Cerebral hemorrhage accurate drainage method and system based on image density analysis

Abstract

The invention provides a cerebral hemorrhage accurate drainage method and system based on image density analysis, and relates to the technical field of medical image processing and cerebral hemorrhage drainage, and the method comprises the steps: obtaining a head CT image, generating a multi-level three-dimensional density distribution diagram through a density analysis model, extracting a hematoma core region, and building a three-dimensional model; layering the hematoma core area, and determining an optimal drainage point and path; and a multi-channel intelligent drainage system is adopted for hematoma drainage, the drainage speed is adjusted according to the density sublayer parameters, hematoma volume and density distribution changes are monitored in real time, and when the current volume parameter is reduced to be below a preset volume value and the density distribution of each density sublayer tends to be uniform, accurate hematoma drainage is completed. According to the method, accurate positioning and personalized drainage of cerebral hemorrhage can be realized, and the treatment effect is improved.

Description

Technical Field

[0001] The present invention relates to medical image processing and cerebral hemorrhage drainage technology, and in particular to a precise cerebral hemorrhage drainage method and system based on image density analysis. Background Art

[0002] Cerebral hemorrhage is a serious neurological disease, and timely and effective hematoma removal is crucial to improving prognosis. Traditional hematoma drainage methods mainly rely on experience to judge the drainage position and speed, lacking precise and quantitative drainage parameter guidance, which can easily lead to complications caused by insufficient drainage or too fast drainage.

[0003] Although the existing image-guided hematoma drainage technology can provide the location information of the hematoma, it cannot make a detailed analysis of the internal structure of the hematoma, and lacks the ability to monitor the dynamic changes of the hematoma in real time during the drainage process. At the same time, due to the uneven density distribution of the hematoma, a single drainage channel is difficult to achieve accurate drainage of different density areas, affecting the treatment effect.

[0004] Therefore, there is an urgent need for a precise drainage method and system for cerebral hemorrhage based on image density analysis, which can finely stratify the internal structure of the hematoma through deep learning, realize the intelligent planning of the optimal drainage point and drainage path, and perform differentiated drainage of different density areas through a multi-channel intelligent drainage system. At the same time, it can dynamically adjust the drainage parameters in combination with real-time monitoring data to improve the accuracy and safety of hematoma removal. Summary of the invention

[0005] The embodiments of the present invention provide a method and system for accurate drainage of cerebral hemorrhage based on image density analysis, which can solve the problems in the prior art.

[0006] According to a first aspect of the embodiments of the present invention,

[0007] A method is provided, comprising:

[0008] Acquire a three-dimensional computer tomography image of the patient's head, calculate the density value and density gradient value of each pixel in the target bleeding area using a density analysis model based on a convolutional neural network, construct an adaptive density threshold curve according to the density value and density gradient value, generate a multi-level three-dimensional density distribution map, use a regional segmentation algorithm to extract the area in the multi-level three-dimensional density distribution map with a density greater than the density threshold curve as the hematoma core area, use a three-dimensional reconstruction algorithm to establish a three-dimensional model of the hematoma core area, and calculate the volume parameters, centroid coordinates, shape parameters and azimuth parameters of the hematoma core area;

[0009] Inputting the three-dimensional model of the hematoma core area into a deep neural network, performing layered processing on the hematoma core area based on the density gradient value, dividing the hematoma core area into a plurality of density sub-layers according to the multi-level three-dimensional density distribution map, calculating the volume ratio, spatial coordinates and fluidity parameters of each density sub-layer, determining the spatial position coordinates of the optimal drainage point by using an optimization algorithm based on the volume ratio, spatial coordinates and fluidity parameters, importing the angiography image and functional area image of the patient, generating an optimal drainage path under the constraints of the angiography image and the functional area image, and inputting the spatial position coordinates of the optimal drainage point and the optimal drainage path into an electromagnetic positioning system;

[0010] Under the guidance of the electromagnetic positioning system, a multi-channel intelligent drainage system including a main drainage tube and multiple branch drainage tubes is used to drain the hematoma, the main drainage tube is inserted to the optimal drainage point, the optimal drainage position of each density sublayer is calculated according to the spatial coordinates and fluidity parameters of the density sublayer, and multiple branch drainage tubes are respectively inserted into the optimal drainage positions, and the pressure data and electroencephalogram data of each density sublayer are collected. The pressure data, electroencephalogram data and volume ratio are input into a pre-constructed neurological function evaluation model to obtain the drainage rate parameter, and the drainage speed of the main drainage tube and the branch drainage tube is adjusted according to the drainage rate parameter, and the current volume parameter of the hematoma core area and the density distribution changes of each density sublayer are monitored in real time. When the current volume parameter drops below the preset volume value and the density distribution of each density sublayer tends to be uniform, the precise drainage of the hematoma is completed.

[0011] In an optional embodiment,

[0012] Obtain a three-dimensional computer tomography image of the patient's head, calculate the density value and density gradient value of each pixel in the target bleeding area using a density analysis model based on a convolutional neural network, construct an adaptive density threshold curve based on the density value and density gradient value, and generate a multi-level three-dimensional density distribution map including:

[0013] Acquire a three-dimensional computer tomography image of a patient's head, divide the three-dimensional computer tomography image into a plurality of overlapping sub-blocks, calculate a three-dimensional grayscale histogram for each overlapping sub-block, perform cropping and restriction on the three-dimensional grayscale histogram to obtain a cumulative distribution function, perform mapping enhancement on the three-dimensional computer tomography image according to the cumulative distribution function to obtain an enhanced image, perform denoising on the enhanced image by using anisotropic diffusion filtering, determine a diffusion coefficient by analyzing a gradient histogram of the enhanced image, and iteratively process the enhanced image according to the diffusion coefficient to obtain a preprocessed three-dimensional computer tomography image;

[0014] Inputting the preprocessed three-dimensional computed tomography image into a density analysis model, the density analysis model comprising a feature extraction subnetwork and a density calculation subnetwork, the feature extraction subnetwork comprising a plurality of three-dimensional convolution units, each of which is provided with a convolution layer with a different expansion coefficient and a three-dimensional attention calculation module, the density calculation subnetwork comprising a plurality of deconvolution layers, a feature fusion connection structure is provided between the feature extraction subnetwork and the density calculation subnetwork, the density analysis model is trained using a combined loss function including a density information constraint term to obtain a feature map of a target bleeding area, the density value of each pixel point in the target bleeding area is calculated based on the feature map, and a density value distribution map is generated;

[0015] A three-dimensional gradient operator is used to calculate the density value distribution map to obtain the density gradient value of each pixel in the target bleeding area, a two-dimensional distribution histogram is established by the density value in the density value distribution map and the density gradient value, a joint probability distribution of the density value and the density gradient value in the two-dimensional distribution histogram is calculated, the joint probability distribution is divided into multiple candidate categories, the intra-class variance and the inter-class variance of each candidate category are calculated, an optimal segmentation threshold combination that maximizes the inter-class variance and minimizes the intra-class variance is found through an iterative optimization method, a control parameter of an optimal segmentation curve is constructed according to the optimal segmentation threshold combination, a curve interpolation is performed on the optimal segmentation curve according to the control parameter to obtain an adaptive density threshold curve, the density value distribution map and the adaptive density threshold curve are input into a density stratification algorithm, the target bleeding area is density stratified, and a multi-level three-dimensional density distribution map is generated.

[0016] In an optional embodiment,

[0017] The region where the density in the multi-level three-dimensional density distribution map is greater than the density threshold curve is extracted using a regional segmentation algorithm as the hematoma core region, and a three-dimensional model of the hematoma core region is established using a three-dimensional reconstruction algorithm. The volume parameters, centroid coordinates, shape parameters, and azimuth parameters of the hematoma core region are calculated, including:

[0018] A regional segmentation algorithm is used to process the multi-level three-dimensional density distribution map, an initial seed point set is determined according to a density threshold curve, a point in the initial seed point set whose density value is greater than the threshold corresponding to the density threshold curve is used as a growth starting point, a directional density gradient value constraint condition is introduced in the regional growth process, the density value and density gradient direction of the candidate voxel are judged, the candidate voxel whose density value is greater than the density threshold and whose density gradient direction is consistent with the growth direction is included in the growth region, and the growth region is morphologically processed to obtain a segmentation boundary point set of the hematoma core area;

[0019] A three-dimensional reconstruction model is established according to the segmentation boundary point set of the hematoma core area, the density gradient value of each point in the segmentation boundary point set is calculated, the grid size of different areas is set based on the density gradient value, the space is gridded by an octree structure, a distance-based weight coefficient is calculated for each grid node, a local weighted least squares equation group is constructed and solved to obtain the implicit surface function value of the grid node, an isosurface point set with zero implicit surface function value is extracted to perform triangular meshing, and a three-dimensional model of the hematoma core area is obtained;

[0020] The morphological parameters of the core area of ​​the hematoma are calculated using the three-dimensional model, the volume value of the hematoma is calculated according to the grid vertex coordinates and vertex normal vectors of the three-dimensional model, the density values ​​in the multi-level three-dimensional density distribution map are correspondingly aligned with the voxel position information in the three-dimensional model, the centroid coordinates of the hematoma are calculated according to the aligned density values ​​and voxel positions, the inertia matrix of the hematoma area is constructed, the eigenvalues ​​of the inertia matrix are calculated, the flatness coefficient, elongation coefficient and sphericity coefficient of the hematoma are calculated according to the eigenvalues, the eigenvector of the inertia matrix is ​​calculated, the spatial deflection angle and azimuth angle of the hematoma are determined according to the angle between the eigenvector and the coordinate axis, and the volume parameters, centroid coordinates, shape parameters and azimuth parameters of the core area of ​​the hematoma are output.

[0021] In an optional embodiment,

[0022] Inputting the three-dimensional model of the hematoma core area into a deep neural network, performing layered processing on the hematoma core area based on the density gradient value, dividing the hematoma core area into multiple density sub-layers according to the multi-level three-dimensional density distribution map, calculating the volume ratio, spatial coordinates and fluidity parameters of each density sub-layer, and determining the spatial position coordinates of the optimal drainage point by using an optimization algorithm based on the volume ratio, spatial coordinates and fluidity parameters, including:

[0023] The three-dimensional model and density gradient value of the hematoma core area are input into a deep neural network, the deep neural network extracts density space features through density weighted convolution and gradient normalization, transmits feature information through density jump connection, and outputs a hierarchical prediction result of the hematoma core area;

[0024] Based on the hierarchical prediction results, the core area of ​​the hematoma is divided into multiple density sub-layers, the density values ​​of the voxels in each density sub-layer are counted, the ratio of the volume of each density sub-layer to the total volume of the hematoma is calculated to obtain the volume ratio, the density-weighted summation of the voxels in each density sub-layer is calculated to obtain the spatial centroid coordinates of the density sub-layer, the mapping relationship between density and viscosity is established according to the density value statistics results to obtain the equivalent viscosity coefficient of each density sub-layer, and the equivalent viscosity coefficient is multiplied by the velocity gradient of each density sub-layer to obtain the fluidity parameter;

[0025] The volume ratio is combined with the distance from the point to be located to the spatial coordinates of each density sublayer to construct a volume effect term, the minimum distance from the point to be located to the hematoma boundary is constructed as a distance constraint term, the fluidity parameter is combined with the angle between the density gradient direction and the spatial coordinate direction of each density sublayer pointed by the point to be located to construct a fluidity constraint term, and the volume effect term, distance constraint term and fluidity constraint term are weightedly combined to obtain the drainage point optimization objective function;

[0026] The objective function is iteratively solved by an optimization algorithm. The search direction and step size are adjusted by using the density gradient direction in each iteration, and a local search is performed on the candidate solution obtained by the current iteration. When the optimized value of the objective function meets the convergence condition, the corresponding candidate solution is used as the spatial position coordinate of the optimal drainage point.

[0027] In an optional embodiment,

[0028] Importing the angiography image and the functional area image of the patient, and generating the optimal drainage path under the constraints of the angiography image and the functional area image comprises:

[0029] Obtaining the patient's angiography image and functional area image, performing vascular enhancement processing on the angiography image to extract vascular direction characteristics, and constructing a vascular three-dimensional feature model in combination with vascular diameter information; reconstructing white matter fiber bundles based on diffusion tensor images to obtain functional area connection characteristics, and marking functional area boundaries in combination with functional magnetic resonance imaging data to obtain functional area spatial distribution;

[0030] Extracting feature points from the angiography image and the functional area image and establishing corresponding relationships, and obtaining a transformation matrix by optimizing the registration objective function; using the transformation matrix to transform the three-dimensional feature model of the blood vessel and the spatial distribution of the functional area into a hematoma model coordinate system, and establishing a unified surgical planning space;

[0031] A vascular distance field is constructed based on the vascular three-dimensional feature model and a vascular safety constraint is calculated; a gradient potential field is constructed based on the spatial distribution of the functional area to obtain the functional area protection constraint; and a path planning cost function is constructed by combining the vascular safety constraint, the functional area protection constraint and the path geometry constraint in combination with the length and curvature constraints of the drainage path;

[0032] A path search tree is constructed with the optimal drainage point as the starting point, a sampling strategy is determined according to the cost function, and the tree is expanded, and a global optimal path is obtained through path optimization and pruning;

[0033] Curvature sampling is performed on the global optimal path to obtain a key control point sequence, and a non-uniform rational parameter spline function is used to interpolate and reconstruct the control point sequence to obtain a smooth drainage trajectory; dynamic step sampling is performed along the smooth drainage trajectory, and the sampling points are substituted into the vascular safety constraints and functional area protection constraints for safety verification; the verified dangerous trajectory segment is iteratively optimized by adjusting the position of the local control points until all safety constraints are met, and the final drainage path is output.

[0034] In an optional embodiment,

[0035] Collecting pressure data and electroencephalogram data of each density sublayer, inputting the pressure data, electroencephalogram data and volume ratio into a pre-constructed neural function evaluation model to obtain a drainage rate parameter, and adjusting the drainage speed of the main drainage tube and the branch drainage tube according to the drainage rate parameter includes:

[0036] Arrange pressure sensors at multiple density sublayers to collect pressure data, wherein the pressure sensors are arranged in a distributed manner, perform median filtering and wavelet denoising on the pressure data to obtain filtered pressure data, and calculate the filtered pressure data between adjacent measuring points to obtain a pressure gradient distribution diagram;

[0037] Collecting EEG signals from the area surrounding the lesion, filtering the EEG signals to obtain filtered EEG signals, extracting time domain features, frequency domain features and nonlinear features of the filtered EEG signals, performing time-frequency analysis on the filtered EEG signals through continuous wavelet transform to obtain a time-frequency feature matrix, and constructing EEG signal quality indicators based on the time domain features, frequency domain features and nonlinear features;

[0038] Constructing a multi-layer deep neural network model, the deep neural network model includes a pressure feature extraction branch, an EEG feature extraction branch and a volume feature extraction branch, wherein the pressure feature extraction branch uses a multi-scale one-dimensional convolutional layer to extract pressure time series features, the EEG feature extraction branch uses a bidirectional long short-term memory network and a self-attention mechanism to fuse multi-lead information, and the volume feature extraction branch uses a residual fully connected layer to process volume change features;

[0039] The pressure gradient distribution map, the time-frequency feature matrix, the EEG signal quality index, and the volume ratio of each density sublayer are input into the deep neural network model, and the features of the pressure feature extraction branch, the EEG feature extraction branch, and the volume feature extraction branch are adaptively weighted combined through the feature fusion layer to output the drainage rate parameter and the drainage risk index;

[0040] According to the drainage rate parameter and the drainage risk index, a fuzzy PID control algorithm is used to calculate the drainage speed control parameter, and the drainage speeds of the main drainage tube and multiple branch drainage tubes are adjusted according to the drainage speed control parameter and the spatial coordinates of each density sublayer. The drainage speed is adjusted in real time based on the pressure gradient distribution map, the EEG signal quality index and the drainage risk index.

[0041] In an optional embodiment,

[0042] According to the drainage rate parameter and the drainage risk index, the drainage speed control parameter is calculated by using the fuzzy PID control algorithm, including:

[0043] The drainage rate parameter and the drainage risk index output by the deep neural network are weightedly combined to obtain a target drainage rate, the actual drainage rate of the drainage device is collected, and the actual drainage rate is compared with the target drainage rate to obtain a control error value;

[0044] Collecting real-time data of the control error value to calculate the control error change rate;

[0045] The control error value is fuzzy processed by using the triangular membership function to obtain the control error membership value, and the control error change rate is fuzzy processed by using the triangular membership function to obtain the error change rate membership value;

[0046] Performing fuzzy reasoning according to the control error membership value, the error change rate membership value and a preset fuzzy rule base, and using a minimum-maximum synthesis method to obtain a fuzzy reasoning result of the drainage speed control parameter;

[0047] Performing a centroid method defuzzification process on the fuzzy inference result to obtain a proportional parameter adjustment amount, an integral parameter adjustment amount and a differential parameter adjustment amount;

[0048] The proportional parameter adjustment amount is added to the initial proportional parameter to obtain the proportional control parameter, the integral parameter adjustment amount is added to the initial integral parameter to obtain the integral control parameter, the differential parameter adjustment amount is added to the initial differential parameter to obtain the differential control parameter, and the proportional control parameter, the integral control parameter and the differential control parameter are combined to obtain the drainage speed control parameter.

[0049] A second aspect of an embodiment of the present invention provides a precise drainage system for cerebral hemorrhage based on image density analysis, comprising:

[0050] The first unit is used to obtain a three-dimensional computer tomography image of a patient's head, calculate the density value and density gradient value of each pixel in the target bleeding area by using a density analysis model based on a convolutional neural network, construct an adaptive density threshold curve according to the density value and density gradient value, generate a multi-level three-dimensional density distribution map, use a regional segmentation algorithm to extract a region in the multi-level three-dimensional density distribution map with a density greater than the density threshold curve as a hematoma core area, use a three-dimensional reconstruction algorithm to establish a three-dimensional model of the hematoma core area, and calculate the volume parameters, centroid coordinates, shape parameters and azimuth parameters of the hematoma core area;

[0051] The second unit is used to input the three-dimensional model of the hematoma core area into a deep neural network, perform layered processing on the hematoma core area based on the density gradient value, divide the hematoma core area into multiple density sub-layers according to the multi-level three-dimensional density distribution map, calculate the volume ratio, spatial coordinates and fluidity parameters of each density sub-layer, determine the spatial position coordinates of the optimal drainage point by using an optimization algorithm based on the volume ratio, spatial coordinates and fluidity parameters, import the patient's angiography image and functional area image, generate an optimal drainage path under the constraints of the angiography image and the functional area image, and input the spatial position coordinates of the optimal drainage point and the optimal drainage path into an electromagnetic positioning system;

[0052] The third unit is used to perform hematoma drainage using a multi-channel intelligent drainage system including a main drainage tube and multiple branch drainage tubes under the guidance of the electromagnetic positioning system, insert the main drainage tube to the optimal drainage point, calculate the optimal drainage position of each density sublayer according to the spatial coordinates and fluidity parameters of the density sublayer, insert multiple branch drainage tubes into the optimal drainage positions respectively, collect pressure data and electroencephalogram data of each density sublayer, input the pressure data, electroencephalogram data and volume ratio into a pre-constructed neurological function evaluation model to obtain a drainage rate parameter, adjust the drainage speed of the main drainage tube and the branch drainage tube according to the drainage rate parameter, and monitor the current volume parameter of the hematoma core area and the density distribution changes of each density sublayer in real time. When the current volume parameter drops below the preset volume value and the density distribution of each density sublayer tends to be uniform, the precise drainage of the hematoma is completed.

[0053] A third aspect of the embodiments of the present invention

[0054] An electronic device is provided, comprising:

[0055] processor;

[0056] a memory for storing processor-executable instructions;

[0057] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0058] According to a fourth aspect of the embodiments of the present invention,

[0059] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0060] In this embodiment, by using deep learning and density analysis technology, the core area of ​​the hematoma can be accurately identified and three-dimensionally reconstructed to achieve accurate positioning and stratification of the hematoma. This method can accurately assess the volume, shape and distribution characteristics of the hematoma, providing an important basis for formulating a personalized drainage plan. Using a multi-channel intelligent drainage system, the optimal drainage point and path are determined according to the density distribution characteristics of the hematoma, and the sublayers of different densities are drained differentially. This precise drainage strategy can effectively avoid important blood vessels and functional areas, minimize surgical trauma, and improve drainage efficiency and safety. By real-time monitoring of physiological parameters such as pressure and EEG during the drainage process, and combining the neurological function evaluation model to dynamically adjust the drainage rate. This closed-loop feedback mechanism can achieve precise control of the drainage process, effectively prevent complications such as rebleeding, and significantly improve patient prognosis. Overall, this method provides a new technical means for the precise diagnosis and treatment of cerebral hemorrhage, and has important clinical application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a schematic diagram of the process of a precise drainage method for cerebral hemorrhage based on image density analysis according to an embodiment of the present invention;

[0062] Figure 2 It is a schematic diagram of the structure of a precise drainage system for cerebral hemorrhage based on image density analysis according to an embodiment of the present invention. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0064] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0065] Figure 1 FIG. 1 is a flow chart of a method for accurate drainage of cerebral hemorrhage based on image density analysis according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0066] S101. Obtain a three-dimensional computer tomography image of the patient's head, calculate the density value and density gradient value of each pixel in the target bleeding area using a density analysis model based on a convolutional neural network, construct an adaptive density threshold curve according to the density value and density gradient value, generate a multi-level three-dimensional density distribution map, use a regional segmentation algorithm to extract the area in the multi-level three-dimensional density distribution map with a density greater than the density threshold curve as the hematoma core area, use a three-dimensional reconstruction algorithm to establish a three-dimensional model of the hematoma core area, and calculate the volume parameters, centroid coordinates, shape parameters and azimuth parameters of the hematoma core area;

[0067] S102. Input the three-dimensional model of the hematoma core area into a deep neural network, perform layered processing on the hematoma core area based on the density gradient value, divide the hematoma core area into multiple density sub-layers according to the multi-level three-dimensional density distribution map, calculate the volume ratio, spatial coordinates and fluidity parameters of each density sub-layer, determine the spatial position coordinates of the optimal drainage point using an optimization algorithm based on the volume ratio, spatial coordinates and fluidity parameters, import the patient's angiography image and functional area image, generate an optimal drainage path under the constraints of the angiography image and the functional area image, and input the spatial position coordinates of the optimal drainage point and the optimal drainage path into the electromagnetic positioning system;

[0068] S103. Under the guidance of the electromagnetic positioning system, a multi-channel intelligent drainage system including a main drainage tube and multiple branch drainage tubes is used to drain the hematoma, the main drainage tube is inserted to the optimal drainage point, the optimal drainage position of each density sublayer is calculated according to the spatial coordinates and fluidity parameters of the density sublayer, and multiple branch drainage tubes are respectively inserted into the optimal drainage positions, and the pressure data and electroencephalogram data of each density sublayer are collected. The pressure data, electroencephalogram data and volume ratio are input into a pre-constructed neurological function evaluation model to obtain a drainage rate parameter, and the drainage speed of the main drainage tube and the branch drainage tube is adjusted according to the drainage rate parameter, and the current volume parameter of the hematoma core area and the density distribution changes of each density sublayer are monitored in real time. When the current volume parameter drops below the preset volume value and the density distribution of each density sublayer tends to be uniform, the precise drainage of the hematoma is completed.

[0069] In an optional embodiment, a three-dimensional computerized tomography image of a patient's head is obtained, a density analysis model based on a convolutional neural network is used to calculate the density value and density gradient value of each pixel in the target bleeding area, and an adaptive density threshold curve is constructed according to the density value and density gradient value to generate a multi-level three-dimensional density distribution map, including:

[0070] Acquire a three-dimensional computer tomography image of a patient's head, divide the three-dimensional computer tomography image into a plurality of overlapping sub-blocks, calculate a three-dimensional grayscale histogram for each overlapping sub-block, perform cropping and restriction on the three-dimensional grayscale histogram to obtain a cumulative distribution function, perform mapping enhancement on the three-dimensional computer tomography image according to the cumulative distribution function to obtain an enhanced image, perform denoising on the enhanced image by using anisotropic diffusion filtering, determine a diffusion coefficient by analyzing a gradient histogram of the enhanced image, and iteratively process the enhanced image according to the diffusion coefficient to obtain a preprocessed three-dimensional computer tomography image;

[0071] Inputting the preprocessed three-dimensional computed tomography image into a density analysis model, the density analysis model comprising a feature extraction subnetwork and a density calculation subnetwork, the feature extraction subnetwork comprising a plurality of three-dimensional convolution units, each of which is provided with a convolution layer with a different expansion coefficient and a three-dimensional attention calculation module, the density calculation subnetwork comprising a plurality of deconvolution layers, a feature fusion connection structure is provided between the feature extraction subnetwork and the density calculation subnetwork, the density analysis model is trained using a combined loss function including a density information constraint term to obtain a feature map of a target bleeding area, the density value of each pixel point in the target bleeding area is calculated based on the feature map, and a density value distribution map is generated;

[0072] A three-dimensional gradient operator is used to calculate the density value distribution map to obtain the density gradient value of each pixel in the target bleeding area, a two-dimensional distribution histogram is established by the density value in the density value distribution map and the density gradient value, a joint probability distribution of the density value and the density gradient value in the two-dimensional distribution histogram is calculated, the joint probability distribution is divided into multiple candidate categories, the intra-class variance and the inter-class variance of each candidate category are calculated, an optimal segmentation threshold combination that maximizes the inter-class variance and minimizes the intra-class variance is found through an iterative optimization method, a control parameter of an optimal segmentation curve is constructed according to the optimal segmentation threshold combination, a curve interpolation is performed on the optimal segmentation curve according to the control parameter to obtain an adaptive density threshold curve, the density value distribution map and the adaptive density threshold curve are input into a density stratification algorithm, the target bleeding area is density stratified, and a multi-level three-dimensional density distribution map is generated.

[0073] For example, first, a three-dimensional computed tomography image of a patient's head is obtained. The obtained three-dimensional image is divided into a plurality of overlapping sub-blocks, each of which is 64×64×64 pixels in size, and the overlap rate between adjacent sub-blocks is 50%. A three-dimensional grayscale histogram is calculated for each overlapping sub-block, and the distribution of grayscale values ​​in the range of 0-255 is counted.

[0074] Next, the three-dimensional grayscale histogram is clipped and restricted. The clipping threshold is set to 1% of the maximum value of the histogram, and the frequency of grayscale values ​​exceeding the threshold is limited to within the threshold. The cumulative distribution function is calculated based on the clipped histogram, which reflects the proportion of pixels with grayscale values ​​less than or equal to a certain value to the total pixels.

[0075] Then, the original 3D CT image is mapped and enhanced according to the cumulative distribution function. The grayscale value in the original image is mapped to a new grayscale value range through the cumulative distribution function, thereby enhancing the image contrast. For example, if the pixel point with a grayscale value of 100 in the original image has a corresponding cumulative distribution function value of 0.4, the grayscale value of the pixel point is mapped to 102 (0.4×255).

[0076] The enhanced image is subjected to denoising by anisotropic diffusion filtering. First, the gradient histogram of the enhanced image is analyzed to determine the diffusion coefficient. The gradient histogram reflects the distribution of edge and texture information in the image. According to the peak value and variance of the gradient histogram, the initial value and range of the diffusion coefficient are set. For example, the initial diffusion coefficient is set to 0.2, and is reduced by 0.01 each iteration until it drops to 0.05.

[0077] The enhanced image is iteratively processed according to the determined diffusion coefficient. In each iteration, the gradient value of each pixel of the image is calculated, and the pixel gray value is adjusted according to the gradient value and the diffusion coefficient. The number of iterations is set to 20 times. After the noise reduction process, the preprocessed three-dimensional computed tomography image is obtained.

[0078] The preprocessed 3D image is input into the density analysis model. The model includes a feature extraction subnetwork and a density calculation subnetwork. The feature extraction subnetwork consists of 5 3D convolution units, each of which contains 3 convolution layers and 1 3D attention calculation module. The expansion coefficients of the convolution layers are 1, 2, and 4, respectively, and the convolution kernel size is 3×3×3. The 3D attention calculation module is used to highlight important features and suppress irrelevant information.

[0079] The density calculation subnetwork contains four deconvolution layers, which are used to upsample the feature map to restore the original image size. A feature fusion connection structure is set between the feature extraction subnetwork and the density calculation subnetwork to fuse features of different scales. The density analysis model is trained using a combined loss function containing density information constraints, which comprehensively considers the density estimation error and the consistency of density distribution.

[0080] After training is completed, the preprocessed image is input into the model to obtain the feature map of the target bleeding area. Based on the feature map, the density value of each pixel in the target area is calculated to generate a density value distribution map. The density value range is usually between 0 and 1, reflecting the severity of the bleeding.

[0081] The density value distribution map is gradient-calculated using a three-dimensional Sobel gradient operator to obtain the density gradient value of each pixel in the target bleeding area. The density gradient value reflects the rate and direction of change of the density value, which helps to identify the bleeding boundary.

[0082] The density values ​​and density gradient values ​​in the density value distribution map are used to establish a two-dimensional distribution histogram. The x-axis of the histogram represents the density value, the y-axis represents the density gradient value, and the value of the rectangular area in each two-dimensional distribution histogram represents the frequency of the corresponding density value and density gradient value combination. Calculate the joint probability distribution of the density value and density gradient value in the two-dimensional distribution histogram.

[0083] The joint probability distribution is divided into multiple candidate categories, and the initial number of categories is set to 5. The intra-class variance and inter-class variance of each candidate category are calculated. The intra-class variance reflects the uniformity of the density distribution within the category, and the inter-class variance reflects the discrimination between different categories.

[0084] The optimal segmentation threshold combination is found through iterative optimization. In each iteration, the number of categories and category boundaries are adjusted, and the intra-class variance and inter-class variance are recalculated. The segmentation scheme that maximizes the inter-class variance and minimizes the intra-class variance is selected as the optimal scheme. The number of iterations is set to 100.

[0085] The control parameters of the optimal segmentation curve are constructed according to the optimal segmentation threshold combination. The control parameters include the coordinates of the starting point, the end point and the intermediate control point of the curve. According to the control parameters, the optimal segmentation curve is interpolated with cubic spline to obtain a smooth adaptive density threshold curve.

[0086] The density value distribution map and the adaptive density threshold curve are input into the density stratification algorithm. The algorithm divides the target bleeding area into multiple density levels according to the threshold curve. The specific steps are as follows: First, each pixel point in the density value distribution map is compared with the threshold curve to determine the density level to which it belongs. Then, the connectivity analysis of adjacent pixels is performed to merge the connected areas belonging to the same density level. Finally, a multi-level three-dimensional density distribution map is generated, and different colors represent different density levels.

[0087] In this embodiment, by preprocessing the three-dimensional computed tomography image, the image quality and contrast are improved, providing a good data basis for subsequent density analysis. The anisotropic diffusion filter used in the preprocessing process can effectively reduce noise while retaining important edge and structural information in the image. The density analysis model based on the convolutional neural network can automatically extract the deep features of the bleeding area and realize the accurate estimation of the density value and density gradient value. The three-dimensional convolution unit and attention mechanism used in the model effectively capture the three-dimensional spatial information and improve the accuracy of density analysis. The feature fusion connection structure realizes the effective combination of multi-scale features and enhances the expression ability of the model. The construction method of the adaptive density threshold curve takes into account the joint distribution characteristics of the density value and the density gradient value, and can automatically adjust the segmentation threshold according to the specific image. The multi-level three-dimensional density distribution map intuitively shows the density distribution of the bleeding area, which helps doctors to make accurate diagnosis and formulate treatment plans. This method has good adaptability and robustness and can be applied to cases of head hemorrhage of different types and degrees.

[0088] In an optional embodiment, a region segmentation algorithm is used to extract a region in the multi-level three-dimensional density distribution map where the density is greater than the density threshold curve as the hematoma core region, a three-dimensional reconstruction algorithm is used to establish a three-dimensional model of the hematoma core region, and the volume parameters, centroid coordinates, shape parameters, and azimuth parameters of the hematoma core region are calculated, including:

[0089] A regional segmentation algorithm is used to process the multi-level three-dimensional density distribution map, an initial seed point set is determined according to a density threshold curve, a point in the initial seed point set whose density value is greater than the threshold corresponding to the density threshold curve is used as a growth starting point, a directional density gradient value constraint condition is introduced in the regional growth process, the density value and density gradient direction of the candidate voxel are judged, the candidate voxel whose density value is greater than the density threshold and whose density gradient direction is consistent with the growth direction is included in the growth region, and the growth region is morphologically processed to obtain a segmentation boundary point set of the hematoma core area;

[0090] A three-dimensional reconstruction model is established according to the segmentation boundary point set of the hematoma core area, the density gradient value of each point in the segmentation boundary point set is calculated, the grid size of different areas is set based on the density gradient value, the space is gridded by an octree structure, a distance-based weight coefficient is calculated for each grid node, a local weighted least squares equation group is constructed and solved to obtain the implicit surface function value of the grid node, an isosurface point set with zero implicit surface function value is extracted to perform triangular meshing, and a three-dimensional model of the hematoma core area is obtained;

[0091] The morphological parameters of the core area of ​​the hematoma are calculated using the three-dimensional model, the volume value of the hematoma is calculated according to the grid vertex coordinates and vertex normal vectors of the three-dimensional model, the density values ​​in the multi-level three-dimensional density distribution map are correspondingly aligned with the voxel position information in the three-dimensional model, the centroid coordinates of the hematoma are calculated according to the aligned density values ​​and voxel positions, the inertia matrix of the hematoma area is constructed, the eigenvalues ​​of the inertia matrix are calculated, the flatness coefficient, elongation coefficient and sphericity coefficient of the hematoma are calculated according to the eigenvalues, the eigenvector of the inertia matrix is ​​calculated, the spatial deflection angle and azimuth angle of the hematoma are determined according to the angle between the eigenvector and the coordinate axis, and the volume parameters, centroid coordinates, shape parameters and azimuth parameters of the core area of ​​the hematoma are output.

[0092] Exemplarily, the multi-level three-dimensional density distribution map is first preprocessed, including image denoising, contrast enhancement, etc. The initial seed point set is determined according to the density threshold curve, and the point with a density value greater than the threshold is selected as the starting point of growth. Directional density gradient value constraints are introduced in the regional growth process to judge the density value and density gradient direction of the candidate voxel. When the density value is greater than the threshold and the gradient direction is consistent with the growth direction, the voxel is included in the growth region. For example, the density threshold can be set to 40HU, and the gradient direction deviation is less than 30°, which is considered consistent. The growth area is morphologically processed, such as opening operations to remove small noise points and closing operations to fill small holes, to obtain the segmentation boundary point set of the hematoma core area.

[0093] A three-dimensional reconstruction model is established based on the segmented boundary point set. The density gradient value of each point in the boundary point set is calculated, and the grid size of different areas is set according to the size of the gradient value. The larger the gradient value, the smaller the grid size. The octree structure is used to adaptively mesh the space. The distance-based weight coefficient is calculated for each grid node, and the weighted least squares equation set is constructed and solved to obtain the implicit surface function value of the grid node. The isosurface point set with zero function value is extracted, and the triangular mesh is divided using the walking cube algorithm to generate a three-dimensional model of the hematoma core area.

[0094] The morphological parameters of the core area of ​​the hematoma are calculated using the three-dimensional model. The Gaussian theorem is used to calculate the volume of the hematoma based on the mesh vertex coordinates and normal vectors of the model. The density values ​​in the multi-level density distribution map are aligned with the voxel position information in the three-dimensional model, and the coordinates of the centroid of the hematoma are calculated based on the aligned density values ​​and voxel positions. The inertia matrix of the hematoma area is constructed, the eigenvalues ​​of the matrix are calculated, and the flatness, elongation, and sphericity coefficients are calculated based on the eigenvalues. The eigenvector of the inertia matrix is ​​calculated, and the spatial deflection angle and azimuth angle of the hematoma are determined based on the angle between the eigenvector and the coordinate axis. Finally, the volume, centroid coordinates, shape parameters, and azimuth parameters of the core area of ​​the hematoma are output.

[0095] For example, a set of CT image data is processed, the density threshold is set to 40HU, and the extracted boundary point set of the hematoma core area contains a total of 5000 points. The reconstructed three-dimensional model contains 10,000 triangular facets. The calculated hematoma volume is 50mL, the centroid coordinates are (100, 80, 60) mm, the flatness coefficient is 0.8, the elongation coefficient is 1.2, the sphericity coefficient is 0.9, the spatial deflection angle is 30°, and the orientation angle is 45°.

[0096] In this embodiment, by introducing the directional density gradient value constraint, the accuracy of the hematoma core area extraction is improved, and the leakage problem in the regional growth process is avoided. The adaptive meshing and weighted least squares fitting are used to improve the accuracy of the three-dimensional reconstruction model, which can better show the detailed characteristics of the hematoma. By calculating a variety of morphological parameters, the three-dimensional morphological characteristics of the hematoma core area are fully characterized, providing an important reference for clinical diagnosis and surgical planning.

[0097] In an optional embodiment, the three-dimensional model of the hematoma core area is input into a deep neural network, the hematoma core area is layered based on the density gradient value, the hematoma core area is divided into a plurality of density sublayers according to the multi-level three-dimensional density distribution map, the volume ratio, spatial coordinates and fluidity parameters of each density sublayer are calculated, and the spatial position coordinates of the optimal drainage point are determined by an optimization algorithm based on the volume ratio, spatial coordinates and fluidity parameters, including:

[0098] The three-dimensional model and density gradient value of the hematoma core area are input into a deep neural network, the deep neural network extracts density space features through density weighted convolution and gradient normalization, transmits feature information through density jump connection, and outputs a hierarchical prediction result of the hematoma core area;

[0099] Based on the hierarchical prediction results, the core area of ​​the hematoma is divided into multiple density sub-layers, the density values ​​of the voxels in each density sub-layer are counted, the ratio of the volume of each density sub-layer to the total volume of the hematoma is calculated to obtain the volume ratio, the density-weighted summation of the voxels in each density sub-layer is calculated to obtain the spatial centroid coordinates of the density sub-layer, the mapping relationship between density and viscosity is established according to the density value statistics results to obtain the equivalent viscosity coefficient of each density sub-layer, and the equivalent viscosity coefficient is multiplied by the velocity gradient of each density sub-layer to obtain the fluidity parameter;

[0100] The volume ratio is combined with the distance from the point to be located to the spatial coordinates of each density sublayer to construct a volume effect term, the minimum distance from the point to be located to the hematoma boundary is constructed as a distance constraint term, the fluidity parameter is combined with the angle between the density gradient direction and the spatial coordinate direction of each density sublayer pointed by the point to be located to construct a fluidity constraint term, and the volume effect term, distance constraint term and fluidity constraint term are weightedly combined to obtain the drainage point optimization objective function;

[0101] The objective function is iteratively solved by an optimization algorithm. The search direction and step size are adjusted by using the density gradient direction in each iteration, and a local search is performed on the candidate solution obtained by the current iteration. When the optimized value of the objective function meets the convergence condition, the corresponding candidate solution is used as the spatial position coordinate of the optimal drainage point.

[0102] Exemplarily, first, the deep neural network adopts the U-Net architecture, which includes two parts: encoder and decoder. The encoder consists of multiple convolution blocks, each of which contains two layers of 3D convolution layers, batch normalization layers, and ReLU activation functions. During the convolution operation, the convolution kernel is weighted according to the input density gradient value, so that the network can better capture the density distribution characteristics. At the same time, maximum pooling is used for downsampling between each convolution block to gradually extract higher-level features. The decoder uses transposed convolution for upsampling, and splices the feature map of the corresponding encoder layer with the upsampling result through jump connections to restore the lost spatial information. Finally, the probability of each voxel belonging to a different density layer is output through a 1x1x1 convolution layer.

[0103] When training the network, the cross entropy loss function is used, and the Adam optimizer is used to update the parameters. In order to enhance the generalization ability of the model, the input data is subjected to data enhancement operations such as random rotation and translation during the training process. After the training is completed, the three-dimensional model of the hematoma core area is input into the trained network to obtain the hierarchical prediction results.

[0104] Based on the hierarchical prediction results, the core area of ​​the hematoma is divided into multiple density sublayers. Density sublayers refer to different density levels formed by dividing the core area of ​​the hematoma according to the density gradient value. The density values ​​of the voxels in each density sublayer are counted, and the ratio of the volume of each density sublayer to the total volume of the hematoma is calculated to obtain the volume ratio. For example, a hematoma is divided into 5 density sublayers, and the volume ratios of each layer are 0.15, 0.25, 0.3, 0.2, and 0.1 respectively.

[0105] The density weighted sum of the voxels in each density sublayer is calculated to obtain the spatial centroid coordinates of the density sublayer. Specifically, the density value of each voxel is first normalized to the range of 0-1, and then the normalized density value is used as a weight, multiplied with the spatial coordinates of the voxel and summed, and finally divided by the weight sum to obtain the weighted average coordinates. For example, the spatial centroid coordinates of a density sublayer are (25.3, 18.7, 42.1) mm.

[0106] Based on the statistical results of density values, the mapping relationship between density and viscosity is established to obtain the equivalent viscosity coefficient of each density sublayer. The mapping relationship can be established by polynomial fitting. The viscosity data of blood of different densities are obtained by consulting relevant literature, and the fitting coefficient is determined by the least squares method. For example, the equivalent viscosity coefficient of a certain density sublayer is 3.5mPa·s.

[0107] The equivalent viscosity coefficient is multiplied by the velocity gradient of each density sublayer to obtain the fluidity parameter. The velocity gradient can be estimated approximately by the density gradient, that is, it is assumed that the flow velocity in the area with higher density is slower and the flow velocity in the area with lower density is faster. Specifically, the modulus of the density gradient is first calculated, then normalized to the range of 0-1, and finally the approximate value of the velocity gradient is obtained by subtracting the normalized result from 1. The fluidity parameter can be obtained by multiplying the velocity gradient by the equivalent viscosity coefficient. For example, the fluidity parameter of a density sublayer is 2.8.

[0108] Next, the drainage point optimization objective function is constructed. First, the volume ratio is combined with the distance from the point to be located to the spatial coordinates of each density sublayer to construct the volume effect term. Specifically, the Euclidean distance from the point to be located to the center of mass of each density sublayer is calculated, and the inverse of the distance is multiplied by the corresponding volume ratio and summed to obtain the volume effect term. The physical meaning of this term is to make the drainage point as close as possible to the density sublayer with a larger volume.

[0109] Then the minimum distance from the point to be located to the hematoma boundary is constructed as a distance constraint. By traversing all points on the hematoma boundary, the distance from the point to be located to these points is calculated, and the minimum value is taken as the distance constraint. The physical meaning of this term is to keep a certain distance between the drainage point and the hematoma boundary to prevent the drainage tube from puncturing the blood vessel.

[0110] Finally, the mobility parameter is combined with the angle between the density gradient direction and the coordinate direction of the point to be located in each density sublayer space to construct the mobility constraint term. Specifically, the cosine value of the angle between the density gradient direction and the direction of the center of mass of each density sublayer space is calculated, and the cosine value is multiplied and summed with the corresponding mobility parameter to obtain the mobility constraint term. The physical meaning of this term is to make the drainage point position conducive to the discharge of hematoma.

[0111] The objective function for optimizing drainage points is obtained by weighted combination of volume effect, distance constraint and mobility constraint. The weight coefficient can be adjusted according to specific needs, for example, it can be set to 0.5, 0.3, or 0.2. The mathematical expression of the objective function is: f(x, y, z) = 0.5 × V(x, y, z) + 0.3 × D(x, y, z) + 0.2 × F(x, y, z), where V, D, and F represent volume effect, distance constraint, and mobility constraint, respectively, and x, y, and z are the spatial coordinates of the point to be located.

[0112] The objective function is iteratively solved using an optimization algorithm. This embodiment uses a particle swarm optimization algorithm (PSO) to solve. First, a certain number of particles are randomly initialized, each particle represents a candidate drainage point location. In each iteration, the search direction and step size are adjusted using the density gradient direction. Specifically, the density gradient direction is used as a component of the particle velocity update, so that the particles tend to move along the density gradient direction.

[0113] At the same time, a local search is performed on the candidate solutions obtained in the current iteration. A simulated annealing strategy is used to accept slightly worse solutions with a certain probability to escape the local optimum. Specifically, several new solutions are randomly generated around the candidate solution. If the new solution is better than the current solution, it is directly accepted. Otherwise, the new solution is accepted with a certain probability. The probability decreases with the increase in the number of iterations, so that the algorithm gradually converges to the optimal solution in the later stage.

[0114] When the optimized value of the objective function meets the convergence condition, the corresponding candidate solution is used as the spatial position coordinate of the optimal drainage point. The convergence condition can be set as the change of the objective function value for several consecutive iterations is less than a certain threshold, such as the relative change rate of the objective function value for 10 consecutive iterations is less than 0.1%. The optimal drainage point spatial position coordinates outputted in the end are the best puncture point for hematoma drainage.

[0115] In this embodiment, the accurate stratification of hematoma is achieved through deep learning, which overcomes the problem that traditional methods are difficult to handle complex density distribution, and improves the accuracy and robustness of stratification. A multi-objective optimization strategy is adopted, and factors such as volume effect, safety distance and fluidity are comprehensively considered, so that the determined drainage point position is more reasonable, which is conducive to the rapid discharge of hematoma and reduces the risk of complications. The introduction of density gradient information to guide the optimization process improves the convergence speed and global search ability of the algorithm, avoids falling into the local optimal solution, and thus obtains a better drainage point position.

[0116] In an optional implementation, importing the angiography image and the functional area image of the patient, and generating the optimal drainage path under the constraints of the angiography image and the functional area image includes:

[0117] Obtaining the patient's angiography image and functional area image, performing vascular enhancement processing on the angiography image to extract vascular direction characteristics, and constructing a vascular three-dimensional feature model in combination with vascular diameter information; reconstructing white matter fiber bundles based on diffusion tensor images to obtain functional area connection characteristics, and marking functional area boundaries in combination with functional magnetic resonance imaging data to obtain functional area spatial distribution;

[0118] Extracting feature points from the angiography image and the functional area image and establishing corresponding relationships, and obtaining a transformation matrix by optimizing the registration objective function; using the transformation matrix to transform the three-dimensional feature model of the blood vessel and the spatial distribution of the functional area into a hematoma model coordinate system, and establishing a unified surgical planning space;

[0119] A vascular distance field is constructed based on the vascular three-dimensional feature model and a vascular safety constraint is calculated; a gradient potential field is constructed based on the spatial distribution of the functional area to obtain the functional area protection constraint; and a path planning cost function is constructed by combining the vascular safety constraint, the functional area protection constraint and the path geometry constraint in combination with the length and curvature constraints of the drainage path;

[0120] A path search tree is constructed with the optimal drainage point as the starting point, a sampling strategy is determined according to the cost function, and the tree is expanded, and a global optimal path is obtained through path optimization and pruning;

[0121] Curvature sampling is performed on the global optimal path to obtain a key control point sequence, and a non-uniform rational parameter spline function is used to interpolate and reconstruct the control point sequence to obtain a smooth drainage trajectory; dynamic step sampling is performed along the smooth drainage trajectory, and the sampling points are substituted into the vascular safety constraints and functional area protection constraints for safety verification; the verified dangerous trajectory segment is iteratively optimized by adjusting the position of the local control points until all safety constraints are met, and the final drainage path is output.

[0122] Exemplarily, first, the patient's angiography image and functional area image are obtained. The angiography image is subjected to vascular enhancement processing to extract the characteristics of the vascular direction. In this step, the original image can be processed using a Frangi filter to enhance the visualization effect of the vascular structure. Then, a 3D feature model of the blood vessel is constructed in combination with the vascular diameter information. The vascular diameter information can be obtained through centerline extraction and cross-sectional analysis. Based on this, a 3D reconstruction algorithm such as the Marching Cubes algorithm is used to construct a 3D mesh model of the blood vessel.

[0123] Next, the white matter fiber bundles are reconstructed based on the diffusion tensor image to obtain the functional area connection characteristics. This step can use a deterministic fiber tracking algorithm, such as the FACT (Fiber Assignment by Continuous Tracking) method. At the same time, the functional area boundaries are marked in combination with functional magnetic resonance imaging data to obtain the spatial distribution of the functional area. The functional area boundaries can be obtained through image segmentation algorithms such as threshold segmentation and region growing.

[0124] Then, feature points are extracted from the angiography image and the functional area image and the corresponding relationship is established. The SIFT (Scale-Invariant Feature Transform) algorithm can be used for feature point extraction, and the RANSAC (Random Sample Consensus) algorithm can be used for corresponding relationship establishment. The transformation matrix is ​​obtained by optimizing the registration objective function. Here, mutual information maximization can be used as the objective function, and the Powell optimization algorithm is used to solve the optimal transformation parameters. The obtained transformation matrix is ​​used to transform the vascular three-dimensional feature model and the functional area spatial distribution to the hematoma model coordinate system to establish a unified surgical planning space.

[0125] Next, the vascular distance field is constructed based on the vascular three-dimensional feature model and the vascular safety constraints are calculated. The distance field can be calculated by the Fast Marching Method. The gradient potential field is constructed based on the spatial distribution of the functional area to obtain the functional area protection constraints. The gradient potential field can be obtained by Gaussian filtering and gradient calculation. Combined with the length and curvature constraints of the drainage path, the vascular safety constraints, functional area protection constraints and path geometry constraints are combined to construct the path planning cost function.

[0126] The path search tree is constructed with the optimal drainage point as the starting point. The optimal drainage point can be determined by analyzing the shape and position of the hematoma. The sampling strategy is determined according to the cost function and the tree is expanded. Here, the Rapidly-exploring Random Tree Star algorithm can be used. The global optimal path is obtained through path optimization and pruning.

[0127] Finally, the curvature sampling of the global optimal path is performed to obtain a sequence of key control points. The sampling interval can be dynamically adjusted according to the path curvature, and the sampling density is higher where the curvature is larger. The control point sequence is interpolated and reconstructed using a non-uniform rational parameter spline function to obtain a smooth drainage trajectory. Dynamic step sampling is performed along the smooth drainage trajectory, and the sampling points are substituted into the vascular safety constraints and functional area protection constraints for safety verification. The verified dangerous trajectory segment is iteratively optimized by adjusting the position of the local control points until all safety constraints are met, and the final drainage path is output.

[0128] Assume that the resolution of a patient's angiography image is 512×512×256, and the voxel size is 0.5mm×0.5mm×1mm. The resolution of the functional area image is 256×256×128, and the voxel size is 1mm×1mm×1mm. After vascular enhancement, 30 main vascular branches are extracted. 500 fiber bundles are reconstructed from white matter fiber bundles. 10 key functional areas are marked by functional area boundaries. 200 pairs of corresponding points are obtained by feature point extraction and matching. The transformation matrix optimization is iterated 50 times to converge. The resolution of the vascular distance field and the functional area gradient potential field is 256×256×256. The path search tree contains 10,000 nodes. The final smooth drainage trajectory is 80 mm long and contains 40 key control points.

[0129] In this embodiment, the optimal drainage path is automatically planned by comprehensively utilizing the angiography image and functional area image information, thereby improving the accuracy and safety of surgical planning. By constructing a unified surgical planning space, multiple constraints such as vascular safety and functional area protection are integrated into the path optimization process, effectively avoiding damage to important blood vessels and functional areas. The use of non-uniform rational parameter spline function reconstruction and local optimization strategies ensures the smoothness and safety of the generated path, thereby improving the feasibility of surgical operations.

[0130] In an optional embodiment, collecting pressure data and electroencephalogram data of each density sublayer, inputting the pressure data, electroencephalogram data and volume ratio into a pre-constructed neural function evaluation model to obtain a drainage rate parameter, and adjusting the drainage speed of the main drainage tube and the branch drainage tube according to the drainage rate parameter includes:

[0131] Arrange pressure sensors at multiple density sublayers to collect pressure data, wherein the pressure sensors are arranged in a distributed manner, perform median filtering and wavelet denoising on the pressure data to obtain filtered pressure data, and calculate the filtered pressure data between adjacent measuring points to obtain a pressure gradient distribution diagram;

[0132] Collecting EEG signals from the area surrounding the lesion, filtering the EEG signals to obtain filtered EEG signals, extracting time domain features, frequency domain features and nonlinear features of the filtered EEG signals, performing time-frequency analysis on the filtered EEG signals through continuous wavelet transform to obtain a time-frequency feature matrix, and constructing EEG signal quality indicators based on the time domain features, frequency domain features and nonlinear features;

[0133] Constructing a multi-layer deep neural network model, the deep neural network model includes a pressure feature extraction branch, an EEG feature extraction branch and a volume feature extraction branch, wherein the pressure feature extraction branch uses a multi-scale one-dimensional convolutional layer to extract pressure time series features, the EEG feature extraction branch uses a bidirectional long short-term memory network and a self-attention mechanism to fuse multi-lead information, and the volume feature extraction branch uses a residual fully connected layer to process volume change features;

[0134] The pressure gradient distribution map, the time-frequency feature matrix, the EEG signal quality index, and the volume ratio of each density sublayer are input into the deep neural network model, and the features of the pressure feature extraction branch, the EEG feature extraction branch, and the volume feature extraction branch are adaptively weighted combined through the feature fusion layer to output the drainage rate parameter and the drainage risk index;

[0135] According to the drainage rate parameter and the drainage risk index, a fuzzy PID control algorithm is used to calculate the drainage speed control parameter, and the drainage speeds of the main drainage tube and multiple branch drainage tubes are adjusted according to the drainage speed control parameter and the spatial coordinates of each density sublayer. The drainage speed is adjusted in real time based on the pressure gradient distribution map, the EEG signal quality index and the drainage risk index.

[0136] Exemplarily, first, pressure sensors are arranged in multiple density sublayers to collect pressure data. The pressure sensors are arranged in a distributed manner, and multiple pressure sensors are arranged at different positions in each density sublayer to obtain the global pressure distribution information of each sublayer. The collected pressure data is subjected to median filtering and wavelet denoising to remove outliers and high-frequency noise to obtain filtered pressure data. Then, the filtered pressure data between adjacent measuring points is calculated to obtain a pressure gradient distribution diagram to reflect the pressure changes inside each sublayer.

[0137] Next, collect EEG signals from the area surrounding the lesion. Use multi-lead EEG electrodes to collect EEG signals from multiple brain regions, and perform 50Hz power frequency filtering and 0.5-30Hz bandpass filtering on the collected EEG signals to obtain filtered EEG signals. Extract time domain features (such as average amplitude, variance, etc.), frequency domain features (such as power spectral density, relative power, etc.) and nonlinear features (such as approximate entropy, Lyapunov exponent, etc.) from the filtered EEG signals. Perform time-frequency analysis on the filtered EEG signals through continuous wavelet transform to obtain a time-frequency feature matrix reflecting the time-frequency characteristics of the signal. Based on the extracted time domain features, frequency domain features and nonlinear features, construct a comprehensive index reflecting the quality of EEG signals.

[0138] Then, a multi-layer deep neural network model is constructed, including pressure feature extraction branch, EEG feature extraction branch and volume feature extraction branch. The pressure feature extraction branch uses a multi-scale one-dimensional convolution layer and uses different convolution kernel sizes to extract local and global features of pressure time series data. The EEG feature extraction branch uses a bidirectional long short-term memory network to capture the long-term dependency of EEG signals and combines the self-attention mechanism to fuse multi-lead information. The volume feature extraction branch uses a residual fully connected layer to process the volume change characteristics of each density sublayer.

[0139] The pressure gradient distribution map, EEG time-frequency feature matrix, EEG signal quality index, volume ratio of each density sublayer, spatial coordinates and fluidity parameters are input into the constructed deep neural network model. The features extracted by the three branches are adaptively weighted and combined through the feature fusion layer, and the pressure, EEG and volume information are comprehensively considered to output the drainage rate parameters and drainage risk indicators.

[0140] Finally, according to the output drainage rate parameters and drainage risk indicators, the fuzzy PID control algorithm is used to calculate the drainage speed control parameters. Based on the drainage speed control parameters and the spatial coordinate information of each density sublayer, the drainage speed of the main drainage tube and multiple branch drainage tubes is adjusted. At the same time, according to the real-time monitored pressure gradient distribution map, EEG signal quality indicators and drainage risk indicators, the drainage speed is dynamically adjusted to ensure the safety and effectiveness of the drainage process.

[0141] Optionally, the pressure sensors are arranged in 6 density sublayers, with 8 measuring points arranged in each layer. The collected raw pressure data is subjected to median filtering (window size 5) and db4 wavelet 3-layer decomposition and denoising to obtain filtered pressure data. The pressure difference between adjacent measuring points is calculated, and a pressure gradient distribution thermodynamic map is drawn.

[0142] 16-lead EEG signals were collected with a sampling rate of 250Hz. The original EEG signals were notched at 50Hz and band-pass filtered at 0.5-30Hz. Time domain features (10 features such as mean, variance, peak-to-peak value), frequency domain features (8 features such as relative power of δ, θ, α, and β bands) and nonlinear features (6 features such as approximate entropy and correlation dimension) were extracted. Morlet wavelet was used for continuous wavelet transform to obtain a 16×250×5 time-frequency feature matrix. Based on the 24 extracted features, principal component analysis was used to construct EEG signal quality indicators.

[0143] The stress feature extraction branch of the deep neural network model uses a one-dimensional convolutional layer with kernel_size of 3, 5, and 7. The EEG feature extraction branch uses a 2-layer bidirectional LSTM with 128 hidden units. The volume feature extraction branch uses a 3-layer residual fully connected layer with 64 neurons in each layer. The feature fusion layer uses an attention mechanism to adaptively allocate the weights of the three branches.

[0144] The fuzzy rule base of the fuzzy PID control algorithm contains 49 rules, the input variables are the pressure error e and the pressure change rate ec, and the output variables are the PID parameter increments ΔKp, ΔKi, and ΔKd. According to the calculated PID parameters, combined with the spatial position information of each density sublayer, the drainage speed of the 7 drainage pipes (1 main pipe and 6 branch pipes) is adjusted.

[0145] In this embodiment, multi-density sublayer distributed pressure monitoring and multi-lead EEG signal analysis are used to comprehensively evaluate the cerebrospinal fluid dynamics and neurological function status, providing a reliable basis for precise drainage. The deep learning model is used to integrate pressure, EEG and volume characteristics, which realizes intelligent analysis of complex cerebrospinal fluid systems and improves the accuracy and individualization of drainage parameter optimization. The fuzzy PID control algorithm is combined to achieve adaptive adjustment of drainage speed, and real-time risk assessment and speed correction are performed based on multi-source information, which significantly improves the safety and effectiveness of the drainage process.

[0146] In an optional implementation, according to the drainage rate parameter and the drainage risk index, using a fuzzy PID control algorithm to calculate the drainage speed control parameter includes:

[0147] The drainage rate parameter and the drainage risk index output by the deep neural network are weightedly combined to obtain a target drainage rate, the actual drainage rate of the drainage device is collected, and the actual drainage rate is compared with the target drainage rate to obtain a control error value;

[0148] Collecting real-time data of the control error value to calculate the control error change rate;

[0149] The control error value is fuzzy processed by using the triangular membership function to obtain the control error membership value, and the control error change rate is fuzzy processed by using the triangular membership function to obtain the error change rate membership value;

[0150] Performing fuzzy reasoning according to the control error membership value, the error change rate membership value and a preset fuzzy rule base, and using a minimum-maximum synthesis method to obtain a fuzzy reasoning result of the drainage speed control parameter;

[0151] Performing a centroid method defuzzification process on the fuzzy inference result to obtain a proportional parameter adjustment amount, an integral parameter adjustment amount and a differential parameter adjustment amount;

[0152] The proportional parameter adjustment amount is added to the initial proportional parameter to obtain the proportional control parameter, the integral parameter adjustment amount is added to the initial integral parameter to obtain the integral control parameter, the differential parameter adjustment amount is added to the initial differential parameter to obtain the differential control parameter, and the proportional control parameter, the integral control parameter and the differential control parameter are combined to obtain the drainage speed control parameter.

[0153] Exemplarily, the drainage rate parameter and drainage risk index output by the deep neural network are first weighted and combined to obtain the target drainage rate. Specifically, the weight of the drainage rate parameter can be set to 0.7, the weight of the drainage risk index can be set to 0.3, and the target drainage rate can be obtained by weighted average. For example, if the drainage rate parameter is 5mL / h and the drainage risk index is 3mL / h, the target drainage rate is 5×0.7+3×0.3=4.4mL / h.

[0154] Next, the actual drainage rate of the drainage device is collected, and the actual drainage rate is compared with the target drainage rate to obtain a control error value. For example, if the actual drainage rate is 4 mL / h, the control error value is 4.4-4=0.4 mL / h.

[0155] Then, collect the real-time data of the control error value and calculate the control error change rate. The difference method can be used for calculation, that is, subtract the control error value at the previous moment from the control error value at the current moment, and then divide it by the sampling time interval. Assuming that the sampling time interval is 0.1s and the control error value at the previous moment is 0.3mL / h, the control error change rate is (0.4-0.3) / 0.1=1mL / h / s.

[0156] Next, the triangular membership function is used to fuzzify the control error value and the control error change rate. For the control error value, 7 fuzzy subsets can be set: NB (negative large), NM (negative medium), NS (negative small), ZO (zero), PS (positive small), PM (positive medium), PB (positive large). For the control error change rate, the same 7 fuzzy subsets can also be set. The membership values ​​of the control error value and the control error change rate on each fuzzy subset are obtained by calculation.

[0157] For example, for a control error value of 0.4 mL / h, possible membership values ​​are: ZO (0.2), PS (0.8). For a control error change rate of 1 mL / h / s, possible membership values ​​are: PS (0.6), PM (0.4).

[0158] Then, fuzzy reasoning is performed based on the control error membership value, error change rate membership value and the preset fuzzy rule base. The fuzzy rule base can contain 49 rules, for example: if the error is NB and the error change rate is NB, the proportional parameter adjustment amount is PB, the integral parameter adjustment amount is NB, and the differential parameter adjustment amount is PS. The minimum-maximum synthesis method is used to obtain the fuzzy reasoning results of the drainage speed control parameters.

[0159] The fuzzy inference result is defuzzified by the centroid method to obtain the proportional parameter adjustment amount, the integral parameter adjustment amount and the differential parameter adjustment amount. For example, the possible results are: proportional parameter adjustment amount 0.2, integral parameter adjustment amount -0.1, differential parameter adjustment amount 0.3.

[0160] Finally, the proportional parameter adjustment amount is added to the initial proportional parameter to obtain the proportional control parameter, the integral parameter adjustment amount is added to the initial integral parameter to obtain the integral control parameter, and the differential parameter adjustment amount is added to the initial differential parameter to obtain the differential control parameter. Assuming that the initial parameters are 1.0, 0.5, and 0.2, respectively, the final drainage velocity control parameters are: proportional control parameter 1.2, integral control parameter 0.4, and differential control parameter 0.5.

[0161] In this embodiment, the drainage rate parameters and drainage risk indicators output by the deep neural network are used as input, and the target drainage rate is obtained by weighted combination, which fully considers the drainage effect and safety, and can achieve more accurate drainage speed control. The fuzzy PID control algorithm is adopted, and the nonlinearity and uncertainty in the drainage process can be effectively handled through fuzzy processing and fuzzy reasoning, which improves the robustness and adaptability of the control. By dynamically adjusting the PID control parameters, real-time optimization can be performed according to the actual drainage situation, which effectively improves the accuracy and stability of the drainage speed control, and is conducive to improving the effect and safety of drainage treatment.

[0162] Figure 2 FIG. 1 is a schematic diagram of a precise drainage system for cerebral hemorrhage based on image density analysis according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0163] The first unit is used to obtain a three-dimensional computer tomography image of a patient's head, calculate the density value and density gradient value of each pixel in the target bleeding area by using a density analysis model based on a convolutional neural network, construct an adaptive density threshold curve according to the density value and density gradient value, generate a multi-level three-dimensional density distribution map, use a regional segmentation algorithm to extract a region in the multi-level three-dimensional density distribution map with a density greater than the density threshold curve as a hematoma core area, use a three-dimensional reconstruction algorithm to establish a three-dimensional model of the hematoma core area, and calculate the volume parameters, centroid coordinates, shape parameters and azimuth parameters of the hematoma core area;

[0164] The second unit is used to input the three-dimensional model of the hematoma core area into a deep neural network, perform layered processing on the hematoma core area based on the density gradient value, divide the hematoma core area into multiple density sub-layers according to the multi-level three-dimensional density distribution map, calculate the volume ratio, spatial coordinates and fluidity parameters of each density sub-layer, determine the spatial position coordinates of the optimal drainage point by using an optimization algorithm based on the volume ratio, spatial coordinates and fluidity parameters, import the patient's angiography image and functional area image, generate an optimal drainage path under the constraints of the angiography image and the functional area image, and input the spatial position coordinates of the optimal drainage point and the optimal drainage path into an electromagnetic positioning system;

[0165] The third unit is used to perform hematoma drainage using a multi-channel intelligent drainage system including a main drainage tube and multiple branch drainage tubes under the guidance of the electromagnetic positioning system, insert the main drainage tube to the optimal drainage point, calculate the optimal drainage position of each density sublayer according to the spatial coordinates and fluidity parameters of the density sublayer, insert multiple branch drainage tubes into the optimal drainage positions respectively, collect pressure data and electroencephalogram data of each density sublayer, input the pressure data, electroencephalogram data and volume ratio into a pre-constructed neurological function evaluation model to obtain a drainage rate parameter, adjust the drainage speed of the main drainage tube and the branch drainage tube according to the drainage rate parameter, and monitor the current volume parameter of the hematoma core area and the density distribution changes of each density sublayer in real time. When the current volume parameter drops below the preset volume value and the density distribution of each density sublayer tends to be uniform, the precise drainage of the hematoma is completed.

[0166] According to a third aspect of the embodiments of the present invention,

[0167] An electronic device is provided, comprising:

[0168] processor;

[0169] a memory for storing processor-executable instructions;

[0170] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0171] According to a fourth aspect of the embodiments of the present invention,

[0172] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0173] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0174] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A precise drainage method for cerebral hemorrhage based on image density analysis, characterized in that: include: Acquire a three-dimensional computer tomography image of the patient's head, calculate the density value and density gradient value of each pixel in the target bleeding area using a density analysis model based on a convolutional neural network, construct an adaptive density threshold curve according to the density value and density gradient value, generate a multi-level three-dimensional density distribution map, use a regional segmentation algorithm to extract the area in the multi-level three-dimensional density distribution map with a density greater than the density threshold curve as the hematoma core area, use a three-dimensional reconstruction algorithm to establish a three-dimensional model of the hematoma core area, and calculate the volume parameters, centroid coordinates, shape parameters and azimuth parameters of the hematoma core area; Inputting the three-dimensional model of the hematoma core area into a deep neural network, performing layered processing on the hematoma core area based on the density gradient value, dividing the hematoma core area into a plurality of density sub-layers according to the multi-level three-dimensional density distribution map, calculating the volume ratio, spatial coordinates and fluidity parameters of each density sub-layer, determining the spatial position coordinates of the optimal drainage point by using an optimization algorithm based on the volume ratio, spatial coordinates and fluidity parameters, importing the angiography image and functional area image of the patient, generating an optimal drainage path under the constraints of the angiography image and the functional area image, and inputting the spatial position coordinates of the optimal drainage point and the optimal drainage path into an electromagnetic positioning system; Under the guidance of the electromagnetic positioning system, a multi-channel intelligent drainage system including a main drainage tube and multiple branch drainage tubes is used to drain the hematoma, the main drainage tube is inserted to the optimal drainage point, the optimal drainage position of each density sublayer is calculated according to the spatial coordinates and fluidity parameters of the density sublayer, and multiple branch drainage tubes are respectively inserted into the optimal drainage positions, and the pressure data and electroencephalogram data of each density sublayer are collected. The pressure data, electroencephalogram data and volume ratio are input into a pre-constructed neurological function evaluation model to obtain the drainage rate parameter, and the drainage speed of the main drainage tube and the branch drainage tube is adjusted according to the drainage rate parameter, and the current volume parameter of the hematoma core area and the density distribution changes of each density sublayer are monitored in real time. When the current volume parameter drops below the preset volume value and the density distribution of each density sublayer tends to be uniform, the precise drainage of the hematoma is completed.

2. The method according to claim 1, characterized in that Obtain a three-dimensional computer tomography image of the patient's head, calculate the density value and density gradient value of each pixel in the target bleeding area using a density analysis model based on a convolutional neural network, construct an adaptive density threshold curve based on the density value and density gradient value, and generate a multi-level three-dimensional density distribution map including: Acquire a three-dimensional computer tomography image of a patient's head, divide the three-dimensional computer tomography image into a plurality of overlapping sub-blocks, calculate a three-dimensional grayscale histogram for each overlapping sub-block, perform cropping and restriction on the three-dimensional grayscale histogram to obtain a cumulative distribution function, perform mapping enhancement on the three-dimensional computer tomography image according to the cumulative distribution function to obtain an enhanced image, perform denoising on the enhanced image by using anisotropic diffusion filtering, determine a diffusion coefficient by analyzing a gradient histogram of the enhanced image, and iteratively process the enhanced image according to the diffusion coefficient to obtain a preprocessed three-dimensional computer tomography image; Inputting the preprocessed three-dimensional computed tomography image into a density analysis model, the density analysis model comprising a feature extraction subnetwork and a density calculation subnetwork, the feature extraction subnetwork comprising a plurality of three-dimensional convolution units, each of which is provided with a convolution layer with a different expansion coefficient and a three-dimensional attention calculation module, the density calculation subnetwork comprising a plurality of deconvolution layers, a feature fusion connection structure is provided between the feature extraction subnetwork and the density calculation subnetwork, the density analysis model is trained using a combined loss function including a density information constraint term to obtain a feature map of a target bleeding area, the density value of each pixel point in the target bleeding area is calculated based on the feature map, and a density value distribution map is generated; A three-dimensional gradient operator is used to calculate the density value distribution map to obtain the density gradient value of each pixel in the target bleeding area, a two-dimensional distribution histogram is established by the density value in the density value distribution map and the density gradient value, a joint probability distribution of the density value and the density gradient value in the two-dimensional distribution histogram is calculated, the joint probability distribution is divided into multiple candidate categories, the intra-class variance and the inter-class variance of each candidate category are calculated, an optimal segmentation threshold combination that maximizes the inter-class variance and minimizes the intra-class variance is found through an iterative optimization method, a control parameter of an optimal segmentation curve is constructed according to the optimal segmentation threshold combination, a curve interpolation is performed on the optimal segmentation curve according to the control parameter to obtain an adaptive density threshold curve, the density value distribution map and the adaptive density threshold curve are input into a density stratification algorithm, the target bleeding area is density stratified, and a multi-level three-dimensional density distribution map is generated.

3. The method according to claim 1, characterized in that The region where the density in the multi-level three-dimensional density distribution map is greater than the density threshold curve is extracted using a regional segmentation algorithm as the hematoma core region, and a three-dimensional model of the hematoma core region is established using a three-dimensional reconstruction algorithm. The volume parameters, centroid coordinates, shape parameters, and azimuth parameters of the hematoma core region are calculated, including: A regional segmentation algorithm is used to process the multi-level three-dimensional density distribution map, an initial seed point set is determined according to a density threshold curve, a point in the initial seed point set whose density value is greater than the threshold corresponding to the density threshold curve is used as a growth starting point, a directional density gradient value constraint condition is introduced in the regional growth process, the density value and density gradient direction of the candidate voxel are judged, the candidate voxel whose density value is greater than the density threshold and whose density gradient direction is consistent with the growth direction is included in the growth region, and the growth region is morphologically processed to obtain a segmentation boundary point set of the hematoma core area; A three-dimensional reconstruction model is established according to the segmentation boundary point set of the hematoma core area, the density gradient value of each point in the segmentation boundary point set is calculated, the grid size of different areas is set based on the density gradient value, the space is gridded by an octree structure, a distance-based weight coefficient is calculated for each grid node, a local weighted least squares equation group is constructed and solved to obtain the implicit surface function value of the grid node, an isosurface point set with zero implicit surface function value is extracted to perform triangular meshing, and a three-dimensional model of the hematoma core area is obtained; The morphological parameters of the core area of ​​the hematoma are calculated using the three-dimensional model, the volume value of the hematoma is calculated according to the grid vertex coordinates and vertex normal vectors of the three-dimensional model, the density values ​​in the multi-level three-dimensional density distribution map are correspondingly aligned with the voxel position information in the three-dimensional model, the centroid coordinates of the hematoma are calculated according to the aligned density values ​​and voxel positions, the inertia matrix of the hematoma area is constructed, the eigenvalues ​​of the inertia matrix are calculated, the flatness coefficient, elongation coefficient and sphericity coefficient of the hematoma are calculated according to the eigenvalues, the eigenvector of the inertia matrix is ​​calculated, the spatial deflection angle and azimuth angle of the hematoma are determined according to the angle between the eigenvector and the coordinate axis, and the volume parameters, centroid coordinates, shape parameters and azimuth parameters of the core area of ​​the hematoma are output.

4. The method according to claim 1, characterized in that: Inputting the three-dimensional model of the hematoma core area into a deep neural network, performing layered processing on the hematoma core area based on the density gradient value, dividing the hematoma core area into multiple density sub-layers according to the multi-level three-dimensional density distribution map, calculating the volume ratio, spatial coordinates and fluidity parameters of each density sub-layer, and determining the spatial position coordinates of the optimal drainage point by using an optimization algorithm based on the volume ratio, spatial coordinates and fluidity parameters, including: The three-dimensional model and density gradient value of the hematoma core area are input into a deep neural network, the deep neural network extracts density space features through density weighted convolution and gradient normalization, transmits feature information through density jump connection, and outputs a hierarchical prediction result of the hematoma core area; Based on the hierarchical prediction results, the core area of ​​the hematoma is divided into multiple density sub-layers, the density values ​​of the voxels in each density sub-layer are counted, the ratio of the volume of each density sub-layer to the total volume of the hematoma is calculated to obtain the volume ratio, the density-weighted summation of the voxels in each density sub-layer is calculated to obtain the spatial centroid coordinates of the density sub-layer, the mapping relationship between density and viscosity is established according to the density value statistics results to obtain the equivalent viscosity coefficient of each density sub-layer, and the equivalent viscosity coefficient is multiplied by the velocity gradient of each density sub-layer to obtain the fluidity parameter; The volume ratio is combined with the distance from the point to be located to the spatial coordinates of each density sublayer to construct a volume effect term, the minimum distance from the point to be located to the hematoma boundary is constructed as a distance constraint term, the fluidity parameter is combined with the angle between the density gradient direction and the spatial coordinate direction of each density sublayer pointed by the point to be located to construct a fluidity constraint term, and the volume effect term, distance constraint term and fluidity constraint term are weightedly combined to obtain the drainage point optimization objective function; The objective function is iteratively solved by an optimization algorithm. The search direction and step size are adjusted by using the density gradient direction in each iteration, and a local search is performed on the candidate solution obtained by the current iteration. When the optimized value of the objective function meets the convergence condition, the corresponding candidate solution is used as the spatial position coordinate of the optimal drainage point.

5. The method according to claim 1, characterized in that Importing the angiography image and the functional area image of the patient, and generating the optimal drainage path under the constraints of the angiography image and the functional area image comprises: Obtaining the patient's angiography image and functional area image, performing vascular enhancement processing on the angiography image to extract vascular direction characteristics, and constructing a vascular three-dimensional feature model in combination with vascular diameter information; reconstructing white matter fiber bundles based on diffusion tensor images to obtain functional area connection characteristics, and marking functional area boundaries in combination with functional magnetic resonance imaging data to obtain functional area spatial distribution; Extracting feature points from the angiography image and the functional area image and establishing corresponding relationships, and obtaining a transformation matrix by optimizing the registration objective function; using the transformation matrix to transform the three-dimensional feature model of the blood vessel and the spatial distribution of the functional area into a hematoma model coordinate system, and establishing a unified surgical planning space; A vascular distance field is constructed based on the vascular three-dimensional feature model and a vascular safety constraint is calculated; a gradient potential field is constructed based on the spatial distribution of the functional area to obtain the functional area protection constraint; and a path planning cost function is constructed by combining the vascular safety constraint, the functional area protection constraint and the path geometry constraint in combination with the length and curvature constraints of the drainage path; A path search tree is constructed with the optimal drainage point as the starting point, a sampling strategy is determined according to the cost function, and the tree is expanded, and a global optimal path is obtained through path optimization and pruning; Curvature sampling is performed on the global optimal path to obtain a key control point sequence, and a non-uniform rational parameter spline function is used to interpolate and reconstruct the control point sequence to obtain a smooth drainage trajectory; dynamic step sampling is performed along the smooth drainage trajectory, and the sampling points are substituted into the vascular safety constraints and functional area protection constraints for safety verification; the verified dangerous trajectory segment is iteratively optimized by adjusting the position of the local control points until all safety constraints are met, and the final drainage path is output.

6. The method according to claim 1, characterized in that Collecting pressure data and electroencephalogram data of each density sublayer, inputting the pressure data, electroencephalogram data and volume ratio into a pre-constructed neural function evaluation model to obtain a drainage rate parameter, and adjusting the drainage speed of the main drainage tube and the branch drainage tube according to the drainage rate parameter includes: Arrange pressure sensors at multiple density sublayers to collect pressure data, wherein the pressure sensors are arranged in a distributed manner, perform median filtering and wavelet denoising on the pressure data to obtain filtered pressure data, and calculate the filtered pressure data between adjacent measuring points to obtain a pressure gradient distribution diagram; Collecting EEG signals from the area surrounding the lesion, filtering the EEG signals to obtain filtered EEG signals, extracting time domain features, frequency domain features and nonlinear features of the filtered EEG signals, performing time-frequency analysis on the filtered EEG signals through continuous wavelet transform to obtain a time-frequency feature matrix, and constructing EEG signal quality indicators based on the time domain features, frequency domain features and nonlinear features; Constructing a multi-layer deep neural network model, the deep neural network model includes a pressure feature extraction branch, an EEG feature extraction branch and a volume feature extraction branch, wherein the pressure feature extraction branch uses a multi-scale one-dimensional convolutional layer to extract pressure time series features, the EEG feature extraction branch uses a bidirectional long short-term memory network and a self-attention mechanism to fuse multi-lead information, and the volume feature extraction branch uses a residual fully connected layer to process volume change features; The pressure gradient distribution map, the time-frequency feature matrix, the EEG signal quality index, and the volume ratio of each density sublayer are input into the deep neural network model, and the features of the pressure feature extraction branch, the EEG feature extraction branch, and the volume feature extraction branch are adaptively weighted combined through the feature fusion layer to output the drainage rate parameter and the drainage risk index; According to the drainage rate parameter and the drainage risk index, a fuzzy PID control algorithm is used to calculate the drainage speed control parameter, and the drainage speeds of the main drainage tube and multiple branch drainage tubes are adjusted according to the drainage speed control parameter and the spatial coordinates of each density sublayer. The drainage speed is adjusted in real time based on the pressure gradient distribution map, the EEG signal quality index and the drainage risk index.

7. The method according to claim 6, characterized in that According to the drainage rate parameter and the drainage risk index, the drainage speed control parameter is calculated by using the fuzzy PID control algorithm, including: The drainage rate parameter and the drainage risk index output by the deep neural network are weightedly combined to obtain a target drainage rate, the actual drainage rate of the drainage device is collected, and the actual drainage rate is compared with the target drainage rate to obtain a control error value; Collecting real-time data of the control error value to calculate the control error change rate; The control error value is fuzzy processed by using the triangular membership function to obtain the control error membership value, and the control error change rate is fuzzy processed by using the triangular membership function to obtain the error change rate membership value; Performing fuzzy reasoning according to the control error membership value, the error change rate membership value and a preset fuzzy rule base, and using a minimum-maximum synthesis method to obtain a fuzzy reasoning result of the drainage speed control parameter; Performing a centroid method defuzzification process on the fuzzy inference result to obtain a proportional parameter adjustment amount, an integral parameter adjustment amount and a differential parameter adjustment amount; The proportional parameter adjustment amount is added to the initial proportional parameter to obtain the proportional control parameter, the integral parameter adjustment amount is added to the initial integral parameter to obtain the integral control parameter, the differential parameter adjustment amount is added to the initial differential parameter to obtain the differential control parameter, and the proportional control parameter, the integral control parameter and the differential control parameter are combined to obtain the drainage speed control parameter.

8. A precise drainage system for cerebral hemorrhage based on image density analysis, used to implement the method according to any one of claims 1 to 7, characterized in that: include: The first unit is used to obtain a three-dimensional computer tomography image of a patient's head, calculate the density value and density gradient value of each pixel in the target bleeding area by using a density analysis model based on a convolutional neural network, construct an adaptive density threshold curve according to the density value and density gradient value, generate a multi-level three-dimensional density distribution map, use a regional segmentation algorithm to extract a region in the multi-level three-dimensional density distribution map with a density greater than the density threshold curve as a hematoma core area, use a three-dimensional reconstruction algorithm to establish a three-dimensional model of the hematoma core area, and calculate the volume parameters, centroid coordinates, shape parameters and azimuth parameters of the hematoma core area; The second unit is used to input the three-dimensional model of the hematoma core area into a deep neural network, perform layered processing on the hematoma core area based on the density gradient value, divide the hematoma core area into multiple density sub-layers according to the multi-level three-dimensional density distribution map, calculate the volume ratio, spatial coordinates and fluidity parameters of each density sub-layer, determine the spatial position coordinates of the optimal drainage point by using an optimization algorithm based on the volume ratio, spatial coordinates and fluidity parameters, import the patient's angiography image and functional area image, generate an optimal drainage path under the constraints of the angiography image and the functional area image, and input the spatial position coordinates of the optimal drainage point and the optimal drainage path into an electromagnetic positioning system; The third unit is used to perform hematoma drainage using a multi-channel intelligent drainage system including a main drainage tube and multiple branch drainage tubes under the guidance of the electromagnetic positioning system, insert the main drainage tube to the optimal drainage point, calculate the optimal drainage position of each density sublayer according to the spatial coordinates and fluidity parameters of the density sublayer, insert multiple branch drainage tubes into the optimal drainage positions respectively, collect pressure data and electroencephalogram data of each density sublayer, input the pressure data, electroencephalogram data and volume ratio into a pre-constructed neurological function evaluation model to obtain a drainage rate parameter, adjust the drainage speed of the main drainage tube and the branch drainage tube according to the drainage rate parameter, and monitor the current volume parameter of the hematoma core area and the density distribution changes of each density sublayer in real time. When the current volume parameter drops below the preset volume value and the density distribution of each density sublayer tends to be uniform, the precise drainage of the hematoma is completed.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images