a mouse ischemia injury database. The physiological parameter dataset consists of initial data collected from mice before ischemia induction.
[0031] Specifically, blood flow data of the target area (such as myocardium and hind limb muscles) is obtained by laser Doppler flowmeter, with the unit being perfusion units (PU); body temperature data is obtained by infrared thermometer or rectal temperature probe, with the unit being degrees Celsius (°C); and heart rate data is obtained by electrocardiogram measurement, with the unit being beats per minute (bpm).
[0032] In one specific embodiment, the corresponding individual vulnerability index is determined by acquiring the physiological feature vector of each mouse. The mean blood flow (BF) is determined based on the arithmetic mean of blood flow data, and BF reflects the individual's basal perfusion level. The mean body temperature (T) is determined based on the arithmetic mean of body temperature data, and T reflects the individual's basal body temperature. The heart rate variability (HRV) is determined based on frequency domain analysis of heart rate data.
[0033] Specifically, the heart rate time-series data were first subjected to a Fast Fourier Transform (HF) to convert the time-domain signal to the frequency domain. High-frequency power (HF) (0.15Hz–0.4Hz) and low-frequency power (LF) (0.04–0.15Hz) were extracted, and heart rate variability (HRV) was calculated: HRV = HF / LF. Here, HF reflects parasympathetic activity, and LF reflects the mixed activity of the sympathetic and parasympathetic nervous systems. A larger HRV ratio indicates stronger autonomic nervous system regulation and potentially better individual tolerance to ischemic stress.
[0034] In this embodiment, the mean blood flow (BF), mean body temperature (T), and heart rate variability (HRV) are used to construct a physiological feature vector [BF, T, HRV]. Based on this physiological feature vector, an individual vulnerability index (IVI) is determined through weighted calculation. The specific calculation formula is: IVI = α·(1 / BF) + β·(|T-Tref| / Tref) + γ·(1 / HRV), where: BF is the mean blood flow, and its reciprocal 1 / BF reflects that the lower the baseline blood flow, the higher the vulnerability; Tref is the reference body temperature value, which is taken as the normal body temperature of mice, 37.0℃. |T-Tref| / Tref reflects the degree of deviation of the individual's body temperature from the normal range. The greater the deviation, the more unstable the physiological state and the higher the vulnerability; 1 / HRV reflects the heart rate variability. The lower the heterogeneity, the higher the vulnerability. α, β, and γ are the vulnerability index weighting coefficients, which can be determined through statistical analysis of historical data. Specifically, samples are selected from a historical database, with the percentage of infarct area as the dependent variable and 1 / BF, |T-Tref| / Tref, and 1 / HRV as independent variables. Multiple linear regression is used to fit the data, and the regression coefficients are used as the initial values for the vulnerability index weighting coefficients. For example, α=0.4, β=0.3, and γ=0.3. A higher IVI value indicates poorer individual tolerance (lower baseline blood flow, larger body temperature fluctuations, and lower heart rate variability), making them more prone to excessive ischemia under the same surgical conditions. A lower IVI value indicates better individual tolerance, making them more prone to insufficient ischemia. This index provides a unified quantitative scale for subsequent similarity matching, enabling mice with different physiological characteristics to be compared on the same dimension.
[0035] Please see Figure 2 As shown, this is a logical decision diagram for constructing a matching sample set in an embodiment of the present invention. Several individual vulnerability indices (IVI) are matched with a specific individual vulnerability index (IVIta) for similarity. Based on the similarity matching results, the corresponding matching individual vulnerability index (IVIth) is determined. A matching sample set is then constructed based on the corresponding matching individual vulnerability index (IVIth), the corresponding matching intermediate monitoring dataset, and the corresponding matching infarct area percentage dataset.
[0036] Specifically, for the current model that needs to be constructed for the target mouse type, its specific individual vulnerability index is denoted as IVIta. The individual vulnerability indices IVIi (i=1,…,N) of all historical mice are obtained from the historical database, and the Euclidean distance between each IVIi and IVIta is calculated as the deviation degree di, with the specific formula: di=|IVIi-IVIta|, where the smaller the deviation degree di, the more similar the physiological characteristics of the historical mouse are to the target mouse.
[0037] In this embodiment, based on a preset deviation δ0, historical mice with di < δ0 are used as matching sample mice, and the individual vulnerability index IVIi corresponding to the matching sample mice is recorded as the corresponding matched individual vulnerability index IVIth. The physiological parameter dataset corresponding to IVIth of the matching sample mice is indexed from the mouse ischemic injury database using a unique identifier. Each historical record in the mouse ischemic injury database has a unique sample identifier, which is associated with all data tables of that sample, including: physiological parameter dataset, intermediate monitoring dataset, infarct area percentage dataset, and the corresponding individual vulnerability index IVI. The corresponding matched intermediate monitoring dataset and the corresponding matched infarct area percentage dataset are determined based on the corresponding matched physiological parameter dataset. The preset deviation δ0 is determined based on the distribution characteristics of all individual vulnerability indices IVI in the historical database. Specifically, the standard deviation σIVI of all individual vulnerability indices IVI in the historical database is calculated, and 0.5·σIVI is taken as the initial value to ensure that the matching sample set can cover a reasonable range similar to the physiological characteristics of the target sample. For example, δ0 can be set to 0.05. The matching sample set was constructed by matching the individual vulnerability index (IVIth) of the matched sample mice, the intermediate monitoring dataset, and the percentage of infarct area.
[0038] Specifically, the intermediate monitoring dataset includes: time-series data on blood flow after ischemia: recording blood flow changes from the onset to the end of ischemia, sampled at a fixed frequency (1Hz), and uniformly downsampled to 100 time points to form a 100-dimensional time-series vector, used to capture dynamic blood flow change patterns; blood flow descent rate: the average rate at which blood flow decreases from the mean blood flow (BF) to the lowest point after the onset of ischemia. The blood flow descent rate reflects the severity of the surgical procedure; a rate that is too fast may indicate that the ligation is too tight, while a rate that is too slow may indicate that the ligation is insufficient; duration of ischemia: the length of time (in minutes) from the onset of ischemia to reperfusion or blood flow stabilization, the value of which directly determines the degree of tissue damage; and the infarct area percentage dataset, which is the percentage of the infarct area relative to the corresponding organ determined by methods such as TTC staining in the historical mouse, used to measure the degree of ischemic damage. According to the requirements of GDF-15 protective effect assessment, if the infarct area is too small, the protective effect cannot be detected, and if it is too large, it may mask the protective effect. Therefore, it is necessary to control the degree of model damage within a moderate range. In this embodiment, samples with an infarct area between 25% and 35% are marked as qualified models, and others are marked as unqualified models.
[0039] In one specific embodiment, the current set of matched samples is used as training samples. The input features of each training sample include: individual vulnerability index (IVI), time-series data of blood flow after ischemia, rate of blood flow decline, and duration of ischemia. The output label is the GDF-15 protective effect assessment result determined based on the infarct area percentage dataset. If the infarct area is between 25% and 35%, it is marked as a qualified model; otherwise, it is marked as an unqualified model.
[0040] Specifically, to fully capture the physiological change patterns at different time scales contained in the post-ischemic blood flow time-series data, this embodiment preferably constructs three parallel convolutional layer branches, each using a one-dimensional convolutional kernel of different sizes to extract multi-scale features from the input features. The input features are a preprocessed and downsampled 100-dimensional post-ischemic blood flow time-series vector, along with other scalar features (individual vulnerability index, blood flow decline rate, and ischemia duration) concatenated with it, resulting in a total feature dimension of 103, including:
[0041] The branch for extracting local detail features uses a convolutional kernel of size 3 to capture local abrupt changes in blood flow time-series data, such as sudden drops or rapid recoveries in blood flow. The branch for extracting medium-scale features uses a convolutional kernel of size 5 to extract morphological features at specific stages of ischemia, such as the overall slope of the descent phase or the rate of increase in the recovery phase. The branch for extracting global trend features uses a convolutional kernel of size 7 to perceive the macroscopic evolution trend of the entire ischemic cycle, distinguishing between different patterns such as continuous descent or descent-recovery. The technique of using convolutional kernels to extract multi-scale features from input features is existing technology and will not be elaborated further.
[0042] Each convolutional layer is followed by a batch normalization layer to accelerate training stabilization, and then a max pooling layer (MaxPooling1D, poolsize=2) is connected to reduce the feature dimension. The outputs of the three convolutional layer branches are then concatenated along the channel dimension to generate a feature map that integrates multi-scale information, with a shape of (time step × 96 channels). Global average pooling and global max pooling are then performed on the feature map to obtain two C-dimensional vectors (C=96). These two vectors are input into a shared fully connected network to generate channel weight vectors (dimension C). After sigmoid activation, these weights are multiplied channel-by-channel with the original feature map to enhance the response of the feature channels most relevant to the pass / fail judgment.
[0043] Subsequently, the channel-weighted feature map is subjected to average pooling and max pooling along the channel dimension to obtain two two-dimensional matrices (time step · 1). After concatenation, a spatial weight map (time step · 1) is generated through a convolutional layer (kernel size = 7). After Sigmoid activation, it is multiplied element-wise with the feature map to enhance the response at key time locations.
[0044] Furthermore, the attention-enhanced feature maps are compressed into one-dimensional feature vectors. Both fully connected layers (64 neurons in the first layer and 32 neurons in the second layer) are activated using the ReLU activation function, and each layer is followed by Dropout (dropout rate 0.3) to prevent overfitting.
[0045] For a single neuron, the Sigmoid activation function is used to output a probability value between 0 and 1, which is the pass / fail probability, indicating the likelihood that the current sample is suitable for the GDF-15 protective effect assessment.
[0046] The network was trained using training samples, employing a binary classification cross-entropy loss function. The network parameters were iteratively updated using the Adam optimizer, with an initial learning rate of 0.001. After training, a well-trained ischemic injury model was obtained, with its network parameters fixed.
[0047] Please see Figure 3 As shown, this is a logic diagram for determining whether an ischemic injury model is qualified in an embodiment of the present invention. The specific individual vulnerability index IVIta of the target mouse, specific intermediate monitoring data (specific post-ischemic blood flow time-series data, specific blood flow decline rate, specific ischemic duration), and a specific infarct area percentage dataset are input into the trained ischemic injury model. After forward computation, the model outputs a probability value, which is the protective effect simulation index, denoted as P. The P value reflects the current model's response characteristics to standard input. The qualified probability output by the model under standard input can be used as a quantitative basis for measuring the quality of the model itself. The qualification of the ischemic injury model is determined based on the comparison between the protective effect simulation index P and the preset protective effect simulation index Pth.
[0048] The specific individual vulnerability index (IVIta), specific intermediate monitoring data, and specific infarct area percentage dataset serve as standardized test inputs for uniformly validating model quality. The specific individual vulnerability index (IVIta) can be calculated by averaging the corresponding individual vulnerability indices (IVI) from all qualified model construction records in the historical database. The specific post-ischemic blood flow time-series data in the specific intermediate monitoring data can be taken from the average time-series curve of the same qualified sample set; the specific blood flow decline rate is taken as the average blood flow decline rate of that sample set; and the specific ischemic duration is taken as the average ischemic duration of that sample set. The specific infarct area percentage dataset can be the median value of the qualified interval.
[0049] For example, the specific individual vulnerability index IVIta is set to 0.58, the specific blood flow decline rate is set to 0.23% / s, the specific ischemia duration is set to 45 minutes, and the specific infarct area percentage is set to 30%. The preset protective effect simulation index Pth is determined based on historical experimental data. Specifically, all model construction records with infarct area between 25% and 35% that have been successfully verified by subsequent experiments are selected from the historical database. The output probability distribution of the model to standard input in these records is statistically analyzed, and the median is taken as the threshold. For example, Pth can be set to 0.7.
[0050] If P≥Pth, it indicates that the probability of the current ischemic injury model meeting the standardized test input and output is at or above the typical level of a successful model (i.e., the preset threshold), and the model performance meets the requirements of GDF-15 protective effect assessment. Therefore, the current ischemic injury model is deemed qualified and can be used for subsequent GDF-15 protective effect assessment. If P < Pth, it indicates that the current model's probability of passing the standardized test input and output is lower than the typical level of a successful model, and it is therefore deemed unqualified, and an optimized model is determined.
[0051] Please see Figure 4 As shown, this is a logical decision diagram for re-matching similarity and building the model in an embodiment of the present invention. When the model is deemed unqualified, it indicates that the current set of matching samples for the target mouse may be insufficient to train a model that can correctly evaluate the sample, and the matching conditions need to be adjusted to include more diverse historical samples.
[0052] Specifically, the mean of intermediate monitoring data for all historical samples in the matched sample set is calculated, including the average blood flow decline rate μr and the average ischemia duration μd. The corresponding mean values in a specific intermediate monitoring dataset include a specific blood flow decline rate μrth and a specific ischemia duration mean μdth. Then, based on the calculated deviation degree D, the degree of deviation of the current sample's ischemia process characteristics from the typical pattern of its peer group (matched samples) is determined. A larger D value indicates a more abnormal ischemia process in the current sample, resulting in a greater deviation from the typical pattern of its peer group, thus leading to a model failure.
[0053] Wherein, the degree of deviation D = w1·|μrth-μr|+w2·|μdth-μd|, where w1 and w2 are deviation weight coefficients, which can be set according to the degree of influence of the index on the model quality. In this embodiment, w1=0.6 and w2=0.4 are preferred.
[0054] In this embodiment, the preset deviation threshold δ0 is adjusted based on the degree of deviation D, and the adjustment amount Δδ is positively correlated with D. The specific formula is: Δδ=k·D, where k is a preset coefficient, preferably set to k=0.5; then the adjusted preset deviation δnew is determined: δnew=δ0+Δδ.
[0055] In one specific embodiment, if the current ischemic injury model is determined to be unqualified, a new set of matching samples is obtained by re-matching similarity using δnew. Then, the ischemic injury model is reconstructed using the new set of matching samples, and simulation is performed again to generate Pnew. Finally, Pnew is compared with Pth. If Pnew≥Pth, the model is deemed qualified, optimization stops, and the current model can be used for subsequent GDF-15 protection effect assessment. If Pnew < Pth, the result is still unqualified. In this case, it is determined whether the current iteration count has reached the preset maximum iteration count (e.g., 3 times). If the maximum number of iterations has not been reached, the mean blood flow descent rate μr' and the mean ischemia duration μd' are recalculated based on the new matching sample set. The new deviation degree Dnew is obtained by comparing it with the corresponding average value in the specific intermediate monitoring dataset of the target mouse, and δ (δnew2=δnew+k·Dnew) is adjusted again to continue the loop.
[0056] If the maximum number of iterations has been reached, the optimization process is terminated, the sample is marked as an "edge sample" and stored in the historical database for subsequent analysis. It is also recommended that the original data and modeling process of the sample be manually reviewed to identify possible anomalies (such as data recording errors, extreme abnormalities in surgical procedures, etc.) to ensure the reliability of the model construction.
[0057] All technologies not mentioned in the above embodiments are applicable to existing technologies. It is understood that no specific limitation is made to any preset parameter or critical parameter in the embodiments of the present invention, and the above values are not limited thereto. Those skilled in the art can adjust the preset parameters or critical parameters accordingly based on actual needs, analysis of historical data, or equipment usage.
[0058] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.
Claims
1. A method for constructing a mouse ischemic injury model to evaluate the protective effect of GDF-15, characterized in that, include: Obtain the physiological parameter dataset from the mouse ischemic injury database, which includes blood flow data, body temperature data, and heart rate data; Physiological feature vectors are determined based on the aforementioned physiological parameter dataset, and individual vulnerability indices for several mice are determined based on the physiological feature vectors. The physiological feature vectors include mean blood flow, mean body temperature, and heart rate variability. The individual vulnerability indices are matched with a specific individual vulnerability index based on similarity, and the corresponding matched individual vulnerability index is determined based on the similarity matching results. A matching sample set is constructed based on the corresponding matched individual vulnerability index, intermediate monitoring dataset, and infarct area percentage dataset. The intermediate monitoring dataset includes time-series data of blood flow after ischemia, rate of blood flow decline, and duration of ischemia, while the infarct area percentage dataset includes data on the percentage of infarct area. An ischemic injury model is constructed using the matching sample set as training data through a convolutional neural network architecture that integrates attention mechanisms and multi-scale features. The specific individual vulnerability index, the specific intermediate monitoring dataset, and the specific infarct area percentage dataset are input into the ischemic injury model for dynamic simulation to output a protective effect simulation index. The specific intermediate monitoring dataset includes specific post-ischemic blood flow time-series data, specific blood flow decline rate, and specific ischemic duration. The specific infarct area percentage dataset includes specific infarct area percentage data. The suitability of the ischemic injury model is determined based on the comparison results between the simulated protective effect index and the preset simulated protective effect index. If the identification result is unqualified, the similarity matching and model are re-constructed based on the degree of deviation determined by several corresponding matching intermediate monitoring datasets and the specific intermediate monitoring dataset.
2. The method for constructing a mouse ischemic injury model for evaluating the protective effect of GDF-15 according to claim 1, characterized in that, The process of determining physiological feature vectors based on the physiological parameter dataset includes: The mean values of the blood flow data and the body temperature data are calculated separately to determine the corresponding mean values of the blood flow and the body temperature. Frequency domain analysis was performed on the heart rate data to determine the corresponding heart rate variability.
3. The method for constructing a mouse ischemic injury model for evaluating the protective effect of GDF-15 according to claim 2, characterized in that, The process of determining the individual vulnerability index of several mice based on physiological feature vectors includes: The corresponding individual vulnerability index is determined by weighting the mean blood flow, mean body temperature, and heart rate variability.
4. The method for constructing a mouse ischemic injury model for evaluating the protective effect of GDF-15 according to claim 3, characterized in that, The process of matching the plurality of individual vulnerability indices with the specific individual vulnerability index includes: Calculate the Euclidean distance between several individual vulnerability indices and the specific individual vulnerability index to determine the corresponding deviation. Based on the comparison result between the deviation and the preset deviation, it is determined whether the current individual vulnerability index is the corresponding matching individual vulnerability index.
5. The method for constructing a mouse ischemic injury model for evaluating the protective effect of GDF-15 according to claim 4, characterized in that, The process of determining whether the current individual vulnerability index is the corresponding matching individual vulnerability index includes: The individual vulnerability index that has a deviation less than a preset deviation is statistically analyzed, and the individual vulnerability index is recorded as the corresponding matching individual vulnerability index.
6. The method for constructing a mouse ischemic injury model for evaluating the protective effect of GDF-15 according to claim 5, characterized in that, The process of constructing the matching sample set includes: Based on the corresponding matched individual vulnerability index, the corresponding matched physiological parameter dataset in the mouse ischemic injury database is determined, and based on the corresponding matched physiological parameter dataset, the corresponding matched intermediate monitoring dataset and the corresponding matched infarct area percentage dataset in the mouse ischemic injury database are determined. The corresponding matching sample set is determined by associating the individual vulnerability index, the intermediate monitoring dataset, and the infarct area percentage dataset with corresponding matching data using unique identifiers.
7. The method for constructing a mouse ischemic injury model for evaluating the protective effect of GDF-15 according to claim 6, characterized in that, The process of constructing the ischemic injury model includes: The convolutional neural network is trained using the set of matched samples, wherein, The matching sample set is used as training samples. The input features of each training sample include the individual vulnerability index, the time series data of blood flow after ischemia, the rate of blood flow decline, and the duration of ischemia. The output is the evaluation result of the protective effect of GDF-15 determined based on the infarct area percentage dataset. Several convolutional layer branches are constructed to extract multi-scale features from the input features in order to generate a multi-scale feature neural network; The multi-scale feature neural network is based on channel attention and spatial attention to enhance the expression of key features, and outputs the pass probability after passing through a fully connected layer and a sigmoid activation function. The parameters of the multi-scale feature neural network are optimized by backpropagation to generate the ischemic injury model.
8. The method for constructing a mouse ischemic injury model for evaluating the protective effect of GDF-15 according to claim 4, characterized in that, The process for determining whether the ischemic injury model is qualified includes: The specific individual vulnerability index, the specific intermediate monitoring dataset, and the specific infarct area percentage dataset are input into the ischemic injury model for dynamic simulation to output the protective effect simulation index used to evaluate whether the ischemic injury model is qualified. The ischemic injury model is deemed qualified based on the comparison results where the protective effect simulation index is greater than or equal to the preset protective effect simulation index. Based on the comparison results where the simulated protective effect index is less than the preset simulated protective effect index, the ischemic injury model is determined to be unqualified.
9. The method for constructing a mouse ischemic injury model for evaluating the protective effect of GDF-15 according to claim 8, characterized in that, The process of re-matching similarity and rebuilding the model includes: The mean is calculated based on the intermediate monitoring dataset that is currently matched to obtain the intermediate monitoring mean set, which includes the average blood flow rate and the average ischemia duration. The difference between the intermediate monitoring mean set and a specific intermediate monitoring mean set is calculated to determine the corresponding degree of deviation, wherein the specific intermediate monitoring mean set includes the mean of a specific blood flow decrease rate and a specific ischemia duration. The preset deviation is adjusted based on the degree of deviation, and similarity matching and model construction are performed again based on the adjusted preset deviation.
10. The method for constructing a mouse ischemic injury model for evaluating the protective effect of GDF-15 according to claim 9, characterized in that, The process of adjusting the preset deviation based on the degree of deviation includes: The increase in the preset deviation is determined based on the degree of deviation, wherein the increase in the preset deviation is positively correlated with the degree of deviation.