Method for training thrombolysis adaptability evaluation model and related product

Abstract

The invention discloses a method for training a thrombolysis adaptability evaluation model and a related product. The method comprises the following steps: acquiring brain perfusion image data, a brain perfusion parameter map, clinical data and treatment data; performing region-of-interest extraction on the cerebral perfusion parameter map to obtain an attention map containing a core infarction region and an ischemic penumbra sensing region; multiplying the brain perfusion image data by the attention map to obtain weighted brain perfusion image data; the first labeling information and the second labeling information are used for labeling whether thrombolysis treatment is carried out or not and labeling the effect grade after thrombolysis treatment is carried out; and inputting the weighted brain perfusion image data, the brain perfusion parameter diagram, the clinical data, the first annotation information and the second annotation information into a thrombolysis adaptability evaluation model as training data so as to train the thrombolysis adaptability evaluation model. By utilizing the scheme disclosed by the invention, the thrombolysis adaptability and treatment effect prediction accuracy can be improved, and a more reliable basis is provided for clinical decision making.

Description

Technical Field

[0001] The present disclosure generally relates to the field of computer vision technology. More specifically, the present disclosure relates to a method, apparatus, and computer-readable storage medium for training a thrombolytic adaptability assessment model. Furthermore, the present disclosure also relates to an apparatus and computer-readable storage medium for assessing thrombolytic adaptability. Background Art

[0002] Acute stroke is a serious threat to human health, and thrombolytic therapy is currently an important treatment option. However, thrombolytic therapy is not suitable for all patients, and incorrect thrombolytic decisions can lead to serious complications, such as hemorrhagic transformation. Traditional assessments of thrombolytic suitability rely primarily on physicians' clinical experience and limited clinical indicators, such as the National Institutes of Health Stroke Scale (NIHSS) score. However, these methods are subject to significant subjectivity and limitations, and assessment results can vary significantly between physicians, making it difficult to accurately determine a patient's suitability for thrombolytic therapy, thereby impacting treatment efficacy and patient prognosis.

[0003] With the development of medical imaging technologies, such as CT perfusion imaging (CTP), detailed information about cerebral blood flow is now available, providing a more comprehensive basis for evaluating thrombolytic therapy. Currently, some deep learning methods based on CTP and clinical features have achieved some success in assessing the suitability of thrombolysis in stroke patients. However, the evaluation results of these existing deep learning methods are not accurate enough, and the model training process does not consider how to integrate thrombolytic suitability assessment with treatment effect prediction. Therefore, further prediction of treatment effect is impossible, resulting in the lack of an end-to-end solution.

[0004] In view of this, there is an urgent need to provide a scheme for training the thrombolytic adaptability assessment model so that the trained thrombolytic adaptability assessment model can accurately evaluate the patient's thrombolytic adaptability and treatment effect, and provide clinicians with a more scientific and objective decision-making basis. Summary of the Invention

[0005] In order to at least solve one or more of the technical problems mentioned above, the present disclosure proposes a solution for evaluating thrombolytic adaptability in the following aspects.

[0006] In a first aspect, the present disclosure provides a method for training a thrombolytic adaptability assessment model, the method comprising: acquiring a patient's cerebral perfusion image data, a cerebral perfusion parameter map, clinical data, and treatment data; performing region of interest extraction on the cerebral perfusion parameter map to obtain an attention map including a core infarct area and an ischemic penumbra sensing area; multiplying the cerebral perfusion image data by the attention map to obtain weighted cerebral perfusion image data; based on the treatment data, using first annotation information and second annotation information to respectively label whether thrombolytic treatment has been performed and the effect grade after thrombolytic treatment; and inputting the weighted cerebral perfusion image data, the cerebral perfusion parameter map, the clinical data, the first annotation information, and the second annotation information as training data into the thrombolytic adaptability assessment model to train it.

[0007] In some embodiments, the thrombolysis adaptability assessment model includes a feature extraction model and a thrombolysis adaptability analysis model; the feature extraction model includes a first extraction model, a second extraction model, and a third extraction model; and the weighted cerebral perfusion image data, the cerebral perfusion parameter map, the clinical data, and the standard information are input as training data into the thrombolysis adaptability assessment model to train it, including: inputting the weighted cerebral perfusion image data, the cerebral perfusion parameter map, and the clinical data into the first extraction model, the second extraction model, and the third extraction model respectively for feature extraction to obtain a first feature map, a second feature map, and a third feature; inputting the first feature map, the second feature map, and the third feature into the thrombolysis adaptability analysis model for thrombolysis adaptability analysis to output a first evaluation result and a second evaluation result, wherein the first evaluation result is the probability of being suitable for thrombolytic therapy, and the second evaluation result is the probability of each effect grade after thrombolytic therapy; determining a loss value based on the first annotation information, the second annotation information, the first evaluation result, and the second evaluation result, and updating the parameters of the thrombolysis adaptability assessment model based on the loss value.

[0008] In some embodiments, the thrombolysis adaptability analysis model includes a 1*1 convolution block and at least one fully connected layer; and the first feature map, the second feature map and the third feature are input into the thrombolysis adaptability analysis model for thrombolysis adaptability analysis to output a first evaluation result and a second evaluation result, including: inputting the first feature map, the second feature map and the third feature into the 1*1 convolution block for fusion to obtain the fused features; inputting the fused features into the at least one fully connected layer for evaluation to obtain the first evaluation result and the second evaluation result.

[0009] In some embodiments, inputting the fused features into the at least one fully connected layer for evaluation to obtain the first evaluation result and the second evaluation result includes: inputting the fused features into the at least one fully connected layer for linear transformation to output the linearly transformed features; based on the linearly transformed features, using a first activation function to obtain the first evaluation result; based on the linearly transformed features, using a second activation function to obtain the second evaluation result.

[0010] In some embodiments, determining the loss value based on the first labeling information, the second labeling information, the first evaluation result and the second evaluation result includes: determining a first loss value using a first loss function based on the first labeling information and the first evaluation result; determining a second loss value using a second loss function based on the second labeling information and the second evaluation result; and determining the weighted sum of the first loss value and the second loss value as the loss value.

[0011] In some embodiments, the clinical data include age, gender, blood pressure, blood sugar, blood oxygen saturation, timing of thrombolytic therapy and stroke scale score, and the treatment data include whether thrombolytic therapy was performed and the stroke scale score after thrombolytic therapy.

[0012] In some embodiments, after obtaining the patient's cerebral perfusion image data, cerebral perfusion parameter map and clinical data, the method further includes: performing noise reduction and normalization on the cerebral perfusion image data and the cerebral perfusion parameter map to standardize the cerebral perfusion image data and the cerebral perfusion parameter map to obtain standardized cerebral perfusion image data and cerebral perfusion parameter map; normalizing the quantitative data or semi-quantitative data in the clinical data respectively to standardize the clinical data to obtain standardized clinical data.

[0013] In a second aspect, the present disclosure provides a device for training a thrombolysis adaptability assessment model, comprising: a processor; and a memory storing program instructions for training a thrombolysis adaptability assessment model, wherein when the program instructions are executed by the processor, the method described in the aforementioned first aspect and its multiple embodiments are implemented.

[0014] In a third aspect, the present disclosure provides a device for evaluating thrombolytic adaptability, characterized in that it includes: a processor; a memory storing program instructions for evaluating thrombolytic adaptability, and when the program instructions are executed by the processor, the device performs the following operations: obtaining brain perfusion image data, brain perfusion parameter map and clinical data of a target patient; performing region of interest extraction on the brain perfusion parameter map to obtain an attention map including a core infarct area and an ischemic penumbra sensing area; multiplying the brain perfusion image data with the attention map to obtain weighted brain perfusion image data; inputting the weighted brain perfusion image data, brain perfusion parameter map and clinical data into a thrombolytic adaptability evaluation model trained according to the method described in the first aspect and multiple embodiments thereof for evaluation to output a first evaluation result and a second evaluation result, wherein the first evaluation result indicates whether the target patient is suitable for thrombolytic treatment, and the second evaluation result indicates the effect grade of the target patient after thrombolytic treatment.

[0015] In a fourth aspect, the present disclosure provides a computer-readable storage medium having stored thereon program instructions for training a thrombolytic adaptability assessment model and / or for assessing thrombolytic adaptability, wherein when the program instructions are executed by a processor, the operations implemented by the method described in the first aspect and its multiple embodiments and / or the device described in the third aspect are implemented.

[0016] The proposed scheme for assessing thrombolytic adaptability, as described above, utilizes a comprehensive and rich source of information during the training phase of the thrombolytic adaptability assessment model by acquiring patient perfusion imaging data, perfusion parameter maps, clinical data, and treatment data. This data allows the model to integrate multiple factors in its judgment. Regions of interest (ROIs) are extracted from the perfusion parameter maps to generate attention maps, highlighting the core infarct region and the ischemic penumbra sensing region. This attention map is then multiplied with the perfusion imaging data to generate weighted perfusion imaging data. This process enhances the model's focus on key regions and helps capture features with greater diagnostic value. Treatment data are annotated to clarify whether thrombolytic therapy was initiated and the grading of treatment efficacy, providing accurate supervision for model training. Inputting this data as training data allows the model to learn the complex relationships between various factors and thrombolytic adaptability and treatment efficacy, thereby improving the model's accuracy in predicting patient adaptability and treatment efficacy, providing a more reliable basis for clinical decision-making.

[0017] During the application phase of the thrombolytic adaptability assessment model, the device used to assess thrombolytic adaptability can efficiently and accurately assess the target patient with the help of the trained thrombolytic adaptability assessment model. By acquiring the target patient's cerebral perfusion imaging data, cerebral perfusion parameter maps, and clinical data and inputting them into the trained model, the device can quickly output a first assessment result on whether the patient is suitable for thrombolytic therapy and a second assessment result on the grading of the effect after thrombolytic therapy. This not only greatly shortens the assessment time and improves the assessment efficiency, but also avoids the subjectivity and uncertainty of human assessment. Based on a model trained with a large amount of data, the assessment results are more objective and accurate, which can provide strong support for clinical practice. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The above and other objects, features and advantages of the exemplary embodiments of the present disclosure will become readily understood by reading the detailed description below with reference to the accompanying drawings. In the accompanying drawings, several embodiments of the present disclosure are shown in an illustrative and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0019] Figure 1 An exemplary structural diagram of a thrombolysis adaptability evaluation model according to an embodiment of the present disclosure is shown;

[0020] Figure 2 An exemplary flow chart of a method for training a thrombolysis adaptability assessment model according to an embodiment of the present disclosure is shown;

[0021] Figure 3 An exemplary flowchart of a process of training a thrombolysis adaptability evaluation model using training data according to an embodiment of the present disclosure is shown;

[0022] Figure 4 An exemplary structural block diagram of a device for training a thrombolysis adaptability assessment model according to an embodiment of the present disclosure is shown;

[0023] Figure 5 An exemplary flow chart of a method for evaluating thrombolysis suitability according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0024] The following will be combined with the accompanying drawings in the embodiments of the present disclosure to clearly and completely describe the technical solutions in the embodiments of the present disclosure. Obviously, the embodiments described are part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments of the present disclosure, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present disclosure.

[0025] It should be understood that the terms “include” and “comprising” used in the specification and claims of the present disclosure indicate the presence of described features, integers, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.

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

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

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

[0029] Figure 1 FIG. 4 shows an exemplary structural diagram of the thrombolysis adaptability evaluation model according to the embodiment of the present disclosure. Figure 1 As shown, the thrombolysis adaptability evaluation model disclosed herein may include a feature extraction model 101 and a thrombolysis adaptability analysis model 102 .

[0030] Given that the disclosed thrombolysis adaptability assessment model needs to process multimodal data consisting of cerebral perfusion imaging data, cerebral perfusion parameter maps, and clinical data, and that each data record contains different information, different feature extraction models can be used to extract the features required by the model from each data item. In one embodiment, the feature extraction model 101 can 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 and the second extraction model 1012 can respectively use convolutional neural networks, such as any one of VGG, Resnet, U-Net, DenseNet or EfficientNet, to process the image data input therein, such as weighted cerebral perfusion image data or cerebral perfusion parameter maps, to extract features of different levels and scales from the image data. The aforementioned third extraction model 1013 can use a multi-layer perceptron (MLP) to learn quantitative data or semi-quantitative data in clinical data for high-dimensional mapping, obtain clinical features, and enhance the expressive power of features. MLP is a neural network containing multiple fully connected layers. In order to increase the robustness of the network and prevent overfitting, ReLU activation functions and Dropout layers can be added between each layer of the network.

[0032] The aforementioned thrombolytic adaptability analysis model 102 may further include a 1x1 convolutional block 1021 and at least one fully connected layer 1022. The 1x1 convolutional block 1021 is configured to fuse multiple features input therein to obtain a fused feature. The at least one fully connected layer 1022 is configured to perform an evaluation based on the fused feature input therein to obtain a first evaluation result and a second evaluation result. The first evaluation result is the probability of suitability for thrombolytic therapy, and the second evaluation result is the probability of each effect grade after thrombolytic therapy.

[0033] In the embodiment of the present disclosure, the aforementioned 1*1 convolution block 1021 can be composed of a 1*1 convolution, a batch normalization layer (BatchNormalization, BN) and an activation function (Rectified Linear Unit, ReLu). The convolution kernel size used by the 1*1 convolution is 1*1, that is, in each convolution operation, the convolution kernel only operates with one pixel on the input feature map and all its corresponding channels. By using a 1*1 convolution kernel to perform a convolution operation on the feature map, the high-dimensional feature map can be mapped to a low-dimensional space, the number of channels can be reduced, and redundant information can be removed, thereby reducing the complexity of the model. The BN layer can normalize the input data, making the distribution of the data more stable, which helps to accelerate the convergence speed of the model. At the same time, the BN layer also has a certain regularization effect, which can improve the generalization ability of the model and reduce the risk of overfitting. The ReLU (Rectified Linear Unit) activation function introduces a nonlinear factor to the model. After 1×1 convolution and batch normalization, the ReLU function sets values ​​less than 0 to 0 and keeps values ​​greater than 0 unchanged, thereby increasing the model's nonlinear expression capabilities. This enables the model to learn more complex feature patterns and improve the model's performance in tasks such as classification and regression.

[0034] The at least one fully connected layer 1022 can perform a linear transformation on the fused features input therein to output the linearly transformed features. In practical applications, more fully connected layers can perform more detailed feature extraction and transformation on the data, enabling the model to learn more advanced and abstract feature representations, helping the model learn more complex feature relationships, thereby improving the model's prediction accuracy. However, at the same time, too many fully connected layers may also lead to problems such as model overfitting, extended training time, and increased computing resource consumption. Therefore, those skilled in the art can select the appropriate specific number of fully connected layers through experiments and tuning to achieve a balance between model performance and efficiency.

[0035] Next, based on the linearly transformed features, a first activation function, such as the Sigmoid activation function, is used to obtain the aforementioned first evaluation result. The Sigmoid function is used to perform binary classification based on the linearly transformed features of the fully connected layer. Specifically, it converts the linearly transformed features of the fully connected layer into probability values, ensuring that the output is between 0 and 1. The output of the Sigmoid function can be directly interpreted as the probability of being suitable for thrombolytic therapy, providing intuitive support for clinical decision-making.

[0036] Furthermore, based on the linearly transformed features, a second activation function, such as the Softmax activation function, can be used to obtain the aforementioned second evaluation result. The Softmax activation function is used for multi-classification based on the linearly transformed features of the fully connected layer. Specifically, the linearly transformed features of the fully connected layer can be converted into the probability of each effect grade after thrombolytic therapy.

[0037] In the disclosed embodiments, the effect grading after thrombolytic therapy can include mild, moderate, severe and extremely severe. Specifically, the NIHSS score corresponding to mild is 0-5 points. This grading represents a mild neurological deficit, a good prognosis, and most functions can usually be restored. The NIHSS score corresponding to moderate is 6-13 points. This grading represents a more obvious neurological deficit, which may require a longer period of rehabilitation and a moderate prognosis. The NIHSS score corresponding to severe is 14-20 points. This grading represents a serious neurological deficit, a poor prognosis, and may require long-term care. The NIHSS score corresponding to extremely severe is 21-42 points. This grading represents an extremely serious neurological deficit, a very poor prognosis, and is often accompanied by a high mortality rate and severe disability.

[0038] In actual operation, different effect levels can be identified using numbers, letters, symbols, or a combination thereof. For example, the letters A, B, C, and D can be used to identify mild, moderate, severe, and very severe in the effect level, respectively. Alternatively, Roman numerals I, II, III, and IV can be used to identify mild, moderate, severe, and very severe in the effect level, respectively. In other words, the embodiments of the present disclosure do not limit the specific symbolic form for identifying different effect levels, as long as the different effect levels can be distinguished.

[0039] Figure 2 An exemplary flow chart of a method 200 for training a thrombolysis adaptability assessment model according to an embodiment of the present disclosure is shown. It is understood that the method 200 can be executed by any appropriate device with data processing capabilities, including but not limited to a terminal device, a processor, and a server.

[0040] Based on this, Figure 2 As shown, at step S201, method 200 can obtain the patient's cerebral perfusion image data, cerebral perfusion parameter map, clinical data and treatment data. Then, at step S202, method 200 can extract the region of interest on the cerebral perfusion parameter map to obtain an attention map containing the core infarct area and the ischemic penumbra sensing area. At step S203, method 200 can multiply the cerebral perfusion image data with the attention map to obtain weighted cerebral perfusion image data. Next, at step S204, method 200 can use the first annotation information and the second annotation information to respectively mark whether thrombolytic therapy has been performed and the effect grade after thrombolytic therapy based on the treatment data.

[0041] Finally, at step S205, the method 200 can input the weighted cerebral perfusion image data, the cerebral perfusion parameter map, the clinical data, the first annotation information and the second annotation information as training data into the thrombolysis adaptability assessment model to train it. Figure 3 The process 300 of training the stroke thrombolysis adaptability assessment model using the training data is described in detail and will not be further described here.

[0042] In the disclosed embodiments, the aforementioned clinical data may include age, gender, blood pressure, blood sugar, blood oxygen saturation, timing of thrombolytic therapy, and stroke scale score before thrombolytic therapy. The aforementioned treatment data may include whether thrombolytic therapy was performed and the stroke scale score after thrombolytic therapy.

[0043] In the disclosed embodiments, cerebral perfusion imaging data and cerebral perfusion parameter maps of patients with acute ischemic stroke (AIS) can be retrospectively collected. The aforementioned cerebral perfusion imaging data refers to multiple CT scans of the same area of ​​interest within a short period of time to obtain dynamic imaging data of the area after contrast agent injection. These data reflect the flow and distribution of the contrast agent in the brain tissue. The cerebral perfusion parameter map is an image generated by extracting specific parameters from the cerebral perfusion image. 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. The cerebral perfusion imaging data provides dynamic blood flow information, while the cerebral perfusion 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.

[0044] In step S202, the attention map is used to highlight the lesion area. Therefore, in the attention map containing the core infarct area and the ischemic penumbra sensing area, different values ​​are used to distinguish different areas. Specifically, the core infarct area can be assigned a value of 3, the ischemic penumbra sensing area can be assigned a value of 2, and other areas can be assigned a value of 1. In actual operation, the size of the attention map remains consistent with the brain perfusion image data and is strictly aligned in space. Using spatially aligned attention maps and brain perfusion image data ensures that the multiplication operation in step S203 ensures that the attention map accurately highlights the information at the corresponding location in the brain perfusion image data.

[0045] In practice, the core infarct area and the ischemic penumbra sensing area can be segmented by setting appropriate thresholds. For example, when the CBV is below a certain threshold (e.g., 30% of normal brain tissue CBV) and the MTT exceeds a certain threshold (e.g., 150% of normal brain tissue MTT), the area can be identified as the core infarct area. When the CBF is below 30%-50% of normal brain tissue and the MTT exceeds 150%-200% of normal brain tissue, the area can be identified as the ischemic penumbra sensing area.

[0046] Furthermore, the segmented core infarct or ischemic penumbra sensing area can be subjected to connected region analysis to remove isolated small regions and retain only the largest connected region as the core infarct or ischemic penumbra sensing area. Furthermore, morphological operations (such as dilation and erosion) can be used to smooth the boundaries of the core infarct or ischemic penumbra sensing area, making the extracted region more consistent with the actual physiological structure. In step S204, a record of whether thrombolytic therapy was performed can be obtained from the treatment data, and the first annotation information can be used to indicate whether thrombolytic therapy was performed. For example, if thrombolytic therapy was performed, the first annotation information can be set to 1; otherwise, the first annotation information can be set to 0. Then, a record of the stroke scale score after thrombolytic therapy can be obtained from the treatment data, and the stroke scale score after thrombolytic therapy can be annotated using the second annotation information. For example, if the stroke scale score is 3, the corresponding effect grade is mild, and the second annotation information can be A; for another example, if the stroke scale score is 40, the corresponding effect grade is extremely severe, and the second annotation information can be D.

[0047] In the disclosed embodiments, to improve the training efficiency of the thrombolytic adaptability assessment model and save time and cost, the aforementioned cerebral perfusion imaging data, clinical data, and follow-up data can be preprocessed before being used for model training to obtain preprocessed data for model training. This can reduce unnecessary computational effort while ensuring sufficient information, enabling more efficient use of computing resources and accelerating the entire training process.

[0048] Specifically, after obtaining the patient's clinical data, the quantitative data or semi-quantitative data in the clinical data can be normalized separately, and the quantitative data or semi-quantitative data can be normalized to between 0 and 1, so as to realize the standardized processing of the clinical data and obtain the standardized clinical data. In the disclosed embodiment, quantitative data refers to continuous data with clear numerical values, has units, and can be accurately measured. For example, age, blood pressure, blood sugar, blood oxygen saturation, timing of thrombolytic therapy, etc. in clinical data. Semi-quantitative data refers to hierarchical data between qualitative and quantitative, which can be expressed by discrete scores, such as gender, stroke scale score, effect grade, etc. in clinical data.

[0049] After obtaining the patient's cerebral perfusion image data and cerebral perfusion parameter map, the cerebral perfusion image data and the cerebral perfusion parameter map can be normalized using methods such as minimum-maximum normalization, Z-score normalization, or normalization based on a specific threshold value, so as to perform standardization on the cerebral perfusion image data and the cerebral perfusion parameter map to obtain the standardized cerebral perfusion image data and the cerebral perfusion parameter map. Normalization of cerebral perfusion image data and cerebral perfusion parameter maps can eliminate dimensional differences between data, reduce data fluctuations, make data comparable, and improve the stability and effect of model training. In some implementation scenarios, spatial domain filtering, frequency domain filtering, or denoising methods based on deep learning can also be used to denoise the cerebral perfusion image data and the cerebral perfusion parameter map, reduce the interference of noise on the training process, improve image quality, and provide more accurate information for subsequent thrombolysis adaptability analysis.

[0050] Combination of the above Figure 2 A method for training a thrombolytic adaptability assessment model is described. By acquiring patient perfusion imaging data, perfusion parameter maps, clinical data, and treatment data, the model is provided with a comprehensive and rich source of information, enabling it to make judgments based on multiple factors. Regions of interest (ROIs) are extracted from the perfusion parameter maps to generate attention maps, highlighting the core infarct region and the ischemic penumbra sensing region. This attention map is then multiplied with the perfusion imaging data to generate weighted perfusion image data. This process enhances the model's focus on key regions and helps capture features with greater diagnostic value. Treatment data are annotated to clarify whether thrombolytic therapy was initiated and the grading of treatment effects, providing accurate supervisory information for model training. Inputting this data as training data allows the model to learn the complex relationships between different factors and thrombolytic adaptability and treatment efficacy, thereby improving the model's accuracy in predicting patient adaptability and treatment efficacy, providing a more reliable basis for clinical decision-making.

[0051] Figure 3 FIG2 shows an exemplary flow chart of a process 300 of training a thrombolysis adaptability assessment model using training data according to an embodiment of the present disclosure. Figure 3 The description is a specific implementation of the above step S205. Figure 2 The features described can apply analogously here.

[0052] like Figure 3As shown, at step S301, the weighted cerebral perfusion image data, the cerebral perfusion parameter map, and the clinical data can be respectively input into the first extraction model, the second extraction model, and the third extraction model for feature extraction to obtain a first feature map (also referred to as a weighted cerebral perfusion feature map), a second feature map (also referred to as a perfusion parameter feature map), and a third feature (also referred to as a clinical feature). Then, at step S302, the first feature map, the second feature map, and the third feature can be input into a thrombolysis adaptability analysis model for thrombolysis adaptability analysis to output a first evaluation result and a second evaluation result.

[0053] Finally, at step S303, a loss value can be determined based on the first annotation information, the second annotation information, the first evaluation result, and the second evaluation result. Based on the loss value, the parameters of each model and each layer in the thrombolysis adaptability evaluation model can be updated through backpropagation. Specifically, backpropagation can be used to calculate the gradient of the error relative to each parameter in the thrombolysis adaptability evaluation model, and stochastic gradient descent (SGD) or adaptive moment estimation (Adam) can be used to update the parameters of each model and each layer based on the gradient.

[0054] In the embodiment of the present disclosure, the loss value determined at step S303 can be a weighted sum of a first loss value and a second loss value, wherein the first loss value can evaluate the accuracy of the model's assessment of whether it is suitable for thrombolytic therapy, and the second loss value can evaluate the accuracy of the model's assessment of the effect grading after thrombolytic therapy. In actual operation, the first loss value can be determined based on the first annotation information and the first evaluation result using a first loss function (such as Binary Cross-Entropy, BCE loss), and the second loss value can be determined based on the second annotation information and the second evaluation result using a second loss function (ategorical Cross-Entropy, CCE loss). It should be understood that the specific values ​​of the proportion of the first loss value and the second loss value in the loss value can be set according to actual needs, and the present disclosure does not make specific limitations on this.

[0055] After completing the training of the thrombolysis adaptability assessment model, the performance of the trained thrombolysis adaptability assessment model can be tested using test data. If it is determined that its performance meets actual needs, the model can be used in actual thrombolysis adaptability assessment tasks.

[0056] In some implementation scenarios, a cross-validation method can be used to evaluate the generalization ability of the model to prevent overfitting. Specifically, the verification data can be divided into multiple subsets, and the model can be verified multiple times on different subsets to evaluate the performance of the model on unseen data, thereby quantifying the generalization ability of the model and avoiding overfitting misjudgments caused by data set division deviations. It is understandable that before the verification data is input into the evaluation model, a pre-processing process (such as normalization and noise reduction) that is completely consistent with the training data needs to be adopted to ensure the consistency of data distribution. In other implementation scenarios, test data that is completely independent of the training data and verification data can also be used to perform a final evaluation on the thrombolytic adaptability evaluation model. Further, accuracy, precision, recall rate, F1-score, etc. can be used to evaluate the model results. It is understandable that before the test data is input into the evaluation model, a pre-processing process (such as normalization and noise reduction) that is completely consistent with the training data needs to be adopted to ensure the consistency of data distribution.

[0057] Combined with the above Figure 2 and Figure 3 The present disclosure also provides a method for training a thrombolysis adaptability assessment model. Figure 4 The device 400 provided in the embodiment of the present application is introduced as an example. Figure 4 As shown, the electronic device 400 of the embodiment of the present application may include a processor 401 , a memory 402 and a communication bus 403 .

[0058] In a specific embodiment, the processor 401 may be at least one of an application-specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field programmable gate array (FPGA), a CPU, a controller, a microcontroller, and a microprocessor. It is understood that for different devices, the electronic device used to implement the functions of the processor may also be other, and this embodiment does not specifically limit this.

[0059] In the embodiment of the present application, the communication bus 403 is used to realize the connection and communication between the processor 401 and the memory 402; the memory 402 stores program instructions for training the thrombolysis adaptability evaluation model; when the processor 401 executes the program instructions stored in the memory 402, the present application is realized. Figure 2 and Figure 3 The described method for training a thrombolysis fitness assessment model.

[0060] Combination of the above Figure 4 An electronic device for training a thrombolysis adaptability assessment model that can be used to execute the present application is described. It should be understood that the device structure or architecture here is merely exemplary, and the implementation method and implementation entity of the present application are not limited thereto, but can be changed without departing from the spirit of the present application. It is understandable that the description of each embodiment in this disclosure emphasizes the differences between the various embodiments, and the same or corresponding parts can be referenced to each other. For the purpose of brevity, this disclosure will not go into details one by one.

[0061] In addition, the present disclosure also provides a device for evaluating the suitability of thrombolysis. The structure or architecture of the device, combined with the above Figure 4 The structure or framework of the device 400 for training the thrombolytic adaptability assessment model is the same as that described above. For the purpose of brevity, the present disclosure will not describe them one by one. In the device for assessing thrombolytic adaptability, the memory stores program instructions for assessing thrombolytic adaptability. When the program instructions are executed by the processor, the device can implement the following combination Figure 5 A method 500 for assessing suitability for thrombolysis is described.

[0062] Figure 5 An exemplary flow chart of a method 500 for evaluating thrombolytic suitability according to an embodiment of the present disclosure is shown. It is understood that the method 500 can be executed by any appropriate device with data processing capabilities, such as but not limited to a processor, a terminal device, and a server.

[0063] like Figure 5 As shown, at step S501, method 500 can obtain the cerebral perfusion image data, cerebral perfusion parameter map and clinical data of the target patient. Then, at step S502, method 500 can extract the region of interest on the cerebral perfusion parameter map to obtain an attention map including the core infarct area and the ischemic penumbra sensing area. At step S503, method 500 can multiply the cerebral perfusion image data with the attention map to obtain weighted cerebral perfusion image data. It should be noted that, in this embodiment, the contents of steps S502 and S503 are consistent with steps S202 and S203 in the previous embodiment, and will not be repeated here.

[0064] Next, at step S504, method 500 may input the weighted cerebral perfusion image data, cerebral perfusion parameter map and clinical data into the trained thrombolysis adaptability assessment model for prediction, so as to output a first assessment result and a second assessment result. Here, the first assessment result may indicate whether the target patient is suitable for thrombolysis treatment, and the second assessment result may indicate the effect grade of the target patient after thrombolysis treatment. It is understood that the trained thrombolysis adaptability assessment model here refers to the one that is used in combination with the present application. Figure 2 and Figure 3 The described method for training a thrombolysis fitness assessment model trains the completed model.

[0065] As can be seen from the foregoing description, the first evaluation result output by the trained thrombolytic adaptability evaluation model is the probability of being suitable for thrombolytic therapy. On this basis, the first evaluation result can be compared with a preset threshold (such as 0.5) to determine whether it is suitable for thrombolytic therapy. Specifically, when the first evaluation result is greater than or equal to 0.5, it can be determined to be suitable for thrombolytic therapy, otherwise, it can be determined to be unsuitable for thrombolytic therapy. Similarly, the second evaluation result output by the thrombolytic adaptability evaluation model is the probability of each effect grade after thrombolytic therapy. On this basis, the probabilities of each effect grade can be compared, and the effect grade corresponding to the maximum probability can be used as the effect grade after thrombolytic therapy for the target patient.

[0066] In the disclosed embodiment, the target patient can be any acute ischemic stroke AIS patient. Before inputting the weighted cerebral perfusion imaging data, cerebral perfusion parameter map and clinical data of the target patient into the trained thrombolysis adaptability assessment model for evaluation, the data can be preprocessed using a preprocessing process (such as normalization and noise reduction) that is completely consistent with the training data to improve the accuracy of the prediction results. Figure 5 A method for assessing thrombolytic suitability is described. By acquiring a target patient's cerebral perfusion imaging data, cerebral perfusion parameter maps, and clinical data and inputting these into a trained model, the method rapidly outputs a first assessment result regarding the patient's suitability for thrombolytic therapy and a second assessment result regarding the grading of the therapeutic effect. This method not only significantly shortens assessment time and improves efficiency, but also avoids the subjectivity and uncertainty inherent in human assessments. Based on a model trained on a large amount of data, the assessment results are more objective and accurate, providing strong support for clinical practice.

[0067] According to the above description in conjunction with the accompanying drawings, those skilled in the art can also understand that the embodiments of the present application can also be implemented by a software program. Therefore, the present application also provides a computer-readable storage medium. The computer-readable storage medium stores program instructions for training a thrombolytic adaptability assessment model and / or for assessing thrombolytic adaptability. The program instructions can be used to implement the present application in conjunction with Figure 2 and Figure 3 The method described for training a thrombolytic fitness assessment model or a combination thereof Figure 5 The method described for assessing suitability for thrombolysis.

[0068] It should be noted that although the operations of the present method are described in a particular order in the accompanying drawings, this does not require or imply that the operations must be performed in that particular order, or that all of the operations shown must be performed to achieve the desired results. Rather, the steps depicted in the flowcharts may be performed in a different order. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into a single step, and / or a single step may be broken down into multiple steps.

[0069] Although multiple embodiments of the present application have been shown and described herein, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art can conceive of many changes, modifications, and alternatives without departing from the thought and spirit of the present application. It should be understood that in the process of practicing the present application, various alternatives to the embodiments of the present application described herein can be adopted. The accompanying claims are intended to define the scope of protection of the present application and therefore cover equivalents or alternatives within the scope of these claims.

[0070] The collection and acquisition of various data in this application complies with relevant laws and regulations and is authorized by the data provider. Any organization or individual that needs to obtain external data must obtain authorization in accordance with the law and ensure data security. They must 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 thrombolysis adaptability assessment model, the method comprising: Obtain the patient's cerebral perfusion imaging data, cerebral perfusion parameter maps, clinical data, and treatment data; performing region of interest extraction on the cerebral perfusion parameter map to obtain an attention map including a core infarct area and an ischemic penumbra sensing area; multiplying the cerebral perfusion image data by the attention map to obtain weighted cerebral perfusion image data; Based on the treatment data, using the first annotation information and the second annotation information, respectively, annotate whether thrombolytic therapy has been performed and the grade of the effect after the thrombolytic therapy has been performed; The weighted cerebral perfusion image data, the cerebral perfusion parameter map, the clinical data, the first annotation information, and the second annotation information are input as training data into the thrombolysis adaptability assessment model to train the model.

2. The method according to claim 1, wherein The thrombolysis adaptability assessment model includes a feature extraction model and a thrombolysis adaptability analysis model; the feature extraction model includes a first extraction model, a second extraction model, and a third extraction model; and inputting the weighted cerebral perfusion image data, the cerebral perfusion parameter map, the clinical data, and the annotation information as training data into the thrombolysis adaptability assessment model to train the model includes: Inputting the weighted cerebral perfusion image data, the cerebral perfusion parameter map, and the clinical data into the first extraction model, the second extraction model, and the third extraction model, respectively, for feature extraction to obtain a first feature map, a second feature map, and a third feature; Inputting the first characteristic graph, the second characteristic graph, and the third characteristic into the thrombolysis adaptability analysis model to perform thrombolysis adaptability analysis to output a first evaluation result and a second evaluation result, wherein the first evaluation result is the probability of being suitable for thrombolytic therapy, and the second evaluation result is the probability of each effect grade after thrombolytic therapy; A loss value is determined based on the first labeling information, the second labeling information, the first evaluation result, and the second evaluation result, and parameters of the thrombolysis adaptability evaluation model are updated based on the loss value.

3. The method according to claim 2, wherein: The thrombolysis adaptability analysis model includes a 1*1 convolution block and at least one fully connected layer; and the first feature map, the second feature map, and the third feature are input into the thrombolysis adaptability analysis model to perform thrombolysis adaptability analysis to output a first evaluation result and a second evaluation result, including: Inputting the first feature map, the second feature map and the third feature into the 1*1 convolution block for fusion to obtain a fused feature; The fused features are input into the at least one fully connected layer for evaluation to obtain the first evaluation result and the second evaluation result.

4. The method according to claim 3, wherein: Inputting the fused features into the at least one fully connected layer for evaluation to obtain the first evaluation result and the second evaluation result includes: Inputting the fused features into the at least one fully connected layer for linear transformation to output the linearly transformed features; Based on the linearly transformed features, using a first activation function to obtain the first evaluation result; Based on the linearly transformed features, a second activation function is used to obtain the second evaluation result.

5. The method according to claim 2, wherein: Determining the loss value based on the first labeling information, the second labeling information, the first evaluation result, and the second evaluation result includes: Determining a first loss value using a first loss function based on the first labeling information and the first evaluation result; Determining a second loss value using a second loss function based on the second labeling information and the second evaluation result; A weighted sum of the first loss value and the second loss value is determined as the loss value.

6. The method according to any one of claims 1 to 5, wherein: The clinical data include age, gender, blood pressure, blood sugar, blood oxygen saturation, timing of thrombolytic therapy and stroke scale score, and the treatment data include whether thrombolytic therapy was performed and the stroke scale score after thrombolytic therapy.

7. The method according to claim 6, wherein: After obtaining the patient's cerebral perfusion image data, cerebral perfusion parameter map, and clinical data, the method further includes: performing noise reduction and normalization on the cerebral perfusion image data and the cerebral perfusion parameter map to perform standardization processing on the cerebral perfusion image data and the cerebral perfusion parameter map, thereby obtaining standardized cerebral perfusion image data and cerebral perfusion parameter map; The quantitative data or semi-quantitative data in the clinical data are normalized respectively to perform standardization processing on the clinical data to obtain standardized clinical data.

8. A device for training a thrombolytic adaptability assessment model, comprising: processor; as well as A memory storing program instructions for training a thrombolysis adaptability assessment model, wherein when the program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

9. A device for evaluating thrombolytic suitability, characterized in that: include: processor; A memory storing program instructions for evaluating thrombolysis suitability, wherein when the program instructions are executed by the processor, the device implements the following operations: Obtain cerebral perfusion imaging data, cerebral perfusion parameter maps and clinical data of target patients; performing region of interest extraction on the cerebral perfusion parameter map to obtain an attention map including a core infarct area and an ischemic penumbra sensing area; multiplying the cerebral perfusion image data by the attention map to obtain weighted cerebral perfusion image data; The weighted cerebral perfusion image data, cerebral perfusion parameter map and clinical data are input into a thrombolysis adaptability assessment model trained according to the method described in any one of claims 1 to 7 for evaluation to output a first assessment result and a second assessment result, wherein the first assessment result indicates whether the target patient is suitable for thrombolytic therapy, and the second assessment result indicates the effect grade of the target patient after thrombolytic therapy.

10. A computer-readable storage medium storing program instructions for training a thrombolytic adaptability assessment model and / or for assessing thrombolytic adaptability, wherein when the program instructions are executed by a processor, the method according to any one of claims 1 to 7 and / or the operations implemented by the device according to claim 9 are implemented.

Images