WO2023168728A1 - 基于多模态影像组学的癫痫药物治疗结局预测方法和装置 - Google Patents

基于多模态影像组学的癫痫药物治疗结局预测方法和装置 Download PDF

Info

Publication number
WO2023168728A1
WO2023168728A1 PCT/CN2022/080826 CN2022080826W WO2023168728A1 WO 2023168728 A1 WO2023168728 A1 WO 2023168728A1 CN 2022080826 W CN2022080826 W CN 2022080826W WO 2023168728 A1 WO2023168728 A1 WO 2023168728A1
Authority
WO
WIPO (PCT)
Prior art keywords
drug treatment
radiomics
features
outcome
magnetic resonance
Prior art date
Application number
PCT/CN2022/080826
Other languages
English (en)
French (fr)
Inventor
蒋典
王海峰
梁栋
Original Assignee
中国科学院深圳先进技术研究院
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 中国科学院深圳先进技术研究院 filed Critical 中国科学院深圳先进技术研究院
Publication of WO2023168728A1 publication Critical patent/WO2023168728A1/zh

Links

Images

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/48—Other medical applications
    • A61B5/4848—Monitoring or testing the effects of treatment, e.g. of medication
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/0033—Features or image-related aspects of imaging apparatus classified in A61B5/00, e.g. for MRI, optical tomography or impedance tomography apparatus; arrangements of imaging apparatus in a room
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/0033—Features or image-related aspects of imaging apparatus classified in A61B5/00, e.g. for MRI, optical tomography or impedance tomography apparatus; arrangements of imaging apparatus in a room
    • A61B5/004—Features or image-related aspects of imaging apparatus classified in A61B5/00, e.g. for MRI, optical tomography or impedance tomography apparatus; arrangements of imaging apparatus in a room adapted for image acquisition of a particular organ or body part
    • A61B5/0042—Features or image-related aspects of imaging apparatus classified in A61B5/00, e.g. for MRI, optical tomography or impedance tomography apparatus; arrangements of imaging apparatus in a room adapted for image acquisition of a particular organ or body part for the brain
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/05—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves 
    • A61B5/055—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves  involving electronic [EMR] or nuclear [NMR] magnetic resonance, e.g. magnetic resonance imaging
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/40—Detecting, measuring or recording for evaluating the nervous system
    • A61B5/4076—Diagnosing or monitoring particular conditions of the nervous system
    • A61B5/4094—Diagnosing or monitoring seizure diseases, e.g. epilepsy
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
    • A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7271—Specific aspects of physiological measurement analysis
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7271—Specific aspects of physiological measurement analysis
    • A61B5/7275—Determining trends in physiological measurement data; Predicting development of a medical condition based on physiological measurements, e.g. determining a risk factor

Definitions

  • the present invention relates to the field of computer-aided diagnosis technology, and specifically relates to a method and device for predicting the outcome of epilepsy drug treatment based on multi-modal radiomics, and terminal equipment.
  • Tuberous sclerosis complex is a rare autosomal dominant disease caused by loss-of-function mutations in the TSC1 or TSC2mTOR pathway genes.
  • TSC is a neuropsychiatric disorder affecting the brain, skin, heart, lungs, kidneys and epilepsy.
  • Epilepsy is the most common and challenging symptom among patients with TSC, affecting approximately 85% of patients, and nearly two-thirds of these patients experience a first seizure around the age of one.
  • early epilepsy treatment can prevent or control epileptic seizures, improve the cognitive neurodevelopment of TSC patients, and improve the patient's quality of life.
  • Neurologic manifestations can be observed on brain imaging in nearly all TSC patients, and magnetic resonance imaging (MRI), with its rich soft tissue contrast, is an advanced imaging tool for clinical diagnosis of TSC.
  • Cortical nodules and subcortical nodules are the main TSC-related brain lesions, and abnormal high or low signals can be clearly observed in the fluid-attenuated inversion recovery (FLAIR) sequence and T2W sequence of MRI.
  • FLAIR fluid-attenuated inversion recovery
  • T1W can observe the patient's brain structure and other lesions.
  • T2W and FLAIR provide higher contrast of lesions and brain.
  • the main way to diagnose the results of epilepsy drug treatment in clinical practice is: after TSC is diagnosed, the patient is treated with anti-epileptic drugs (AEDs), and the doctor follows up to observe whether the patient still has epileptic seizures within a year to determine the outcome of drug treatment. If Patients who still have epileptic seizures within one year are classified as drug-refractory patients, otherwise they are drug-controlled patients. Patients who are refractory to medication require alternative treatments, such as surgery.
  • This method has the following disadvantages: high labor cost. Due to the particularity of drug treatment, professionally trained doctors are required to treat patients; high time cost, and the outcome of drug treatment generally requires more than one year of treatment. Only then can we know whether the patient is drug-resistant, and then change the treatment plan for the drug-resistant patient. This is very detrimental to the patient's treatment and may delay the patient's optimal treatment time.
  • studying an intelligent model to predict the outcome of epilepsy drug treatment to distinguish between drug-controlled epilepsy and uncontrolled (drug-refractory) TSC epilepsy patients can assist doctors in formulating targeted treatment plans for the two types of patients. It is of great significance to reduce the patient's mortality and improve the patient's quality of life.
  • the present invention provides a method and device for predicting the outcome of epilepsy drug treatment based on multi-modal imaging to solve the problem of how to predict the outcome of epilepsy drug treatment for TSC epilepsy patients, and can quickly distinguish drug treatment control Type and uncontrolled (drug-refractory) TSC epilepsy patients.
  • one aspect of the present invention is to provide a method for predicting the outcome of epilepsy drug treatment based on multi-modal radiomics, which includes the steps:
  • TSC patients are randomly divided into a training set and a test set in proportion, the training set is used to train the prediction model, and the test set is used to verify the performance of the prediction model;
  • the multiple modal magnetic resonance images include T1-weighted images, T2-weighted images and fluid-attenuated inversion recovery images among magnetic resonance images.
  • the inclusion criteria for TSC patients need to meet at least the following three conditions: 1) The patient has undergone multiple modal magnetic resonance imaging scans before being treated with anti-epileptic drugs; 2) The patient has been treated with anti-epileptic drugs for 1 year Above; 3) The patient did not undergo lesion resection surgery.
  • the ratio of the number of the training set and the test set is 8:2 or 7:3.
  • the segmentation contours of the images are manually inspected and modified to obtain the corresponding images of each modality. area of interest.
  • the high-dimensional radiomics features include at least the following three types of features: 1) three-dimensional morphological features used to describe the size and shape of nodule lesions; 2) first-order statistical features used to describe the intensity distribution of the lesion area; 3 ) Texture features used to describe the spatial distribution information of the lesion area, including gray level co-occurrence matrix, gray level run length matrix, gray level area size matrix, adjacent gray level difference matrix and gray level co-occurrence matrix.
  • the steps of analyzing and screening high-dimensional radiomics features and obtaining the target radiomics features after dimensionality reduction include: first, using a bivariate analysis algorithm to preliminary screen the high-dimensional radiomics features: calculating each The p-value of the Spearman correlation coefficient between the omics features and the treatment outcome of drug treatment was screened out to select the omics features with a p-value ⁇ 0.05 to obtain the preliminary radiomics features; then the lasso algorithm was used to analyze the preliminary radiomics features.
  • Features are further screened to select radiomics features that are significantly related to the classification of lesions, and the target radiomics features are obtained.
  • the steps of using a machine learning algorithm to train a prediction model for the target radiomic features in the training set, constructing a prediction model for epilepsy drug treatment outcome, and validating the model in the test set include: using multiple machine learning algorithms to build models respectively , obtain multiple types of prediction models; the various machine learning algorithms include support vector machine algorithm, random forest algorithm, logistic regression analysis algorithm, Ada Boost algorithm, Gradient Boosting algorithm and Decision Tree algorithm; for each type of prediction model , using ten-fold cross-validation for training.
  • the training process uses a grid search algorithm to select the optimal hyperparameters; based on the determined optimal hyperparameters, various types of prediction models are trained on the entire training set, and the candidate prediction models are obtained after training. Test and verify on the test set; perform performance evaluation of the candidate prediction model based on the performance parameters of AUC, accuracy, sensitivity and specificity, and select the candidate prediction model with the best performance as the epilepsy drug treatment outcome prediction model.
  • another aspect of the present invention is to provide a device for predicting the outcome of epilepsy drug treatment based on multi-modal radiomics, which includes:
  • the image acquisition module is used to acquire multiple modal magnetic resonance images of TSC patients before anti-epileptic drug treatment, and preprocess the multiple modal magnetic resonance images;
  • a grouping module used to randomly divide TSC patients into a training set and a test set in proportion, the training set is used to train the prediction model, and the test set is used to verify the performance of the prediction model;
  • the image segmentation module is used to perform regional segmentation on preprocessed magnetic resonance images of each modality based on the U-net++ network, and obtain the regions of interest corresponding to each modal magnetic resonance image;
  • the feature extraction module is used to extract features from each region of interest in each modality magnetic resonance image and obtain the high-dimensional radiomics features corresponding to each region of interest;
  • a feature screening module is used to analyze and screen the high-dimensional radiomics features, and obtain the target radiomics features after dimensionality reduction;
  • the model building module is used to use machine learning algorithms to train the prediction model on the target radiomic features in the training set, build a prediction model for the outcome of epilepsy drug treatment, and validate the model in the test set;
  • the treatment outcome prediction module is used to predict the target radiomics characteristics of patients to be treated by constructing a prediction model to obtain the outcome of epilepsy drug treatment, and obtain the predicted outcome of epilepsy drug treatment.
  • the present invention also provides a terminal device, which includes a memory, a processor and a computer program stored in the memory and executable on the processor.
  • the feature is that when the processor executes the program, The steps of the method for predicting the outcome of epilepsy drug treatment based on multimodal radiomics as described above.
  • the method and device for predicting the outcome of epilepsy drug treatment provided by the embodiments of the present invention accurately segment multiple modal magnetic resonance images based on the U-net++ network, and then extract the radiomics features of the area of interest.
  • Feature-based prediction models can quickly and effectively predict the outcome of drug treatment for TSC epilepsy patients.
  • MRI images can be used to predict whether patients are drug-resistant before drug treatment begins. This can assist doctors in making more accurate clinical decisions and buy time for patients. With more appropriate treatment, patients do not need drug resistance testing for more than a year, which can greatly reduce doctors’ labor costs and patients’ time costs.
  • Figure 1 is a flow diagram of a method for predicting the outcome of epilepsy drug treatment in an embodiment of the present invention
  • FIG. 2 is a structural diagram of the U-net++ network in the embodiment of the present invention.
  • Figure 3 is a structural diagram of an epilepsy drug treatment outcome prediction device in an embodiment of the present invention.
  • Figure 4 is a structural diagram of a terminal device in an embodiment of the present invention.
  • Figure 1 is a flow chart of a method for predicting the outcome of epilepsy drug treatment based on multi-modal radiomics in an embodiment of the present invention.
  • the epilepsy drug treatment outcome prediction method based on multimodal radiomics of the present application is applied to a terminal device, wherein the terminal device can be a server or a mobile device, or it can also be a mutual interaction between the server and the mobile device.
  • the terminal device can be a server or a mobile device, or it can also be a mutual interaction between the server and the mobile device.
  • a coordinated system e.g., various parts included in the terminal device, such as each unit, sub-unit, module, and sub-module, can all be set up in the server, or all can be set up in the mobile device, or can be set up in the server and the mobile device respectively.
  • the terminal device is, for example, a computer device.
  • the above-mentioned server may be hardware or software.
  • the server When the server is hardware, it can be implemented as a distributed server cluster composed of multiple servers or as a single server.
  • the server is software, it can be implemented as multiple software or software modules, such as software or software modules used to provide distributed servers, or it can be implemented as a single software or software module.
  • this embodiment provides a method for predicting the outcome of epilepsy drug treatment based on multi-modal radiomics, which mainly includes constructing a prediction model for the outcome of epilepsy drug treatment and using the construction to obtain the prediction model for the outcome of epilepsy drug treatment for patients to be treated. Predict two parts.
  • the part of building a prediction model for epilepsy drug treatment outcomes includes the following steps:
  • Step S1 Obtain multi-modal magnetic resonance images of TSC patients before anti-epileptic drug treatment, and preprocess the multi-modal magnetic resonance images.
  • the multiple modal magnetic resonance images include T1-weighted images (T1W), T2-weighted images (T2W) and fluid-attenuated inversion recovery images (FLAIR) in magnetic resonance images.
  • the enrollment criteria for TSC patients need to meet at least the following three conditions: 1) The patient has undergone multiple modal magnetic resonance imaging scans (including at least T1W, T2W) before treatment with anti-epileptic drugs and FLAIR three sequences); 2) The patient has been treated with anti-epileptic drugs for more than 1 year; 3) The patient has not undergone lesion resection surgery.
  • the treatment principles are determined by experienced tuberous sclerosis experts based on experience and guidelines.
  • epilepsy drug treatment outcomes of the enrolled TSC patients were labeled and divided into a control group and an uncontrolled group.
  • Epilepsy pharmacotherapy outcomes were defined according to the 1981 ILAE classification: patients were considered controls if they were clinically seizure-free within 1 year of AED treatment; patients were considered controls if they had at least one seizure or died within 1 year. Treat this as an uncontrolled group.
  • the specific preprocessing of the multiple modal magnetic resonance images is: using the deep learning tool HD-bet (or preprocessing tools such as FSL and SPM) to remove the differences in multi-contrast MRI (T1W, T2W and FLAIR) images.
  • the lesion is unrelated to the skull.
  • Step S2 Randomly divide TSC patients into a training set and a test set in proportion.
  • the training set is used to train the prediction model
  • the test set is used to verify the performance of the prediction model.
  • the ratio of the number of the training set and the test set is preferably 8:2 or 7:3.
  • the T1W, T2W and FLAIR data of 300 TSC patients from a children's hospital were enrolled as a data set, including 240 patients in the training set and 60 patients in the test set. That is, the ratio of training set to test set is 8:2.
  • Step S3 Perform region segmentation on the preprocessed magnetic resonance image of each modality based on the U-net++ network to obtain the region of interest corresponding to each modality of the magnetic resonance image.
  • the lesions of tuberous sclerosis are defined as cortical and subcortical nodules.
  • the U-net++ network is used to segment the nodular lesions of tuberous sclerosis and obtain the corresponding regions of interest.
  • the structure of the U-net++ network is shown in Figure 2.
  • the network structure is an end-to-end structure, that is, the original two-dimensional multi-contrast image data is input and the segmented nodules are output.
  • U-net++ is a fully convolutional neural network without fully connected layers. It still has good segmentation results when the data set is small.
  • the U-net++ network's image segmentation process mainly involves three steps: upsampling, downsampling and feature splicing.
  • the first layer of downsampling network is: the input image size is (512, 512), the image is convolved and activated twice, the convolution kernel size is (3, 3), and the activation function is relu function:
  • the pooling kernel size is set to (2, 2), and the step size is set to 2.
  • the feature map size is halved, that is, the downsampling operation is completed.
  • the network structure of the remaining layers of downsampling is the same as that of the first layer. The only difference is the number of convolution kernels.
  • the number of convolution kernels in each layer is 64, 128, 256, and 512 respectively.
  • the purpose of upsampling is to restore and decode the abstract features to the size of the original image, and finally obtain the segmentation result.
  • the convolution kernel size is set to (2, 2) and the step size is set to 2, that is, the image feature size is doubled after upsampling.
  • the feature splicing operation is to splice the previous feature map to the back of the current feature map, that is, the concatenate operation, and then perform two convolutions and activations.
  • the convolution kernel size is (3, 3)
  • the activation function is relu
  • padding is same
  • Output the image of (512, 512).
  • the SGD algorithm stocastic gradient descent algorithm
  • a 1 ⁇ 1 convolution kernel is added after the feature map. This convolution is activated using the sigmoid function:
  • the loss function uses the cross-entropy loss function:
  • the segmentation contours of the images are then manually inspected and modified, and finally Obtain the region of interest corresponding to each modal magnetic resonance image. That is, in the preferred embodiment of the present invention, a semi-automatic segmentation method is used to segment each modality magnetic resonance image.
  • Step S4 Perform feature extraction on each region of interest in each modality magnetic resonance image to obtain high-dimensional radiomic features corresponding to each region of interest.
  • pyradiomics is an open source python software package that can be used as a feature extraction tool for radiomics feature extraction of medical images.
  • Feature extraction can be performed by specifying image categories, feature categories, and custom feature extraction parameters for specific filters.
  • the high-dimensional radiomics features extracted in step S4 include at least the following three types of features: 1) three-dimensional morphological features used to describe the size and shape of nodule lesions; 2) first-order features used to describe the intensity distribution of the lesion area Statistical features; 3) Texture features used to describe the spatial distribution information of the lesion area, including gray level co-occurrence matrix (GLCM), gray level run length matrix (GLRLM), gray level area size matrix (GLSZM), adjacent gray level difference matrix ( NGTDM) and gray level co-occurrence matrix (GLDM) and other features.
  • GLCM gray level co-occurrence matrix
  • GLRLM gray level run length matrix
  • GLSZM gray level area size matrix
  • NGTDM adjacent gray level difference matrix
  • GLDM gray level co-occurrence matrix
  • Step S5 Analyze and screen the high-dimensional radiomics features, and obtain the target radiomics features after dimensionality reduction.
  • omics features extracted from one modality in step S4 above there are thousands of omics features extracted from one modality in step S4 above, and the number of features for the three modalities may reach thousands.
  • step S5 includes:
  • the Lasso algorithm is used to further screen the preliminary radiomics features, screen out the radiomics features that are significantly related to the classification of lesions, and obtain the target radiomics features.
  • Step S6 Use a machine learning algorithm to train a prediction model on the target radiomic features in the training set, build a prediction model for the outcome of epilepsy drug treatment, and verify the model in the test set.
  • step S6 includes the following sub-steps:
  • the multiple machine learning algorithms include support vector machine algorithm, random forest algorithm, logistic regression analysis algorithm, Ada Boost algorithm, Gradient Boosting algorithm and Decision Tree algorithm.
  • ROC receiver operating curve
  • Optimal model parameters based on the determined optimal hyperparameters, train various types of prediction models on the entire training set, train the candidate prediction models, and test and verify on the test set.
  • the embodiment of the present invention constructs and obtains a multimodal radiomics-based epilepsy drug treatment outcome prediction model.
  • the part of predicting patients to be treated includes the following steps:
  • Step S7 Use the constructed prediction model to obtain the outcome of epilepsy drug treatment to predict the target radiomics characteristics of the patient to be treated, and obtain the predicted outcome of epilepsy drug treatment.
  • the doctor collects T1W, T2W and FLAIR sequence images of TSC patients, extracts the target radiomic features corresponding to the three modalities and inputs them into the epilepsy drug treatment outcome prediction model for prediction, and obtains the predicted epilepsy drug treatment outcome.
  • the device 100 includes an image acquisition module 1, a grouping module 2, an image segmentation module 3, a feature extraction module 4, Feature screening module 5, model building module 6 and treatment outcome prediction module 7. in,
  • the image acquisition module is used to acquire multiple modal magnetic resonance images of TSC patients before anti-epileptic drug treatment, and preprocess the multiple modal magnetic resonance images; that is, corresponding to step S1 in Embodiment 1 work process.
  • the grouping module is used to randomly divide TSC patients into training sets and test sets in proportion, the training set is used to train the prediction model, and the test set is used to verify the performance of the prediction model; that is, step S2 in Embodiment 1 corresponds to working process.
  • the image segmentation module is used to perform regional segmentation on the preprocessed magnetic resonance images of each modality based on the U-net++ network, and obtain the regions of interest corresponding to the magnetic resonance images of each modality; that is, the steps in Embodiment 1 The working process corresponding to S3.
  • the feature extraction module is used to extract features from each region of interest in each modality magnetic resonance image, and obtain high-dimensional radiomic features corresponding to each region of interest; that is, corresponding to step S4 in Embodiment 1 work process.
  • the feature screening module is used to analyze and screen the high-dimensional radiomics features, and obtain the target radiomics features after dimensionality reduction; that is, the working process corresponding to step S5 in Embodiment 1.
  • the model building module is used to use a machine learning algorithm to train a prediction model on the target radiomic features in the training set, construct a prediction model for epilepsy drug treatment outcome, and verify the model in the test set; that is, the work corresponding to step S6 in Embodiment 1 process.
  • the treatment outcome prediction module is used to predict the target radiomic features of the patient to be treated by constructing an epilepsy drug treatment outcome prediction model and obtain the predicted epilepsy drug treatment outcome; that is, the working process corresponding to step S7 in Embodiment 1.
  • this embodiment provides a terminal device, as shown in Figure 4.
  • the terminal device includes: a processor 10, a memory 20, The input device 30 and the output device 40 are provided with a GPU in the processor 10.
  • the number of the processors 10 may be one or more. In FIG. 2, one processor 10 is taken as an example.
  • the processor 10, memory 20, input device 30 and output device 40 in the terminal device may be connected through a bus or other means.
  • the memory 20 serves as a computer-readable storage medium and can be used to store software programs, computer executable programs and modules.
  • the processor 10 executes software programs, instructions and modules stored in the memory 20 to execute various functional applications and data processing of the device, that is, to implement the multi-modal imaging-based epilepsy described in the previous embodiments of the present invention. Steps in a method for predicting drug treatment outcomes.
  • the input device 30 is used to receive image data, input numeric or character information, and generate key signal input related to user settings and function control of the device.
  • the output device 40 may include a display device such as a display screen, for example, used to display images.
  • the method and device for predicting the outcome of epilepsy drug treatment and the corresponding terminal equipment provided in the above embodiments of the present invention perform accurate regional segmentation of multiple modal magnetic resonance images based on the U-net++ network, and then extract the radiomics of the area of interest.
  • a prediction model is established based on radiomics features, which can quickly and effectively predict the drug treatment outcomes of TSC epilepsy patients.
  • MRI images can be used to predict whether patients are drug-resistant before drug treatment begins, which can assist doctors in making more accurate clinical decisions. , giving patients time for more appropriate treatment, and patients do not need drug resistance testing for more than a year, which can greatly reduce doctors’ labor costs and patients’ time costs.

Landscapes

  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Veterinary Medicine (AREA)
  • Public Health (AREA)
  • Animal Behavior & Ethology (AREA)
  • Biophysics (AREA)
  • Pathology (AREA)
  • Biomedical Technology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Medical Informatics (AREA)
  • Surgery (AREA)
  • Artificial Intelligence (AREA)
  • Physiology (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Signal Processing (AREA)
  • Psychiatry (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Neurology (AREA)
  • Evolutionary Computation (AREA)
  • Fuzzy Systems (AREA)
  • Mathematical Physics (AREA)
  • Neurosurgery (AREA)
  • High Energy & Nuclear Physics (AREA)
  • Magnetic Resonance Imaging Apparatus (AREA)

Abstract

一种癫痫药物治疗结局预测方法和装置，包括：获取TSC患者在进行抗癫痫药物治疗前的多种模态磁共振影像(S1)；把TSC患者按比例随机分为训练集和测试集(S2)；基于U-net++网络对每一种模态磁共振影像进行区域分割，获取感兴趣区域(S3)；对每一个感兴趣区域进行特征提取，获取高维影像组学特征(S4)；对高维影像组学特征进行分析筛选得到目标影像组学特征(S5)；对训练集中的目标影像组学特征运用机器学习算法训练预测模型，构建获得癫痫药物治疗结局预测模型并在验证模型(S6)；利用构建获得癫痫药物治疗结局预测模型对待治疗患者的目标影像组学特征进行预测，获得预测的癫痫药物治疗结局(S7)。该方法能够快速、有效地预测癫痫患者的药物治疗结局，辅助医生制定更好的治疗方案。

Description

基于多模态影像组学的癫痫药物治疗结局预测方法和装置 技术领域
本发明涉及计算机辅助诊断技术领域，具体涉及一种基于多模态影像组学的癫痫药物治疗结局预测方法和装置、终端设备。
背景技术
结节性硬化症(TSC)是一种罕见的常染色体显性遗传病，由TSC1或TSC2mTOR通路基因功能丧失突变引起。TSC是一种影响脑、皮肤、心脏、肺、肾脏和癫痫的神经精神疾病。癫痫是TSC患者中最普遍和最具挑战性的症状，影响了大约85％的患者，而且其中将近三分之二的患者在一岁左右会伴随着癫痫首次发作。在TSC诊断后，患者尽早进行癫痫治疗可以预防或控制癫痫发作，改善TSC患者的认知神经发育并提高患者的生活质量。
在所有TSC患者中，几乎都可以在脑部影像上观察到神经系统表现，而磁共振成像(MRI)具有丰富的软组织对比度，是用于临床诊断TSC的先进的成像工具。皮质结节和皮质下结节是主要的TSC相关的脑部病灶，在MRI的液体衰减反转恢复(FLAIR)序列中和T2W序列中能够清晰观察到异常高或低信号。T1W能够观察患者的脑部结构和其他病变、T2W和FLAIR提供了较高的病变和大脑对比度，这三个序列通常包含在常规TSC诊断的MRI扫描方案中。在MRI上不仅可以清晰的观察到脑部病灶，也能够观察患者的脑部结构，是目前临床上认可的成像方法。
目前临床上癫痫药物治疗结果诊断的主要途径是：确诊TSC后，对患者进行抗癫痫药物(AEDs)治疗，由医生随访观察患者是否一年之内仍有癫痫发作，来判断药物治疗结局，如果一年之内仍有癫痫发作，则为药物难治型患者，否则为药物控制型患者。药物难治型患者需要更换治疗方法，比如手术。这种方式有以下几个缺点：人力成本高，由于药物治疗的特殊性，所以需要经过专业训练的医师才能够对患者进行药物治疗；时间成本高，药物治疗结局一般需要一年以上的治疗，才能知道患者是否耐药，然后再对耐药患者更换治疗方案，这对患者治疗很不利，可能会耽误患者的最佳治疗时间。
因此，研究一种预测癫痫药物治疗结局的智能模型来区分药物治疗控制型癫痫和未控制型(药物难治)的TSC癫痫患者，能够辅助医生对两种类型的患者制定针对性的治疗方案，降低患者的死亡率以及提高患者的生活质量，具有重要的意义。
发明内容
有鉴于此，本发明提供了一种基于多模态影像组学的癫痫药物治疗结局预测方法和装置，以解决如何对TSC癫痫患者的癫痫药物治疗结局进行预测的问题，能够快速区分药物治疗控制型和未控制型(药物难治)的TSC癫痫患者。
为了解决上述技术问题，本发明的一方面是提供一种基于多模态影像组学的癫痫药物治疗结局预测方法，其包括步骤：
获取TSC患者在进行抗癫痫药物治疗前的多种模态磁共振影像，并对所述多种模态磁共振影像进行预处理；
把TSC患者按比例随机分为训练集和测试集，所述训练集用于训练预测模型，所述测试集用于验证预测模型的性能；
基于U-net++网络对预处理后的每一种模态磁共振影像进行区域分割，获取每一种模态磁共振影像对应的感兴趣区域；
对每一种模态磁共振影像的每一个感兴趣区域进行特征提取，获取每一个感兴趣区域对应的高维影像组学特征；
对所述高维影像组学特征进行分析筛选，降维后得到目标影像组学特征；
对训练集中的目标影像组学特征运用机器学习算法训练预测模型，构建获得癫痫药物治疗结局预测模型，并在测试集中验证模型；
利用构建获得癫痫药物治疗结局预测模型对待治疗患者的目标影像组学特征进行预测，获得预测的癫痫药物治疗结局。
具体地，所述多种模态磁共振影像包括磁共振影像中的T1加权影像、T2加权影像和液体衰减反转恢复影像。
具体地，所述TSC患者的入组标准需要至少满足以下3个条件：1)患者在使用抗癫痫药物治疗前进行了多种模态磁共振影像扫描；2)患者使用抗癫痫药物治疗1年以上；3)患者未进行病灶切除手术。
具体地，所述训练集和所述测试集的数量比例为8:2或7:3。
具体地，在基于U-net++网络对预处理后的每一种模态磁共振影像进行区域分割之后，通过人工检查并修改完善影像的分割轮廓，进而获取每一种模态磁共振影像对应的感兴趣区域。
具体地，所述高维影像组学特征至少包括以下3类特征：1)用于描述结节病灶尺寸和形状的三维形态特征；2)用于描述病灶区域强度分布的一阶统计特征；3)用于描述病灶区域空间分布信息的纹理特征，包括灰度共生矩阵、灰度游程矩阵、灰度区域大小矩阵、相邻灰度差分矩阵和灰度共生矩阵。
具体地，所述对高维影像组学特征进行分析筛选，降维后得到目标影像组学特征的步骤包括：首先采用了双变量分析算法对高维影像组学特征进行初步筛选：计算每个组学特征与药物治疗的治疗结局之间的斯皮尔曼相关系数p值，筛选出p值<0.05的组学特征，获得初步影像组学特征；然后使用套索算法对所述初步影像组学特征进行进一步筛选，筛选出与病灶分类有显著关系的影像组学特征，获得所述目标影像组学特征。
具体地，所述对训练集中的目标影像组学特征运用机器学习算法训练预测模型，构建获得癫痫药物治疗结局预测模型，并在测试集中验证模型的步骤包括：采用多种机器学习算法分别建立模型，获得多种类型的预测模型；所述多种机器学习算法包括支持向量机算法、随机森林算法、Logistic回归分析算法、Ada Boost算法、Gradient Boosting算法和Decision Tree算法；对于每一类型的预测模型，采用十折交叉验证进行训练，训练过程使用网格搜索算法选择最优超参数；根据确定的最优超参数，在整个训练集上训练各个类型的预测模型，训练得到候选预测模型，并在测试集上测试验证；基于AUC、准确性、敏感性和特异性的性能参数对候选预测模型进行性能评估，选取性能最佳的候选预测模型作为所述癫痫药物治疗结局预测模型。
为了解决上述技术问题，本发明的另一方面是提供一种基于多模态影像组学的癫痫药物治疗结局预测装置，其包括：
影像获取模块，用于获取TSC患者在进行抗癫痫药物治疗前的多种模态磁共振影像，并对所述多种模态磁共振影像进行预处理；
分组模块，用于把TSC患者按比例随机分为训练集和测试集，所述训练集用于训练预测模型，所述测试集用于验证预测模型的性能；
影像分割模块，用于基于U-net++网络对预处理后的每一种模态磁共振影像进行区域分割，获取每一种模态磁共振影像对应的感兴趣区域；
特征提取模块，用于对每一种模态磁共振影像的每一个感兴趣区域进行特征提取，获取每一个感兴趣区域对应的高维影像组学特征；
特征筛选模块，用于对所述高维影像组学特征进行分析筛选，降维后得到目标影像组学特征；
模型构建模块，用于对训练集中的目标影像组学特征运用机器学习算法训练预测模型，构建获得癫痫药物治疗结局预测模型，并在测试集中验证模型；
治疗结局预测模块，用于利用构建获得癫痫药物治疗结局预测模型对待治疗患者的目标影像组学特征进行预测，获得预测的癫痫药物治疗结局。
为了解决上述技术问题，本发明还提供一种终端设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，其特征在于，所述处理器执行所述程序时实现如上所述的基于多模态影像组学的癫痫药物治疗结局预测方法的步骤。
本发明实施例提供的癫痫药物治疗结局预测方法和装置，通过基于U-net++网络对多种模态磁共振影像进行精确的区域分割，再提取感兴趣区域的影像组学特征，根据影像组学特征建立预测模型，能够快速、有效地预测TSC癫痫患者的药物治疗结局，在药物治疗开始之前就通过MRI影像来预测患者是否耐药，能够辅助医生进行更准确的临床决策，给患者争取时间进行更合适的治疗，患者不需要长达一年以上的耐药性测试，能够极大地降低医生的人力成本和患者的时间成本。
附图说明
图1是本发明实施例中的癫痫药物治疗结局预测方法的流程图示；
图2是本发明实施例中的U-net++网络的结构图示
图3是本发明实施例中的癫痫药物治疗结局预测装置的结构图示；
图4是本发明实施例中的一种终端设备的结构图示。
具体实施方式
为使本发明的目的、技术方案和优点更加清楚，下面结合附图对本发明的具体实施方式进行详细说明。这些优选实施方式的示例在附图中进行了例示。附图中所示和根据附图描述的本发明的实施方式仅仅是示例性的，并且本发明并不限于这些实施方式。
在此，还需要说明的是，为了避免因不必要的细节而模糊了本发明，在附图中仅仅示出了与根据本发明的方案密切相关的结构和/或处理步骤，而省略了与本发明关系不大的其他细节。
图1是本发明实施例中的基于多模态影像组学的癫痫药物治疗结局预测方法的流程图示。
本申请的基于多模态影像组学的癫痫药物治疗结局预测方法应用于一种终端设备，其中，所述的终端设备可以为服务器，也可以为移动设备，还可以为由服务器和移动设备相互配合的系统。相应地，终端设备包括的各个部分，例如各个单元、子单元、模块、子模块可以全部设置于服务器中，也可以全部设置于移动设备中，还可以分别设置于服务器和移动设备中。所述终端设备例如是计算机设备。
进一步地，上述服务器可以是硬件，也可以是软件。当服务器为硬件时，可以实现成由多个服务器组成的分布式服务器集群，也可以实现成单个服务器。当服务器为软件时，可以实现成多个软件或软件模块，例如用来提供分布式服务器的软件或软件模块，也可以实现成单个软件或软件模块。
实施例1
参阅图1，本实施例提供的一种基于多模态影像组学的癫痫药物治疗结局预测方法，其中主要包括构建癫痫药物治疗结局预测模型以及利用构建获得癫痫药物治疗结局预测模型对待治疗患者进行预测两大部分。
其中，构建癫痫药物治疗结局预测模型的部分包括以下步骤：
步骤S1、获取TSC患者在进行抗癫痫药物治疗前的多种模态磁共振影像，并对所述多种模态磁共振影像进行预处理。
在本发明具体实施例中，所述多种模态磁共振影像包括磁共振影像中的T1加权影像(T1W)、T2加权影像(T2W)和液体衰减反转恢复影像(FLAIR)。
在本发明具体实施例中，所述TSC患者的入组标准需要至少满足以下3个 条件：1)患者在使用抗癫痫药物治疗前进行了多种模态磁共振影像扫描(至少包括T1W、T2W和FLAIR三个序列的影像)；2)患者使用抗癫痫药物治疗1年以上；3)患者未进行病灶切除手术。其中，治疗原则由资深的结节性硬化症专家根据经验和指南确定。
进一步地，对入组的TSC患者的癫痫药物治疗结局进行标签，分为控制组和未控制组。癫痫药物治疗结局根据1981年的ILAE分类来定义：如果患者在AED治疗的1年之内没有临床癫痫发作，则将其视为控制组；如果患者在一年内至少有一次癫痫发作或死亡，则将其视为未控制组。
其中，对所述多种模态磁共振影像进行预处理具体是：使用深度学习工具HD-bet(或者FSL、SPM等预处理工具)去除了多对比度MRI(T1W、T2W和FLAIR)影像中与病灶无关的颅骨。
步骤S2、把TSC患者按比例随机分为训练集和测试集，所述训练集用于训练预测模型，所述测试集用于验证预测模型的性能。
具体地，所述训练集和所述测试集的数量比例优选为8:2或7:3。例如，在本发明的具体实施例中，入组了某儿童医院的300个TSC患者的T1W、T2W和FLAIR的数据作为数据集，其中，训练集240个患者，测试集60个患者。即，训练集和测试集的比例为8:2。
步骤S3、基于U-net++网络对预处理后的每一种模态磁共振影像进行区域分割，获取每一种模态磁共振影像对应的感兴趣区域。
结节性硬化症(TSC)的病灶定义为皮质和皮质下结节，本发明实施例中采用U-net++网络来对结节性硬化症的结节病灶进行分割，获取对应的感兴趣区域。
U-net++网络的结构如图2所示，网络结构是端到端的结构，即原始二维多对比影像数据输入，输出分割后的结节。U-net++是一个全卷积的神经网络，无全连接层，在数据集小的时候仍然有很好的分割效果。U-net++网络的对影像的分割过程主要涉及三个步骤：上采样、下采样和特征拼接。
在具体实施例中，下采样的第一层网络为：输入图片大小为(512，512)，对图片进行两次卷积和激活，卷积核大小为(3，3)，激活函数用relu函数：
Figure PCTCN2022080826-appb-000001
然后进行最大池化，池化核大小设置为(2，2)，步长设置为2，池化后特征图大小减半，即完成下采样操作。下采样其余各层网络结构与第一层一致，唯一区别是卷积核个数，各层卷积核数依次分别为64、128、256、512。
上采样的目的就是把抽象的特征再还原解码到原图的尺寸，最终得到分割结果。仅一步操作，即为进行转置卷积，卷积核大小设置为(2，2)，步长设置为2，即上采样后图像特征大小加倍。
特征拼接操作则是将前面的特征图拼接到当前特征图后面即concatenate操作，然后进行两次卷积和激活，卷积核大小为(3，3)，激活函数为relu，padding为same，最后输出(512，512)的图像。
基于以上的操作，到此整个网络搭建完成，接下来就是训练。采用SGD算法(随机梯度下降算法)进行训练网络，在特征图后面加一个1×1的卷积核，此卷积使用sigmoid函数激活：
Figure PCTCN2022080826-appb-000002
损失函数使用交叉熵损失函数：
Figure PCTCN2022080826-appb-000003
对前面每个1×1卷积分别计算loss后求和为此次前向传播的损失值，然后在结节病灶数据集上进行训练，更新迭代优化参数。
在优选的方案中，为确保病灶分割正确，在基于U-net++网络对预处理后的每一种模态磁共振影像进行自动分割完毕之后，再通过人工检查并修改完善影像的分割轮廓，最终获取每一种模态磁共振影像对应的感兴趣区域。即，本发明优选实施例中是采用半自动分割的方式对每一种模态磁共振影像进行分割。
步骤S4、对每一种模态磁共振影像的每一个感兴趣区域进行特征提取，获取每一个感兴趣区域对应的高维影像组学特征。
具体地，由于MRI影像的层厚和像素间距等设备参数的差异，首先将所有影像都被重采样到1×1×1mm 3。然后分别对T1W和T2W以及FLAIR影像采用pyradiomics软件包提取影像组学特征。
其中，pyradiomics是一个开源的python软件包，可以作为一种特征提取的工具，用于医学图像的影像组学特征提取。可以通过指定图像类别、特征类别、特定滤波器自定义特征提取参数进行特征提取。
在步骤S4中提取获得的所述高维影像组学特征至少包括以下3类特征：1)用于描述结节病灶尺寸和形状的三维形态特征；2)用于描述病灶区域强度分布的一阶统计特征；3)用于描述病灶区域空间分布信息的纹理特征，包括灰度共生矩阵(GLCM)、灰度游程矩阵(GLRLM)、灰度区域大小矩阵(GLSZM)、相邻灰度差分矩阵(NGTDM)和灰度共生矩阵(GLDM)等特征。
步骤S5、对所述高维影像组学特征进行分析筛选，降维后得到目标影像组学特征。
具体地，如上步骤S4中从一种模态提取的组学特征就上千个，对于三种模态的特征可能数量达到几千。为了降低预测模型过拟合风险，提高模型预测性能，所以需要对高维影像组学特征进行分析筛选，降维处理。
在具体的方案，所述步骤S5包括：
首先采用了双变量分析算法对高维影像组学特征进行初步筛选：计算每个组学特征与药物治疗的治疗结局之间的斯皮尔曼相关系数p值，筛选出p值<0.05(p值<0.05被认为有统计意义)的组学特征，获得初步影像组学特征；
然后使用套索算法(Lasso算法)对所述初步影像组学特征进行进一步筛选，筛选出与病灶分类有显著关系的影像组学特征，获得所述目标影像组学特征。
步骤S6、对训练集中的目标影像组学特征运用机器学习算法训练预测模型，构建获得癫痫药物治疗结局预测模型，并在测试集中验证模型。
在本发明实施例中，所述步骤S6包括以下子步骤：
S61、采用多种机器学习算法分别建立模型，获得多种类型的预测模型；所述多种机器学习算法包括支持向量机算法、随机森林算法、Logistic回归分析算法、Ada Boost算法、Gradient Boosting算法和Decision Tree算法。
S62、对于每一类型的预测模型，采用十折交叉验证进行训练，训练过程使用网格搜索算法选择最优超参数，具体是根据受试者工作曲线(ROC)下面积(AUC)逆向选择最佳模型参数；根据确定的最优超参数，在整个训练集上训练各个类型的预测模型，训练得到候选预测模型，并在测试集上测试验证。
S63、基于AUC、准确性、敏感性和特异性的性能参数对候选预测模型进行性能评估，选取性能最佳的候选预测模型作为所述癫痫药物治疗结局预测模型。
基于以上步骤S1至S6，本发明实施例构建获得一种基于多模态影像组学的癫痫药物治疗结局预测模型。
本发明实施例提供的预测方法中，对待治疗患者进行预测的部分包括以下步骤：
步骤S7、利用构建获得癫痫药物治疗结局预测模型对待治疗患者的目标影像组学特征进行预测，获得预测的癫痫药物治疗结局。
具体地，医生采集TSC患者的T1W、T2W和FLAIR序列影像，提取三种模态对应的目标影像组学特征输入到所述癫痫药物治疗结局预测模型中进行预测，获得预测的癫痫药物治疗结局。
实施例2
本实施例提供一种基于多模态影像组学的癫痫药物治疗结局预测装置，如图3所示，该装置100包括影像获取模块1、分组模块2、影像分割模块3、特征提取模块4、特征筛选模块5、模型构建模块6和治疗结局预测模块7。其中，
所述影像获取模块用于获取TSC患者在进行抗癫痫药物治疗前的多种模态磁共振影像，并对所述多种模态磁共振影像进行预处理；即实施例1中步骤S1对应的工作过程。
所述分组模块用于把TSC患者按比例随机分为训练集和测试集，所述训练集用于训练预测模型，所述测试集用于验证预测模型的性能；即实施例1中步骤S2对应的工作过程.
所述影像分割模块用于基于U-net++网络对预处理后的每一种模态磁共振影像进行区域分割，获取每一种模态磁共振影像对应的感兴趣区域；即实施例1中步骤S3对应的工作过程。
所述特征提取模块用于对每一种模态磁共振影像的每一个感兴趣区域进行特征提取，获取每一个感兴趣区域对应的高维影像组学特征；即实施例1中步骤S4对应的工作过程。
所述特征筛选模块用于对所述高维影像组学特征进行分析筛选，降维后得到目标影像组学特征；即实施例1中步骤S5对应的工作过程。
所述模型构建模块用于对训练集中的目标影像组学特征运用机器学习算法训练预测模型，构建获得癫痫药物治疗结局预测模型，并在测试集中验证模型；即实施例1中步骤S6对应的工作过程。
所述治疗结局预测模块用于利用构建获得癫痫药物治疗结局预测模型对待治疗患者的目标影像组学特征进行预测，获得预测的癫痫药物治疗结局；即实施例1中步骤S7对应的工作过程。
实施例3
基于如上实施例提供的基于多模态影像组学的癫痫药物治疗结局预测方法，本实施例提供了一种终端设备，如图4所示，所述终端设备包括：处理器10、存储器20、输入装置30和输出装置40，处理器10中设置有GPU，处理器10的数量可以是一个或多个，图2中以一个处理器10为例。终端设备中的处理器10、存储器20、输入装置30和输出装置40可以通过总线或其他方式连接。
其中，存储器20作为一种计算机可读存储介质，可用于存储软件程序、计算机可执行程序以及模块。处理器10通过运行存储在存储器20中的软件程序、指令以及模块，从而执行设备的各种功能应用以及数据处理，即实现本发明前述实施例中所述的基于多模态影像组学的癫痫药物治疗结局预测方法的步骤。输入装置30用于接收图像数据、输入的数字或字符信息，以及产生与设备的用户设置以及功能控制有关的键信号输入。输出装置40可包括显示屏等显示设备，例如是用于显示图像。
本发明如上实施例提供的癫痫药物治疗结局预测方法和装置以及相应的终端设备，通过基于U-net++网络对多种模态磁共振影像进行精确的区域分割，再提取感兴趣区域的影像组学特征，根据影像组学特征建立预测模型，能够快速、有效地预测TSC癫痫患者的药物治疗结局，在药物治疗开始之前就通过MRI影像来预测患者是否耐药，能够辅助医生进行更准确的临床决策，给患者争取时间进行更合适的治疗，患者不需要长达一年以上的耐药性测试，能够极大地 降低医生的人力成本和患者的时间成本。
需要指出的是，上述实施例仅为说明本发明的技术构思及特点，其目的在于让熟悉此项技术的人士能够了解本发明的内容并据以实施，并不能以此限制本发明的保护范围。凡根据本发明精神实质所作的等效变化或修饰，都应涵盖在本发明的保护范围之内。

Claims (10)

  1. 一种基于多模态影像组学的癫痫药物治疗结局预测方法，其中，包括步骤：
    获取TSC患者在进行抗癫痫药物治疗前的多种模态磁共振影像，并对所述多种模态磁共振影像进行预处理；
    把TSC患者按比例随机分为训练集和测试集，所述训练集用于训练预测模型，所述测试集用于验证预测模型的性能；
    基于U-net++网络对预处理后的每一种模态磁共振影像进行区域分割，获取每一种模态磁共振影像对应的感兴趣区域；
    对每一种模态磁共振影像的每一个感兴趣区域进行特征提取，获取每一个感兴趣区域对应的高维影像组学特征；
    对所述高维影像组学特征进行分析筛选，降维后得到目标影像组学特征；
    对训练集中的目标影像组学特征运用机器学习算法训练预测模型，构建获得癫痫药物治疗结局预测模型，并在测试集中验证模型；
    利用构建获得癫痫药物治疗结局预测模型对待治疗患者的目标影像组学特征进行预测，获得预测的癫痫药物治疗结局。
  2. 根据权利要求1所述的癫痫药物治疗结局预测方法，其中，所述多种模态磁共振影像包括磁共振影像中的T1加权影像、T2加权影像和液体衰减反转恢复影像。
  3. 根据权利要求1或2所述的癫痫药物治疗结局预测方法，其中，所述TSC患者的入组标准需要至少满足以下3个条件：1)患者在使用抗癫痫药物治疗前进行了多种模态磁共振影像扫描；2)患者使用抗癫痫药物治疗1年以上；3)患者未进行病灶切除手术。
  4. 根据权利要求1所述的癫痫药物治疗结局预测方法，其中，所述训练集和所述测试集的数量比例为8:2或7:3。
  5. 根据权利要求1所述的癫痫药物治疗结局预测方法，其中，在基于U-net++网络对预处理后的每一种模态磁共振影像进行区域分割之后，通过人工检查并 修改完善影像的分割轮廓，进而获取每一种模态磁共振影像对应的感兴趣区域。
  6. 根据权利要求1所述的癫痫药物治疗结局预测方法，其中，所述高维影像组学特征至少包括以下3类特征：1)用于描述结节病灶尺寸和形状的三维形态特征；2)用于描述病灶区域强度分布的一阶统计特征；3)用于描述病灶区域空间分布信息的纹理特征，包括灰度共生矩阵、灰度游程矩阵、灰度区域大小矩阵、相邻灰度差分矩阵和灰度共生矩阵。
  7. 根据权利要求1所述的癫痫药物治疗结局预测方法，其中，所述对高维影像组学特征进行分析筛选，降维后得到目标影像组学特征的步骤包括：
    首先采用了双变量分析算法对高维影像组学特征进行初步筛选：计算每个组学特征与药物治疗的治疗结局之间的斯皮尔曼相关系数p值，筛选出p值<0.05的组学特征，获得初步影像组学特征；
    然后使用套索算法对所述初步影像组学特征进行进一步筛选，筛选出与病灶分类有显著关系的影像组学特征，获得所述目标影像组学特征。
  8. 根据权利要求1或7所述的癫痫药物治疗结局预测方法，其中，所述对训练集中的目标影像组学特征运用机器学习算法训练预测模型，构建获得癫痫药物治疗结局预测模型，并在测试集中验证模型的步骤包括：
    采用多种机器学习算法分别建立模型，获得多种类型的预测模型；所述多种机器学习算法包括支持向量机算法、随机森林算法、Logistic回归分析算法、Ada Boost算法、Gradient Boosting算法和Decision Tree算法；
    对于每一类型的预测模型，采用十折交叉验证进行训练，训练过程使用网格搜索算法选择最优超参数；根据确定的最优超参数，在整个训练集上训练各个类型的预测模型，训练得到候选预测模型，并在测试集上测试验证；
    基于AUC、准确性、敏感性和特异性的性能参数对候选预测模型进行性能评估，选取性能最佳的候选预测模型作为所述癫痫药物治疗结局预测模型。
  9. 一种基于多模态影像组学的癫痫药物治疗结局预测装置，其中，包括：
    影像获取模块，用于获取TSC患者在进行抗癫痫药物治疗前的多种模态磁共振影像，并对所述多种模态磁共振影像进行预处理；
    分组模块，用于把TSC患者按比例随机分为训练集和测试集，所述训练集 用于训练预测模型，所述测试集用于验证预测模型的性能；
    影像分割模块，用于基于U-net++网络对预处理后的每一种模态磁共振影像进行区域分割，获取每一种模态磁共振影像对应的感兴趣区域；
    特征提取模块，用于对每一种模态磁共振影像的每一个感兴趣区域进行特征提取，获取每一个感兴趣区域对应的高维影像组学特征；
    特征筛选模块，用于对所述高维影像组学特征进行分析筛选，降维后得到目标影像组学特征；
    模型构建模块，用于对训练集中的目标影像组学特征运用机器学习算法训练预测模型，构建获得癫痫药物治疗结局预测模型，并在测试集中验证模型；
    治疗结局预测模块，用于利用构建获得癫痫药物治疗结局预测模型对待治疗患者的目标影像组学特征进行预测，获得预测的癫痫药物治疗结局。
  10. 一种终端设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，其中，所述处理器执行所述程序时实现如权利要求1-8任一项所述的基于多模态影像组学的癫痫药物治疗结局预测方法的步骤。
PCT/CN2022/080826 2022-03-10 2022-03-15 基于多模态影像组学的癫痫药物治疗结局预测方法和装置 WO2023168728A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210228783.3 2022-03-10
CN202210228783.3A CN114711717A (zh) 2022-03-10 2022-03-10 基于多模态影像组学的癫痫药物治疗结局预测方法和装置

Publications (1)

Publication Number Publication Date
WO2023168728A1 true WO2023168728A1 (zh) 2023-09-14

Family

ID=82237164

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/080826 WO2023168728A1 (zh) 2022-03-10 2022-03-15 基于多模态影像组学的癫痫药物治疗结局预测方法和装置

Country Status (2)

Country Link
CN (1) CN114711717A (zh)
WO (1) WO2023168728A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116825382A (zh) * 2023-02-23 2023-09-29 深圳市儿童医院 一种基于多模态融合的癫痫药物有效性检测方法及装置

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106875401A (zh) * 2017-01-10 2017-06-20 中国科学院深圳先进技术研究院 多模态影像组学的分析方法、装置及终端
US20180211012A1 (en) * 2017-01-25 2018-07-26 Ucb Biopharma Sprl Method and system for predicting optimal epilepsy treatment regimes
CN111462116A (zh) * 2020-05-13 2020-07-28 吉林大学第一医院 基于影像组学特征的多模态参数模型优化融合方法
CN112927187A (zh) * 2021-01-27 2021-06-08 张凯 —种自动识别定位局灶性皮质发育不良癫痫病灶的方法
CN113222915A (zh) * 2021-04-28 2021-08-06 浙江大学 基于多模态磁共振影像组学的pd诊断模型的建立方法
CN113571195A (zh) * 2021-07-20 2021-10-29 南京脑科医院 基于小脑功能连接特征的阿尔茨海默病早期的预测模型

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106875401A (zh) * 2017-01-10 2017-06-20 中国科学院深圳先进技术研究院 多模态影像组学的分析方法、装置及终端
US20180211012A1 (en) * 2017-01-25 2018-07-26 Ucb Biopharma Sprl Method and system for predicting optimal epilepsy treatment regimes
CN111462116A (zh) * 2020-05-13 2020-07-28 吉林大学第一医院 基于影像组学特征的多模态参数模型优化融合方法
CN112927187A (zh) * 2021-01-27 2021-06-08 张凯 —种自动识别定位局灶性皮质发育不良癫痫病灶的方法
CN113222915A (zh) * 2021-04-28 2021-08-06 浙江大学 基于多模态磁共振影像组学的pd诊断模型的建立方法
CN113571195A (zh) * 2021-07-20 2021-10-29 南京脑科医院 基于小脑功能连接特征的阿尔茨海默病早期的预测模型

Also Published As

Publication number Publication date
CN114711717A (zh) 2022-07-08

Similar Documents

Publication Publication Date Title
Cui et al. RNN-based longitudinal analysis for diagnosis of Alzheimer’s disease
Chen et al. Computer‐aided grading of gliomas combining automatic segmentation and radiomics
Yan et al. Cortical surface biomarkers for predicting cognitive outcomes using group l2, 1 norm
Sethi et al. [Retracted] Classification of Alzheimer’s Disease Using Gaussian‐Based Bayesian Parameter Optimization for Deep Convolutional LSTM Network
Mehmood et al. Prioritization of brain MRI volumes using medical image perception model and tumor region segmentation
RU2681280C2 (ru) Процесс обработки медицинских изображений
Diniz et al. Detection of white matter lesion regions in MRI using SLIC0 and convolutional neural network
Bennai et al. A stochastic multi-agent approach for medical-image segmentation: application to tumor segmentation in brain MR images
Rahim et al. Prediction of Alzheimer's progression based on multimodal deep-learning-based fusion and visual explainability of time-series data
Dong et al. Identifying carotid plaque composition in MRI with convolutional neural networks
WO2023168728A1 (zh) 基于多模态影像组学的癫痫药物治疗结局预测方法和装置
Nizamani et al. Advance Brain Tumor segmentation using feature fusion methods with deep U-Net model with CNN for MRI data
Kim et al. Web applicable computer-aided diagnosis of glaucoma using deep learning
Coupet et al. A multi-sequences MRI deep framework study applied to glioma classfication
Haider et al. Exploring deep feature-blending capabilities to assist glaucoma screening
Alshehri et al. A few-shot learning-based ischemic stroke segmentation system using weighted MRI fusion
Li et al. AMD-Net: Automatic subretinal fluid and hemorrhage segmentation for wet age-related macular degeneration in ocular fundus images
Asgharzadeh-Bonab et al. An alzheimer’s disease classification method using fusion of features from brain magnetic resonance image transforms and deep convolutional networks
Iqbal et al. Hybrid Parallel Fuzzy CNN Paradigm: Unmasking Intricacies for Accurate Brain MRI Insights
Berkley et al. Clinical capability of modern brain tumor segmentation models
Rodrigues et al. DermaDL: advanced convolutional neural networks for automated melanoma detection
Mehmood et al. Utilizing Siamese 4D-AlzNet and Transfer Learning to Identify Stages of Alzheimer’s Disease
CN110163195A (zh) 肝癌分群预测模型、其预测系统以及肝癌分群判断方法
Guo et al. Diabetic retinopathy lesion segmentation using deep multi-scale framework
Vasilevski Meta-learning for Clinical and Imaging Data Fusion for Improved Deep Learning Inference

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22930365

Country of ref document: EP

Kind code of ref document: A1