Liver image three-dimensional reconstruction system and method based on artificial intelligence technology

Abstract

The present application relates to the field of image reconstruction, and specifically discloses a liver image three-dimensional reconstruction system and method based on artificial intelligence technology, which collects liver images of patients, then performs standardization processing to unify the image format, and then removes noise in the image to improve clarity. The segmentation boundary of the liver is optimized through morphological adjustment, and then the two-dimensional image is converted into a three-dimensional liver model using three-dimensional reconstruction technology. Finally, the model is displayed on the screen. These processes improve the accuracy and efficiency of image processing, provide intuitive and clear three-dimensional views for medical diagnosis and surgical planning, and enhance the accuracy and reliability of medical decision-making.

Description

Technical Field

[0001] The present application relates to the field of image reconstruction, and more specifically, to a system and method for three-dimensional reconstruction of liver images based on artificial intelligence technology. Background Art

[0002] In modern medicine, accurate diagnosis and treatment planning of liver diseases are highly dependent on high-quality liver image analysis. With the rapid development of medical imaging technologies, such as ultrasound, CT (computed tomography) and MRI (magnetic resonance imaging), more and more detailed liver image information can be obtained. However, these images are inevitably interfered by various noises during the generation and transmission process, which seriously affects the subsequent image analysis and diagnostic accuracy.

[0003] Traditional image denoising methods, such as those based on filters (such as Gaussian filtering, median filtering, etc.), can remove some noise to a certain extent, but they often have limitations. They may blur the edges and details of liver images while removing noise. In addition, it is difficult to accurately remove noise for complex liver tissue structures and lesion characteristics, and cannot meet the requirements of modern medicine for high-precision analysis of liver images.

[0004] Therefore, an optimized three-dimensional reconstruction scheme for liver images is desired. Summary of the invention

[0005] In order to solve the above technical problems, this application is proposed.

[0006] According to one aspect of the present application, a liver image three-dimensional reconstruction system based on artificial intelligence technology is provided, which includes:

[0007] A liver image acquisition module, used for acquiring a liver image of a target patient;

[0008] A standardization module, used for performing standardization processing on the liver image to obtain a standardized liver image;

[0009] a denoising module, configured to perform denoising processing on the standardized liver image to obtain a denoised liver image, wherein the denoising module comprises: a standardized image discrete feature extraction unit, configured to perform discrete transformation and liver frequency feature extraction on the standardized liver image to obtain a sequence of liver sub-band frequency features; a liver sub-band frequency aggregation unit, configured to perform graph autocorrelation gated significant aggregation on the sequence of liver sub-band frequency features to obtain a liver sub-band frequency significant aggregation representation; and a denoised liver image generation unit, configured to generate the denoised liver image based on the liver sub-band frequency significant aggregation representation;

[0010] A semantic segmentation module, used for performing semantic segmentation on the denoised liver image to obtain a semantic segmentation image of the liver;

[0011] An optimization segmentation module, used for optimizing the segmentation boundary of the liver semantic segmentation image by morphological adjustment to obtain a boundary-optimized liver semantic segmentation image;

[0012] A three-dimensional reconstruction module, used for applying a three-dimensional reconstruction technology to perform three-dimensional reconstruction on the boundary-optimized liver semantic segmentation image to obtain a three-dimensional liver model;

[0013] The display module is used to display the three-dimensional liver model on a display screen.

[0014] Furthermore, the standardization process includes size adjustment, gray value normalization and histogram equalization.

[0015] Furthermore, the standardized image discrete feature extraction unit comprises:

[0016] A liver image discrete wavelet transform subunit, used for performing a two-dimensional discrete wavelet transform on the standardized liver image to generate a plurality of liver sub-band images;

[0017] The liver sub-band frequency feature extraction sub-unit is used to pass the multiple liver sub-band images through a liver sub-band feature extractor based on a CNN-LSTM model to obtain a sequence of liver sub-band frequency feature vectors as the sequence of liver sub-band frequency features.

[0018] Furthermore, the liver sub-band frequency aggregation unit comprises:

[0019] A liver subband frequency hyperbolic mapping subunit, used for mapping each liver subband frequency feature vector in the sequence of liver subband frequency feature vectors to a hyperbolic space to obtain a sequence of liver subband frequency feature vectors after hyperbolic space mapping;

[0020] A liver sub-band frequency graph walk topology feature calculation subunit, used to calculate the graph walk topology feature of the sequence of liver sub-band frequency feature vectors after the hyperbolic space mapping to obtain a liver sub-band frequency graph walk topology feature matrix;

[0021] A liver subband frequency feature semantic enhancement subunit, used for performing a global context walk of a graph convolution model on the liver subband frequency graph walk topological feature matrix and the sequence of liver subband frequency feature vectors after hyperbolic space mapping to obtain a sequence of liver subband frequency context semantic enhancement feature vectors;

[0022] The liver sub-band frequency feature significant aggregation sub-unit is used to calculate the significant aggregation representation of the sequence of liver sub-band frequency feature vectors based on the sequence of liver sub-band frequency context semantic reinforcement feature vectors to obtain a liver sub-band frequency significant aggregation representation vector as the liver sub-band frequency significant aggregation representation.

[0023] Furthermore, the liver sub-band frequency graph wandering topological feature calculation subunit is used to:

[0024] Calculating the Poincare distance between any two liver subband frequency feature vectors after hyperbolic space mapping in the sequence of liver subband frequency feature vectors after hyperbolic space mapping to obtain a liver subband frequency graph walk topology matrix;

[0025] The liver sub-band frequency graph walk topology matrix is ​​subjected to dilated convolution coding to obtain the liver sub-band frequency graph walk topology feature matrix.

[0026] Furthermore, the liver sub-band frequency characteristic significant aggregation sub-unit is used for:

[0027] Performing autocorrelation gating processing on each group of corresponding liver sub-band frequency context semantic enhancement feature vectors and liver sub-band frequency feature vectors in the sequence of liver sub-band frequency context semantic enhancement feature vectors and the sequence of liver sub-band frequency feature vectors to obtain a sequence of liver sub-band frequency autocorrelation gated significant confidence factors;

[0028] Performing a normalization process based on a Softmax function on the sequence of liver sub-band frequency autocorrelation gated significant confidence factors to obtain a sequence of liver sub-band frequency autocorrelation gated significant confidence weight factors;

[0029] Based on the sequence of liver sub-band frequency autocorrelation gated significant confidence weight factors, a position-weighted sum of the sequence of liver sub-band frequency feature vectors is calculated to obtain the liver sub-band frequency significant aggregate representation vector.

[0030] Furthermore, the denoised liver image generating unit is used to: after adding generation prompt information at the tail of the liver sub-band frequency significant aggregation representation vector, input it together with the standardized liver image into the AIGC-based liver image denoising generator to obtain the denoised liver image, and the generation prompt information is: denoising the standardized liver image based on the input corpus.

[0031] Furthermore, the semantic segmentation module is used to: perform semantic segmentation on the denoised liver image using a semantic segmentation model based on a convolutional neural network to obtain the semantic segmentation image of the liver.

[0032] According to another aspect of the present application, a method for three-dimensional reconstruction of liver images based on artificial intelligence technology is provided, comprising:

[0033] Acquire a liver image of a target patient; perform standardization processing on the liver image to obtain a standardized liver image; perform denoising processing on the standardized liver image to obtain a denoised liver image; perform semantic segmentation on the denoised liver image to obtain a liver semantic segmentation image; perform morphological adjustment on the liver semantic segmentation image to optimize the segmentation boundary to obtain a boundary-optimized liver semantic segmentation image; perform three-dimensional reconstruction on the boundary-optimized liver semantic segmentation image using a three-dimensional reconstruction technology to obtain a liver three-dimensional model; and display the liver three-dimensional model on a display screen;

[0034] The step of performing denoising on the standardized liver image to obtain a denoised liver image includes:

[0035] Performing discrete transformation and liver feature extraction on the standardized liver image to obtain a sequence of liver sub-band frequency features;

[0036] Performing graph autocorrelation gating significant aggregation on the sequence of liver sub-band frequency features to obtain a liver sub-band frequency significant aggregation representation;

[0037] The denoised liver image is generated based on the liver sub-band frequency significant aggregation representation.

[0038] Compared with the prior art, the present application provides a liver image three-dimensional reconstruction system and method based on artificial intelligence technology, which collects the patient's liver image, then performs standardization processing to unify the image format, and then removes the noise in the image to improve clarity. The segmentation boundary of the liver is optimized through morphological adjustment, and then the two-dimensional image is converted into a three-dimensional liver model using three-dimensional reconstruction technology. Finally, the model is displayed on the screen. These processes improve the accuracy and efficiency of image processing, provide intuitive and clear three-dimensional views for medical diagnosis and surgical planning, and enhance the accuracy and reliability of medical decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0040] Figure 1 4 is a block diagram of a liver image three-dimensional reconstruction system based on artificial intelligence technology according to an embodiment of the present application.

[0041] Figure 2 It is a block diagram of a denoising module in a liver image three-dimensional reconstruction system based on artificial intelligence technology according to an embodiment of the present application.

[0042] Figure 3 It is a block diagram of a standardized image discrete feature extraction unit in a liver image three-dimensional reconstruction system based on artificial intelligence technology according to an embodiment of the present application.

[0043] Figure 4 4 is a block diagram of a liver sub-band frequency aggregation unit in a liver image three-dimensional reconstruction system based on artificial intelligence technology according to an embodiment of the present application.

[0044] Figure 5 The present invention is a flowchart of a method for three-dimensional reconstruction of liver images based on artificial intelligence technology according to an embodiment of the present application. DETAILED DESCRIPTION

[0045] The embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments described herein, but rather these embodiments are provided for a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0046] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0047] It is worth noting that in this application, all actions to obtain signals, information or data are carried out in compliance with the relevant data protection laws and policies of the country where they are located and with the authorization given by the owner of the corresponding device.

[0048] In response to the problems in the above-mentioned background technology, the present application proposes a three-dimensional reconstruction system of liver images based on artificial intelligence technology, which achieves more precise and effective processing of liver images by introducing advanced AI technology and mathematical models. Figure 1 FIG. 1 is a block diagram of a liver image three-dimensional reconstruction system based on artificial intelligence technology according to an embodiment of the present application. Specifically, Figure 1As shown, according to an embodiment of the present application, a liver image three-dimensional reconstruction system 100 based on artificial intelligence technology includes: a liver image acquisition module 110, used to acquire a liver image of a target patient object; a standardization module 120, used to perform standardization processing on the liver image to obtain a standardized liver image; a denoising module 130, used to perform denoising processing on the standardized liver image to obtain a denoised liver image; a semantic segmentation module 140, used to perform semantic segmentation on the denoised liver image to obtain a liver semantic segmentation image; an optimized segmentation module 150, used to optimize the segmentation boundary of the liver semantic segmentation image through morphological adjustment to obtain a boundary-optimized liver semantic segmentation image; a three-dimensional reconstruction module 160, used to perform three-dimensional reconstruction on the boundary-optimized liver semantic segmentation image using a three-dimensional reconstruction technology to obtain a liver three-dimensional model; and a display module 170, used to display the liver three-dimensional model on a display screen.

[0049] Specifically, the liver image acquisition module 110 is used to acquire a liver image of a target patient object. It should be understood that the liver image of the target patient object is a visual image representation of the liver part obtained by a medical imaging device. The source of the image can be varied, such as through an ultrasound imaging device, a CT (computed tomography) scanner, or an MRI (magnetic resonance imaging) device. These devices use different physical principles to generate liver images. Ultrasound imaging uses the reflection of ultrasound to construct an image, which has the advantages of real-time and non-invasiveness; CT scanning uses X-rays to perform tomography of the human body, which can provide high-resolution liver structure information; MRI uses magnetic fields and radio frequency pulse imaging, and has a strong ability to resolve soft tissues.

[0050] Specifically, first of all, the medical team will choose the most suitable imaging method for different conditions or examination purposes. For example, if you need to quickly obtain the general structural information of the liver, ultrasound imaging is a good choice because it is real-time and non-invasive; for more detailed anatomical structure analysis, especially when high soft tissue contrast is required, MRI may be used. When it is necessary to observe calcification points or other high-density materials, CT becomes the first choice. Once the imaging method is determined, the technician will prepare the corresponding equipment and set the necessary parameters to ensure image quality.

[0051] Next, before specific imaging is performed, the upcoming process must be explained to the patient to ensure that they understand and agree to the examination. At the same time, the technician will also instruct the patient on how to cooperate, such as whether they need to maintain a specific body position, hold their breath, and other actions, all of which are to ensure that the quality of the image is not affected. For some special types of imaging, such as enhanced CT or MRI, it may be necessary to inject contrast agents into the patient, which helps to better display blood vessels and other internal structures. Throughout the process, medical staff always pay attention to the patient's condition to ensure that they complete the examination in a safe and comfortable environment.

[0052] As the device is turned on, imaging begins. At this point, the device uses different physical principles to generate images of the liver. For example, ultrasound equipment constructs images by transmitting ultrasound waves into the human body and receiving the reflected information; CT scanners use X-rays to perform a series of thin-layer scans on the human body, and then the computer processes the data to form cross-sectional images; MRI relies on strong magnetic fields and radio wave pulses to excite hydrogen nuclei in the body to produce signals, and then undergoes complex mathematical conversions to obtain a detailed picture of the internal structure of the human body. Regardless of the technology used, the ultimate goal is to capture clear and accurate images of the liver. After imaging is completed, the resulting raw image data will be transmitted to a dedicated image processing workstation.

[0053] Specifically, the standardization module 120 is used to perform standardization processing on the liver image to obtain a standardized liver image. Specifically, in the embodiment of the present application, the standardization processing includes size adjustment, gray value normalization and histogram equalization. Accordingly, due to factors such as different imaging devices, imaging parameters and individual differences of patients, the liver image will have differences in size, gray range, spatial coordinates, etc. The standardization processing is to eliminate these differences so that the subsequent processing steps can be performed on the basis of a relatively uniform image.

[0054] During the resizing stage, the goal is to make all acquired liver images have the same physical size or resolution. This is because different imaging devices may generate images of different sizes, and these size differences will affect the results of subsequent processing. For example, some algorithms may perform better for inputs of a specific size, or in order to facilitate the comparison of data from different sources, all images need to be resized to the same standard size. In actual operation, technicians will scale the original image according to the preset target size while ensuring that the image proportions remain unchanged to avoid deformation. In addition, for some special clinical needs, such as local magnification observation, it may be necessary to retain the original size while increasing the resolution of a specific area.

[0055] Grayscale normalization is the process of adjusting the brightness and contrast of an image. Due to different imaging conditions, even images of the same part may have significant brightness differences, which not only affects the visual effect, but more importantly, interferes with the accuracy of the computer-aided diagnosis system. Through normalization, the pixel grayscale values ​​of all images can be mapped to a fixed range, usually an integer value from 0 to 255. This process helps to reduce errors caused by factors such as lighting conditions and device sensitivity, so that images acquired at different time points or with different devices can be compared at the same scale. Common methods for achieving grayscale normalization include linear transformation, nonlinear transformation (such as logarithmic transformation), and other methods based on statistical models. Which method to choose depends on the specific image characteristics and the expected application scenario.

[0056] Histogram equalization is a technology that enhances image contrast. It redistributes the frequency of occurrence of each gray level in the image to make the grayscale distribution of the entire image more uniform, thereby improving the visibility of details. Simply put, it makes the dark part brighter and the bright part maintains an appropriate brightness, so that the information in the originally darker or brighter area can be better displayed. This is particularly important for identifying the internal structure of the liver, especially small lesions. Histogram equalization can not only improve the overall look and feel of the image, but also to a certain extent make up for the shortcomings caused by poor imaging quality. However, excessive equalization may cause image distortion, so a balance needs to be found during the implementation process, that is, it can effectively improve the image quality without introducing unnecessary artifacts or noise.

[0057] Specifically, the denoising module 130 is used to perform denoising on the standardized liver image to obtain a denoised liver image. The standardized liver image may contain various noises. These noises may be caused by factors such as electronic interference of the imaging device itself, body movement of the patient or imaging environment. The purpose of denoising is to remove these noises while retaining useful information of the liver image as much as possible, such as the contour of the liver, internal vascular structure, etc.

[0058] Accordingly, in the denoising module 130, the technical concept of the present application is to use the image analysis and processing technology based on artificial intelligence technology to perform discrete wavelet transform and liver frequency feature extraction on the standardized liver image, so as to intelligently generate a denoised liver image based on the graph autocorrelation gated aggregation representation between the frequency features of each liver sub-band. The present application can maximize the retention of the edges and other important details of the liver image while removing noise, and extract those frequency components that are critical to the liver anatomical structure, so as to improve the quality of the final image and make it closer to the real anatomical morphology. Moreover, compared with traditional filters, it has stronger generalization ability and better adaptability, and can adapt to noises of different types and intensities, as well as various complex liver lesion characteristics.

[0059] Figure 2 FIG. 1 is a block diagram of a denoising module in a liver image 3D reconstruction system based on artificial intelligence technology according to an embodiment of the present application. Specifically, Figure 2 As shown, the denoising module 130 includes: a standardized image discrete feature extraction unit 131, used to perform discrete transformation and liver frequency feature extraction on the standardized liver image to obtain a sequence of liver sub-band frequency features; a liver sub-band frequency aggregation unit 132, used to perform graph autocorrelation gating significant aggregation on the sequence of liver sub-band frequency features to obtain a liver sub-band frequency significant aggregation representation; a denoised liver image generation unit 133, used to generate the denoised liver image based on the liver sub-band frequency significant aggregation representation.

[0060] In the embodiment of the present application, the standardized image discrete feature extraction unit 131 is used to perform discrete transformation and liver frequency feature extraction on the standardized liver image to obtain a sequence of liver sub-band frequency features. Figure 3 FIG. 1 is a block diagram of a standardized image discrete feature extraction unit in a liver image three-dimensional reconstruction system based on artificial intelligence technology according to an embodiment of the present application. Specifically, Figure 3 As shown, the standardized image discrete feature extraction unit 131 includes: a liver image discrete wavelet transform subunit 1311, used to perform a two-dimensional discrete wavelet transform on the standardized liver image to generate a plurality of liver sub-band images; a liver sub-band frequency feature extraction subunit 1312, used to pass the plurality of liver sub-band images through a liver sub-band feature extractor based on a CNN-LSTM model to obtain a sequence of liver sub-band frequency feature vectors as a sequence of liver sub-band frequency features.

[0061] It should be understood that, considering that in medical images, the low-frequency sub-band mainly reflects the overall structure of the image, while the high-frequency sub-band captures details and edge information, and the frequency components displayed in different pathological structures are different. Therefore, in the technical solution of the present application, the standardized liver image is subjected to a two-dimensional discrete wavelet transform to decompose the frequency information of the image according to different scales to generate multiple liver sub-band images. It should be understood that the wavelet transform can separate the different frequency components contained in the original liver image (such as low-frequency liver contour and other general structural information and high-frequency liver texture, small blood vessels and other detailed information). This helps to analyze and process information with different frequency characteristics more targeted in the future, because different frequency parts are important for understanding the overall morphology and fine structure of the liver. For example, the low-frequency part can reflect the overall shape of the liver, the direction of the main trunk of large blood vessels, etc., and the high-frequency part reflects more delicate blood vessel branches, tissue texture differences, etc. In particular, in different liver sub-band images, the differences in frequency characteristics of lesions (if any), normal liver tissue, and noise will be more obvious. For example, certain liver diseases may cause changes in the texture frequency of local tissues. In the specific subband after wavelet transform, this change will be more obvious than in the original image, making it easier to capture these abnormal features when extracting liver frequency features later. It also facilitates the distinction between true liver tissue features and interference features caused by noise, providing a clearer basis for denoising.

[0062] Accordingly, considering the complex physiological structure of the liver, its image features include both spatial texture, shape and other features, and there are interrelated feature relationships between different frequency sub-bands. Based on this, in the technical solution of the present application, the multiple liver sub-band images are processed by a liver sub-band feature extractor based on the CNN-LSTM model to obtain a sequence of liver sub-band frequency feature vectors. Specifically, the CNN in the CNN-LSTM model has a powerful local feature extraction capability. It performs convolution operations by sliding the convolution kernel on the image, and can automatically learn the local features of different positions in the liver sub-band image, such as the local morphology of vascular texture in the liver sub-band image, the local changes in the liver tissue boundary and other features. For each liver sub-band image, which has spatial structural information, CNN can effectively capture the rich spatial feature patterns, and these features are crucial for accurately characterizing the state of the liver at different frequency levels. At the same time, the features in the liver image are often correlated, especially the multiple liver sub-band images obtained by wavelet transform, and the different sub-bands are sequential and interrelated in frequency. LSTM is good at processing data with sequence characteristics. It can remember the relevant information in the previous sub-band image features and combine it with the current sub-band image features, solving the gradient vanishing or gradient exploding problems that are prone to occur in traditional neural networks when processing long sequence data, thereby better modeling the relationship between the features of multiple liver sub-band images. Therefore, the CNN-LSTM model can more fully mine the complex feature information in the liver sub-band images and extract accurate liver sub-band frequency feature vectors from multiple aspects such as the spatial structure of different liver sub-bands and the sequential association between them.

[0063] In an embodiment of the present application, the liver sub-band frequency aggregation unit 132 is used to perform graph autocorrelation gating significant aggregation on the sequence of liver sub-band frequency features to obtain a liver sub-band frequency significant aggregation representation. Furthermore, considering that in the sequence of liver sub-band frequency feature vectors, the different sub-band frequency features represented by each vector are not independent of each other, but have complex internal correlations. For example, there is a potential dependency between the overall contour features of the liver in the low-frequency sub-band and the liver texture detail features in the high-frequency sub-band, and they jointly reflect the complete structure and state of the liver. And different features have different degrees of importance for accurately describing the liver and distinguishing liver tissue, diseased tissue (if any) and noise, that is, there are significant differences. Therefore, in order to capture this complex correlation structure and at the same time dig out those features that are more critical and more significant for liver image representation, in the technical solution of the present application, the sequence of liver sub-band frequency features is significantly aggregated by graph autocorrelation gating to obtain a liver sub-band frequency significant aggregation representation.

[0064] Figure 4FIG. 4 is a block diagram of a liver sub-band frequency aggregation unit in a liver image three-dimensional reconstruction system based on artificial intelligence technology according to an embodiment of the present application. Specifically, Figure 4 As shown, the liver subband frequency aggregation unit 132 includes: a liver subband frequency hyperbolic mapping subunit 1321, which is used to map each liver subband frequency feature vector in the sequence of the liver subband frequency feature vectors to the hyperbolic space to obtain a sequence of liver subband frequency feature vectors after the hyperbolic space mapping; a liver subband frequency graph walk topology feature calculation subunit 1322, which is used to calculate the graph walk topology feature of the sequence of liver subband frequency feature vectors after the hyperbolic space mapping to obtain a liver subband frequency graph walk topology feature matrix; a liver subband frequency feature semantic enhancement subunit 1321, which is used to map each liver subband frequency feature vector in the sequence of the liver subband frequency feature vectors to the hyperbolic space to obtain a liver subband frequency graph walk topology feature matrix; a liver subband frequency feature semantic enhancement subunit 1322, which is used to calculate the graph walk topology feature of the sequence of liver subband frequency feature vectors after the hyperbolic space mapping to obtain a liver subband frequency graph walk topology feature matrix; a liver subband frequency feature semantic enhancement subunit 1321, which is used to map each liver subband frequency feature vector in the sequence of the liver subband frequency feature vectors to ... Unit 1323 is used to perform a global context walk of a graph convolution model on the liver subband frequency graph walk topological feature matrix and the sequence of liver subband frequency feature vectors after hyperbolic space mapping to obtain a sequence of liver subband frequency context semantic enhancement feature vectors; a liver subband frequency feature significant aggregation subunit 1324 is used to calculate a significant aggregation representation of the sequence of liver subband frequency feature vectors based on the sequence of liver subband frequency context semantic enhancement feature vectors to obtain a liver subband frequency significant aggregation representation vector as the liver subband frequency significant aggregation representation.

[0065] Specifically, the processing process of the liver sub-band frequency hyperbolic mapping sub-unit 1321 can be expressed by the formula:

[0066] ;

[0067] ;

[0068] in, is the sequence of liver sub-band frequency feature vectors, , , , and are the first, second, and third sub-band frequency feature vectors in the sequence of the liver sub-band. , and liver subband frequency eigenvectors, and are the first weight matrix in hyperbolic space and the second weight matrix in hyperbolic space respectively, is the first in the sequence of liver subband frequency feature vectors after hyperbolic space mapping Liver subband frequency eigenvector after hyperbolic space mapping.

[0069] It should be understood that in order to effectively characterize the complex relationship between different areas of the liver, each liver sub-band frequency feature vector is mapped to the hyperbolic space to obtain a sequence of liver sub-band frequency feature vectors after hyperbolic space mapping. That is, the hyperbolic space has unique geometric properties and can better process hierarchical data. The sub-band frequency feature vectors of the liver image often have complex hierarchical and nested relationships, and mapping them to the hyperbolic space can better retain these inherent hierarchical information.

[0070] Specifically, in the embodiment of the present application, the liver sub-band frequency graph wandering topological feature calculation subunit 1322 is used to:

[0071] The Poincare distance between any two liver subband frequency feature vectors after hyperbolic space mapping in the sequence of liver subband frequency feature vectors after hyperbolic space mapping is calculated to obtain the liver subband frequency graph walk topology matrix. The process can be expressed by the formula:

[0072] ;

[0073] in, and are the first and second frequency eigenvectors of the liver subband after hyperbolic space mapping. and The liver subband frequency eigenvector after hyperbolic space mapping, is the square of the vector's magnitude, is the inverse hyperbolic cosine function, For calculation and The Poincare distance between is the first in the liver sub-band frequency map walk topology matrix The eigenvalues ​​of the positions;

[0074] The liver sub-band frequency graph walk topology matrix is ​​subjected to dilated convolution coding to obtain the liver sub-band frequency graph walk topology feature matrix. The process can be expressed by the formula:

[0075] ;

[0076] in, is the liver sub-band frequency map walk topology matrix, is the dilated convolutional coding, It is the topological characteristic matrix of the liver sub-band frequency map.

[0077] Accordingly, the Poincare distance between any two liver subband frequency feature vectors after hyperbolic space mapping is calculated to construct the liver subband frequency graph walk topology matrix. In other words, the Poincare distance is a distance metric suitable for hyperbolic space, which can quantify the degree of difference between two nodes (here, liver subbands), reflecting the similarities and differences between different liver subband frequency features, and providing basic connection structure information for subsequent graph walk operations.

[0078] It should be understood that the liver sub-band frequency graph wandering topology matrix is ​​subjected to dilated convolutional coding to mine feature relationships at different levels in the topology matrix and context information in a larger range, thereby obtaining the liver sub-band frequency graph wandering topology feature matrix.

[0079] Specifically, the processing process of the liver sub-band frequency feature semantic enhancement sub-unit 1323 can be expressed by the formula:

[0080] ;

[0081] in, is the first in the sequence of liver subband frequency feature vectors after hyperbolic space mapping The liver subband frequency eigenvector after hyperbolic space mapping, is the topological characteristic matrix of the liver sub-band frequency map, is the graph convolutional coding, is the first in the sequence of liver subband frequency context semantic enhancement feature vectors The frequency context semantic enhancement feature vector of liver sub-bands.

[0082] It should be understood that the global context walk of the graph convolution model on the topological feature matrix of the liver subband frequency graph and the sequence of liver subband frequency feature vectors after hyperbolic space mapping fully utilizes the context information in the graph structure, so that each feature can fuse information from other related features, thereby strengthening its semantic representation and obtaining a sequence of liver subband frequency context semantically enhanced feature vectors.

[0083] Specifically, in the embodiment of the present application, the liver sub-band frequency feature significant aggregation sub-unit 1324 is used to:

[0084] Autocorrelation gating is performed on each group of corresponding liver sub-band frequency context semantic enhancement feature vectors and liver sub-band frequency feature vectors in the sequence of liver sub-band frequency context semantic enhancement feature vectors and the sequence of liver sub-band frequency feature vectors to obtain a sequence of liver sub-band frequency autocorrelation gated significant confidence factors. The process can be expressed by the formula:

[0085] ;

[0086] in, is the first in the sequence of the liver subband frequency feature vectors liver subband frequency eigenvectors, is the first in the sequence of liver subband frequency context semantic enhancement feature vectors liver subband frequency context semantic enhancement feature vector, is the normalization function, To subtract by position point, It is calculated and The autocorrelation gated significance confidence factor between is the number of vectors in the sequence of liver subband frequency eigenvectors, is the sequence of the liver subband frequency autocorrelation gated significant confidence factors The confidence factor of the gated significance of the autocorrelation of the liver sub-band frequency;

[0087] The sequence of the liver sub-band frequency autocorrelation gated significant confidence factors is normalized based on the Softmax function to obtain a sequence of liver sub-band frequency autocorrelation gated significant confidence weight factors. The process can be expressed by the formula:

[0088] ;

[0089] in, is the sequence of the liver subband frequency autocorrelation gated significant confidence factors The liver sub-band frequency autocorrelation gated significant confidence factor, is the sequence of the liver subband frequency autocorrelation gated significant confidence factors The liver sub-band frequency autocorrelation gated significant confidence factor, represents the exponential function value with the natural constant e as the base, is the number of liver sub-band frequency autocorrelation gated significant confidence factors in the sequence of liver sub-band frequency autocorrelation gated significant confidence factors, is the sequence of liver subband frequency autocorrelation gated significant confidence weight factors The confidence weight factor of the gated significance of the autocorrelation of the liver sub-band frequency;

[0090] Based on the sequence of the liver sub-band frequency autocorrelation gated significant confidence weight factors, the position-weighted sum of the sequence of the liver sub-band frequency feature vectors is calculated to obtain the liver sub-band frequency significant aggregation representation vector. This process can be expressed by the formula:

[0091] ;

[0092] in, is the first in the sequence of the liver subband frequency feature vectors liver subband frequency eigenvectors, is the sequence of liver subband frequency autocorrelation gated significant confidence weight factors The liver sub-band frequency autocorrelation gated significant confidence weight factor, is the number of vectors in the sequence of liver subband frequency eigenvectors, For point-by-position multiplication, is the frequency-significant aggregation representation vector of the liver sub-bands.

[0093] Accordingly, each group of corresponding liver subband frequency context semantic enhancement feature vectors and liver subband frequency feature vectors are subjected to autocorrelation gating processing to adaptively measure the correlation and importance between the new feature vector of each liver subband frequency feature vector after context enhancement and the original feature vector, and obtain a sequence of liver subband frequency autocorrelation gated significant confidence factors.

[0094] Next, each liver sub-band frequency autocorrelation gated significant confidence factor is normalized based on the Softmax function to utilize the Softmax function to convert a set of numerical values ​​into a form representing probability distribution, thereby obtaining a sequence of liver sub-band frequency autocorrelation gated significant confidence weight factors.

[0095] Finally, the sequence of liver subband frequency feature vectors is aggregated according to the sequence of liver subband frequency autocorrelation gated significant confidence weight factors by means of position-weighted sum to form a more representative and significant liver subband frequency significant aggregate representation vector.

[0096] In an embodiment of the present application, the denoised liver image generation unit 133 is used to generate the denoised liver image based on the liver sub-band frequency significant aggregation representation. Specifically, in an embodiment of the present application, the denoised liver image generation unit 133 is used to: after adding generation prompt information at the tail of the liver sub-band frequency significant aggregation representation vector, it is input together with the standardized liver image into the AIGC-based liver image denoising generator to obtain the denoised liver image, and the generation prompt information is: denoise the standardized liver image based on the input corpus. That is, the AIGC model has learned a lot of image-related knowledge and patterns during the training process and has a strong generalization ability. By inputting the liver sub-band frequency significant aggregation representation vector (which contains the key feature information extracted from the liver image through multi-step analysis) together with the prompt information with clear denoising, the model can combine its own learned general image denoising knowledge and the key features unique to liver images to better understand the current denoising needs of liver images, such as the electronic noise generated by the equipment, artifacts in the imaging process, etc., while retaining the useful information of the liver's own tissue structure, vascular texture, etc. as completely as possible, thereby using its powerful generation ability to generate denoised liver images that meet the requirements.

[0097] Specifically, this process first involves preparing the liver subband frequency-significant aggregate representation vector. Next, the generation hint information is added to the tail of the liver subband frequency-significant aggregate representation vector. This hint information is essentially an instruction to the generator, clearly informing it that it needs to perform denoising on the input standardized liver image, and emphasizes that the operation should be based on the input corpus. The "input corpus" here refers to the preprocessed liver image and its associated feature information, which together constitute the context of the denoising process. In this way, the generation hint information acts as a bridge, allowing the generator to combine its own learned general image processing knowledge and the background information unique to the current image to more accurately understand and respond to denoising needs.

[0098] Then, the liver subband frequency significant aggregation representation vector with generation hint information is input into the liver image denoising generator based on AIGC together with the normalized liver image. This generator is a deep neural network model that has been trained extensively and has strong generalization and adaptability. It can automatically analyze the information contained in the input data while receiving it, including but not limited to the spatial structure, frequency distribution and possible noise patterns of the image. On this basis, the generator uses its internal knowledge base to adopt appropriate algorithms and techniques to remove noise, such as adjusting the filter strength or selecting a specific denoising method through the learned characteristics of different types of noise. At the same time, the generator will pay special attention to protecting those details that are critical to the anatomical structure of the liver, such as edge contours, vascular textures, etc., to avoid the loss of useful information due to excessive processing.

[0099] To ensure the denoising effect, the generator continuously evaluates the quality of the output results throughout the processing and adjusts parameter settings as needed. For example, when it detects that some areas may be over-smoothed, the denoising intensity will be appropriately reduced; in the face of high noise density areas, the processing intensity will be increased to ensure that the interference is completely removed. In addition, the generator will also consider the unique artifact problems introduced by different imaging devices, such as metal artifacts in CT scans or motion artifacts in MRI, and design solutions specifically for these problems to improve the quality of the final image.

[0100] Finally, after a series of complex and sophisticated operations, the generator outputs a denoised liver image. This image not only eliminates various noises in the original image, but also completely preserves the tissue structure and characteristic information of the liver itself. High-quality denoised images lay a good foundation for further medical analysis and provide doctors with a clearer and more accurate basis for diagnosis.

[0101] It should be understood that, here, in the case where each liver subband frequency feature vector in the sequence of liver subband frequency feature vectors respectively represents the image semantic features of the corresponding liver subband image based on the short-range-long-range bidirectional correlation between local image semantics, when performing feature sequence aggregation analysis based on graph walk autocorrelation gating, the misalignment of graph walk autocorrelations between local image semantic domains will also cause the richness of the sequence aggregation distribution of the liver subband frequency significant aggregation representation vector, resulting in repetition of unstructured probability mapping, affecting the accuracy of the generated results.

[0102] In a preferred example, after adding generation prompt information to the tail of the liver subband frequency significant aggregation representation vector, inputting it together with the standardized liver image into the AIGC-based liver image denoising generator to obtain the denoised liver image includes:

[0103] Determine the number of non-zero eigenvalues ​​in the liver subband frequency significant aggregation representation vector, and subtract the number of non-zero eigenvalues ​​from the length of the liver subband frequency significant aggregation representation vector to obtain the liver subband frequency significant aggregation zero-dimensional value ;

[0104] Calculate the liver subband frequency significant aggregation zero differentiable value after subtracting one from the liver subband frequency significant aggregation zero-dimensional value The first liver subband frequency significant aggregation field fitting value obtained by multiplying and dividing by the liver subband frequency significant aggregation zero-dimensional value and the second liver sub-band frequency significantly aggregated field fitting value ;

[0105] Calculate the square root of the sum of squares of all eigenvalues ​​of the liver sub-band frequency significant aggregation representation vector to obtain the liver sub-band frequency significant aggregation pattern representation value ;

[0106] The liver subband frequency significantly aggregates zero differentiable values The liver subband frequency significant aggregation representation vector is calculated as an index for each eigenvalue The power function , and multiply it by the significant aggregate field fitted value of the first liver sub-band frequency To obtain the local representation value of liver sub-band frequency significant aggregation ;

[0107] Calculate each eigenvalue of the liver subband frequency significant aggregation representation vector The field fitting value with significant aggregation with the second liver sub-band frequency and the significant aggregation pattern characterization values ​​of the liver sub-band frequencies The product of the liver sub-band frequency significant aggregation center directivity value is obtained ;

[0108] The liver sub-band frequency is significantly aggregated into local representation values The liver sub-band frequency significantly aggregates the central pointing value The weighted difference The composition of the optimized liver sub-band frequency-significant aggregate representation vector;

[0109] After adding generation prompt information to the tail of the optimized liver subband frequency significant aggregation representation vector, it is input into an AIGC-based liver image denoising generator together with the normalized liver image to obtain the denoised liver image.

[0110] Therefore, based on the vector field fitting law of high-dimensional manifold, the isolated zeros in the liver subband frequency-significant aggregation representation vector are used as the differentiable dimension of the high-dimensional manifold, so as to fix the normal direction pointing to the distribution center near the local coordinates represented by the eigenvalue of the liver subband frequency-significant aggregation representation vector, thereby avoiding invalid repetition in the probability mapping process caused by the unstructured characteristics of the rich distribution of the liver subband frequency-significant aggregation representation vector by effectively aligning the characteristic distribution pattern of the liver subband frequency-significant aggregation representation vector to the probability density differentiable field, thereby improving the accuracy of the generated result obtained by inputting it into the AIGC-based liver image denoising generator.

[0111] In summary, the denoising module 130 is explained, which uses image analysis and processing technology based on artificial intelligence technology to perform discrete wavelet transform and liver frequency feature extraction on the standardized liver image, so as to intelligently generate a denoised liver image based on the graph autocorrelation gated aggregation representation between the frequency features of each liver sub-band. The present application can retain the edges and other important details of the liver image to the maximum extent while removing noise, and extract those frequency components that are crucial to the liver anatomical structure, so as to improve the quality of the final image and make it closer to the real anatomical morphology. Moreover, compared with traditional filters, it has stronger generalization ability and better adaptability, and can adapt to noise of different types and intensities, as well as various complex liver lesion characteristics.

[0112] Specifically, the semantic segmentation module 140 is used to perform semantic segmentation on the denoised liver image to obtain a semantic segmentation image of the liver. Specifically, in an embodiment of the present application, the semantic segmentation module 140 is used to: use a semantic segmentation model based on a convolutional neural network to perform semantic segmentation on the denoised liver image to obtain the semantic segmentation image of the liver. It should be understood that semantic segmentation is to divide each pixel in the liver image according to the category to which it belongs, such as distinguishing liver tissue, blood vessels, and diseased tissue (if any). The semantic segmentation model based on the convolutional neural network can learn the characteristics of different liver tissues and structures, and then perform segmentation prediction on the new liver image to obtain a semantic segmentation image of the liver, in which each pixel is assigned a specific category label.

[0113] The following are specific steps for performing semantic segmentation on the denoised liver image using a semantic segmentation model based on a convolutional neural network:

[0114] First, we need to build an effective deep learning framework. Convolutional neural networks have become the tool of choice because of their powerful feature extraction capabilities. They can automatically learn visual patterns of different tissue types from a large number of annotated liver image datasets. Specifically, CNN captures local features in the image through multi-layer convolution operations and reduces spatial dimensions through pooling layers while retaining important information. These features are then passed to fully connected layers or other structures for the final classification decision. In order to ensure that the model can handle the complex liver anatomical structure and possible lesion changes, researchers will adopt more complex network architectures, such as U-Net, DeepLab, etc., which introduce mechanisms such as skip connections or hole convolutions to enhance the spatial resolution and contextual understanding capabilities of the model.

[0115] Once a suitable network architecture is prepared, the next step is to collect and preprocess a large number of high-quality liver image datasets. This includes obtaining images containing various types of liver conditions (healthy and diseased states) from multiple sources and annotating them, that is, assigning the correct label to each pixel in each image. This annotation process is usually completed by experienced radiologists, who accurately mark the liver, blood vessels, and any abnormal areas based on their medical expertise. Due to the high degree of professionalism and complexity of liver images, high-quality datasets are particularly critical for training semantic segmentation models with good performance. In addition, the original images need to be standardized, such as resizing, grayscale value normalization, and histogram equalization as mentioned above, to ensure that all images input to the model are under the same standard, thereby improving the generalization ability and stability of the model.

[0116] After that, the semantic segmentation model was trained. During the training process, the model continuously receives labeled liver images as input and tries to predict the category to which each pixel belongs. Then, the difference between the predicted result and the true label is calculated, and the network weights are adjusted using the back-propagation algorithm to gradually optimize the model parameters until the loss function converges or the predetermined stopping condition is met. It is worth noting that during training, the researchers will also adopt a variety of strategies to enhance the performance of the model, such as data augmentation technology (rotation, flipping, scaling and other transformations) to make the model more robust; and transfer learning methods, using models that have been pre-trained on large-scale general images to initialize some network layers, accelerate convergence and improve the final effect.

[0117] Once the training is completed, the resulting semantic segmentation model can be applied to new denoised liver images. At this stage, the model takes new, unseen images as input and outputs one or more probability maps of the same size, where each pixel corresponds to the probability value of belonging to a certain category. In order to obtain the final segmentation result, threshold processing or maximum likelihood estimation is usually applied to these probability maps, and the most likely category is selected as the label of the pixel. That is, a complete semantic segmentation image of the liver is generated, in which different colors or grayscale levels represent different tissue types. For some cases where the boundaries are blurred or difficult to determine, morphological operations can be combined to further optimize the segmentation boundaries to ensure that the results are both accurate and smooth.

[0118] Specifically, the optimization segmentation module 150 is used to optimize the segmentation boundary by morphological adjustment of the semantic segmentation image of the liver to obtain a boundary-optimized semantic segmentation image of the liver. Accordingly, the semantic segmentation image of the liver obtained by semantic segmentation may have inaccuracies or jaggedness on the segmentation boundary. The optimization segmentation module optimizes these boundaries through morphological adjustment. Morphological operations include corrosion, expansion, opening and closing operations. For example, the corrosion operation can shrink the boundary inward to remove small protrusions or isolated small areas on the boundary; the expansion operation expands the boundary outward to fill small depressions on the boundary or connect adjacent areas. Through the combination of these morphological operations, a boundary-optimized semantic segmentation image of the liver with smoother and more accurate boundaries can be obtained.

[0119] After receiving the semantic segmentation image of the liver from the semantic segmentation module, the first challenge is how to effectively remove the noise points and small areas that may exist in the image. These unnecessary details not only affect the visual effect, but also interfere with the subsequent 3D reconstruction process. To solve this problem, the optimized segmentation module uses the erosion operation, which is a process of shrinking the white (or specified color) area in the image, which can eliminate small protrusions or isolated small spots while maintaining the integrity of larger structures. Specifically, the erosion operation defines a structural element (usually a circle or square), scans the image according to the shape of the element, and retains only those pixels that are completely contained in the white area. This method helps to clean up the burrs on the edges of the image, making the liver contour clearer and smoother.

[0120] However, simple erosion may cause some important structures to shrink or even disappear, especially when dealing with thinner blood vessels or other fine structures. Therefore, after the initial denoising, the dilation operation is applied as a supplement. Dilation is the opposite of erosion. It is the process of expanding the white area in the image. It can restore the previously compressed part by filling the depression or connecting adjacent but unconnected small areas. Dilation also relies on structural elements, but this time it is to increase rather than reduce the white area. By setting the degree of dilation reasonably, the slight loss caused by erosion can be repaired without affecting the overall structure, and the segmentation boundary can be made more natural and smooth.

[0121] In addition to using erosion and dilation separately, the optimized segmentation module also combines them to form more complex morphological operations - opening and closing operations. The opening operation is a process of corrosion followed by dilation, which is mainly used to remove small objects in the image without significantly changing the size of larger objects; while the closing operation is a process of dilation followed by corrosion, which is used to close small holes in the image and smooth the contours of larger objects. These two combined operations are particularly suitable for processing liver images with complex boundaries because they can effectively improve the quality of segmentation results while preserving key anatomical structures. For example, when faced with tiny lesions or blood vessel branches on the surface of the liver, the opening operation can help remove artifacts, while the closing operation ensures that these subtle structures are not misclassified as background or other tissue types.

[0122] In order to further improve the optimization effect, conditional morphological operations can also be introduced, that is, whether to perform a certain morphological transformation can be selected according to specific conditions. This requires a set of rules to be defined in advance to determine which parts need special treatment. For example, for areas close to the edge of the liver, it may be more inclined to use gentler operations to avoid excessive modifications; while for internal structures, parameters can be flexibly adjusted according to actual conditions to achieve the best optimization effect. In addition, contextual information, such as the color distribution and texture characteristics of neighboring pixels, can be combined to guide the direction and intensity of morphological adjustments, making the entire optimization process more intelligent and adaptable.

[0123] During the entire optimization process, morphological adjustments are not static, but are continuously iterated and improved according to actual needs. Each morphological operation will produce a new image version, which will be evaluated by technicians to check whether there is over-segmentation or under-segmentation, and the parameters will be adjusted accordingly until they are satisfactory.

[0124] Specifically, the three-dimensional reconstruction module 160 is used to apply three-dimensional reconstruction technology to perform three-dimensional reconstruction on the boundary-optimized liver semantic segmentation image to obtain a three-dimensional liver model. That is, the semantic segmentation image of the liver after optimization and segmentation is three-dimensionally reconstructed. It uses computer graphics and image processing technology to convert a two-dimensional liver semantic segmentation image sequence into a three-dimensional model. This can help doctors more intuitively observe the overall morphology of the liver, vascular distribution, and the position and shape of the lesion in three-dimensional space. There are many methods for three-dimensional reconstruction, such as a voxel-based method, which stacks pixels in a two-dimensional image according to spatial positions to construct a three-dimensional voxel model; or a surface reconstruction method, which constructs a three-dimensional model by extracting the surface contour of the liver tissue.

[0125] After receiving the semantic segmentation image of the liver with optimized boundaries, the 3D reconstruction module first preprocesses it to prepare a data format suitable for 3D modeling.

[0126] Next, we enter the core 3D reconstruction stage. According to different application scenarios and technical routes, commonly used 3D reconstruction methods mainly include voxel-based methods and surface-based methods. The voxel-based method is similar to the process of stacking building blocks. It stacks up the pixels in each layer of the liver image one by one according to their spatial position information to form a 3D volume data set composed of countless small cubes (voxels). The advantage of this method is that it can directly obtain density information from the original image, which is suitable for representing the overall volume and internal tissue distribution of the liver; however, because the voxel model is relatively rough and it is difficult to finely display complex structures, further processing is usually required to achieve the ideal visual effect.

[0127] In contrast, surface-based methods focus more on the extraction and reconstruction of the liver surface contour. Specifically, this technology first looks for a series of key points on the liver image, which constitute the main characteristic curves of the liver surface. Then, using surface fitting algorithms such as triangulation or polygonization, these discrete key points are connected to form a closed and smooth three-dimensional surface. This process may involve complex mathematical operations, such as least squares method, radial basis function interpolation, etc., to ensure that the surface is as close to the actual liver shape as possible. Compared with the voxel method, the surface-based method can more delicately express the geometric shape of the liver, and is particularly suitable for observing subtle changes on the organ surface or locating lesions.

[0128] In order to improve the quality of reconstruction, the 3D reconstruction module also uses a variety of enhancement technologies. For example, texture mapping can project the color and brightness information in the 2D image onto the surface of the 3D model, making the model look more real and natural; lighting simulation simulates the effect of liver light reflection under different lighting conditions by setting a virtual light source, increasing the three-dimensional sense and layering. In addition, transparency control allows users to customize the transparency of different parts of the liver, making it convenient to view the internal structure without blocking the line of sight. These advanced features not only improve the aesthetics of the model, but also provide more valuable perspectives for clinical analysis. It is worth noting that in actual applications, 3D reconstruction does not exist in isolation, but is closely integrated with other medical imaging technologies and tools to serve the goal of precision medicine. For example, combined with virtual reality (VR) or augmented reality (AR) devices, doctors can explore the liver model in an immersive environment and even simulate the surgical process.

[0129] Finally, the liver model obtained after 3D reconstruction will be output into standard file formats, such as STL, OBJ, etc., to facilitate compatibility with other software platforms and support printing of physical models for teaching demonstrations.

[0130] Specifically, the display module 170 is used to display the three-dimensional liver model on a display screen. This allows doctors or related medical personnel to easily view the three-dimensional structure of the liver. Through appropriate visualization technology, such as adjusting the transparency, color, and lighting effects of the model, the detailed structure inside the liver can be better displayed, providing an intuitive image reference for medical behaviors such as disease diagnosis and surgical planning.

[0131] When the 3D model of the liver is ready, the first thing to do is to choose a suitable rendering engine to process this data. Modern computer graphics provides a wealth of options, such as OpenGL, DirectX, etc., which can efficiently manage and draw a large number of geometric shapes and texture details. In this application, a rendering engine that supports hardware acceleration and has good cross-platform compatibility is usually selected to ensure smooth operation on both high-performance workstations and ordinary clinical computers. In addition, in order to improve the visual quality, advanced lighting models such as global illumination, shadow casting, and reflection mapping are also used to make the surface of the liver present a realistic texture, helping doctors better judge tissue characteristics.

[0132] Next, in order to allow users to easily explore and operate the liver model, the display module integrates a variety of interactive functions. Through mouse or touch screen gestures, doctors can easily rotate, zoom, and pan the model to view its internal structure from all angles. For example, when examining the distribution of blood vessels, you can zoom in on a specific area for a closer look; when assessing the extent of the lesion, you may need to zoom out to get the overall layout. At the same time, it also supports multi-view switching, allowing users to examine the liver from different angles, including common orientations such as the front, side, and top, and even customize any perspective to meet personalized needs. For complex cases, you can also enable the section mode to cut the liver along any plane to reveal the information hidden inside it.

[0133] In addition to basic browsing operations, the display module also provides a wealth of annotation and measurement tools. Doctors can mark important anatomical locations or lesions on the model and add annotations for subsequent reference and communication. For tasks that require high precision, such as surgical planning, built-in measurement tools can also be used to calculate distances, angles, volumes and other parameters to assist in formulating treatment plans. In order to improve work efficiency, all operations are simplified to one-click completion as much as possible, reducing tedious steps and allowing users to focus on core business.

[0134] Considering the different device configurations that may be encountered in actual application scenarios, the display module pays special attention to performance optimization. It automatically adjusts the rendering accuracy according to the current system resource status to ensure good response speed even in a lower configuration environment. For example, when insufficient hardware performance is detected, the system will reduce the model resolution or reduce the use of special effects to prioritize interactive smoothness. At the same time, for high-end devices, its computing power is fully utilized to enable more advanced rendering options to present the most delicate picture effects. This intelligent adaptation mechanism ensures that the display module can run stably on a wide range of hardware platforms and adapt to the needs of various clinical environments.

[0135] In summary, a liver image three-dimensional reconstruction system 100 based on artificial intelligence technology according to an embodiment of the present application is described, which collects liver images of patients, then performs standardization processing to unify the image format, and then removes noise in the image to improve clarity. The segmentation boundary of the liver is optimized by morphological adjustment, and then the two-dimensional image is converted into a three-dimensional liver model using three-dimensional reconstruction technology. Finally, the model is displayed on the screen. These processes improve the accuracy and efficiency of image processing, provide intuitive and clear three-dimensional views for medical diagnosis and surgical planning, and enhance the accuracy and reliability of medical decision-making.

[0136] As described above, the liver image three-dimensional reconstruction system 100 based on artificial intelligence technology according to the embodiment of the present application can be implemented in various wireless terminals, such as a server with a liver image three-dimensional reconstruction algorithm based on artificial intelligence technology. In a possible implementation, the liver image three-dimensional reconstruction system 100 based on artificial intelligence technology according to the embodiment of the present application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the liver image three-dimensional reconstruction system 100 based on artificial intelligence technology can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the liver image three-dimensional reconstruction system 100 based on artificial intelligence technology can also be one of the many hardware modules of the wireless terminal.

[0137] Alternatively, in another example, the artificial intelligence technology-based three-dimensional reconstruction system for liver images 100 and the wireless terminal may also be separate devices, and the artificial intelligence technology-based three-dimensional reconstruction system for liver images 100 may be connected to the wireless terminal via a wired and / or wireless network, and transmit interactive information in accordance with an agreed data format.

[0138] Figure 5 FIG. 1 is a flow chart of a method for three-dimensional reconstruction of liver images based on artificial intelligence technology according to an embodiment of the present application. Figure 5 As shown, according to the embodiment of the present application, the method for three-dimensional reconstruction of liver images based on artificial intelligence technology includes: S110, collecting a liver image of a target patient object; S120, performing standardization processing on the liver image to obtain a standardized liver image; S130, performing denoising processing on the standardized liver image to obtain a denoised liver image; S140, performing semantic segmentation on the denoised liver image to obtain a liver semantic segmentation image; S150, performing morphological adjustment on the liver semantic segmentation image to optimize the segmentation boundary to obtain a boundary-optimized liver semantic segmentation image; S160, applying three-dimensional reconstruction The technology performs three-dimensional reconstruction on the boundary-optimized liver semantic segmentation image to obtain a three-dimensional liver model; S170, the three-dimensional liver model is displayed on a display screen; wherein, step S130, denoising the standardized liver image to obtain a denoised liver image, including: performing discrete transformation and liver feature extraction on the standardized liver image to obtain a sequence of liver sub-band frequency features; performing graph autocorrelation gated significant aggregation on the sequence of liver sub-band frequency features to obtain a liver sub-band frequency significant aggregation representation; based on the liver sub-band frequency significant aggregation representation, generating the denoised liver image.

[0139] Here, those skilled in the art can understand that the specific operations of each step in the above-mentioned liver image three-dimensional reconstruction method based on artificial intelligence technology have been referred to above. Figures 1 to 4The description of the liver image three-dimensional reconstruction system based on artificial intelligence technology has been introduced in detail, and therefore, its repeated description will be omitted.

[0140] The above has described various implementations of the present disclosure, and the above description is exemplary and not exhaustive. It is not limited to the disclosed implementations, and many modifications and changes are obvious to those of ordinary skill in the art without departing from the scope and spirit of the described implementations. The choice of terms used in this article is intended to best explain the principles of each implementation, practical application or improvement of technology in the market, or to enable other ordinary technicians in this technical field to understand the various embodiments disclosed herein.

Claims

1. A liver image three-dimensional reconstruction system based on artificial intelligence technology, characterized in that: include: A liver image acquisition module, used for acquiring a liver image of a target patient; A standardization module, used for performing standardization processing on the liver image to obtain a standardized liver image; a denoising module, configured to perform denoising processing on the standardized liver image to obtain a denoised liver image, wherein the denoising module comprises: a standardized image discrete feature extraction unit, configured to perform discrete transformation and liver frequency feature extraction on the standardized liver image to obtain a sequence of liver sub-band frequency features; a liver sub-band frequency aggregation unit, configured to perform graph autocorrelation gated significant aggregation on the sequence of liver sub-band frequency features to obtain a liver sub-band frequency significant aggregation representation; and a denoised liver image generation unit, configured to generate the denoised liver image based on the liver sub-band frequency significant aggregation representation; A semantic segmentation module, used for performing semantic segmentation on the denoised liver image to obtain a semantic segmentation image of the liver; An optimization segmentation module, used for optimizing the segmentation boundary of the liver semantic segmentation image by morphological adjustment to obtain a boundary-optimized liver semantic segmentation image; A three-dimensional reconstruction module, used for applying a three-dimensional reconstruction technology to perform three-dimensional reconstruction on the boundary-optimized liver semantic segmentation image to obtain a three-dimensional liver model; A display module, used for displaying the three-dimensional liver model on a display screen; Wherein, the liver sub-band frequency aggregation unit comprises: A liver subband frequency hyperbolic mapping subunit, used for mapping each liver subband frequency feature vector in the sequence of liver subband frequency feature vectors to a hyperbolic space to obtain a sequence of liver subband frequency feature vectors after hyperbolic space mapping; A liver sub-band frequency graph walk topology feature calculation subunit, used to calculate the graph walk topology feature of the sequence of liver sub-band frequency feature vectors after the hyperbolic space mapping to obtain a liver sub-band frequency graph walk topology feature matrix; A liver subband frequency feature semantic enhancement subunit, used for performing a global context walk of a graph convolution model on the liver subband frequency graph walk topological feature matrix and the sequence of liver subband frequency feature vectors after hyperbolic space mapping to obtain a sequence of liver subband frequency context semantic enhancement feature vectors; A liver sub-band frequency feature significant aggregation sub-unit is used to calculate a significant aggregation representation of the sequence of liver sub-band frequency feature vectors based on the sequence of liver sub-band frequency context semantic reinforcement feature vectors to obtain a liver sub-band frequency significant aggregation representation vector as the liver sub-band frequency significant aggregation representation; Wherein, the liver sub-band frequency characteristic significant aggregation sub-unit is used for: Performing autocorrelation gating processing on each group of corresponding liver sub-band frequency context semantic enhancement feature vectors and liver sub-band frequency feature vectors in the sequence of liver sub-band frequency context semantic enhancement feature vectors and the sequence of liver sub-band frequency feature vectors to obtain a sequence of liver sub-band frequency autocorrelation gated significant confidence factors; Performing a normalization process based on a Softmax function on the sequence of liver sub-band frequency autocorrelation gated significant confidence factors to obtain a sequence of liver sub-band frequency autocorrelation gated significant confidence weight factors; Based on the sequence of liver sub-band frequency autocorrelation gated significant confidence weight factors, a position-weighted sum of the sequence of liver sub-band frequency feature vectors is calculated to obtain the liver sub-band frequency significant aggregate representation vector.

2. The liver image three-dimensional reconstruction system based on artificial intelligence technology according to claim 1, characterized in that: The standardization process includes size adjustment, gray value normalization and histogram equalization.

3. The liver image three-dimensional reconstruction system based on artificial intelligence technology according to claim 2, characterized in that: The standardized image discrete feature extraction unit comprises: A liver image discrete wavelet transform subunit, used for performing a two-dimensional discrete wavelet transform on the standardized liver image to generate a plurality of liver sub-band images; The liver sub-band frequency feature extraction sub-unit is used to pass the multiple liver sub-band images through a liver sub-band feature extractor based on a CNN-LSTM model to obtain a sequence of liver sub-band frequency feature vectors as the sequence of liver sub-band frequency features.

4. The liver image three-dimensional reconstruction system based on artificial intelligence technology according to claim 3, characterized in that: The liver sub-band frequency graph wandering topological feature calculation subunit is used for: Calculating the Poincare distance between any two liver subband frequency feature vectors after hyperbolic space mapping in the sequence of liver subband frequency feature vectors after hyperbolic space mapping to obtain a liver subband frequency graph walk topology matrix; The liver sub-band frequency graph walk topology matrix is ​​subjected to dilated convolution coding to obtain the liver sub-band frequency graph walk topology feature matrix.

5. The liver image three-dimensional reconstruction system based on artificial intelligence technology according to claim 4, characterized in that: The denoised liver image generating unit is used to: after adding generation prompt information at the tail of the liver sub-band frequency significant aggregation representation vector, input it together with the standardized liver image into the AIGC-based liver image denoising generator to obtain the denoised liver image, wherein the generation prompt information is: denoising the standardized liver image based on the input corpus.

6. The liver image three-dimensional reconstruction system based on artificial intelligence technology according to claim 5, characterized in that: The semantic segmentation module is used to: use a semantic segmentation model based on a convolutional neural network to perform semantic segmentation on the denoised liver image to obtain the semantic segmentation image of the liver.

7. A method for three-dimensional reconstruction of liver images based on artificial intelligence technology, characterized in that: include: acquiring a liver image of a target patient subject; performing standardization processing on the liver image to obtain a standardized liver image; Denoising the standardized liver image to obtain a denoised liver image; semantically segmenting the denoised liver image to obtain a liver semantic segmentation image; morphologically adjusting the liver semantic segmentation image to optimize the segmentation boundary to obtain a boundary-optimized liver semantic segmentation image; and applying a three-dimensional reconstruction technique to perform three-dimensional reconstruction on the boundary-optimized liver semantic segmentation image to obtain a three-dimensional liver model; Displaying the three-dimensional liver model on a display screen; The step of performing denoising on the standardized liver image to obtain a denoised liver image includes: Performing discrete transformation and liver feature extraction on the standardized liver image to obtain a sequence of liver sub-band frequency features; Performing graph autocorrelation gating significant aggregation on the sequence of liver sub-band frequency features to obtain a liver sub-band frequency significant aggregation representation; generating the denoised liver image based on the liver sub-band frequency significant aggregation representation; The step of performing graph autocorrelation gating significant aggregation on the sequence of liver sub-band frequency features to obtain a significant aggregation representation of liver sub-band frequencies includes: Mapping each liver sub-band frequency feature vector in the sequence of liver sub-band frequency feature vectors to a hyperbolic space to obtain a sequence of liver sub-band frequency feature vectors after hyperbolic space mapping; Calculating the graph walk topology characteristics of the sequence of liver sub-band frequency feature vectors after the hyperbolic space mapping to obtain a liver sub-band frequency graph walk topology feature matrix; Performing a global context walk of a graph convolution model on the liver subband frequency graph walk topological feature matrix and the sequence of liver subband frequency feature vectors after hyperbolic space mapping to obtain a sequence of liver subband frequency context semantic enhancement feature vectors; Based on the sequence of liver sub-band frequency context semantic enhancement feature vectors, calculating a significant aggregation representation of the sequence of liver sub-band frequency feature vectors to obtain a liver sub-band frequency significant aggregation representation vector as the liver sub-band frequency significant aggregation representation; Wherein, based on the sequence of the liver sub-band frequency context semantic enhancement feature vectors, calculating the significant aggregation representation of the sequence of the liver sub-band frequency feature vectors to obtain a liver sub-band frequency significant aggregation representation vector as the liver sub-band frequency significant aggregation representation, comprises: Performing autocorrelation gating processing on each group of corresponding liver sub-band frequency context semantic enhancement feature vectors and liver sub-band frequency feature vectors in the sequence of liver sub-band frequency context semantic enhancement feature vectors and the sequence of liver sub-band frequency feature vectors to obtain a sequence of liver sub-band frequency autocorrelation gated significant confidence factors; Performing a normalization process based on a Softmax function on the sequence of liver sub-band frequency autocorrelation gated significant confidence factors to obtain a sequence of liver sub-band frequency autocorrelation gated significant confidence weight factors; Based on the sequence of liver sub-band frequency autocorrelation gated significant confidence weight factors, a position-weighted sum of the sequence of liver sub-band frequency feature vectors is calculated to obtain the liver sub-band frequency significant aggregate representation vector.

Images