Dura mater three-dimensional modeling method, system, device, storage medium and program product

Abstract

The application relates to the technical field of medical imaging, and provides a dura mater three-dimensional modeling method, a dura mater three-dimensional modeling system, a dura mater three-dimensional modeling device, a storage medium and a program product, which comprise the following steps: acquiring a first weighted structure image and a second weighted structure image of a target user and performing spatial registration; segmenting the first weighted structure image to obtain a brain parenchyma mask and constructing a brain parenchyma smooth outer mask; constructing a dura mater candidate search space according to the three-dimensional physical distance from a spatial voxel to the brain parenchyma smooth outer mask; adaptively determining an individualized segmentation threshold according to the image intensity distribution characteristics; performing weighted calculation in combination with signal intensity, spatial distance prior and the individualized segmentation threshold to obtain a voxel attribution probability graph; segmenting a dura mater region according to the voxel attribution probability graph and performing three-dimensional reconstruction to determine an individualized dura mater three-dimensional model. The application can improve the individualization degree and segmentation precision of dura mater three-dimensional modeling, and provides a reliable data basis for preoperative planning and electrode implantation.

Description

Technical Field

[0001] This application relates to the field of medical image processing technology, and in particular to a method, system, device, storage medium and program product for three-dimensional modeling of the dura mater. Background Technology

[0002] Currently, in invasive brain-computer interface applications, the accuracy of the segmentation boundaries, surface continuity, and local detail representation of the 3D model of the dura mater directly affect the preoperative judgment of the implantation area, craniotomy range, instrument placement path, and electrode fit.

[0003] In existing technologies, the automation and accuracy of constructing a three-dimensional model of the whole brain dura mater based on individual user structural images are limited, making it difficult to fully reflect individual differences in intracranial brain anatomy. Inaccurate segmentation boundaries or insufficient surface continuity will reduce the reliability and safety of surgical planning. In addition, the localization of brain regions of interest often relies on manual annotation or population activation statistics, which may lead to deviations in the localization of individual brain regions and make it difficult to fully adapt to the differences of individual users. This, in turn, affects the coverage of the electrode array, the selection of implantation sites, and the subsequent neural signal acquisition and decoding effects.

[0004] Therefore, there is an urgent need for a technical solution that can improve the individualization and segmentation accuracy of 3D modeling of the dura mater. Summary of the Invention

[0005] To address the limitations of existing technologies in terms of the degree of individualization and segmentation accuracy in 3D modeling of the dura mater, this application provides a 3D modeling method, system, electronic device, and storage medium for the dura mater, enabling high-precision and individualized 3D modeling of the dura mater and providing a reliable data foundation for preoperative planning, navigation and positioning, and electrode implantation.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A method for three-dimensional modeling of the dura mater includes: acquiring a first weighted structural image and a second weighted structural image of a target user; registering the second weighted structural image to the three-dimensional voxel space of the first weighted structural image; segmenting the first weighted structural image to obtain a brain parenchyma mask; constructing a smooth outer mask of the brain parenchyma based on the brain parenchyma mask; constructing a dura mater candidate search space based on the three-dimensional physical distance from the spatial voxels to the smooth outer mask of the brain parenchyma; adaptively determining an individualized segmentation threshold based on the image intensity distribution characteristics corresponding to the target user; performing weighted calculations based on the signal intensity of the first weighted structural image, the signal intensity of the second weighted image, and the spatial distance prior corresponding to the dura mater candidate search space, combined with the individualized segmentation threshold, to obtain a voxel assignment probability map; segmenting the dura mater region based on the voxel assignment probability map, and performing three-dimensional reconstruction on the segmentation results to determine an individualized three-dimensional model of the dura mater.

[0008] In this way, by constructing a candidate search space for the dura mater to provide anatomical constraints, combining individualized threshold estimation to adaptively determine segmentation parameters, and then making decisions by integrating multimodal information through multi-feature probability fusion, high-precision and individualized 3D modeling of the dura mater can be achieved, solving the problem of limited individualization and segmentation accuracy in existing dura mater modeling technologies.

[0009] In some embodiments, the step of segmenting the first weighted structural image to obtain a brain parenchyma mask and constructing a smooth outer mask of the brain parenchyma based on the brain parenchyma mask includes: performing outlier truncation and normalization on the voxel intensity based on the statistical quantile values ​​of the global voxel intensity of the first weighted structural image; extracting initial candidate masks based on the normalized voxel intensity, removing small outlier noise points through three-dimensional morphological opening operations, and performing connected component hole filling processing; extracting the connected component with the largest number of three-dimensional voxels as the basic mask, first reducing boundary residue and then restoring the boundary on the outer contour of the basic mask to determine the brain parenchyma mask; performing morphological closing operations and hole filling on the brain parenchyma mask to eliminate brain sulci depressions and obtain the smooth outer mask of the brain parenchyma.

[0010] In this way, by eliminating sulci and depressions in the brain through morphological closure operations and hole filling, a smooth outer mask of macroscopically continuous brain parenchyma is constructed, providing an accurate structural prior basis for the subsequent construction of the dura mater candidate search space.

[0011] In some embodiments, constructing the dura mater candidate search space based on the three-dimensional physical distance from the spatial voxel to the smooth outer mask of the brain parenchyma includes: calculating the outer distance from the spatial voxel to the outer side of the smooth outer mask of the brain parenchyma, wherein the outer distance is the minimum physical distance from the voxel to the smooth outer mask of the brain parenchyma, and for voxels belonging to the smooth outer mask of the brain parenchyma, the outer distance is zero; constructing the dura mater candidate search space based on the outer distance and a preset medial dura mater gap and dura mater thickness, wherein the medial dura mater gap represents the gap distance between the medial surface of the dura mater and the surface of the brain parenchyma, and the dura mater thickness represents the thickness parameter of the dura mater.

[0012] In this way, by constructing a search shell using lateral distance and anatomical parameters, the spatial range in which the dura mater may exist is limited, reducing the probability of missegmentation far from the target area.

[0013] In some embodiments, the step of adaptively determining the individualized segmentation threshold based on the image intensity distribution characteristics corresponding to the target user includes: extracting the signal intensity percentile of the first weighted structural image within the brain parenchyma mask to determine a first signal reference threshold; extracting the signal intensity percentage of the first weighted structural image within the brain parenchyma mask and the dura mater candidate search space to determine a first lower signal threshold and a first upper signal threshold; and extracting the signal intensity percentage within the dura mater candidate search space to determine a second upper signal threshold.

[0014] In this way, the individualized segmentation threshold is adaptively determined based on the image intensity distribution corresponding to a single target user, transforming the fixed threshold judgment into an individualized statistical judgment, enabling the system to adapt to the differences in image brightness, contrast, and tissue signals of different target users.

[0015] In some embodiments, the first signal lower threshold includes the maximum value of a preset low quantile among the percentages of signal intensity of the first weighted structural images within the brain parenchyma mask and the dura mater candidate search space; the first signal reference threshold includes the percentage value of signal intensity of the first weighted structural images within the brain parenchyma mask; and the first signal upper threshold includes the maximum value of a preset high quantile among the percentages of signal intensity of the first weighted structural images within the brain parenchyma mask and the dura mater candidate search space.

[0016] In this way, the lower threshold is determined by the low quantile value to exclude low signal regions, and the upper threshold is determined by the high quantile value to reduce the probability of high signal regions entering the candidate layer, thereby improving the segmentation accuracy.

[0017] In some embodiments, the method for obtaining the voxel assignment probability map includes: calculating a first signal score based on the signal intensity of the first weighted structural image, wherein the first signal score includes a first signal lower limit score and a high signal upper limit score; calculating a second signal score based on the signal intensity of the second weighted structural image and a second signal upper limit threshold; calculating a distance prior score based on a spatial distance prior; and performing a weighted calculation on the first signal score, the second signal score, and the distance prior score to obtain the voxel assignment probability map.

[0018] In this way, by combining T1 features, T2 features, and spatial distance priors, the candidate probabilities of the dura mater are calculated in a weighted manner, making the determination of candidate dura mater regions more robust.

[0019] In some embodiments, the first signal score is calculated as follows: based on the relationship between the signal intensity of the first weighted structural image, a first signal lower limit threshold, and a first signal reference threshold, the first signal lower limit score is calculated using a smooth step function; based on the relationship between the signal intensity of the first weighted structural image and a first signal upper limit threshold, the high signal upper limit score is calculated using the smooth step function; the first signal lower limit score and the high signal upper limit score are multiplied to obtain the first signal score; the second signal score is calculated as follows: based on the relationship between the signal intensity of the second weighted structural image and a second signal upper limit threshold, the second signal score is calculated using a smooth step function; the distance prior score is calculated as follows: based on the minimum physical distance from the voxel to the smooth outer mask of the brain parenchyma, the distance prior score is calculated using an inverse linear decay function.

[0020] In this way, soft decision-making is achieved through the Sigmoid smooth step function, avoiding the boundary discontinuity problem caused by hard threshold, while the distance prior makes the candidate probability preferentially distributed near the inner side of the search shell.

[0021] In some embodiments, the weighted probability fusion calculation based on the individualized segmentation threshold is implemented by multiplying the first signal score, the second signal score, and the distance prior score by their respective first weight, second weight, and third weight, and then summing them.

[0022] In this way, the contribution of each feature can be flexibly adjusted by weighting parameters to adapt to different image qualities and individual differences.

[0023] In some embodiments, segmenting the dura mater region according to the voxel assignment probability map and performing three-dimensional reconstruction on the segmentation results to determine an individualized dura mater three-dimensional model includes: marking voxels with probability values ​​higher than a preset segmentation judgment threshold in the voxel assignment probability map as target voxels, and binarizing them to determine a coarse dura mater mask; performing a spatial topological connectivity check on the coarse dura mater mask, filtering out isolated misjudged regions with volume sizes smaller than a preset tolerance, and repairing local broken boundaries to determine a continuous fine dura mater mask; extracting surface isosurfaces from the fine dura mater mask and performing mesh rendering to determine the individualized dura mater three-dimensional model.

[0024] In this way, post-processing operations can remove isolated small areas, smooth boundaries, and repair local fractures, ensuring surface continuity and improving the quality of the 3D model.

[0025] In some embodiments, registering the second weighted structural image to the three-dimensional voxel space of the first weighted structural image includes: extracting the spatial geometric relationship between the first weighted structural image and the second weighted structural image; iteratively optimizing the rotational spatial parameters and translational spatial parameters in six degrees of freedom with the optimization objective of maximizing the global pixel mutual information between the two structural images; and resampling the second weighted structural image using the optimized rotational spatial parameters and translational spatial parameters to complete cross-modal rigid registration.

[0026] This solves the problem of inconsistencies in spatial resolution, voxel coordinates, and physical coordinates between different image sequences, enabling two modal images to correspond to the same anatomical location under the same voxel grid.

[0027] In some embodiments, the method further includes the step of determining candidate implantation regions for subdural electrodes based on task-based functional magnetic resonance imaging (fMRI), specifically including: acquiring the fMRI signal of the target user and task time-series information corresponding to the fMRI signal; constructing a design matrix based on the task time-series information and performing statistical model estimation on the fMRI signal to obtain an activation statistics map related to the target task; registering the activation statistics map to the space where the first weighted structural image is located to determine an individual functional activation map; projecting the individual functional activation map onto the individual cortical surface to determine the spatial distribution of the target functional area on the individual cortical surface, and determining candidate implantation regions for subdural electrodes based on the spatial distribution on the individual cortical surface.

[0028] This avoids relying solely on standard brain templates, population average activation maps, or human experience to locate target brain regions, and more accurately reflects the functional area distribution of the target user, providing individualized reference for electrode implantation.

[0029] In some embodiments, before acquiring the functional magnetic resonance imaging (fMRI) signal of the target user and the task timing information corresponding to the fMRI signal, the method further includes: determining a fMRI experimental paradigm based on decoding requirements, wherein the fMRI experimental paradigm includes a task stimulus sequence matching the decoding requirements; the fMRI experimental paradigm consists of multiple task cycles, each task cycle including a fixation phase, a task cueing phase, a task execution phase, and a rest phase, wherein the content of the task execution phase corresponds to the decoding requirements; performing task-oriented fMRI scans on the target user based on the fMRI experimental paradigm, acquiring the fMRI signal, and recording the task timing information corresponding to the fMRI experimental paradigm; the decoding requirements include at least one of hand movement decoding requirements or language decoding requirements, wherein when the decoding requirement is a hand movement decoding requirement, the task execution phase is a motor imagery phase, and when the decoding requirement is a language decoding requirement, the task execution phase is a word imagery phase.

[0030] In this way, by designing a matching experimental paradigm based on the decoding requirements, the acquired BOLD signal can effectively activate the target functional area corresponding to the decoding requirements, ensuring the pertinence and effectiveness of the functional positioning.

[0031] In some embodiments, determining the candidate implantation region for the subdural electrode based on the spatial distribution on the individual cortical surface includes: determining the medial surface of the dura mater based on the dural candidate search space; calculating the minimum distance from each vertex on the medial surface of the dura mater to the spatial distribution of the target functional area on the individual cortical surface; mapping the functional activation thermal values ​​on the individual cortical surface to the medial surface of the dura mater based on the minimum distance, wherein a distance attenuation parameter controls the attenuation rate of the mapping weight as the distance increases; and filtering the functional mapping intensity on the medial surface of the dura mater based on preset activation thresholds and cluster thresholds to determine the candidate implantation region for the subdural electrode.

[0032] Thus, achieving the spatial transformation from "brain function activation location" to "subdural electrode coverage location" is an important bridge connecting functional positioning and implantation planning.

[0033] In some embodiments, the method further includes: determining an avoidance mask based on the dura mater candidate search space and the brain parenchyma mask; wherein the avoidance mask represents a region located within the dura mater mask but not belonging to the dura mater mask or the brain parenchyma mask, and the avoidance mask is used as an avoidance region or a safety constraint region in the step of determining the subdura mater electrode candidate implantation region based on task-based functional magnetic resonance imaging.

[0034] This provides a safe constraint area for subsequent electrode placement, instrument path planning, or candidate area selection, thereby improving surgical safety.

[0035] Furthermore, this invention also provides a 3D modeling system for the dura mater, comprising: a data acquisition module for acquiring a first weighted structural image and a second weighted structural image of a target user; a spatial registration module for spatially registering the second weighted structural image; a brain parenchyma segmentation module for segmenting brain parenchyma based on the first weighted structural image to obtain a brain parenchyma mask; a search space construction module for constructing a smooth outer mask of brain parenchyma based on the brain parenchyma mask, and constructing a candidate search space for the dura mater based on the smooth outer mask of brain parenchyma; a threshold estimation module for adaptively determining an individualized segmentation threshold based on the image intensity distribution corresponding to the target user; a probability fusion module for performing weighted calculations based on the signal intensity of the first weighted structural image, the signal intensity of the second weighted structural image, and the spatial distance prior corresponding to the candidate search space for the dura mater, combined with the individualized segmentation threshold, to obtain a voxel assignment probability map; and a 3D model determination module for segmenting the dura mater region based on the voxel assignment probability map, and performing 3D reconstruction on the segmentation results to determine an individualized 3D model of the dura mater.

[0036] In this way, the system achieves high-precision, individualized 3D modeling of the dura mater through the collaborative work of its various modules.

[0037] Furthermore, the present invention also provides an electronic device, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to implement the method described in any of the preceding embodiments when executing the instructions.

[0038] In this way, the electronic device executes the stored instructions through the processor to realize the three-dimensional modeling method of the dura mater.

[0039] In addition, the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the method described in any of the above-mentioned embodiments.

[0040] In this way, when the program instructions stored in the storage medium are executed by the processor, a three-dimensional modeling method for the dura mater is realized.

[0041] In addition, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the method described in any of the above-mentioned embodiments.

[0042] In this way, when the computer program product is executed by the processor, a three-dimensional modeling method for the dura mater is realized.

[0043] Compared with the prior art, this application has the following beneficial effects:

[0044] 1. This invention performs whole-brain dura mater modeling based on the first and second weighted structural images of individual users. It fully utilizes the complementary information in tissue contrast from different MRI sequences, thereby improving the accuracy of identifying the boundaries of the dura mater and its adjacent tissues. Compared to methods relying on standard brain templates, population average models, or manual experience-based delineation, this invention directly completes image registration, brain parenchyma segmentation, candidate search space construction, and dura mater probability estimation within the target user's individual space. This better reflects individual intracranial anatomical differences, reduces localization bias caused by template matching errors or manual annotation differences, and enhances the individualization and clinical usability of the dura mater 3D model.

[0045] 2. This invention employs individualized threshold estimation and multi-feature probability fusion mechanisms to avoid the problem of insufficient adaptability of fixed MRI intensity thresholds under different target users, different scanning devices, or different scanning parameters. Instead of directly using a uniform T1 or T2 intensity threshold, the system calculates quantile thresholds within the target user's own brain parenchyma region and the dura mater candidate search space. Furthermore, it combines T1 low signal exclusion, T1 high signal penalty, T2 low signal features, and distance priors to form a dura mater candidate probability map. This method can simultaneously consider image intensity features and anatomical spatial constraints, making the determination of dura mater candidate regions more robust and reducing the risk of missegmentation caused by a single modality, single threshold, or single morphological rule.

[0046] 3. This invention can also combine task-based fMRI data to achieve individualized functional area localization. The system obtains an individual functional activation map related to the target task through task design, BOLD signal analysis, design matrix construction, and generalized linear model estimation. This map is then registered to the T1 individual space and projected onto the cortical surface, and further mapped to the dura mater candidate layer or the medial surface of the dura mater. Therefore, this invention not only provides an anatomically meaningful three-dimensional model of the dura mater but also establishes a spatial correspondence between functional activation areas and candidate implantation areas of subdural electrodes, providing an individualized basis for assessing electrode coverage in key functional areas such as the sensorimotor, language, and visual cortices.

[0047] 4. This invention provides a more reliable preoperative planning basis for invasive brain-computer interfaces and neurosurgical applications. Through individualized dura mater models and functional area mapping results, physicians or planning systems can more intuitively assess the craniotomy area, dura mater incision area, electrode array coverage, electrode fit, instrument entry path, and spatial relationship with key functional areas, thereby helping to reduce the risks of implantation deviation, tissue damage, and functional impairment. Overall, this invention combines multimodal medical image processing, individualized 3D modeling, probabilistic segmentation, and functional area mapping, offering advantages such as high automation, strong individual adaptability, high visualization of output results, and ease of integration with preoperative planning and navigation systems. It can significantly improve the accuracy, safety, and practical value of dura mater and subdura mater electrode implantation planning. Attached Figure Description

[0048] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, wherein:

[0049] Figure 1 This is a schematic flowchart of a three-dimensional modeling method for the dura mater provided in one embodiment of this application;

[0050] Figure 2 This is a schematic diagram of the process of registering a second weighted structural image to the three-dimensional voxel space of a first weighted structural image according to one embodiment of this application;

[0051] Figure 3 This is a schematic diagram of the process for constructing a smooth outer mask of brain parenchyma according to one embodiment of this application;

[0052] Figure 4 This is a schematic diagram of the process for constructing a candidate search space for the dura mater provided in one embodiment of this application;

[0053] Figure 5 This is a schematic diagram of the process for determining an individualized segmentation threshold provided in one embodiment of this application;

[0054] Figure 6 This is a flowchart illustrating the determination of voxel assignment probability map provided in one embodiment of this application;

[0055] Figure 7 This is a flowchart illustrating the process of determining a personalized three-dimensional model of the dura mater, provided in one embodiment of this application.

[0056] Figure 8 This is a flowchart illustrating the process of determining a candidate implantation region for a subdural electrode, provided in one embodiment of this application.

[0057] Figure 9 This is a schematic diagram of an experimental paradigm provided in one embodiment of this application;

[0058] Figure 10 This is a schematic diagram of the structure of a three-dimensional modeling system for the dura mater provided in one embodiment of this application.

[0059] In the picture:

[0060] 1. Data acquisition module; 2. Spatial registration module; 3. Brain parenchyma segmentation module; 4. Search space construction module; 5. Threshold estimation module; 6. Probability fusion module; 7. 3D model determination module. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0062] The technical solutions of the various embodiments of this application can be combined with each other, but only if they are based on the ability of a person skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by this application.

[0063] This solution does not aim to obtain disease diagnosis results or health status. It is a system that processes users' medical imaging data and uses it to achieve three-dimensional modeling of the dura mater. All steps are information processing methods implemented by computers and other devices.

[0064] It should be fully understood that the user information involved in this application (including but not limited to T1 weighted structured images, T2 weighted structured images, etc.) is information and data authorized by the user or fully authorized by all parties. The use of user information should comply with industry privacy policies and practices that are generally considered to meet or exceed the requirements for maintaining user privacy. The collection, use and processing of related data should comply with relevant laws, regulations and standards, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0065] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, specific embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0066] Example 1: As Figures 1-9 As shown, this embodiment provides a method for three-dimensional modeling of the dura mater. This method provides anatomical constraints by constructing a candidate search space for the dura mater, adaptively determines segmentation parameters by combining individualized threshold estimation, and then makes decisions by integrating multimodal information through multi-feature probability fusion, thereby achieving high-precision, individualized three-dimensional modeling of the dura mater. Specifically, it includes the following steps:

[0067] S100: Obtain the first weighted structural image and the second weighted structural image of the target user, and register the second weighted structural image to the three-dimensional voxel space of the first weighted structural image.

[0068] Specifically, the first weighted structural image referred to in this embodiment is the T1-weighted structural image, which serves as the primary image and provides high-quality information on the individual brain anatomy. On the image, tissues with shorter T1 relaxation times appear as high signal intensity, adipose tissue is brighter, and tissues with higher water content or cerebrospinal fluid are darker. The second weighted structural image refers to the T2-weighted structural image, which serves as an auxiliary image and supplements the contrast information of different tissues near water content, cerebrospinal fluid, and intracranial boundaries. On the image, tissues with longer T2 relaxation times appear as high signal intensity, and areas with higher water content, cerebrospinal fluid, or inflammation are generally brighter.

[0069] Specifically, the formula for obtaining the T1 and T2 weighted structural images can be:

[0070]

[0071]

[0072]

[0073] ,

[0074] In the formula, This represents a T1-weighted structural image. This represents the three-dimensional voxel space where the T1-weighted structure image resides. Represents the first in the three-dimensional voxel space T1 signal intensity of individual units. This represents a T2-weighted structural image. This represents the three-dimensional voxel space where the T2-weighted structure image resides. Represents the first in the three-dimensional voxel space T2 signal intensity of individual units. This indicates the size of the T1-weighted structured image. This indicates the size of the T2-weighted structured image. These represent the voxel dimensions in three spatial directions (left and right, front and back, and up and down), typically in millimeters (e.g., ), These represent the voxel dimensions in three spatial directions (left and right, front and back, and up and down), typically in millimeters (e.g., This means that the image is clearer than T1 on the cross-section. Since T1 and T2 weighted structural images may have different spatial resolutions, voxel coordinates, and physical coordinates, their voxel indices cannot directly correspond to the same anatomical location. Therefore, this step uses spatial transformation to register the T2 weighted structural image into the three-dimensional voxel space of the T1 weighted structural image, so that the two modal images correspond to the same anatomical location under the same voxel grid, solving the problem of spatial inconsistency across modal images. It should be understood that although this embodiment uses T1 and T2 weighted structural images as an example, in other embodiments, the first and second weighted structural images can also be other medical image sequences that can reflect brain anatomical structures and tissue contrast information, as long as they can provide complementary tissue feature information.

[0075] S200, segment the first weighted structural image to obtain a brain parenchyma mask, and construct a smooth outer mask of the brain parenchyma based on the brain parenchyma mask.

[0076] Specifically, this step first isolates the brain parenchyma from T1-weighted structural images, excluding the scalp, skull, air background, and other non-brain tissues, to obtain a brain parenchyma mask. This brain parenchyma mask reflects the anatomical boundaries of the individual brain. Subsequently, a smooth outer mask of the brain parenchyma is constructed based on this mask. This smooth outer mask is not the final dura mater result, but rather a structural prior for the subsequent construction of the dura mater candidate search space. Its purpose is to eliminate local discontinuities on the boundaries of the brain parenchyma mask caused by sulci indentation, constructing a macroscopically continuous outer surface of the brain parenchyma, thereby preventing the subsequently constructed dura mater candidate search area from indenting inward along the sulci, making the search area closer to the macroscopically continuous inner surface of the dura mater. By constructing the smooth outer mask, an accurate anatomical reference benchmark is provided for the spatial localization of the dura mater.

[0077] S300: Construct a candidate search space for the dura mater based on the three-dimensional physical distance from the spatial voxel to the smooth outer mask of the brain parenchyma.

[0078] Specifically, this step calculates the three-dimensional physical distance (e.g., lateral Euclidean distance) from each spatial voxel to the smooth outer mask of the brain parenchyma in the T1 individual space, and constructs a candidate search space for the dura mater based on this distance and preset anatomical parameters (e.g., medial dura mater space and dura mater thickness). This search space is a shell region that limits the possible spatial range of the dura mater; it is not the final segmentation result, but rather a spatial prior region used to limit subsequent probability calculations. By constructing the search space, the candidate region for dura mater segmentation is restricted to a specific range near the outer surface of the brain parenchyma, thereby significantly reducing the probability of voxels far from the target region (e.g., scalp, lateral skull) being missegmented as the dura mater, improving segmentation efficiency and accuracy. It should be understood that the construction of the search space depends on the smooth outer mask obtained in the previous step, reflecting the design concept of "anatomical constraints."

[0079] S400, based on the image intensity distribution characteristics corresponding to the target user, adaptively determine the individualized segmentation threshold.

[0080] Specifically, MRI image intensity typically lacks a consistent absolute physical scale across target users and devices. Image brightness, contrast, and tissue signal vary significantly among different target users, scanning devices, or scanning parameters. Using a fixed threshold is insufficient to accommodate these individual differences. Therefore, this step does not employ a fixed threshold but instead adaptively determines an individualized segmentation threshold based on the image intensity distribution characteristics (e.g., statistical quantiles of signal intensity) within the brain parenchyma and dura mater candidate search space, using the image data of a single target user. This threshold includes a lower limit for excluding low-signal regions, a reference threshold for determining effective signal scores, and an upper limit for reducing the probability of high-signal regions (such as scalp fat and blood vessels) entering the candidate layer. By estimating the individualized threshold, the fixed threshold judgment is transformed into an individualized statistical judgment, enabling the system to adapt to the image differences among different target users and improving the robustness of segmentation.

[0081] S500, based on the signal intensity of the first weighted structural image, the signal intensity of the second weighted structural image, and the spatial distance prior corresponding to the dura mater candidate search space, and combined with the individualized segmentation threshold, a weighted calculation is performed to obtain the voxel assignment probability map.

[0082] Specifically, this step integrates multimodal image features and spatial distance priors to calculate the probability of each spatial voxel within the dura mater candidate search space. Specifically, the signal intensity features of the first weighted structural image are used to calculate a first signal score, which comprehensively considers low-signal exclusion (excluding air, background, or extremely low-signal areas) and high-signal penalty (reducing the probability of high-signal areas such as scalp fat and blood vessels entering the candidate layer); the signal intensity features of the second weighted structural image are used to calculate a second signal score, which utilizes the sensitivity of T2-weighted images to high-signal tissues such as cerebrospinal fluid to help distinguish the dura mater from adjacent cerebrospinal fluid spaces; and the spatial distance prior is used to calculate a distance prior score, which is based on the distance from the voxel to the smooth outer mask of the brain parenchyma, preferentially distributing candidate probabilities near the inner side of the search shell (i.e., regions closer to the brain parenchyma surface). Subsequently, the above three types of scores are weighted and calculated together with an individualized segmentation threshold to obtain a voxel attribution probability map. This probability map reflects the probability value of each candidate voxel belonging to the dura mater. By integrating multi-feature probability fusion, T1 features, T2 features, and spatial constraints, "multimodal decision-making" is achieved, making the determination of dura mater candidate regions more robust and reducing the risk of missegmentation caused by a single modality or a single feature.

[0083] S600, the dura mater region is segmented according to the voxel assignment probability map, and three-dimensional reconstruction is performed on the segmentation results to determine an individualized three-dimensional model of the dura mater.

[0084] Specifically, this step, based on a voxel assignment probability map, determines voxels within the candidate search space of the dura mater using a preset probability threshold. Voxels with probability values ​​higher than the threshold are marked as dura mater regions, thus determining the dura mater mask. Subsequently, 3D reconstruction is performed on the segmentation results. For example, the dura mater mask is converted into a 3D surface or mesh model using an isosurface extraction algorithm, determining an individualized 3D dura mater model. This 3D model can visually display the individual dura mater morphology and can be used for preoperative navigation, instrument path planning, electrode fit analysis, and craniotomy area design. It should be understood that although this embodiment describes segmentation using a probability threshold, in other embodiments, the segmentation results can be further post-processed (e.g., removing isolated small regions, smoothing boundaries, and repairing local breaks) to ensure surface continuity and improve the quality of the 3D model.

[0085] Through steps S100 to S600, this embodiment establishes a complete technical chain of "search space construction → individualized threshold estimation → multi-feature probability fusion". Among them, search space construction provides spatial constraints, individualized threshold estimation provides adaptive parameters, and probability fusion provides multimodal decision-making. The three work together to solve the problem of limited individualization and segmentation accuracy in dura mater modeling in the prior art, and realize high-precision, individualized 3D modeling of dura mater.

[0086] In some embodiments, such as Figure 1 and Figure 3 As shown, in step S200, the brain parenchyma mask is obtained by segmenting the first weighted structural image, and a smooth outer mask of the brain parenchyma is constructed based on the brain parenchyma mask, including the following sub-steps:

[0087] S201, based on the statistical quantile values ​​of the global voxel intensity of the first weighted structural image, outlier truncation and normalization are performed on the voxel intensity. Specifically, the formula for calculating the statistical quantile values ​​of the global voxel intensity of the first weighted structural image can be:

[0088]

[0089]

[0090] ,

[0091] In the formula, This represents the set of effective voxels constructed from the T1 image intensity. Indicates the first T1 signal intensity of individual voxels This represents the three-dimensional voxel space where the T1-weighted structure image resides. express The position of voxels in the middle, Indicates the effective mask. This indicates the calculation of percentiles. This represents the low quantile T1 intensity value. This indicates the lower quantile (e.g., the 5th percentile). This represents the high quantile T1 intensity value. This indicates the high percentile (e.g., the 95th percentile).

[0092] Subsequently, based on the calculated low quantile intensity values and high quantile intensity values After truncating the T1 intensity and normalizing the intensity, the calculation formula can be:

[0093]

[0094]

[0095] In the formula, Indicates the first The intensity of the T1 signal after truncation of the individual units. Indicates the first Normalized T1 signal intensity of voxels. This step aims to eliminate the influence of air background and invalid voxels, and reduce the impact of different scanning devices, different acquisition parameters, and abnormal intensity values ​​on subsequent segmentation. For example, the low quantile intensity value (e.g., the 5th percentile) and high quantile intensity value (e.g., the 95th percentile) of the effective region of the T1-weighted structural image can be calculated. Voxels with intensity values ​​below the low quantile are truncated to low quantile values, and voxels with intensity values ​​above the high quantile are truncated to high quantile values. Then, the truncated intensity values ​​are normalized to the interval between 0 and 1.

[0096] S202 extracts initial candidate masks based on normalized voxel intensities, removes outliers (e.g., small-scale isolated noise and fine artifacts) through 3D morphological opening operations, and performs connected component hole filling (e.g., detecting boundary edges in the mesh, identifying closed hole loops, triangulating each hole, and determining new triangular facets to fill the holes). Specifically, the initial candidate mask... The extraction calculation formula can be:

[0097] ,

[0098] In the formula, Indicates the first Initial candidate mask values ​​for individual pixels. Indicates the normalized intensity threshold. This represents the effective mask. Specifically, morphological opening operations consist of erosion operations (e.g., scanning an image using a structuring element to eliminate bright details or noise smaller than the structuring element, causing object boundaries to shrink) and dilation operations (e.g., dilating the eroded image to restore the original size of the object, but eliminating noise will not be restored). It can remove bright details or noise smaller than the structuring element while maintaining the overall shape of the object. The specific formula for morphological opening operations can be:

[0099]

[0100] ,

[0101] In the formula, Represents the candidate mask for brain parenchyma after the opening operation. This represents the initial candidate mask for brain parenchyma. This represents the structure element of the opening operation. This indicates a corrosion operation. This indicates an expansion operation. This represents a candidate mask for brain parenchyma after the holes have been filled. This refers to the hole-filling operation. For example, an opening operation can be performed using a three-dimensional spherical structuring element with a radius of 1-2 voxels, iterating 1-2 times to remove small outliers near the scalp and skull. The hole-filling operation is used to fill closed cavities inside the candidate mask, ensuring the integrity of the brain parenchyma region.

[0102] S203, extract the connected component with the largest number of three-dimensional voxels as the basic mask. First, reduce boundary residue on the outer contour of the basic mask, then restore the boundary to determine the brain parenchyma mask. Specifically, the calculation formula for the brain parenchyma mask can be:

[0103] ,

[0104] In the formula, This represents the brain parenchyma mask value of the i-th voxel. This represents the largest connected component. This step reduces residual skull, meninges, or other boundaries through a slight erosion operation, followed by a slight dilation operation to restore the brain parenchyma boundaries, thus obtaining an accurate brain parenchyma mask. For example, the erosion and dilation operations can be performed once each using a three-dimensional spherical structuring element with a radius of 1 voxel.

[0105] S204, perform morphological closing operations and hole filling on the brain parenchyma mask to eliminate sulci depressions and obtain a smooth outer mask of the brain parenchyma. The morphological closing operation consists of dilation and erosion operations, which can fill small holes inside the object and connect adjacent objects while maintaining the overall shape of the object. Specifically, the specific formula for the morphological closing operation can be:

[0106]

[0107] ,

[0108] In the formula, This represents the mask after the closing operation. Represents a mask for brain parenchyma. This represents a structure element that performs a closing operation. This indicates a corrosion operation. This indicates an expansion operation. This represents a smooth outer mask of brain parenchyma after the holes have been filled. This step represents the hole-filling operation. The core objective of this step is to eliminate local discontinuities at the boundaries of the brain parenchyma mask caused by sulci depressions. Since the brain parenchyma mask is directly segmented from T1-weighted structural images, its boundaries often depression inward along the sulci. If the dura mater candidate search space is directly constructed based on this, the search area would also depression inward along the sulci, which does not conform to the macroscopically continuous anatomical morphology of the dura mater. Therefore, morphological closing operations can fill in the sulci depression areas, constructing a macroscopically continuous outer surface of the brain parenchyma. For example, a three-dimensional spherical structural element with a radius of 3-5 voxels can be used for closing operations, iterating 1-2 times to effectively eliminate sulci depressions with widths smaller than the structural element's radius. Subsequently, a three-dimensional hole-filling operation is performed on the result of the closing operation to further fill any possible internal voids, ultimately obtaining a smooth outer mask of the brain parenchyma. This smooth outer mask provides an accurate structural prior for the subsequent construction of the dura mater candidate search space, ensuring that the search area is closer to the macroscopically continuous inner surface of the dura mater.

[0109] In some embodiments, such as Figure 1 and Figure 4 As shown, in step S300, based on the three-dimensional physical distance from the spatial voxel to the smooth outer mask of the brain parenchyma, a candidate search space for the dura mater is constructed, including the following sub-steps:

[0110] S301, calculate the outer distance from each spatial voxel to the smooth outer mask of the brain parenchyma, where the outer distance is the minimum physical distance from the voxel to the smooth outer mask of the brain parenchyma, and for voxels belonging to the smooth outer mask of the brain parenchyma, the outer distance is zero. This step calculates the outer Euclidean distance from each spatial voxel to the smooth outer mask of the brain parenchyma in the T1 individual space. Specifically, the... Lateral distance from voxel to smooth outer mask of brain parenchyma The calculation formula can be expressed as:

[0111] ,

[0112] ,

[0113] In the formula, This represents a smooth outer mask of the brain parenchyma. Indicates the spatial coordinates of the current voxel. This represents the spatial coordinates of a voxel belonging to the smooth outer mask of brain parenchyma. express and The Euclidean distance between two points. For voxels belonging to the smooth outer mask of brain parenchyma, the outer distance is defined as 0. It reflects the physical distance of a voxel from the outer surface of the brain parenchyma and is a key geometric parameter for constructing the candidate search space for the dura mater.

[0114] S302, based on the lateral distance and the preset medial dura mater gap and dura mater thickness, a candidate dura mater search space is constructed, wherein the medial dura mater gap represents the distance between the medial surface of the dura mater and the surface of the brain parenchyma, and the dura mater thickness represents the thickness parameter of the dura mater. This step constructs the candidate dura mater search space based on the lateral distance and anatomical parameters. Specifically, the candidate dura mater search space... It can be defined as a set of voxels that satisfies the following conditions:

[0115] ,

[0116] In the formula, Indicates the medial dura mater space. Indicates the thickness of the dura mater. Indicates the voxels that meet the criteria. Medial dura mater This reflects the anatomical space between the medial surface of the dura mater and the surface of the brain parenchyma. This space typically corresponds to areas such as the cerebrospinal fluid space and the subarachnoid space, and its value generally ranges from 1 to 3 millimeters, for example, it can be set to 2 millimeters. Dura mater thickness. The thickness parameter reflects the dura mater tissue, and its value generally ranges from 0.5 to 2 mm, for example, it can be set to 1 mm. It should be understood that the specific value of the above parameter can be adjusted according to individual anatomical differences or clinical needs, and this embodiment does not strictly limit it. By constructing a candidate search space for the dura mater, the candidate region for dura mater segmentation is restricted to a specific shell layer near the outer surface of the brain parenchyma, thereby significantly reducing the probability of voxels far from the target region (such as the scalp or lateral skull) being missegmented as the dura mater, improving segmentation efficiency and accuracy. This search space embodies the design concept of "anatomical constraints," providing a spatial prior basis for subsequent probabilistic fusion calculations.

[0117] This embodiment details the construction process of the smooth outer mask of the brain parenchyma and the construction process of the candidate search space for the dura mater. Morphological closure and hole-filling operations effectively eliminate sulci depressions, constructing a macroscopically continuous outer surface of the brain parenchyma; lateral distance calculation and anatomical parameter definition accurately limit the possible spatial range of the dura mater. These two processes work together to provide accurate spatial constraints for the 3D modeling of the dura mater, further improving the individualization of the modeling and the segmentation accuracy.

[0118] Specifically, MRI image intensity typically lacks a consistent absolute physical scale across target users and devices. Image brightness, contrast, and tissue signal vary significantly among different target users, scanning devices, or scanning parameters. Using a fixed threshold is insufficient to accommodate these individual differences, potentially leading to inaccurate segmentation results. Therefore, this embodiment does not employ a fixed threshold but instead utilizes the image data of a single target user.

[0119] In some embodiments, such as Figure 1 and Figure 5 As shown, in step S400, an individualized segmentation threshold is adaptively determined based on the image intensity distribution characteristics corresponding to the target user. This process specifically includes the following sub-steps:

[0120] S401, extract the percentile of signal intensity of the first weighted structural image within the brain parenchyma mask to determine a first signal reference threshold. This step aims to extract a reference value from the brain parenchyma region of the T1-weighted structural image that reflects the signal intensity of normal brain tissue. Specifically, the statistical percentile of the T1 signal intensity of all voxels within the brain parenchyma mask, such as the 50th percentile (i.e., the median), can be calculated as the first signal reference threshold. This threshold reflects the typical signal level of the target user's brain parenchyma region and is used to subsequently determine the effective T1 signal score. It should be understood that although this embodiment uses the 50th percentile as an example, in other embodiments, other percentiles that can reflect the typical signal intensity of brain parenchyma, such as the 40th percentile or the 60th percentile, can also be used, as long as they can serve as a reference benchmark for the effective signal.

[0121] S402, extract the percentage of signal intensity of the first weighted structural image within the brain parenchyma mask and the dura mater candidate search space to determine a first lower signal threshold and a first upper signal threshold. This step aims to determine the threshold boundaries used to exclude low-signal regions and penalize high-signal regions. Specifically, the first lower signal threshold is used to exclude air, background, or extremely low-signal regions, and can be determined by extracting the low quantile value of the T1 signal intensity within the brain parenchyma mask and the dura mater candidate search space, for example, the 5th percentile. When the T1 signal intensity of a voxel is below this threshold, its probability of belonging to the dura mater will be significantly reduced. The first upper signal threshold is used to reduce the probability of scalp fat, high-signal blood vessels, or locally abnormally high-signal regions entering the candidate layer, and can be determined by extracting the high quantile value of the T1 signal intensity within the brain parenchyma mask and the dura mater candidate search space, for example, the 95th percentile. When the T1 signal intensity of a voxel is above this threshold, its probability of belonging to the dura mater will be penalized and reduced. It should be understood that the specific values ​​of the above percentile values ​​can be adjusted according to image quality or segmentation requirements. For example, the lower percentile value can be set to any value between the 1st and 10th percentiles, and the higher percentile value can be set to any value between the 90th and 99th percentiles.

[0122] S403, extract the percentage of signal intensity within the dura mater candidate search space to determine the second upper limit threshold. This step aims to extract reference values ​​reflecting the T2 signal characteristics of normal brain tissue from the dura mater candidate search space of the registered T2-weighted structural image. Specifically, the statistical quantile of the T2 signal intensity of all voxels within the brain parenchyma mask, such as the 50th percentile, can be calculated as the second brain parenchyma mask signal quantile value. This second upper limit threshold is used for subsequent calculation of the T2 signal score to help distinguish the dura mater from the adjacent cerebrospinal fluid space.

[0123] In some embodiments, the first signal lower threshold includes the maximum value of a preset low quantile among the signal intensity percentages of the first weighted structural images within the brain parenchyma mask and the dura mater candidate search space; the first signal reference threshold includes the signal intensity percentage of the first weighted structural images within the brain parenchyma mask; and the first signal upper threshold includes the maximum value of a preset high quantile among the signal intensity percentages of the first weighted structural images within the brain parenchyma mask and the dura mater candidate search space. Specifically, this embodiment defines the individualized segmentation threshold using the following formula:

[0124]

[0125]

[0126]

[0127] In the formula, The first term representing the T1 signal intensity in brain parenchyma regions percentile, This indicates the calculation of percentiles. Indicates the first T1 signal intensity of individual voxels Represents a mask for brain parenchyma. The first term represents the T1 signal intensity within the candidate search space of the dura mater. percentile, For the dura mater candidate search space, The first term representing the T2 signal intensity in brain parenchyma regions percentile, No. The T2 signal intensity after registration of individual voxels. The first term represents the T2 signal intensity within the candidate search space of the dura mater. percentile.

[0128] The formula for calculating the lower limit threshold of the first signal can be:

[0129] ,

[0130] In the formula, This represents the lower threshold of the first signal. This represents the preset low quantile (e.g., the fifth percentile) of the signal intensity percentage of the first weighted structural image within the dura mater candidate search space. This represents a predetermined low quantile (e.g., the 5th percentile) in the percentage of signal intensity in the first weighted structural image within the brain parenchyma mask. When the voxel T1 signal intensity is below this threshold, the score approaches 0, thus excluding air, background, or extremely low signal regions.

[0131] The formula for calculating the first signal reference threshold can be:

[0132]

[0133] In the formula, Indicates the first signal reference threshold. This represents the percentage of signal intensity (e.g., the 50th percentile) of the first weighted structural image within the brain parenchyma mask. This threshold is used to construct a transition range for the lower limit score of the T1 signal, when the voxel T1 signal intensity is between... and During this period, the score smoothly transitions from 0 to 1.

[0134] The formula for calculating the upper limit threshold of the first signal can be:

[0135]

[0136] In the formula, Indicates the upper limit threshold of the first signal. This represents the preset high quantile (e.g., the 95th percentile) of the signal intensity percentage of the first weighted structural image within the dura mater candidate search space. This represents a preset high quantile (e.g., the 95th percentile) in the percentage of signal intensity of the first weighted structural image within the brain parenchyma mask. This threshold is used to construct the upper limit score for T1 high signal intensity. When the T1 signal intensity of a voxel is higher than this threshold, the score approaches 0, thereby reducing the probability of high signal regions such as scalp fat and blood vessels entering the candidate layer.

[0137] The formula for calculating the upper limit threshold of the second signal can be:

[0138]

[0139] In the formula, Indicates the upper limit threshold of the second signal. This represents the percentage of signal intensity within the candidate search space for the dura mater (e.g., the 95th percentile).

[0140] Through steps S401 to S403 described above, this embodiment adaptively determines three types of individualized segmentation thresholds based on the image intensity distribution corresponding to a single target user. The lower threshold is used to exclude low-signal regions, the reference threshold is used to determine the effective signal score, and the upper threshold is used to reduce the probability of high-signal regions entering the candidate layer. These three thresholds work together to transform fixed threshold judgment into individualized statistical judgment, enabling the system to adapt to image differences among different target users and improving the robustness and accuracy of segmentation.

[0141] In some embodiments, such as Figure 1 and Figure 6 As shown in S500, the specific method for obtaining the voxel assignment probability map by weighting the signal intensity of the first weighted structural image, the signal intensity of the second weighted structural image, and the spatial distance prior corresponding to the dura mater candidate search space, combined with the individualized segmentation threshold, includes the following sub-steps:

[0142] S501, based on the signal intensity of the first weighted structural image, a first signal score is calculated. The first signal score includes a lower limit score and an upper limit score. This step aims to assess the likelihood of a voxel belonging to the dura mater from the perspective of T1-weighted structural images. Since the dura mater exhibits moderate signal intensity on T1 images, it should not be pure background, air, or extremely low signal areas (such as cerebrospinal fluid spaces), nor should it be extremely high signal areas (such as scalp fat or blood vessels). Therefore, the first signal score is designed as a dual screening mechanism: on the one hand, areas with excessively low signal intensity are excluded through the lower limit score, and on the other hand, areas with excessively high signal intensity are suppressed through the upper limit score. The final first signal score is derived from the combination of these two scores; only voxels with signal intensity within a reasonable range can obtain a higher score.

[0143] S502, calculate the second signal score based on the signal intensity of the second weighted structural image and the second signal upper limit threshold. This step aims to assist in the evaluation from the perspective of T2-weighted structural images. Since the dura mater and the surrounding area of ​​the inner table of the skull typically show lower signal intensity than cerebrospinal fluid on T2 images, appearing as intermediate or low signal, while areas with higher water content such as cerebrospinal fluid and inflammation appear as high signal, the second signal score is mainly used to distinguish the dura mater from the adjacent high-signal cerebrospinal fluid space. When the T2 signal intensity of the voxel is significantly higher than the second signal upper limit threshold, its score will decrease, thereby reducing the probability of cerebrospinal fluid areas being misidentified as the dura mater.

[0144] S503, Calculate the distance prior score based on spatial distance priors. This step aims to introduce anatomical spatial constraints. Based on the anatomical location characteristics of the dura mater, it is more likely to be located in a proximal region to the outer surface of the brain parenchyma, rather than on the outer edge of the brain parenchyma (such as the scalp or the outer side of the skull). Therefore, the distance prior score is calculated based on the three-dimensional physical distance from the voxel to the smooth outer mask of the brain parenchyma; the closer the voxel, the higher the score, and the farther the voxel, the lower the score. This score preferentially distributes candidate probabilities near the inner side of the search shell, consistent with the anatomical fact that the dura mater is closely attached to the surface of the brain parenchyma.

[0145] S504, the first signal score, the second signal score, and the distance prior score are weighted and calculated to obtain the voxel assignment probability map. This step involves weighting and summing the three types of scores according to preset weight parameters to obtain the comprehensive probability value of each spatial voxel belonging to the dura mater. This probability map reflects the comprehensive judgment result of multimodal features and spatial constraints, providing a robust decision basis for subsequent segmentation.

[0146] In some embodiments, the first signal score, the second signal score, and the distance prior score are calculated in the following ways:

[0147] First, this embodiment defines a smooth step function (Sigmoid function) to implement the soft decision mechanism. The function is defined as follows:

[0148] ,

[0149] In the formula, For input variables, This is a scaling parameter used to control the smoothness of the function's transitions. The characteristic of this function is that when the input variable... When the value is much greater than 0, the function output approaches 1; when When the value is much less than 0, the function output approaches 0; in Near the threshold of 0, the function output transitions smoothly. Compared to traditional hard threshold judgment (such as directly determining the signal strength as 1 if it is higher than a certain threshold, and 0 otherwise), the Sigmoid function avoids the jagged or abrupt segmentation boundary problems caused by discontinuities in the threshold boundary, making the segmentation result smoother and more natural. It should be understood that although this embodiment uses the Sigmoid function as the preferred implementation of the smooth step function, other functions with similar smooth transition characteristics, such as the hyperbolic tangent function or piecewise linear function, can also be used in other embodiments, as long as a soft decision mechanism can be achieved.

[0150] The calculation process for the first signal score specifically includes:

[0151] Based on the relationship between the signal strength of the first weighted structural image, the first signal lower limit threshold, and the first signal reference threshold, the first signal lower limit score is calculated using a smooth step function. Specifically, the first signal lower limit score... The calculation formula can be expressed as:

[0152] ,

[0153] In the formula, Indicates the first T1 signal intensity of voxels This represents the lower threshold of the first signal. Indicates the first signal reference threshold. This represents the parameter controlling the transition width. The technical implication of this score is: when the T1 signal intensity of the voxel is significantly lower than the lower threshold... When the signal strength is between (e.g., corresponding to air, background, or extremely low signal areas), the score approaches 0, thus excluding these areas; when the signal strength is between and When the signal strength is between 0 and 1, the score smoothly transitions from 0 to 1; when the signal strength is higher than 1... When the score approaches 1, it indicates that the voxel has passed the low-signal exclusion screening. This is achieved by introducing a first signal reference threshold. As the upper limit of the transition range, it avoids the hard decision boundary problem caused by a single threshold.

[0154] Based on the relationship between the signal intensity of the first weighted structural image and the first signal upper limit threshold, the high signal upper limit score is calculated using the smooth step function. Specifically, the high signal upper limit score... The calculation formula can be expressed as:

[0155] ,

[0156] In the formula, Indicates the upper limit threshold of the first signal. This parameter represents the width of the penalty transition. The technical implication of this score is: when the T1 signal intensity of the voxel is significantly higher than the upper threshold... When the signal intensity is low (e.g., corresponding to high signal intensity in scalp fat, blood vessels, or local abnormally high signal areas), the score approaches 0, thus reducing the probability of these areas entering the candidate layer; when the signal intensity is lower than... When the score approaches 1, it indicates that the voxel has not been penalized for high signal intensity. This high signal penalty mechanism effectively suppresses the interference of high signal artifacts on the segmentation results.

[0157] The first signal score is obtained by multiplying the lower limit score of the first signal by the upper limit score of the high signal. Specifically, the formula for calculating the first signal score S_1(x) can be expressed as:

[0158] ,

[0159] In the formula, `clip` represents the truncation function, restricting the result to the range of 0 to 1. This step uses multiplication instead of addition to combine the two scores. The technical principle behind this is that low signal exclusion and high signal penalty are two independent screening conditions. A voxel must simultaneously meet both conditions—"signal intensity not lower than the lower limit" and "signal intensity not higher than the upper threshold"—to be considered a reasonable candidate voxel for the dura mater. If addition were used, there might be a situation where one aspect scores extremely low but another aspect scores extremely high, resulting in a still high overall score, making it impossible to effectively exclude unqualified voxels. However, with multiplication, as long as either aspect's score approaches 0, the overall score also approaches 0, thus achieving a strict constraint of double screening. This design reflects the deeper consideration of "why not something else," ensuring the effectiveness of the first signal score screening.

[0160] The calculation process for the second signal score specifically includes:

[0161] Based on the relationship between the signal intensity of the second weighted structural image and the second signal upper threshold, the second signal score is calculated using a smooth step function. Specifically, the second signal score... The calculation formula can be expressed as:

[0162]

[0163] In the formula, This indicates the T2 signal strength after registration. Indicates the upper limit threshold of the second signal. This represents the parameter controlling the transition width. The technical implication of this score is: when the T2 signal intensity of the voxel is significantly higher than the second signal upper limit threshold (i.e., higher than...), it indicates that... When the T2 signal intensity is below 0, the score approaches 0, thus excluding areas with high signal intensity such as cerebrospinal fluid; when the T2 signal intensity is below 0... When the score approaches 1, it indicates that the voxel matches the low signal characteristics of the dura mater on a T2 image. It should be understood that, although this embodiment uses... As a reference benchmark, other quantile values ​​that can reflect the T2 signal characteristics of brain parenchyma can also be used in other embodiments, as long as they can effectively distinguish between the dura mater and high-signal cerebrospinal fluid.

[0164] The calculation process for the distance prior score specifically includes:

[0165] The distance prior score is calculated using an inverse linear decay function based on the minimum physical distance from the voxel to the smooth outer mask of the brain parenchyma. Specifically, the distance prior score... The calculation formula can be expressed as:

[0166] ,

[0167] in Voxel representation The distance to the outer side of the smooth outer mask of the brain parenchyma. Indicates the medial dura mater space. This indicates the thickness of the dura mater. The technical meaning of this score is: when the lateral distance of the voxel... equal When the lateral distance is near the medial surface of the dura mater, the score is highest at 1; as the lateral distance increases, the score decreases linearly; when the lateral distance reaches... + When the distance is close to the lateral surface of the dura mater, the score drops to 0. This reverse linear decay design preferentially distributes candidate probabilities near the inner side of the search shell, consistent with the anatomical fact that the dura mater is closely attached to the surface of the brain parenchyma, thereby reducing the probability of voxels far from the surface of the brain parenchyma (such as the lateral skull or scalp) being misidentified as the dura mater. It should be understood that although this embodiment uses a linear decay function, in other embodiments, a non-linear decay function (such as exponential decay) can also be used, as long as the effect of higher scores for closer objects can be achieved.

[0168] In some embodiments, the weighted probability fusion calculation based on the individualized segmentation threshold is implemented in the following manner:

[0169] The first signal score, the second signal score, and the distance prior score are multiplied by their respective first, second, and third weights, and then summed. Specifically, the voxel assignment probability... The calculation formula can be expressed as:

[0170] ,

[0171] in Indicates the first weight. Indicates the second weight. This represents the third weight. This step flexibly adjusts the contribution of each feature using weight parameters. In a preferred embodiment, the values ​​of the weight parameters are determined based on the following: First weight... Set to 0.4, second weight Set to 0.3, third weight The weight is set to 0.3. The technical principle behind this value is as follows: T1-weighted structural images provide the most important anatomical structural information, and the first signal score includes a dual screening mechanism of low signal exclusion and high signal penalty, making its contribution to dura mater segmentation most critical; therefore, it is assigned the highest weight of 0.4. T2-weighted structural images are mainly used to assist in distinguishing high-signal areas such as cerebrospinal fluid; their contribution is relatively minor, so a weight of 0.3 is assigned. Distance prior scores provide spatial constraints; although important, their role is mainly reflected in limiting the search space range, and their contribution to probabilistic fusion is comparable to that of T2 features, so a weight of 0.3 is assigned. It should be understood that the specific values ​​of the above weight parameters can be adjusted according to image quality, individual differences, or clinical needs. For example, when the T2 image quality is poor, the weight can be reduced. And correspondingly improve or This embodiment does not impose strict limitations on it.

[0172] Furthermore, for the candidate search space located in the dura mater The probability of voxel assignment to other voxels Setting it directly to 0 ensures that the segmentation results are strictly limited to the preset search space.

[0173] Finally, based on the voxel assignment probability map, this embodiment introduces a preset probability threshold. The formula for segmentation determination can be:

[0174] ,

[0175] In the formula, Indicates the dura mater mask value. This represents the probability of voxel assignment. This represents a preset probability threshold. In a preferred embodiment, the probability threshold... Set to 0.5. The technical principle behind this value is as follows: A value of 0.5 corresponds to the decision logic of "if the probability is greater than 50%, it is determined to be dura mater," which is a neutral threshold balancing sensitivity and specificity. When Setting the value too high may result in the dura mater region being missed during segmentation; when Setting the value too low may result in the missegmentation of non-dura mater areas. A probability threshold of 0.5 ensures both accuracy and completeness in segmentation results. It should be understood that this probability threshold... The specific value can also be adjusted according to actual needs. For example, when a more conservative segmentation result is required, it can be... Increase to 0.6 or 0.7; when more complete coverage is needed, it can be... The value is reduced to 0.4, but this embodiment does not impose a strict limit on it.

[0176] Through S501 to S504 above, this embodiment details the specific algorithm implementation of multi-feature probability fusion. Specifically, the Sigmoid smooth step function implements a soft decision mechanism, avoiding the boundary discontinuity problem caused by hard thresholding; the multiplicative synthesis mechanism of the first signal score implements strict constraints for dual screening; the differentiated setting of weight parameters reflects the reasonable allocation of the contribution of each feature; and the balanced value of the probability threshold takes into account both the accuracy and completeness of segmentation. Overall, this embodiment, through the design concept of "multimodal decision-making," integrates T1 features, T2 features, and spatial constraints, making the determination of dura mater candidate regions more robust, effectively reducing the risk of missegmentation caused by a single modality or single feature, and further improving the individualization of modeling and segmentation accuracy.

[0177] In some embodiments, such as Figure 1 and Figure 7 As shown, in step S600, the dura mater region is segmented according to the voxel assignment probability map, and three-dimensional reconstruction is performed on the segmentation results to determine an individualized three-dimensional model of the dura mater, including the following sub-steps:

[0178] S601, voxels with probability values ​​higher than a preset segmentation threshold in the voxel assignment probability map are marked as target voxels, and binarized to determine the dura mater coarse mask. This step performs preliminary segmentation determination based on the voxel assignment probability map. Specifically, the preset segmentation threshold... It can be set to 0.5, which is the probability of voxel assignment. A value greater than or equal to 0.5 is marked as a target voxel (value 1); otherwise, it is marked as a non-target voxel (value 0). The probability map is converted into a coarse dura mater mask through binarization. It should be understood that although this embodiment uses… Taking 0.5 as an example, but in other embodiments, the threshold... The specific value can be adjusted according to the segmentation requirements. For example, when a more conservative segmentation result is needed, the value can be adjusted accordingly. Increase to 0.6 or 0.7; when more complete coverage is needed, it can be... The value is reduced to 0.4, but this embodiment does not impose a strict limitation on it. This coarse mask of the dura mater initially reflects the spatial distribution of the dura mater, but due to the possibility of local noise or boundary discontinuities in the probabilistic fusion calculation, the coarse mask may contain isolated small regions or local breaks, requiring further post-processing.

[0179] S602, a spatial topological connectivity check is performed on the coarse dura mater mask to filter out isolated misjudged regions with a volume size smaller than a preset tolerance, and to repair local broken boundaries to determine a continuous fine dura mater mask. This step aims to remove isolated misjudged regions in the coarse dura mater mask and repair local broken boundaries to ensure the spatial continuity of the dura mater mask. Specifically, the spatial topological connectivity check can be implemented using a three-dimensional connected component labeling algorithm. This algorithm can identify all connected components in the coarse dura mater mask and calculate the number of voxels or physical volume of each connected component. The preset tolerance can be set to isolated regions with a volume size smaller than 100 cubic millimeters. For example, when the physical volume corresponding to the number of voxels of a certain connected component is less than 100 cubic millimeters, it is determined to be an isolated misjudged region (possibly caused by noise or local misjudgment) and filtered out. It should be understood that the specific value of the preset tolerance can be adjusted according to image resolution or clinical needs. For example, in high-resolution images, the tolerance can be set smaller (e.g., 50 cubic millimeters) to retain more details; in low-resolution images, the tolerance can be set larger (e.g., 200 cubic millimeters) to more thoroughly remove noise. This embodiment does not strictly limit this. After filtering out isolated misjudged areas, this step further repairs local fracture boundaries. The repair method can employ morphological closing operations, such as using a three-dimensional spherical structural element with a radius of 1-2 voxels to perform closing operations on the coarse dura mater mask to fill local fractures or holes with a width smaller than the radius of the structural element, while maintaining the overall shape of the dura mater. Through topological inspection and boundary repair, a continuous fine dura mater mask is determined. This mask has better spatial continuity and topological integrity, providing a reliable data foundation for subsequent three-dimensional reconstruction.

[0180] S603, extract surface isosurfaces from the fine dura mater mask and perform mesh rendering to determine the individualized dura mater 3D model. This step converts the fine dura mater mask into a visualized 3D surface model. Specifically, the surface isosurface extraction can use the MarchingCubes algorithm, a classic 3D reconstruction algorithm. This algorithm determines the topology of the isosurfaces and the triangular facet mesh by traversing each cube cell in the voxel mesh and based on the mask value (0 or 1) of the cell vertices. For the fine dura mater mask, the isosurface threshold can be set to 0.5, i.e., extracting the boundary surface between voxels with a mask value of 1 and voxels with a mask value of 0. The MarchingCubes algorithm can determine a smooth and continuous 3D surface mesh, accurately reflecting the anatomical morphology of the dura mater. Subsequently, mesh rendering is performed on the determined triangular facet mesh, for example, using a 3D visualization library such as OpenGL or VTK, to determine the individualized dura mater 3D model. This 3D model can intuitively display the morphology, location, and spatial relationships of an individual's dura mater, and can be used for preoperative navigation, instrument path planning, electrode fit analysis, and craniotomy area design. It should be understood that although this embodiment uses the MarchingCubes algorithm as an example, other isosurface extraction algorithms, such as the Cuberille algorithm or the SurfaceNets algorithm, can also be used in other embodiments, as long as a continuous 3D surface mesh can be determined.

[0181] Through steps S601 to S603 described above, this embodiment details the post-processing steps of dural segmentation and 3D reconstruction. Specifically, binarization thresholding converts the probability map to a mask, topological connectivity checks filter out isolated misclassified regions, morphological closing operations repair local broken boundaries, and the MarchingCubes algorithm determines a continuous 3D surface mesh. Overall, this embodiment ensures the spatial continuity and topological integrity of the dural mask through post-processing operations, improves the quality of the 3D model, makes it more suitable for clinical applications, and provides a reliable data foundation for preoperative planning.

[0182] In some embodiments, such as Figure 1 and Figure 2 As shown, in step S100, registering the second weighted structural image to the three-dimensional voxel space of the first weighted structural image includes the following sub-steps:

[0183] S101, Extract the spatial geometric relationship between the first weighted structured image and the second weighted structured image. This step aims to obtain the initial spatial mapping information between the two modal images. Specifically, the mapping relationship between voxel indices and physical coordinates can be extracted by reading the spatial affine matrix carried in the image file header information. The mapping formula can be:

[0184] ,

[0185] In the formula, This represents the spatial affine matrix carried in the T1-weighted structured image scan file, which describes the mapping relationship between image voxel indices and real physical coordinates. Indicates voxel index, express This is converted into their corresponding physical spatial coordinates. This allows for a preliminary assessment of the differences between the two modal images in terms of spatial location, orientation, and scale, providing an initial estimate for subsequent registration parameter optimization. It should be understood that although this embodiment extracts the initial relationship using an affine matrix, in other embodiments, the initial correspondence can also be extracted by manually marking anatomical feature points or using an automatic feature detection algorithm, as long as it provides an initial reference for registration.

[0186] S102, with the optimization objective of maximizing the global pixel mutual information between the two modal structural images, iteratively optimizes the rotation and translation spatial parameters within six degrees of freedom. This step is the core of registration, aiming to find the optimal spatial transformation parameters (i.e., rotation and translation spatial parameters) to achieve the best spatial alignment between the T2-weighted structural image and the T1-weighted structural image.

[0187] This embodiment uses mutual information as a similarity metric. Mutual information is a concept in information theory used to measure the statistical correlation between two random variables. In medical image registration, mutual information can effectively measure the degree of similarity between images of different modalities. Even if the gray-level distributions of two images differ greatly (such as gray-level inversion between T1 and T2 images), mutual information can still reflect the spatial correspondence of anatomical structures. Maximizing mutual information means that after spatial alignment, the entropy of the joint gray-level distribution of the two images is minimized, that is, there is the strongest statistical dependency between the gray-level values ​​of the two images, thus achieving optimal registration.

[0188] This embodiment employs a six-degree-of-freedom rigid transformation model, including three rotation parameters (corresponding to rotation angles along the X, Y, and Z axes, respectively) and three translation parameters (corresponding to translation distances along the X, Y, and Z axes, respectively). The rigid transformation preserves the relative position and shape within the image, changing only the overall spatial position and orientation. The technical principle behind this design is that both T1 and T2 weighted structural images originate from the same target user's head. During scanning, although there may be slight differences in head position, the brain tissue itself does not deform. Therefore, accurate registration can be achieved using a rigid transformation, eliminating the need for complex calculations and potential deformation risks associated with non-rigid transformations (such as affine or elastic transformations). It should be understood that while this embodiment uses a six-degree-of-freedom rigid transformation, in other embodiments, if there are scale differences in the images, it can be extended to a seven-degree-of-freedom or nine-degree-of-freedom affine transformation, as long as spatial alignment can be achieved.

[0189] Regarding the optimization algorithm, this embodiment can also employ the Powell algorithm for iterative optimization. The Powell algorithm is a direct search optimization algorithm that does not require derivative calculation. It approximates the optimal solution by performing a linear search along a series of directions in the parameter space. Specifically, the algorithm first initializes a set of search directions (usually coordinate axis directions), then searches for the step size that maximizes mutual information in each direction, updating the transformation parameters. Subsequently, it adjusts the direction set based on the search results, determines new conjugate directions, and continues iteratively until the convergence condition is met (e.g., the change in mutual information is less than a preset threshold or the number of iterations reaches an upper limit). The advantage of the Powell algorithm is that it does not require calculating the gradient of mutual information, avoiding the difficulty of gradient calculation caused by the complex gray-scale distribution of multimodal images. It also has a relatively fast convergence speed, making it suitable for parameter optimization in rigid registration. It should be understood that although this embodiment uses the Powell algorithm as an example, other optimization algorithms, such as gradient descent, simulated annealing, or genetic algorithms, can be used in other embodiments, as long as they can maximize mutual information.

[0190] S103, the second weighted structured image is resampled using the rotational and translational spatial parameters obtained through optimization, thus completing cross-modal rigid registration. Specifically, the formula for rigid registration can be:

[0191]

[0192] In the formula, Indicates a rigid transformation. This represents the input physical coordinates. Represents spatial transformation parameters, This represents the three-dimensional translation. This step aims to transform the T2-weighted structured image into the three-dimensional voxel space of the T1-weighted structured image based on the optimized transformation parameters.

[0193] In some embodiments, the resampling process includes two key steps: spatial coordinate transformation and grayscale interpolation. Spatial coordinate transformation constructs a rigid transformation matrix based on the rotation and translation parameters obtained through optimization, mapping the coordinates of each spatial voxel in space T1 back to the corresponding coordinates in the original space T2. Since the transformed coordinates typically do not fall at integer positions in the original T2 voxel grid, grayscale interpolation is needed to calculate the grayscale value at that position.

[0194] In some embodiments, trilinear interpolation is used for grayscale interpolation. Trilinear interpolation is a commonly used interpolation method in three-dimensional space. It obtains the grayscale value of the interpolation point by calculating the weighted average of the grayscale values ​​of the eight neighboring voxels around the target coordinates. The weights are determined based on the distance between the target coordinates and each neighboring voxel; the closer the distance, the greater the weight. Trilinear interpolation can determine a smooth grayscale transition, avoiding grayscale jumps or block effects caused by nearest-neighbor interpolation. It also has high computational efficiency, making it suitable for the resampling requirements of medical image registration. It should be understood that although this embodiment uses trilinear interpolation, other interpolation methods, such as nearest-neighbor interpolation (suitable for label images) or cubic interpolation (suitable for higher precision requirements), can also be used in other embodiments, as long as the transformed grayscale value can be accurately calculated.

[0195] Through S101 to S103 above, this embodiment details the specific algorithm implementation for cross-modal rigid registration. The mutual information maximization objective function effectively measures the similarity between multimodal images, overcoming the registration difficulties caused by differences in grayscale distribution; the six-degree-of-freedom rigid transformation model accurately reflects the spatial relationships of the same user's head images, avoiding the deformation risks associated with non-rigid transformations; the Powell algorithm efficiently optimizes parameters without gradient calculation; and trilinear interpolation ensures the smoothness of grayscale after resampling. Overall, this embodiment, through the rigid registration algorithm, accurately registers the T2-weighted structural image to the three-dimensional voxel space of the T1-weighted structural image, solving the problem of spatial inconsistency between different modalities. This provides a unified spatial benchmark for subsequent brain parenchyma segmentation, search space construction, and multi-feature probability fusion, ensuring the spatial correspondence of the processing results in each step, thereby improving the accuracy and reliability of dura mater 3D modeling.

[0196] In some embodiments, such as Figure 1 , Figure 8 and Figure 9 As shown, the method further includes a step of determining candidate implantation regions for subdural electrodes based on task-oriented functional magnetic resonance imaging, specifically including the following sub-steps:

[0197] S700, Determine the functional magnetic resonance imaging (fMRI) experimental paradigm based on decoding needs. This paradigm includes a task stimulus sequence that matches the decoding needs. This step aims to design a suitable fMRI experimental paradigm for the target user based on actual decoding needs or clinical judgment. Specifically, the fMRI experimental paradigm consists of multiple task cycles. Each task cycle includes a fixation phase, a task cueing phase, a task execution phase, and a rest phase, where the content of the task execution phase corresponds to the decoding needs. For example, when the decoding need is for hand movement, the task execution phase is the motor imagery phase, where the target user needs to imagine left hand movement in fMRI; when the decoding need is for language decoding, the task execution phase is the word imagery phase, where the target user needs to imagine specific words in fMRI. As shown, the motor imagery process sequentially passes through fixation point, auditory task cue, motor imagery, and rest phase; the word imagery process sequentially passes through fixation point, textual task cue, word imagery, and rest phase. It should be understood that although this embodiment uses motion imagery and word imagery as examples for illustration, in other embodiments, the experimental paradigm can be designed according to specific decoding requirements, such as visual stimulus tasks, language comprehension tasks, or emotion judgment tasks, as long as the target functional area corresponding to the decoding requirements can be activated.

[0198] S701, Perform a task-based functional magnetic resonance imaging (fMRI) scan on the target user according to the stated functional magnetic resonance imaging (fMRI) experimental paradigm, acquire the fMRI signals, and record the task timing information corresponding to the fMRI experimental paradigm. This step aims to perform fMRI scans on the target user according to the experimental paradigm determined in S700, acquire BOLD signals reflecting brain functional activity, and simultaneously record the task timing information corresponding to the experimental paradigm. Specifically, functional magnetic resonance imaging utilizes the blood oxygen level dependent (BOLD) effect to indirectly reflect neural activity by detecting changes in the magnetization rate of hemoglobin in brain tissue. When the target user performs a specific task according to the experimental paradigm, neuronal activity in the relevant brain regions increases, leading to increased local cerebral blood flow, which in turn causes changes in T2-weighted signal intensity. This step simultaneously records task timing information, i.e., the time sequence of task start, duration, and end, while acquiring BOLD signals. For example, in an experimental paradigm of hand motor imagery, the task timing information may include: 2 seconds of fixation, 1 second of auditory task cues, 6 seconds of motor imagery, 4 seconds of rest, and repeated multiple times.

[0199] S702, construct a design matrix based on the task time-series information and perform statistical model estimation on the functional magnetic resonance imaging (fMRI) signal to obtain an activation statistical map related to the target task. This step is the core of fMRI data analysis, aiming to extract activation components related to the target task from the BOLD signal. Specifically, the design matrix is ​​a key input to the generalized linear model (GLM) used to describe the time series of experimental conditions. Each column of the design matrix corresponds to an experimental condition (such as task state, rest state) or control variable (such as head movement parameters, linear drift), and each row corresponds to a time point. By fitting the design matrix to the BOLD signal, the GLM can estimate the response amplitude (i.e., β parameter) of each spatial voxel to different experimental conditions. For example, for a hand movement imagery task, the design matrix may contain a column of square wave sequence representing the task state (1 during the task period, 0 during the rest period) and several columns of nuisance variables representing head movement correction parameters. After GLM estimation, the β parameter of each spatial voxel can be obtained, reflecting the signal change amplitude of that voxel during the task. Subsequently, activation statistics (such as t-tests or F-tests) for each spatial voxel are calculated to determine the activation statistics map. This activation statistics map reflects the response intensity of different brain regions to the target task, and the regions with high activation statistics are the functional areas related to the target task. It should be understood that although this embodiment uses GLM for statistical model estimation, other fMRI analysis methods, such as independent component analysis (ICA) or correlation analysis, can also be used in other embodiments, as long as the task-related activation signals can be extracted.

[0200] S703, the activation statistics map is registered to the space of the first weighted structural image to determine the individual functional activation map. This step aims to align the functional activation results with the structural image space. Since fMRI images typically have low spatial resolution and potential geometric distortions, they need to be registered to the high-resolution T1-weighted structural image space to accurately locate the anatomical position of functional areas. Specifically, the registration process can employ a rigid registration method similar to that in Example 6, but due to the significant difference in grayscale distribution between fMRI and T1 images, non-rigid transformations (such as affine or elastic transformations) may need to be introduced to correct local geometric distortions. After registration, the activation statistics map is resampled to the T1 individual space to determine the individual functional activation map. This individual functional activation map is directly superimposed on the individual anatomical structure, accurately reflecting the spatial distribution of the target functional area within the target user's own brain, avoiding individual difference errors caused by using a standard brain template.

[0201] S704, Project the individual functional activation map onto the individual cortical surface to determine the spatial distribution of the target functional area on the individual cortical surface. This step aims to convert voxel-level activation information into an activation distribution on the cortical surface for subsequent mapping to the medial surface of the dura mater. Specifically, the individual cortical surface can be determined by surface extraction algorithms (such as MarchingCubes or LevelSet) on a brain parenchyma mask, or by using cortical segmentation software (such as FreeSurfer) to determine a high-precision cortical surface model. The projection process can employ nearest-neighbor mapping or weighted average mapping, assigning the voxel activation values ​​closest to the cortical surface in the individual functional activation map to the corresponding cortical vertices. Through projection, the spatial distribution of the target functional area on the individual cortical surface can be determined. For example, for a hand motor imagery task, the target functional area may be distributed in specific regions of the precentral gyrus and postcentral gyrus. It should be understood that the extraction accuracy of the cortical surface directly affects the accuracy of functional localization; therefore, this step preferably uses a high-precision cortical segmentation algorithm.

[0202] S705, determine the candidate implantation area for the subdural electrode based on the spatial distribution on the individual's cortical surface. This step aims to convert the functionally activated cortical area into an implantable area for the subdural electrode. Since the subdural electrode is usually placed below the dura mater, close to the surface of the cerebral cortex, it is necessary to establish a spatial correspondence between the cortical surface and the medial surface of the dura mater. The specific implementation of this step will be detailed in subsequent steps S706 to S707.

[0203] In some embodiments, determining the candidate implantation region for subdural electrodes based on the spatial distribution on the individual's cortical surface includes the following sub-steps:

[0204] S706, determine the medial dura mater surface based on the dura mater candidate search space, and calculate the minimum distance from each vertex on the medial dura mater surface to the spatial distribution of the target functional area on the individual cortical surface. This step aims to establish the geometric relationship between the medial dura mater surface and the cortical functional activation area. Specifically, the medial dura mater surface can be determined by extracting the medial boundary of the dura mater candidate search space (i.e., corresponding to...). The isosurface of the location is obtained. For each vertex on the medial surface of the dura mater, the minimum Euclidean distance to all vertices of the target functional area on the individual cortical surface is calculated. This minimum distance reflects the spatial proximity of a point on the medial surface of the dura mater to the target functional area; the smaller the distance, the closer the point is to the target functional area, and the more suitable it is as a candidate site for electrode implantation. It should be understood that although this embodiment uses the minimum Euclidean distance as a metric, other distance metrics may be used in other embodiments, such as geometric distance along the cortical surface or weighted distance considering the depth of the sulci, as long as they can reflect the spatial proximity relationship.

[0205] S707, based on the minimum distance, the functional activation thermal values ​​on the individual cortical surface are mapped to the medial dura mater surface, wherein the distance attenuation parameter controls the attenuation rate of the mapping weight as the distance increases. This step aims to transfer the activation intensity of the cortical surface to the medial dura mater surface, determining the functional mapping intensity distribution on the medial dura mater surface. Through distance attenuation mapping, a spatial transformation from "cortical functional activation" to "medial dura mater surface functional mapping" is achieved, ensuring the spatial correspondence between the candidate implantation area of ​​the subdural electrode and the target functional area. It should be understood that although this embodiment uses an exponential decay function for mapping, other attenuation functions, such as linear decay or Gaussian decay, can also be used in other embodiments, as long as the effect of higher mapping weight with closer distance is achieved.

[0206] S708, based on preset activation and cluster thresholds, the functional mapping intensity on the inner surface of the dura mater is screened to determine the candidate implantation region for the subdural electrode. Specifically, the screening formula can be:

[0207]

[0208]

[0209] ,

[0210] In the formula, Indicates the target functional cortical surface region. The first on the surface of the individual's cortex Individual factors, This indicates the activation value of that vertex in the function activation heatmap. This indicates the preset activation threshold. The medial side of the dura mater indicates the medial surface of the dura mater. Functional mapping strength of individual elements Indicates the first on the dura mater The coordinates of the individual elements Indicates the first functional cortical surface region The coordinates of the individual elements This represents the distance decay parameter, which determines the decay rate of the distance term. When this parameter is large, as the distance increases, the mapping weight decreases rapidly. Only dura mater voxels very close to the target functional cortical region will obtain high functional mapping values. The mask representing the target dura mater region needs to satisfy a value greater than the functional mapping value. Greater than the function mapping threshold And cluster value Greater than the cluster threshold .

[0211] This step aims to screen candidate regions that meet the implantation criteria from the functional mapping intensity distribution on the medial surface of the dura mater. Specifically, the activation threshold... To filter vertices with high function mapping strength, it can be set to, for example, the 90th percentile of the function mapping strength distribution or a fixed threshold (e.g., (Value greater than 3.0). Only when the function mapping strength is higher than the activation threshold. Only the vertices of the clusters are considered potential electrode implantation sites. This is used to further filter consecutive activation regions, preventing isolated single vertices from being misclassified as candidate regions. For example, cluster thresholding. This can be set to the minimum cluster area (e.g., 50 square millimeters) or the minimum number of vertices (e.g., 10 vertices). Only when the function mapping strength is above the activation threshold... And the number of vertices is greater than the cluster threshold. Only continuous regions were identified as candidate implantation areas for subdural electrodes. This dual threshold screening ensured that candidate implantation areas possessed both high functional relevance and sufficient area to accommodate the electrode array. It should be understood that the activation threshold... and cluster threshold The specific value can be adjusted according to electrode size, implantation precision or clinical needs, and this embodiment does not impose strict limitations on it.

[0212] In some embodiments, the method further includes:

[0213] S709, Based on the dura mater candidate search space and the brain parenchyma mask, determine an avoidance mask; wherein, the avoidance mask represents a region located within the dura mater mask but not belonging to either the dura mater mask or the brain parenchyma mask, and the avoidance mask is used as an avoidance region or a safety constraint region in the step of determining the subdural electrode candidate implantation region based on task-oriented functional magnetic resonance imaging. This step aims to define a safety constraint region for electrode implantation planning. Specifically, the definition formula of the avoidance mask can be expressed as:

[0214] ,

[0215] In the formula, Avoid mask values, It is the intradural mask value, representing the area enclosed by the inner surface of the dura mater (i.e., within the dural envelope). Indicates the dura mater mask value. This represents the brain parenchyma mask value. The spatial region corresponding to the avoidance mask typically includes the cerebrospinal fluid space, subarachnoid space, blood vessels, or other non-target tissue regions located between the dura mater and brain parenchyma. In electrode implantation planning, these regions need to be avoided or used as safety constraint boundaries to prevent electrodes from penetrating blood vessels, damaging brain tissue, or entering the cerebrospinal fluid space, which could lead to signal quality degradation. For example, when determining candidate implantation regions for subdural electrodes, the avoidance mask can be used as a negative constraint to exclude regions that overlap with the avoidance mask, thereby ensuring the safety of the electrode implantation location. It should be understood that the definition of the avoidance mask depends on the segmentation results of the aforementioned dural candidate search space and brain parenchyma mask, reflecting the collaborative application of structural modeling and functional localization.

[0216] Through steps S700 to S709 above, this embodiment details the complete process of determining candidate implantation regions for subdural electrodes based on task-oriented functional magnetic resonance imaging (fMRI). Specifically, the experimental paradigm design step determines a matching fMRI experimental paradigm based on decoding requirements, ensuring that the acquired BOLD signals effectively activate the target functional areas corresponding to the decoding needs; the fMRI signal processing flow (design matrix → GLM estimation → activation statistics map) accurately extracts individual functional activation signals; the registration and projection steps map the functional activation results to the individual anatomical space; the distance mapping formula and activation threshold screening establish a spatial correspondence between cortical functional activation and the medial surface of the dura mater; and the definition of mask avoidance provides a safe constraint area for electrode implantation. Overall, this embodiment, through the combination of structural modeling and functional localization, achieves a spatial transformation from "brain functional activation location" to "subdural electrode coverage location," providing individualized references for electrode implantation in key functional areas such as the sensorimotor, language, and visual cortices, significantly improving the accuracy and safety of preoperative planning.

[0217] Example 2:

[0218] like Figure 10 As shown, this embodiment provides a 3D modeling system for the dura mater. Through the collaborative work of its various modules, this system achieves high-precision, individualized 3D modeling of the dura mater, providing a reliable data foundation for preoperative planning, navigation and positioning, and electrode implantation. Specifically, it includes the following modules:

[0219] Data acquisition module 1 is used to acquire the first and second weighted structural images of the target user. This module is the input terminal of the system and is responsible for data interaction with medical imaging scanning equipment (such as an MRI scanner) to acquire raw DICOM or NIfTI format image data. Specifically, data acquisition module 1 can read the target user's T1-weighted and T2-weighted structural images through a network interface (such as a hospital PACS system) or local storage media (such as an optical disc or external hard drive). The T1-weighted structural image serves as the primary image, providing high-quality individual brain anatomical structure information; the T2-weighted structural image serves as an auxiliary image, supplementing the image contrast information of different tissues near water content, cerebrospinal fluid, and intracranial boundaries. It should be understood that although this embodiment uses T1 and T2-weighted structural images as examples, in other embodiments, data acquisition module 1 can also acquire other medical image sequences that can reflect brain anatomical structure and tissue contrast information, as long as they can provide complementary tissue feature information. In addition, the data acquisition module 1 can also acquire metadata related to the image data, such as voxel size, spatial resolution, and scanning parameters, providing basic data for the subsequent processing by the spatial registration module 2.

[0220] Spatial registration module 2 is used to spatially register the second weighted structural image. This module is connected to data acquisition module 1, receives the first and second weighted structural images as input, and is responsible for resolving inconsistencies in spatial resolution, voxel coordinates, and physical coordinates between different image sequences. Specifically, spatial registration module 2 executes the rigid registration algorithm described in Example 6, aiming to maximize the global pixel mutual information between the two modal structural images. It iteratively optimizes the rotation and translation spatial parameters within six degrees of freedom, and resamples the second weighted structural image using the optimized parameters, registering it to the three-dimensional voxel space of the first weighted structural image. Through spatial registration, the two modal images correspond to the same anatomical location under the same voxel grid, providing a unified spatial reference for subsequent processing by brain parenchyma segmentation module 3. It should be understood that spatial registration module 2 can be implemented in software (such as a registration algorithm based on the ITK library) or accelerated in hardware (such as GPU parallel computing) to improve registration efficiency.

[0221] The brain parenchyma segmentation module 3 is used to segment the brain parenchyma mask based on the first weighted structural image. This module is connected to the spatial registration module 2, receives the registered first weighted structural image as input, and is responsible for separating the brain parenchyma region from the T1 weighted structural image, excluding scalp, skull, air background, and other non-brain tissues. Specifically, the brain parenchyma segmentation module 3 executes the segmentation process described in Example 2, performs outlier truncation and normalization based on the statistical quantile values ​​of the global voxel intensity of the first weighted structural image, extracts initial candidate masks, removes small outliers through three-dimensional morphological opening operations, performs connected component hole filling, extracts the connected component with the largest number of three-dimensional voxels as the base mask, and performs erosion and dilation correction on the outer contour of the base mask to finally determine the brain parenchyma mask. This brain parenchyma mask reflects the anatomical boundaries of the individual brain, providing a structural basis for the subsequent processing of the search space construction module 4. It should be understood that the brain parenchyma segmentation module 3 can integrate existing brain parenchyma segmentation algorithms (such as BET or FSL) or adopt the adaptive segmentation process described in this invention to improve the individualization of the segmentation.

[0222] Search space construction module 4 is used to construct a smooth outer mask of the brain parenchyma based on the brain parenchyma mask, and to construct a candidate search space for the dura mater based on the smooth outer mask. This module is connected to the brain parenchyma segmentation module 3, receives the brain parenchyma mask as input, and is responsible for constructing the possible spatial range of the dura mater. Specifically, search space construction module 4 first performs morphological closure operations and hole filling on the brain parenchyma mask to eliminate sulci depressions and obtain a smooth outer mask of the brain parenchyma; then, it calculates the lateral distance from each spatial voxel to the smooth outer mask of the brain parenchyma, and constructs a candidate search space for the dura mater based on the lateral distance and the preset medial dura mater gap and dura mater thickness. This search space is a shell region that limits the possible spatial range of the dura mater, used to restrict the spatial prior region for subsequent probability calculations, reflecting the design concept of "anatomical constraints". It should be understood that search space construction module 4 can preset default anatomical parameters (such as... =2mm, =1mm), or it can receive individualized anatomical parameter input through the user interface to adapt to the anatomical differences of different target users.

[0223] The threshold estimation module 5 is used to adaptively determine the individualized segmentation threshold based on the image intensity distribution corresponding to the target user. This module is connected to the search space construction module 4 and receives the first weighted structural image, the second weighted structural image, the brain parenchyma mask, and the dura mater candidate search space as input. It is responsible for adaptively determining the individualized segmentation threshold based on the image intensity distribution characteristics corresponding to a single target user. Specifically, the threshold estimation module 5 extracts the signal intensity percentile of the first weighted structural image within the brain parenchyma mask to determine the first signal reference threshold; extracts the signal intensity percentage of the first weighted structural image within the brain parenchyma mask and the dura mater candidate search space to determine the first lower limit threshold and the first upper limit threshold; extracts the signal intensity percentage of the second weighted structural image within the brain parenchyma mask to determine the second brain parenchyma mask signal quantile value; and extracts the signal intensity percentage within the dura mater candidate search space to determine the second dura mater candidate signal quantile value. Through individualized threshold estimation, the fixed threshold judgment is transformed into an individualized statistical judgment, enabling the system to adapt to the image differences of different target users, embodying the design concept of "adaptive parameters." It should be understood that the threshold estimation module 5 can be implemented using a statistical calculation unit (such as a sorting algorithm for calculating quantiles), and its calculation results can be directly output to the probability fusion module 6.

[0224] The probability fusion module 6 is used to perform weighted calculations based on the signal intensity of the first weighted structural image, the signal intensity of the second weighted structural image, and the spatial distance prior corresponding to the dura mater candidate search space, combined with the individualized segmentation threshold, to obtain a voxel assignment probability map. This module is connected to the threshold estimation module 5 and receives the individualized segmentation threshold, the first weighted structural image, the second weighted structural image, and the dura mater candidate search space as input. It is responsible for integrating multimodal image features and spatial distance priors to perform probability calculations for each spatial voxel within the dura mater candidate search space. Specifically, the probability fusion module 6 executes a probability fusion process to calculate a first signal score (including a lower limit score and a higher limit score) based on the signal intensity of the first weighted structural image, a second signal score based on the signal intensity of the second weighted structural image, and a distance prior score based on the spatial distance prior. The three scores are then weighted to obtain the voxel assignment probability map. This probability map reflects the probability value of each candidate voxel belonging to the dura mater, embodying the design concept of multimodal decision-making. It should be understood that the probability fusion module 6 can use matrix operation units (such as GPU or DSP) to achieve efficient voxel-level probability calculation, and its calculation results are output to the 3D model determination module 7.

[0225] The 3D model determination module 7 is used to segment the dura mater region based on the voxel assignment probability map and perform 3D reconstruction on the segmentation results to determine an individualized 3D model of the dura mater. This module is connected to the probability fusion module 6, receives the voxel assignment probability map as input, and is responsible for determining the final 3D model based on the probability map. Specifically, the 3D model determination module 7 performs a segmentation and reconstruction process, marking voxels with probability values ​​higher than a preset segmentation judgment threshold in the voxel assignment probability map as target voxels, and binarizing them to determine a coarse dura mater mask; it performs a spatial topological connectivity check on the coarse dura mater mask, filters out isolated misjudged regions with volume sizes smaller than a preset tolerance, and repairs local broken boundaries to determine a continuous fine dura mater mask; it extracts surface isosurfaces from the fine dura mater mask and performs mesh rendering to determine an individualized 3D model of the dura mater. This 3D model can intuitively display the individual dura mater morphology and can be used for preoperative navigation, instrument path planning, electrode adhesion analysis, and craniotomy area design. It should be understood that the 3D model determination module 7 can integrate a 3D visualization engine (such as VTK or OpenGL) to output the determined 3D model in a standard mesh format (such as STL, OBJ or PLY) for use by the subsequent surgical planning system.

[0226] Through the collaborative work of modules 1 to 7, the system in this embodiment realizes a complete technical chain of "search space construction → individualized threshold estimation → multi-feature probability fusion". Specifically, the search space construction module 4 provides spatial constraints, the threshold estimation module 5 provides adaptive parameters, and the probability fusion module 6 provides multimodal decision-making. These three modules work together to solve the problem of limited individualization and segmentation accuracy in dura mater modeling in existing technologies. It should be understood that although this embodiment divides the system into seven modules, in other embodiments, modules can be merged or split according to functional requirements. For example, the threshold estimation module 5 and the probability fusion module 6 can be merged into a single comprehensive calculation module, as long as the functions described in this invention can be achieved. Furthermore, data flow between modules can be achieved through internal buses, shared memory, or network interfaces; this embodiment does not strictly limit these methods.

[0227] Example 3:

[0228] This embodiment provides an electronic device, a computer-readable storage medium, and a computer program product for implementing the three-dimensional modeling method of the dura mater described in Embodiment 1 above. The electronic device executes stored computer program instructions through a processor to complete a series of processing steps, including image data acquisition, spatial registration, brain parenchyma segmentation, search space construction, threshold estimation, probability fusion, and three-dimensional reconstruction, thereby determining an individualized three-dimensional model of the dura mater.

[0229] Specifically, the electronic device includes a processor and a memory. The processor can be a general-purpose processor, such as a central processing unit (CPU), digital signal processor (DSP), network processor (NP), etc., or a dedicated processor, such as a graphics processing unit (GPU), embedded processor, microcontroller (MCU), or field-programmable gate array (FPGA). The processor executes computer program instructions stored in the memory to implement the dura mater 3D modeling method described in this invention. It should be understood that although this embodiment uses a general-purpose processor as an example, in other embodiments, a dedicated processor or a parallel computing architecture composed of multiple processors (such as a multi-core CPU or GPU cluster) can also be used to improve the computational efficiency of image processing and 3D reconstruction. The memory stores processor-executable instructions. The memory can include volatile memory (such as random access memory RAM) and non-volatile memory (such as read-only memory ROM, flash memory, hard disk, solid-state drive SSD, etc.). The volatile memory is used to temporarily store the currently executing program instructions and intermediate processing data (such as image data, probability maps, mask data, etc.), while the non-volatile memory is used to permanently store the computer program code, system configuration files, and the final determined dura mater 3D model file. It should be understood that the specific type and capacity of the memory can be configured according to the scale of the image data and processing requirements. For example, for high-resolution three-dimensional medical image data, a large capacity RAM (such as 16GB or 32GB) and a high-speed SSD can be configured to ensure data read and write speed and processing efficiency.

[0230] Furthermore, the electronic device may also include a bus architecture for connecting the processor, memory, and other peripheral devices (such as input / output interfaces, network interfaces, display devices, etc.). The bus architecture can be a standard bus, such as a PCIe bus, USB bus, or SATA bus, or a dedicated bus, such as an on-chip bus. Through the bus architecture, the processor can read computer program instructions from memory and output the processing results (such as a three-dimensional model of the dura mater) to a display device or transmit them to a surgical planning system via a network interface. It should be understood that although this embodiment describes the basic components of a bus architecture, in other embodiments, the electronic device may also adopt a distributed architecture (such as a cloud computing platform), connecting multiple computing nodes through a network to achieve parallel processing of large-scale image data.

[0231] The processor is configured to implement the method described in Embodiment 1 when executing the instructions. Specifically, when the processor loads and executes computer program instructions from memory, it will sequentially perform the following steps: First, the processor acquires the T1-weighted structural image and T2-weighted structural image of the target user through an input / output interface, and registers the T2-weighted structural image to the three-dimensional voxel space of the T1-weighted structural image (corresponding to S100 and the registration algorithm in Embodiment 1); then, the processor segments the T1-weighted structural image to obtain a brain parenchyma mask, and constructs a smooth outer mask of the brain parenchyma based on the brain parenchyma mask (corresponding to S200 and the segmentation process in Embodiment 1); next, the processor constructs a dura mater candidate search space based on the three-dimensional physical distance from the spatial voxel to the smooth outer mask of the brain parenchyma (corresponding to S300 and the search process in Embodiment 1). The process involves several steps: 1) Spatial construction process; 2) The processor adaptively determines an individualized segmentation threshold based on the image intensity distribution characteristics corresponding to the target user (corresponding to S400 and the threshold estimation process in Example 1); 3) The processor performs weighted calculations based on the signal intensity of the T1-weighted structural image, the signal intensity of the T2-weighted structural image, and the spatial distance prior corresponding to the dura mater candidate search space, combined with the individualized segmentation threshold, to obtain a voxel assignment probability map (corresponding to S500 and the probability fusion process in Example 1); 4) The processor segments the dura mater region based on the voxel assignment probability map and performs 3D reconstruction on the segmentation results to determine the individualized dura mater 3D model (corresponding to S600 and the segmentation and reconstruction process in Example 1). It should be understood that the specific logic of the processor executing the above steps has been detailed in Example 1, and this example will not repeat the details. By executing program instructions, the electronic device can automatically complete the entire process of dura mater 3D modeling without manual intervention, significantly improving modeling efficiency and individualization.

[0232] Furthermore, this embodiment also provides a computer-readable storage medium storing computer program instructions. When these computer program instructions are executed by a processor, they implement the method described in Embodiment 1 above. Specifically, the computer-readable storage medium can be a magnetic storage medium (such as a hard disk or magnetic tape), an optical storage medium (such as an optical disc, DVD, or CD-ROM), a semiconductor storage medium (such as flash memory or SSD), or any other physical medium capable of storing computer program code. The computer program instructions are stored in the medium in the form of binary code or high-level language code. When the processor reads and executes these instructions, the three-dimensional modeling method of the dura mater described in this invention will be implemented. It should be understood that the computer-readable storage medium can exist independently of the electronic device (such as a portable storage disk) or can be integrated inside the electronic device (such as a built-in hard disk). By distributing computer-readable storage media, the technical solution of this invention can be conveniently deployed on different computing platforms (such as hospital servers, personal workstations, or cloud computing nodes), improving the applicability and flexibility of the technical solution.

[0233] Furthermore, this embodiment also provides a computer program product, including a computer program. When executed by a processor, this computer program implements the method described in Embodiment 1 above. Specifically, the computer program product may be a software package, application program, or software module, containing all the program code and related configuration files required to implement the method of the present invention. The computer program product can be provided to users through network download (e.g., download from a software distribution server), physical media distribution (e.g., CD-ROM or USB flash drive), or pre-installed on electronic devices. When a user installs and runs the computer program product on an electronic device, the processor executes the program code to implement the dura mater three-dimensional modeling method. It should be understood that the computer program product may also include a user interface module for displaying image data, intermediate processing results (e.g., brain parenchyma mask, dura mater candidate search space), and the final three-dimensional model, and providing a parameter setting interface (e.g., adjusting anatomical parameters). , or weight parameters , , This allows users to adjust the modeling process according to their actual needs. By providing a computer program product, the technical solution of this invention can be flexibly integrated into existing medical image processing systems or surgical planning systems in software form, reducing hardware deployment costs and improving the convenience of clinical application.

[0234] Through the implementation of the aforementioned electronic devices, computer-readable storage media, and computer program products, the dura mater 3D modeling method of the present invention can run efficiently on a general-purpose computing platform. The collaborative work of the processor and memory provides powerful computing and data storage capabilities, the bus architecture ensures high-speed data transmission, and the computer program instructions transform complex method logic into executable code, achieving automated processing. Overall, this embodiment provides a universal hardware and software carrier for the dura mater 3D modeling method, enabling its widespread application in various scenarios such as hospital radiology departments, neurosurgical operating rooms, or research laboratories, significantly improving the practical value and ease of promotion of the technical solution.

[0235] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. For example, although the foregoing embodiments use T1-weighted structural images and T2-weighted structural images as examples, other medical image sequences that can reflect brain anatomical structures and tissue contrast information can also be used in other embodiments; although the foregoing embodiments use the Sigmoid function as a smoothing step function and the MarchingCubes algorithm for isosurface extraction, equivalent algorithms with similar functions can also be used in other embodiments; although the foregoing embodiments provide specific threshold percentiles, weighting coefficients, and anatomical parameter values, these parameters can be adjusted according to image quality, individual differences, or clinical needs. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0236] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0237] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for three-dimensional modeling of the dura mater, characterized in that, include: Obtain the first weighted structural image and the second weighted structural image of the target user, and register the second weighted structural image to the three-dimensional voxel space of the first weighted image; The brain parenchyma mask is obtained by segmenting the first weighted structural image, and a smooth outer mask of the brain parenchyma is constructed based on the brain parenchyma mask. Based on the three-dimensional physical distance from the spatial voxel to the smooth outer mask of the brain parenchyma, a candidate search space for the dura mater is constructed. Based on the image intensity distribution characteristics corresponding to the target user, an individualized segmentation threshold is adaptively determined. The adaptive determination of the individualized segmentation threshold includes: extracting the signal intensity percentile of the first weighted structural image within the brain parenchyma mask to determine a first signal reference threshold; extracting the signal intensity percentage of the first weighted structural image within the brain parenchyma mask and within the dura mater candidate search space to determine a first lower signal threshold and a first upper signal threshold; and extracting the signal intensity percentage within the dura mater candidate search space to determine a second upper signal threshold. Based on the signal intensity of the first weighted structural image, the signal intensity of the second weighted structural image, and the spatial distance prior corresponding to the dura mater candidate search space, a weighted calculation is performed in conjunction with the individualized segmentation threshold to obtain the voxel assignment probability map. The dura mater region is segmented based on the voxel assignment probability map, and three-dimensional reconstruction is performed on the segmentation results to determine an individualized three-dimensional model of the dura mater.

2. The method according to claim 1, characterized in that, The process of segmenting the first weighted structural image to obtain a brain parenchyma mask, and constructing a smooth outer mask of the brain parenchyma based on the brain parenchyma mask, includes: Based on the statistical quantile values ​​of the global voxel intensity of the first weighted structural image, outlier truncation and normalization are performed on the voxel intensity. Initial candidate masks are extracted based on the normalized voxel intensity, outlier noise is removed by 3D morphological opening operation, and connected component hole filling is performed. The connected component with the largest number of three-dimensional voxels is extracted as the basic mask. The boundary residue of the outer contour of the basic mask is reduced and the boundary is restored to determine the brain parenchyma mask. Morphological closure and hole filling operations are performed on the brain parenchyma mask to eliminate brain sulci depressions, resulting in a smooth outer mask of the brain parenchyma.

3. The method according to claim 1, characterized in that, The process of constructing a candidate search space for the dura mater based on the three-dimensional physical distance from the spatial voxel to the smooth outer mask of the brain parenchyma includes: Calculate the outer distance from the spatial voxel to the smooth outer mask of the brain parenchyma, wherein the outer distance is the minimum physical distance from the voxel to the smooth outer mask of the brain parenchyma, and the outer distance is zero for voxels belonging to the smooth outer mask of the brain parenchyma. The candidate search space for the dura mater is constructed based on the lateral distance, the preset medial dura mater gap, and the dura mater thickness, wherein the medial dura mater gap represents the gap distance between the medial surface of the dura mater and the surface of the brain parenchyma, and the dura mater thickness represents the thickness parameter of the dura mater.

4. The method according to claim 1, characterized in that, The first signal lower limit threshold includes the maximum value of the preset low quantile of the percentage of signal intensity of the first weighted structural image within the brain parenchyma mask and the dura mater candidate search space; The first signal reference threshold includes the percentage of signal intensity of the first weighted structural image within the brain parenchyma mask; The first signal upper limit threshold includes the maximum value of the preset high quantile of the percentage of signal intensity of the first weighted structural image within the brain parenchyma mask and the dura mater candidate search space.

5. The method according to claim 4, characterized in that, The method for obtaining the voxel assignment probability map includes: Based on the signal intensity of the first weighted structured image, a first signal score is calculated, which includes a first signal lower limit score and a high signal upper limit score. The second signal score is calculated based on the signal strength of the second weighted structured image and the second signal upper limit threshold. Calculate the distance prior score based on the spatial distance prior; The first signal score, the second signal score, and the distance prior score are weighted and calculated to obtain the voxel assignment probability map.

6. The method according to claim 5, characterized in that, The first signal score is calculated in the following way: Based on the relationship between the signal strength of the first weighted structural image, the first signal lower limit threshold and the first signal reference threshold, the first signal lower limit score is calculated using a smooth step function. Based on the relationship between the signal intensity of the first weighted structural image and the first signal upper limit threshold, the high signal upper limit score is calculated using the smooth step function; The first signal score is obtained by multiplying the lower limit score of the first signal by the upper limit score of the high signal. The second signal score is calculated as follows: Based on the relationship between the signal intensity of the second weighted structural image and the upper threshold of the second signal, the score of the second signal is calculated using a smooth step function; The distance prior score is calculated in the following way: The distance prior score is calculated using an inverse linear decay function based on the minimum physical distance from the voxel to the smooth outer mask of the brain parenchyma.

7. The method according to claim 6, characterized in that, The weighted calculation based on the individualized segmentation threshold is achieved in the following way: The first signal score, the second signal score, and the distance prior score are multiplied by their respective first weight, second weight, and third weight, and then summed.

8. The method according to claim 1, characterized in that, The step of segmenting the dura mater region based on the voxel assignment probability map and performing three-dimensional reconstruction on the segmentation results to determine an individualized three-dimensional model of the dura mater includes: Voxels whose probability values ​​in the voxel assignment probability map are higher than the preset segmentation judgment threshold are marked as target voxels, and binarization is used to determine the dura mater coarse mask. A spatial topological connectivity check is performed on the coarse dura mater mask to filter out isolated misjudged regions with a volume size smaller than a preset tolerance, and local broken boundaries are repaired to determine a continuous fine dura mater mask. The surface isosurface of the dura mater is extracted from the fine mask and meshed rendering is performed to determine the individualized 3D model of the dura mater.

9. The method according to claim 1, characterized in that, Registering the second weighted structural image to the three-dimensional voxel space of the first weighted structural image includes: Extract the spatial geometric relationship between the first weighted structured image and the second weighted structured image; With the goal of maximizing the global pixel mutual information between the two modal structured images, the rotation space parameters and translation space parameters are iteratively optimized within six degrees of freedom. The second weighted structural image is resampled using the rotational and translational spatial parameters obtained through optimization to complete cross-modal rigid registration.

10. The method according to claim 1, characterized in that, It also includes the step of determining candidate implantation regions for subdural electrodes based on task-based functional magnetic resonance imaging, specifically including: Acquire the functional magnetic resonance imaging (fMRI) signal of the target user and the corresponding task timing information; A design matrix is ​​constructed based on the task time series information, and a statistical model is performed on the functional magnetic resonance signal to obtain an activation statistics map related to the target task. The activation statistics map is registered to the space where the first weighted structural image is located to determine the individual functional activation map; The individual functional activation map is projected onto the individual cortical surface to determine the spatial distribution of the target functional area on the individual cortical surface, and the candidate implantation area of ​​the subdural electrode is determined based on the spatial distribution on the individual cortical surface.

11. The method according to claim 10, characterized in that, Prior to the step of determining candidate implantation regions for subdural electrodes based on task-oriented functional magnetic resonance imaging, the method further includes: The functional magnetic resonance imaging (fMRI) experimental paradigm is determined based on the decoding requirements, and the fMRI experimental paradigm includes a task stimulus sequence that matches the decoding requirements. The functional magnetic resonance experimental paradigm consists of multiple task cycles, each of which includes a gaze phase, a task cueing phase, a task execution phase, and a rest phase, wherein the content of the task execution phase corresponds to the decoding requirements. According to the functional magnetic resonance experimental paradigm, the target user is subjected to task-state functional magnetic resonance scanning, the functional magnetic resonance signals are collected, and the task timing information corresponding to the functional magnetic resonance experimental paradigm is recorded. The decoding requirement includes at least one of hand movement decoding requirement or language decoding requirement; in response to the decoding requirement being a hand movement decoding requirement, the task execution stage is a motor imagery stage, and in response to the decoding requirement being a language decoding requirement, the task execution stage is a word imagery stage.

12. The method according to claim 10, characterized in that, The step of determining the candidate implantation region for subdural electrodes based on the spatial distribution on the individual's cortical surface includes: The inner surface of the dura mater is determined based on the candidate search space of the dura mater, and the minimum distance from each vertex on the inner surface of the dura mater to the spatial distribution of the target functional area on the individual cortical surface is calculated. Based on the minimum distance, the functional activation thermometric values ​​on the individual cortical surface are mapped to the inner surface of the dura mater, wherein the distance decay parameter controls the decay rate of the mapping weight as the distance increases. Based on preset activation thresholds and cluster thresholds, the functional mapping intensity on the inner surface of the dura mater is screened to determine the candidate implantation area of ​​the subdural electrode.

13. The method according to claim 10, characterized in that, Also includes: Based on the dura mater candidate search space and the brain parenchyma mask, an avoidance mask is determined; The avoidance mask refers to the area located within the dura mater mask but not belonging to the dura mater mask or the brain parenchyma mask. The avoidance mask can be used as an avoidance area or a safety constraint area in the step of determining the candidate implantation area of ​​the subdural electrode based on task-based functional magnetic resonance imaging.

14. A three-dimensional modeling system for the dura mater, characterized in that, include: The data acquisition module is used to acquire the first weighted structural image and the second weighted structural image of the target user; A spatial registration module is used to perform spatial registration on the second weighted structural image; The brain parenchyma segmentation module is used to segment the brain parenchyma mask based on the first weighted structural image. The search space construction module is used to construct a smooth outer mask of brain parenchyma based on the brain parenchyma mask, and to construct a candidate search space for dura mater based on the smooth outer mask of brain parenchyma. The threshold estimation module is used to adaptively determine an individualized segmentation threshold based on the image intensity distribution corresponding to the target user. The adaptive determination of the individualized segmentation threshold includes: extracting the signal intensity percentile of the first weighted structural image within the brain parenchyma mask to determine a first signal reference threshold; extracting the signal intensity percentage of the first weighted image within the brain parenchyma mask and within the dura mater candidate search space to determine a first lower signal threshold and a first upper signal threshold; and extracting the signal intensity percentage within the dura mater candidate search space to determine a second upper signal threshold. The probability fusion module is used to perform weighted calculations based on the signal intensity of the first weighted structural image, the signal intensity of the second weighted structural image, and the spatial distance prior corresponding to the dura mater candidate search space, combined with the individualized segmentation threshold, to obtain a voxel assignment probability map. The three-dimensional model determination module is used to segment the dura mater region according to the voxel assignment probability map, and perform three-dimensional reconstruction on the segmentation results to determine an individualized three-dimensional model of the dura mater.

15. An electronic device, characterized in that, include: processor; Memory is used to store processor-executable instructions; The processor is configured to implement the method of any one of claims 1 to 13 when executing the instructions.

16. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method of any one of claims 1 to 13.

17. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 13.

Images