Funnel chest operation simulation test method and device based on deep learning and finite element analysis, electronic equipment and storage medium

Abstract

The invention discloses a funnel chest simulation test method and device based on deep learning and finite element analysis, electronic equipment and a storage medium. An accurate funnel chest operation simulation test result is achieved. The invention discloses a funnel chest operation simulation test method based on deep learning and finite element analysis. The method comprises the following steps: (1) acquiring a chest CT image of a testee; (2) carrying out denoising processing on the CT image by adopting a deep learning technology, and removing image noise through an attention enhancement module and a Transform module; (3) converting the denoised CT image into a 3D thoracic cavity model by utilizing MIMICs software, and optimizing model details through an iterative enhancement technology, including hole filling, model smoothing, model refining and tissue segmentation; and (4) performing an operation simulation test on the 3D model by applying a finite element analysis technology, calculating a Haller index, determining an intercostal gap position, calculating stress and strain, and determining configuration of an NUSS operation rod.

Description

Technical Field

[0001] The present invention relates to the fields of medical image processing, surgical simulation testing technology, and the application of deep learning in the medical field, and specifically to a pectus excavatum surgical simulation testing technology based on deep learning and finite element analysis. Background Art

[0002] With the development of society and technology, surgical simulation technology has gradually entered large hospitals, schools and medical research institutions, helping doctors, students and researchers to improve surgical skills and conduct scientific research. By analyzing the results of surgical simulation tests, the level of medical care can be improved.

[0003] Pectus excavatum, commonly known as "depressed chest" or "funnel chest," is the most common congenital chest wall deformity that typically presents in infancy and often worsens during the rapid growth associated with puberty. As a deformity, pectus excavatum not only affects the person's appearance, but the deformity can also compress vital structures including the heart and lungs, leading to reduced exercise tolerance and breathing problems.

[0004] There is no simulation test of related pectus excavatum surgery at present. When managers use manual calculation and statistical methods to obtain surgical simulation test results, there may be subjective factors, and no quantitative analysis is performed on the obtained surgical parameters, which will make the surgical simulation test results of the subjects inaccurate.

[0005] When we try to conduct simulation tests, the CT images of the subjects are used as the basis for simulation. Due to factors such as the instability of X-ray sources and detectors, quantum effects, and low-dose scanning, the noise that often exists in existing CT images may affect the accurate identification of key details of the chest wall and surrounding structures, thereby hindering the simulation test results.

[0006] Therefore, in view of the limitations of the current lack of simulation testing methods for pectus excavatum and the problem of noise in CT images, the present invention aims to propose a simulation testing method, device, electronic device and storage medium for pectus excavatum based on deep learning and finite element analysis, so as to achieve accurate simulation test results. Summary of the invention

[0007] The technical problem to be solved by the present invention is that in view of the deficiencies in the above-mentioned prior art, the present invention discloses a pectus excavatum simulation test method, device, electronic device and storage medium based on deep learning and finite element analysis to achieve accurate pectus excavatum surgery simulation test results.

[0008] In order to achieve the above object, the present invention adopts the following technical solutions:

[0009] In a first aspect, the present invention discloses a simulation test method for pectus excavatum surgery based on deep learning and finite element analysis.

[0010] The following steps are involved:

[0011] (1) Obtain a chest CT image of the subject; (2) Use deep learning technology to denoise the CT image and remove image noise through an attention enhancement module and a Transformer module;

[0012] (3) MIMICs software was used to convert the denoised CT images into a 3D chest model, and the model details were optimized through iterative enhancement technology, including filling holes, model smoothing, model refinement, and tissue segmentation. (4) Finite element analysis technology was used to perform surgical simulation tests on the 3D model, calculate the Haller index, determine the position of the intercostal space, calculate stress and strain, and determine the configuration of the NUSS surgical rod. The virtual surgical simulation test was performed in combination with ANSYS software. The test results of the surgical simulation of the subject were obtained.

[0013] As an improvement of the pectus excavatum surgery simulation test method based on deep learning and finite element analysis disclosed in the present invention,

[0014] The deep learning denoising of step (2) above includes the following steps:

[0015] S1: Preprocessing and data enhancement of CT image data, including noise addition, contrast adjustment, and edge enhancement;

[0016] S2: Use the Ridnet architecture to build a deep learning denoising model, integrating the Enhanced Attention Module (EAM) and Transformer module to improve denoising capabilities;

[0017] S3: Feature extraction and reconstruction are performed through the cascaded attention enhancement module (EAM) and Transformer module to output the denoised CT image.

[0018] As an improvement of the pectus excavatum surgery simulation test method based on deep learning and finite element analysis disclosed in the present invention,

[0019] In the above step (2), the key parameters in the deep learning denoising step include:

[0020] Use the Adam optimizer with a momentum of 0.99;

[0021] The initial learning rate is 0.001, and the maximum number of training cycles is 200;

[0022] The batch size is 16, and the number of convolutional layers and filter stacks is 3 pairs;

[0023] The kernel depth of each filter stack is 64, 128, 256;

[0024] The convolution kernel size is 3×3, and the activation function is ReLU;

[0025] The maximum pooling layer uses a 2×2 pooling kernel;

[0026] There are 2 layers of fully connected layers for potential features;

[0027] The L2 regularization coefficient is 0.0001;

[0028] The input image size is 64×64 pixels;

[0029] The loss function is cross entropy;

[0030] One dropout layer with a dropout rate of 20%.

[0031] As an improvement of the pectus excavatum surgery simulation test method based on deep learning and finite element analysis disclosed in the present invention,

[0032] In the above step (3), the construction of the 3D chest model includes the following steps:

[0033] S10: Import the denoised CT image into the MIMICs software;

[0034] S11: Perform thresholding, editing masks, image segmentation, smart filling, smoothing, and thinning to build a highly realistic chest model.

[0035] As an improvement of the pectus excavatum surgery simulation test method based on deep learning and finite element analysis disclosed in the present invention,

[0036] In the above step (3), the key parameters in the modeling step of the 3D chest model include:

[0037] During threshold processing, the lowest threshold was set to 120HU and the highest threshold was set to 2000HU;

[0038] The “Region Growing” function was applied for image segmentation, and the minimum and maximum deviations were set to 50 HU;

[0039] Use the Smart Fill feature to fill gaps and inconsistencies in your images;

[0040] Apply the “Smoothing” function to smooth and refine the 3D chest model.

[0041] As an improvement of the pectus excavatum surgery simulation test method based on deep learning and finite element analysis disclosed in the present invention,

[0042] In the above step (4), the finite element analysis includes the following steps:

[0043] S101: Calculate the Haller index to assess the severity of chest wall deformity;

[0044] S102: Determine the position of the intercostal space for subsequent surgical simulation;

[0045] S103: Calculate stress and strain to optimize the configuration of the NUSS surgical rod;

[0046] S104: Use ANSYS software to perform virtual surgery simulation and iteratively adjust to achieve the best correction effect.

[0047] As an improvement of the pectus excavatum surgery simulation test method based on deep learning and finite element analysis disclosed in the present invention,

[0048] In the above step (4), the key parameters in the finite element analysis step include:

[0049] When applying meshing, pay special attention to the mesh density and type of the thorax;

[0050] Use the "shrink wrap" function to smooth the edges of the rib cage;

[0051] Mark key simulation points on the thorax, including the upper and lower spine and the anterior ribs of the first intercostal space;

[0052] Design the NUSS rod specific to the subject, and choose the material as metal rod;

[0053] The designed NUSS rod geometry was loaded into the model, with the density set to 1.6E5, Young's modulus to 4.6E5, Poisson's ratio to 0.3, bulk modulus to 3.79E8, and shear modulus to 1.75E8;

[0054] Define fixed points and stress analysis, and set the stretch range to 20mm to 45mm;

[0055] By exploring different intercostal spaces and lift levels, iterative adjustments are made until the best configuration is found for the subject, resulting in the best test results.

[0056] In a second aspect, the present invention also discloses a simulation test device for pectus excavatum surgery based on deep learning and finite element analysis.

[0057] include:

[0058] A first acquisition module is used to acquire a CT image of the chest of the subject;

[0059] The second acquisition module is used for CT image denoising, feature extraction and reconstruction, and outputs the denoised CT image;

[0060] The third acquisition module is used to construct a three-dimensional chest model of the subject after inputting the denoised CT image;

[0061] The calculation module is used for surgical simulation testing. By calculating the Haller index, stress and strain, a virtual surgical simulation test is performed in combination with ANSYS software to obtain the test results of the surgical simulation of the subject performing the surgery.

[0062] In a third aspect, the present invention also discloses an electronic device, comprising a processor, a storage medium and a bus, wherein the storage medium stores machine-readable instructions executable by the processor. When the electronic device is running, the processor and the storage medium communicate via the bus, and the processor executes the machine-readable instructions to perform any step of the above method.

[0063] In a fourth aspect, the present invention further discloses a storage medium having a computer program stored thereon, wherein the computer program executes any one of the steps of the above method when executed by a processor.

[0064] The beneficial effects of the present invention are:

[0065] The present invention can avoid the phenomenon of obtaining surgical simulation test results by artificial calculation and statistics, and can not only quantitatively analyze the test parameters of each subject, but also improve the accuracy of the surgical simulation test results of the subject. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 It is a deep learning network architecture designed in the present invention for CT image denoising;

[0067] Figure 2 It is the Enhanced Attention (EAM) module in the network for CT image denoising in the present invention;

[0068] Figure 3 It is a schematic diagram of the calculation of the Haller index in the present invention;

[0069] Figure 4 It is a schematic diagram for determining the intercostal space in the present invention;

[0070] Figure 5-1 This is a schematic diagram of a CT image before denoising in the present invention;

[0071] Figure 5-2 yes Figure 5-1 Schematic diagram of CT image after denoising. DETAILED DESCRIPTION

[0072] The specific implementation of the present invention is described below in conjunction with the accompanying drawings and embodiments:

[0073] Embodiment 1:

[0074] A simulation test method for pectus excavatum surgery based on deep learning and finite element analysis.

[0075] The following steps are involved:

[0076] (1) Obtain a chest CT image of the subject; (2) Use deep learning technology to denoise the CT image, and remove image noise through the attention enhancement module and the Transformer module; (3) Use MIMICs software to convert the denoised CT image into a 3D chest model, and optimize the model details through iterative enhancement technology, including filling holes, model smoothing, model refinement, and tissue segmentation; (4) Use finite element analysis technology to perform surgical simulation tests on the 3D model, calculate the Haller index, determine the position of the intercostal space, calculate stress and strain, and determine the configuration of the NUSS surgical rod, and use ANSYS software to perform virtual surgical simulation tests; and obtain the test results of the surgical simulation of the subject.

[0077] As a preferred embodiment of the above-mentioned pectus excavatum surgery simulation test method based on deep learning and finite element analysis,

[0078] The deep learning denoising of step (2) above includes the following steps:

[0079] S1: Preprocessing and data enhancement of CT image data, including noise addition, contrast adjustment, and edge enhancement;

[0080] S2: Use the Ridnet architecture to build a deep learning denoising model, integrating the Enhanced Attention Module (EAM) and Transformer module to improve denoising capabilities;

[0081] S3: Feature extraction and reconstruction are performed through the cascaded attention enhancement module (EAM) and Transformer module to output the denoised CT image.

[0082] As a preferred embodiment of the above-mentioned pectus excavatum surgery simulation test method based on deep learning and finite element analysis,

[0083] In the above step (2), the key parameters in the deep learning denoising step include:

[0084] Use the Adam optimizer with a momentum of 0.99;

[0085] The initial learning rate is 0.001, and the maximum number of training cycles is 200;

[0086] The batch size is 16, and the number of convolutional layers and filter stacks is 3 pairs;

[0087] The kernel depth of each filter stack is 64, 128, 256;

[0088] The convolution kernel size is 3×3, and the activation function is ReLU;

[0089] The maximum pooling layer uses a 2×2 pooling kernel;

[0090] There are 2 layers of fully connected layers for potential features;

[0091] The L2 regularization coefficient is 0.0001;

[0092] The input image size is 64×64 pixels;

[0093] The loss function is cross entropy;

[0094] One dropout layer with a dropout rate of 20%.

[0095] As an improvement of the pectus excavatum surgery simulation test method based on deep learning and finite element analysis disclosed in the present invention,

[0096] In the above step (3), the construction of the 3D chest model includes the following steps:

[0097] S10: Import the denoised CT image into the MIMICs software;

[0098] S11: Perform thresholding, editing masks, image segmentation, smart filling, smoothing, and thinning to build a highly realistic chest model.

[0099] As an improvement of the pectus excavatum surgery simulation test method based on deep learning and finite element analysis disclosed in the present invention,

[0100] In the above step (3), the key parameters in the modeling step of the 3D chest model include:

[0101] During threshold processing, the lowest threshold was set to 120HU and the highest threshold was set to 2000HU;

[0102] The “Region Growing” function was applied for image segmentation, and the minimum and maximum deviations were set to 50 HU;

[0103] Use the Smart Fill feature to fill gaps and inconsistencies in your images;

[0104] Apply the “Smoothing” function to smooth and refine the 3D chest model.

[0105] As a preferred embodiment of the above-mentioned pectus excavatum surgery simulation test method based on deep learning and finite element analysis,

[0106] In the above step (4), the finite element analysis includes the following steps:

[0107] S101: Calculate the Haller index to assess the severity of chest wall deformity;

[0108] S102: Determine the position of the intercostal space for subsequent surgical simulation;

[0109] S103: Calculate stress and strain to optimize the configuration of the NUSS surgical rod;

[0110] S104: Use ANSYS software to perform virtual surgery simulation and iteratively adjust to achieve the best correction effect.

[0111] As a preferred embodiment of the above-mentioned pectus excavatum surgery simulation test method based on deep learning and finite element analysis,

[0112] In the above step (4), the key parameters in the finite element analysis step include:

[0113] When applying meshing, pay special attention to the mesh density and type of the thorax;

[0114] Use the "shrink wrap" function to smooth the edges of the rib cage;

[0115] Mark key simulation points on the thorax, including the upper and lower spine and the anterior ribs of the first intercostal space;

[0116] Design the NUSS rod specific to the subject, and choose the material as metal rod;

[0117] The designed NUSS rod geometry was loaded into the model, with the density set to 1.6E5, Young's modulus to 4.6E5, Poisson's ratio to 0.3, bulk modulus to 3.79E8, and shear modulus to 1.75E8;

[0118] Define fixed points and stress analysis, and set the stretch range to 20mm to 45mm;

[0119] By exploring different intercostal spaces and lift levels, iterative adjustments are made until the best configuration is found for the subject, resulting in the best test results.

[0120] Embodiment 2: A simulation test device for pectus excavatum surgery based on deep learning and finite element analysis.

[0121] include:

[0122] A first acquisition module is used to acquire a CT image of the chest of the subject;

[0123] The second acquisition module is used for CT image denoising, feature extraction and reconstruction, and outputs the denoised CT image;

[0124] The third acquisition module is used to construct a three-dimensional chest model of the subject after inputting the denoised CT image;

[0125] The calculation module is used for surgical simulation testing. By calculating the Haller index, stress and strain, a virtual surgical simulation test is performed in combination with ANSYS software to obtain the test results of the surgical simulation of the subject performing the surgery.

[0126] Embodiment 3: An electronic device comprises a processor, a storage medium and a bus, wherein the storage medium stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform any step of the above method.

[0127] Embodiment 4: A storage medium having a computer program stored thereon, wherein the computer program executes any one of the steps of the above method when executed by a processor.

[0128] In the above embodiment 1:

[0129] The inherent noise in CT images can significantly blur key anatomical details, making the visibility and distinction of these structures challenging. If these noises are not processed, the clarity required for medical diagnosis and treatment planning will be seriously impeded. Therefore, in order to better diagnose and treat pectus excavatum, the first thing that needs to be solved is to denoise the CT images. In the step of CT image denoising, the present invention proposes a CT image denoising method based on deep learning, which is committed to effectively removing image noise and improving image quality, providing a solid and reliable foundation for all subsequent stages. The CT image denoising of the present invention will first introduce the basic data, then the overall model architecture, and then the key design EAM and Transformer modules in the denoising link, and then the model configuration, and finally the denoising process as a whole.

[0130] (1) Data introduction and preprocessing:

[0131] The CT image data was collected from the OSIC Pulmonary Fibrosis Progression Challenge. As a public dataset, this dataset has been widely used in the study of medical image analysis. The collected dataset contains CT images of 200 subjects, providing rich data samples for subsequent analysis. Before model training, the CT images were preprocessed, including noise addition, contrast adjustment, edge enhancement, etc., to improve image quality and highlight anatomical features. At the same time, in order to improve the generalization ability of the model, data enhancement techniques such as image rotation, flipping, and scaling were applied to the dataset to simulate different imaging conditions and directions. After preprocessing and data enhancement, the CT image data has high resolution and can clearly show the subject's chest structure, including ribs, sternum, and other related anatomical details. At the same time, the dataset reflects the diversity of chest structures of different subjects, including different degrees of pectus excavatum deformity and other possible abnormalities, and because the dataset provides detailed information required to assess the severity of chest wall deformity and plan surgery, the collected data is highly relevant to the simulation test results.

[0132] (2) Model architecture:

[0133] Based on the CT image data after preprocessing and data enhancement, the present invention utilizes the existing Ridnet architecture, innovates on its basis, and builds a deep learning denoising model that combines the attention mechanism. Ridnet itself is a deep learning architecture widely used in the field of image denoising, and it already has good denoising performance. Based on this architecture, the present invention further enhances the model's capabilities by integrating the Transformer module. The specific deep learning network model is as follows: Figure 1 shown.

[0134] The overall network architecture shown in the figure includes: input layer, multiple processing layers and output layer. The input layer inputs the pre-processed CT image data; the multiple processing layers in the middle first perform a feature extraction module, which consists of a convolution layer to extract preliminary features from the input image; followed by a residual learning module, which consists of a cascaded attention enhancement module (EAM). EAM uses residual connections and feature attention mechanisms to enhance the model's ability to capture image details to improve denoising capabilities. At the same time, Transformer modules are added between EAMs to use the self-attention mechanism to process the global context information of the image, allowing the model to consider a wider context when processing images, rather than just local features; the last layer of the processing layer is a reconstruction module, and the model converts the processed feature map to the original data space through a convolution layer; the last layer of the entire network architecture is the output layer, which outputs the denoised CT image data.

[0135] (3) Key components:

[0136] After clarifying the overall denoising process, the present invention will focus on two key components, namely the attention enhancement module (EAM) and the Transformer module. The EAM in the denoising framework proposed in the present invention adopts the "residual of residual connections" (RoR) architecture. This structural design combines local skip connections and short skip connections to provide a stable framework that supports deep network structures, thereby significantly improving the performance of denoising tasks. The EAM in the present invention uses three groups of cascades to optimize the model to maintain a balance between performance and computational efficiency. The specific structure of EAM is as follows Figure 2 As shown, the specific process of EAM is divided into two steps:

[0137] (a) The first step is implemented by a “merge-run” unit, starting with a branch of input features, which are then passed through two parallel dilated convolutional layers. These layers allow for a wider receptive field, capturing more contextual information without significantly increasing the number of parameters, while the outputs after these layers are concatenated and passed through an additional convolutional layer to merge and refine the feature maps.

[0138] (b) Step 2: Further processing in a residual block containing two convolutional layers. Compression of learned features is achieved through an enhanced residual block (ERB), which contains three convolutional layers, where the last layer in ERB uses convolution kernels to flatten features in preparation for the final feature attention stage. ERB not only reduces the dimensionality of the feature space, but also retains important information, optimizes the speed and efficiency of the network, and at the same time, the output of ERB is added to the input of the next EAM module, forming a feedback loop that enhances the network's learning and denoising capabilities.

[0139] The overall architecture of the model proposed in the present invention includes three such EAM blocks, which are interspersed with two Transformer blocks to further refine the feature attention process. The Transformer module has advantages in processing long-distance dependencies and rich contextual information due to its self-attention mechanism, enabling the model to more effectively identify and remove image noise. In addition, EAM can better preserve the key edge details in the image and can completely preserve the structure of the CT image. Its edge focusing attention mechanism can be reflected by the following formula:

[0140] EAM(x)=σ(Conv(x))x

[0141] Where σ is an activation function, x is the image data input to EAM, Conv(x) represents the convolution operation used to emphasize the edge, and EAM(x) is the output result of EAM. This model can remove noise while retaining edge details. This image processing quality can provide a clear data basis for subsequent NUSS surgical planning.

[0142] (4) Model configuration:

[0143] In terms of model configuration, for the convolutional layers, the kernel size is uniformly set to 3×3, while the final convolution in ERB and the convolution in the feature attention unit use a 1×1 kernel to refine feature selection and reduce dimensionality. The convolutional layers are standardized to 64 channels per layer. The last convolutional layer of the network is designed to output three or one feature map according to the specific requirements of the input data. Some parameter configurations in other deep learning denoising models are shown in Table 1.

[0144] Key parameters of deep learning models Optimal parameters Solver Name Adam Momentum 0.99 Initial Learning Rate Single,0.001 Epoches(Maxim) 200 Mini batch size 16 Ppairs of conv.Layers and filter stacks 3 Kernel depth(size of each filter stack) 64,128,256 Kernel size 3×3 ReLU layers 3 Pooling type MaxPool(2×2) Fully connected layers for latent features 2 Number of filter stacks 3 L2-regularization 0.0001 Input image size 64×64 Loss function Cross Entropy Number of dropout layers 1 Dropout rate(%) 20

[0145] Table 1. Model configuration in deep learning denoising model

[0146] (5) Denoising process:

[0147] Based on the denoising network architecture proposed in the present invention, the following will describe the process of applying the network architecture to denoise CT images. The specific denoising process is as follows: first, the CT image is input into the network, and the data after the input layer processing is input into the feature extraction module for feature extraction. The feature extraction module in the network architecture of the present invention includes a convolutional layer for feature extraction, which is composed of 16 filters for feature extraction from the input CT image, as shown in the following formula:

[0148] f0=M c (x)

[0149] Among them, M c (x) represents the convolution operation on the input noisy image, and the convolution operation result f0 is subsequently passed to the learning residual on the residual learning module, as shown in the following formula:

[0150] f r =M fl (f0)

[0151] where f r represents the learned features, M fl(f0) indicates residual feature learning for the result of the previous convolution operation, and the residual feature learning module is composed of cascaded EAMs. The network architecture proposed by the present invention has a small depth, but the kernel expansion of the initial two branch convolutions of each EAM provides a wide receptive field, which enables the model to effectively capture the global information in the image even at a shallow network level. The output features of a layer in the last EAM are input to the reconstruction module in the network architecture, which is also composed of a convolution layer, as shown in the following formula:

[0152]

[0153] Among them, M r (f r ) means putting the learned features into the reconstruction module. It represents the image data remapped back to the space of the original image data, and finally passed to the output layer to output the denoised CT image data.

[0154] The method for 3D modeling of the chest cavity in the above-mentioned embodiment 1:

[0155] Based on the denoised CT images, it is necessary to restore the two-dimensional CT images to a three-dimensional 3D chest model to enrich the data details and improve the data accuracy, so as to facilitate the subsequent positioning of the chest. MIMICs is a mature software for converting CT images into 3D models, so the present invention mainly performs 3D construction and iterative simulation based on MIMICs software.

[0156] The first is 3D modeling of the chest cavity. Unlike directly using MIMICs software for 3D modeling, the present invention combines the use of software with iterative enhancement technology, and repeatedly adjusts the threshold parameters to better capture the contrast between the chest cavity and surrounding tissues, thereby enhancing the visibility of key anatomical features. MIMICs can more accurately depict the edges of the ribs and costal cartilages through the parameters provided by the present invention, thereby improving the overall structural fidelity of the model. In addition, for complex intersections and discontinuous areas in the skeletal structure, MINICs cannot automatically model, and manual intervention is required to further correct artifacts and fill in inaccurate areas caused by automatic segmentation. The present invention improves the accuracy of the model based on MINICs in 3D modeling of the chest cavity, and specifically achieves refinement of the model by solving the following problems, which are as follows:

[0157] (1) Filling holes: Due to image resolution limitations or noise that is not completely eliminated during the denoising process, small gaps or holes often appear in the generated model.

[0158] (2) Model smoothing: The details of the model after filling are not enough to meet the high level required for surgery, so the irregularities on the model surface need to be smoothed;

[0159] (3) Model refinement: To improve accuracy, the geometric structure of the model needs to be enhanced, that is, the mesh density of the area that requires higher resolution is increased to depict finer anatomical features;

[0160] (4) Segmentation: After the thoracic cavity model is refined, different tissues in the thoracic cavity need to be distinguished and segmented. This is achieved by assigning different densities and colors to different tissue areas in the 3D model. This segmentation is crucial for surgical simulation.

[0161] Through repeated experimental verification and parameter adjustment, the present invention provides a set of operating procedures and key parameters for 3D modeling of the chest cavity in MINICs, as shown in Table 2. The denoised CT image is imported, and the steps and parameters in the table can be used to more accurately realize the 3D modeling of the chest cavity, preparing for subsequent simulation verification.

[0162]

[0163] Table 2. Steps and key parameters for 3D modeling of the thorax using MINICs

[0164] In the above embodiment 1, the medical image mining process based on finite element analysis is as follows:

[0165] Based on the 3D modeled chest cavity model, the present invention further performs image information mining through Finite Element Analysis (FEA) to assist and optimize subsequent tests.

[0166] Since existing image mining technology has certain technical limitations in planning and execution, the present invention will innovatively perform surgical simulation on a 3D chest model based on the finite element analysis technology that has been maturely applied in fields such as mechanics, in order to simulate and predict the biomechanical behavior during surgery.

[0167] Finite element analysis is a numerical simulation method used to solve engineering problems in the fields of structure, fluid, heat conduction, etc. Finite element analysis divides a complex structure or system into many small finite element units, then models and analyzes each unit, and finally obtains the behavior of the entire system through combination. This method can be used to simulate and predict the physical properties of the structure, such as stress, deformation, temperature, flow field, etc. Therefore, applying finite element analysis to NUSS surgery analysis and combining it with a 3D chest model can better optimize NUSS surgery from a mechanical perspective.

[0168] The specific finite element analysis image information mining method is divided into the following key steps and indicators, including:

[0169] (1) Calculation of the Haller Index: The Haller Index is calculated by measuring the ratio of the transverse diameter (the distance between the rib surfaces) to the anterior thoracic distance (the shortest distance between the vertebrae and the sternum). This index helps to assess the severity of chest deformity. By calculating the Haller Index, it can help to understand the degree of inward depression of the ribs and the degree to which they need to be lifted during surgery. The higher the Haller Index, the more severe the deformity, and usually a more extensive surgical approach is required to effectively correct the rib structure. The Haller Index calculation based on the sternum is as follows: Figure 3 Incorporating the Haller index into FEA analysis makes it possible to simulate the surgical procedures required to correct the ribs based on this index, which helps to improve the precision and accuracy of surgical simulation.

[0170] (2) Determine the location of the intercostal space: After determining the Haller index, a detailed anatomical assessment of the subject's ribs is required, which requires determining the location of the intercostal space for subsequent simulation. Figure 4 As shown. The determination process first requires checking the 3D model of the chest cavity, and each intercostal space site must be located in the 3D model. Secondly, it is necessary to use the NUSS surgical rod defined in ANSYS for simulation. The NUSS surgical rod needs to be inserted into each determined gap site in the 3D model in turn, and the crossbar is inserted into each intercostal space. Through continuous simulation, the position, insertion angle and force of the crossbar during the operation are repeatedly adjusted, and the effect is determined based on the changes in the rib shape and the overall contour of the chest wall.

[0171] (3) Stress and strain calculation: Based on the sternum model with the NUSS surgical rod distributed, a virtual surgical scene needs to be created to calculate the stress and strain of specific ribs and cartilage during the simulation. The simulation results are quantitatively analyzed. As shown in the figure, the ribs are first measured in the virtual surgery, then the stress analysis points are fixed, and finally the stress and strain of the ribs are analyzed. By analyzing the stress and strain in different simulation processes, the technology can be adjusted to minimize risks.

[0172] In practical operation, ANSYS software is used to perform virtual surgery iteration simulation and optimization, and finally achieve assistance and optimization. The present invention provides the operation flow and key parameters of the virtual surgery iteration simulation and optimization part of the 3D model, as shown in Table 3, imports the 3D rendering file STL in MINICs, and uses the steps and parameters in the table to more accurately realize the NUSS surgery simulation and refinement of the chest model, so as to achieve a better correction effect of pectus excavatum.

[0173]

[0174] Table 3. Iterative simulation, detailed operations and key parameters of virtual surgery using ANSYS

[0175] In addition, in the above-mentioned embodiment 1: data preprocessing and data enhancement, it is necessary to first collect data in a public data set. The selected data set is OSIC Pulmonary Fibrosis Progression. 200 CT images of subjects with pectus excavatum were screened from the public data set as the image data used in this example. Then the data will be preprocessed. First, advanced image processing techniques are applied to the selected images, including noise reduction, contrast enhancement and edge sharpening. In addition, the preprocessing also standardizes the image data to ensure the consistency of the data set. Standardization involves adjusting the image's brightness, contrast, color and other attributes to meet unified standards, thereby reducing errors caused by differences in imaging conditions or equipment. Then, the data is enhanced to a certain extent. By randomly rotating and flipping the image, the model can learn the noise characteristics in different directions and positions, which helps to improve the model's generalization ability for various noise situations.

[0176] Secondly, the obtained CT images are denoised using a deep learning framework, and the preprocessed data is input into the deep learning network proposed in the present invention for denoising. During the experiment, the training data set is split, 70% of the data is used for training (238 frames of images), and 30% of the data is used for testing (101 frames of images). During the training of the deep learning model, the parameter selection of each model is shown in Table 1. Finally, for the denoised CT image, the comparison before and after denoising is shown in Figure 5.

[0177] Based on the denoised CT images, MIMICS software needs to be used to perform 3D modeling of the chest cavity. The specific 3D modeling steps and key parameters are shown in Table 2.

[0178] Finally, the 3D chest model and the denoised CT image were mined for optimal surgical sites based on finite element analysis, and NUSS surgical simulation and experiments were continuously performed in combination with ANSYS software. The specific operation steps and key parameters are shown in Table 3. Finally, the best NUSS surgical rod insertion node was found in multiple surgical simulations, and the pectus excavatum correction simulation was performed. Finally, a set of schematic diagrams of pectus excavatum corrected by the pectus excavatum correction method proposed by the present invention were given, as shown in Figure 3. Figure 5-2 shown.

[0179] The preferred embodiments of the present invention are described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above embodiments. Various changes can be made within the knowledge scope of ordinary technicians in the field without departing from the purpose of the present invention. These changes involve related technologies well known to those skilled in the art, which all fall within the scope of protection of the patent of the present invention.

[0180] Many other changes and modifications may be made without departing from the concept and scope of the present invention.It should be understood that the present invention is not limited to the specific embodiments, and the scope of the present invention is defined by the appended claims.

Claims

1. Simulation test method of pectus excavatum surgery based on deep learning and finite element analysis, It is characterized in that The following steps are involved: (1) Obtaining a chest CT image of the subject; (2) Deep learning technology is used to denoise CT images, and image noise is removed through the attention enhancement module and the Transformer module; (3) MIMICs software was used to convert the denoised CT images into 3D chest models, and the model details were optimized through iterative enhancement techniques, including hole filling, model smoothing, model refinement, and tissue segmentation; (4) Finite element analysis technology was used to perform surgical simulation tests on the 3D model, calculate the Haller index, determine the position of the intercostal space, calculate stress and strain, and determine the configuration of the NUSS surgical rod. The virtual surgical simulation test was performed in combination with ANSYS software to obtain the test results of the surgical simulation of the subject.

2. The method according to claim 1, It is characterized in that The deep learning denoising in step (2) comprises the following steps: S1: Preprocessing and data enhancement of CT image data, including noise addition, contrast adjustment, and edge enhancement; S2: Use the Ridnet architecture to build a deep learning denoising model, integrating the Enhanced Attention Module (EAM) and Transformer module to improve denoising capabilities; S3: Feature extraction and reconstruction are performed through the cascaded attention enhancement module (EAM) and Transformer module to output the denoised CT image.

3. The method according to claim 2, It is characterized in that In step (2), the key parameters in the deep learning denoising step include: Use the Adam optimizer with a momentum of 0.99; The initial learning rate is 0.001, and the maximum number of training cycles is 200; The batch size is 16, and the number of convolutional layers and filter stacks is 3 pairs; The kernel depth of each filter stack is 64, 128, 256; The convolution kernel size is 3×3, and the activation function is ReLU; The maximum pooling layer uses a 2×2 pooling kernel; There are 2 layers of fully connected layers for potential features; The L2 regularization coefficient is 0.0001; The input image size is 64×64 pixels; The loss function is cross entropy; One dropout layer with a dropout rate of 20%.

4. The method according to claim 1, It is characterized in that In step (3), the construction of the 3D chest model includes the following steps: S10: Import the denoised CT image into the MIMICs software; S11: Perform thresholding, editing masks, image segmentation, smart filling, smoothing, and thinning to build a highly realistic chest model.

5. The method according to claim 4, It is characterized in that In step (3), the key parameters in the modeling step of the 3D chest model include: During threshold processing, the lowest threshold was set to 120HU and the highest threshold was set to 2000HU; The "Region Growing" function was used for image segmentation, and the minimum and maximum deviations were set to 50 HU; Use the "Smart Fill" feature to fill gaps and inconsistencies in your images; Apply the "Smoothing" function to smooth and refine the 3D chest model.

6. The method according to claim 1, It is characterized in that In the step (4), the finite element analysis comprises the following steps: S101: Calculate the Haller index to assess the severity of chest wall deformity; S102: Determine the position of the intercostal space for subsequent surgical simulation; S103: Calculate stress and strain to optimize the configuration of the NUSS surgical rod; S104: Use ANSYS software to perform virtual surgery simulation and iteratively adjust to achieve the best correction effect.

7. The method according to claim 6, It is characterized in that In step (4), the key parameters in the finite element analysis step include: When applying meshing, pay special attention to the mesh density and type of the thorax; Use the "shrink wrap" function to smooth the edges of the rib cage; Mark key simulation points on the thorax, including the upper and lower spine and the anterior ribs of the first intercostal space; Design the NUSS rod specific to the subject, and choose the material as metal rod; The designed NUSS rod geometry was loaded into the model, with the density set to 1.6E5, Young's modulus to 4.6E5, Poisson's ratio to 0.3, bulk modulus to 3.79E8, and shear modulus to 1.75E8; Define fixed points and stress analysis, and set the stretch range to 20mm to 45mm; By exploring different intercostal spaces and lift levels, iterative adjustments are made until the best configuration is found for the subject, resulting in the best test results.

8. Funnel chest surgery simulation test device based on deep learning and finite element analysis, It is characterized in that include: A first acquisition module is used to acquire a CT image of the chest of the subject; The second acquisition module is used for CT image denoising, feature extraction and reconstruction, and outputs the denoised CT image; The third acquisition module is used to construct a three-dimensional chest model of the subject after inputting the denoised CT image; The calculation module is used for surgical simulation testing. It calculates the Haller index, stress and strain, and combines ANSYS software to perform virtual surgical simulation testing. A test result of the surgical simulation of the subject performing the surgery is obtained.

9. An electronic device comprising a processor, a storage medium and a bus, characterized in that: The storage medium stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the storage medium via a bus, and the processor executes the machine-readable instructions to perform the steps of the method as described in any one of claims 1-6.

10. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are executed.

Images