A method for segmenting and finite element modeling of human skull tissue based on complete CT images

Abstract

The present invention belongs to the technical field of finite element modeling, and discloses a method for human skull tissue separation and finite element modeling based on complete CT images. The present invention obtains a total loss function value according to the mean square error loss function value and the maximum a posteriori loss function value through a skull CT image denoising processing method, and finally iteratively trains a convolutional neural network based on the calculated total loss function value until the training termination condition is met, outputs the trained skull CT image denoising model, and denoises the skull CT image to be denoised according to the skull CT image denoising model. Therefore, by iteratively training the convolutional neural network in combination with the mean square error loss function value and the maximum a posteriori loss function value, the effect of removing real noise can be greatly improved; at the same time, an accurate skull three-dimensional model can be constructed by constructing a skull three-dimensional model method.

Description

Technical Field

[0001] The invention belongs to the technical field of finite element modeling, and in particular relates to a method for separating and finite element modeling of human skull tissue based on complete CT images. Background Art

[0002] CT (Computed Tomography), or electronic computer tomography, uses precisely collimated X-ray beams, gamma rays, ultrasound, etc., together with highly sensitive detectors to perform cross-sectional scans around a certain part of the human body one by one. It has the characteristics of fast scanning time and clear images, and can be used to examine a variety of diseases; according to the different rays used, it can be divided into: X-ray CT (X-CT) and gamma ray CT (γ-CT), etc.; the finite element model is a calculation model for finite element analysis, which provides all the necessary raw data for finite element calculation. The process of establishing a finite element model is called finite element modeling, which is the key to the entire finite element analysis process. Whether the model is reasonable or not will directly affect the accuracy of the calculation results, the length of the calculation time, the size of the storage capacity, and whether the calculation process can be completed; however, the existing methods of tissue separation and finite element modeling based on the complete CT image of the human skull cannot effectively remove the noise on the real skull CT image; at the same time, the construction of the three-dimensional model of the skull is not accurate.

[0003] Through the above analysis, the problems and defects of the prior art are as follows:

[0004] (1) Existing methods for tissue segmentation and finite element modeling based on intact CT images of the human head cannot effectively remove noise from real head CT images.

[0005] (2) The three-dimensional model of the skull is not accurate. Summary of the invention

[0006] In view of the problems existing in the prior art, the present invention provides a method for separating and finite element modeling of human skull tissue based on complete CT images.

[0007] The present invention is implemented as follows: a human skull tissue separation and finite element modeling based on complete CT images includes:

[0008] Step 1: configure CT equipment parameters, perform positioning scanning on the target head through the CT equipment; obtain the head CT image, and perform denoising on the head CT image;

[0009] Step 2: Segment the head CT image to obtain the head structure;

[0010] Step 3: construct a three-dimensional head model according to the head CT image through a three-dimensional modeling program; assemble the constructed three-dimensional head model in the three-dimensional CAD system control software, scale and assemble it according to the actual size, and save it as a file in a specific format;

[0011] Step 4: Import the specific format file of the CAD three-dimensional solid model into the mesh pre-processing software, perform volume mesh-hexahedral unit segmentation, manually adjust the mesh accuracy on the irregular surface, and finally obtain the three-dimensional finite element model of the human head.

[0012] Furthermore, the method for denoising the head CT image is as follows:

[0013] (1) constructing a head CT image set, storing the acquired head CT images in the head CT image set; acquiring a training sample set, wherein the training sample set includes a noisy head CT image sample set and a noise-free head CT image sample set corresponding to the noisy head CT image sample set;

[0014] (2) inputting the noisy head CT image sample set into a convolutional neural network for training, and outputting a denoised head CT image sample set obtained by training; calculating the corresponding mean square error loss function value and maximum a posteriori loss function value according to the noise-free head CT image sample set and the denoised head CT image sample set, respectively, and obtaining a total loss function value according to the mean square error loss function value and the maximum a posteriori loss function value;

[0015] The step of respectively calculating the corresponding mean square error loss function value and maximum a posteriori loss function value according to the noise-free head CT image sample set and the denoised head CT image sample set comprises:

[0016] The mean square error loss function value is calculated according to the noise-free head CT image sample set and the denoised head CT image sample set, specifically:

[0017] The square value of each pixel difference between each frame of noise-free head CT image sample in the noise-free head CT image sample set and the denoised head CT image sample corresponding to each frame of noise-free head CT image sample in the denoised head CT image sample set is calculated respectively, and the average of the square values ​​of each pixel difference calculated is used as the mean square error loss function value; and the maximum a posteriori loss function value is calculated according to the noise-free head CT image sample set and the denoised head CT image sample set, specifically:

[0018] Calculate the posterior probability that each noise-free head CT image sample in the noise-free head CT image sample set has Poisson noise according to each noise-free head CT image sample in the noise-free head CT image sample set and a Poisson noise head CT image after Poisson noise is added to each noise-free head CT image sample;

[0019] According to the Poisson noise head CT image obtained by adding Poisson noise to each noise-free head CT image sample and the Gaussian noise head CT image obtained by adding Gaussian noise to each noise-free head CT image sample, the posterior probability that each Poisson noise head CT image has Gaussian noise is calculated;

[0020] According to the posterior probability that each noise-free head CT image sample has Poisson noise and the posterior probability that each Poisson noise head CT image has Gaussian noise, the maximum posterior probability that each noise-free head CT image sample has Poisson Gaussian noise is calculated;

[0021] Calculate the negative logarithmic function value with the maximum posterior probability of Poisson-Gaussian noise in each noise-free head CT image sample as the independent variable, and use the negative logarithmic function value as the maximum posterior loss function value;

[0022] The convolutional neural network is iteratively trained based on the calculated total loss function value until a training termination condition is met, the trained head CT image denoising model is output, and the head CT image to be denoised is denoised according to the head CT image denoising model.

[0023] Furthermore, the step of obtaining a training sample set includes:

[0024] Acquire multiple frames of noisy head CT image samples continuously shot by a head CT image acquisition device based on configured head CT image acquisition parameters to construct a noisy head CT image sample set, wherein the head CT image acquisition parameters include head CT image exposure parameters and shooting frequency parameters;

[0025] For each frame of noisy head CT image sample in the noisy head CT image sample set, the pixel value of each pixel point in the frame of noisy head CT image sample is sorted, and according to the sorting result, the abnormal pixel points whose pixel value is less than the first preset pixel value and whose pixel value is greater than the second preset pixel value are removed to obtain the noisy head CT image sample after the abnormal pixel points are removed;

[0026] Calculate the average value of each pixel of the noisy head CT image sample after removing the abnormal pixels of each frame;

[0027] According to the relationship between the preset noisy head CT image samples and the corresponding noise-free head CT image samples and the average value of each pixel of the noisy head CT image samples after removing the abnormal pixels of each frame, a noise-free head CT image sample set corresponding to the noisy head CT image sample set is obtained.

[0028] Furthermore, the step of inputting the noisy head CT image sample set into a convolutional neural network for training and outputting the denoised head CT image sample set obtained through training comprises:

[0029] Inputting the noisy head CT image sample set into a convolutional neural network, and sequentially extracting head CT image feature information of each noisy head CT image sample in the noisy head CT image sample set through each convolutional layer in the convolutional neural network;

[0030] For each convolution layer, the head CT image feature information of each noisy head CT image sample extracted by the convolution layer is sent to the deconvolution layer symmetrically arranged in the convolution layer through the connection layer, and the deconvolution layer generates a de-noised head CT image sample set after deconvolution according to the head CT image feature information of each noisy head CT image sample.

[0031] Further, the step of obtaining a total loss function value according to the mean square error loss function value and the maximum a posteriori loss function value comprises:

[0032] Calculate a first weighted loss function value of the mean square error loss function value and a second weighted loss function value of the maximum a posteriori loss function value respectively according to a preset weight ratio;

[0033] The total loss function value is obtained according to the first weight loss function value and the second weight loss function value.

[0034] Further, the method for constructing the three-dimensional model of the skull is as follows:

[0035] 1) acquiring a plurality of CT images of a target head, wherein the plurality of CT images are CT images obtained by a CT image capturing device capturing the target head at a plurality of different angles; constructing an initial three-dimensional model of the target head according to the plurality of CT images; and identifying at least one reference object on the initial three-dimensional model in response to a recognition operation;

[0036] 2) determining a scaling ratio of the three-dimensional model based on at least one real dimension of at least one of the reference objects and at least one scanned dimension of at least one of the reference objects in the initial three-dimensional model; and converting the initial three-dimensional model into a target three-dimensional model based on the scaling ratio of the three-dimensional model, wherein the target three-dimensional model is used to reflect the real dimension of the target head.

[0037] Further, constructing the initial three-dimensional model of the target head according to the multiple CT images includes:

[0038] The multiple CT images are processed based on a three-dimensional reconstruction algorithm to obtain an initial three-dimensional model of the target skull, wherein the three-dimensional reconstruction algorithm at least includes feature point matching and incremental reconstruction.

[0039] Furthermore, the processing of the plurality of CT images based on a three-dimensional reconstruction algorithm to obtain an initial three-dimensional model of the target skull includes:

[0040] Performing CT image segmentation on each captured CT image, wherein the CT image segmentation is used to filter out background area CT images that are irrelevant to the target head and the reference object in the captured CT image;

[0041] The plurality of photographed CT images after CT image segmentation are processed based on the three-dimensional reconstruction algorithm.

[0042] Further, at least one real size of the reference object and at least one scanned size of at least one of the reference objects in the initial three-dimensional model, determining the scaling ratio of the three-dimensional model comprises:

[0043] In response to a size input operation, obtaining at least one real size of at least one of the reference objects;

[0044] Calculating at least one scan size of at least one of the reference objects according to the number of voxels and the voxel size of at least one of the reference objects on the initial three-dimensional model;

[0045] A scaling ratio of the three-dimensional model is determined according to at least one real dimension of at least one of the reference objects and at least one scanned dimension of at least one of the reference objects.

[0046] Further, determining the scaling ratio of the three-dimensional model according to at least one real size of at least one of the reference objects and at least one scanned size of at least one of the reference objects in the initial three-dimensional model comprises:

[0047] Determining the scaling ratio of the three-dimensional model according to at least one real size of the reference object and at least one scanned size of the reference object in the initial three-dimensional model; and / or,

[0048] The scaling ratio of the three-dimensional model is determined according to a real size of at least one of the reference objects and a scanned size of at least one of the reference objects in the initial three-dimensional model.

[0049] In combination with the above technical solutions and the technical problems solved, please analyze the advantages and positive effects of the technical solutions to be protected by the present invention from the following aspects:

[0050] First, the present invention uses a method for denoising head CT images to input a noisy head CT image sample set into a convolutional neural network for training, outputs a denoised head CT image sample set obtained through training, calculates the corresponding mean square error loss function value and maximum a posteriori loss function value according to the noise-free head CT image sample set and the denoised head CT image sample set, and obtains a total loss function value according to the mean square error loss function value and the maximum a posteriori loss function value, and finally iteratively trains the convolutional neural network based on the calculated total loss function value until the training termination condition is met, outputs the trained head CT image denoising model, and denoises the denoised head CT image according to the head CT image denoising model. Thus, by iteratively training the convolutional neural network in combination with the mean square error loss function value and the maximum a posteriori loss function value, the removal effect of real noise can be greatly improved; at the same time, an accurate three-dimensional head model can be constructed by constructing a three-dimensional head model method. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a flow chart of a method for separating human skull tissue and finite element modeling based on complete CT images provided by an embodiment of the present invention.

[0052] Figure 2 It is a flow chart of a method for denoising a head CT image provided by an embodiment of the present invention.

[0053] Figure 3 It is a flow chart of a method for constructing a three-dimensional skull model provided by an embodiment of the present invention. DETAILED DESCRIPTION

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

[0055] like Figure 1 As shown, the present invention provides a method for separating and finite element modeling of human skull tissue based on complete CT images, comprising the following steps:

[0056] S101, configuring CT equipment parameters, performing positioning scanning on the target head through the CT equipment; acquiring the head CT image, and performing denoising processing on the head CT image;

[0057] S102, performing image segmentation on the head CT image to obtain the head structure;

[0058] S103, constructing a three-dimensional head model according to the head CT image through a three-dimensional modeling program; assembling the constructed three-dimensional head model in a three-dimensional CAD system control software, scaling and assembling it according to the actual size, and saving it as a file in a specific format;

[0059] S104, importing the specific format file of the CAD three-dimensional solid model into the meshing pre-processing software, performing volume mesh-hexahedral unit segmentation, manually adjusting the mesh accuracy on the irregular surface, and finally obtaining a three-dimensional finite element model of the human head.

[0060] like Figure 2 As shown, the method for denoising a head CT image provided by the present invention is as follows:

[0061] S201, constructing a head CT image set, storing the acquired head CT images in the head CT image set; acquiring a training sample set, the training sample set comprising a noisy head CT image sample set and a noise-free head CT image sample set corresponding to the noisy head CT image sample set;

[0062] S202, inputting the noisy head CT image sample set into a convolutional neural network for training, and outputting a denoised head CT image sample set obtained by training; calculating the corresponding mean square error loss function value and maximum a posteriori loss function value according to the noise-free head CT image sample set and the denoised head CT image sample set, respectively, and obtaining a total loss function value according to the mean square error loss function value and the maximum a posteriori loss function value;

[0063] The step of respectively calculating the corresponding mean square error loss function value and maximum a posteriori loss function value according to the noise-free head CT image sample set and the denoised head CT image sample set comprises:

[0064] The mean square error loss function value is calculated according to the noise-free head CT image sample set and the denoised head CT image sample set, specifically:

[0065] Respectively calculating the square value of each pixel difference between each frame of noise-free head CT image sample in the noise-free head CT image sample set and the denoised head CT image sample corresponding to each frame of noise-free head CT image sample in the denoised head CT image sample set, and taking the average of the calculated square values ​​of each pixel difference value as the mean square error loss function value; and

[0066] The maximum a posteriori loss function value is calculated according to the noise-free head CT image sample set and the denoised head CT image sample set, specifically:

[0067] Calculate the posterior probability that each noise-free head CT image sample in the noise-free head CT image sample set has Poisson noise according to each noise-free head CT image sample in the noise-free head CT image sample set and a Poisson noise head CT image after Poisson noise is added to each noise-free head CT image sample;

[0068] According to the Poisson noise head CT image obtained by adding Poisson noise to each noise-free head CT image sample and the Gaussian noise head CT image obtained by adding Gaussian noise to each noise-free head CT image sample, the posterior probability that each Poisson noise head CT image has Gaussian noise is calculated;

[0069] According to the posterior probability that each noise-free head CT image sample has Poisson noise and the posterior probability that each Poisson noise head CT image has Gaussian noise, the maximum posterior probability that each noise-free head CT image sample has Poisson Gaussian noise is calculated;

[0070] Calculate the negative logarithmic function value with the maximum posterior probability of Poisson-Gaussian noise in each noise-free head CT image sample as the independent variable, and use the negative logarithmic function value as the maximum posterior loss function value;

[0071] The convolutional neural network is iteratively trained based on the calculated total loss function value until a training termination condition is met, the trained head CT image denoising model is output, and the head CT image to be denoised is denoised according to the head CT image denoising model.

[0072] The step of obtaining a training sample set provided by the present invention includes:

[0073] Acquire multiple frames of noisy head CT image samples continuously shot by a head CT image acquisition device based on configured head CT image acquisition parameters to construct a noisy head CT image sample set, wherein the head CT image acquisition parameters include head CT image exposure parameters and shooting frequency parameters;

[0074] For each frame of noisy head CT image sample in the noisy head CT image sample set, the pixel value of each pixel point in the frame of noisy head CT image sample is sorted, and according to the sorting result, the abnormal pixel points whose pixel value is less than the first preset pixel value and whose pixel value is greater than the second preset pixel value are removed to obtain the noisy head CT image sample after the abnormal pixel points are removed;

[0075] Calculate the average value of each pixel of the noisy head CT image sample after removing the abnormal pixels of each frame;

[0076] According to the relationship between the preset noisy head CT image samples and the corresponding noise-free head CT image samples and the average value of each pixel of the noisy head CT image samples after removing the abnormal pixels of each frame, a noise-free head CT image sample set corresponding to the noisy head CT image sample set is obtained.

[0077] The step of inputting the noisy head CT image sample set into a convolutional neural network for training and outputting the denoised head CT image sample set obtained through training provided by the present invention comprises:

[0078] Inputting the noisy head CT image sample set into a convolutional neural network, and sequentially extracting head CT image feature information of each noisy head CT image sample in the noisy head CT image sample set through each convolutional layer in the convolutional neural network;

[0079] For each convolution layer, the head CT image feature information of each noisy head CT image sample extracted by the convolution layer is sent to the deconvolution layer symmetrically arranged in the convolution layer through the connection layer, and the deconvolution layer generates a de-noised head CT image sample set after deconvolution according to the head CT image feature information of each noisy head CT image sample.

[0080] The step of obtaining a total loss function value according to the mean square error loss function value and the maximum a posteriori loss function value provided by the present invention comprises:

[0081] Calculate a first weighted loss function value of the mean square error loss function value and a second weighted loss function value of the maximum a posteriori loss function value respectively according to a preset weight ratio;

[0082] The total loss function value is obtained according to the first weight loss function value and the second weight loss function value.

[0083] like Figure 3 As shown, the method for constructing a three-dimensional head model provided by the present invention is as follows:

[0084] S301, acquiring a plurality of CT images of a target head, wherein the plurality of CT images are CT images obtained by a CT image capturing device capturing the target head at a plurality of different angles; constructing an initial three-dimensional model of the target head according to the plurality of CT images; and identifying at least one reference object on the initial three-dimensional model in response to a recognition operation;

[0085] S302, determining a scaling ratio of the three-dimensional model according to at least one real dimension of at least one of the reference objects and at least one scanned dimension of at least one of the reference objects in the initial three-dimensional model; and converting the initial three-dimensional model into a target three-dimensional model according to the scaling ratio of the three-dimensional model, wherein the target three-dimensional model is used to reflect the real dimension of the target head.

[0086] The method of constructing the initial three-dimensional model of the target skull according to the plurality of CT images provided by the present invention comprises:

[0087] The multiple CT images are processed based on a three-dimensional reconstruction algorithm to obtain an initial three-dimensional model of the target skull, wherein the three-dimensional reconstruction algorithm at least includes feature point matching and incremental reconstruction.

[0088] The method provided by the present invention processes the plurality of CT images to obtain an initial three-dimensional model of the target skull based on a three-dimensional reconstruction algorithm, including:

[0089] Performing CT image segmentation on each captured CT image, wherein the CT image segmentation is used to filter out background area CT images that are irrelevant to the target head and the reference object in the captured CT image;

[0090] The plurality of photographed CT images after CT image segmentation are processed based on the three-dimensional reconstruction algorithm.

[0091] At least one real size of the reference object provided by the present invention, and at least one scanned size of at least one of the reference objects in the initial three-dimensional model, determining the scaling ratio of the three-dimensional model comprises:

[0092] In response to a size input operation, obtaining at least one real size of at least one of the reference objects;

[0093] Calculating at least one scan size of at least one of the reference objects according to the number of voxels and the voxel size of at least one of the reference objects on the initial three-dimensional model;

[0094] A scaling ratio of the three-dimensional model is determined according to at least one real dimension of at least one of the reference objects and at least one scanned dimension of at least one of the reference objects.

[0095] The method of determining the scaling ratio of the three-dimensional model according to at least one real size of at least one reference object and at least one scanned size of at least one reference object in the initial three-dimensional model provided by the present invention includes:

[0096] Determining the scaling ratio of the three-dimensional model according to at least one real size of the reference object and at least one scanned size of the reference object in the initial three-dimensional model; and / or,

[0097] The scaling ratio of the three-dimensional model is determined according to a real size of at least one of the reference objects and a scanned size of at least one of the reference objects in the initial three-dimensional model.

[0098] The present invention uses a head CT image denoising processing method to input a noisy head CT image sample set into a convolutional neural network for training, outputs a denoised head CT image sample set obtained by training, calculates the corresponding mean square error loss function value and maximum a posteriori loss function value according to the noise-free head CT image sample set and the denoised head CT image sample set, and obtains a total loss function value according to the mean square error loss function value and the maximum a posteriori loss function value, and finally iteratively trains the convolutional neural network based on the calculated total loss function value until the training termination condition is met, outputs the trained head CT image denoising model, and denoises the denoised head CT image according to the head CT image denoising model. Thus, by iteratively training the convolutional neural network in combination with the mean square error loss function value and the maximum a posteriori loss function value, the removal effect of real noise can be greatly improved; at the same time, an accurate head three-dimensional model can be constructed by constructing a head three-dimensional model method.

[0099] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. It can be understood by a person of ordinary skill in the art that the above-mentioned devices and methods can be implemented using computer executable instructions and / or contained in a processor control code, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. Such code is provided on the carrier medium. The device and its modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, and can also be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0100] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principle of the present invention should be covered by the protection scope of the present invention.

Claims

1. A method for separating and finite element modeling of human skull tissue based on complete CT images, characterized in that: The following steps are involved: Step 1: configure CT equipment parameters, perform positioning scanning on the target head through the CT equipment; obtain the head CT image, and perform denoising on the head CT image; Step 2: Segment the head CT image to obtain the head structure; Step 3: construct a three-dimensional head model according to the head CT image through a three-dimensional modeling program; assemble the constructed three-dimensional head model in the three-dimensional CAD system control software, scale and assemble it according to the actual size, and save it as a file in a specific format; Step 4: Import the specific format file of the CAD three-dimensional solid model into the mesh pre-processing software, perform volume mesh-hexahedral unit segmentation, manually adjust the mesh accuracy on the irregular surface, and finally obtain the three-dimensional finite element model of the human head; The method for denoising the head CT image is as follows: (1) constructing a head CT image set, storing the acquired head CT images in the head CT image set; acquiring a training sample set, wherein the training sample set includes a noisy head CT image sample set and a noise-free head CT image sample set corresponding to the noisy head CT image sample set; (2) inputting the noisy head CT image sample set into a convolutional neural network for training, and outputting a denoised head CT image sample set obtained by training; calculating the corresponding mean square error loss function value and maximum a posteriori loss function value according to the noise-free head CT image sample set and the denoised head CT image sample set, respectively, and obtaining a total loss function value according to the mean square error loss function value and the maximum a posteriori loss function value; The step of respectively calculating the corresponding mean square error loss function value and maximum a posteriori loss function value according to the noise-free head CT image sample set and the denoised head CT image sample set comprises: The mean square error loss function value is calculated according to the noise-free head CT image sample set and the denoised head CT image sample set, specifically: Respectively calculating the square value of each pixel difference between each frame of noise-free head CT image sample in the noise-free head CT image sample set and the denoised head CT image sample corresponding to each frame of noise-free head CT image sample in the denoised head CT image sample set, and taking the average of the calculated square values ​​of each pixel difference value as the mean square error loss function value; and The maximum a posteriori loss function value is calculated according to the noise-free head CT image sample set and the denoised head CT image sample set, specifically: Calculate the posterior probability that each noise-free head CT image sample in the noise-free head CT image sample set has Poisson noise according to each noise-free head CT image sample in the noise-free head CT image sample set and a Poisson noise head CT image after Poisson noise is added to each noise-free head CT image sample; According to the Poisson noise head CT image obtained by adding Poisson noise to each noise-free head CT image sample and the Gaussian noise head CT image obtained by adding Gaussian noise to each noise-free head CT image sample, the posterior probability that each Poisson noise head CT image has Gaussian noise is calculated; According to the posterior probability that each noise-free head CT image sample has Poisson noise and the posterior probability that each Poisson noise head CT image has Gaussian noise, the maximum posterior probability that each noise-free head CT image sample has Poisson Gaussian noise is calculated; Calculate the negative logarithmic function value with the maximum posterior probability of Poisson-Gaussian noise in each noise-free head CT image sample as the independent variable, and use the negative logarithmic function value as the maximum posterior loss function value; Iteratively training the convolutional neural network based on the calculated total loss function value until a training termination condition is met, outputting the trained head CT image denoising model, and performing denoising processing on the head CT image to be denoised according to the head CT image denoising model; The method for constructing the three-dimensional head model is as follows: 1) acquiring a plurality of CT images of a target head, wherein the plurality of CT images are CT images obtained by a CT image capturing device capturing the target head at a plurality of different angles; constructing an initial three-dimensional model of the target head according to the plurality of CT images; and identifying at least one reference object on the initial three-dimensional model in response to a recognition operation; 2) determining a scaling ratio of the three-dimensional model based on at least one real dimension of at least one of the reference objects and at least one scanned dimension of at least one of the reference objects in the initial three-dimensional model; and converting the initial three-dimensional model into a target three-dimensional model based on the scaling ratio of the three-dimensional model, wherein the target three-dimensional model is used to reflect the real dimension of the target head.

2. The method for separating and finite element modeling of human skull tissue based on complete CT images as claimed in claim 1, characterized in that: The step of obtaining a training sample set includes: Acquire multiple frames of noisy head CT image samples continuously shot by a head CT image acquisition device based on configured head CT image acquisition parameters to construct a noisy head CT image sample set, wherein the head CT image acquisition parameters include head CT image exposure parameters and shooting frequency parameters; For each frame of noisy head CT image sample in the noisy head CT image sample set, the pixel value of each pixel point in the frame of noisy head CT image sample is sorted, and according to the sorting result, the abnormal pixel points whose pixel value is less than the first preset pixel value and whose pixel value is greater than the second preset pixel value are removed to obtain the noisy head CT image sample after the abnormal pixel points are removed; Calculate the average value of each pixel of the noisy head CT image sample after removing the abnormal pixels of each frame; According to the relationship between the preset noisy head CT image samples and the corresponding noise-free head CT image samples and the average value of each pixel of the noisy head CT image samples after removing the abnormal pixels of each frame, a noise-free head CT image sample set corresponding to the noisy head CT image sample set is obtained.

3. The method for separating and finite element modeling of human skull tissue based on complete CT images as claimed in claim 1, characterized in that: The step of inputting the noisy head CT image sample set into a convolutional neural network for training and outputting the denoised head CT image sample set obtained through training comprises: Inputting the noisy head CT image sample set into a convolutional neural network, and sequentially extracting head CT image feature information of each noisy head CT image sample in the noisy head CT image sample set through each convolutional layer in the convolutional neural network; For each convolution layer, the head CT image feature information of each noisy head CT image sample extracted by the convolution layer is sent to the deconvolution layer symmetrically arranged in the convolution layer through the connection layer, and the deconvolution layer generates a de-noised head CT image sample set after deconvolution according to the head CT image feature information of each noisy head CT image sample.

4. The method for separating and finite element modeling of human skull tissue based on complete CT images as claimed in claim 1, characterized in that: The step of obtaining a total loss function value according to the mean square error loss function value and the maximum a posteriori loss function value comprises: Calculate a first weighted loss function value of the mean square error loss function value and a second weighted loss function value of the maximum a posteriori loss function value respectively according to a preset weight ratio; The total loss function value is obtained according to the first weight loss function value and the second weight loss function value.

5. The method for separating and finite element modeling of human skull tissue based on complete CT images as claimed in claim 4, characterized in that: The step of constructing the initial three-dimensional model of the target skull according to the plurality of CT images comprises: The multiple CT images are processed based on a three-dimensional reconstruction algorithm to obtain an initial three-dimensional model of the target skull, wherein the three-dimensional reconstruction algorithm at least includes feature point matching and incremental reconstruction.

6. The method for separating and finite element modeling of human skull tissue based on complete CT images as claimed in claim 5, characterized in that: The step of processing the plurality of CT images based on a three-dimensional reconstruction algorithm to obtain an initial three-dimensional model of the target skull includes: Performing CT image segmentation on each captured CT image, wherein the CT image segmentation is used to filter out background area CT images that are irrelevant to the target head and the reference object in the captured CT image; The plurality of photographed CT images after CT image segmentation are processed based on the three-dimensional reconstruction algorithm.

7. The method for separating and finite element modeling of human skull tissue based on complete CT images as claimed in claim 4, characterized in that: At least one real size of the reference object, and at least one scanned size of at least one of the reference objects in the initial three-dimensional model, determining the scaling ratio of the three-dimensional model comprises: In response to a size input operation, obtaining at least one real size of at least one of the reference objects; Calculating at least one scan size of at least one of the reference objects according to the number of voxels and the voxel size of at least one of the reference objects on the initial three-dimensional model; A scaling ratio of the three-dimensional model is determined according to at least one real size of at least one of the reference objects and at least one scanned size of at least one of the reference objects in the initial three-dimensional model.

8. The method for separating and finite element modeling of human skull tissue based on complete CT images as claimed in claim 4, characterized in that: Determining the scaling ratio of the three-dimensional model according to at least one real size of at least one of the reference objects and at least one scanned size of at least one of the reference objects in the initial three-dimensional model comprises: The scaling ratio of the three-dimensional model is determined according to at least one real dimension of the reference object and at least one scanned dimension of the reference object in the initial three-dimensional model.

Images