Method and system for quantitative feature selection of epilepsy lesion based on 18F-FDG PET imaging

Abstract

The invention belongs to but is not limited to the technical field of epileptic focus positioning, and discloses an epilepsy lesion quantitative feature selection method and system based on 18F-FDG PET imaging, and the method comprises the steps: collecting 18F-FDG PET imaging data of a patient through an image collection module; the image preprocessing module is used for preprocessing an original PET image; the feature extraction module is used for extracting potential features of epilepsy lesions by using multiple algorithms; the feature selection module performs feature selection by using a machine learning algorithm; the epileptic focus positioning and analysis module analyzes the PET data of the patient through a classification model in combination with the selected features, generates a probability heat map of the epileptic focus and displays the position of the epileptic focus; and the result display and report module visualizes the analysis result and generates a detailed analysis report. According to the method, the limitation of subjectivity and single-dimensional analysis in the prior art is effectively solved, and the positioning precision and clinical application value of the epilepsy lesion are remarkably improved.

Description

Technical Field

[0001] The present invention belongs to but is not limited to the technical field of epileptic focus localization, and in particular relates to a method and system for selecting quantitative features of epileptic focus based on 18F-FDG PET imaging. Background Art

[0002] Among the current methods for localizing epileptogenic foci, quantitative analysis based on 18F-FDG PET imaging is an important means, especially in the treatment of drug-resistant epilepsy, where accurate localization of the epileptogenic foci is crucial for surgical resection. Existing technologies usually rely on visual assessment or simple quantitative parameters (such as SUV values) to determine the epileptogenic foci, but these methods have significant limitations:

[0003] 1. Visual assessment is highly subjective: Doctors determine the location of the epileptogenic focus by observing the brightness changes of PET imaging data with the naked eye. This method relies on the doctor's experience and is highly subjective, which can easily lead to inconsistent results. In particular, when the epileptogenic focus is small or presents a complex metabolic pattern, the visual assessment effect is poor.

[0004] 2. Single dimension of quantitative analysis: Traditional quantitative methods, such as standardized uptake value (SUV), can only reflect the metabolic status of local brain tissue, but the manifestation of epileptogenic focus is affected by many factors, such as the complexity of brain structure, metabolic status of adjacent areas, etc. Therefore, a single quantitative parameter is not enough to provide complete information about the epileptogenic focus, resulting in inaccurate identification of the location and characteristics of the epileptogenic focus.

[0005] 3. Lack of automated processing and feature selection: The existing technology lacks an effective way to combine PET images with data-driven methods such as machine learning, making it difficult to automatically identify and select image features that are highly correlated with epileptic foci.

[0006] In order to improve the accuracy of localization of epileptogenic foci based on 18F-FDG PET imaging, the following problems need to be urgently addressed:

[0007] How to objectify and automate feature extraction and analysis to reduce errors caused by subjective judgment.

[0008] How to construct a multi-dimensional quantitative feature selection model to fully utilize the metabolic characteristics of epileptic foci implicit in 18F-FDG PET data.

[0009] How to achieve efficient computing architecture and system integration to transform complex data analysis processes into clinically usable tools. Summary of the invention

[0010] In view of the problems existing in the prior art, the present invention provides a method and system for selecting quantitative features of epileptogenic foci based on 18F-FDG PET imaging.

[0011] The present invention is implemented as follows: a quantitative feature selection system for epileptogenic focus based on 18F-FDG PET imaging, comprising:

[0012] Image acquisition module, used to collect the patient's 18F-FDG PET imaging data and perform registration with structural images such as MRI to improve the accuracy of spatial positioning;

[0013] Image preprocessing module, used to preprocess the original PET images, including standardization, denoising and normalization;

[0014] The feature extraction module is used to extract the potential features of the epileptogenic focus using a variety of algorithms. The feature dimensions include standardized uptake value (SUV), spatial texture features, and local metabolic features.

[0015] Feature selection module, used to perform feature selection using machine learning algorithms;

[0016] The epileptic focus localization and analysis module is used to combine the selected features and analyze the patient's PET data through a classification model (such as a support vector machine (SVM) or a random forest) to generate a probability heat map of the epileptic focus and display the location of the epileptic focus;

[0017] The result display and reporting module is used to visualize the analysis results, show the location of the epileptic focus to clinicians in the form of images, and generate a detailed analysis report, including the location of the epileptic focus and its confidence; the features involved in the decision-making and their weights; and the corresponding surgical resection recommendations.

[0018] Furthermore, the image acquisition module uses standard PET scanning equipment to obtain metabolic imaging data of the patient's brain to ensure the clarity and integrity of the imaging; uses MRI scanning to obtain high-resolution structural images to provide accurate spatial information of various brain regions; and aligns the PET image with the MRI image based on an image registration algorithm (such as rigid registration and elastic registration) to ensure the spatial consistency of the two modal images.

[0019] Furthermore, the preprocessing in the image preprocessing module includes:

[0020] Image registration: Multimodal image registration using PET and MRI data, using algorithms such as mutual information (MI) and normalized cross entropy to ensure spatial consistency;

[0021] Image normalization: PET data is normalized by a normalization algorithm to adjust the image data to the same range and eliminate differences caused by equipment and imaging conditions. Maximum and minimum normalization or Z-Score normalization can be used;

[0022] Noise removal: Use Gaussian filter or median filter to smooth the noise in the image and improve image clarity.

[0023] Furthermore, the feature dimensions include:

[0024] Standardized uptake value (SUV) extraction: Calculate the SUV value of each brain area, which represents the metabolic activity of each brain area; it can be performed based on the division of the region of interest (ROI) using the standardized uptake value calculation method.

[0025] Spatial texture feature extraction: The texture features of the brain area are extracted through the gray-level co-occurrence matrix (GLCM) to reflect the metabolic state of the local brain tissue; the extracted features include contrast, energy, homogeneity, etc.

[0026] Local metabolic characteristics: Calculate the metabolic differences between adjacent brain regions, analyze the changes in metabolic levels between different regions, and find hot spots of metabolic abnormalities; use spatial convolution algorithms or gradient-based analysis algorithms to calculate metabolic differences.

[0027] Furthermore, the feature selection module performs feature selection including:

[0028] Initial feature screening: Use statistical methods (such as chi-square test, t-test, etc.) to preliminarily screen the extracted features, remove features that are not related to the epileptic focus, and retain highly relevant features;

[0029] Recursive Feature Elimination (RFE): Using the RFE algorithm, the model is trained step by step, unimportant features are removed, and feature screening is repeated. The most important feature set is retained by removing less influential features in each iteration.

[0030] L1 regularization (LASSO) selection: LASSO regression is used for further feature compression.

[0031] Furthermore, the results are visualized in the result display and reporting module: through image processing technology, the epileptic focus localization results are superimposed on the MRI image and displayed to the clinician in a visual way.

[0032] Report generation: Use natural language processing (NLP) algorithms to generate detailed reports based on model analysis results; the report includes confidence analysis of epileptic foci, feature weight distribution used in decision-making, epileptic foci location, and resection recommendations.

[0033] Another object of the present invention is to provide a quantitative feature selection method for epilepsy focus based on 18F-FDG PET imaging, comprising the following steps:

[0034] Step 1: Data acquisition and registration: Obtain the patient's brain 18F-FDG PET image through the PET scanning device, and use image registration technology to spatially align the PET image with structural images such as MRI;

[0035] Step 2, image preprocessing: standardize, denoise, and normalize PET images to ensure data consistency and quality in subsequent analysis;

[0036] Step 3: Multi-dimensional feature extraction: Use image processing algorithms to extract different types of image features, including metabolic features, texture features, and spatial features;

[0037] Step 4, feature selection: Through statistical screening, recursive feature elimination and L1 regularization, the feature space is gradually reduced to finally obtain the optimal feature set for identifying epileptic foci;

[0038] Step 5: Model training and prediction: Use machine learning models such as support vector machines or random forests for training, and predict the location of the epileptogenic focus in the patient's brain based on the selected feature set;

[0039] Step 6: Result visualization and report generation: The analysis results are presented in the form of images, and a detailed epileptic focus localization report is generated to provide diagnostic support for clinicians.

[0040] Another object of the present invention is to provide a computer device, the computer device comprising a memory and a processor, the memory storing a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method for quantitative feature selection of epilepsy focus based on 18F-FDG PET imaging.

[0041] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to execute the steps of the method for quantitative feature selection of epileptogenic foci based on 18F-FDG PET imaging.

[0042] Another object of the present invention is to provide an information data processing terminal, which includes the quantitative feature selection system for epileptic focus based on 18F-FDG PET imaging.

[0043] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows:

[0044] The system of the quantitative feature selection method of epileptogenic focus based on 18F-FDG PET imaging of the present invention adopts a modular design and integrates modules such as image acquisition, preprocessing, feature extraction, feature selection, epilepsy focus localization analysis, and result display and report generation.

[0045] The present invention improves positioning accuracy: compared with traditional SUV value analysis or manual visual evaluation, the system can capture more metabolic features related to epileptic foci through multi-dimensional feature extraction, significantly improving the positioning accuracy of epileptic foci.

[0046] The present invention automates and objectively analyzes: the system automatically processes data and generates positioning results, reducing the influence of doctors' subjective judgment. Through machine learning algorithms, the system can automatically identify the most relevant features, ensuring the objectivity and consistency of the analysis.

[0047] Real-time feedback and optimization of the present invention: Combining feature selection with machine learning models, the system can quickly analyze new patient data and generate accurate epileptic focus localization results, greatly improving the efficiency of clinical applications.

[0048] The present invention provides more comprehensive diagnostic support: through in-depth analysis of different features, the system can not only locate the epileptic focus, but also provide a deep understanding of the formation mechanism of the epileptic focus, which is helpful to formulate more effective surgical or treatment plans.

[0049] In summary, the present invention effectively solves the limitations of subjectivity and single-dimensional analysis in the prior art through systematic feature selection and machine learning algorithms, and significantly improves the accuracy and clinical application value of epileptic focus localization. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 18F-FDG PET imaging-based quantitative feature selection system for epilepsy epileptogenic focus provided by an embodiment of the present invention;

[0051] Figure 2 18F-FDG PET imaging-based quantitative feature selection system for epilepsy epileptogenic focus provided by an embodiment of the present invention;

[0052] Figure 3 is a flow chart of a method for quantitative feature selection of epileptogenic foci based on 18F-FDG PET imaging provided by an embodiment of the present invention;

[0053] Figure 4 It is a specific flow chart of the method for quantitative feature selection of epileptogenic focus based on 18F-FDG PET imaging provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0055] like Figure 1 As shown, an embodiment of the present invention provides a quantitative feature selection system for epilepsy focus based on 18F-FDG PET imaging, comprising:

[0056] Image acquisition module, used to collect the patient's 18F-FDG PET imaging data and perform registration with structural images such as MRI to improve the accuracy of spatial positioning;

[0057] Image preprocessing module, used to preprocess the original PET images, including standardization, denoising and normalization;

[0058] The feature extraction module is used to extract the potential features of the epileptogenic focus using a variety of algorithms. The feature dimensions include standardized uptake value (SUV), spatial texture features, and local metabolic features.

[0059] Feature selection module, used to perform feature selection using machine learning algorithms;

[0060] The epileptic focus localization and analysis module is used to combine the selected features and analyze the patient's PET data through a classification model (such as a support vector machine (SVM) or a random forest) to generate a probability heat map of the epileptic focus and display the location of the epileptic focus;

[0061] The result display and reporting module is used to visualize the analysis results, show the location of the epileptic focus to clinicians in the form of images, and generate a detailed analysis report, including the location of the epileptic focus and its confidence; the features involved in the decision-making and their weights; and the corresponding surgical resection recommendations.

[0062] The embodiment of the present invention provides a quantitative feature selection system for epilepsy focus based on 18F-FDG PET imaging, which aims to accurately locate the epilepsy focus of epilepsy patients by combining PET image data with machine learning algorithms. The system includes six modules: image acquisition, preprocessing, feature extraction, feature selection, epilepsy focus localization and analysis, and result display and reporting. The following is a detailed working principle:

[0063] 1. Image acquisition module

[0064] This module is mainly used to collect 18F-FDG PET imaging data of patients. PET imaging can reflect the metabolic activity of the brain. In order to improve the spatial localization accuracy of the epileptic focus, the system also combines structural images such as MRI for registration. MRI images can provide clear brain anatomical structure, while PET images can provide metabolic information. By using image registration algorithms (such as mutual information-based registration algorithms), PET and MRI images are accurately aligned, so that metabolic activities are matched with specific anatomical structures, providing a more accurate spatial reference for subsequent epileptic focus localization.

[0065] 2. Image preprocessing module

[0066] After image acquisition, the system will preprocess the original PET image data. The preprocessing steps include standardization, denoising and normalization. First, the standardization process is to eliminate the impact of different scanning conditions and standardize the data values ​​of the PET image to comparable values; secondly, the denoising process removes the noise signal in the image through filtering technology (such as Gaussian filtering or wavelet transform) to ensure that the extracted features have higher credibility; finally, the normalization process is to convert the image data into the same dimensional range, making the subsequent feature extraction and model training more stable.

[0067] 3. Feature Extraction Module

[0068] After preprocessing, the system will extract features from PET images through a variety of algorithms, including standardized uptake value (SUV), spatial texture features, and local metabolic features. SUV is an important parameter in PET imaging, representing the metabolic activity of a region to 18F-FDG, and is often used to measure metabolic changes in the epileptogenic region. Spatial texture features capture the spatial distribution and texture patterns of images through image processing algorithms (such as gray-level co-occurrence matrix or local binary pattern) to help identify abnormal metabolic areas. Local metabolic features further distinguish normal tissue from epileptogenic areas by analyzing the distribution of local metabolic activity in PET images.

[0069] 4. Feature Selection Module

[0070] After extracting a large number of features, the system uses machine learning algorithms to select these features, with the goal of screening out the features that are most relevant to the epileptogenic focus. Commonly used algorithms in this module include support vector machine recursive feature elimination (SVM-RFE) and random forest feature importance analysis. Through these algorithms, the system can effectively reduce feature dimensions, remove redundant and irrelevant features, and retain core features that have diagnostic value for the epileptogenic focus. This step helps to improve the training efficiency of the classification model and improve the accuracy of localizing the epileptic focus.

[0071] 5. Epilepsy focus localization and analysis module

[0072] After completing feature selection, the system inputs the selected features into a classification model, such as a support vector machine (SVM) or random forest, to analyze the patient's PET data. Through these classification models, the system can generate a probability heat map of the epileptic focus, showing the location of the epileptic focus. The training data of the classification model is based on a large number of known cases, and the classification ability of the model is optimized through supervised learning. The generated probability heat map shows the probability of different areas being epileptic focuses, helping doctors to intuitively identify potential lesion areas.

[0073] 6. Results display and reporting module

[0074] Finally, the system will present the analysis results of epileptic focus localization to clinicians through visualization. The images generated by this module include an overlay of PET data and probability heat maps, which clearly mark the location of the epileptic focus with high confidence. In addition, the system also generates a detailed analysis report, which includes the location of the epileptic focus and its confidence, the features involved in the decision and their weight distribution, and provides corresponding surgical resection recommendations based on the analysis results. This visualization and reporting function provides clinicians with more intuitive diagnostic support, helping them make more accurate treatment decisions.

[0075] The embodiment of the present invention provides a quantitative feature selection system for epilepsy epileptogenic focus based on 18F-FDG PET imaging technology, and its working principle mainly includes the following key modules: image acquisition module, image preprocessing module, feature extraction module and feature selection module. The function and workflow of each module will be introduced in detail below.

[0076] First, the main task of the image acquisition module is to obtain the patient's 18F-FDG PET scan data. 18F-FDGPET is a functional imaging technology that can reflect the glucose metabolism of the brain and is often used to identify areas of abnormal metabolic activity, such as epileptogenic foci. In this module, in order to improve the spatial positioning accuracy of the imaging data, the acquired PET image data will be registered with structural images such as MRI. This step is to spatially align images of different imaging modalities through software algorithms to ensure accurate correspondence between functional and anatomical structures.

[0077] Next, the image preprocessing module processes the original PET images for subsequent analysis and feature extraction. Preprocessing steps usually include standardization, denoising, and normalization. Standardization refers to adjusting the image data to the same scale and range to eliminate the impact of equipment differences; denoising is used to reduce random noise in the imaging process and improve image quality; normalization is to adjust the pixel values ​​of the PET image to a uniform range for comparison and further analysis.

[0078] The feature extraction module is the core part of the system, responsible for extracting multidimensional data features that can describe the characteristics of epileptogenic foci from preprocessed images. These features include standardized uptake value (SUV), which is a common indicator of radiotracer concentration in tissues; spatial texture features, which are used to describe the distribution and organization of patterns in images; and local metabolic features, which involve the evaluation of metabolic activity in local brain regions. By applying a variety of algorithms and statistical methods, useful information can be systematically extracted from large amounts of data, providing a basis for subsequent analysis.

[0079] Finally, the feature selection module uses machine learning algorithms to select the most diagnostically valuable features from the many extracted features. The purpose of this step is to reduce the data dimension and improve the predictive performance and interpretability of the model. Common feature selection methods include recursive feature elimination and model-based selection. The selected features will be used to build a model to accurately locate and quantitatively analyze epileptogenic foci, thereby supporting clinical diagnosis and treatment decisions.

[0080] Through the coordinated work of these four modules, the present invention can effectively extract valuable information from complex PET imaging data, assisting doctors to better understand and treat epilepsy.

[0081] The epilepsy focus localization and analysis module and the result display and reporting module of the present invention are key components of the epilepsy analysis system based on 18F-FDG PET imaging data. These two modules together provide a complete workflow from data processing to clinical decision support. The functions and working principles of these two modules are described in detail below.

[0082] The main task of the epileptic focus localization and analysis module is to locate and analyze epileptic foci using the key features selected from the feature selection module. This module uses advanced machine learning algorithms, such as support vector machines (SVM) or random forests, which are able to process high-dimensional data and identify potential epileptic focus locations. Using these algorithms, the module analyzes the patient's PET scan data and generates a heat map by calculating the probability of epileptogenicity for each brain region. This heat map uses different color intensities to indicate the high and low probability of epileptogenicity, providing clinicians with an intuitive visual reference to help them identify and evaluate potential epileptic foci.

[0083] The main function of the result display and reporting module is to present the analysis results in the form of images and text, so that clinicians can intuitively understand the analysis output and make clinical decisions based on it. This module not only displays the location heat map of the epileptogenic focus, but also includes a detailed analysis report. The report contains key information, such as the location of the epileptogenic focus and its confidence, the features involved in the decision and their weights, and surgical resection recommendations based on the analysis results. This information provides doctors with a scientific basis to help them develop more accurate treatment plans.

[0084] Through this comprehensive analysis and reporting method, doctors can more effectively use PET imaging data to guide the diagnosis and treatment of epilepsy. The module is designed with the needs of users in mind, ensuring ease of operation and accuracy of results, greatly improving the efficiency of clinical operations and the success rate of treatment. The application of this system can significantly improve the management and treatment outcomes of epilepsy patients and improve the quality of life of patients.

[0085] like Figure 2 As shown, the following is the specific implementation plan:

[0086] (1) Image acquisition module

[0087] Functional description: The core task of the image acquisition module is to obtain the patient's 18F-FDG PET imaging data and combine it with images of other modalities (such as MRI) for registration to improve the accuracy of spatial localization of epileptic foci.

[0088] Input: 18F-FDG PET data, MRI image data.

[0089] Implementation method:

[0090] Standard PET scanning equipment is used to obtain metabolic imaging data of the patient's brain to ensure the clarity and integrity of the imaging.

[0091] MRI scans are used to obtain high-resolution structural images that provide precise spatial information about various brain regions.

[0092] The PET images are registered with the MRI images based on image registration algorithms (such as rigid registration and elastic registration) to ensure the spatial consistency of the two modality images.

[0093] Output: registered PET and MRI image data for use by subsequent modules.

[0094] (2) Image preprocessing module

[0095] Functional description: The purpose of the preprocessing module is to improve the quality of image data and ensure the accuracy of subsequent feature extraction and analysis.

[0096] Input: Registered PET and MRI image data.

[0097] Implementation method:

[0098] Image registration: Multimodal image registration is performed using PET and MRI data, using algorithms such as mutual information (MI) and normalized cross entropy to ensure spatial consistency.

[0099] Image normalization: PET data is normalized by a normalization algorithm to adjust the image data to the same range and eliminate differences caused by equipment and imaging conditions. Maximum and minimum normalization or Z-Score normalization can be used.

[0100] Noise removal: Use Gaussian filter or median filter to smooth the noise in the image and improve image clarity.

[0101] Output: preprocessed, standardized, and denoised PET image data.

[0102] (3) Feature extraction module

[0103] Functional description: The feature extraction module extracts relevant features of epileptic foci through a variety of algorithms, providing rich image information for subsequent analysis.

[0104] Input: Preprocessed PET image data.

[0105] Implementation method:

[0106] Standardized uptake value (SUV) extraction: Calculate the SUV value of each brain region, which represents the metabolic activity of each brain region. This can be performed based on the division of the region of interest (ROI) using the standardized uptake value calculation method.

[0107] Spatial texture feature extraction: The texture features of the brain region are extracted through the gray level co-occurrence matrix (GLCM) to reflect the metabolic state of the local brain tissue. The extracted features include contrast, energy, homogeneity, etc.

[0108] Local metabolic characteristics: Calculate the metabolic differences between adjacent brain regions, analyze the changes in metabolic levels between different regions, and find hot spots of metabolic abnormalities. Use spatial convolution algorithms or gradient-based analysis algorithms to calculate metabolic differences.

[0109] Output: A multidimensional feature vector containing standardized uptake values, spatial texture features, and local metabolic features.

[0110] (4) Feature Selection Module

[0111] Functional description: This module uses machine learning algorithms to select features, eliminate redundant information, and retain the most useful features for localizing epileptogenic foci.

[0112] Input: The extracted multi-dimensional feature vector.

[0113] Implementation method:

[0114] Initial feature screening: The extracted features are preliminarily screened by statistical methods (such as chi-square test, t-test, etc.), and features irrelevant to the epileptic focus are removed, while highly relevant features are retained.

[0115] Recursive Feature Elimination (RFE): Using the RFE algorithm, the model is trained step by step, and unimportant features are removed to repeatedly filter features. By removing less influential features in each iteration, the most important feature set is finally retained.

[0116] L1 regularization (LASSO) selection: LASSO regression is introduced for further feature compression. L1 regularization retains the most representative features and reduces the complexity of the model to avoid overfitting through sparsity constraints.

[0117] Output: An optimized feature set containing the most relevant epileptic focus features.

[0118] (5) Epilepsy focus localization and analysis module

[0119] Function description: Combine the selected features, use the classification model to accurately locate the epileptic focus, and generate a probability heat map.

[0120] Input: optimized feature set, MRI and PET imaging data.

[0121] Implementation method:

[0122] Classification model training and prediction: Use support vector machine (SVM) or random forest model to train based on the extracted and selected feature set. The model optimizes parameters through methods such as cross-validation to ensure the accuracy of prediction.

[0123] Probabilistic heat map generation: Based on the output of the classification model, a probability heat map of the epileptic focus is generated to display the epileptic focus area. The accuracy of spatial positioning is ensured by combining the results with MRI images.

[0124] Output: Spatial probability heat map of epileptic foci, identifying epileptogenic areas.

[0125] (6) Result display and report module

[0126] Function description: Visualize the localization results of epileptic focus and generate a detailed analysis report to provide reference for doctors.

[0127] Input: epileptic focus probability heat map, classification model analysis results.

[0128] Implementation method:

[0129] Result visualization: Through image processing technology, the results of epileptic focus localization are superimposed on the MRI image and displayed to clinicians in a visual manner.

[0130] Report generation: Using natural language processing (NLP) algorithms, a detailed report is generated based on the model analysis results. The report includes confidence analysis of epileptic foci, feature weight distribution used in decision-making, epileptic foci location, and resection recommendations.

[0131] Output: Epilepsy focus localization results and detailed reports in image form for doctors’ reference.

[0132] Specific significant technical progress brought by the system provided by the embodiment of the present invention

[0133] 1. Significantly improved accuracy: Compared with traditional methods based on visual assessment or simple SUV values, this system can significantly improve the accuracy of epileptic focus identification and positioning by extracting and optimizing multi-dimensional features and combining machine learning algorithms.

[0134] 2. Automated analysis: The system uses automated image processing and feature selection methods to reduce the impact of subjective judgment on the results and ensure the objectivity and consistency of the analysis process.

[0135] 3. Real-time and efficiency improvement: Combined with efficient preprocessing and analysis algorithms, the system can quickly process large amounts of data and generate high-precision epileptic focus localization results, greatly shortening the time from data acquisition to result output.

[0136] 4. Multi-dimensional feature analysis: By introducing multi-dimensional features such as texture features and spatial metabolic differences, the system can capture richer information related to the epileptogenic focus and provide more in-depth analysis support.

[0137] 5. Visual reports support clinical decision-making: The generated visual reports can intuitively display the location of the epileptic focus and provide detailed analysis and resection recommendations, which greatly supports the doctor's clinical decision-making.

[0138] like Figure 4 As shown, an embodiment of the present invention provides a system for selecting a quantitative feature of an epilepsy focus based on 18F-FDG PET imaging, and a method for selecting a quantitative feature of an epilepsy focus based on 18F-FDG PET imaging, comprising the following steps:

[0139] Step 1: Data acquisition and registration: Obtain the patient's brain 18F-FDG PET image through the PET scanning device, and use image registration technology to spatially align the PET image with structural images such as MRI;

[0140] Step 2, image preprocessing: standardize, denoise, and normalize PET images to ensure data consistency and quality in subsequent analysis;

[0141] Step 3: Multi-dimensional feature extraction: Use image processing algorithms to extract different types of image features, including metabolic features, texture features, and spatial features;

[0142] Step 4, feature selection: Through statistical screening, recursive feature elimination and L1 regularization, the feature space is gradually reduced to finally obtain the optimal feature set for identifying epileptic foci;

[0143] Step 5: Model training and prediction: Use machine learning models such as support vector machines or random forests for training, and predict the location of the epileptogenic focus in the patient's brain based on the selected feature set;

[0144] Step 6: Result visualization and report generation: The analysis results are presented in the form of images, and a detailed epileptic focus localization report is generated to provide diagnostic support for clinicians.

[0145] like Figure 3As shown, the following is the specific implementation plan for each step.

[0146] S1, Data acquisition and registration

[0147] Objective: To collect 18F-FDG PET imaging data of patients and spatially align them with structural images such as MRI to ensure the consistency of spatial position in subsequent analysis.

[0148] Data collection: PET scanning equipment is used to obtain 18F-FDG PET images of the patient's brain and record the metabolic data of each brain region. MRI data is collected through corresponding MRI equipment to provide high-resolution structural images to show the anatomical structure of the brain.

[0149] Image registration: Use a rigid registration or elastic registration algorithm based on image features to ensure the spatial consistency of PET images and MRI images. Mutual information (MI) or cross entropy methods can be used to align multimodal images. This registration method can ensure that the spatial coordinates of PET and MRI are completely consistent while maintaining the integrity of the original image data, facilitating subsequent analysis.

[0150] Output: Registered PET and MRI image data, which serve as input for subsequent analysis.

[0151] S2, image preprocessing

[0152] Objective: To perform necessary preprocessing on PET images, including standardization, denoising and normalization, to ensure the consistency and reliability of the data.

[0153] Image normalization: Use normalization algorithms to adjust PET images to the same data range, such as using maximum and minimum normalization or Z-Score normalization, to remove differences caused by different scanning instruments and imaging conditions and ensure that the data is processed under the same scale.

[0154] Noise removal: Use denoising techniques such as Gaussian filtering or median filtering to smooth PET images, remove noise from the image, enhance image quality, and facilitate subsequent feature extraction.

[0155] Quality check: Perform quality assessment on preprocessed images to ensure image clarity and consistency and avoid noise or imaging errors that may affect the analysis results.

[0156] Output: preprocessed normalized and denoised PET image data.

[0157] S3, multi-dimensional feature extraction

[0158] Features of multiple dimensions related to epileptic foci are extracted from preprocessed PET images, including metabolic features, texture features, and spatial features.

[0159] Metabolic feature extraction: Extract the standardized uptake value (SUV) and calculate the average uptake value of each region of interest (ROI). SUV can reflect the local metabolic level, especially in the epileptic focus area, where metabolic activity is usually significantly different from that of surrounding tissues.

[0160] Texture feature extraction: Use image processing algorithms such as gray-level co-occurrence matrix (GLCM) to extract texture features. Features include contrast, energy, homogeneity, etc. Through this texture information, the system can evaluate the metabolic complexity of brain areas and help identify epileptic foci.

[0161] Spatial feature extraction: Analyze the metabolic differences between adjacent brain regions and identify hot spots of local metabolic abnormalities. Through spatial convolution algorithms or gradient-based analysis, extract the spatial relationship features of each brain region to further improve the accuracy of epileptic focus localization.

[0162] Output: The extracted multi-dimensional features, including metabolic, texture, and spatial features, form a feature vector.

[0163] S4, feature selection

[0164] Through machine learning methods, the feature space is gradually reduced, redundant features are removed, and the most relevant feature sets are retained to optimize the accuracy of epileptic focus identification.

[0165] Initial feature screening: The extracted features are preliminarily screened by statistical methods (such as chi-square test or t-test) to eliminate features that are irrelevant to the epileptic focus or have little contribution, thereby reducing the dimension of the feature space.

[0166] Recursive Feature Elimination (RFE): Use the RFE algorithm to perform recursive feature elimination to gradually remove unimportant features. In each iteration, a simple machine learning model (such as SVM or decision tree) is trained to remove the least relevant features according to their weights until the most useful features remain.

[0167] L1 regularization (LASSO) selection: Further feature selection is performed through LASSO regression. L1 regularization automatically eliminates redundant features and prevents model overfitting by imposing sparsity constraints.

[0168] Output: The optimal feature set after screening for model training and prediction.

[0169] S5, model training and prediction

[0170] Based on the selected optimal feature set, the machine learning model is trained to predict the location of epileptic foci in the patient's brain images.

[0171] Model training: Use common supervised learning models such as support vector machines (SVM) or random forests for training. Optimize model parameters through cross-validation and other techniques to ensure the generalization ability of the model.

[0172] Prediction and verification: The data after feature selection is input into the trained model, and the model predicts the location of epileptic foci in the PET images of new patients based on the learned rules and patterns. By generating probability distribution, the model can predict the nature of each area as an epileptic focus.

[0173] Output: Generate a probability heat map of epileptic foci and identify the locations of epileptic foci.

[0174] S6, Result visualization and report generation

[0175] Purpose: To present the analysis results in an intuitive graphical form and generate a detailed diagnostic report to help doctors make clinical decisions.

[0176] Specific implementation:

[0177] Result visualization: The probability heat map of the epileptic focus is superimposed on the MRI image through image processing technology to generate an epileptic focus localization map in three-dimensional space, showing the specific location and nature of the epileptic focus.

[0178] Report generation: Generate detailed reports based on prediction results using natural language processing (NLP) algorithms. The report content includes:

[0179] 1. The location of the epileptic focus and its confidence level.

[0180] 2. The features involved in decision making and their weights, explaining how the model reached its conclusion.

[0181] 3. Corresponding surgical resection recommendations or other clinical operation recommendations.

[0182] Output: intuitive epileptic focus localization map and detailed diagnostic report for clinicians' reference.

[0183] The significant technical progress achieved by the method provided in the embodiment of the present invention is:

[0184] 1. Automation and objectivity: The system automatically processes image data, reducing the errors caused by doctors’ subjective judgment and improving the consistency and accuracy of analysis.

[0185] 2. Efficient feature extraction and selection: Through the combination of multi-dimensional feature extraction and machine learning algorithms, the system can extract the most informative features, significantly improving the accuracy of epileptic focus localization.

[0186] 3. Real-time feedback and result display: The system can quickly generate a probability heat map of epileptic focus location and display it intuitively to doctors, greatly improving the efficiency of clinical diagnosis.

[0187] 4. Personalized report generation: The system provides personalized diagnostic reports, showing feature decision weights and model explanations, to help doctors better understand positioning results and optimize clinical operations.

[0188] The following is a specific implementation plan based on the above modules, which combines specific algorithms and programming processes to describe in detail the working principles and steps of each module.

[0189] 1) Image acquisition module-implementation solution

[0190] Algorithm: Image registration algorithm (such as mutual information-based registration algorithm)

[0191] Programming implementation:

[0192] -Registration of PET and MRI images using the `SimpleITK` library in Python. `SimpleITK` provides simple image processing functions, especially suitable for image registration tasks.

[0193]

[0194]

[0195] 2) Image preprocessing module - implementation algorithm: Programming implementation using standardization, denoising and normalization processing techniques:

[0196]

[0197]

[0198] 3) Feature extraction module-implementation solution

[0199] Algorithms: Extract SUV (standardized uptake value), texture features (such as gray-level co-occurrence matrix GLCM), local metabolic features

[0200] Programming implementation:

[0201]

[0202]

[0203] 4) Feature selection module-implementation solution

[0204] Algorithm: Feature importance analysis based on support vector machine recursive feature elimination (SVM-RFE) or random forest

[0205] Programming implementation:

[0206]

[0207]

[0208] 5) Epilepsy focus localization and analysis module - implementation algorithm: Based on the support vector machine (SVM) or random forest classification algorithm, generate probability heat map programming implementation:

[0209]

[0210]

[0211] 6) Result display and reporting module-implementation plan

[0212] Algorithm: Visualize results and generate detailed analysis reports showing feature weights, confidence levels, and surgical recommendations

[0213] Programming implementation:

[0214]

[0215] Through the above implementation scheme, the system can perform image acquisition, data processing, feature extraction, feature selection and positioning analysis, and generate the final analysis report. Each module is combined with specific algorithms and programming implementations, so that the entire system can accurately analyze and locate epileptic foci, providing valuable diagnostic support for clinicians.

[0216] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. It can be understood by a person of ordinary skill in the art that the above-mentioned devices and methods can be implemented using computer executable instructions and / or contained in a processor control code, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. Such code is provided on the carrier medium. The device and its modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, and can also be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0217] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principle of the present invention should be covered by the protection scope of the present invention.

Claims

1. A quantitative feature selection system for epileptogenic focus based on 18F-FDG PET imaging, characterized in that: include: Image acquisition module, used to collect the patient's 18F-FDG PET imaging data and perform registration with structural images such as MRI to improve the accuracy of spatial positioning; Image preprocessing module, used to preprocess the original PET images, including standardization, denoising and normalization; Feature extraction module, used to extract potential features of epileptogenic focus using multiple algorithms, the feature dimensions include standardized uptake value SUV, spatial texture features, local metabolic features; Feature selection module, used to perform feature selection using machine learning algorithms; An epileptic focus localization and analysis module is used to analyze the patient's PET data by combining the selected features through a classification model to generate a probability heat map of the epileptic focus and display the location of the epileptic focus; the classification model is a support vector machine SVM or a random forest; The result display and reporting module is used to visualize the analysis results, show the location of the epileptogenic focus to clinicians in the form of images, and generate a detailed analysis report, including the location of the epileptogenic focus and its confidence level; Characteristics of participation in decision making and their weight; Corresponding surgical resection recommendations.

2. The system for quantitative feature selection of epileptogenic focus based on 18F-FDG PET imaging according to claim 1, characterized in that: The image acquisition module uses standard PET scanning equipment to obtain metabolic imaging data of the patient's brain to ensure the clarity and integrity of the imaging; uses MRI scanning to obtain high-resolution structural images to provide accurate spatial information of various brain regions; and aligns the PET image with the MRI image based on an image registration algorithm, which includes rigid registration and elastic registration to ensure the spatial consistency of the two modal images.

3. The system for quantitative feature selection of epileptogenic focus based on 18F-FDG PET imaging according to claim 1, characterized in that: The preprocessing in the image preprocessing module includes: Image registration: Multimodal image registration is performed using PET and MRI data, and algorithms such as mutual information MI and normalized cross entropy are used to ensure spatial consistency; Image normalization: PET data is normalized by a normalization algorithm to adjust the image data to the same range and eliminate differences caused by equipment and imaging conditions. Maximum and minimum normalization or Z-Score normalization can be used; Noise removal: Use Gaussian filter or median filter to smooth the noise in the image and improve image clarity.

4. The system for quantitative feature selection of epilepsy focus based on 18F-FDG PET imaging according to claim 1, characterized in that: The feature dimensions extracted by the feature extraction module include: Standardized uptake value SUV extraction: Calculate the SUV value of each brain region to represent the metabolic activity of each brain region; This can be performed based on the division of the region of interest (ROI) using the standardized uptake value calculation method; Spatial texture feature extraction: The texture features of the brain region are extracted through the gray-level co-occurrence matrix GLCM to reflect the metabolic state of the local brain tissue; the extracted features include contrast, energy, and homogeneity; Local metabolic characteristics: Calculate the metabolic differences between adjacent brain regions, analyze the changes in metabolic levels between different regions, and find hot spots of metabolic abnormalities; use spatial convolution algorithms or gradient-based analysis algorithms to calculate metabolic differences.

5. The system for quantitative feature selection of epileptogenic focus based on 18F-FDG PET imaging according to claim 1, characterized in that: The feature selection module performs feature selection including: Initial feature screening: The extracted features were preliminarily screened using statistical methods, including chi-square test and t-test, to remove features that were not related to the epileptic focus and retain highly relevant features; Recursive Feature Elimination (RFE): Using the RFE algorithm, the model is gradually trained and unimportant features are removed to repeatedly screen features. The most important feature set is retained by removing features with less influence in each iteration. L1 regularized LASSO selection: LASSO regression is used for further feature compression.

6. The system for selecting quantitative features of epileptogenic foci based on 18F-FDG PET imaging according to claim 1, characterized in that: Results visualization in the result display and report module: Through image processing technology, the results of epileptic focus localization are superimposed on the MRI image and displayed to clinicians in a visual way; Report generation: Use the natural language processing algorithm NLP to generate a detailed report based on the model analysis results; the report includes confidence analysis of the epileptic focus, feature weight distribution used in decision-making, epileptic focus location and resection recommendations.

7. A method for selecting quantitative features of epileptic focus based on 18F-FDG PET imaging according to any one of claims 1 to 6, characterized in that: The following steps are involved: Step 1: Data acquisition and registration: Obtain the patient's brain 18F-FDG PET image through the PET scanning device, and use image registration technology to spatially align the PET image with the MRI structural image; Step 2, image preprocessing: standardize, denoise, and normalize PET images to ensure data consistency and quality in subsequent analysis; Step 3: Multi-dimensional feature extraction: Use image processing algorithms to extract different types of image features, including metabolic features, texture features, and spatial features; Step 4, feature selection: Through statistical screening, recursive feature elimination and L1 regularization, the feature space is gradually reduced to obtain the optimal feature set for identifying epileptic foci; Step 5: Model training and prediction: Use support vector machine or random forest machine learning model for training and predict the location of the epileptogenic focus in the patient's brain based on the selected feature set; Step 6: Result visualization and report generation: The analysis results are presented in the form of images, and a detailed epileptic focus localization report is generated to provide diagnostic support for clinicians.

8. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method for quantitative feature selection of epileptogenic focus based on 18F-FDG PET imaging as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the method for quantitative feature selection of epileptogenic focus based on 18F-FDG PET imaging as described in any one of claims 1 to 6.

10. An information data processing terminal, characterized in that: The information data processing terminal includes the quantitative feature selection system for epilepsy focus based on 18F-FDG PET imaging as described in claim 7.

Images