Repetitive Transcranial Magnetic Stimulation Intervention System

Abstract

This application relates to the technical field of repetitive transcranial magnetic stimulation. Specifically, it discloses a repetitive transcranial magnetic stimulation intervention system, which includes: a brain region initial positioning data acquisition module, a basic near-infrared spectroscopy data acquisition module, an updated infrared spectroscopy data acquisition module, and a target brain region positioning detection module, which is used to determine whether the positioning accuracy of the target brain region meets the requirements based on the comparison between the updated infrared spectroscopy data and the basic near-infrared spectroscopy data; a positioning adjustment module. In this way, it is possible to automatically judge the positioning accuracy of the target brain region and detect whether there is a deviation based on the infrared spectroscopy semantics and cerebral blood hemodynamics changes before and after rTMS stimulation, and send a positioning adjustment prompt signal when there is a deviation. In this way, it is possible to ensure the positioning accuracy of the target brain region for each stimulation in an automated manner during the repetitive transcranial magnetic stimulation process, thereby improving the intelligent level of the repetitive transcranial magnetic stimulation intervention system.

Description

Technical Field

[0001] This application relates to the technical field of repetitive transcranial magnetic stimulation, and more specifically, to a repetitive transcranial magnetic stimulation intervention system. Background Art

[0002] Repetitive Transcranial Magnetic Stimulation (rTMS) is a non-invasive brain stimulation method that affects the activities of cortical neurons by changing the magnetic field above the scalp. rTMS has shown potential in treating neurological diseases such as depression and Parkinson's disease and has been used to study the relationship between brain function and structure. However, the effect of rTMS highly depends on the accurate localization of the target brain region. If the stimulation position is inaccurate, the expected treatment purpose may not be achieved.

[0003] Traditional methods for localizing the target brain region usually set the position of the coil based on pre-acquired imaging data (such as MRI) before treatment. However, during the actual treatment process, the head posture of the patient may change, which may cause the predetermined stimulation position to deviate from the target brain region. In addition, the brain is a dynamic organ, and its internal structure and activity pattern also change over time, while static localization cannot respond to this. That is to say, most rTMS localization techniques lack a real-time monitoring mechanism, which means that even if the initial localization is correct, any changes during the treatment process (such as patient movement or brain state change) may affect the final stimulation effect. Summary of the Invention

[0004] This application provides a repetitive transcranial magnetic stimulation intervention system, which can automatically judge the localization accuracy of the target brain region and detect whether there is a deviation based on the infrared spectral semantics and cerebral hemodynamic changes before and after rTMS stimulation, and send a positioning adjustment prompt signal when there is a deviation, so as to ensure the localization accuracy of the target brain region for each stimulation in an automated manner during the repetitive transcranial magnetic stimulation process, thereby improving the intelligent level of the repetitive transcranial magnetic stimulation intervention system.

[0005] In a first aspect, there is provided a repetitive transcranial magnetic stimulation intervention system, comprising:

[0006] A brain region initial positioning data acquisition module for recording the initial positioning data of the target brain region;

[0007] A basic near-infrared spectroscopy data acquisition module for using a NIRS monitoring device to acquire the basic near-infrared spectroscopy data of the target brain region without any stimulation;

[0008] An updated infrared spectrum data acquisition module is used to control the rTMS device to stimulate the target brain region, and at the same time, the NIRS monitoring device is used to continuously monitor the near-infrared spectrum changes in the target brain region to obtain updated infrared spectrum data;

[0009] A target brain region positioning detection module is used to determine whether the positioning accuracy of the target brain region meets the requirements based on the comparison between the updated infrared spectrum data and the basic near-infrared spectrum data;

[0010] A positioning adjustment module is used to generate a target brain region positioning adjustment prompt in response to the positioning accuracy of the target brain region not meeting the requirements.

[0011] A repetitive transcranial magnetic stimulation intervention system provided by the present application processes the basic near-infrared spectrum data and the updated infrared spectrum data together through a data processing and analysis algorithm based on artificial intelligence and signal processing, so as to capture the semantics of the basic near-infrared spectrum image and the updated near-infrared spectrum image, as well as the fine-grained contrast feature representation information between the two, thereby judging the positioning accuracy of the target brain region to determine whether the positioning accuracy of the target brain region meets the requirements. In this way, it is possible to automatically judge the positioning accuracy of the target brain region and detect whether there is a deviation based on the infrared spectrum semantics and cerebral blood hemodynamics changes before and after rTMS stimulation, and send a positioning adjustment prompt signal when there is a deviation, so as to ensure the positioning accuracy of the target brain region for each stimulation in an automated manner during the repetitive transcranial magnetic stimulation process, thereby improving the intelligent level of the repetitive transcranial magnetic stimulation intervention system. Description of the Drawings

[0012] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings of the embodiments of the present application will be briefly introduced below. Obviously, the drawings described below only relate to some embodiments of the present application and do not limit the present application.

[0013] Figure 1 It is a schematic block diagram of the repetitive transcranial magnetic stimulation intervention system according to the embodiment of the present application.

[0014] Figure 2 It is a schematic block diagram of the repetitive transcranial magnetic stimulation intervention system according to the embodiment of the present application.

[0015] Figure 3 It is a schematic diagram of the data flow of the target brain region positioning detection module in the repetitive transcranial magnetic stimulation intervention system according to the embodiment of the present application.

[0016] Figure 4 It is a schematic block diagram of the near-infrared spectrum dominant feature enhancement processing unit in the repetitive transcranial magnetic stimulation intervention system according to the embodiment of the present application.

[0017] Figure 5 It is a schematic block diagram of the near-infrared spectrum interaction coding unit in the repetitive transcranial magnetic stimulation intervention system according to an embodiment of the present application.

[0018] Figure 6 It is a schematic block diagram of the near-infrared spectrum fine-grained semantic sharing compensation subunit in the repetitive transcranial magnetic stimulation intervention system according to an embodiment of the present application. Detailed implementation manners

[0019] Next, the technical solutions in the embodiments of the present application will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all embodiments. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall also fall within the protection scope of the present application.

[0020] Near-infrared spectroscopy (NIRS) is a technology that can measure the change in hemoglobin concentration in the brain tissue, thereby reflecting the metabolic activity of brain regions. NIRS has high temporal resolution and good spatial resolution, and can monitor the hemodynamic changes during brain activities in real time. This immediacy is crucial for ensuring the accuracy of the stimulation location. Therefore, it has unique advantages in evaluating the impact of rTMS on specific brain regions and the localization detection of target brain regions.

[0021] Based on this, in the technical solution of the present application, a repetitive transcranial magnetic stimulation intervention system is proposed. The system aims to improve the localization accuracy of the target brain region during the rTMS intervention process. By combining NIRS monitoring and rTMS stimulation, the system can compare and analyze the physiological responses of the target brain region before and after the stimulation to ensure the correctness of the stimulation location. Specifically, the system will record the initial localization data of the target brain region and use the NIRS monitoring device to obtain the baseline near-infrared spectrum data in the non-stimulated state as a reference. Then, during the rTMS stimulation process, it continuously monitors and collects the updated NIRS data. By comparing these two sets of data, it can be determined whether the actually stimulated brain region is consistent with the predetermined target region and whether the location of the stimulated target brain region deviates.

[0022] Specifically, as Figure 1As shown, the repetitive transcranial magnetic stimulation intervention system 1 includes: a brain region initial positioning data acquisition module 10 for recording the initial positioning data of the target brain region; a basic near-infrared spectroscopy data acquisition module 20 for using an NIRS monitoring device to acquire the basic near-infrared spectroscopy data of the target brain region without any stimulation; an updated infrared spectroscopy data acquisition module 30 for controlling the rTMS device to stimulate the target brain region, and simultaneously using the NIRS monitoring device to continuously monitor the near-infrared spectroscopy changes of the target brain region to obtain updated infrared spectroscopy data; a target brain region positioning detection module 40 for determining whether the positioning accuracy of the target brain region meets the requirements based on the comparison between the updated infrared spectroscopy data and the basic near-infrared spectroscopy data; and a positioning adjustment module 50 for generating a target brain region positioning adjustment prompt in response to the positioning accuracy of the target brain region not meeting the requirements.

[0023] Exemplarily, in the brain region initial positioning data acquisition module 10, the initial positioning data of the target brain region is recorded. It should be understood that recording the initial positioning data of the target brain region is crucial for repetitive transcranial magnetic stimulation (rTMS) treatment because it ensures that the stimulation can accurately act on the predetermined target brain region, which is essential for achieving effective treatment results and reducing potential side effects. Since the brain is a dynamic organ and its internal structure and activity patterns change over time, static positioning methods may not be able to respond to these changes. Therefore, recording the initial positioning data not only provides a precise starting point for treatment but also allows real-time monitoring and adjustment of any position deviation during the treatment process, thus maintaining the consistency and effectiveness of each stimulation.

[0024] In one embodiment, recording the initial positioning data of the target brain region includes: First, by using structural imaging techniques such as MRI (magnetic resonance imaging), high-resolution images of the patient's brain can be obtained, providing detailed anatomical structure information to help accurately locate the brain region to be stimulated. To ensure the consistency of positioning among different patients, a standard brain coordinate system (such as MNI space or Talairach space) is often used to map the individualized MRI scan results into a common spatial framework for cross-individual comparison and positioning. In addition, technicians will set several reference points on the patient's scalp and associate these reference points with the corresponding positions on the MRI image through special devices (such as laser pointers, robotic arms, etc.). This step ensures that even if the patient's head moves slightly, the position of the rTMS coil can be quickly recalibrated. Sometimes, electrophysiological signals such as EEG (electroencephalogram) are also combined as an auxiliary means. Especially when looking for the cortical positions related to specific functions, the EEG activity patterns during certain cognitive tasks can help identify the brain regions responsible for performing that task.

[0025] Exemplarily, in the basic near-infrared spectroscopy data acquisition module 20, a NIRS monitoring device is used to acquire the basic near-infrared spectroscopy data of the target brain region without any stimulation. It should be understood that using a NIRS (near-infrared spectroscopy imaging) monitoring device to acquire the basic near-infrared spectroscopy data of the target brain region without any stimulation is crucial for ensuring the accuracy and effectiveness of repetitive transcranial magnetic stimulation (rTMS) treatment. This process not only provides a reliable benchmark for the subsequent rTMS stimulation effect, but also helps to improve the positioning accuracy, monitor brain region activities, and reduce the influence of interference factors. Specifically, the basic NIRS data collected without any external stimulation is used as a benchmark to compare the changes after subsequent rTMS stimulation. This comparison helps to evaluate the impact of rTMS on specific brain regions and confirm whether the expected effect has been achieved. At the same time, by analyzing physiological parameters such as hemoglobin concentration and oxygenation level in the resting state, the location of the target brain region can be determined more precisely, providing an important reference for the placement of the rTMS coil, thereby improving the accuracy of the stimulation location.

[0026] In addition, NIRS can monitor the hemodynamic changes during brain activities in real time, reflecting the metabolic activity of brain regions. This information can help identify which regions are more active in the natural state and which may need intervention, providing support for personalized treatment plans. Since the brain is a dynamic organ, its internal structure and activity patterns change over time, and the basic NIRS data provides a stable reference point for continuously tracking the changes in brain regions throughout the treatment course, helping the system to flexibly respond to these changes and ensuring that each stimulation acts on the correct brain region. The data obtained under non-stimulated conditions excludes the influence of external intervention, making the subsequent comparison more reliable and enabling a clearer distinction of the specific changes caused by rTMS rather than the results caused by other variables.

[0027] The process of acquiring the basic near-infrared spectroscopy data of the target brain region without any stimulation first requires ensuring that the patient is in a relaxed state and avoiding any factors that may cause abnormal brain activities, such as tension or thinking about complex problems. Usually, the patient is required to keep quiet and close their eyes and rest for a few minutes to obtain data closest to the natural state. Then, the NIRS monitoring device is correctly worn on the patient's head, ensuring that the sensors (light sources and detectors) are located above the target brain region and are properly calibrated to ensure the accuracy of the measurement results. The position of the NIRS probe is corresponded to the position of the target brain region previously determined by MRI or other imaging techniques using a standardized coordinate system (such as MNI or Talairach space) to achieve precise spatial positioning.

[0028] Before starting the recording, allow the device to have a warm-up period to stabilize the signals, and then start the data acquisition program. Continuously monitor the changes in hemoglobin concentration and other relevant parameters in the target brain region within a certain period of time. During the acquisition process, ensure that the environmental conditions are constant, such as light intensity, temperature, etc., to reduce the interference of external factors on the measurement results. After the data acquisition is completed, use a specially designed software platform to process the raw data, remove noise and extract useful information, such as calculating the changes in light absorption at different wavelengths, and then inferring the changes in hemoglobin concentration. The obtained basic NIRS data can be further used for feature extraction to generate an image semantic coding feature map for subsequent comparison with the data collected after stimulation. All the collected basic NIRS data should be properly stored and marked with relevant information, such as the acquisition date, time, patient ID, etc., for future reference and analysis. These data will serve as an important reference during the rTMS treatment process to help determine whether the actually stimulated brain region is consistent with the predetermined target area and whether the position of the stimulated target brain region deviates.

[0029] Exemplarily, in the updated infrared spectrum data acquisition module 30, control the rTMS device to stimulate the target brain region, and at the same time use the NIRS monitoring device to continuously monitor the near-infrared spectrum changes in the target brain region to obtain updated infrared spectrum data. It should be understood that controlling the rTMS (repetitive transcranial magnetic stimulation) device to stimulate the target brain region and simultaneously using the NIRS (near-infrared spectroscopy imaging) device to continuously monitor the near-infrared spectrum changes in the target brain region to obtain updated data is an important means to ensure the treatment effect and safety of rTMS. This process allows real-time assessment of the impact of rTMS on the brain, ensures the accuracy of the stimulation location, and monitors potential physiological responses. Specifically: The basic NIRS data collected without any external stimulation provides a reference point in the resting state. When rTMS starts to act, changes in cerebral hemodynamics and metabolic activity will occur, and these changes can be captured in real time by NIRS. By comparing the situations before and after stimulation, it can be accurately determined whether rTMS has reached the expected target brain region, whether there is any deviation, and specific physiological responses can be observed. This real-time monitoring is crucial for dynamically adjusting rTMS parameters and optimizing the treatment plan. In addition, it can also help identify possible side effects or adverse reactions caused by improper stimulation, so as to take timely measures to correct them.

[0030] In one embodiment, the control rTMS device stimulates the target brain region, while the NIRS monitoring device continuously monitors the near-infrared spectral changes in the target brain region to obtain updated infrared spectral data, including: First, ensure that the patient is in a relaxed state and avoid external factors interference, such as nervousness or complex thinking activities. Correctly wear the NIRS monitoring device on the patient's head so that the sensors accurately cover the target brain region, and complete the necessary calibration steps, including using a standardized coordinate system (such as MNI or Talairach space) to correspond the position of the NIRS probe with the previously determined target brain region position to ensure accurate spatial positioning. Next, start the rTMS device and apply stimulation to the target brain region according to the predetermined parameter settings. At the same time, the NIRS monitoring device begins to continuously record the changes in hemoglobin concentration and other relevant physiological parameters in the target brain region. During this process, keep the environmental conditions stable, such as light intensity and temperature, to reduce the influence of external factors on the measurement results. The NIRS monitoring device should have a sufficient sampling frequency to capture the rapidly occurring hemodynamic changes and ensure the integrity and accuracy of the data. As the rTMS stimulation progresses, the NIRS monitoring device will continuously generate new data, which reflect the immediate response of the target brain region after being stimulated. By analyzing these updated near-infrared spectral data, information about blood flow, oxygenation level, etc. can be extracted, and then the effect of rTMS can be evaluated. All the collected updated infrared spectral data should be properly stored and marked with relevant information, such as acquisition timestamp, rTMS stimulation parameters, etc., for subsequent analysis. A specially designed software platform is used to process these data, remove noise and extract useful information, such as calculating the changes in light absorption at different wavelengths and inferring the changes in hemoglobin concentration. Further, these data can be compared with the previously established baseline NIRS data to generate an image semantic coding feature map to visually display the differences before and after stimulation.

[0031] Exemplarily, in the target brain region localization detection module 40, based on the comparison between the updated infrared spectral data and the basic near-infrared spectral data, it is determined whether the localization accuracy of the target brain region meets the requirements. It should be understood that in the above repetitive transcranial magnetic stimulation intervention system, the step of determining whether the localization accuracy of the target brain region meets the requirements is crucial. It is the key to ensuring that rTMS stimulation acts on the expected target brain region, avoiding mis-stimulation of other functional regions in the brain, and reducing potential risks and side effects, which is crucial for protecting the safety of patients. Specifically, in the process of determining whether the localization accuracy of the target brain region meets the requirements, the technical concept of the present application is to process the basic near-infrared spectral data and the updated infrared spectral data together through a data processing and analysis algorithm based on artificial intelligence and signal processing, so as to capture the semantics of the basic near-infrared spectral image and the updated near-infrared spectral image, as well as the fine-grained contrast feature representation information between the two, thereby judging the localization accuracy of the target brain region to determine whether the localization accuracy of the target brain region meets the requirements. In this way, it is possible to automatically judge the localization accuracy of the target brain region and detect whether there is a deviation based on the infrared spectral semantics and cerebral hemodynamics changes before and after rTMS stimulation, and send a localization adjustment prompt signal when there is a deviation, so that it is possible to ensure the localization accuracy of the target brain region in an automated manner during repetitive transcranial magnetic stimulation, thereby improving the intelligent level of the repetitive transcranial magnetic stimulation intervention system.

[0032] In one embodiment, as Figure 2 and Figure 3 shown, the target brain region localization detection module 40 includes: a near-infrared spectral feature extraction unit 41, configured to perform near-infrared spectral-based feature extraction on the updated infrared spectral data and the basic near-infrared spectral data respectively to obtain an updated near-infrared spectral image semantic encoding feature map and a basic near-infrared spectral image semantic encoding feature map; a near-infrared spectral dominant feature enhancement processing unit 42, configured to perform dominant feature enhancement processing based on spatial domain transformation on the updated near-infrared spectral image semantic encoding feature map and the basic near-infrared spectral image semantic encoding feature map respectively to obtain an updated near-infrared spectral image semantic enhanced encoding feature map and a basic near-infrared spectral image enhanced semantic encoding feature map; a near-infrared spectral interactive encoding unit 43, configured to perform interactive encoding based on fine-grained semantic sharing compensation on the updated near-infrared spectral image semantic enhanced encoding feature map and the basic near-infrared spectral image enhanced semantic encoding feature map to obtain an updated near-infrared spectral - basic near-infrared spectral semantic fine-grained contrast feature map; a localization accuracy determination unit 44, configured to determine whether the localization accuracy of the target brain region meets the requirements based on the updated near-infrared spectral - basic near-infrared spectral semantic fine-grained contrast feature map.

[0033] Exemplarily, in the near-infrared spectroscopy feature extraction unit 41, feature extraction based on near-infrared spectroscopy is respectively performed on the updated infrared spectroscopy data and the basic near-infrared spectroscopy data to obtain an updated near-infrared spectroscopy image semantic encoding feature map and a basic near-infrared spectroscopy image semantic encoding feature map. It should be understood that through feature extraction, complex NIRS data can be converted into a feature representation form with clear biological significance, and these features can better reflect the activity patterns and physiological changes of the brain in different states. Specifically, the basic near-infrared spectroscopy data represents the hemodynamic characteristics of the target brain region in the resting state, while the updated infrared spectroscopy data reflects the immediate response after rTMS stimulation. Comparing these two sets of features can help accurately evaluate whether rTMS has reached the expected target brain region and whether there are any deviations or abnormalities.

[0034] In one embodiment, the near-infrared spectroscopy feature extraction unit is configured to: input the updated infrared spectroscopy data and the basic near-infrared spectroscopy data into a near-infrared spectroscopy feature extractor based on the Mobile-Former model to obtain the updated near-infrared spectroscopy image semantic encoding feature map and the basic near-infrared spectroscopy image semantic encoding feature map. It should be understood that Mobile-Former is a lightweight but efficient neural network architecture that combines the advantages of convolutional neural networks (CNNs) and Transformers and can capture complex spatial and temporal features, which is crucial for analyzing subtle changes in NIRS data. Therefore, through encoding by the near-infrared spectroscopy feature extractor based on the Mobile-Former model, the infrared spectroscopy image semantic feature information in the updated infrared spectroscopy data and the basic near-infrared spectroscopy data can be respectively extracted.

[0035] Exemplarily, in the near-infrared spectrum dominant feature enhancement processing unit 42, the updated near-infrared spectrum image semantic encoding feature map and the basic near-infrared spectrum image semantic encoding feature map are respectively subjected to dominant feature enhancement processing based on spatial domain transformation to obtain an updated near-infrared spectrum image semantic enhanced encoding feature map and a basic near-infrared spectrum image enhanced semantic encoding feature map. Considering that the updated near-infrared spectrum image semantic encoding feature map and the basic near-infrared spectrum image semantic encoding feature map respectively contain the semantic features of the infrared spectrum images before and after rTMS stimulation, the features of these two respectively reflect the specific positions and patterns of the changes in regional cerebral blood dynamics and metabolic activity and other target brain region activities. However, the original updated near-infrared spectrum image semantic encoding feature map and the basic near-infrared spectrum image semantic encoding feature map will contain some pixel coupling relationships and feature information with low contribution to the semantic comparison of the two infrared spectrum images and the subsequent target brain region localization detection task. The influence of this information will result in the lack of pertinence and distinguishability in the semantic comparison of the infrared spectrum images before and after stimulation. Therefore, in order to enhance the fine-grained feature comparison accuracy and robustness between the semantics of these two infrared spectrum images, in the technical solution of this application, the updated near-infrared spectrum image semantic encoding feature map and the basic near-infrared spectrum image semantic encoding feature map are further respectively subjected to dominant feature enhancement processing based on spatial domain transformation to obtain an updated near-infrared spectrum image semantic enhanced encoding feature map and a basic near-infrared spectrum image enhanced semantic encoding feature map.

[0036] In one embodiment, as Figure 4 shown, the near-infrared spectrum dominant feature enhancement processing unit 42 includes: a near-infrared spectrum pixel granularity feature decoupling sub-unit 421, configured to perform pixel granularity feature decoupling on the updated near-infrared spectrum image semantic encoding feature map to obtain a set of updated near-infrared spectrum pixel granularity decoupled feature vectors; a near-infrared spectrum structural semantic saliency quantum unit 422, configured to calculate the fine-grained structural semantic saliency measurement coefficients of each updated near-infrared spectrum pixel granularity decoupled feature vector in the set of updated near-infrared spectrum pixel granularity decoupled feature vectors to obtain an updated near-infrared spectrum structural semantic saliency distribution matrix; a near-infrared spectrum feature dominant hierarchical precipitation processing sub-unit 423, configured to perform feature dominant hierarchical precipitation processing on the updated near-infrared spectrum structural semantic saliency distribution matrix based on a multi-layer mask function to obtain an updated near-infrared spectrum structural semantic saliency distribution weight matrix; a point-by-position multiplication processing sub-unit 424, configured to calculate the point-by-position multiplication between the updated near-infrared spectrum structural semantic saliency distribution weight matrix and each feature matrix along the channel dimension of the updated near-infrared spectrum image semantic encoding feature map to obtain the updated near-infrared spectrum image semantic enhanced encoding feature map.

[0037] Exemplarily, in the near-infrared spectral pixel granularity feature decoupling subunit 421, the updated near-infrared spectral image semantic encoding feature map is subjected to pixel granularity feature decoupling to obtain a set of updated near-infrared spectral pixel granularity decoupled feature vectors. Specifically, this process can be expressed by the formula:

[0038] decouple(F) = {v1, v2,..., v i ,..., v n}

[0039] where F is the updated near-infrared spectral image semantic encoding feature map, decouple(F) is the pixel granularity feature decoupling of F, and v1, v2, v i and v n are the 1st, 2nd, i-th, and n-th updated near-infrared spectral pixel granularity decoupled feature vectors in the set of updated near-infrared spectral pixel granularity decoupled feature vectors, respectively.

[0040] That is, by finely analyzing the feature information at each pixel level, the redundancy and unnecessary coupling relationships in the original feature map are eliminated, so that the features of each pixel can be processed independently. This not only reduces the redundancy of the feature space but also enhances the model's ability to understand local features. Specifically, during the rTMS treatment process, changes in cerebral hemodynamics and metabolic activity will be reflected in the near-infrared spectral data. When these changes are converted into image form, each pixel actually represents the physiological response at a specific location. However, the original updated near-infrared spectral image semantic encoding feature map may contain some pixel coupling relationships and feature information with low contribution to the target brain region localization detection task. These information may interfere with subsequent analysis, resulting in insufficient pertinence and distinguishability in the semantic comparison of infrared spectral images before and after stimulation. Through pixel granularity feature decoupling, the system can effectively remove these interfering factors, making the features of each pixel point more independent and representative. This means that for rTMS treatment, it becomes possible to more accurately identify the specific location and pattern of changes in target brain region activity. For example, this decoupling operation can help the system focus on those local patterns and features that are crucial for the accuracy of target brain region localization, ensuring that the stimulation acts on the correct target area. In addition, the set of decoupled feature vectors provides a cleaner data basis for subsequent steps, such as explicit feature enhancement processing, interactive coding, etc. These steps further improve the quality of feature representation, ensuring that the output feature map can highlight the most representative and distinguishable feature information, providing a solid basis for verifying the accuracy of target brain region localization.

[0041] In one embodiment, the near-infrared spectral structure semantic saliency quantum unit 422 is configured to: input each updated near-infrared spectral pixel granularity decoupled feature vector in the set of updated near-infrared spectral pixel granularity decoupled feature vectors into a hyperbolic space mapper to obtain a set of updated near-infrared spectral hyperbolic space modulated pixel granularity decoupled feature vectors; calculate the fine-grained structure semantic saliency measurement coefficients of each updated near-infrared spectral hyperbolic space modulated pixel granularity decoupled feature vector in the set of updated near-infrared spectral hyperbolic space modulated pixel granularity decoupled feature vectors to obtain the updated near-infrared spectral structure semantic saliency distribution matrix arranged by a plurality of updated near-infrared spectral fine-grained structure semantic saliency measurement coefficients. Specifically, this process can be expressed by the formula as follows:

[0042] h i = W1v i W2

[0043]

[0044] where W1 and W2 are the first weight matrix and the second weight matrix respectively, h i is the updated near-infrared spectral hyperbolic space modulated pixel granularity decoupled feature vector corresponding to v i , ||·|| 2 is the square of the one-norm of the vector, log2 represents the logarithmic function value with base 2, and s i is the updated near-infrared spectral fine-grained structure semantic saliency measurement coefficient corresponding to h i in the updated near-infrared spectral structure semantic saliency distribution matrix.

[0045] That is, first, by inputting the decoupled feature vectors into the hyperbolic space mapper, the spatial transformation of feature representation can be achieved. The hyperbolic space is a non-Euclidean geometric space, which is particularly suitable for representing hierarchical or tree-structured data and is used here to better capture the complex relationships between different regions of the brain. This mapping helps to reveal subtle differences that are not easily detectable in the original feature space, enabling the system to focus on truly important local features. The set of feature vectors after hyperbolic space mapping, that is, the set of updated near-infrared spectroscopy hyperbolic space modulated pixel granularity decoupled feature vectors, provides a more refined and discriminative data basis for subsequent analysis. Next, calculate the fine-grained structural semantic significance measurement coefficients of these feature vectors, aiming to quantify the importance of each feature and construct an updated near-infrared spectroscopy structural semantic significance distribution matrix arranged by multiple updated near-infrared spectroscopy fine-grained structural semantic significance measurement coefficients. This matrix reflects the response intensity and relative importance of each position point in the target brain region after being stimulated, helping the system identify local patterns and features that are crucial for the positioning accuracy of the target brain region. For example, certain specific regions may show higher changes in hemoglobin concentration or other physiological indicators, indicating that these regions have been more strongly activated or inhibited. In this way, the system can more accurately determine whether the actually stimulated brain region is consistent with the predetermined target region and promptly detect any deviations.

[0046] Exemplarily, in the near-infrared spectroscopy feature dominance hierarchical precipitation processing subunit 423, perform feature dominance hierarchical precipitation processing on the updated near-infrared spectroscopy structural semantic significance distribution matrix based on a multi-layer mask function to obtain an updated near-infrared spectroscopy structural semantic significance distribution weight matrix. Specifically, this process can be expressed by the formula:

[0047]

[0048] where θ is a preset threshold, mask is a masking operation, and s i ’ is the updated near-infrared spectroscopy structural semantic significance distribution weight value corresponding to h i in the updated near-infrared spectroscopy structural semantic significance distribution weight matrix.

[0049] That is, during rTMS treatment, changes in cerebral hemodynamics and metabolic activity are reflected in near-infrared spectroscopy data. Through the previous steps, an updated near-infrared spectroscopy structural semantic significance distribution matrix has been obtained, which details the response intensity and relative importance of each position point after being stimulated. However, some features with relatively low relevance or redundancy to the target brain region localization detection task may still be included in this information. To further enhance the model's ability to understand key features and reduce the influence of irrelevant information, feature explicit hierarchical precipitation processing based on a multi-layer masking function is introduced. This processing process applies a series of masking functions to the significance distribution matrix, layer by layer screening and emphasizing those features that best represent the activity change pattern of the target brain region. Each layer of the masking function selectively retains certain specific features according to preset thresholds or rules, while suppressing other unimportant parts. As the number of layers increases, this selectivity and emphasis gradually accumulate, finally forming an updated near-infrared spectroscopy structural semantic significance distribution weight matrix. This weight matrix not only highlights the most representative and discriminative feature information but also provides a cleaner data basis for subsequent analysis. For example, in actual operation, certain position points may show higher changes in hemoglobin concentration or other physiological indicators than other regions, indicating that they have received stronger activation or inhibition. Through the processing of the multi-layer masking function, the features of these key position points will be assigned higher weights in the final weight matrix, ensuring that they are fully emphasized in subsequent localization detection and effect evaluation. On the contrary, those position points that have less impact on the localization accuracy of the target brain region will be appropriately weakened to avoid interfering with the analysis results.

[0050] Exemplarily, in the position-wise multiplication processing subunit 424, calculate the position-wise multiplication between each feature matrix along the channel dimension of the updated near-infrared spectroscopy structural semantic significance distribution weight matrix and the updated near-infrared spectroscopy image semantic encoding feature map to obtain the updated near-infrared spectroscopy image semantic enhanced encoding feature map. Specifically, this process can be represented by the formula:

[0051]

[0052] where S is the updated near-infrared spectroscopy structural semantic significance distribution weight matrix, is matrix multiplication, and F f is the updated near-infrared spectroscopy image semantic enhanced encoding feature map.

[0053] That is, during the rTMS treatment, an updated near-infrared spectroscopy structural semantic saliency distribution weight matrix and an updated near-infrared spectroscopy image semantic encoding feature map have been obtained through a series of processing steps. The former reflects the response intensity and its relative importance of each position point after being stimulated, while the latter contains the initially analyzed spatial and temporal information. To further improve the quality of these feature representations, it is necessary to effectively combine the two. By calculating the element-wise multiplication operation between the two matrices, that is, multiplying each element, it is possible to weight and adjust each feature in the original encoded feature map according to the weight values in the saliency distribution weight matrix while maintaining the spatial position correspondence. This weighted fusion makes the features of the key position points emphasized by the saliency distribution weight matrix more prominent in the finally generated updated near-infrared spectroscopy image semantic enhanced encoding feature map, while the relatively unimportant features are appropriately weakened. As a result, the generated feature map not only retains the rich details in the original data but also enhances the ability to understand the activity change patterns of the target brain regions.

[0054] In summary, the explicit feature enhancement processing based on spatial domain transformation performs feature enhancement by using the explicit modeling and hierarchical mask modulation of the fine-grained features of each infrared spectral image pixel in the feature map, effectively improving the quality of the semantic encoded feature map of the updated near-infrared spectral image and the semantic encoded feature map of the basic near-infrared spectral image. Specifically, for the semantic encoded feature map of the updated near-infrared spectral image, the explicit feature enhancement processing based on spatial domain transformation can eliminate the unnecessary coupling relationships in the original semantic encoded feature map of the updated near-infrared spectral image through pixel-level feature decoupling operation, enabling the features of each pixel to be processed individually. This not only reduces the redundancy of the feature space but also enhances the model's ability to understand local features. For rTMS treatment, this means more accurately identifying the specific locations and patterns of changes in target brain region activities. Further, the hyperbolic space mapping and calculation of the fine-grained structure semantic significance measurement coefficient can quantify the importance of each fine-grained feature in the feature map, providing a scientific basis for subsequent feature selection. In the context of rTMS, this step helps the system focus on identifying the local patterns and features crucial for the positioning accuracy of the target brain region, ensuring that the stimulation acts on the correct target area. Then, selective amplification of features at different levels is achieved through a multi-layer mask function. This process is similar to the attention mechanism, allowing the system to dynamically adjust the degree of attention to the semantic features of different pixel levels in the updated near-infrared spectral image. For rTMS treatment, this helps to highlight the key features that have a significant impact on the treatment effect while suppressing noise or irrelevant features, improving the quality of the final feature representation. Finally, element-wise multiplication is used to generate the final semantic enhanced encoded feature map of the updated near-infrared spectral image, ensuring that the output feature map can highlight the most representative and discriminative feature information, providing a solid foundation for subsequent verification of the accuracy of target brain region positioning.

[0055] Exemplarily, in the near-infrared spectrum interactive encoding unit 43, an interactive encoding based on fine-grained semantic sharing compensation is performed on the updated near-infrared spectrum image semantic enhanced encoding feature map and the basic near-infrared spectrum image enhanced semantic encoding feature map to obtain an updated near-infrared spectrum - basic near-infrared spectrum semantic fine-grained contrast feature map. It should be understood that since the updated near-infrared spectrum image semantic enhanced encoding feature map and the basic near-infrared spectrum image enhanced semantic encoding feature map contain the semantic feature information of the infrared spectrum images before and after stimulation after explicit feature enhancement, some of the feature semantics between the two are common, while some are different feature semantics. These semantics are crucial for the semantic comparison and localization detection tasks of the external spectrum images before and after rTMS stimulation. However, traditional feature comparison and interaction methods often focus on the combination and comparison of overall features, ignoring the subtle differences in local regions of the feature map. This may lead to the loss of key details between the semantic of the external spectrum images before and after rTMS stimulation, affecting the accuracy of the final localization detection result. In addition, traditional methods mainly rely on the information provided by the input features themselves and are powerless for potential semantic information that is not explicitly expressed, possibly ignoring implicit features crucial for the target brain region localization and recognition task. Based on this, in the technical solution of this application, an interactive encoding based on fine-grained semantic sharing compensation is further performed on the updated near-infrared spectrum image semantic enhanced encoding feature map and the basic near-infrared spectrum image enhanced semantic encoding feature map to obtain the updated near-infrared spectrum - basic near-infrared spectrum semantic fine-grained contrast feature map.

[0056] In one embodiment, as Figure 5 shown, the near-infrared spectrum interactive encoding unit 43 includes: a near-infrared spectrum feature dissociation unit 431, configured to perform feature dissociation on the updated near-infrared spectrum image semantic enhanced encoding feature map and the basic near-infrared spectrum image enhanced semantic encoding feature map based on the channel dimension to obtain a set of updated near-infrared spectrum image semantic local feature matrices and a set of basic near-infrared spectrum image semantic local feature matrices; a near-infrared spectrum fine-grained semantic sharing compensation subunit 432, configured to perform fine-grained semantic sharing compensation and interaction on the set of updated near-infrared spectrum image semantic local feature matrices and the set of basic near-infrared spectrum image semantic local feature matrices to obtain the updated near-infrared spectrum - basic near-infrared spectrum semantic fine-grained contrast feature map.

[0057] Exemplarily, in the near-infrared spectral feature decoupling unit 431, the updated near-infrared spectral image semantic enhanced encoding feature map and the basic near-infrared spectral image enhanced semantic encoding feature map are respectively subjected to feature decoupling based on the channel dimension to obtain a set of updated near-infrared spectral image semantic local feature matrices and a set of basic near-infrared spectral image semantic local feature matrices. Specifically, this process can be expressed by the formula as follows:

[0058] Decouple(F1)={F 11 ,F 12 ,...,F 1i ,...,F 1n}

[0059] Decouple(F2)={F 21 ,F 22 ,...,F 2i ,...,F 2n}

[0060] where F1 and F2 are the updated near-infrared spectral image semantic enhanced encoding feature map and the basic near-infrared spectral image enhanced semantic encoding feature map respectively, Decouple(·) is the operation of performing feature decoupling on the feature map, F 11 , F 12 , F 1i and F 1n are respectively the 1st, 2nd, ith, and nth updated near-infrared spectral image semantic local feature matrices in the set of updated near-infrared spectral image semantic local feature matrices, and F 21 , F 22 , F 2i and F 2n are respectively the 1st, 2nd, ith, and nth basic near-infrared spectral image semantic local feature matrices in the set of basic near-infrared spectral image semantic local feature matrices.

[0061] That is, through the feature dissociation process, the original enhanced encoded feature map is decomposed into multiple local feature matrices according to different channel dimensions. Each local feature matrix captures the information within a specific spatial position or frequency range, reflecting the change patterns of physiological parameters such as hemoglobin concentration and oxygenation level in that area. This decomposition method allows for a more focused view of the specific details of the internal brain structure and activity patterns, providing more accurate data support for subsequent fine-grained comparison and localization detection. Specifically, in the rTMS intervention system, the updated near-infrared spectroscopy image semantic enhanced encoded feature map records the state of the target brain area after stimulation, while the basic near-infrared spectroscopy image enhanced semantic encoded feature map represents the baseline state of the same area without stimulation. By performing feature dissociation based on the channel dimension on these two sets of data, the complex brain activity signals can be split into several relatively independent but interrelated components. This not only helps to identify the specific changes caused by rTMS stimulation but also enables the differentiation of natural fluctuations and other non-specific interference factors.

[0062] In one embodiment, as Figure 6 shown, the near-infrared spectroscopy fine-grained semantic sharing compensation subunit 432 includes: a semantic sharing processing secondary subunit 4321 for performing semantic sharing on each corresponding channel dimension of the updated near-infrared spectroscopy image semantic local feature matrix set and the basic near-infrared spectroscopy image semantic local feature matrix set in the updated near-infrared spectroscopy image semantic local feature matrix set and the basic near-infrared spectroscopy image semantic local feature matrix set to obtain a set of updated near-infrared spectroscopy - basic near-infrared spectroscopy fine-grained local shared semantic feature matrices; a semantic feature compensation processing secondary subunit 4322 for performing semantic feature compensation on the set of updated near-infrared spectroscopy - basic near-infrared spectroscopy fine-grained local shared semantic feature matrices to obtain a set of updated near-infrared spectroscopy - basic near-infrared spectroscopy semantic compensation text semantic encoded feature matrices; a semantic enhancement fusion processing secondary subunit 4323 for performing semantic enhancement fusion processing on the set of updated near-infrared spectroscopy - basic near-infrared spectroscopy fine-grained local shared semantic feature matrices based on the set of updated near-infrared spectroscopy - basic near-infrared spectroscopy semantic compensation text semantic encoded feature matrices to obtain the updated near-infrared spectroscopy - basic near-infrared spectroscopy semantic fine-grained comparison feature map.

[0063] Exemplarily, in the semantic sharing processing secondary subunit 4321, the updated near-infrared spectral image semantic local feature matrix set and the basic near-infrared spectral image semantic local feature matrix set are subjected to semantic sharing for each group of corresponding channel dimensions of the updated near-infrared spectral image semantic local feature matrix and the basic near-infrared spectral image semantic local feature matrix to obtain a set of updated near-infrared spectral - basic near-infrared spectral fine-grained local shared semantic feature matrices. Specifically, this process can be expressed by the formula as follows:

[0064]

[0065] where ⊙ and are pointwise addition, pointwise multiplication, and pointwise subtraction by position respectively, Concat(·;·;·) is concatenation processing, Cov 1×1 (·) is point convolutional coding, sigmoid(·) is the sigmoid function, and S i is the updated near-infrared spectral - basic near-infrared spectral fine-grained local shared semantic feature matrix between F 1i and F 2i .

[0066] That is to say, semantic sharing processing is not just simple superposition or averaging, but deeply mines the information in each local feature matrix to find out those features that remain consistent before and after stimulation (i.e., commonalities), and those features that change significantly (i.e., differences). For the consistent part, it indicates that these regions may not be affected by rTMS stimulation or are stable functional regions; while for the changed part, it may indicate changes in neuron activity levels or other physiological responses. In this way, the system can more accurately identify which regions are truly affected by the stimulation and which regions remain unchanged, so as to confirm whether the stimulation accurately acts on the expected target brain region. The generated set of updated near-infrared spectral - basic near-infrared spectral fine-grained local shared semantic feature matrices provides more detailed data support for subsequent localization detection. These matrices not only retain the rich details in the original data but also highlight the most critical changes before and after stimulation, enabling the system to evaluate the effect of rTMS at a finer granularity.

[0067] In one embodiment, the semantic feature compensation processing secondary subunit 4322 is configured to: input each updated near-infrared spectrum - basic near-infrared spectrum fine-grained local shared semantic feature matrix in the set of updated near-infrared spectrum - basic near-infrared spectrum fine-grained local shared semantic feature matrices into a semantic compensation decoding module based on a large language model to obtain a set of updated near-infrared spectrum - basic near-infrared spectrum semantic compensation text descriptions; and input each updated near-infrared spectrum - basic near-infrared spectrum semantic compensation text description in the set of updated near-infrared spectrum - basic near-infrared spectrum semantic compensation text descriptions into a semantic encoder based on a text convolutional neural network model to obtain a set of updated near-infrared spectrum - basic near-infrared spectrum semantic compensation text semantic encoding feature matrices. Specifically, this process can be represented by the formula:

[0068] T i = LLM{S i}

[0069] M i = TextCNN{T i}

[0070] where LLM is the semantic compensation decoding operation, T i is the updated near-infrared spectrum - basic near-infrared spectrum semantic compensation text description corresponding to S i , TextCNN is the text convolutional encoding, and M i is the updated near-infrared spectrum - basic near-infrared spectrum semantic compensation text semantic encoding feature matrix corresponding to S i .

[0071] That is, through a series of processes on the updated and baseline near-infrared spectroscopy images, a set of fine-grained local shared semantic feature matrices has been obtained. These matrices capture the subtle changes in the target brain region before and after rTMS stimulation. However, they are still in the form of numerical feature representations and lack direct interpretability. To better understand the biological significance behind these changes and supplement potential important information, a semantic compensation decoding module based on a large language model is introduced. By inputting each fine-grained local shared semantic feature matrix into the semantic compensation decoding module, corresponding semantic compensation text descriptions can be generated. This step utilizes the powerful expressive ability of the large language model to generate biologically interpretable text descriptions based on the context information provided by the feature matrix. For example, for a specific location point, if its hemoglobin concentration increases significantly, the decoding module may generate a textual description indicating enhanced neuronal activity or vasodilation in that region. This transformation not only makes the features more intuitive and understandable but also provides a rich semantic background for subsequent analysis. Next, these semantic compensation text descriptions are input into a semantic encoder based on the Text Convolutional Neural Network (TextCNN) model. TextCNN is good at extracting meaningful features from text data and converting them into a compact and efficient vector representation form. After being processed by the encoder, a set of updated near-infrared spectroscopy - baseline near-infrared spectroscopy semantic compensation text semantic encoding feature matrices is obtained. These encoded feature matrices fuse the spatio-temporal information of the original NIRS data with the semantic descriptions generated by the large language model, forming a more rich and comprehensive data representation. The generated semantic compensation text semantic encoding feature matrices not only retain the key information in the original feature matrices but also enhance the understanding of physiological changes through text descriptions. This helps to more accurately identify which regions are truly affected by rTMS stimulation and which regions remain unchanged, thereby confirming whether the stimulation accurately acts on the expected target brain region.

[0072] In one embodiment, the semantic enhancement fusion processing secondary subunit 4323 is configured to: perform fine-grained semantic interaction compensation on each pair of corresponding updated near-infrared spectroscopy - baseline near-infrared spectroscopy fine-grained local shared semantic feature matrices and updated near-infrared spectroscopy - baseline near-infrared spectroscopy semantic compensation text semantic encoding feature matrices in the set of updated near-infrared spectroscopy - baseline near-infrared spectroscopy fine-grained local shared semantic feature matrices and the set of updated near-infrared spectroscopy - baseline near-infrared spectroscopy semantic compensation text semantic encoding feature matrices to obtain a set of updated near-infrared spectroscopy - baseline near-infrared spectroscopy fine-grained local interaction semantic enhancement feature matrices; aggregate the set of updated near-infrared spectroscopy - baseline near-infrared spectroscopy fine-grained local interaction semantic enhancement feature matrices along the channel dimension to obtain the updated near-infrared spectroscopy - baseline near-infrared spectroscopy semantic fine-grained contrast feature map. Specifically, this process can be represented by the formula:

[0073]

[0074] F b = couple{S b1 , S b2 ,..., S bi ,..., S bn}

[0075] where, M i T is the transpose matrix of M i , is matrix multiplication, D is the scale of M , that is, the width of the matrix multiplied by the height of the matrix, softmax(·) is the softmax function, both α and β are weighting coefficients, S i , S b1 , S b2 , S bi and S bn are the 1st, 2nd, ith, and nth updated near-infrared spectrum - base near-infrared spectrum fine-grained local interaction semantic enhancement feature matrices in the set of updated near-infrared spectrum - base near-infrared spectrum fine-grained local interaction semantic enhancement feature matrices respectively, n is the number of feature matrices in the set of the updated near-infrared spectrum - base near-infrared spectrum fine-grained local interaction semantic enhancement feature matrices, couple{·, ·,..., ·} is the aggregation along the channel dimension for the set of feature matrices, F b is the updated near-infrared spectrum - base near-infrared spectrum semantic fine-grained contrast feature map.

[0076] That is, semantic interaction compensation is not just simple superposition or averaging, but delves deep into the relationship between each fine-grained local shared semantic feature matrix and its corresponding semantic compensation text semantic coding feature matrix. In this way, the feature performance of the same position point in different states can be analyzed more meticulously, and the features that remain consistent before and after the stimulus (i.e., commonalities) and those that change significantly (i.e., differences) can be identified. For the consistent parts, it indicates that these regions may not be affected by rTMS stimulation or are stable functional regions; while for the changing parts, it may indicate changes in neuron activity levels or other physiological responses. The generated updated near-infrared spectroscopy - basic near-infrared spectroscopy fine-grained local interaction semantic enhancement feature matrix not only retains the key information in the original feature matrix but also enhances the understanding of physiological changes through semantic descriptions. For example, certain specific position points may show higher hemoglobin concentration changes or changes in other physiological indicators, and semantic descriptions can provide additional explanations, such as the specific mechanisms of enhanced neuron activity or vasodilation in this region. Through the analysis of these interaction semantic enhancement feature matrices, the rTMS intervention system can more accurately judge whether the actually stimulated brain region is consistent with the predetermined target region and promptly detect any deviations. Finally, to form an overall representation, these interaction semantic enhancement feature matrices are aggregated along the channel dimension, and finally, the updated near-infrared spectroscopy - basic near-infrared spectroscopy semantic fine-grained contrast feature map is obtained.

[0077] In summary, the interactive coding method based on fine-grained semantic sharing compensation can utilize prompt learning to perform fine-grained compensation during the feature interaction between the semantics of the updated near-infrared spectroscopy image and the semantics of the basic near-infrared spectroscopy image, thereby using the prior information of the large model to conduct fine-grained semantic compensation for feature interaction contrast, so as to use prior knowledge to supplement potential important information to generate a richer and more accurate infrared spectroscopy semantic contrast feature representation. Moreover, the introduction of prompt learning also improves the interpretability of feature interaction. This method ensures that even the slightest semantic change in the infrared spectroscopy image will not be overlooked, enabling a more meticulous analysis of the interactive contrast of the semantics of the infrared spectroscopy image before and after rTMS stimulation and improving the accuracy of subsequent localization detection tasks.

[0078] Exemplarily, in the positioning accuracy determination unit 44, based on the updated near-infrared spectrum - basic near-infrared spectrum semantic fine-grained contrast feature map, it is determined whether the positioning accuracy of the target brain region meets the requirements. In one embodiment, the positioning accuracy determination unit is configured to: input the updated near-infrared spectrum - basic near-infrared spectrum semantic fine-grained contrast feature map into a classifier-based positioning recognizer to obtain a positioning result, and the positioning result is used to indicate whether the positioning accuracy of the target brain region meets the requirements. That is, the fine-grained contrast feature information between the updated near-infrared spectrum image semantics and the basic near-infrared spectrum semantics is used for classification processing, so as to perform positioning recognition, thereby determining whether the positioning accuracy of the target brain region meets the requirements. In this way, it is possible to automatically judge the positioning accuracy of the target brain region and detect whether there is a deviation based on the infrared spectrum semantics and the hemodynamic changes in the brain region before and after rTMS stimulation, and send a positioning adjustment prompt signal when there is a deviation, so that it is possible to ensure the positioning accuracy of the target brain region for each stimulation in an automated manner during repetitive transcranial magnetic stimulation, thereby improving the intelligent level of the repetitive transcranial magnetic stimulation intervention system. In one embodiment, the classifier-based positioning recognizer uses an SVM model.

[0079] Preferably, inputting the updated near-infrared spectrum - basic near-infrared spectrum semantic fine-grained contrast feature map into a classifier-based positioning recognizer to obtain a positioning result includes:

[0080] Determine the median eigenvalue, the maximum eigenvalue, and the minimum eigenvalue in the updated near-infrared spectrum - basic near-infrared spectrum semantic fine-grained contrast feature map, and divide the median eigenvalue by the difference between the maximum eigenvalue and the minimum eigenvalue to obtain the infrared spectrum semantic fine-grained contrast distribution probability value, that is, p = f mid / (f max -f min ), where f mid , f max , f min respectively represent the median eigenvalue, the maximum eigenvalue, and the minimum eigenvalue in the updated near-infrared spectrum - basic near-infrared spectrum semantic fine-grained contrast feature map, and p represents the infrared spectrum semantic fine-grained contrast distribution probability value;

[0081] Perform maximum normalization on the updated near-infrared spectrum - basic near-infrared spectrum semantic fine-grained contrast feature map to obtain the infrared spectrum semantic fine-grained contrast probability feature map F;

[0082] Calculate the power function of each eigenvalue of the infrared spectrum semantic fine-grained contrast probability feature map with one minus the infrared spectrum semantic fine-grained contrast distribution probability value as the exponent to obtain the first infrared spectrum semantic fine-grained contrast convergence feature map F1 = F ⊙(1-p), where F represents the infrared spectrum semantic fine-grained contrast probability feature map, (·) ⊙(1-p) represents calculating the power function with the exponent of one minus the infrared spectrum semantic fine-grained contrast distribution probability value, and F1 represents the first infrared spectrum semantic fine-grained contrast convergence feature map;

[0083] Calculating the power function of each eigenvalue of the point difference feature map between the unit feature map and the infrared spectrum semantic fine-grained contrast probability feature map with the infrared spectrum semantic fine-grained contrast distribution probability value as the exponent to obtain the second infrared spectrum semantic fine-grained contrast convergence feature map where represents subtraction by position, F I represents the unit feature map, F represents the infrared spectrum semantic fine-grained contrast probability feature map, (·) ⊙p Calculating the power function with the infrared spectrum semantic fine-grained contrast distribution probability value as the exponent, and F2 represents the second infrared spectrum semantic fine-grained contrast convergence feature map;

[0084] Taking the dot product of the difference between the infrared spectrum semantic fine-grained contrast probability feature map and one minus the infrared spectrum semantic fine-grained contrast distribution probability value to obtain the first infrared spectrum semantic fine-grained contrast limited feature map F3 = F⊙(1 - p), where F represents the infrared spectrum semantic fine-grained contrast probability feature map, p represents the infrared spectrum semantic fine-grained contrast distribution probability value, ⊙ represents dot product, and F3 represents the first infrared spectrum semantic fine-grained contrast limited feature map;

[0085] Taking the dot product of the point difference feature map and the infrared spectrum semantic fine-grained contrast distribution probability value to obtain the second infrared spectrum semantic fine-grained contrast limited feature map where F represents the infrared spectrum semantic fine-grained contrast probability feature map, represents subtraction by position, F I represents the unit feature map, p represents the infrared spectrum semantic fine-grained contrast distribution probability value, ⊙ represents dot product, and F4 represents the second infrared spectrum semantic fine-grained contrast limited feature map;

[0086] Taking the dot product of the first infrared spectrum semantic fine-grained contrast convergence feature map and the second infrared spectrum semantic fine-grained contrast convergence feature map, and then taking the dot product with the first infrared spectrum semantic fine-grained contrast limited feature map and the second infrared spectrum semantic fine-grained contrast limited feature map to obtain the optimized updated near-infrared spectrum - basic near-infrared spectrum semantic fine-grained contrast feature map Among them, F1 represents the first infrared spectrum semantic fine-grained contrast convergence feature map, F2 represents the second infrared spectrum semantic fine-grained contrast convergence feature map, F3 represents the first infrared spectrum semantic fine-grained contrast limiting feature map, F4 represents the second infrared spectrum semantic fine-grained contrast limiting feature map, ⊙ represents dot multiplication, represents addition by position, and F' represents the optimized updated near-infrared spectrum - basic near-infrared spectrum semantic fine-grained contrast feature map;

[0087] Input the optimized updated near-infrared spectrum - basic near-infrared spectrum semantic fine-grained contrast feature map into the classifier-based localization recognizer to obtain a localization result.

[0088] Here, considering that the updated near-infrared spectrum image semantic enhancement coding feature map and the basic near-infrared spectrum image enhanced semantic coding feature map respectively represent the infrared spectrum features of the updated infrared spectrum data of the target brain region after being stimulated and the infrared spectrum features of the basic near-infrared spectrum data of the target brain region without any stimulation, when performing feature enhancement interaction coding based on cross-domain fine-grained semantic compensation, due to the cross-domain fine-grained semantic compensation difference caused by the infrared spectrum semantic feature distribution difference, the updated near-infrared spectrum - basic near-infrared spectrum semantic fine-grained contrast feature map will have probability convergence divergence based on different feature enhancement interactivity, thus affecting the accuracy of the localization result obtained by inputting into the classifier-based localization recognizer.

[0089] Based on this, by taking the cross-entropy form power series of the distribution analysis probability of the updated near-infrared spectrum - basic near-infrared spectrum semantic fine-grained contrast feature map as the probability distribution convergence limit, on the basis of the combination of the feature set distribution and the probability density distribution of the updated near-infrared spectrum - basic near-infrared spectrum semantic fine-grained contrast feature map, perform the class probability convergence approximation of the updated near-infrared spectrum - basic near-infrared spectrum semantic fine-grained contrast feature map, so as to, on the premise of avoiding the fatal potential divergence of the updated near-infrared spectrum - basic near-infrared spectrum semantic fine-grained contrast feature map in conservative convergence control, further use the probability distribution cross-entropy of the updated near-infrared spectrum - basic near-infrared spectrum semantic fine-grained contrast feature map as the objective function to guide the limiting recovery strategy, realize the common agility of the convergence of the feature set of the probability convergence divergence of the updated near-infrared spectrum - basic near-infrared spectrum semantic fine-grained contrast feature map to the probability density distribution space, and improve the accuracy of the localization result obtained by inputting the updated near-infrared spectrum - basic near-infrared spectrum semantic fine-grained contrast feature map into the classifier-based localization recognizer.

[0090] Exemplarily, in the positioning adjustment module 50, in response to the positioning accuracy of the target brain region not meeting the requirements, a positioning adjustment prompt for the target brain region is generated. It should be understood that during the rTMS treatment process, accurate positioning of the target brain region is crucial. If the stimulation position is inaccurate, the expected therapeutic effect may not be achieved, and even side effects or ineffective treatment may occur. Therefore, real-time monitoring and timely adjustment of the positioning are essential for maintaining the effectiveness of the treatment and reducing potential risks. Through dynamic monitoring and timely adjustment, the needs of personalized treatment can be better met, the pertinence and effectiveness of the treatment can be improved, and reliable data support can also be provided for scientific research. Specifically, if it is found that the positioning deviation exceeds the acceptable range, the system will immediately generate a positioning adjustment prompt. This prompt can be conveyed to the operator in the form of vision, audition, or touch. According to the prompt information, the technician can manually or with the aid of automated tools recalibrate the position of the rTMS coil until it returns to the correct stimulation position. Some advanced systems also have the ability of automatic adjustment and can complete this process without manual intervention. After the adjustment is completed, short-term monitoring is carried out again to confirm whether the new positioning is accurate. All relevant operations and data should be completely recorded for future review and further analysis.

[0091] In summary, the repetitive transcranial magnetic stimulation intervention system according to the embodiments of the present application is elucidated. It processes the basic near-infrared spectroscopy data and the updated infrared spectroscopy data together through data processing and analysis algorithms based on artificial intelligence and signal processing, so as to capture the semantics of the basic near-infrared spectroscopy image and the updated near-infrared spectroscopy image, as well as the fine-grained contrast feature representation information between the two, thereby judging the positioning accuracy of the target brain region to determine whether the positioning accuracy of the target brain region meets the requirements. In this way, it is possible to automatically judge the positioning accuracy of the target brain region and detect whether there is deviation based on the infrared spectroscopy semantics and the hemodynamic changes in the brain region before and after rTMS stimulation, and send out a positioning adjustment prompt signal when there is deviation. In this way, it is possible to ensure the positioning accuracy of the target brain region for each stimulation in an automated manner during the repetitive transcranial magnetic stimulation process, thereby improving the intelligent level of the repetitive transcranial magnetic stimulation intervention system.

[0092] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or by a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0093] It should be understood that the specific examples in this text are only to help those skilled in the art better understand the embodiments of the present application, rather than limiting the scope of the embodiments of the present application.

[0094] It should also be understood that in various embodiments of the present application, the magnitude of the serial numbers of each process does not mean the sequence of execution, and the execution sequence of each process should be determined according to its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0095] It should also be understood that the various embodiments described in this specification can be implemented alone or in combination, and the embodiments of the present application do not limit this.

[0096] Unless otherwise specified, all technical and scientific terms used in the embodiments of the present application have the same meaning as commonly understood by those skilled in the technical field of the present application. The terms used in the present application are only for the purpose of describing specific embodiments and are not intended to limit the scope of the present application. The term "and / or" used in the present application includes any and all combinations of one or more of the related listed items. The singular forms of "a", "above-mentioned" and "the" used in the embodiments of the present application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. In addition, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0097] In several embodiments provided by the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only illustrative. For example, the division of the units is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point, the displayed or discussed coupling or direct coupling or communication connection to each other can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be in electrical, mechanical or other forms.

[0098] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0099] In addition, in each embodiment of the present application, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit.

[0100] When the above-mentioned function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of this application. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs.

[0101] As described above, the above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed by this application can easily think of changes or substitutions, which should all be covered within the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

Claims

1. A repetitive transcranial magnetic stimulation intervention system, characterized in that: include: A brain region initial positioning data acquisition module is used to record the initial positioning data of the target brain region; A basic near-infrared spectral data acquisition module, used to acquire basic near-infrared spectral data of the target brain area without any stimulation using a near-infrared spectral imaging NIRS monitoring device; An updated infrared spectrum data acquisition module is used to control the repetitive transcranial magnetic stimulation (rTMS) device to stimulate the target brain area, and at the same time, the NIRS monitoring device is used to continuously monitor the near-infrared spectrum changes of the target brain area to obtain updated infrared spectrum data; A target brain region positioning detection module, used to determine whether the positioning accuracy of the target brain region meets the requirements based on the comparison between the updated infrared spectrum data and the basic near-infrared spectrum data; A positioning adjustment module, configured to generate a target brain region positioning adjustment prompt in response to the positioning accuracy of the target brain region not meeting the requirement; Wherein, the target brain area positioning detection module includes: A near infrared spectrum feature extraction unit, used for performing feature extraction based on the near infrared spectrum on the updated infrared spectrum data and the basic near infrared spectrum data respectively to obtain an updated near infrared spectrum image semantic coding feature map and a basic near infrared spectrum image semantic coding feature map; A near infrared spectrum explicit feature enhancement processing unit is used to perform explicit feature enhancement processing based on spatial domain conversion on the updated near infrared spectrum image semantic coding feature map and the basic near infrared spectrum image semantic coding feature map to obtain an updated near infrared spectrum image semantic enhancement coding feature map and a basic near infrared spectrum image enhanced semantic coding feature map; A near infrared spectrum interactive coding unit, used for interactively coding the updated near infrared spectrum image semantic enhancement coding feature map and the basic near infrared spectrum image enhanced semantic coding feature map based on fine-grained semantic shared compensation to obtain an updated near infrared spectrum - basic near infrared spectrum semantic fine-grained contrast feature map, wherein the near infrared spectrum interactive coding unit comprises: a near infrared spectrum feature decomposition unit, used for performing feature decomposition based on the channel dimension on the updated near infrared spectrum image semantic enhancement coding feature map and the basic near infrared spectrum image enhanced semantic coding feature map respectively to obtain a set of updated near infrared spectrum image semantic local feature matrices and a set of basic near infrared spectrum image semantic local feature matrices; a near infrared spectrum fine-grained semantic shared compensation subunit, used for performing fine-grained semantic shared compensation and interaction on the set of updated near infrared spectrum image semantic local feature matrices and the set of basic near infrared spectrum image semantic local feature matrices to obtain the updated near infrared spectrum - basic near infrared spectrum semantic fine-grained contrast feature map; The positioning accuracy determination unit is used to determine whether the positioning accuracy of the target brain area meets the requirements based on the updated near-infrared spectrum-basic near-infrared spectrum semantic fine-grained comparison feature map.

2. The repetitive transcranial magnetic stimulation intervention system according to claim 1, characterized in that: The near infrared spectrum feature extraction unit is used to: input the updated infrared spectrum data and the basic near infrared spectrum data into a near infrared spectrum feature extractor based on a Mobile-Former model to obtain the updated near infrared spectrum image semantic coding feature map and the basic near infrared spectrum image semantic coding feature map.

3. The repetitive transcranial magnetic stimulation intervention system according to claim 2, characterized in that: The near-infrared spectrum dominant feature enhancement processing unit comprises: The near infrared spectrum pixel granularity feature decoupling subunit is used to perform pixel granularity feature decoupling on the updated near infrared spectrum image semantic coding feature map to obtain a set of updated near infrared spectrum pixel granularity decoupling feature vectors. The process can be expressed as follows: decouple(F)={v1,v2,...,v i ,...,v n } Wherein, F is the updated near infrared spectral image semantic coding feature map, decouple(F) is the pixel granularity feature decoupling of F, v1, v2, v i and v n The first, second, i-th and n-th updated near-infrared spectrum pixel granularity decoupling feature vectors in the set of updated near-infrared spectrum pixel granularity decoupling feature vectors are respectively updated; The near infrared spectrum structural semantic saliency measurement subunit is used to calculate the fine-grained structural semantic saliency measurement coefficient of each updated near infrared spectrum pixel granularity decoupling feature vector in the set of updated near infrared spectrum pixel granularity decoupling feature vectors to obtain an updated near infrared spectrum structural semantic saliency distribution matrix. The process can be expressed as follows: h i =W1v i W2 Where W1 and W2 are the first weight matrix and the second weight matrix respectively, h i v i The corresponding updated near-infrared spectral hyperbolic spatial modulation pixel granularity decoupling feature vector, ||·|| 2 is the square of the norm of the vector, log2 represents the logarithmic function value with base 2, s i is to update the semantic saliency distribution matrix of the near infrared spectral structure h i The corresponding updated near-infrared spectral fine-grained structure semantic saliency metric coefficient; The near infrared spectrum feature explicit hierarchical precipitation processing subunit is used to perform feature explicit hierarchical precipitation processing based on a multi-layer mask function on the updated near infrared spectrum structure semantic saliency distribution matrix to obtain an updated near infrared spectrum structure semantic saliency distribution weight matrix. The process can be expressed as follows: Among them, θ is the preset threshold, mask is the masking operation, and s i ' is to update the near infrared spectral structure semantic saliency distribution weight matrix h i The corresponding updated near-infrared spectral structure semantic significance distribution weight value; The position point multiplication processing subunit is used to calculate the position point multiplication between the updated near-infrared spectral structure semantic significance distribution weight matrix and the various feature matrices along the channel dimension of the updated near-infrared spectral image semantic encoding feature map to obtain the updated near-infrared spectral image semantic enhancement encoding feature map.

4. The repetitive transcranial magnetic stimulation intervention system according to claim 3, characterized in that: The near infrared spectral structure semantic saliency measurement subunit is used to: Inputting each updated near infrared spectrum pixel granularity decoupling feature vector in the set of updated near infrared spectrum pixel granularity decoupling feature vectors into a hyperbolic space mapper to obtain a set of updated near infrared spectrum hyperbolic space modulation pixel granularity decoupling feature vectors; The fine-grained structural semantic significance measurement coefficient of each updated near-infrared spectrum hyperbolic spatial modulation pixel granularity decoupling feature vector in the set of the updated near-infrared spectrum hyperbolic spatial modulation pixel granularity decoupling feature vectors is calculated to obtain the updated near-infrared spectrum structural semantic significance distribution matrix formed by arranging multiple updated near-infrared spectrum fine-grained structural semantic significance measurement coefficients.

5. The repetitive transcranial magnetic stimulation intervention system according to claim 4, characterized in that: The near infrared spectroscopy fine-grained semantic sharing compensation subunit includes: The semantic sharing processing secondary subunit is used to semantically share the updated near infrared spectrum image semantic local feature matrix and the basic near infrared spectrum image semantic local feature matrix of each corresponding channel dimension in the set of the updated near infrared spectrum image semantic local feature matrix and the set of the basic near infrared spectrum image semantic local feature matrix to obtain a set of updated near infrared spectrum-basic near infrared spectrum fine-grained local shared semantic feature matrices. The process can be expressed by the formula: Among them, F 1i is the ith updated near infrared spectral image semantic local feature matrix in the set of updated near infrared spectral image semantic local feature matrices, F 2i is the i-th basic near infrared spectral image semantic local feature matrix in the set of basic near infrared spectral image semantic local feature matrices, ⊙ and They are point-by-point addition, point-by-point multiplication, and point-by-point subtraction. Concat(·;·;·) is cascade processing. Cov 1×1 (·) is the point convolutional coding, sigmoid(·) is the sigmoid function, S i Yes F 1i and F 2i Updated NIR-basic NIR fine-grained local shared semantic feature matrix between them; The semantic feature compensation processing secondary subunit is used to perform semantic feature compensation on the set of updated near infrared spectrum-basic near infrared spectrum fine-grained local shared semantic feature matrices to obtain a set of updated near infrared spectrum-basic near infrared spectrum semantic compensation text semantic encoding feature matrices. The process can be expressed as follows: T i =LLM{S i } M i =TextCNN{T i } Among them, LLM is the semantic compensation decoding operation, T i For S i The corresponding updated near infrared spectrum-based near infrared spectrum semantic compensation text description, TextCNN is text convolutional coding, M i For S i The corresponding updated near infrared spectrum-basic near infrared spectrum semantic compensation text semantic encoding feature matrix; The semantic enhancement fusion processing secondary subunit is used to perform semantic enhancement fusion processing on the set of updated near infrared spectrum-basic near infrared spectrum fine-grained local shared semantic feature matrices based on the set of updated near infrared spectrum-basic near infrared spectrum semantic compensation text semantic encoding feature matrices to obtain the updated near infrared spectrum-basic near infrared spectrum semantic fine-grained comparison feature map. The process can be expressed by the formula: F b =couple{S b1 ,S b2 ,...,S bi ,...,S bn } Among them, M i T M i The transposed matrix of is matrix multiplication, D is M i The scale is the matrix width multiplied by the matrix height, softmax(·) is the softmax function, α and β are weighting coefficients, S b1 , S b2 , S bi and S bn are the first, second, i-th and n-th updated near infrared spectrum - basic near infrared spectrum fine-grained local interaction semantic reinforcement feature matrices in the set of updated near infrared spectrum - basic near infrared spectrum fine-grained local interaction semantic reinforcement feature matrices, n is the number of feature matrices in the set of updated near infrared spectrum - basic near infrared spectrum fine-grained local interaction semantic reinforcement feature matrices, couple{·,·,...,·} is the aggregation of the set of feature matrices along the channel dimension, F b It is the updated near infrared spectrum-basic near infrared spectrum semantic fine-grained contrast feature map.

6. The repetitive transcranial magnetic stimulation intervention system according to claim 5, characterized in that: The semantic feature compensation processing secondary subunit is used to: Inputting each updated near infrared spectrum-basic near infrared spectrum fine-grained local shared semantic feature matrix in the set of updated near infrared spectrum-basic near infrared spectrum fine-grained local shared semantic feature matrix into a semantic compensation decoding module based on a large language model to obtain a set of updated near infrared spectrum-basic near infrared spectrum semantic compensation text description; Each updated near infrared spectrum - basic near infrared spectrum semantic compensation text description in the set of updated near infrared spectrum - basic near infrared spectrum semantic compensation text description is input into the semantic encoder based on the text convolutional neural network model to obtain the set of updated near infrared spectrum - basic near infrared spectrum semantic compensation text semantic encoding feature matrix.

7. The repetitive transcranial magnetic stimulation intervention system according to claim 6, characterized in that: The semantic enhancement fusion processing secondary sub-unit is used to: Perform fine-grained semantic interaction compensation on each corresponding group of updated near infrared spectrum - basic near infrared spectrum fine-grained local shared semantic feature matrices and updated near infrared spectrum - basic near infrared spectrum semantic compensation text semantic encoding feature matrices in the set of the updated near infrared spectrum - basic near infrared spectrum fine-grained local shared semantic feature matrices and the set of the updated near infrared spectrum - basic near infrared spectrum semantic compensation text semantic encoding feature matrices to obtain a set of updated near infrared spectrum - basic near infrared spectrum fine-grained local interactive semantic reinforcement feature matrices; The set of the updated near infrared spectrum-basic near infrared spectrum fine-grained local interactive semantic reinforcement feature matrices is aggregated along the channel dimension to obtain the updated near infrared spectrum-basic near infrared spectrum semantic fine-grained comparison feature map.

8. The repetitive transcranial magnetic stimulation intervention system according to claim 7, characterized in that: The positioning accuracy determination unit is used to: input the updated near-infrared spectrum-basic near-infrared spectrum semantic fine-grained comparison feature map into a classifier-based positioning identifier to obtain a positioning result, and the positioning result is used to indicate whether the positioning accuracy of the target brain area meets the requirements.

Images