Medical image processing system and method for transcranial magnetic stimulation

Abstract

The application relates to the technical field of transcranial magnetic stimulation medical image processing, and discloses a transcranial magnetic stimulation medical image processing system and method. The system comprises an image acquisition module, which is used for collecting multi-modal medical image data and defining image partitions according to clinical diagnosis requirements; a feature analysis module, which is used for extracting stimulation-related features and constructing a feature mapping relationship between the image partitions and image positioning stimulation areas; a region fusion module, which is used for screening core regions and analyzing parameter fusion matching degrees to obtain parameter focus fusion information; a mapping relationship updating module, which is used for updating the feature mapping relationship based on the parameter focus fusion information; and a processing execution module, which is used for determining processing targets and processes and generating a processing execution library according to the updated mapping relationship. The method corresponds to the steps of the modules of the system. The scheme can improve the accuracy and efficiency of transcranial magnetic stimulation medical image processing and meet the clinical diagnosis requirements.

Description

Technical Field

[0001] This invention relates to the field of transcranial magnetic stimulation (TMS) medical image processing technology, specifically to a TMS medical image processing system and method. Background Technology

[0002] Transcranial magnetic stimulation (TMS), as a non-invasive neuromodulation technique, is increasingly widely used in the diagnosis and treatment of neuropsychiatric disorders. With the development of this technology, the acquisition and analysis of multimodal medical imaging data during TMS has become crucial. However, there are currently many problems in the processing of medical images related to TMS.

[0003] In terms of feature parsing and mapping relationship construction, existing technologies struggle to efficiently retrieve corresponding parameters and states from medical image data when extracting stimulus-related features. This results in generating unrelated parameter parsing results and accurately determining whether they represent the target parameter parsing results, thus hindering the identification of image-localized stimulus regions. Furthermore, when constructing feature mapping relationships between image partitions and image-localized stimulus regions, it may fail to fully utilize information from the temporal feature domain, spatial feature domain, and associated feature domain within the image partition, as well as the descriptive information and category of the image-localized stimulus regions. This leads to incomplete mapping relationships, impacting subsequent analysis and processing of the image-localized stimulus regions.

[0004] The implementation of the region fusion module also has shortcomings. Traditional methods, when performing cluster analysis on image partitions and image-localized stimulus areas to screen core regions, may fail to accurately cluster according to stimulus type, functional type, or effect, leading to inaccurate identification of core regions. Furthermore, the methods used to calculate feature similarity between core features and fusion matching degree between parameters may be unscientific, resulting in fusion matching degree calculations that do not accurately reflect the actual mapping relationship between parameters, thus affecting the acquisition of parameter-focused fusion information.

[0005] When updating the feature mapping relationship between parameter focusing fusion information and image partitions and image localization stimulation areas, the mapping relationship update module may not be able to accurately extract the temporal distribution probability of each parameter, set the target path and perform fitting, resulting in inaccurate calculation of the fusion occurrence probability. Consequently, when comparing with the reference situation of the feature mapping relationship between image partitions and image localization stimulation areas, it is unable to accurately mark the difference value and classify and update the mapping relationship, affecting the timely update and optimization of the feature mapping relationship.

[0006] When determining processing targets and procedures and generating a processing execution library, the processing execution module may fail to accurately extract processing targets and sort them by probability of occurrence based on the updated feature mapping relationship. This results in an inadequately designed processing flow and a processing execution library that does not adequately meet practical application needs. These problems severely restrict the accuracy and efficiency of transcranial magnetic stimulation (TMS) technology in medical image processing, necessitating a more advanced and comprehensive medical image processing system and method to address these issues. Summary of the Invention

[0007] The purpose of this invention is to provide a medical image processing system and method for transcranial magnetic stimulation to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a transcranial magnetic stimulation medical image processing system, the system comprising:

[0009] The image acquisition module is used to collect multimodal medical image data generated during transcranial magnetic stimulation and to define the image partitions of the medical image data relative to the nerve stimulation area according to clinical diagnostic needs.

[0010] The feature parsing module is used to extract stimulus-related features from medical image data and construct a feature mapping relationship between image partitions and image-localized stimulus areas.

[0011] The region fusion module is used to select core regions from image partitions and image localization stimulation areas, and analyze the fusion matching degree between multiple parameters within the image localization stimulation area based on the core regions to obtain parameter focusing fusion information;

[0012] The mapping relationship update module is used to perform information verification on the image localization stimulation area based on the parameter focusing fusion information, identify the update status of the feature mapping relationship between the image partition and the image localization stimulation area, and update the feature mapping relationship between the image partition and the image localization stimulation area according to the update status of the feature mapping relationship.

[0013] The processing execution module is used to determine the processing objectives and processes of medical image data based on the feature mapping relationship between the updated image partitions and image localization stimulation areas, and to generate a processing execution library.

[0014] Preferably, the image acquisition module is implemented in the following ways:

[0015] For medical imaging data at any moment during transcranial magnetic stimulation, obtain the modality recognition model corresponding to the medical imaging data;

[0016] Use a modality recognition model to classify medical image data by modality, and obtain at least one modality category;

[0017] Identify the temporal, spatial, and correlational features in medical image data under the corresponding modality category to form an image feature set. A modality of medical image data corresponds to an image feature set containing temporal, spatial, and correlational features.

[0018] Each individual modality category is processed separately, and the temporal features, spatial features, and correlation features in the image feature set are analyzed separately. The temporal feature domain, spatial feature localization domain, and correlation feature association domain corresponding to the image feature set are obtained in sequence, and used as image partitions of medical image data relative to the neural stimulation region.

[0019] Preferably, the implementation methods for obtaining the temporal feature domain, spatial feature localization domain, and associated feature association domain corresponding to the image feature set also include:

[0020] The temporal, spatial, and correlation features in the image feature set are integrated according to modal categories to obtain a multimodal integration result;

[0021] Extract the temporal feature groups from the modality integration results, and compare the temporal feature groups with the feature library to obtain the temporal feature domain;

[0022] The localization error rate of spatial features and the correlation tightness of associated features are extracted from the modal integration results. Based on the localization error rate of spatial features and the correlation tightness of associated features, the modal integration results are divided to obtain the spatial feature localization domain and the associated feature association domain.

[0023] Preferably, the implementation of defining image partitions of medical image data relative to the neural stimulation region also includes:

[0024] The imaging regions are defined, and the frequency and duration of neural stimulation images within each region are analyzed. The imaging regions are then fitted according to their frequency and duration to construct a mapping relationship between the imaging regions and actual clinical needs.

[0025] Preferably, the feature parsing module is implemented in the following ways:

[0026] The parameters and states corresponding to the medical image data are called to generate multiple unrelated parameter parsing results. The unrelated parameter parsing results represent parameters and states that are not associated with features in the image partition.

[0027] Determine whether the parsing results of multiple unrelated parameters are the parsing results of the target parameter. If they are the parsing results of the target parameter, then the parsing results of the target parameter are regarded as the image localization stimulation area of ​​the medical image data.

[0028] The target parameter analysis results must meet two core conditions: First, they must be able to accurately match the temporal feature domain, spatial feature localization domain, and related feature association domain of the image partition, such as the parameter time series fluctuating in accordance with the temporal feature domain, and the spatial parameters coinciding with the coordinate range of the localization domain; Second, they must meet the clinical diagnostic criteria for defining the image localization stimulation area, including that the parameter values ​​are within the clinically effective range (such as the stimulation frequency and intensity conforming to the treatment guidelines for the corresponding disease), and that the anatomical location corresponding to the analysis results is consistent with the clinical target brain region.

[0029] Preferably, the method for constructing the feature mapping relationship between image partitions and image localization stimulus areas includes: using the information representation of the temporal feature domain, spatial feature localization domain, and associated feature association domain existing in the image partition, the descriptive information of the image localization stimulus area, and the category of the image localization stimulus area to construct the feature mapping relationship between the image partition and the image localization stimulus area.

[0030] Preferably, the implementation methods of the regional fusion module include:

[0031] The image partitions and image localization stimulation areas are clustered according to stimulation type, functional type, and function. The largest cluster center after the clustering analysis is set as the core region.

[0032] Extract the core features of the core region, calculate the feature similarity between each core feature, and set a common sequence related to the feature similarity between each core feature;

[0033] By utilizing the common sequences related to the feature similarity between the core features, parameters existing in the common sequences are extracted, and the longest common subsequence between the parameters is set. The length of the longest common subsequence is then set as the fusion matching degree between the parameters.

[0034] The degree of fusion and matching between each parameter is determined according to the time distribution probability of each parameter, and the parameter is set to focus on fusion information.

[0035] Preferably, the implementation of the mapping relationship update module includes:

[0036] Extract the temporal distribution probability of each parameter from the parameter-focused fusion information; set the target path of the parameter-focused fusion information according to the time period corresponding to the temporal distribution probability of each parameter.

[0037] The target paths of each parameter in the parameter-focused fusion information are fitted to obtain the fitted target paths. The probability values ​​of the fitted target paths in each time period are set as the fusion occurrence probability of the parameter-focused fusion information.

[0038] The probability of fusion of parameter-focused fusion information is compared with the reference of the feature mapping relationship between image partitions and image localization stimulation areas. The difference values ​​are marked. The difference values ​​are used to reflect the update status of the feature mapping relationship between image partitions and image localization stimulation areas. The feature mapping relationship between image partitions and image localization stimulation areas is classified according to the difference values, and the update of the feature mapping relationship between image partitions and image localization stimulation areas is completed.

[0039] Preferably, the implementation method of the processing execution module is as follows:

[0040] Based on the feature mapping relationship between the updated image partitions and image localization stimulation areas, processing targets are extracted from the medical image data. The processing targets are sorted according to their occurrence probability to obtain the processing flow in the medical image data. The processing targets and processing flow are combined in a structured form to obtain the processing execution library.

[0041] Preferably, the present invention further includes a transcranial magnetic stimulation (TMS) medical image processing method, applied to the aforementioned TMS medical image processing system, the method comprising the following steps:

[0042] Multimodal medical imaging data generated during transcranial magnetic stimulation are collected, and image partitions of the medical imaging data relative to the nerve stimulation area are defined according to clinical diagnostic needs.

[0043] Extract stimulus-related features from medical imaging data and construct a feature mapping relationship between image partitions and image-localized stimulus areas;

[0044] Core regions are selected from image partitions and image localization stimulation areas, and the fusion matching degree between multiple parameters within the image localization stimulation area is analyzed based on the core regions to obtain parameter focusing fusion information;

[0045] Based on the parametric focusing fusion information, information verification is performed on the image localization stimulation area, the update status of the feature mapping relationship between the image partition and the image localization stimulation area is identified, and the feature mapping relationship between the image partition and the image localization stimulation area is updated according to the update status of the feature mapping relationship.

[0046] Based on the updated feature mapping relationship between image partitions and image localization stimulation areas, the processing objectives and processes of medical image data are determined, and a processing execution library is generated.

[0047] Compared with the prior art, the beneficial effects of the present invention are:

[0048] In the image acquisition stage, medical image data at any moment during transcranial magnetic stimulation is used to obtain the corresponding modality recognition model. The data is then classified into modalities, and temporal, spatial, and related features are identified to form an image feature set. Further analysis yields the temporal feature domain, spatial feature localization domain, and related feature association domain, which serve as image partitions. Simultaneously, the frequency and duration of nerve stimulation images within the image partitions are analyzed to construct a mapping relationship to actual clinical needs. This allows for precise definition of image partitions based on clinical diagnostic requirements, making the partitions more closely match the actual nerve stimulation areas and providing accurate basic data for subsequent processing.

[0049] The feature parsing module calls the parameters and status of medical image data to generate unrelated parameter parsing results, determines whether it is the target result, and identifies the image localization stimulus area. It then constructs a feature mapping relationship based on the temporal, spatial, and associated feature domain information of the image partition, as well as the descriptive information and category of the image localization stimulus area. This achieves efficient extraction and accurate mapping of stimulus-related features, ensuring that the determination of the image localization stimulus area and the construction of the mapping relationship are scientific and reasonable, and providing a reliable basis for subsequent analysis.

[0050] The region fusion module clusters image partitions and image localization stimulus areas according to stimulus type, functional type, and function to determine core regions, extracts core features to calculate similarity, and sets common sequences and longest common subsequences to determine the fusion matching degree between parameters, thereby obtaining parameter focusing fusion information. This method can accurately screen core regions, scientifically analyze the fusion matching degree between parameters, and make parameter focusing fusion information more consistent with the actual situation, thus improving the accuracy of the analysis of parameter mapping relationships within the image localization stimulus area.

[0051] The mapping relationship update module extracts the temporal distribution probability of parameters from the parameter focusing fusion information, sets the target path and fits it to obtain the fusion occurrence probability. It compares the reference status of the feature mapping relationship between the image partition and the image localization stimulation area, marks the difference value and classifies and updates the mapping relationship. This realizes the dynamic update and optimization of the feature mapping relationship between the image partition and the image localization stimulation area, ensuring the timeliness and accuracy of the feature mapping relationship, and enabling the system to better adapt to changes during transcranial magnetic stimulation.

[0052] The processing execution module extracts processing targets based on the updated feature mapping relationship and sorts them according to their occurrence probability to obtain the processing flow. This flow is then combined in a structured form to generate a processing execution library, ensuring that the determination of processing targets and flows is scientifically sound. The generated processing execution library can efficiently guide the processing of medical image data, improving processing efficiency and accuracy, and meeting the actual needs of clinical diagnosis. In summary, this invention, through the collaborative work of its various modules, comprehensively improves the accuracy, efficiency, and adaptability of transcranial magnetic stimulation (TMS) medical image processing, providing stronger technical support for the diagnosis and treatment of related diseases. Attached Figure Description

[0053] Figure 1 This is a schematic diagram illustrating the working principle of the transcranial magnetic stimulation medical image processing system and method described in this invention.

[0054] Figure 2 This is a schematic diagram illustrating the working principle of the image acquisition module.

[0055] Figure 3 This is a schematic diagram illustrating the working principle of the regional fusion module.

[0056] Figure 4 A diagram illustrating the working principle of the mapping relationship update module. Detailed Implementation

[0057] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] Please see Figures 1-4 This invention relates to a medical image processing system for transcranial magnetic stimulation, the core components of which include an image acquisition module, a feature parsing module, a region fusion module, a mapping relationship update module, and a processing execution module. Specific implementation methods are as follows:

[0059] The image acquisition module is used to collect multimodal medical image data generated during transcranial magnetic stimulation and to define the image partitions of the medical image data relative to the nerve stimulation area according to clinical diagnostic needs.

[0060] The feature parsing module is used to extract stimulus-related features from medical image data and construct a feature mapping relationship between image partitions and image-localized stimulus areas.

[0061] The region fusion module is used to select core regions from image partitions and image localization stimulation areas, and analyze the fusion matching degree between multiple parameters within the image localization stimulation area based on the core regions to obtain parameter focusing fusion information;

[0062] The mapping relationship update module is used to perform information verification on the image localization stimulation area based on the parameter focusing fusion information, identify the update status of the feature mapping relationship between the image partition and the image localization stimulation area, and update the feature mapping relationship between the image partition and the image localization stimulation area according to the update status of the feature mapping relationship.

[0063] The processing execution module is used to determine the processing objectives and processes of medical image data based on the feature mapping relationship between the updated image partitions and image localization stimulation areas, and to generate a processing execution library.

[0064] Example 1:

[0065] The specific implementation of the image acquisition module is as follows: This module is used to acquire multimodal medical image data generated during transcranial magnetic stimulation (TMS) and define image partitions of the medical image data relative to the nerve stimulation area according to clinical diagnostic needs. In practical applications, for medical image data at any given moment during TMS, it is necessary to acquire the modality recognition model corresponding to that medical image data. This modality recognition model can be trained using a large amount of labeled multimodal medical image data, and it can accurately classify different types of medical image data modally.

[0066] The acquired modality recognition model is used to classify medical image data into at least one modality category. Common modality categories include magnetic resonance imaging (MRI), functional magnetic resonance imaging (fMRI), positron emission tomography (PET), and computed tomography (CT). Different modality categories reflect the physiological and pathological characteristics of the nerve-stimulated region from different perspectives. After completing the modality classification, it is necessary to identify the temporal, spatial, and correlational features present in the medical image data under the corresponding modality category to form an image feature set.

[0067] The identification of temporal features primarily involves analyzing the changing patterns of medical imaging data over time. For example, in fMRI images, the temporal series characteristics of neuronal activity are obtained by monitoring changes in blood oxygenation level-dependent signals, including fluctuations in signal intensity over time and oscillations at specific frequencies. The identification of spatial features involves determining the geometric attributes of various tissues and structures in medical images, such as their location, shape, and size. For instance, image segmentation techniques are used to separate the neural stimulation region from surrounding tissues, thereby obtaining its spatial coordinates, volume, shape descriptors, and other features. The identification of associative features focuses on the corresponding mapping relationships between different modalities of images and the connection strength between the neural stimulation region and surrounding tissues. For example, when fusing MRI and PET images, it is necessary to establish a spatial registration mapping relationship between the two, while simultaneously analyzing the functional connectivity strength between the neural stimulation region and other brain regions.

[0068] After forming the image feature set, it is necessary to analyze the temporal, spatial, and correlation features within the set to sequentially obtain the corresponding temporal feature domain, spatial feature localization domain, and correlation feature association domain. These domains are then used as image partitions of the medical image data relative to the neural stimulation region. When analyzing the temporal features, time series analysis methods, such as Fourier transform and wavelet transform, can be used to convert the temporal features to the frequency or time-frequency domain for analysis, thereby obtaining their distribution patterns over time and determining the temporal feature domain. For example, the frequency components of the temporal features can be obtained through Fourier transform, thus determining the main oscillation frequency range as part of the temporal feature domain.

[0069] When analyzing spatial features, techniques from image recognition and computer vision, such as edge detection, contour extraction, and 3D reconstruction, are used to accurately determine the location and extent of the spatial features, thus obtaining the spatial feature localization domain. For example, for a neural stimulation region in an MRI image, its boundary is extracted using an edge detection algorithm, and then 3D reconstruction technology is used to construct its specific location and shape in 3D space, thereby determining the spatial feature localization domain.

[0070] When analyzing association features, the degree of association and transmission paths are analyzed by calculating the number of phase mapping relationships and constructing graphical models, thus forming the association domain of association features. For example, when analyzing functional connectivity, the number of time-series phase mapping relationships between neural stimulation regions and other brain regions is calculated to construct a functional connectivity network, thereby determining the association strength and transmission paths between each brain region, which serve as the content of the association domain of association features.

[0071] Furthermore, when defining the image partitions of medical imaging data relative to the neural stimulation region, further analysis of these image partitions is required. Specifically, this involves analyzing the image frequency and duration of neural stimulation within each image partition. Image frequency refers to the number of times a neural stimulus appears in that image partition within a given time frame; image duration refers to the duration of each neural stimulus within that image partition.

[0072] Image partitions are fitted according to image frequency and duration to construct a mapping relationship between image partitions and actual clinical needs. Various mathematical methods can be used in the fitting process, such as linear regression, nonlinear regression, and neural network fitting. This mapping relationship allows for the correspondence between image partitions and specific clinical diagnostic needs; for example, it identifies which image partitions are associated with specific neurological dysfunctions or treatment effects, thus providing more targeted information for clinical diagnosis.

[0073] In practical applications, this image acquisition module can interface with various medical imaging devices, such as MRI scanners and PET scanners, to acquire multimodal medical image data in real time during transcranial magnetic stimulation. Furthermore, this module can flexibly adjust the definition criteria and methods of image partitions according to different clinical diagnostic needs, to meet the requirements of different disease diagnoses and treatments. For example, in diagnosing Parkinson's disease, more attention may be paid to image partitions related to the motor cortex; while in diagnosing depression, more attention may be paid to image partitions related to the limbic system.

[0074] Through the above series of operations, the image acquisition module can accurately collect multimodal medical image data and define reasonable image partitions according to clinical diagnostic needs, providing high-quality data support for subsequent feature analysis, region fusion and other modules, thereby ensuring that the entire transcranial magnetic stimulation medical image processing system can effectively assist clinical diagnosis and treatment.

[0075] Example 2:

[0076] The specific implementation of the feature parsing module is as follows: This module is used to extract stimulus-related features from medical image data and construct a feature mapping relationship between image partitions and image-localized stimulus areas. In actual operation, the feature parsing module first needs to call the parameters and states corresponding to the medical image data. These parameters and states cover multiple dimensions, such as parameters of the medical image itself, including the intensity values ​​of each pixel or voxel, image resolution, noise level, etc.; they also include relevant parameters during transcranial magnetic stimulation, such as stimulation frequency, intensity, pulse width, stimulation duration, etc., and also involve the working status parameters of the stimulation device, such as the position and angle of the coil.

[0077] After invoking these parameters and states, the system processes them independently, generating multiple unrelated parameter analysis results. "Unrelated" here means that these results have not yet been correlated with the defined temporal feature domain, spatial feature location domain, and associated feature domain within the image partition. For example, when processing a stimulus intensity parameter, only the numerical range or variation curve of the parameter may be obtained, without relating it to the spatial location or temporal series characteristics of a specific area within the image partition. Each parameter analysis result is an independent analysis of a single parameter or state, without involving cross-correlation with other features.

[0078] The system needs to determine whether the analysis results of these unrelated parameters are the analysis results of the target parameters. The determination process is based on preset multi-dimensional conditions, which are set according to clinical diagnostic needs and the application scenario of transcranial magnetic stimulation. For example, for the analysis results of stimulation frequency parameters, target conditions may include whether the frequency value is within the effective range for treating a certain type of disease (e.g., high-frequency stimulation for treating depression is usually set above 10Hz), and whether its trend conforms to the preset stimulation pattern (e.g., whether it is a continuous and stable frequency output). For the analysis results of image intensity parameters, target conditions may involve whether the intensity value exceeds the normal physiological range, and whether there are abnormal local high-intensity areas. If a parameter analysis result meets all the preset target conditions, it is considered as the image localization stimulation area of ​​the medical imaging data. For example, when a parameter analysis result shows that the stimulation intensity is within a specific treatment range, and the corresponding image area has abnormal neural activity signals, that area will be identified as the image localization stimulation area.

[0079] When constructing the feature mapping relationship between image partitions and image-localized stimulation areas, it is necessary to use the existing temporal feature domain, spatial feature localization domain, and associated feature domain within the image partition as the basic representation. The temporal feature domain contains the feature distribution of medical image data in the time dimension, such as the periodic fluctuation pattern of neural activity and the time delay of stimulus response. The spatial feature localization domain clarifies the specific location and range of each feature in three-dimensional space, such as the coordinate boundaries and volume of the neural stimulation area. The associated feature domain records the connection mapping relationship and mutual influence between different features, such as the functional connectivity strength between different brain regions and the transmission path of stimulus signals.

[0080] The descriptive information of the image-localized stimulation area includes its anatomical name (e.g., dorsolateral prefrontal cortex) and physiological response characteristics during stimulation (e.g., changes in blood oxygen levels). The categories of image-localized stimulation areas are based on their functional attributes, such as motor cortex image-localized stimulation areas, sensory cortex image-localized stimulation areas, and limbic system image-localized stimulation areas. The system establishes a multi-dimensional mapping mechanism to match the temporal feature domain information of the image partition with the temporal-related descriptive information of the image-localized stimulation area. For example, it correlates the high-frequency oscillation characteristics of neural activity within a certain time period in the image partition with the stimulation frequency parameters of the image-localized stimulation area during that time period. It also spatially registers the coordinate range of the spatial feature localization domain with the anatomical location of the image-localized stimulation area, ensuring consistency in their positions in three-dimensional space. Furthermore, it maps the functional connectivity strength in the association feature domain to the interaction mapping relationship between the image-localized stimulation area and other brain regions. For example, it maps the high connectivity strength characteristics between the image partition and the language center to the description of the influence of the image-localized stimulation area on language function.

[0081] In practical implementation, this mapping relationship can be achieved by constructing a data table or a graph structure model. In the data table, rows and columns correspond to the features of the image partition and the image-localized stimulation area, respectively, and the values ​​in the table represent the degree of correlation between the two. In the graph structure model, nodes represent the features of the image partition or the image-localized stimulation area, and the weights of the edges represent the strength of the correlation between the features. For example, for a specific image partition, its spatial feature localization domain identifies a region located in the left dorsolateral prefrontal cortex, the temporal feature domain shows a significant increase in blood oxygen levels in this region 200ms after stimulation, and the correlation feature correlation domain indicates a strong functional connection between this region and the hippocampus. The description information of the image-localized stimulation area is the left dorsolateral prefrontal cortex, the stimulation category is cognitive function regulation, and the stimulation parameters include a stimulation frequency of 10Hz and a stimulation intensity of 1.5T. At this time, the system will match the spatial location of the image partition with the anatomical location of the image-localized stimulation area, correspond the time point of blood oxygen increase in the temporal features with the time series of stimulation parameters, and correlate the hippocampal connectivity strength in the correlation features with the description of the effect of the image-localized stimulation area on memory function, thereby establishing a complete feature mapping relationship.

[0082] In practical applications, the feature parsing module needs to work closely with the image acquisition module to ensure the accuracy of the acquired image feature sets. Simultaneously, this module also needs to dynamically adjust the judgment conditions for the target parameter parsing results and the rules for constructing feature mapping relationships based on different transcranial magnetic stimulation application scenarios and clinical diagnostic needs. For example, in treating obsessive-compulsive disorder, more attention may be paid to the image localization stimulation area related to the orbitofrontal cortex; in this case, the target conditions will emphasize the matching of image features in this area with specific stimulation parameters. Conversely, in treating post-stroke sequelae, more attention may be paid to the remodeling of the motor cortex; the feature mapping relationship will emphasize the association between the image localization stimulation area and image features related to motor function.

[0083] Example 3:

[0084] The specific implementation of the region fusion module is as follows: This module is used to select core regions from image partitions and image localization stimulation areas, and analyze the fusion matching degree between multiple parameters within the image localization stimulation area based on the core regions to obtain parameter focusing fusion information. In practice, the region fusion module needs to perform cluster analysis on image partitions and image localization stimulation areas according to multiple dimensions. The relevant data are categorized according to three main dimensions: stimulus type, functional type, and effect function. Stimulation type includes different stimulus patterns such as single-pulse stimulation, repetitive-pulse stimulation, and burst stimulation, each with different time parameters and energy release characteristics; functional type involves the physiological function categories affected by neural stimulation, such as motor function, cognitive function, sensory function, and emotion regulation function; effect function refers to the specific effects of stimulation on neural tissue, such as excitatory effect, inhibitory effect, and plasticity regulation effect.

[0085] During cluster analysis, a clustering algorithm suitable for the characteristics of medical imaging data and stimulus parameters is used to process the feature vectors of image partitions and image-localized stimulus regions. For example, for each image partition, its feature vector may include statistical parameters of the temporal feature domain, geometric parameters of the spatial feature domain, and connection strength values ​​of the association feature domain; for each image-localized stimulus region, the feature vector may include stimulus type parameters, functional type identifiers, functional indicators, and related anatomical location parameters. By calculating the similarity or distance between these feature vectors, data points with similar characteristics are grouped into the same cluster. After clustering, the largest cluster center is determined, i.e., the central location of the cluster containing the most data points, and this is set as the core region. This core region represents the most representative part of the image partitions and image-localized stimulus regions, reflecting the common characteristics of most data points.

[0086] After identifying the core region, its core features need to be extracted. Core feature extraction is based on the feature vector of the core region, covering multiple aspects: morphological features include geometric attributes such as shape descriptors (e.g., roundness, complexity), volume, and surface area; physiological features involve physiological indicators such as signal intensity, metabolic level, and blood oxygen saturation in medical images; and stimulus parameter features include parameters such as stimulus frequency, intensity, pulse width, and stimulus duration corresponding to the core region. After extracting these core features, the feature similarity between each core feature is calculated. The method for calculating feature similarity varies depending on the feature type. For numerical features (e.g., stimulus intensity, volume), Euclidean distance or cosine similarity can be used; for categorical features (e.g., stimulus type, functional type), chi-square distance or mutual information can be used.

[0087] Based on the feature similarity among the core features, a common sequence related to the feature similarity among the core features is defined. This common sequence can be understood as a feature sequence segment shared by all core features, or a combination of features that reflects the common attributes of the core features. For example, if one core feature includes a stimulation frequency of 10Hz, a stimulation type of repetitive pulse stimulation, and an excitatory effect, while another core feature includes a stimulation frequency of 12Hz, a stimulation type of repetitive pulse stimulation, and an excitatory effect, then "repetitive pulse stimulation" and "excitatory effect" constitute the common sequence of these two core features.

[0088] By leveraging the common sequences related to feature similarity among core features, parameters are extracted from these common sequences. These parameters may include stimulus type parameters, function type parameters, effect function parameters, and related auxiliary parameters. The longest common subsequence among the parameters is defined, which is the longest subsequence appearing in two or more parameter sequences. The solution process employs dynamic programming, using a two-dimensional array to record solutions to subproblems, ultimately yielding the length of the longest common subsequence. This length is set as the fusion matching degree among the parameters. The magnitude of the fusion matching degree reflects the degree of correlation and fusion probability between parameters; a larger value indicates more commonalities among the parameters, making fusion processing easier.

[0089] The fusion matching degree between various parameters is set according to the temporal distribution probability of each parameter to define the parameter focusing fusion information. The temporal distribution probability of each parameter is obtained through statistical analysis, that is, by statistically analyzing the frequency or probability density of each parameter at different time points during transcranial magnetic stimulation. For example, the probability of a certain stimulation intensity parameter appearing within 0-10 seconds after the start of stimulation is 0.3, and the probability of appearing within 10-20 seconds is 0.5, etc. Based on the temporal distribution probability, the fusion matching degree is weighted so that parameters that are more likely to appear in time and their fusion mapping relationship occupy a more important position in the parameter focusing fusion information. The parameter focusing fusion information obtained in this way considers both the intrinsic correlation between parameters and their temporal distribution characteristics, and can more accurately reflect the actual fusion situation between multiple parameters within the image localization stimulation area.

[0090] In practical applications, the region fusion module needs to work closely with the feature parsing module to ensure the accuracy of the acquired features of image partitions and image-localized stimulation areas. Simultaneously, this module also needs to adjust the dimensions of cluster analysis, the extraction methods of core features, and the calculation rules of fusion matching degree according to different transcranial magnetic stimulation (TMS) treatment plans and clinical diagnostic needs. For example, in TMS treatment for depression, the focus may be more on the emotion regulation function parameters of image partitions and image-localized stimulation areas related to the limbic system. In this case, cluster analysis will emphasize the functional type and functional dimension, and the extraction of core features will focus more on physiological indicators and stimulation parameters related to emotion regulation. Conversely, in the treatment of Parkinson's disease, the focus may be more on the motor function parameters of image partitions and image-localized stimulation areas related to the motor cortex. Cluster analysis and core feature extraction will accordingly revolve around the motor function dimension.

[0091] Example 4:

[0092] The specific implementation of the mapping relationship update module is as follows: This module is used to verify the information of the image localization stimulation area based on the parameter focusing fusion information, update the feature mapping relationship between the image partition and the image localization stimulation area, and update the feature mapping relationship between the image partition and the image localization stimulation area according to the update status of the feature mapping relationship. In actual operation, the mapping relationship update module needs to extract the temporal distribution probability of each parameter from the parameter focusing fusion information. The parameter focusing fusion information contains multiple parameters within the image localization stimulation area and their fusion matching degree information, while the temporal distribution probability of each parameter reflects the probability of these parameters appearing at different time periods during transcranial magnetic stimulation.

[0093] The feature mapping relationship between the image partition and the image-localized stimulation area is constructed by the feature parsing module through a series of operations. The feature parsing module calls the parameters and states corresponding to the medical image data to generate multiple unrelated parameter parsing results that are not associated with the image partition features. These parameters cover the image's own pixel intensity and resolution, process parameters such as the frequency and intensity of transcranial magnetic stimulation, and device operating status parameters. The system determines whether these unrelated parameter parsing results are target parameter parsing results. The target parameter parsing results must accurately match the temporal feature domain, spatial feature localization domain, and associated feature association domain of the image partition, and meet the clinical diagnostic criteria for defining the image-localized stimulation area, including parameter values ​​within the clinically effective range and anatomical location consistent with the clinical target brain region. Those meeting these conditions are identified as image-localized stimulation areas. Using the temporal feature domain, spatial feature localization domain, and correlation feature association domain information in the image partition as representations, combined with the descriptive information and category of the image localization stimulus area, a multi-dimensional mapping mechanism is established to correspond the image partition features with the image localization stimulus area features, thereby forming a mapping relationship in the image localization stimulus area. These mapping relationships cover multiple dimensions such as spatial positional correspondence, temporal sequence association, and functional causal relationship.

[0094] For example, in a specific transcranial magnetic stimulation (TMS) scenario, the parameter focusing fusion information may involve three parameters: stimulation frequency, stimulation intensity, and changes in blood oxygen level in the imaging region. Specifically, the probability of a stimulation frequency of 10Hz within 0-10 seconds is 0.6, and the probability of 15Hz within 10-20 seconds is 0.7; the probability of a stimulation intensity of 1.2T within 0-10 seconds is 0.5, and the probability of 1.5T within 10-20 seconds is 0.8; and the probability of a blood oxygen level increase of 0.5% within 0-10 seconds is 0.4, and the probability of an increase of 0.8% within 10-20 seconds is 0.6. These temporal probability distributions are obtained through statistical analysis of historical data from multiple TMS procedures and reflect the temporal distribution patterns of the parameters.

[0095] The target path for parameter-focused fusion information is set according to the time period corresponding to the time distribution probability of each parameter. Setting the target path requires comprehensive consideration of the probability distribution of each parameter in different time periods, and the parameter combination with the highest probability is taken as the state of the target path within that time period. For example, in the 0-10 second time period, if the stimulation frequency is 10Hz (probability 0.6), the stimulation intensity is 1.2T (probability 0.5), and the blood oxygen level changes by 0.5% (probability 0.4), then the high-probability combination of stimulation frequency and stimulation intensity might be taken as the target path segment for that time period. In the 10-20 second time period, if the stimulation frequency is 15Hz (probability 0.7), the stimulation intensity is 1.5T (probability 0.8), and the blood oxygen level changes by 0.8% (probability 0.6), then the high-probability combination of these three parameters would be taken as the target path segment for that time period.

[0096] The target paths of each parameter in the parameter-focused fusion information are fitted to obtain the fitted target path. Various methods can be used for fitting, such as spline curve fitting and polynomial fitting, with the appropriate fitting method selected based on the time distribution characteristics of the parameters. For example, for the stimulus frequency parameter, whose target values ​​are 10Hz in 0-10 seconds and 15Hz in 10-20 seconds, a smooth transition curve from 10Hz to 15Hz can be obtained using linear fitting; a similar fitting method can be used for the stimulus intensity parameter, changing from 1.2T to 1.5T. The fitted target path can more smoothly reflect the time-varying trend of the parameters, facilitating subsequent analysis.

[0097] The probability value of the fitted target path in each time period is set as the fusion occurrence probability of the parameter-focused fusion information. The fusion occurrence probability represents the likelihood of the parameter-focused fusion information appearing within that time period, and it is obtained by combining the temporal distribution probabilities of each parameter within that time period. For example, in the 0-10 second time period, the fusion occurrence probability can be obtained by calculating the weighted average of the probabilities of the three parameters—stimulation frequency, stimulation intensity, and blood oxygen level change—with the weights determined according to the importance of each parameter in the parameter-focused fusion information; in the 10-20 second time period, the weighted average of the probabilities of each parameter is also calculated as the fusion occurrence probability.

[0098] The probability of fusion of parameter-focused fusion information is compared with the reference frequency of feature mapping relationships between image partitions and image localization stimulus regions. The reference frequency of feature mapping relationships between image partitions and image localization stimulus regions can be recorded in a table, which includes information such as mapping relationship type, mapping relationship description, historical reference frequency, and current association strength. Below is a specific example table:

[0099] Table 1. Comparison and Judgment Table of Unrelated Parameter Parsing Results and Target Parameter Standards in Feature Parsing Module

[0100]

[0101] In this example, historical reference frequency indicates how often the mapping has been used in past transcranial magnetic stimulation cases, ranging from 0 to 1; current association strength indicates the degree of association of the mapping with other mappings in the current parameter-focused fusion information, also ranging from 0 to 1.

[0102] The probability of fusion occurrence is compared with the historical reference frequency and current association strength in the table, and the difference value is marked. The difference value is used to reflect the update status of the feature mapping relationship between image partitions and image localization stimulus areas. For example, suppose that in the time period of 0-10 seconds, the probability of fusion occurrence is 0.75, while the historical reference frequency of "temporal association" is 0.8 and the current association strength is 0.78, the difference value may be small; while the historical reference frequency of "structural association" is 0.6 and the current association strength is 0.55, the difference between the probability of fusion occurrence and the historical reference frequency may be large. The calculation method of the difference value can be set according to specific circumstances, such as absolute difference, relative difference, etc.

[0103] The feature mapping relationship between image partitions and image localization stimulus areas is classified according to the observed differences, thus updating the feature mapping relationship between image partitions and image localization stimulus areas. The classification criteria can be set as follows: when the difference value is less than a certain threshold (e.g., 0.1), the mapping relationship is in a "stable" state and does not need to be updated; when the difference value is between 0.1 and 0.2, the mapping relationship is in a "needs adjustment" state, requiring fine-tuning of the association strength; when the difference value is greater than 0.2, the mapping relationship is in a "needs reconstruction" state, requiring the mapping relationship to be re-established.

[0104] For example, in the above example, the difference value of "structural association" may be greater than 0.2, and it is classified as "reconstruction required". In this case, it is necessary to re-analyze the boundary overlap mapping relationship between the image partition and the image localization stimulus area. This mapping relationship may be updated by re-segmenting the image and adjusting the spatial registration parameters. The difference value of "spatial association" may be between 0.1 and 0.2, and it is classified as "adjustment required". Its association strength needs to be fine-tuned. The difference values ​​of "temporal association" and "functional association" are small, and they are classified as "stable" and remain unchanged.

[0105] In practical applications, the mapping update module needs to work closely with the region fusion module to ensure that the acquired parameters accurately reflect the fusion information. Simultaneously, this module also needs to adjust the calculation method for the probability of fusion, the standard for calculating the difference value, and the classification threshold for the mapping relationship based on different transcranial magnetic stimulation treatment plans and clinical diagnostic needs. For example, in treating depression, more attention may be paid to "functional association" and "temporal association," with stricter thresholds set for the difference values ​​of these mapping relationships; while in treating stroke sequelae, more attention may be paid to "spatial association" and "structural association," with corresponding adjustments to the thresholds and classification standards.

[0106] Through the above series of operations, the mapping update module can verify the information of the image localization stimulation area based on the parameter focusing fusion information, identify the mapping relationships that need to be updated, and update the feature mapping relationship between the image partition and the image localization stimulation area according to the update status. These updated mapping relationships can more accurately reflect the real correlation between image data and image localization stimulation area during transcranial magnetic stimulation, providing a more reliable basis for subsequent processing execution modules, thereby improving the adaptability and accuracy of the entire medical image processing system and better assisting clinical diagnosis and treatment.

[0107] In practical implementation, analysis can be conducted by incorporating more parameters and mapping relationship types, such as the position parameters of the stimulation coil and the connection mapping relationships of nerve fiber bundles, making the mapping relationship updates more comprehensive and accurate. Simultaneously, as transcranial magnetic stimulation cases accumulate, historical citation frequency and correlation strength data will be continuously updated, enabling the mapping relationship update module to dynamically adapt to new situations and continuously optimize feature mapping relationships.

[0108] Example 5:

[0109] The specific implementation of the processing execution module is as follows: This module is used to determine the processing target and processing flow of medical image data based on the updated feature mapping relationship between image partitions and image localization stimulation areas, and to generate a processing execution library. In practical applications, the operation of the processing execution module needs to be based on the updated result of the feature mapping relationship, which includes the corresponding association between the temporal, spatial, and correlation features of the image partitions and the parameters, functions, and anatomical information of the image localization stimulation areas.

[0110] Processing objectives are extracted from medical imaging data based on the updated feature mapping. These objectives encompass multiple dimensions, such as image preprocessing objectives (e.g., noise reduction, artifact removal), feature extraction objectives (e.g., lesion identification, functional area localization), and stimulation effect evaluation objectives (e.g., neural response analysis, treatment effect prediction). Taking a specific transcranial magnetic stimulation (TMS) scenario as an example, suppose the updated feature mapping shows that the left dorsolateral prefrontal cortex region in the image partition exhibits a significant increase in blood oxygenation level under 10Hz stimulation, and this region has a strong functional connection with the default network. In this case, processing objectives might include: denoising the fMRI image of this region, extracting the blood oxygenation change characteristics of this region, and evaluating the modulatory effect of 10Hz stimulation on the default network.

[0111] After extracting the processing targets, they need to be sorted according to their probability of occurrence to determine the processing flow of medical imaging data. The probability of occurrence of a processing target is determined by a combination of factors, including the urgency of clinical diagnostic needs, the frequency of execution of the target in historical cases, and the feature matching degree of the current medical imaging data. For example, in the context of depression treatment, clinical diagnosis focuses more on the impact of stimuli on brain regions related to mood regulation. If the current imaging data shows a strong correlation between the image-localized stimulus area and the amygdala, then the target "assessing the impact of stimuli on amygdala activity" will have a higher probability of occurrence. Below is an example table of processing targets and their probabilities of occurrence:

[0112] Table 2. Calculation of Difference Values ​​and Classification Update Strategies for Feature Mapping Relationships of Image Localization Stimulus Regions

[0113]

[0114] In this example, the clinical need weight is set according to the standard treatment protocol for depression, the historical execution frequency is based on statistics of 100 similar cases in the past, and the feature matching degree is determined by the degree of correlation between the current image data and the target features. The overall occurrence probability is calculated by weighting, such as: Overall occurrence probability = Clinical need weight × 0.4 + Historical execution frequency × 0.3 + Feature matching degree × 0.3.

[0115] The processing flow is obtained by ranking the processing targets according to their overall occurrence probability. In the example above, the processing flow is as follows: head motion correction of fMRI images (probability 0.75) → extraction of blood oxygen level change features in the image-localized stimulation area (probability 0.82) → analysis of the impact of stimulation on the functional connectivity of the default mode network (probability 0.81) → visualization of the three-dimensional connectivity between the image-localized stimulation area and related brain regions (probability 0.58). It should be noted that the order of the processing flow must consider the dependency mapping relationship between the targets. For example, "feature extraction" should be performed after "image preprocessing". If a target with a high occurrence probability depends on a target with a low occurrence probability, their execution order should be adjusted.

[0116] After defining the processing goals and workflow, the two are combined in a structured format to generate a processing execution library. The structured format uses a standardized data format, such as JSON or XML, to ensure data storability and callability. Below is an example of a processing execution library in JSON format:

[0117] {"Processing Execution Library Identifier":"TMS_Processing_20250622","Processing Target List":[{"Target ID":"T001","Target Description":"Head Motion Correction for fMRI Images","Target Category":"Image Preprocessing","Execution Probability":0.75,"Dependent Target":[]},{"Target ID":"T002","Target Description":"Feature Extraction of Blood Oxygen Level Changes in Image Localization Stimulation Area","Target Category":"Feature Extraction","Execution Probability":0.82,"Dependent Target":["T001"]},{"Target ID":"T003","Target Description":"Analysis of the Impact of Stimuli on Default Functional Network Connectivity"] {"Target Category":"Effect Evaluation","Execution Probability":0.81,"Dependent Target":["T002"]},{"Target ID":"T004","Target Description":"3D connectivity visualization between image-localized stimulation area and related brain regions","Target Category":"Visualization","Execution Probability":0.58,"Dependent Target":["T002","T003"]}],"Processing Flow":["T001","T002","T003","T004"],"Association Parameters":{"Stimulation Frequency":"10Hz","Stimulation Intensity":"1.5T","Image Modality":"fMRI","Image Localization Stimulation Area":"Left dorsolateral prefrontal cortex"}}

[0118] In the JSON structure described above, each processing target includes a unique identifier, description, category, execution probability, and dependency mapping. The processing flow lists the target IDs sequentially, and the associated parameters record key information about the current transcranial magnetic stimulation. This structured format facilitates system calls to processing targets and processes, and also simplifies subsequent maintenance and updates.

[0119] In practical applications, the processing execution module needs to work closely with the mapping update module to ensure that processing targets and processes are generated based on the latest feature mapping relationships. Simultaneously, this module must dynamically adjust the extraction rules for processing targets, the methods for calculating their occurrence probabilities, and the structured parameter configurations according to different clinical application scenarios (such as treatment of depression, Parkinson's disease, and stroke) and medical imaging modalities (such as fMRI, PET, and MRI). For example, in Parkinson's disease treatment, the processing target may focus more on feature extraction of the motor cortex and assessment of the impact of stimulation on motor function; in PET image processing, the processing target may include the distribution analysis of radioactive tracers.

[0120] Furthermore, the processing execution library is not static after generation; it can be iteratively optimized based on feedback from the actual processing. For example, if the current algorithm's correction effect is found to be unsatisfactory when performing "head motion correction of fMRI images," the execution parameters for that target can be manually adjusted or the algorithm can be changed, and the processing execution library updated. This dynamic optimization mechanism can improve the adaptability and effectiveness of the processing execution library.

[0121] Through the above steps, the processing execution module transforms abstract feature mapping relationships into specific processing objectives and executable processes, generating a structured processing execution library. This library provides clear operational guidelines for transcranial magnetic stimulation (TMS) medical image processing, making the entire process operable and traceable, thereby assisting clinicians in developing more efficient diagnostic and treatment plans. In practical implementation, the processing execution library can also be integrated with the hospital's Picture Archiving and Communication System (PACS) and Electronic Medical Record (EMR) system to achieve end-to-end management of medical image data and processing results.

[0122] It should be noted that, in this document, mapping terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual mapping relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus.

[0123] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A medical image processing system for transcranial magnetic stimulation, characterized in that, include: The image acquisition module is used to collect multimodal medical image data generated during transcranial magnetic stimulation and to define the image partitions of the medical image data relative to the nerve stimulation area according to clinical diagnostic needs. The feature parsing module is used to extract stimulus-related features from medical image data and construct a feature mapping relationship between image partitions and image-localized stimulus areas. The region fusion module is used to select core regions from image partitions and image localization stimulation areas, and analyze the fusion matching degree between multiple parameters within the image localization stimulation area based on the core regions to obtain parameter focusing fusion information; The mapping relationship update module is used to perform information verification on the image localization stimulation area based on the parameter focusing fusion information, identify the update status of the feature mapping relationship between the image partition and the image localization stimulation area, and update the feature mapping relationship between the image partition and the image localization stimulation area according to the update status of the feature mapping relationship. The processing execution module is used to determine the processing objectives and processes of medical image data based on the feature mapping relationship between the updated image partitions and image localization stimulation areas, and to generate a processing execution library. The implementation methods of the regional fusion module include: The image partitions and image localization stimulation areas are clustered according to stimulation type, functional type, and function. The largest cluster center after the clustering analysis is set as the core region. Extract the core features of the core region, calculate the feature similarity between each core feature, and set a common sequence related to the feature similarity between each core feature; By utilizing the common sequences related to the feature similarity between the core features, parameters existing in the common sequences are extracted, and the longest common subsequence between the parameters is set. The length of the longest common subsequence is then set as the fusion matching degree between the parameters. The degree of fusion and matching between each parameter is determined according to the time distribution probability of each parameter, and the parameter is set to focus on fusion information.

2. The medical image processing system for transcranial magnetic stimulation according to claim 1, characterized in that, The image acquisition module can be implemented in the following ways: For medical imaging data at any moment during transcranial magnetic stimulation, obtain the modality recognition model corresponding to the medical imaging data; Use a modality recognition model to classify medical image data by modality, and obtain at least one modality category; Identify the temporal, spatial, and correlational features in medical image data under the corresponding modality category to form an image feature set. A modality of medical image data corresponds to an image feature set containing temporal, spatial, and correlational features. Each individual modality category is processed separately, and the temporal features, spatial features, and correlation features in the image feature set are analyzed separately. The temporal feature domain, spatial feature localization domain, and correlation feature association domain corresponding to the image feature set are obtained in sequence, and used as image partitions of medical image data relative to the neural stimulation region.

3. The medical image processing system for transcranial magnetic stimulation according to claim 2, characterized in that, The methods for obtaining the temporal feature domain, spatial feature localization domain, and associated feature association domain corresponding to the image feature set include: The temporal, spatial, and correlation features in the image feature set are integrated according to modal categories to obtain a multimodal integration result; Extract the temporal feature groups from the modality integration results, and compare the temporal feature groups with the feature library to obtain the temporal feature domain; The localization error rate of spatial features and the correlation tightness of associated features are extracted from the modal integration results. Based on the localization error rate of spatial features and the correlation tightness of associated features, the modal integration results are divided to obtain the spatial feature localization domain and the associated feature association domain.

4. The medical image processing system for transcranial magnetic stimulation according to claim 1, characterized in that, The image acquisition module is also used for: The imaging regions are defined, and the frequency and duration of neural stimulation images within each region are analyzed. The imaging regions are then fitted according to their frequency and duration to construct a mapping relationship between the imaging regions and actual clinical needs.

5. The medical image processing system for transcranial magnetic stimulation according to claim 1, characterized in that, The feature parsing module can be implemented in the following ways: The parameters and states corresponding to the medical image data are called to generate multiple unrelated parameter parsing results. The unrelated parameter parsing results represent parameters and states that are not associated with features in the image partition. Determine whether the parsing results of multiple unrelated parameters match the parsing results of the target parameter. If they match, then the parsing results of the target parameter are considered as the image localization stimulation area of ​​the medical image data.

6. The medical image processing system for transcranial magnetic stimulation according to claim 3, characterized in that, The implementation of the feature mapping relationship between image partitions and image localization stimulus areas includes: using the information of temporal feature domain, spatial feature localization domain and related feature association domain existing in the image partition as representation, and combining the descriptive information of the image localization stimulus area and the category of the image localization stimulus area, to construct the feature mapping relationship between the image partition and the image localization stimulus area.

7. The medical image processing system for transcranial magnetic stimulation according to claim 1, characterized in that, The implementation methods of the mapping relationship update module include: Extract the temporal distribution probability of each parameter from the parameter-focused fusion information; set the target path of the parameter-focused fusion information according to the time period corresponding to the temporal distribution probability of each parameter. The target paths of each parameter in the parameter-focused fusion information are fitted to obtain the fitted target paths. The probability values ​​of the fitted target paths in each time period are set as the fusion occurrence probability of the parameter-focused fusion information. The probability of fusion of parameter-focused fusion information is compared with the reference of the feature mapping relationship between image partitions and image localization stimulation areas. The difference values ​​are marked. The difference values ​​are used to reflect the update status of the feature mapping relationship between image partitions and image localization stimulation areas. The feature mapping relationship between image partitions and image localization stimulation areas is classified according to the difference values, and the update of the feature mapping relationship between image partitions and image localization stimulation areas is completed.

8. The medical image processing system for transcranial magnetic stimulation according to claim 1, characterized in that, The implementation method of the execution module is as follows: Based on the feature mapping relationship between the updated image partitions and image localization stimulation areas, the processing targets in the medical image data are extracted, and the processing targets are sorted according to their occurrence probability to obtain the processing flow in the medical image data. The processing targets and processes are combined in a structured manner to obtain the processing execution library.

9. A method for processing medical images using transcranial magnetic stimulation (TMS), applied to the TMS medical image processing system according to any one of claims 1 to 8, characterized in that, Includes the following steps: Multimodal medical imaging data generated during transcranial magnetic stimulation are collected, and image partitions of the medical imaging data relative to the nerve stimulation area are defined according to clinical diagnostic needs. Extract stimulus-related features from medical imaging data and construct a feature mapping relationship between image partitions and image-localized stimulus areas; Core regions are selected from image partitions and image localization stimulation areas, and the fusion matching degree between multiple parameters within the image localization stimulation area is analyzed based on the core regions to obtain parameter focusing fusion information; Based on the parametric focusing fusion information, information verification is performed on the image localization stimulation area, the update status of the feature mapping relationship between the image partition and the image localization stimulation area is identified, and the feature mapping relationship between the image partition and the image localization stimulation area is updated according to the update status of the feature mapping relationship. Based on the updated feature mapping relationship between image partitions and image localization stimulation areas, the processing objectives and processes of medical image data are determined, and a processing execution library is generated.

Images