A method, device, equipment and medium for intelligent prediction of epilepsy drug resistance
The drug resistance prediction model is established through deep learning methods, and the features of MRI images are automatically extracted, which solves the high cost and inefficiency problems caused by manual extraction of features in the prior art, and achieves more efficient and accurate prediction of epilepsy drug treatment results.
Patent Information
- Application Number
- CN202111364731.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-11-17
- Publication Date
- 2025-04-01
- Estimated Expiration
- 2041-11-17
AI Technical Summary
The prior art relies on manual extraction of MRI characteristics when predicting epilepsy drug treatment results, resulting in high cost, low efficiency, and may lead to errors.
Using deep learning method, a drug resistance prediction model is established through two three-dimensional convolutional neural networks and fully connected networks, the feature vectors of the FLAIR sequence diagram and the T2W sequence diagram are automatically extracted, and feature styling is performed for prediction.
It improves the efficiency and accuracy of epilepsy resistance prediction, reduces the cost and error of manual extraction of features, and achieves more efficient prediction results.
Smart Images
Figure CN114267420B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of epilepsy drug resistance prediction, and in particular to an intelligent prediction method, device, equipment and medium for epilepsy drug resistance. Background Art
[0002] At present, the main way to diagnose the results of epilepsy drug treatment in clinical practice is: after confirming TSC (tuberonosclerosis), the patient is treated with AEDs (antiepileptic drugs) for more than one year. The doctor observes whether the patient still has epileptic seizures after taking the medicine for one year to judge the results of drug treatment, and divides the patients into drug-controlled and drug-refractory patients. Then, the treatment method for drug-refractory patients is changed, such as surgery. This method has the following disadvantages: high labor cost. Due to the particularity of drug treatment, it requires professionally trained physicians to treat patients with drugs; high time cost. The outcome of drug treatment generally requires more than one year of treatment to know whether the patient is drug-resistant, and then the treatment plan for drug-resistant patients is changed, which is very unfavorable for patient treatment and may delay the best treatment time for patients.
[0003] Based on the above problems, related technologies use manual extraction of lesion location, lesion type and number from MRI (Magnetic Resonance Imaging) of some TSC epilepsy patients, and then use a variety of machine learning methods to model and predict the results of epilepsy drug treatment. However, since MRI features are usually manually extracted, they are not objective, and manual extraction is costly and inefficient.
[0004] Therefore, how to provide a solution to the above technical problems is a problem that those skilled in the art need to solve at present. Summary of the invention
[0005] The first purpose of this application is to provide an intelligent prediction method for epilepsy drug resistance, which can improve the efficiency of epilepsy drug resistance prediction and reduce labor costs.
[0006] The above-mentioned application objective 1 of the present application is achieved through the following technical solutions:
[0007] An intelligent prediction method for epilepsy drug resistance, comprising:
[0008] Acquire an MRI image training set of epilepsy drug resistance, and extract a FLAIR sequence diagram and a T2W sequence diagram from each MRI image in the MRI image training set; wherein the MRI image training set includes a control group image and an uncontrolled group image;
[0009] A drug resistance prediction model to be trained is established according to a two-way three-dimensional convolutional neural network and a fully connected network; wherein the two-way three-dimensional convolutional neural networks are used to extract feature vectors of the FLAIR sequence diagram and the T2W sequence diagram, respectively, and feature splicing is performed on the two feature vectors;
[0010] Using all of the FLAIR sequence diagrams and the T2W sequence diagrams to train the drug resistance prediction model to be trained, to obtain a drug resistance prediction model;
[0011] The MRI image to be predicted is acquired, and a prediction result of epilepsy drug resistance is obtained according to the MRI image to be predicted and the drug resistance prediction model.
[0012] By adopting the above technical scheme, this scheme uses deep learning to establish a drug resistance prediction model, and automatically extracts and splices the feature vectors of the FLAIR sequence diagram and the T2W sequence diagram, avoiding the errors caused by manual extraction, and also improving the extraction efficiency. The spliced features are then input into the fully connected network for prediction, and the prediction results are output. The whole process is low-cost and high-efficiency.
[0013] In a preferred example, the present application can be further configured as follows:
[0014] The method for obtaining an MRI image training set for epilepsy drug resistance comprises:
[0015] Acquire initial control group images and initial uncontrolled group images;
[0016] Performing affine transformation on the initial control group image and the initial uncontrolled group image to obtain an expanded control group image and an expanded uncontrolled group image;
[0017] The initial control group images and the expanded control group images are determined as the control group images, and the initial uncontrolled group images and the expanded uncontrolled group images are determined as the uncontrolled group images, so as to obtain the MRI image training set including the control group images and the uncontrolled group images.
[0018] By adopting the above technical solution, this solution increases the number of images in the trained epilepsy drug-resistant MRI image set by performing affine transformation on the initial control group images and the initial uncontrolled group images, thereby enhancing the robustness of the drug-resistant prediction model.
[0019] In a preferred example, the present application can be further configured as follows:
[0020] The step of performing affine transformation on the initial control group image and the initial uncontrolled group image to obtain an expanded control group image and an expanded uncontrolled group image includes:
[0021] removing interference region images from the initial control group image and the initial uncontrolled group image;
[0022] The affine transformation is performed on the initial control group image and the initial uncontrolled group image after removing the interference area image to obtain the expanded control group image and the expanded uncontrolled group image.
[0023] By adopting the above technical solution, this solution helps to classify by removing the interference area in the initial image, and at the same time can reduce the complexity of model training and improve training efficiency.
[0024] In a preferred example, the present application can be further configured as follows:
[0025] The method of training the drug resistance prediction model to be trained by using all the FLAIR sequence diagrams and the T2W sequence diagrams to obtain the drug resistance prediction model comprises:
[0026] Training the drug resistance prediction model to be trained according to all the FLAIR sequence diagrams and the T2W sequence diagrams to obtain an initial drug resistance prediction model;
[0027] evaluating the initial drug resistance prediction model;
[0028] After passing the evaluation, the initial drug resistance prediction model is determined to be the drug resistance prediction model.
[0029] By adopting the above technical scheme, this scheme evaluates the initial drug resistance prediction model after initial training, and when it fails the evaluation, continues training until it passes the evaluation; only after passing the evaluation can it be used as a drug resistance prediction model, thereby improving the reliability of the drug resistance prediction model.
[0030] In a preferred example, the present application can be further configured as follows:
[0031] The evaluating of the initial drug resistance prediction model comprises:
[0032] Evaluating the initial drug resistance prediction model using a cross-validation approach;
[0033] Correspondingly, the prediction result of epilepsy drug resistance is obtained according to the MRI to be predicted and the drug resistance prediction model, including:
[0034] Using the drug resistance prediction model to predict the image to be predicted, and obtaining a plurality of predicted values of drug resistance results, wherein the number of the predicted values of drug resistance results corresponds to the number of folds of the cross-validation method;
[0035] Comparing the average value of the predicted value of the drug resistance result with a preset drug resistance threshold to obtain a comparison result;
[0036] The prediction result is determined according to the comparison result.
[0037] By adopting the above technical scheme, this scheme evaluates the model through cross-validation, which greatly improves the reliability of the drug resistance prediction model. When the cross-validation method is adopted, in the actual prediction process, after each model outputs the predicted value of the drug resistance result, the average value of the predicted value of the drug resistance result is compared with the preset drug resistance threshold. When the drug resistance threshold is reached, the predicted result is determined to be drug-resistant. When the drug resistance threshold is not reached, the predicted result is determined to be not drug-resistant.
[0038] In a preferred example, the present application can be further configured as follows:
[0039] The method of training the drug resistance prediction model to be trained by using all the FLAIR sequence diagrams and the T2W sequence diagrams to obtain the drug resistance prediction model comprises:
[0040] Inputting the FLAIR sequence image into a first three-dimensional convolutional neural network, performing convolution, pooling, and full connection on the FLAIR sequence image in sequence to obtain a first feature vector;
[0041] Inputting the T2W sequence graph into a second three-dimensional convolutional neural network, performing convolution, pooling, and full connection on the T2W sequence graph in sequence to obtain a second feature vector;
[0042] The first feature vector and the second feature vector are feature spliced, and the spliced feature data is output to the fully connected network until the training of the drug resistance prediction model to be trained by all the FLAIR sequence images and the T2W sequence images is completed to obtain the drug resistance prediction model.
[0043] By adopting the above technical scheme, this scheme establishes a complete drug resistance prediction model to be trained, which includes a three-dimensional convolutional neural network composed of a first three-dimensional convolutional neural network and a second three-dimensional convolutional neural network, and a fully connected network connected to the three-dimensional convolutional neural network. The whole model building process is simple and easy to operate, avoiding the problems of long training time and complex creation caused by training two networks separately.
[0044] In a preferred example, the present application can be further configured as follows:
[0045] Also includes:
[0046] Obtain a new MRI image training set;
[0047] Training the drug resistance prediction model according to the new MRI image training set to obtain a new drug resistance prediction model;
[0048] Use the new drug resistance prediction model to predict new MRI images to be predicted;
[0049] The step of acquiring a new MRI image training set is performed until the number of iterations reaches a set threshold or the prediction accuracy reaches a set value.
[0050] Based on the above technical solution, this solution uses the new MRI image training set to train the current drug resistance prediction model. Since the new MRI image training set is continuously used to iteratively train the previous drug resistance prediction model, the accuracy of drug resistance prediction is continuously improved. The deep neural network algorithm is used to ensure the quality of each training, improve the accuracy of drug resistance prediction, and improve the user experience.
[0051] The second purpose of this application is to provide an intelligent prediction device for epilepsy drug resistance, comprising:
[0052] An image acquisition module, used to acquire an MRI image training set of epilepsy drug resistance, and extract a FLAIR sequence diagram and a T2W sequence diagram from each MRI image in the MRI image training set; wherein the MRI image training set includes a control group image and an uncontrolled group image;
[0053] A model building module, used to establish a drug resistance prediction model to be trained based on two three-dimensional convolutional neural networks and a fully connected network; wherein the two three-dimensional convolutional neural networks are used to extract feature vectors of the FLAIR sequence diagram and the T2W sequence diagram, respectively, and feature splicing is performed on the two feature vectors;
[0054] A model training module, used for training the drug resistance prediction model to be trained by using all the FLAIR sequence diagrams and the T2W sequence diagrams to obtain a drug resistance prediction model;
[0055] The prediction module is used to obtain the MRI image to be predicted, and obtain the prediction result of epilepsy drug resistance according to the MRI image to be predicted and the drug resistance prediction model.
[0056] By adopting the above technical scheme, this scheme uses deep learning to establish a drug resistance prediction model, and automatically extracts and splices the feature vectors of the FLAIR sequence diagram and the T2W sequence diagram, avoiding the errors caused by manual extraction, and also improving the extraction efficiency. The spliced features are then input into the fully connected network for prediction, and the prediction results are output. The whole process is low-cost and high-efficiency.
[0057] The third object of the present application is to provide an electronic device, comprising:
[0058] One or more processors;
[0059] Memory;
[0060] One or more applications, wherein the one or more applications are stored in a memory and configured to be executed by one or more processors, and the one or more programs are configured to: execute an intelligent prediction method for epilepsy drug resistance shown in any possible implementation of the first aspect.
[0061] By adopting the above technical scheme, this scheme uses deep learning to establish a drug resistance prediction model, and automatically extracts and splices the feature vectors of the FLAIR sequence diagram and the T2W sequence diagram, avoiding the errors caused by manual extraction, and also improving the extraction efficiency. The spliced features are then input into the fully connected network for prediction, and the prediction results are output. The whole process is low-cost and high-efficiency.
[0062] The fourth application objective of this application is achieved through the following technical solutions:
[0063] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the above-mentioned intelligent prediction method for epilepsy drug resistance.
[0064] By adopting the above technical scheme, this scheme uses deep learning to establish a drug resistance prediction model, and automatically extracts and splices the feature vectors of the FLAIR sequence diagram and the T2W sequence diagram, avoiding the errors caused by manual extraction, and also improving the extraction efficiency. The spliced features are then input into the fully connected network for prediction, and the prediction results are output. The whole process is low-cost and high-efficiency. BRIEF DESCRIPTION OF THE DRAWINGS
[0065] Figure 1 A schematic diagram of a process flow of an intelligent prediction method for epilepsy drug resistance provided in an embodiment of the present application;
[0066] Figure 2 A schematic diagram of a process for acquiring an epilepsy drug-resistant MRI image set provided in an embodiment of the present application;
[0067] Figure 3 A schematic diagram of a training process of a drug resistance prediction model to be trained provided in an embodiment of the present application;
[0068] Figure 4 A schematic diagram of a three-layer three-dimensional convolutional neural network structure provided in an embodiment of the present application;
[0069] Figure 5 A schematic diagram of the structure of a drug resistance prediction model provided in an embodiment of the present application;
[0070] Figure 6 A schematic diagram of an iterative training process provided in an embodiment of the present application;
[0071] Figure 7 It is a structural block diagram of an intelligent prediction device for epilepsy drug resistance provided in an embodiment of the present application;
[0072] Figure 8 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0073] The present application is further described in detail below in conjunction with the accompanying drawings.
[0074] This specific embodiment is merely an explanation of the present application and is not a limitation of the present application. After reading this specification, a person skilled in the art may make non-creative modifications to the present embodiment as needed, but such modifications are protected by the patent law as long as they are within the scope of the present application.
[0075] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0076] In addition, the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article, unless otherwise specified, generally means that the associated objects before and after are in an "or" relationship.
[0077] The embodiments of the present application are further described in detail below in conjunction with the drawings in the specification.
[0078] The embodiment of the present application provides an intelligent prediction method for epilepsy drug resistance, which is executed by an electronic device, which can be a server or a terminal device, wherein the server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The terminal device can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc., but is not limited thereto. The terminal device and the server can be directly or indirectly connected via wired or wireless communication, and the embodiment of the present application does not limit this. Figure 1 As shown, Figure 1 A schematic flow chart of an intelligent prediction method for epilepsy drug resistance provided in an embodiment of the present application, the method comprising:
[0079] S110 , obtaining an MRI image training set of epilepsy drug resistance, and extracting a FLAIR sequence image and a T2W sequence image from each MRI image in the MRI image training set.
[0080] The MRI image training set includes control group images and uncontrolled group images.
[0081] In TSC patients, neurological manifestations can almost be observed in brain images, and magnetic resonance imaging has rich soft tissue contrast and is an advanced imaging tool for clinical diagnosis of TSC. Cortical tubers and subcortical tubers are the main TSC-related brain lesions. Not only can brain lesions be clearly observed on MRI images, but also the patient's brain structure can be observed. Specifically, abnormal high or low signals can be clearly observed in the fluid attenuated inversion recovery (FLAIR) sequence and T2W sequence of MRI images. The FLAIR sequence diagram and T2W sequence diagram used in this embodiment are used as training diagrams for each patient. This embodiment combines the FLAIR sequence diagram and the T2W sequence diagram for deep learning to achieve an optimal prediction effect.
[0082] Among them, a FLAIR (fluid attenuated inversion recovery) sequence diagram and a T2W sequence diagram corresponding to the FLAIR sequence diagram are extracted from each MRI image in the MRI image training set of epilepsy drug resistance. It can be understood that each image contains a label of whether it is drug resistant.
[0083] The MRI image training set includes control group images and uncontrolled group images. The epilepsy drug-resistant MRI image set of this embodiment selects patients diagnosed with TSC-related epilepsy upon admission. The patient inclusion criteria may include: fluid-attenuated inversion recovery images and T2-weighted images, i.e., T2W image scans, before AEDs treatment; AEDs treatment for at least one year; The patient did not undergo lesion resection surgery, wherein the treatment principle is determined by senior tuberous sclerosis experts based on experience and guidelines. The results of epilepsy drug treatment are defined according to the 1981 ILAE (International League Against Epilepsy) classification, and all patients are divided into a control group and an uncontrolled group. If the patient has no clinical epileptic seizure within at least 1 year of AEDs treatment, he is regarded as a control group, and the corresponding image is a control group image; if the patient has at least one epileptic seizure or dies within one year, he is regarded as an uncontrolled group, and the corresponding image is an uncontrolled group image. It is worth noting that the control group images are images taken by each patient in the control group before receiving epilepsy drug treatment, and similarly, the uncontrolled group images are images taken by each patient in the uncontrolled group before receiving epilepsy drug treatment.
[0084] S120. Establish a drug resistance prediction model to be trained based on a two-way three-dimensional convolutional neural network and a fully connected network.
[0085] Among them, two three-dimensional convolutional neural networks are used to extract the feature vectors of FLAIR sequence images and T2W sequence images respectively, and perform feature splicing on the two feature vectors.
[0086] The purpose of this step is to build a drug resistance prediction model to be trained. A two-way three-dimensional convolutional neural network is used to automatically extract and fuse features, and a fully connected network is used for drug resistance prediction.
[0087] The three-dimensional convolutional neural network is further elaborated. In this embodiment, two three-dimensional convolutional neural networks are used to extract feature vectors of the FLAIR sequence image and the T2W sequence image, and then the features are fused. This improves the extraction efficiency while also improving the accuracy of feature extraction, avoiding the low efficiency and targeted extraction problems caused by manual extraction. Specifically, the three-dimensional convolutional neural network method in deep learning is used to automatically extract features of images. This method can extract features from the FLAIR sequence image and the T2W sequence image at different levels, and learn some intermediate and advanced features of the FLAIR sequence image and the T2W sequence image. These features can better express the original image features, and the three-dimensional weights can be used to extract the FLAIR sequence image and the T2W sequence image. Figure 3 The 3D convolutional neural network can extract important information in three dimensions, which makes up for the defect that the traditional 2D convolutional neural network cannot utilize 3D spatial information, thereby improving the accuracy of feature extraction. This embodiment does not limit the number of convolutional layers of the 3D convolutional neural network, and the user can set it according to actual needs. Preferably, the number of convolutional layers is 3.
[0088] It can be understood that the structure of the entire drug resistance prediction model to be trained is an end-to-end structure, that is, the three-dimensional FLAIR image and T2W image are input, and the prediction result of whether the epilepsy caused by TSC is drug-resistant is output.
[0089] Furthermore, a random dropout regularization mechanism can be applied to the three-dimensional convolutional neural network to prevent overfitting. The random dropout regularization mechanism is used during training to improve the generalization ability of the three-dimensional convolutional neural network.
[0090] In an achievable implementation, the drug resistance monitoring model to be trained needs to train two three-dimensional convolutional neural networks and a fully connected network together. It is worth noting that the parameters in the drug resistance prediction model to be trained can be customized as long as the purpose of this embodiment can be achieved.
[0091] S130 , training the drug resistance prediction model to be trained using all FLAIR sequence images and T2W sequence images to obtain a drug resistance prediction model.
[0092] The two-way three-dimensional convolutional neural network and the fully connected network are used as a model for training to obtain a trained drug resistance prediction model. In this way, the entire model building process is simple and easy to operate.
[0093] S140, obtaining an MRI image to be predicted, and obtaining a prediction result of epilepsy drug resistance according to the MRI image to be predicted and a drug resistance prediction model.
[0094] Obtain the MRI image to be predicted, which is the MRI image of the current patient before drug treatment. The MRI image to be predicted specifically includes FLAIR sequence diagram and T2W sequence diagram. Use the deep learning drug resistance prediction model to predict the predicted results of drug treatment of epilepsy caused by TSC. The predicted results include drug resistance and non-drug resistance, which reduces the manpower cost of doctors. Doctors only need to collect the FLAIR sequence and T2W sequence of TSC patients and input the two sequence diagrams into the drug resistance prediction model to quickly predict the drug treatment outcome of epilepsy patients, and assist doctors in formulating better treatment plans for TSC patients. At the same time, it reduces the patient's time cost. Before the start of drug treatment, MRI images are used to predict whether the patient is drug-resistant, giving the patient time for more appropriate treatment, and the patient does not need drug resistance tests for more than one year.
[0095] Based on the above technical solution, this embodiment uses deep learning to establish a drug resistance prediction model, and automatically extracts and splices the feature vectors of the FLAIR sequence diagram and the T2W sequence diagram, avoiding the errors caused by manual extraction and improving the extraction efficiency. The spliced features are then input into a fully connected network for prediction, and the prediction results are output. The whole process is low-cost and high-efficiency.
[0096] Based on the above embodiment, since the data set of medical images is usually small, and deep learning training requires a large data set, when deep learning is performed on a small data set, fitting phenomenon is prone to occur, and the robustness of the entire model is poor. Therefore, in order to solve the above problem, this embodiment improves the data set by performing affine transformation on the image, please refer to Figure 2 , Figure 2 A schematic diagram of a process for acquiring an epilepsy drug-resistant MRI image set provided in an embodiment of the present application includes:
[0097] S111, acquiring an initial control group image and an initial non-control group image;
[0098] In this embodiment, the sizes of the initial control group images and the initial non-control group images are the same, and each FLAIR image and T2W image may be an image with a size of 256×256×256 and normalized.
[0099] S112, performing affine transformation according to the initial control group image and the initial uncontrolled group image to obtain an expanded control group image and an expanded uncontrolled group image;
[0100] S113, determining the initial control group images and the expanded control group images as the control group images, and determining the initial uncontrolled group images and the expanded uncontrolled group images as the uncontrolled group images, so as to obtain an MRI image training set including the control group images and the uncontrolled group images.
[0101] Among them, due to the particularity of medical images, it is difficult to obtain MRI images of epilepsy drug resistance. In the case of less MRI image data, overfitting is prone to occur during deep learning, and the robustness of the drug resistance prediction model obtained by training is poor. Therefore, in this embodiment, the training images are expanded using affine transformation, wherein the affine transformation is to perform any one or more of translation, rotation, scaling, shearing, and symmetry on the initial control group images and the initial uncontrolled group images, and finally obtain a sufficient number of training sets, that is, composed of the actual collected three-dimensional FLAIR and T2W data, and the data set is expanded by means of image shearing, rotation, flipping, etc., to enhance the robustness of the model.
[0102] Based on the above technical solution, this embodiment increases the number of images in the trained epilepsy drug-resistant MRI image set by performing affine transformation on the initial control group images and the initial non-control group images, thereby enhancing the robustness of the drug-resistant prediction model.
[0103] Based on the above embodiment, in order to improve the model training efficiency and improve the prediction accuracy, an affine transformation is performed on the initial control group image and the initial uncontrolled group image to obtain an expanded control group image and an expanded uncontrolled group image, including:
[0104] Remove the interference area images in the initial control group image and the initial non-control group image;
[0105] Affine transformation is performed on the initial control group image and the initial uncontrolled group image after removing the interference area image to obtain an expanded control group image and an expanded uncontrolled group image.
[0106] The MRI images taken by the patient include interference region images corresponding to parts not associated with brain structures. For example, in the case of the first image in the initial control group image, the first image includes a three-dimensional first FLAIR image and a first T2W image, and both the first FLAIR image and the first T2W image include an effective region image and an interference region image, and the interference region image may correspond to the ear region, skull or eye region.
[0107] This embodiment removes the interference region image, and no longer needs to perform operations such as feature extraction on the interference region image, so as to reduce the complexity of model training, thereby improving the efficiency of model training and the accuracy of the model. Correspondingly, when using the trained drug resistance prediction model for actual prediction, the image to be predicted also needs to remove the interference region image.
[0108] As for the method of removing interfering area images, the deep learning tool HD-bet can be used to remove images that are not related to the lesion. Preprocessing tools such as FSL software and SPM12 (Software-Statistical Parametric Mapping) can also be used to remove interfering areas such as the skull. Among them, FSL software is a powerful brain imaging analysis tool that can analyze functional images, structural images, etc.
[0109] Based on the above technical solution, this embodiment removes interference areas in the initial image, which helps to perform classification and can also reduce the complexity of model training and improve training efficiency.
[0110] Based on the above embodiment, in order to improve the accuracy of the model, step S130 includes:
[0111] The drug resistance prediction model to be trained is trained according to all FLAIR sequence diagrams and T2W sequence diagrams to obtain an initial drug resistance prediction model;
[0112] Evaluation of initial resistance prediction models;
[0113] After passing the evaluation, the initial drug resistance prediction model is determined as the drug resistance prediction model.
[0114] In this embodiment, the drug resistance prediction model to be trained is evaluated after the model is trained. Only if the model passes the evaluation can it be determined as an applicable drug resistance prediction model. If the model fails the evaluation, the initial drug resistance prediction model is continuously trained until the model passes the evaluation.
[0115] Based on the above technical solution, this embodiment evaluates the initial drug resistance prediction model after initial training. If it fails the evaluation, the training is continued until it passes the evaluation. Only if it passes the evaluation can it be used as a drug resistance prediction model, thereby improving the reliability of the drug resistance prediction model.
[0116] The present embodiment provides a specific method for evaluating the initial drug resistance prediction model, including: testing the initial drug resistance prediction model using an epilepsy drug resistance test image set to obtain a test success rate; determining whether the test success rate reaches a set threshold, and if so, determining that the evaluation has been passed; if not, determining that the evaluation has not been passed.
[0117] This embodiment provides another specific method for evaluating the initial drug resistance prediction model, including:
[0118] The initial drug resistance prediction model was evaluated using cross-validation;
[0119] Correspondingly, the prediction results of epilepsy drug resistance are obtained according to the MRI to be predicted and the drug resistance prediction model, including:
[0120] Using the drug resistance prediction model to predict the image to be predicted, a plurality of predicted values of drug resistance results are obtained, wherein the number of predicted values of drug resistance results corresponds to the number of folds of the cross-validation method;
[0121] Comparing the average value of the predicted value of the drug resistance result with the preset drug resistance threshold value to obtain a comparison result;
[0122] The prediction result is determined based on the comparison result.
[0123] This embodiment does not limit the cross-validation method, and it can be a five-fold cross-validation method or a ten-fold cross-validation method.
[0124] Taking the five-fold cross validation method as an example, after the training is completed, the parameters of the five models using the five-fold cross validation are tested on the test set. The prediction score of each patient is determined by the average of the prediction scores of the five models. The specific training and testing process of the five-fold cross validation method includes: dividing the training data set into five parts, selecting one of them as the test set, and the other four as the training set for training. For example, the training set is a, b, c, d, e. The first model is trained using b, c, d, e to obtain the model, and then tested using a to finally obtain the test result; the second model is trained using a, c, d, e to obtain the model, and then tested using b to finally obtain the test result; then, the third model, the fourth model, and the fifth model all obtain the corresponding test results in the above manner, and the mean of the test results is used as the final test result, and the final test result is compared with the test threshold. When the comparison is passed, it is determined that the test passes. Furthermore, the prediction performance of each model in the cross validation method can also be evaluated by calculating AUC (Area under the Curve of ROC, model evaluation index), accuracy, sensitivity, and specificity. Model training, validation, and testing can be implemented in Python (version 3.8.10) and PyTorch (version 1.9.0).
[0125] For the cross-validation method, in actual use, when predicting the MRI image to be predicted, multiple predicted values of drug resistance results will be output. The system has a preset drug resistance threshold. The comparison result of the average value of the predicted value of the drug resistance result and the preset drug resistance threshold is used to determine whether there is drug resistance.
[0126] For example, when the cross-validation method is a five-fold cross-validation method, five models will be generated by using the five-fold cross-validation method for model training and verification. Among the five models, one will be used as a drug resistance prediction model. When predicting the image to be predicted, the five models will output five values, and the average of the five values will be used as the drug resistance value. For example, 0 represents controlled epilepsy, and 1 represents uncontrolled epilepsy. When the values obtained are 0, 1, 1, 0, and 1 in sequence, the drug resistance value is 0.6, and the preset drug resistance threshold is set to 0.5. The prediction result is uncontrolled, that is, drug resistance. For another example, 1 represents controlled epilepsy, and 0 represents uncontrolled epilepsy. When the values obtained are 0, 1, 1, 0, and 1 in sequence, the drug resistance value is 0.6, and the preset drug resistance threshold is set to 0.5. The prediction result is controlled, that is, not drug resistant. The specific setting can be made according to actual needs, and this embodiment will no longer be limited. Anything that can achieve the purpose of this embodiment belongs to the protection scope of this embodiment.
[0127] Based on the above technical scheme, this implementation evaluates the model through cross-validation, which greatly improves the reliability of the drug resistance prediction model. When the cross-validation method is adopted, in the actual prediction process, after each model outputs the predicted value of the drug resistance result, the predicted value of the drug-resistant result is compared with the preset drug resistance threshold. When the drug resistance threshold is reached, the predicted result is determined to be drug-resistant. When the drug resistance threshold is not reached, the predicted result is determined to be not drug-resistant.
[0128] Based on any of the above embodiments, please refer to Figure 3 , Figure 3 A schematic diagram of a training process of a drug resistance prediction model to be trained provided in an embodiment of the present application includes:
[0129] S131, input the FLAIR sequence image into the first three-dimensional convolutional neural network, perform convolution, pooling, and full connection on the FLAIR sequence image in sequence to obtain a first eigenvector.
[0130] S132, input the T2W sequence graph into the second three-dimensional convolutional neural network, perform convolution, pooling, and full connection on the T2W sequence graph in sequence to obtain a second feature vector.
[0131] Three-dimensional convolutional neural network The three-dimensional convolutional neural network includes the first-dimensional convolutional neural network and the second-dimensional convolutional neural network.
[0132] S133, concatenating the first eigenvector and the second eigenvector, and outputting the concatenated feature data to a fully connected network, until the training of the drug resistance prediction model to be trained by all FLAIR sequence images and T2W sequence images is completed, and the drug resistance prediction model is obtained.
[0133] 3D convolutional neural networks can directly extract features in the spatiotemporal domain and are applied to the field of medical image and video processing. They still have good results when the data set is small. The first-pass 3D convolutional neural network or the second-pass 3D convolutional neural network mainly involves three steps: convolution, pooling, and full connection. Please refer to Figure 4 , Figure 4A schematic diagram of a three-layer three-dimensional convolutional neural network structure provided for an embodiment of the present application, after the image is input, a 3*3*3 convolution and Relu activation are performed once, and an initial first image is output; 2*2*2 maximum pooling is performed on the initial first image, and the first image is output; a second 3*3*3 convolution and Relu activation are performed on the first image, and an initial second image is output; 2*2*2 maximum pooling is performed on the initial second image, and the second image is output; a third 3*3*3 convolution and Relu activation are performed on the second image, and an initial third image is output; 2*2*2 maximum pooling is performed on the initial third image, and the third image is output; a first full connection processing is performed on the third image to obtain a fourth image, and a second full connection processing is performed on the fourth image to obtain a fifth image, and the fifth image is a feature image extracted from the input image.
[0134] In general, the three-dimensional FLAIR sequence and T2W sequence are input into the three-dimensional convolutional neural network to extract feature vectors, and then the extracted feature vectors are fused to predict the results of TSC-induced epilepsy patients. For the epilepsy drug treatment outcome prediction model, that is, the drug resistance prediction model, please refer to Figure 5 , Figure 5 A schematic diagram of the structure of a drug resistance prediction model provided in an embodiment of the present application.
[0135] For example, the size of the input three-dimensional FLAIR and T2W images is (256, 256, 256), and the images are convolved and activated three times. The convolution kernel size can be (3, 3, 3), the number of convolution kernels can be (16, 32, 64), the step size is set to 1, and the activation function uses the ReLU function as shown in formula (1):
[0136] (1);
[0137] Where X is the input feature vector.
[0138] After each convolution, the maximum pooling is performed, the pooling kernel size is set to (2, 2, 2), the step size is set to 2, and the size of the feature map is halved after pooling, that is, the downsampling operation is completed; the network structure of each layer of downsampling is consistent with the first layer, the only difference is the number of pooling kernels, and the number of pooling kernels in each layer is (16, 32, 64) respectively; then the output of the previous pooling network is flattened into a one-dimensional vector through two fully connected networks, the number of fully connected neurons is 256 and 128 respectively, and the dropout parameter is 0.7; finally, the extracted FLAIR and T2W image features are fused into a comprehensive feature through feature concatenation. Then the fully connected layer network is input, and the fully connected layer network has three layers, and the number of fully connected neurons is 32, 16, and 2 respectively. The activation function uses the softmax function, and the output can be 0 or 1, where 0 represents controlled epilepsy and 1 represents uncontrolled epilepsy.
[0139] The user can set the training method according to actual needs, and can use 5-fold cross validation to evaluate the model. The Adam algorithm is used to train the network, the learning rate is set to 0.001, the epoch is set to 100, and the loss function uses the cross entropy loss function, please refer to formula (2):
[0140] (2);
[0141] Among them, loss is the loss function, represents the label of sample i, the positive class is 1, and the negative class is 0; yi represents the probability that sample i is predicted to be a positive class.
[0142] Based on the above technical solution, this embodiment establishes a complete drug resistance prediction model to be trained, which includes a three-dimensional convolutional neural network composed of a first three-dimensional convolutional neural network and a second three-dimensional convolutional neural network, and a fully connected network connected to the three-dimensional convolutional neural network. The entire model establishment process is simple and easy to operate, avoiding the problems of long training time and complex creation caused by training two networks separately.
[0143] Based on any of the above embodiments, please refer to Figure 6 , Figure 6 A schematic diagram of an iterative training process provided in an embodiment of the present application includes:
[0144] S150, obtaining a new MRI image training set;
[0145] The purpose of this step is to continue to collect MRI images of patients participating in drug treatment from the hospital to obtain a new MRI image training set, which includes new control group images and uncontrolled group images.
[0146] S160, training the drug resistance prediction model according to the new MRI image training set to obtain a new drug resistance prediction model;
[0147] S170, predicting a new MRI image to be predicted using the new drug resistance prediction model;
[0148] S180, executing the step of acquiring a new MRI image training set until the number of iterations reaches a set threshold or the prediction accuracy reaches a set value.
[0149] Due to the problem of small number of training sets in the pharmaceutical industry, after obtaining the drug resistance prediction model that can be applied for prediction, this embodiment continues to accumulate training images, uses the new MRI image training set to train the current drug resistance prediction model, and obtains a new drug resistance prediction model until the number of iterations reaches a set threshold or the prediction accuracy reaches a set value, and stops iterative training. During the training process, the accuracy of the drug resistance prediction model is continuously improved, the user experience is improved, and the technical problem of low accuracy caused by always using a fixed prediction model for prediction is avoided.
[0150] The following is an introduction to an intelligent prediction device for epilepsy drug resistance provided by an embodiment of the present application. The intelligent prediction device for epilepsy drug resistance described below and the intelligent prediction method for epilepsy drug resistance described above can be referred to each other. The intelligent prediction device for epilepsy drug resistance of this embodiment is set in an electronic device, and the intelligent prediction method for epilepsy drug resistance described above can be referred to each other. Figure 7 , Figure 7 This is a structural block diagram of an intelligent prediction device for epilepsy drug resistance provided in an embodiment of the present application, comprising:
[0151] The image set acquisition module 100 is used to acquire an MRI image training set of epilepsy drug resistance, and extract a FLAIR sequence diagram and a T2W sequence diagram from each MRI image in the MRI image training set; wherein the MRI image training set includes a control group image and an uncontrolled group image;
[0152] The model building module 200 is used to establish a drug resistance prediction model to be trained based on a two-way three-dimensional convolutional neural network and a fully connected network; wherein the two-way three-dimensional convolutional neural network is used to extract feature vectors of the FLAIR sequence diagram and the T2W sequence diagram respectively, and perform feature splicing on the two feature vectors; the three-dimensional convolutional neural network three-dimensional convolutional neural network model training module 300 is used to train the drug resistance prediction model to be trained using all the FLAIR sequence diagrams and the T2W sequence diagrams to obtain the drug resistance prediction model;
[0153] The prediction module 400 is used to obtain the MRI image to be predicted, and obtain the prediction result of epilepsy drug resistance according to the MRI image to be predicted and the drug resistance prediction model.
[0154] Preferably, the image set acquisition module 100 includes:
[0155] An initial image acquisition unit, used for acquiring an initial control group image and an initial non-control group image;
[0156] An expansion unit, used for performing affine transformation on the initial control group image and the initial uncontrolled group image to obtain an expanded control group image and an expanded uncontrolled group image;
[0157] The determination unit is used to determine the initial control group image and the expanded control group image as the control group image, and to determine the initial uncontrolled group image and the expanded uncontrolled group image as the uncontrolled group image, so as to obtain an MRI image training set including the control group image and the uncontrolled group image.
[0158] Preferably, the expansion unit comprises:
[0159] A removal subunit, used for removing interference region images from an initial control group image and an initial non-control group image;
[0160] The expansion subunit is used to perform affine transformation on the initial control group image and the initial uncontrolled group image after removing the interference area image, so as to obtain the expanded control group image and the expanded uncontrolled group image.
[0161] Preferably, the model training module 300 includes:
[0162] The first training unit is used to train the drug resistance prediction model to be trained according to all FLAIR sequence diagrams and T2W sequence diagrams to obtain an initial drug resistance prediction model;
[0163] An evaluation unit, used for evaluating the initial drug resistance prediction model;
[0164] The determination unit is used to determine the initial drug resistance prediction model as the drug resistance prediction model after passing the evaluation.
[0165] Preferably, the evaluation unit comprises:
[0166] An evaluation subunit, used for evaluating the initial drug resistance prediction model by cross-validation;
[0167] Correspondingly, the prediction module 400 includes:
[0168] a drug resistance value obtaining unit, used to predict the image to be predicted using the drug resistance prediction model to obtain a plurality of predicted values of drug resistance results, wherein the number of predicted values of drug resistance results corresponds to the fold number of the cross-validation method;
[0169] A comparison unit, used for comparing the average value of the predicted value of the drug resistance result with a preset drug resistance threshold value to obtain a comparison result;
[0170] The prediction result obtaining unit is used to determine the prediction result according to the comparison result.
[0171] Preferably, the model training module 300 includes:
[0172] The FLAIR training feature image acquisition unit is used to input the FLAIR sequence image into the first three-dimensional convolutional neural network, and perform convolution, pooling, and full connection on the FLAIR sequence image in sequence to obtain a first feature vector;
[0173] A T2W training feature image acquisition unit is used to input the T2W sequence image into the second three-dimensional convolutional neural network, and perform convolution, pooling, and full connection on the T2W sequence image in sequence to obtain a second feature vector;
[0174] The second training unit is used for performing feature concatenation on the first eigenvector and the second eigenvector, and outputting the concatenated feature data to a fully connected network until the training of the drug resistance prediction model to be trained by all FLAIR sequence images and T2W sequence images is completed to obtain the drug resistance prediction model.
[0175] Preferably, it also includes:
[0176] A new image training set acquisition module, used to acquire a new MRI image training set;
[0177] An iterative training module is used to train the drug resistance prediction model according to the new MRI image training set to obtain a new drug resistance prediction model;
[0178] A new prediction module, used to predict new MRI images to be predicted using a new drug resistance prediction model;
[0179] The stop module is used to execute the step of acquiring a new MRI image training set until the number of iterations reaches a set threshold or the prediction accuracy reaches a set value.
[0180] An electronic device provided in an embodiment of the present application is introduced below. The electronic device described below and the intelligent prediction method for epilepsy drug resistance described above can refer to each other.
[0181] An electronic device is provided in an embodiment of the present application, such as Figure 8 As shown, Figure 8 The electronic device 300 shown includes: a processor 301 and a memory 303. The processor 301 and the memory 303 are connected, such as through a bus 302. Optionally, the electronic device 300 may also include a transceiver 304. It should be noted that in actual applications, the transceiver 304 is not limited to one, and the structure of the electronic device 300 does not constitute a limitation on the embodiments of the present application.
[0182] The processor 301 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. It may implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of this application. The processor 301 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.
[0183] The bus 302 may include a path to transmit information between the above components. The bus 302 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The bus 302 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 8 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.
[0184] The memory 303 may be a ROM (Read Only Memory) or other types of static storage devices that can store static information and instructions, a RAM (Random Access Memory) or other types of dynamic storage devices that can store information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory) or other optical disk storage, optical disk storage (including compressed optical disk, laser disk, optical disk, digital versatile disk, Blu-ray disk, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto.
[0185] The memory 303 is used to store the application code for executing the solution of the present application, and the execution is controlled by the processor 301. The processor 301 is used to execute the application code stored in the memory 303 to implement the contents shown in the above method embodiment.
[0186] Among them, electronic devices include but are not limited to: mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 8 The electronic device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.
[0187] A computer-readable storage medium provided in an embodiment of the present application is introduced below. The computer-readable storage medium described below and the method described above can be referenced to each other.
[0188] An embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned intelligent prediction method for epilepsy drug resistance are implemented.
[0189] Since the embodiments of the computer-readable storage medium part correspond to the embodiments of the method part, please refer to the description of the embodiments of the method part for the embodiments of the computer-readable storage medium part, which will not be repeated here.
[0190] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.
[0191] The above are only some implementation methods of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.
Claims
1. An intelligent prediction method for epilepsy drug resistance, characterized in that: include: Obtain an MRI image training set of epilepsy drug resistance, and extract FLAIR sequence diagrams and T2W sequence diagrams from each MRI image in the MRI image training set; wherein the MRI image training set includes control group images and uncontrolled group images, the control group images are images taken before treatment of control group patients who have not had clinical epileptic seizures within at least one year after receiving anti-epileptic drug treatment, and the uncontrolled group images are images taken before treatment of uncontrolled group patients who have had one epileptic seizure or died within at least one year after receiving anti-epileptic drug treatment; A drug resistance prediction model to be trained is established according to a two-way three-dimensional convolutional neural network and a fully connected network; wherein the two-way three-dimensional convolutional neural networks are used to extract feature vectors of the FLAIR sequence diagram and the T2W sequence diagram, respectively, and feature splicing is performed on the two feature vectors; Using all of the FLAIR sequence diagrams and the T2W sequence diagrams to train the drug resistance prediction model to be trained, to obtain a drug resistance prediction model; An MRI image to be predicted is acquired, and a prediction result of epilepsy drug resistance is obtained according to the MRI image to be predicted and the drug resistance prediction model.
2. The intelligent prediction method for epilepsy drug resistance according to claim 1, characterized in that: The method for obtaining an MRI image training set for epilepsy drug resistance comprises: Acquire initial control group images and initial uncontrolled group images; Performing affine transformation on the initial control group image and the initial uncontrolled group image to obtain an expanded control group image and an expanded uncontrolled group image; The initial control group images and the expanded control group images are determined as the control group images, and the initial uncontrolled group images and the expanded uncontrolled group images are determined as the uncontrolled group images, so as to obtain the MRI image training set including the control group images and the uncontrolled group images.
3. The intelligent prediction method for epilepsy drug resistance according to claim 2, characterized in that: The step of performing affine transformation on the initial control group image and the initial uncontrolled group image to obtain an expanded control group image and an expanded uncontrolled group image includes: removing interference region images from the initial control group image and the initial uncontrolled group image; Affine transformation is performed on the initial control group image and the initial uncontrolled group image after removing the interference area image to obtain the expanded control group image and the expanded uncontrolled group image.
4. The intelligent prediction method for epilepsy drug resistance according to claim 1, characterized in that: The method of training the drug resistance prediction model to be trained by using all the FLAIR sequence diagrams and the T2W sequence diagrams to obtain the drug resistance prediction model comprises: Training the drug resistance prediction model to be trained according to all the FLAIR sequence diagrams and the T2W sequence diagrams to obtain an initial drug resistance prediction model; evaluating the initial drug resistance prediction model; After passing the evaluation, the initial drug resistance prediction model is determined to be the drug resistance prediction model.
5. The intelligent prediction method for epilepsy drug resistance according to claim 4, characterized in that: The evaluating of the initial drug resistance prediction model comprises: Evaluating the initial drug resistance prediction model using a cross-validation approach; Correspondingly, the prediction result of epilepsy drug resistance is obtained according to the MRI image to be predicted and the drug resistance prediction model, including: Using the drug resistance prediction model to predict the MRI image to be predicted, and obtaining a plurality of predicted values of drug resistance results, wherein the number of predicted values of drug resistance results corresponds to the fold number of the cross-validation method; Comparing the average value of the predicted value of the drug resistance result with a preset drug resistance threshold to obtain a comparison result; The prediction result is determined according to the comparison result.
6. The intelligent prediction method for epilepsy drug resistance according to claim 1, characterized in that: The method of training the drug resistance prediction model to be trained by using all the FLAIR sequence diagrams and the T2W sequence diagrams to obtain the drug resistance prediction model comprises: Inputting the FLAIR sequence image into a first three-dimensional convolutional neural network, performing convolution, pooling, and full connection on the FLAIR sequence image in sequence to obtain a first feature vector; Inputting the T2W sequence graph into a second three-dimensional convolutional neural network, performing convolution, pooling, and full connection on the T2W sequence graph in sequence to obtain a second feature vector; The first feature vector and the second feature vector are feature spliced, and the spliced feature data is output to the fully connected network until the training of the drug resistance prediction model to be trained by all the FLAIR sequence images and the T2W sequence images is completed to obtain the drug resistance prediction model.
7. The intelligent prediction method for epilepsy drug resistance according to any one of claims 1 to 6, characterized in that: Also includes: Obtain a new MRI image training set; Training the drug resistance prediction model according to the new MRI image training set to obtain a new drug resistance prediction model; Use the new drug resistance prediction model to predict new MRI images to be predicted; The step of acquiring a new MRI image training set is performed until the number of iterations reaches a set threshold or the prediction accuracy reaches a set value.
8. An intelligent prediction device for epilepsy drug resistance, characterized in that: include: An image acquisition module, used to acquire an MRI image training set of epilepsy drug resistance, and extract FLAIR sequence diagrams and T2W sequence diagrams from each MRI image in the MRI image training set; wherein the MRI image training set includes control group images and uncontrolled group images, wherein the control group images are images taken before treatment of control group patients who have not had clinical epileptic seizures within at least one year after receiving anti-epileptic drug treatment, and the uncontrolled group images are images taken before treatment of uncontrolled group patients who have had one epileptic seizure or died within at least one year after receiving anti-epileptic drug treatment; A model building module, used to establish a drug resistance prediction model to be trained based on two three-dimensional convolutional neural networks and a fully connected network; wherein the two three-dimensional convolutional neural networks are used to extract feature vectors of the FLAIR sequence diagram and the T2W sequence diagram, respectively, and feature splicing is performed on the two feature vectors; A model training module, used for training the drug resistance prediction model to be trained by using all the FLAIR sequence diagrams and the T2W sequence diagrams to obtain a drug resistance prediction model; The prediction module is used to obtain the MRI image to be predicted, and obtain the prediction result of epilepsy drug resistance according to the MRI image to be predicted and the drug resistance prediction model.
9. An electronic device, characterized in that: include: One or more processors; Memory; One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, and the one or more programs are configured to: execute the intelligent prediction method for epilepsy drug resistance according to any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that: A computer program is stored on a computer-readable storage medium, and when the computer program is executed by a processor, the steps of the intelligent prediction method for epilepsy drug resistance according to any one of claims 1 to 7 are implemented.