Method for training stroke recurrence risk prediction model and related product

Abstract

The invention discloses a method for training a stroke recurrence risk prediction model and a related product. The method comprises the following steps: acquiring cerebral perfusion image data, clinical data and follow-up visit data of a patient; the follow-up visit data comprises follow-up visit time and a survival state of the follow-up visit time; the survival state comprises whether stroke relapse occurs or not, whether sequelae exists or not, whether complications exist or not and whether death exists or not; data integration is carried out on the follow-up time and the survival state of the follow-up time to obtain at least one survival time-event indication pair, the event indication is used for identifying whether a preset event occurs in the survival time, and the preset event is stroke relapse; and inputting the brain perfusion image data, the clinical data, the follow-up visit data and the at least one survival time-event indication pair into the stroke recurrence risk prediction model as training data to train the stroke recurrence risk prediction model. By using the scheme disclosed by the invention, the prediction precision of the stroke recurrence risk can be improved.

Description

Technical Field

[0001] This disclosure generally relates to the field of computer vision technology. More specifically, this disclosure relates to a method, device, and computer-readable storage medium for training a stroke recurrence risk prediction model. Further, this disclosure also relates to a device and computer-readable storage medium for predicting stroke recurrence risk. Background Art

[0002] Stroke recurrence is one of the main factors leading to patient death and disability. Despite the significant progress of modern medical technology, how to accurately predict the risk of stroke recurrence remains a major challenge in the medical field. Currently, the risk assessment of stroke recurrence is usually based on clinical scoring scales, such as the NIHSS stroke scale. However, this scoring method cannot consider the imaging characteristics and long-term follow-up data of patients, resulting in limited prediction accuracy.

[0003] In recent years, with the development of medical imaging and artificial intelligence technologies, cerebral perfusion imaging (Computed Tomography Perfusion, CTP) images, as a non-invasive method, have been widely used in stroke diagnosis. By combining imaging characteristics and clinical data, a more accurate assessment of stroke recurrence risk can be achieved. However, this assessment method still does not consider the long-term follow-up data of patients, thus ignoring the health trends of patients over time and resulting in relatively low prediction accuracy.

[0004] In view of this, there is an urgent need to provide a solution for training a stroke recurrence risk prediction model so that the trained stroke recurrence risk prediction model can fully utilize the rich information in CTP image data, clinical data, and follow-up data, capture the health trends of patients over time, enhance the time sensitivity of the model to stroke recurrence risk, and improve the prediction accuracy of the model. Summary of the Invention

[0005] To at least solve one or more of the above-mentioned technical problems, this disclosure proposes a solution for predicting stroke recurrence risk in the following aspects.

[0006] In a first aspect, the present disclosure provides a method for training a stroke recurrence risk prediction model, the method comprising: obtaining cerebral perfusion imaging data, clinical data, and follow-up data of a patient; the follow-up data including a follow-up time and a survival status at the follow-up time; the survival status including whether a stroke recurrence has occurred, whether there are sequelae, whether there are complications, and whether death has occurred; integrating the follow-up time and the survival status at the follow-up time to obtain at least one survival time-event indication pair, the event indication being used to identify whether a preset event has occurred at the survival time, the preset event being a stroke recurrence; and inputting the cerebral perfusion imaging data, the clinical data, the follow-up data, and the at least one survival time-event indication pair as training data into the stroke recurrence risk prediction model to train it.

[0007] In some embodiments, the stroke recurrence risk prediction model comprises a feature extraction model and a survival analysis model; the feature extraction model comprises a first extraction model, a second extraction model, and a third extraction model; the survival analysis model comprises a first output layer and a second output layer.

[0008] In some embodiments, inputting the cerebral perfusion imaging data, the clinical data, the follow-up data, and the at least one survival time-event indication pair as training data into the stroke recurrence risk prediction model to train it comprises: inputting the cerebral perfusion imaging data, the clinical data, and the follow-up data into the first extraction model, the second extraction model, and the third extraction model respectively for feature extraction to obtain a first feature, a second feature, and a third feature; inputting the first feature, the second feature, and the third feature into the survival analysis model for survival analysis, so that the first output layer outputs a predicted survival risk ratio and the second output layer outputs a predicted stroke recurrence probability; determining a loss value based on the predicted survival risk ratio, the predicted stroke recurrence probability, and the at least one survival time-event indication pair, and updating the parameters of the stroke recurrence risk prediction model based on the loss value.

[0009] In some embodiments, the first output layer is a fully connected layer, and the second output layer is a fully connected layer and a Sigmoid function; and the first output layer outputting the predicted survival risk ratio comprises: the fully connected layer outputting a risk score, and exponentiating the risk score to obtain the predicted survival risk ratio.

[0010] In some embodiments, the cerebral perfusion imaging data includes cerebral perfusion images and cerebral perfusion imaging parameter maps; the clinical data includes basic information, stroke-related information, past medical history, treatment methods, and clinical examination data, and the stroke-related information includes the initial stroke time and the stroke severity; the follow-up data further includes follow-up examination data.

[0011] In some embodiments, before inputting the clinical data into the stroke recurrence risk prediction model, the method further includes: dividing the clinical data into continuous clinical data and discrete clinical data; normalizing the continuous clinical data to obtain normalized continuous clinical data; and numerically encoding the discrete clinical data to obtain numerical discrete clinical data.

[0012] In some embodiments, the method further includes: normalizing the cerebral perfusion images and the follow-up examination data respectively to obtain normalized cerebral perfusion images and follow-up examination data; and before performing normalization on the cerebral perfusion images, performing at least one of denoising and data augmentation on the cerebral perfusion images to preprocess the cerebral perfusion images and obtain preprocessed cerebral perfusion images.

[0013] In a second aspect, the present disclosure provides an apparatus for training a stroke recurrence risk prediction model, including: a processor; and a memory that stores program instructions for training a stroke recurrence risk prediction model, and when the program instructions are executed by the processor, the methods described in the first aspect and its various embodiments are implemented.

[0014] In a third aspect, the present disclosure provides an apparatus for predicting the risk of stroke recurrence, characterized by including: a processor; a memory that stores program instructions for predicting the risk of stroke recurrence, and when the program instructions are executed by the processor, the apparatus is caused to perform the following operations: obtaining cerebral perfusion imaging data, clinical data, and follow-up data of a target patient; inputting the cerebral perfusion imaging data, the clinical data, and the follow-up data into a stroke recurrence risk prediction model trained according to the methods described in the first aspect and its various embodiments for prediction, so as to output the survival risk ratio and whether the stroke recurs for the target patient.

[0015] In a fourth aspect, the present disclosure provides a computer-readable storage medium, on which program instructions for training a stroke recurrence risk prediction model and / or for predicting the risk of stroke recurrence are stored, and when the program instructions are executed by a processor, the operations implemented by the methods described in the first aspect and its various embodiments and / or the apparatus described in the third aspect are implemented.

[0016] As described above, in the training stage of the stroke recurrence risk prediction model, by using cerebral perfusion imaging data, clinical data, and follow-up data as training data, the model can comprehensively understand the patient's condition from multiple perspectives. Cerebral perfusion imaging data provides detailed information on the blood perfusion in the patient's brain, which helps to detect potential vascular lesions and blood flow abnormalities; clinical data covers the patient's basic information, medical history, treatment situation, etc., providing rich background knowledge for the model; follow-up data records the patient's survival status at different time points, reflecting the dynamic change process of the disease. The integration of such multi-modal data can more comprehensively capture the characteristics of the patient and the development law of the disease, thereby improving the prediction accuracy of the model for stroke recurrence risk. In addition, by integrating data on follow-up time and survival status to generate survival time-event indicator pairs, the model can consider the patient's health trend over time. This solves the problem of ignoring time sensitivity in traditional evaluation methods, enabling the model to better understand the development process of the disease and thus more accurately predict the stroke recurrence risk.

[0017] In the application stage of the stroke recurrence risk prediction model, it can effectively integrate the patient's multi-modal data to achieve accurate assessment of the stroke recurrence risk. By analyzing the patient's cerebral perfusion imaging data, clinical data, and follow-up data, the model can identify high-risk individuals who may be overlooked by traditional evaluation methods. This helps doctors to take targeted intervention measures in a timely manner, such as strengthening drug treatment, increasing the follow-up frequency, etc., thereby reducing the patient's recurrence risk. In addition, the survival risk output by the model, such as the probability of stroke recurrence, can be used for risk stratification management of patients. Patients are divided into high, medium, and low risk groups, so as to develop personalized treatment and follow-up plans for patients with different risk levels. This risk stratification management can avoid waste of medical resources, not only improving the utilization efficiency of medical resources, but also ensuring that high-risk patients receive timely attention and treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] By referring to the following detailed description with reference to the accompanying drawings, the above and other objects, features, and advantages of the exemplary embodiments of the present disclosure will become readily understood. In the drawings, several embodiments of the present disclosure are shown in an exemplary rather than restrictive manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0019] Figure 1 Shows an exemplary structural diagram of the stroke recurrence risk prediction model of the embodiment of the present disclosure;

[0020] Figure 2 Shows an exemplary flowchart of the method for training the stroke recurrence risk prediction model of the embodiment of the present disclosure;

[0021] Figure 3An exemplary flowchart showing the process of training a stroke recurrence risk prediction model using training data according to an embodiment of the present disclosure;

[0022] Figure 4 An exemplary structural block diagram of a device for training a stroke recurrence risk prediction model according to an embodiment of the present disclosure;

[0023] Figure 5 An exemplary flowchart of a method for predicting stroke recurrence risk according to an embodiment of the present disclosure. Detailed Description of the Invention

[0024] The technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present disclosure. Obviously, the described embodiments are some, but not all, of the embodiments of the present disclosure. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present disclosure without creative efforts shall fall within the protection scope of the present disclosure.

[0025] It should be understood that the terms "including" and "comprising" used in the specification and claims of the present disclosure indicate the presence of the described features, wholes, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or their combinations.

[0026] It should also be understood that the terms used in the specification of the present disclosure are only for the purpose of describing specific embodiments and are not intended to limit the present disclosure. As used in the specification and claims of the present disclosure, unless the context clearly indicates otherwise, the singular forms "a", "an", and "the" are intended to include the plural forms. It should be further understood that the term "and / or" used in the specification and claims of the present disclosure refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0027] As used in this specification and the claims, the term "if" may be interpreted as "when", "once", "in response to determining", or "in response to detecting" depending on the context. Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted as meaning "once determined", "in response to determining", "once detected [the described condition or event]", or "in response to detecting [the described condition or event]" depending on the context.

[0028] The specific embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings.

[0029] Figure 1An exemplary structural diagram of the stroke recurrence risk prediction model according to an embodiment of the present disclosure is shown. As Figure 1 shown, the stroke recurrence risk prediction model of the present disclosure may include a feature extraction model 101 and a survival analysis model 102.

[0030] Since the stroke recurrence risk prediction model of the present disclosure needs to process multi-modal data composed of cerebral perfusion imaging data, clinical data, and follow-up data, and the information recorded in each item of data is different. Therefore, different feature extraction models can be used to extract the features required by the model from each item of data. In one embodiment, the feature extraction model 101 may further include a first extraction model 1011, a second extraction model 1012, and a third extraction model 1013.

[0031] The aforementioned first extraction model 1011 may use a convolutional neural network backbone network CNN Backbone, such as VGG, Resnet, DenseNet, or EfficientNet, etc., to process the cerebral perfusion imaging data to extract features at different levels and scales from the cerebral perfusion imaging data. The aforementioned second extraction model 1012 may adopt a multi-layer perceptron (MLP) to learn complex features and patterns in the clinical data. MLP is a neural network containing multiple fully connected layers. In order to increase the robustness of the network and prevent overfitting, a ReLU activation function and a Dropout layer may be added between each layer of the network.

[0032] The aforementioned third extraction model 1013 may adopt a Mamba Vision Block, which can effectively process the follow-up data by combining a linear function, convolution, and a sigmoid function, as well as a state space model SSM, capture the dynamic changes in the time series, and be able to output a high-level feature representation that fuses time series and spatial information. Here, the linear function is used to perform a linear transformation on the input data to adjust the dimension and scale of the data. Convolution extracts local features in the time series through convolution operations to capture short-term dependencies in the time series. The sigmoid function is used to limit the output value between 0 and 1, which is suitable for probability prediction tasks. The state space model SSM models the time series data as an evolution process of states through a state transition matrix and an observation matrix.

[0033] In the embodiment of the present disclosure, the output of the third extraction model 1013 may be connected to a fully connected layer, and the fully connected layer can reduce the dimension of the features, reduce the computational complexity, and at the same time retain the most important feature information, so as to map the high-level features to a suitable feature space for subsequent stroke recurrence risk prediction tasks.

[0034] In addition, to achieve the purpose of simultaneously outputting the survival risk ratio and the probability of stroke recurrence, the survival analysis model 102 of the present disclosure may further include a first output layer 1021 and a second output layer 1022. In some implementation scenarios, the first output layer 1021 is used to output the survival risk ratio, and the second output layer 1022 is used to output the probability of stroke recurrence. Here, the survival risk ratio is a relative risk indicator, which represents the multiple of the risk of an event (such as stroke recurrence) for an individual or group relative to a reference individual or group. The higher the survival risk ratio, the higher the risk of stroke recurrence. The probability of stroke recurrence is an absolute probability value, which represents the likelihood of an individual having a stroke recurrence at a preset time point (such as within 30 days, 60 days, 90 days, and 120 days, etc.). It can be understood that those skilled in the art can select the specific value of the preset time point according to actual needs, and the present disclosure does not make specific limitations on this.

[0035] The aforementioned first output layer 1021 can be a fully connected layer. To achieve the output of an interpretable survival risk ratio HR, it is necessary to design such that the output of the fully connected layer directly learns the linear risk factor βX. Here, X is the input feature vector, which can include imaging features, clinical features, and follow-up information, etc. β is the regression coefficient of the model, which represents the contribution of each feature to the stroke recurrence risk of the patient. Next, by taking the exponent of the output of the fully connected layer, the survival risk ratio HR can be obtained, that is, HR = e βX . This design enables the deep learning model to follow the mathematical form of the Cox proportional hazards model, thereby outputting an interpretable survival risk ratio.

[0036] The aforementioned second output layer 1022 can be a fully connected layer and a Sigmoid function. The value range of the fully connected layer is all real numbers, and the Sigmoid function can convert the linear output of the fully connected layer into a probability value, ensuring that the output is between 0 and 1. The output of the Sigmoid function can be directly interpreted as the probability of an event (such as stroke recurrence) occurring, providing intuitive support for clinical decision-making.

[0037] Figure 2 Fig. shows an exemplary flowchart of a method 200 for training a stroke recurrence risk model according to an embodiment of the present disclosure. It can be understood that the method 200 can be executed by any suitable device with data processing capabilities, such as but not limited to terminal devices, processors, and servers, etc.

[0038] Based on this, as Figure 2As shown, at step S201, method 200 can obtain the cerebral perfusion imaging data, clinical data, and follow-up data of the patient. Next, at step S202, method 200 can integrate the follow-up time and the survival status at the follow-up time to obtain at least one survival time-event indication pair. Finally, at step S203, method 200 can input the cerebral perfusion imaging data, clinical data, follow-up data, and at least one survival time-event indication pair as training data into the stroke recurrence risk prediction model for training. To facilitate understanding of how to train the stroke recurrence risk prediction model, it will be described in detail later in conjunction with Figure 3 The process 300 of training the stroke recurrence risk prediction model using the training data will be described in detail, and will not be elaborated here.

[0039] In the embodiments of the present disclosure, the aforementioned cerebral perfusion imaging data may include cerebral perfusion images and cerebral perfusion imaging parameter maps. The cerebral perfusion image refers to performing CT scans on the same region of interest multiple times within a short period to obtain the dynamic imaging data of the region after the injection of a contrast agent. These data reflect the flow and distribution of the contrast agent in the brain tissue. The cerebral perfusion imaging parameter map is an image generated by extracting specific parameters from the cerebral perfusion image, and each parameter map corresponds to a specific hemodynamic parameter, such as cerebral blood flow CBF, cerebral blood volume CBV, mean transit time MTT, and maximum peak time Tmax, etc. The cerebral perfusion image provides dynamic blood flow information, while the cerebral perfusion imaging parameter map provides quantitative blood flow parameters. The combination of the two can more comprehensively evaluate the cerebral blood perfusion situation and provide strong support for clinical diagnosis and treatment decisions. In the embodiments of the present disclosure, the cerebral perfusion images of acute ischemic stroke (AIS) patients can be retrospectively collected, usually within a time window of 24 hours, that is, the image data of AIS patients who received cerebral perfusion imaging within 24 hours after onset is collected.

[0040] The aforementioned clinical data may include the patient's basic information (such as age, gender, and weight), clinical examination data (blood glucose, cholesterol, coagulation function, etc.), past medical history (such as hypertension, diabetes, transient ischemic attack TIA history, etc.), treatment methods (such as thrombolytic therapy TPA, mechanical thrombectomy MT, antiplatelet therapy, vascular intervention, etc.), and stroke-related information. Here, the stroke-related information may include the time of the first stroke and the severity of the stroke, and the severity of the stroke can be divided into low, medium, and high.

[0041] The foregoing follow-up data may include the follow-up time (such as 30 days, 60 days, 90 days, 120 days, etc.) and the survival status at the follow-up time. Here, the follow-up time refers to the difference between the time of the first stroke and the follow-up time point. In particular, if the patient dies before the follow-up, the time of death can be used as the follow-up time point to ensure that the follow-up data can accurately reflect the patient's survival status. The survival status may include whether a stroke recurrence occurs, whether there are sequelae (such as hemiplegia, speech disorder, etc.), whether there are complications (such as bleeding, infection, etc.), and whether the patient has died. In actual operation, a binary representation can be used to indicate the survival status of whether a stroke recurrence occurs and whether the patient has died. For example, 0 is used to indicate no stroke recurrence, and 1 is used to indicate a stroke recurrence. Similarly, 0 is used to indicate survival, and 1 is used to indicate death. Additionally or optionally, if a basic examination or imaging examination is performed during the patient's follow-up, the patient's follow-up data may also include follow-up examination data (such as blood pressure, blood sugar, blood oxygen saturation, etc.) and cerebral perfusion imaging data during the follow-up period.

[0042] At step S202 described above, method 200 may integrate the follow-up time and the survival status at the follow-up time to obtain at least one survival time-event indicator pair. Here, the follow-up time can be used as the survival time, and the event indicator is used to identify whether a preset event has occurred at the survival time. The preset event may be a stroke recurrence and death. In practical applications, the event indicator can be represented by 0 to indicate no preset event, 1 to indicate a stroke recurrence, and 2 to indicate death.

[0043] In actual operation, there may be a situation where a certain piece of data in the collected clinical data is missing, or a follow-up record for a certain day in the follow-up data is missing. In this case, directly discarding the clinical data or follow-up data will result in a waste of data resources. To collect richer data resources and improve the quality of the model, for the missing data items in the clinical data, methods such as filling with the mean, median, or most frequent value can be used to complete them, and for the missing follow-up records in the follow-up data, data interpolation methods can be used to complete them.

[0044] In the embodiments of this disclosure, to improve the training efficiency of the stroke recurrence risk prediction model and save time costs, before using the foregoing cerebral perfusion imaging data, clinical data, and follow-up data for model training, these data can be preprocessed first to obtain preprocessed data for model training. Thus, on the premise of ensuring sufficient information, unnecessary computational amounts can be reduced, computing resources can be utilized more efficiently, and the entire training process can be accelerated.

[0045] Specifically, before inputting clinical data into the stroke recurrence risk prediction model, the clinical data can be first divided into continuous clinical data and discrete clinical data. As an example, age and weight in the basic information, as well as clinical examination data (such as blood glucose, cholesterol, coagulation function, etc.) can be divided into continuous clinical data. Conversely, gender, treatment method, stroke severity, etc. in the basic information can be divided into discrete clinical data. It can be understood that this division is only exemplary and illustrative, and does not limit the present disclosure solution. Based on this example, those skilled in the art can achieve the division of other clinical data.

[0046] After dividing the clinical data into continuous clinical data and discrete clinical data, the continuous clinical data can be normalized, such as normalizing it to the interval [0, 1] or standardizing it to a distribution with a mean of 0 and a variance of 1, to obtain the normalized continuous clinical data. In addition, the discrete clinical data can also be numerically encoded, such as converting the discrete clinical data into one-hot encoding or label encoding, to obtain the numerical discrete clinical data, which is convenient for the model to understand and process.

[0047] Similarly, in some implementation scenarios, before inputting the follow-up data into the stroke recurrence risk prediction model, the follow-up examination data in the follow-up data can also be normalized to obtain the normalized follow-up examination data. In other implementation scenarios, before inputting the cerebral perfusion imaging data into the stroke recurrence risk prediction model, the gray values of the cerebral perfusion imaging can also be normalized so as to normalize its gray values to the interval [0, 1] to eliminate the differences between different scanning devices, scanning conditions, and patients.

[0048] Furthermore, in order to improve the quality of the training data, before normalizing the cerebral perfusion imaging, the noise in the cerebral perfusion imaging can also be removed using a denoising method (such as the denoising autoencoder DAE) to reduce the interference of the noise on the training process and improve the training efficiency. The denoising autoencoder DAE is a deep learning model composed of an encoder and a decoder, which are responsible for data compression and reconstruction respectively, and is mainly used to recover the original data from noisy data.

[0049] In addition, to improve the generalization ability of the stroke recurrence risk prediction model, data augmentation operations such as rotation, flipping, scaling, and adding noise can also be performed on the cerebral perfusion images to generate more cerebral perfusion images under different morphologies and conditions. This helps the stroke recurrence risk prediction model learn more diverse features during training, so that it can better handle various complex cerebral perfusion images in practical applications, improving the adaptability and generalization ability of the model to data collected from different patients and devices. In this way, during actual stroke recurrence risk prediction, the stroke recurrence risk prediction model can work more stably, being less affected or not affected by interference factors such as noise and blurring, thereby improving the reliability of the prediction results.

[0050] The above combination Figure 2 describes a method for training a stroke recurrence risk model. By using cerebral perfusion image data, clinical data, and follow-up data as training data, the model can comprehensively understand the patient's condition from multiple perspectives. Cerebral perfusion image data provides detailed information on the blood perfusion in the patient's brain, helping to detect potential vascular lesions and blood flow abnormalities; clinical data covers the patient's basic information, medical history, treatment conditions, etc., providing rich background knowledge for the model; follow-up data records the survival status of the patient at different time points, reflecting the dynamic change process of the disease. The fusion of such multi-modal data can more comprehensively capture the characteristics of the patient and the development law of the disease, thereby improving the prediction accuracy of the model for stroke recurrence risk. In addition, by integrating the follow-up time and survival status to generate survival time-event indicator pairs, the model can consider the patient's health trend over time. This solves the problem that traditional evaluation methods ignore time sensitivity, enabling the model to better understand the development process of the disease and thus more accurately predict the stroke recurrence risk.

[0051] Figure 3 shows an exemplary flowchart of process 300 for training a stroke recurrence risk prediction model using training data according to an embodiment of the present disclosure. It can be understood that the description below in combination with Figure 3 is a specific implementation of the foregoing step S203. Therefore, the features described above in combination with Figure 2 can be similarly applied here.

[0052] As Figure 3As shown, at step S301, cerebral perfusion imaging data, clinical data, and follow-up data can be respectively input into the first extraction model, the second extraction model, and the third extraction model for feature extraction to obtain the first feature, the second feature, and the third feature. Then, at step S302, the first feature, the second feature, and the third feature can be input into the survival analysis model for survival analysis, so that the first output layer outputs the predicted survival risk ratio, and the second output layer outputs the predicted stroke recurrence probability. Next, at step S303, the loss value can be determined based on the predicted survival risk ratio, the predicted stroke recurrence probability, and at least one survival time-event indicator, and based on the loss value, the parameters of each model and the output layer in the stroke recurrence risk prediction model can be updated through backpropagation.

[0053] In the embodiment of the present disclosure, the loss value determined at step S303 can be the sum of the survival analysis loss and the cross-entropy loss, where the survival analysis loss can evaluate the prediction accuracy of the model for the survival time, and the cross-entropy loss can evaluate the classification accuracy of the model for whether recurrence occurs at the preset time point.

[0054] In actual operation, the survival analysis loss L can be determined using the following formula:

[0055]

[0056] where, βX i is the output result of the first output layer, β is the regression coefficient of the model, X i is the covariate of the i-th individual, represents the survival risk ratio, represents the sum of the survival risk ratios of all patients in R i where the risk set R i refers to the set of all individuals with a survival time at least as long as that of the i-th individual when the event occurs to the i-th individual. ∑ i∈enents represents the sum over all individuals who have experienced an event (such as stroke recurrence).

[0057] In actual operation, the classification of whether recurrence occurs at the preset time point can be obtained by converting the aforementioned predicted stroke recurrence probability into a binary classification result. In one embodiment, a fixed threshold (such as 0.5) can be selected. Then, if the predicted stroke recurrence probability is greater than or equal to 0.5, it is determined that the classification of whether recurrence occurs at the preset time point is recurrence (labeled as 1), otherwise, it is determined that the classification of whether recurrence occurs at the preset time point is non-recurrence (labeled as 0).

[0058] After completing the training of the stroke recurrence risk prediction model, the performance of the trained stroke recurrence risk prediction model can be tested using test data first. When it is determined that its performance meets the actual needs, the model can be used for actual stroke recurrence risk prediction tasks. In actual operation, the preparation method of test data is the same as that of the aforementioned training data, and the preparation of test data can be referred to the description above. For the sake of brevity, the preparation process of test data will not be elaborated in this disclosure. In one embodiment, the aforementioned steps S201 and S202 can be used to prepare data, and the prepared data can be divided into training data and test data according to a preset ratio. The specific value of the preset ratio can be selected according to actual needs, such as 8:3, 7:3, etc., and this disclosure does not make specific limitations on this.

[0059] In the embodiments of this disclosure, the performance of the trained stroke recurrence risk prediction model can be evaluated by calculating indicators such as C-index value, calibration curve, ROC-AUC value, F1-Score, etc. Calculating the C-index value can measure the ranking consistency between the predicted survival data and the actual survival data. The value range of the C-index value is between 0 and 1. The higher the C-index value, the more consistent the order of the survival risks predicted by the model is with the order of the actual observed survival results. For example, if the model predicts that the survival risk ratio of patient A is higher than that of patient B, and in fact patient A does experience the preset event earlier than patient B, then the consistency of this pair of patients is reflected.

[0060] In addition, the calibration curve is a tool for evaluating the agreement between the predicted probability of the model and the actual occurrence probability. Through the calibration curve, we can intuitively see whether the predicted stroke recurrence probability of the model accurately reflects the probability of the actual event occurring. ROC-AUC (Receiver Operating Characteristic-Area Under Curve) evaluates the classification performance of the model by calculating the area under the ROC curve. The AUC value ranges between 0.5 and 1, and the larger the value, the better the performance of the model. F1-Score is the harmonic mean of precision and recall, which is used to comprehensively consider the performance of the model in identifying and predicting the positive class. The higher the F1-Score, the better the comprehensive performance of the model. In actual applications, ROC-AUC can reflect the ability of the model to distinguish between positive and negative examples, while F1-Score finds a balance between precision and recall and is suitable for dealing with imbalanced datasets. Through these two indicators, we can more accurately evaluate the reliability and accuracy of the model in actual applications.

[0061] As described above in combination with Figure 2and Figure 3 describes a method for training a stroke recurrence risk prediction model. Accordingly, the present disclosure also provides a device 400 for training a stroke recurrence risk prediction model. Next, with reference to Figure 4 an exemplary introduction to the device 400 provided by the embodiments of the present application will be given. As Figure 4 shown, the electronic device 400 of the embodiments of the present application may include a processor 401, a memory 402, and a communication bus 403.

[0062] In the process of specific embodiments, the above-mentioned processor 401 may be at least one of an application specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing image processing device (DSPD), a programmable logic image processing device (PLD), a field programmable gate array (FPGA), a CPU, a controller, a microcontroller, and a microprocessor. It can be understood that for different devices, the electronic devices for implementing the above-mentioned processor functions may also be others, and specific limitations are not made in this embodiment.

[0063] In the embodiments of the present application, the above-mentioned communication bus 403 is used to implement the connection and communication between the processor 401 and the memory 402; the memory 402 stores program instructions for training a stroke recurrence risk prediction model; when the above-mentioned processor 401 executes the program instructions stored in the memory 402, the method for training a stroke recurrence risk prediction model described in the present application in combination with Figure 2 and Figure 3 is implemented.

[0064] The above combination with Figure 4 describes the electronic device that can be used to execute the stroke recurrence risk prediction model training of the present application. It should be understood that the device structure or architecture here is only exemplary, and the implementation manner and implementation entity of the present application are not limited by it, but can be changed without departing from the spirit of the present application. It can be understood that the present disclosure emphasizes the differences between the various embodiments, and the same or corresponding parts can be referred to each other. For the sake of brevity, the present disclosure will not elaborate one by one.

[0065] In addition, the present disclosure also provides a device for predicting the risk of stroke recurrence. The structure or architecture of this device is the same as that described above in combination withFigure 4 The structure or framework of the device 400 described for training the stroke recurrence risk prediction model is the same. For the sake of brevity, the present disclosure will not elaborate on each one. In the device for predicting the stroke recurrence risk, a memory stores program instructions for predicting the stroke recurrence risk. When the program instructions are executed by a processor, the device can implement the following combination Figure 5 The method 500 for predicting the stroke recurrence risk described.

[0066] Figure 5 An exemplary flowchart of the method 500 for predicting the stroke recurrence risk according to an embodiment of the present disclosure is shown. It can be understood that the method 500 can be executed by any suitable device with data processing capabilities, such as, for example, but not limited to, a processor, a terminal device, and a server, etc.

[0067] As Figure 5 Shown, at step S501, the method 500 can obtain the cerebral perfusion imaging data, clinical data, and follow-up data of the target patient. Then, at step S502, the method 500 can input the cerebral perfusion imaging data, clinical data, and follow-up data into the trained stroke recurrence risk prediction model for prediction to output the survival risk ratio and whether the stroke recurs for the target patient. It can be understood that the trained stroke recurrence risk prediction model here refers to the model trained by using the method for training the stroke recurrence risk prediction model described in combination with Figure 2 And Figure 3 The method described.

[0068] As can be known from the foregoing description, the second output layer of the trained stroke recurrence risk prediction model can output the stroke recurrence probability. On this basis, the stroke recurrence probability can be compared with a preset threshold (such as 0.5) to determine whether the stroke recurs. Specifically, in the case where the stroke recurrence probability is greater than or equal to 0.5, it can be determined that the stroke will recur; otherwise, it can be determined that the stroke will not recur.

[0069] In the embodiment of the present disclosure, the target patient can be any acute ischemic stroke (AIS) patient. Before inputting the cerebral perfusion imaging data, clinical data, and follow-up data of the target patient into the trained stroke recurrence risk prediction model for prediction, these data can be preprocessed first to improve the accuracy of the prediction result. It can be understood that here, the preprocessing method for the cerebral perfusion imaging data, clinical data, and follow-up data of the target patient is the same as the preprocessing method for the cerebral perfusion imaging data, clinical data, and follow-up data used as training data described above and can be referred to each other. For the sake of brevity, the present disclosure will not elaborate on each one.

[0070] The above combinationFigure 5 A method for predicting the risk of stroke recurrence is described, which can effectively integrate multi-modal data of patients and achieve accurate assessment of the risk of stroke recurrence. By analyzing the cerebral perfusion imaging data, clinical data, and follow-up data of patients, the model can identify high-risk individuals who may be overlooked by traditional assessment methods. For example, although a patient has a low NIHSS score, their cerebral perfusion imaging shows local blood perfusion abnormalities, and the follow-up data indicates poor blood pressure control. The model can integrate this information and identify the patient as a high-risk patient. This helps doctors take targeted intervention measures in a timely manner, such as strengthening drug treatment and increasing the frequency of follow-up, thereby reducing the recurrence risk of patients. In addition, the survival risk ratio and the probability of stroke recurrence output by the model can be used for risk stratification management of patients. The patients are divided into high, medium, and low-risk groups, so as to develop personalized treatment and follow-up plans for patients with different risk levels. This risk stratification management can avoid waste of medical resources, not only improve the utilization efficiency of medical resources, but also ensure that high-risk patients receive timely attention and treatment.

[0071] According to the above description with reference to the accompanying drawings, those skilled in the art can also understand that the embodiments of the present application can also be implemented by software programs. Therefore, the present application also provides a computer-readable storage medium. The computer-readable storage medium stores program instructions for training a stroke recurrence risk prediction model and / or for predicting the risk of stroke recurrence, and the program instructions can be used to implement the method for training a stroke recurrence risk prediction model described in the present application in combination with Figure 2 and Figure 3 or the method for predicting the risk of stroke recurrence described in the present application in combination with Figure 5 .

[0072] It should be noted that although the operations of the method of the present application are described in a specific order in the drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. On the contrary, the order of the steps depicted in the flowchart can be changed. Additionally or alternatively, some steps can be omitted, multiple steps can be combined into one step for execution, and / or one step can be decomposed into multiple steps for execution.

[0073] Although multiple embodiments of the present application have been shown and described herein, it is obvious to those skilled in the art that such embodiments are provided by way of example only. Many variations, changes, and alternative ways can be thought of by those skilled in the art without departing from the spirit and scope of the present application. It should be understood that various alternative embodiments of the present application described herein can be adopted in the practice of the present application. The appended claims are intended to define the scope of protection of the present application and thus cover equivalents or alternatives within the scope of these claims.

[0074] In this application, the collection and acquisition of various data comply with relevant laws and regulations and are authorized by the data providers. Any organization or individual that needs to obtain external data shall obtain authorization in accordance with the law and ensure data security, and shall not illegally collect, use, process, or transmit unauthorized or unprotected data, nor illegally buy, sell, provide, or disclose unauthorized or unprotected data.

Claims

1. A method for training a stroke recurrence risk prediction model, the method comprising: Obtaining cerebral perfusion imaging data, clinical data, and follow-up data of a patient; The follow-up data includes follow-up time and the survival status at the follow-up time; the survival status includes whether a stroke recurrence has occurred, whether there are sequelae, whether there are complications, and whether death has occurred; Integrating the follow-up time and the survival status at the follow-up time to obtain at least one survival time-event indicator pair, where the event indicator is used to identify whether a preset event has occurred at the survival time, and the preset event is stroke recurrence; Inputting the cerebral perfusion imaging data, the clinical data, the follow-up data, and the at least one survival time-event indicator pair as training data into the stroke recurrence risk prediction model to train it.

2. The method according to claim 1, wherein The stroke recurrence risk prediction model includes a feature extraction model and a survival analysis model; the feature extraction model includes a first extraction model, a second extraction model, and a third extraction model; the survival analysis model includes a first output layer and a second output layer.

3. The method according to claim 2, wherein Inputting the cerebral perfusion imaging data, the clinical data, the follow-up data, and the at least one survival time-event indicator pair as training data into the stroke recurrence risk prediction model to train it includes: Inputting the cerebral perfusion imaging data, the clinical data, and the follow-up data into the first extraction model, the second extraction model, and the third extraction model respectively for feature extraction to obtain a first feature, a second feature, and a third feature; Inputting the first feature, the second feature, and the third feature into the survival analysis model for survival analysis, so that the first output layer outputs a predicted survival risk ratio, and the second output layer outputs a predicted stroke recurrence probability; Determining a loss value based on the predicted survival risk ratio, the predicted stroke recurrence probability, and the at least one survival time-event indicator pair, and updating the parameters of the stroke recurrence risk prediction model based on the loss value.

4. The method according to claim 3, wherein, The first output layer is a fully connected layer, and the second output layer is a fully connected layer and a Sigmoid function; And the first output layer outputs a predicted survival risk ratio including: The fully connected layer outputs a risk score, and the risk score is exponentiated to obtain the predicted survival risk ratio.

5. The method according to any one of claims 1-4, wherein, The cerebral perfusion imaging data includes cerebral perfusion images and cerebral perfusion image parameter maps; the clinical data includes basic information, stroke-related information, past medical history, treatment methods, and clinical examination data, and the stroke-related information includes the initial stroke time and the stroke severity; the follow-up data also includes follow-up examination data.

6. The method according to claim 5, wherein, Before inputting the clinical data into the stroke recurrence risk prediction model, the method further includes: Dividing the clinical data into continuous clinical data and discrete clinical data; Normalizing the continuous clinical data to obtain normalized continuous clinical data; Numerically encoding the discrete clinical data to obtain numerically encoded discrete clinical data.

7. The method according to claim 6 further comprises: Normalizing the cerebral perfusion image and the follow-up examination data respectively to obtain a normalized cerebral perfusion image and follow-up examination data; And Before normalizing the cerebral perfusion image, performing at least one of denoising and data augmentation on the cerebral perfusion image to preprocess the cerebral perfusion image and obtain a preprocessed cerebral perfusion image.

8. A device for training a stroke recurrence risk prediction model, comprising: A processor; And A memory storing program instructions for training a stroke recurrence risk prediction model, which when executed by the processor enables the implementation of the method according to any one of claims 1-7.

9. A device for predicting the risk of stroke recurrence, characterized in that, Comprising: A processor; A memory storing program instructions for predicting the risk of stroke recurrence, which when executed by the processor enables the device to perform the following operations: Obtaining cerebral perfusion image data, clinical data, and follow-up data of a target patient; Inputting the cerebral perfusion image data, the clinical data, and the follow-up data into a stroke recurrence risk prediction model trained by the method according to any one of claims 1-7 for prediction to output the survival risk ratio and whether stroke recurs for the target patient.

10. A computer-readable storage medium storing program instructions for training a stroke recurrence risk prediction model and / or for predicting the risk of stroke recurrence, which when executed by a processor implements the method according to any one of claims 1-7 and / or the operations implemented by the device according to claim 9.

Images