Intelligent chest CT image processing system

Abstract

The invention discloses an intelligent chest CT image processing system which comprises an image acquisition module, a multi-scale filtering processing module, a frequency domain edge enhancement module, an edge fusion correction module, an enhancement loss design module, a chest CT image segmentation model establishment module and a chest CT image processing module. The invention belongs to the field of image processing, and particularly relates to an intelligent chest CT (computed tomography) image processing system, which is characterized in that differentiated processing is performed in different frequency spectrum sections according to noise characteristics, key points are guided to be reserved, and blood vessel walking is enhanced; through edge fusion correction, adaptive reinforcement of small-range nodules and large-range lesions is considered; the branch A is used for refining the nodule edge and artifact weak contrast, and the branch B is used for strengthening the lung lobe macroscopic consistency; the decoding layer inhibits a learned region, focuses a blood vessel and a lung parenchyma junction through interface focusing; the loss is introduced into boundary smoothing, the overall segmentation contour is optimized, and false positive burr-shaped noise is reduced; and the chest CT image processing accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the field of image processing, and specifically refers to an intelligent chest CT image processing system. Background Art

[0002] A chest CT image processing system is a system that performs operations such as noise reduction, enhancement, preprocessing of correction, segmentation and recognition of different tissue organs, three-dimensional reconstruction, measurement and analysis, and lesion detection on chest CT scan images. However, general chest CT image processing systems have problems such as insufficient sensitivity to small blood vessels, improper processing of frequency-domain noise and artifacts, and lack of specialized guidance based on the anatomical structures of blood vessels and bronchi, resulting in poor final processing effects; general chest CT image processing systems have problems such as insufficient response to multi-scale lesions and lack of a mechanism for specifically focusing on boundaries, resulting in over-smoothing or breakage of the segmentation of the blood vessel, lung parenchyma interface and slender branches, and poor accuracy of the processing results. Summary of the Invention

[0003] In view of the above situation, to overcome the defects of the prior art, the present invention provides an intelligent chest CT image processing system. Aiming at the problems of general chest CT image processing systems, such as insufficient sensitivity to small blood vessels, improper processing of frequency-domain noise and artifacts, and lack of specialized guidance based on the anatomical structures of blood vessels and bronchi, resulting in poor final processing effects, this solution calculates gradients at voxel intervals of 1 mm and 2 mm respectively, and can sensitively respond to small nodules and large-range pleural edges; performs differential processing on different frequency bands according to the noise characteristics, guides the key retention and enhancement of the blood vessel course, while suppressing high-frequency noise and restoring large low-frequency structures; through edge fusion correction, takes into account the adaptive enhancement of small-range nodules and large-range lesions; thereby improving the chest CT image processing effect. Aiming at the problems of general chest CT image processing systems, such as insufficient response to multi-scale lesions and lack of a mechanism for specifically focusing on boundaries, resulting in over-smoothing or breakage of the segmentation of the blood vessel, lung parenchyma interface and slender branches, and poor accuracy of the processing results, this solution refines the nodule edge and weak contrast of artifacts through branch A, and strengthens the macroscopic consistency of the lung lobe through branch B; through five-layer parallel refinement + global common downsampling, takes into account small-range ground-glass nodules and large-range lung lobe structures at the same time; the decoding layer focuses on the interface, suppresses the learned area, and focuses on the blood vessel, lung parenchyma junction; the loss introduces boundary smoothness, optimizes the overall segmentation contour, and reduces false positive spiky noise; thereby improving the accuracy of chest CT image processing.

[0004] The technical solution adopted by the present invention is as follows: An intelligent chest CT image processing system provided by the present invention includes an image acquisition module, a multi-scale filtering processing module, a frequency-domain edge enhancement module, an edge fusion correction module, an enhanced loss design module, a chest CT image segmentation model establishment module, and a chest CT image processing module;

[0005] The image acquisition module acquires historical chest CTs and constructs a chest CT sample set;

[0006] The multi-scale filtering processing module calculates the change intensity maps for chest CT images at voxel intervals of 1 mm and 2 mm respectively, and generates a multi-scale change intensity map through upsampling alignment and scale-weighted fusion;

[0007] The frequency-domain edge enhancement module performs 2D FFT on each layer of the multi-scale change intensity map to separate it into three frequency spectrum segments of low / mid / high frequencies, extracts multi-scale residual features using 5×5 and 1×1 convolutions, and after integrating the vascular response into the mid-frequency residuals, reconstructs to obtain the repaired chest CT gradient map;

[0008] The edge fusion correction module splices the original input features of the chest CT image and the repaired gradient map, and successively fuses the features through channel attention and attention map weighted fusion, and then outputs the chest CT correction feature map through three layers of 3×3 convolutions and residual connections;

[0009] The enhanced loss design module constructs the total loss function of the enhancement network based on high-quality chest CT reference images;

[0010] The chest CT image segmentation model establishment module constructs an encoder-bottleneck-decoder network and trains the chest CT image segmentation model with normalized active contour and boundary length smoothing loss;

[0011] The chest CT image processing module processes real-time chest CT images based on the multi-scale filtering processing module, the frequency-domain edge enhancement module, the edge fusion correction module, and the chest CT image segmentation model.

[0012] Furthermore, the image acquisition module acquires historical chest CT images; normalizes the intensity values; annotates high-quality chest CT reference images; annotates all historical chest CT images; and obtains a chest CT sample set.

[0013] Furthermore, the multi-scale filtering processing module selects multi-scale Sobel filtering for chest CT images to estimate the gradient map , expressed as: ; performs upsampling alignment and weighted fusion on the Sobel outputs with voxel intervals of 1 mm and 2 mm respectively, expressed as: ; where , and are the gradient components of the filtered output; is the fused multi-scale change intensity map; is the scale weighting coefficient; and are the change intensity maps calculated at voxel intervals of 1 mm and 2 mm respectively.

[0014] Furthermore, on the basis of correcting the amplitude spectrum in the frequency domain, the frequency domain edge enhancement module adds multi-spectrum band analysis and Frangi-vessel enhancement guidance; performs 2D FFT on each layer of CT gradient map and separates it into three spectrum bands of low frequency / middle frequency / high frequency, which is expressed as: ; ; where is the Fourier transform of the 2D slice; A is the frequency domain separation result, , and respectively represent the spectrum band components of low frequency, middle frequency and high frequency; the multi-scale features are extracted through the convolutional network in the spatial domain, which is expressed as: ; where is the multi-scale feature residual, and k is the spectrum band index; is to use a 5×5 convolutional kernel on the 2D plane; is used for channel integration; is the activation function; for the vascular structure, the Frangi filter is used to generate the vascular response map V and fuse it with , which is expressed as: ; where is the middle frequency spectrum band component after the vascular response fusion; γ is the vascular enhancement coefficient; the reconstruction fusion is expressed as: ; where is the finally enhanced and repaired gradient map; IFFT is the inverse Fourier transform reconstruction; P is the phase spectrum.

[0015] Furthermore, in addition to the channel attention, the edge fusion correction module adds spatial-slice attention to dynamically adjust the gradient and the chest CT image features; and obtains through the joint features and dual attention: ; the channel attention is expressed as: ; ; where is the original input feature map; U is the joint feature after splicing; is the channel attention vector; and are the two split channel attention sub-vectors; is the linear transformation; LayerNorm is the layer normalization; is the segmentation; Sigmoid is the Sigmoid activation function; perform 3D convolution + Sigmoid processing on U to generate the spatial attention map , which is expressed as: ; the feature fusion is expressed as: ; ; finally, it is output through three layers of 3×3 convolution + residual, which is expressed as: ; where is a 3D convolution; S is the feature fusion result; is the weighted feature map; , and are three convolution operations; is the final enhanced output feature map.

[0016] Furthermore, the enhancement loss design module performs enhancement processing on the chest CT sample set based on the multi-scale filtering processing module, the frequency domain edge enhancement module, and the edge fusion correction module, makes a comparison based on the high-quality chest CT reference image, and obtains the chest CT segmentation sample set through loss training; adds edge loss; pixel consistency loss is expressed as: ; edge consistency loss , is expressed as: ; the final total loss function is expressed as: ; where H and W are the height and width of the chest CT image; is the number of channels and rows and columns; is the enhanced pixel value; is the pixel value of the high-quality chest CT reference image; is the smoothing term; is the edge detection operator; is the loss weight; is the gradient constraint term.

[0017] Furthermore, the chest CT image segmentation model establishment module establishes a segmentation model based on the chest CT segmentation sample set. The overall process includes three major stages: encoding, bottleneck, and decoding. It introduces double-branch refined feature extraction in the encoding layer, applies multi-scale hybrid convolutional attention in the core compression layer, combines the reverse attention boundary module in the decoding layer, and finally strengthens the boundary learning with the boundary accuracy loss; specifically includes the following:

[0018] Architecture overview; input the enhanced chest CT image; after initial convolution, enter the five-layer encoder → bottleneck → five-layer decoder; there are five lossless transfer channels between the corresponding encoding layer and decoding layer; and strengthen the blood vessel–lung parenchyma junction edge at the deepest layer;

[0019] Encoder design; introduce double-branch parallel extraction; branch A is expressed as: ; where, is the output of branch A; is the depthwise separable convolution, 3×3 is the convolution kernel; a is the balanced residual coefficient; is the i-th layer of the encoder; is the instance normalization; branch B is expressed as: ; where, Is the output of branch B; Is the axial depthwise separable convolution with a 7×7 convolutional kernel; concatenation, expressed as: ; where, , and Are the outputs of the i-th layer of branch A, branch B, and the concatenated encoder respectively; Is the concatenation operation; the number of channels is doubled, followed by downsampling;

[0020] Core compression layer design; for chest CT, a hybrid convolution + attention process is adopted, expressed as: ; ; The attention weight is expressed as: ; ; ; Segmentation output Is expressed as: ; where, Is the output of the hybrid convolution; Is the output of the encoder; Is the pointwise convolution; Is the pointwise summation output; d is the quantization response intensity, highlighting the key lesion area; Is The mean value; v is the response mean value; Is the attention weight; Is the balance term;

[0021] Boundary strengthening; in the decoding stage, a double mechanism of skip connection feature fusion and reverse attention boundary is introduced; the first three levels of decoding refine the microstructure of the lung tissue, expressed as: ; where, and Are the feature maps after upsampling of the (i + 1)-th layer and the i-th layer decoders respectively; BAM is the boundary attention; for the deepest two layers of decoding, a prior prediction → reverse attention mechanism is introduced; the prior prediction generation is expressed as: ; ; The reverse attention boundary fusion is expressed as: ; ; where, Is the boundary prior; Is the reverse attention weight; Is the feature after reverse attention weighting; Is the skip connection feature; Is the fusion coefficient;

[0022] Loss function design; a boundary length smoothing term is added to the boundary accuracy loss; the segmentation loss function is expressed as: ; ; ; where, is the total segmentation loss function; is the boundary accuracy loss; is the boundary length smoothing term; is the loss weight; g and u are pixel coordinates; is the predicted segmentation result; is the ground truth segmentation label; is the predicted segmentation gradient.

[0023] Furthermore, the chest CT image processing module real-time collects chest CT images. After enhancement based on the multi-scale filtering processing module, frequency domain edge enhancement module, and edge fusion correction module, the images are input into the chest CT image segmentation model, and the segmentation result output by the model is used as the chest CT image processing result.

[0024] The beneficial effects achieved by the present invention using the above solution are as follows:

[0025] (1) Aiming at the problems existing in general chest CT image processing systems, such as insufficient sensitivity to small blood vessels, improper processing of frequency domain noise and artifacts, and lack of specialized guidance based on the anatomical structures of blood vessels and bronchi, resulting in poor final processing effects. In this solution, gradients are calculated at voxel intervals of 1mm and 2mm respectively, which can sensitively respond to both small nodules and large-scale pleural edges; different spectral segments are differentially processed according to the noise characteristics, guiding the key retention and enhancement of the blood vessel trajectories, while suppressing high-frequency noise and restoring low-frequency large structures; through edge fusion correction, the adaptive enhancement of small-scale nodules and large-scale lesions is taken into account; thereby improving the chest CT image processing effect.

[0026] (2) Aiming at the problems existing in general chest CT image processing systems, such as insufficient response to multi-scale lesions and lack of a specialized mechanism for focusing on boundaries, resulting in easy over-smoothing or breakage of the segmentation of the blood vessel, lung parenchyma interface, and slender branches, and poor accuracy of the processing results. In this solution, branch A refines the nodule edge and the weak contrast of artifacts, and branch B enhances the macroscopic consistency of the lung lobe; through five-layer parallel refinement + global common downsampling, both small-scale ground-glass nodules and large-scale lung lobe structures are taken into account; the decoding layer focuses on the interface, suppresses the learned areas, and focuses on the blood vessel, lung parenchyma junction; the loss introduces boundary smoothness to optimize the overall segmentation contour and reduce false positive spiky noise; thereby improving the chest CT image processing accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is a schematic flowchart of an intelligent chest CT image processing system provided by the present invention;

[0028] Figure 2 is a schematic flowchart of the chest CT image segmentation model establishment module.

[0029] The accompanying drawings are used to provide a further understanding of the present invention and form a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention. Detailed Embodiments

[0030] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0031] In the description of the present invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc. indicating the orientation or position relationship are based on the orientation or position relationship shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as a limitation to the present invention.

[0032] Embodiment 1, referring to Figure 1 , an intelligent chest CT image processing system provided by the present invention includes an image acquisition module, a multi-scale filtering processing module, a frequency domain edge enhancement module, an edge fusion and correction module, an enhancement loss design module, a chest CT image segmentation model establishment module, and a chest CT image processing module;

[0033] The image acquisition module acquires historical chest CTs and constructs a chest CT sample set;

[0034] The multi-scale filtering processing module calculates the change intensity map for the chest CT image at voxel intervals of 1 mm and 2 mm respectively, and generates a multi-scale change intensity map through upsampling alignment and scale-weighted fusion;

[0035] The frequency domain edge enhancement module performs 2D FFT on each layer of the multi-scale change intensity map to separate it into three frequency spectrum segments of low / middle / high frequencies, extracts multi-scale residual features using 5×5 and 1×1 convolutions, and after integrating the vascular response into the middle frequency residuals, reconstructs to obtain the repaired chest CT gradient map;

[0036] The edge fusion and correction module splices the original input features of the chest CT image with the repaired gradient map, successively fuses the features through channel attention and attention map weighted fusion, and then outputs the chest CT correction feature map through three layers of 3×3 convolutions and residual connections;

[0037] The enhancement loss design module constructs the total loss function of the enhancement network based on the high-quality chest CT reference image;

[0038] The chest CT image segmentation model building module constructs an encoder-bottleneck-decoder network and trains the chest CT image segmentation model with normalized active contour and boundary length smoothing loss;

[0039] The chest CT image processing module processes real-time chest CT images based on the multi-scale filtering processing module, frequency domain edge enhancement module, edge fusion correction module, and chest CT image segmentation model.

[0040] Embodiment 2, refer to Figure 1 , this embodiment is based on the above embodiment. The image acquisition module acquires historical chest CT images; standardizes the intensity values; truncates to the HU value range of [−1000, 400] and linearly normalizes to [0, 1]; annotates high-quality chest CT reference images; annotates all historical chest CT images with lung lobe and blood vessel segmentation labels; obtains a chest CT sample set.

[0041] Embodiment 3, refer to Figure 1 , this embodiment is based on the above embodiment. Since there are large differences in the sizes of blood vessels, trachea, and lesions in chest CT images, the multi-scale filtering processing module selects multi-scale Sobel filtering to estimate the gradient map , expressed as: ; Upsample and align the Sobel outputs with voxel spacings of 1mm and 2mm respectively and then perform weighted fusion to enhance the edge sensitivity to lesions of different sizes, expressed as: ; where , and are the gradient components of the filtered output; is the multi-scale change intensity map after fusion; is the scale weighting coefficient; and are the change intensity maps calculated at voxel spacings of 1mm and 2mm respectively; Considering the continuity between slices, the processing effect of blood vessel trajectories and bronchial branches is better; Multi-scale fusion takes into account the gradient responses of large-range pleural edges and micro-nodules.

[0042] Embodiment 4, refer to Figure 1 , this embodiment is based on the above embodiment. On the basis of correcting the amplitude spectrum in the frequency domain, the frequency domain edge enhancement module adds multi-spectrum analysis and Frangi–vessel enhancement guidance according to the characteristics of CT noise and artifacts; performs 2D FFT on each layer of CT gradient map and separates it into three frequency spectrum segments of low frequency / mid frequency / high frequency, expressed as: ; ; where is the Fourier transform of the 2D slice; A is the frequency domain separation result, , and represent the spectral components of low frequency, medium frequency, and high frequency respectively; the multi-scale features are extracted from the spatial domain through a convolutional network and are expressed as: ; where is the multi-scale feature residual, and k is the spectral band index; is a 5×5 convolutional kernel used on the 2D plane; is used for channel integration; is the activation function; for the vascular structure, the Frangi filter is used to generate the vascular response map V, which is fused with and is expressed as: ; where is the medium frequency spectral component after the vascular response fusion; γ is the vascular enhancement coefficient; the reconstruction fusion is expressed as: ; where is the finally enhanced and repaired gradient map; IFFT is the inverse Fourier transform reconstruction; P is the phase spectrum; through multi-spectrum band processing, the weak gradients of small blood vessels and bronchioles are finely restored, and the Frangi guidance focuses on retaining the vascular structure, suppressing noise while taking into account anatomical details.

[0043] Example 5, refer to Figure 1 , this example is based on the above example. The edge fusion correction module further adds spatial-slice attention in addition to channel attention to dynamically adjust the gradient and the chest CT image features; through the joint features and dual attention, we get: ; the channel attention is expressed as: ; ; where is the original input feature map; U is the joint feature after splicing; is the channel attention vector; and are the two split channel attention sub-vectors; is the linear transformation; LayerNorm is the layer normalization; is the segmentation; Sigmoid is the Sigmoid activation function; perform 3D convolution + Sigmoid processing on U to generate the spatial attention map , which is expressed as: ; the feature fusion is expressed as: ; ; finally, it is output through three layers of 3×3 convolution + residual, which is expressed as: ; where is the 3D convolution; S is the result of feature fusion; is the weighted feature map; , and are the three convolution operations; is the final enhanced output feature map; by slicing the dimension attention, it ensures that the gradient information can also be dynamically emphasized or suppressed along the axial direction, further improving the integrity of 3D continuous lesions; the channel + spatial dual attention enables both small nodules and large lesions to be adaptively enhanced.

[0044] Example Six. Refer to Figure 1 , this example is based on the above example. The enhanced loss design module performs enhancement processing on the chest CT sample set based on the multi-scale filtering processing module, the frequency domain edge enhancement module, and the edge fusion correction module, makes a comparison based on the high-quality chest CT reference image, and obtains the chest CT segmentation sample set through loss training; adds edge loss to ensure that the enhanced effect of the chest CT image has high-quality semantic and edge information; pixel consistency loss is expressed as: ; edge consistency loss helps to restore the clear boundary of the lesion area and is expressed as: ; the final total loss function is expressed as: ; where H and W are the height and width of the chest CT image; is the number of channels and rows and columns; is the pixel value after enhancement; is the pixel value of the high-quality chest CT reference image; is the smoothing term; is the edge detection operator; is the loss weight; is the gradient constraint term.

[0045] By performing the above operations, for the problems existing in the general chest CT image processing system, such as insufficient sensitivity to small blood vessels, improper processing of frequency domain noise and artifacts, and lack of specialized guidance based on the anatomical structures of blood vessels and bronchi, resulting in poor final processing effects, this solution calculates the gradient at voxel intervals of 1mm and 2mm respectively, and can be sensitive to both small nodules and large pleural edges; performs differential processing on the noise characteristics in different frequency bands, guides the key retention and enhancement of the blood vessel course, while suppressing high-frequency noise and restoring low-frequency large structures; through edge fusion correction, takes into account the adaptive enhancement of small nodules and large lesions; thereby improving the chest CT image processing effect.

[0046] Example Seven. Refer to Figure 1 and Figure 2, based on the above embodiment, the chest CT image segmentation model establishment module establishes a segmentation model based on the chest CT segmentation sample set. The overall process includes three major stages: encoding, bottleneck, and decoding. Double-branch refined feature extraction is introduced in the encoding layer, multi-scale hybrid convolutional attention is applied in the core compression layer, and the reverse attention boundary module is combined in the decoding layer. Finally, boundary learning is strengthened with boundary accuracy loss. Specifically, it includes the following:

[0047] Architectural overview; The enhanced chest CT is input; After initial convolution, it enters a five-layer encoder → bottleneck → five-layer decoder; There are five lossless transfer channels between the corresponding encoding layer and decoding layer to ensure the simultaneous recovery of lobe details and global structure; And the blood vessel–lung parenchyma junction edge is emphasized in the deepest layer;

[0048] Encoder design; To take into account both the fine nodules of ground-glass opacities and the overall anatomical structure of the lung lobes in chest CT, double-branch parallel extraction is introduced; Branch A captures the edges of small nodules, fine textures, and weak contrasts of artifacts; Branch A is expressed as: ; where, is the output of Branch A; is depthwise separable convolution, 3×3 is the convolution kernel; a is the balanced residual coefficient; is the i-th layer of the encoder; is instance normalization; Branch B strengthens the macrostructure consistency of the lung lobes and suppresses artifacts and noise; Branch B is expressed as: ; where, is the output of Branch B; is axial depthwise separable convolution, convolution kernel 7×7; Concatenation, expressed as: ; where, , and are the outputs of the i-th layer of Branch A, Branch B, and the concatenated encoder respectively; is the concatenation operation; The number of channels is doubled, and then downsampling is performed; By simultaneously refining the edges of lung nodules and robustly modeling the lung lobe contours through double branches, the segmentation quality that takes into account both early lesions and late lesions is improved;

[0049] Core compression layer design; For the multi-scale segmentation requirements of coexisting micro-nodules and large-scale lesions in chest CT, hybrid convolution + attention processing is adopted, expressed as: ; ; The attention weight is expressed as: ; ; ; Segmentation output is expressed as: ; where, is the output of hybrid convolution; is the output of the encoder; is pointwise convolution; is point-by-point summation output; d is the quantization response intensity, highlighting the key lesion areas; is the mean value; v is the response mean value; is the attention weight; is the balance term; adaptively highlight the key lesion response, suppress background artifacts, and output features with rich multi-scale and spatial information in the core compression layer;

[0050] Boundary enhancement; introduce a dual mechanism of skip connection feature fusion and reverse attention boundary in the decoding stage; the first three levels of decoding refine the microstructure of lung tissue, expressed as: ; where, and are the feature maps after upsampling of the (i + 1)-th layer and the i-th layer decoders respectively; BAM is the boundary attention; highlight the cell structure of lung tissue; for the deepest two layers of decoding, introduce a prior prediction → reverse attention mechanism, with the boundary as the focus of attention, by suppressing the learned area and focusing on the blood vessel–lung parenchyma junction, the segmentation accuracy of slender structures and interfaces can be significantly improved, and the boundary blur caused by over-smoothing can be reduced; the prior prediction generation is expressed as: ; ; the reverse attention boundary fusion is expressed as: ; ; where, is the boundary prior; is the reverse attention weight; is the feature after reverse attention weighting; is the skip connection feature; is the fusion coefficient;

[0051] Loss function design; add a boundary length smoothing term to the boundary accuracy loss; the segmentation loss function is expressed as: ; ; ; where, is the total segmentation loss function; is the boundary accuracy loss; is the boundary length smoothing term; is the loss weight; g and u are pixel coordinates; is the predicted segmentation result; is the true segmentation label; is the predicted segmentation gradient.

[0052] By performing the above operations, in view of the problems existing in the general chest CT image processing system, such as insufficient response to multi-scale lesions, lack of a mechanism specifically focusing on boundaries, resulting in easy over-smoothing or breakage of the segmentation of blood vessels, the interface between the lung parenchyma, and slender branches, and poor accuracy of the processing results, this solution refines the nodule edges and weak artifact contrast through branch A, and strengthens the macroscopic consistency of the lung lobes through branch B; it simultaneously takes into account small-scale ground-glass nodules and large-scale lung lobe structures through five-layer parallel refinement + global common downsampling; the decoding layer focuses on the interface, suppresses the learned regions, and focuses on the junction of blood vessels and the lung parenchyma; the loss introduces boundary smoothness to optimize the overall segmentation contour and reduce false-positive spiky noises; thereby improving the accuracy of chest CT image processing.

[0053] Embodiment Eight. Refer to Figure 1 , based on the above embodiment, the chest CT image processing module real-time collects chest CT images. After enhancement based on the multi-scale filtering processing module, the frequency-domain edge enhancement module, and the edge fusion correction module, the images are input into the chest CT image segmentation model, and the segmentation result output by the model is used as the chest CT image processing result.

[0054] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principle and spirit of the present invention.

[0055] The above describes the present invention and its implementation manners. This description is not restrictive, and only one of the implementation manners of the present invention is shown in the drawings. The actual structure is not limited thereto. Generally speaking, if those of ordinary skill in the art are inspired by it and design similar structural manners and embodiments without creative efforts without departing from the purpose of the present invention, they shall fall within the protection scope of the present invention.

Claims

1. An intelligent chest CT image processing system, characterized in that: The system includes an image acquisition module, a multi-scale filtering processing module, a frequency-domain edge enhancement module, an edge fusion correction module, an enhancement loss design module, a chest CT image segmentation model establishment module, and a chest CT image processing module; The image acquisition module acquires historical chest CTs to construct a chest CT sample set; The multi-scale filtering processing module calculates the change intensity maps for chest CT images at voxel intervals of 1 mm and 2 mm respectively, and generates a multi-scale change intensity map through upsampling alignment and scale-weighted fusion; The frequency-domain edge enhancement module performs 2D FFT on each layer of the multi-scale change intensity map to separate it into three frequency spectrum segments of low / mid / high frequencies, extracts multi-scale residual features using 5×5 and 1×1 convolutions, and after integrating the vascular response into the mid-frequency residuals, reconstructs to obtain the repaired chest CT gradient map; The edge fusion correction module splices the original input features of the chest CT image with the repaired gradient map, sequentially fuses the features through channel attention and attention map weighted fusion, and then outputs the chest CT correction feature map through three layers of 3×3 convolutions and residual connections; The enhancement loss design module constructs the total loss function of the enhancement network based on high-quality chest CT reference images; The chest CT image segmentation model establishment module constructs an encoder-bottleneck-decoder network and trains the chest CT image segmentation model with normalized active contour and boundary length smoothing loss; The chest CT image processing module processes real-time chest CT images based on the multi-scale filtering processing module, the frequency-domain edge enhancement module, the edge fusion correction module, and the chest CT image segmentation model.

2. The intelligent chest CT image processing system according to claim 1, wherein: The multi-scale filtering processing module uses multi-scale Sobel filtering for chest CT images to estimate the gradient map ; After upsampling and aligning the Sobel outputs with voxel spacings of 1 mm and 2 mm respectively, they are weighted and fused.

3. An intelligent chest CT image processing system according to claim 2, characterized in that: The frequency-domain edge enhancement module, on the basis of correcting the amplitude spectrum in the frequency domain, adds multi-spectrum segment analysis and vascular enhancement guidance; performs 2D FFT on each layer of the CT gradient map and separates it into three frequency spectrum segments of low / mid / high frequencies; extracts multi-scale features through a convolutional network in the spatial domain; For the vascular structure, use the Frangi filter to generate the vascular response map V, and fuse it with Fuse; reconstruct the fused result to obtain the final enhanced and repaired gradient map.

4. An intelligent chest CT image processing system according to claim 3, wherein: The edge fusion correction module adds spatial-slice attention in addition to channel attention to dynamically adjust the gradient and chest CT image features; U is obtained by combining features and dual attention; 3D convolution + Sigmoid processing is performed on U to generate a spatial attention map. Finally, the final enhanced output feature map is obtained through three layers of 3×3 convolution + residual output.

5. An intelligent chest CT image processing system according to claim 4, wherein: The enhancement loss design module enhances the chest CT sample set based on the multi-scale filtering processing module, the frequency-domain edge enhancement module, and the edge fusion correction module, makes a comparison based on high-quality chest CT reference images, and obtains a chest CT segmentation sample set through loss training.

6. An intelligent chest CT image processing system according to claim 5, characterized in that: The chest CT image segmentation model specifically includes the following: Architectural overview; the input is the enhanced chest CT image; after initial convolution, it enters a five-layer encoder → bottleneck → five-layer decoder; there are five lossless transfer channels between the corresponding encoding layer and decoding layer; And intensively strengthens the blood vessel-lung parenchyma junction edge at the deepest layer; Encoder design; introduces dual-branch parallel extraction; doubles the number of channels and then performs downsampling; Core compression layer design; uses hybrid convolution + attention processing for chest CT to obtain the segmentation output; Boundary strengthening; introduces a dual mechanism of skip connection feature fusion and reverse attention boundary in the decoding stage; the first three levels of decoding refine the microstructure of the lung tissue; for the deepest two levels of decoding, introduces prior prediction → reverse attention mechanism; Loss function design; adds a boundary length smoothing term to the boundary accuracy loss.

7. An intelligent chest CT image processing system according to claim 6, characterized in that: The image acquisition module acquires historical chest CT images; normalizes the intensity values; annotates high-quality chest CT reference images; annotates all historical chest CT images; and obtains a chest CT sample set.

8. An intelligent chest CT image processing system according to claim 7, characterized in that: The chest CT image processing module real-time acquires chest CT images. After enhancement based on the multi-scale filtering processing module, frequency domain edge enhancement module, and edge fusion and correction module, the images are input into the chest CT image segmentation model, and the segmentation result output by the model is used as the chest CT image processing result.

Images