Intelligent tumor image processing system

Abstract

The invention discloses an intelligent tumor image processing system which comprises an image acquisition module, a tumor image enhancement module, a tumor image segmentation module and a tumor image processing module. The invention belongs to the field of image processing, and particularly relates to an intelligent tumor image processing system, which introduces an attenuation effect and noise characteristics in an imaging process to enhance the repairing capability of a tumor image; structural loss and boundary loss are introduced, boundary details of a tumor area are enhanced, details of a tumor image are gradually optimized by alternately enhancing global features and local features of the tumor area, and the detection capability of the tumor area is improved; by means of multi-scale feature enhancement and bilateral filtering, in combination with multi-scale convolution and pooling, the edge of a tumor area is reserved, background noise is removed, and meanwhile by means of adjacent feature aggregation, the precision of the boundary of the tumor area is improved, and interference of a fuzzy area is reduced; tumor details are recovered, the boundary definition is enhanced, and the tumor image processing effect is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to an intelligent tumor image processing system. Background Art

[0002] The tumor image processing system is an image processing tool based on computer vision and artificial intelligence technology, which aims to highlight the key areas of tumor images. However, general tumor image processing systems have problems such as loss of details and blurred boundaries when processing low-quality and noise-interfered tumor images, as well as poor processing accuracy caused by noise from imaging equipment; general tumor image processing systems have problems such as difficulty in accurately capturing boundaries for small-sized tumors, and poor processing effects for blurred and unclear tumor areas, which in turn leads to poor tumor image processing effects. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides an intelligent tumor image processing system. In view of the problems that general tumor image processing systems are prone to detail loss and boundary blur when processing low-quality and noise-interfered tumor images, and the noise caused by the imaging device leads to poor processing accuracy, the present invention utilizes structural loss and background light estimation to remove noise and restore details; introduces the attenuation effect and noise characteristics of the imaging process to enhance the ability to repair tumor images; introduces structural loss and boundary loss to enhance the boundary details of the tumor area and prevents artifacts and noise in the image from affecting subsequent processing; thereby improving the accuracy of tumor image processing; in view of the problems that general tumor image processing systems have for small-sized tumors It is difficult to accurately capture the boundaries of tumors, and the effect of processing blurred and unclear tumor areas is poor, which leads to the problem of poor tumor image processing effect. This scheme gradually optimizes the details of tumor images by alternately enhancing the global features and local features of the tumor area, thereby improving the detection ability of the tumor area; global features are used to capture the overall morphology of the tumor, while local features focus on the details, especially the tumor boundaries; through multi-scale feature enhancement and bilateral filtering, combined with multi-scale convolution and pooling, the edges of the tumor area are retained and the background noise is removed. At the same time, through the aggregation of adjacent features, the accuracy of the tumor area boundary is improved and the interference of the blurred area is reduced; thus, the tumor details are restored, the boundary clarity is enhanced, and the tumor image processing effect is improved.

[0004] The technical solution adopted by the present invention is as follows: an intelligent tumor image processing system provided by the present invention comprises an image acquisition module, a tumor image enhancement module, a tumor image segmentation module and a tumor image processing module;

[0005] The image acquisition module acquires historical tumor image data;

[0006] The tumor image enhancement module enhances the low-quality tumor image, simulates the degradation process, and generates a high-quality tumor image;

[0007] The tumor image segmentation module segments the enhanced tumor image through progressive enhancement, feature aggregation and degenerate feature selection;

[0008] The tumor image processing module performs enhanced segmentation on the tumor image collected in real time, thereby realizing tumor image processing.

[0009] Furthermore, in the image acquisition module, the historical tumor image data includes low-quality tumor images and high-quality tumor images; and the tumor area is delineated for the historical tumor image data.

[0010] Furthermore, the tumor image enhancement module is based on a generative adversarial network, inputs a low-quality tumor image I, and outputs a high-quality tumor image J as a target image; specifically, it includes the following contents:

[0011] Tumor image degradation simulation unit; introduces the attenuation effect of the tumor image during the imaging process and the noise characteristics of the imaging device; the tumor image degradation simulation is expressed as: ; ; Wherein, T is the transmission diagram, which is used to describe the attenuation degree of the tumor image during the imaging process; A is the background light; is the noise of the imaging device; and is the weighting coefficient of noise; is Gaussian noise; It’s salt and pepper noise; and is the noise probability;

[0012] Tumor image enhancement inversion unit; accurately estimate the background light according to the noise distribution of the tumor image; introduce low-frequency background noise by estimating the transmission map, and the background light estimation is expressed as: ; The transmission graph estimation is expressed as: ;in, is the estimated background light during the enhancement process; GB(·) is the Gaussian blur; is the estimated transmission map during the enhancement process; is the Sigmoid activation function; and They are 7×7 convolution and 3×3 convolution respectively; and is the adjustment parameter; It is low-frequency background noise;

[0013] Tumor image degradation estimation unit; the transmission map and background light estimation during the degradation process are expressed as: ; ;in, and are the estimated transmission map and background light in the degradation process, respectively; Max(·) is the maximum pooling operation; are features extracted from high-quality tumor images J; It is residual learning;

[0014] Loss function design unit; construct adversarial loss, expressed as: ; ; Construct cycle consistency loss, expressed as: ; Construct the structural loss, expressed as: ; ; Construct tumor boundary loss , expressed as: Final loss It is expressed as: ; Where L1 and L2 are the adversarial losses of I and J respectively; D(·) is the discriminator; is a low-quality image enhanced by the generator; It is a high-quality image after being processed by the degradation network; and is the image restored by enhancing the degradation process; is the L1 norm; and The loss is obtained by weighted averaging all images in the low-quality tumor image and high-quality tumor image datasets; and are the estimated transmission map and the loss of background light, respectively; and They are the estimated transmission diagrams during the enhancement process and the degradation process, respectively; and are the estimated background light during the enhancement process and the degradation process, respectively; , , and is the loss weight; r is the pixel index; is the gradient difference between I and J at pixel r.

[0015] Furthermore, the tumor image segmentation module processes the enhanced tumor image, specifically including the following contents:

[0016] Progressive enhancement unit; by alternately enhancing the global features of tumor location and morphology and the local features of tumor boundaries, and switching the processing of global and local features by adjusting the input order of feature maps; progressive enhancement is expressed as: ; ; ;in, and are the global feature enhancement results and local feature enhancement results generated in the iterative process, k is the number of iterations; α and β are parameters for adjusting the fusion of global and local features; s(·) is the fused feature; F(·) is the fusion operation; R(·) is the transformation operation; gd and ad represent the global feature and local feature, respectively;

[0017] Feature aggregation unit; combine feature aggregation with context information; introduce multi-scale feature enhancement and bilateral filtering; feature aggregation is expressed as: ; ; ; ;in, It is the result of aggregating the current feature and the feature of the previous layer through adjacent features; and is the weight coefficient of feature map fusion; is the feature of the current layer; is another adjacent feature of the current layer; It is the feature map obtained in the previous layer; It is the output feature map after the pooling operation; It is a feature fusion operation; It is the feature map after upsampling; is an upsampling operation; Concat(·) is concatenation and fusion; EP(·) is bilateral filtering; , and is the adjustment coefficient; U and Q are the total number of scales of convolution and pooling, respectively, and u and q are the corresponding indices; and They are the convolution and pooling operations of corresponding scales respectively; and is the scale weighting coefficient;

[0018] Degenerate feature selection unit; repair tumor image and maintain phase information; remove interference through Fourier transform; restore structural information through phase spectrum; expressed as: ; The frequency domain characteristics are expressed as: ; Amplitude spectrum and phase spectrum representation: ; ; Introducing wavelet transform, the amplitude reconstruction is expressed as: ; Amplitude error It is expressed as: ; Wherein, T(·) is the image after Fourier transformation, u and v are the frequency domain coordinates; f(·) is the pixel value of the tumor image in the spatial domain, x and y are the spatial coordinates; M and N are the spatial dimensions; R(·) and I(·) are the real and imaginary parts of the frequency domain, respectively; A(·) is the amplitude spectrum; P(·) is the phase spectrum; is the reconstructed amplitude spectrum; is the convolution kernel; j is the imaginary unit; is the wavelet reconstruction enhancement coefficient; WT(·) is the wavelet transform operation;

[0019] Segmentation loss function construction unit; in the tumor image segmentation task, each iteration includes: gradually optimizing the global and local features of the tumor image through progressive enhancement; refining the features of the tumor area through feature aggregation; repairing the noise and image degradation in the tumor image based on the degraded feature selection, and restoring the structural information; converting the loss after the iteration into With each iteration loss Weighted segmentation loss function , and completed the training based on historical tumor image data; the segmentation loss function is expressed as: ; ; ; Where S is the total number of iterations and s is the iteration index; is the iteration weight; is the iteration adjustment factor; is the current average iteration loss; y is the true tumor region label, i.e., whether it is a tumor region; p is the predicted tumor region label; A is the segmented region; and B is the true tumor region.

[0020] Furthermore, the tumor image processing module acquires tumor image data in real time, enhances and segments the tumor image, and realizes tumor image processing after the training of the tumor image enhancement module and the tumor image segmentation module is completed.

[0021] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0022] (1) In view of the problems that general tumor image processing systems are prone to detail loss and blurred boundaries when processing low-quality and noisy tumor images, as well as the noise caused by imaging equipment, which leads to poor processing accuracy, this scheme uses structural loss and background light estimation to remove noise and restore details; introduces the attenuation effect and noise characteristics in the imaging process to enhance the ability to repair tumor images; introduces structural loss and boundary loss to enhance the boundary details of the tumor area and prevent artifacts and noise in the image from affecting subsequent processing; thereby improving the accuracy of tumor image processing.

[0023] (2) In view of the problem that general tumor image processing systems have difficulty in accurately capturing the boundaries of small-sized tumors and have poor effects in processing blurred and unclear tumor areas, which leads to poor tumor image processing effects, this scheme alternately enhances the global features and local features of the tumor area, gradually optimizes the details of the tumor image, and improves the detection ability of the tumor area; global features are used to capture the overall morphology of the tumor, while local features focus on the details, especially the tumor boundaries; through multi-scale feature enhancement and bilateral filtering, combined with multi-scale convolution and pooling, the edges of the tumor area are retained and background noise is removed. At the same time, by aggregating adjacent features, the accuracy of the tumor area boundary is improved and the interference of the blurred area is reduced; thus, the tumor details are restored, the boundary clarity is enhanced, and the tumor image processing effect is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 A schematic diagram of the process of an intelligent tumor image processing system provided by the present invention;

[0025] Figure 2 This is a flow chart of the tumor image enhancement module;

[0026] Figure 3 Schematic diagram of the process of tumor image segmentation module.

[0027] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0028] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0029] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and 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 direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0030] Example 1, see Figure 1 , the present invention provides an intelligent tumor image processing system, including an image acquisition module, a tumor image enhancement module, a tumor image segmentation module and a tumor image processing module;

[0031] The image acquisition module acquires historical tumor image data; and sends the data to the tumor image enhancement module;

[0032] The tumor image enhancement module enhances the low-quality tumor image, simulates the degradation process, and generates a high-quality tumor image; and sends the data to the tumor image segmentation module;

[0033] The tumor image segmentation module segments the enhanced tumor image through progressive enhancement, feature aggregation and degenerate feature selection; and sends the data to the tumor image processing module;

[0034] The tumor image processing module performs enhanced segmentation on the tumor image collected in real time, thereby realizing tumor image processing.

[0035] Example 2, see Figure 1 This embodiment is based on the above embodiment. In the image acquisition module, the historical tumor image data includes low-quality tumor images and high-quality tumor images; and the tumor area is delineated for the historical tumor image data.

[0036] Example 3, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. The tumor image enhancement module is based on a generative adversarial network, inputs a low-quality tumor image I, and outputs a high-quality tumor image J as a target image. Specifically, it includes the following contents:

[0037] Tumor image degradation simulation unit; introduces the attenuation effect of the tumor image during the imaging process and the noise characteristics of the imaging device; the tumor image degradation simulation is expressed as: ; ; Wherein, T is the transmission diagram, which is used to describe the attenuation degree of the tumor image during the imaging process; A is the background light, which represents the background light or noise in the image that is not related to the tumor; is the noise of the imaging device; and is the weighting coefficient of noise; is Gaussian noise; It’s salt and pepper noise; and is the noise probability;

[0038] Tumor image enhancement inversion unit; accurately estimating the background light according to the noise distribution of the tumor image helps to generate a clear image after denoising; by estimating the transmission map, it helps to improve the boundary blur of the tumor area and enhance the visibility of the tumor tissue; introducing low-frequency background noise, the background light estimation is expressed as: ; The transmission graph estimation is expressed as: ;in, is the estimated background light during the enhancement process; GB(·) is the Gaussian blur; is the estimated transmission map during the enhancement process; is the Sigmoid activation function; and They are 7×7 convolution and 3×3 convolution respectively; and is the adjustment parameter; It is low-frequency background noise;

[0039] Tumor image degradation estimation unit; by simulating the degradation process, it is possible to generate a more realistic low-quality tumor image and promote the generation of higher-quality tumor images in the enhancement process; the transmission map and background light estimation in the degradation process are respectively expressed as: ; ;in, and are the estimated transmission map and background light in the degradation process, respectively; Max(·) is the maximum pooling operation; are features extracted from high-quality tumor images J; It is residual learning;

[0040] Loss function design unit; by increasing the weight of structural loss, the tumor area can be better restored; and the structural loss can help remove artifacts and noise in the tumor image and improve the clarity of the tumor image, especially in low-quality tumor images, it can restore finer tumor area details; construct adversarial loss, expressed as: ; ; Construct cycle consistency loss, expressed as: ; Construct the structural loss, expressed as: ; ; Construct tumor boundary loss , expressed as: Final loss It is expressed as: ; Where L1 and L2 are the adversarial losses of I and J respectively; D(·) is the discriminator; is a low-quality image enhanced by the generator; It is a high-quality image after being processed by the degradation network; and is the image restored by enhancing the degradation process; is the L1 norm; and The loss is obtained by weighted averaging all images in the low-quality tumor image and high-quality tumor image datasets; and are the estimated transmission map and the loss of background light, respectively; and They are the estimated transmission diagrams during the enhancement process and the degradation process, respectively; and are the estimated background light during the enhancement process and the degradation process, respectively; , , and is the loss weight; r is the pixel index; is the gradient difference between I and J at pixel r.

[0041] By performing the above operations, the general tumor image processing system is prone to detail loss and boundary blur when processing low-quality and noise-interfered tumor images, and the noise caused by the imaging equipment leads to poor processing accuracy. This scheme uses structural loss and background light estimation to remove noise and restore details; introduces the attenuation effect and noise characteristics in the imaging process to enhance the ability to repair tumor images; introduces structural loss and boundary loss to enhance the boundary details of the tumor area and prevent artifacts and noise in the image from affecting subsequent processing; thereby improving the accuracy of tumor image processing.

[0042] Example 4, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. The tumor image segmentation module processes the enhanced tumor image, and specifically includes the following contents:

[0043] Progressive enhancement unit; Tumor images usually contain complex edges and small-size features. By alternately enhancing the global features of tumor location and morphology and the local features of tumor boundaries, the detection capability of tumor areas is enhanced; and by adjusting the input order of feature maps, the processing of global and local features is switched, so that the edge and detail features of tumors are better captured; expressed as: ; ; ;in, and are the global feature enhancement results and local feature enhancement results generated in the iterative process, k is the number of iterations; α and β are parameters for adjusting the fusion of global and local features; s(·) is the fused feature; F(·) is the fusion operation; R(·) is the transformation operation; gd and ad represent the global feature and local feature, respectively;

[0044] Feature aggregation unit; for small, blurred and early-stage tumors, multi-scale processing can help capture subtle morphological changes and insignificant tumor features; tumor images often contain irregular shapes, and adjacent feature aggregation can improve the accuracy of tumor region boundaries and reduce interference in blurred areas; capture tiny tumor features and perform feature aggregation and context information combination; introduce multi-scale feature enhancement and bilateral filtering to retain the edges of the tumor region and reflect the details of the tumor image; feature aggregation is expressed as: ; ; ; ;in, It is the result of aggregating the current feature and the feature of the previous layer through adjacent features; and is the weight coefficient of feature map fusion; is the feature of the current layer; is another adjacent feature of the current layer; It is the feature map obtained in the previous layer; It is the output feature map after the pooling operation; It is a feature fusion operation; It is the feature map after upsampling; is an upsampling operation; Concat(·) is concatenation and fusion; EP(·) is bilateral filtering; , and is the adjustment coefficient; U and Q are the total number of scales of convolution and pooling, respectively, and u and q are the corresponding indices; and They are the convolution and pooling operations of corresponding scales respectively; and is the scale weighting coefficient;

[0045] Degenerate feature selection unit; repair tumor images and maintain phase information; for low-quality or noise-interfered tumor images, remove interference through Fourier transform to improve image quality and make tumor segmentation more accurate; restore structural information through phase spectrum to enhance the morphological characteristics of tumors, especially when details are missing; expressed as: ; The frequency domain characteristics are expressed as: ; Amplitude spectrum and phase spectrum representation: ; ; Introducing wavelet transform, the amplitude reconstruction is expressed as: ; Amplitude error It is expressed as: ; Wherein, T(·) is the image after Fourier transformation, u and v are the frequency domain coordinates; f(·) is the pixel value of the tumor image in the spatial domain, x and y are the spatial coordinates; M and N are the spatial dimensions; R(·) and I(·) are the real and imaginary parts of the frequency domain, respectively; A(·) is the amplitude spectrum; P(·) is the phase spectrum; is the reconstructed amplitude spectrum; is the convolution kernel; j is the imaginary unit; is the wavelet reconstruction enhancement coefficient; WT(·) is the wavelet transform operation;

[0046] Segmentation loss function construction unit; in the tumor image segmentation task, each iteration includes: gradually optimizing the global and local features of the tumor image through progressive enhancement; refining the features of the tumor area through feature aggregation; repairing the noise and image degradation in the tumor image based on the degraded feature selection, and restoring the structural information; converting the loss after the iteration into With each iteration loss Weighted segmentation loss function , and completed the training based on historical tumor image data; the segmentation loss function is expressed as: ; ; ; Where S is the total number of iterations and s is the iteration index; is the iteration weight; is the iteration adjustment factor; is the current average iteration loss; y is the true tumor region label, i.e., whether it is a tumor region; p is the predicted tumor region label; A is the segmented region; and B is the true tumor region.

[0047] By performing the above operations, the general tumor image processing system has the problem that it is difficult to accurately capture the boundaries of small-sized tumors, and the effect of processing blurred and unclear tumor areas is poor, which leads to poor tumor image processing effect. This scheme alternately enhances the global features and local features of the tumor area, gradually optimizes the details of the tumor image, and improves the detection ability of the tumor area; the global features are used to capture the overall morphology of the tumor, while the local features focus on the details, especially the tumor boundaries; through multi-scale feature enhancement and bilateral filtering, combined with multi-scale convolution and pooling, the edges of the tumor area are retained and the background noise is removed. At the same time, through the aggregation of adjacent features, the accuracy of the tumor area boundary is improved and the interference of the blurred area is reduced; thereby restoring the tumor details, enhancing the boundary clarity, and improving the tumor image processing effect.

[0048] Example 5, see Figure 1 This embodiment is based on the above embodiment. After the training of the tumor image enhancement module and the tumor image segmentation module is completed, the tumor image data is acquired in real time, and the tumor image is enhanced and segmented to realize tumor image processing.

[0049] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0050] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0051] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. An intelligent tumor image processing system, characterized in that: The system includes an image acquisition module, a tumor image enhancement module, a tumor image segmentation module and a tumor image processing module; The image acquisition module acquires historical tumor image data; The tumor image enhancement module enhances the low-quality tumor image, simulates the degradation process, and generates a high-quality tumor image; The tumor image segmentation module segments the enhanced tumor image through progressive enhancement, feature aggregation and degenerate feature selection; The tumor image processing module performs enhanced segmentation on the tumor image collected in real time, thereby realizing tumor image processing; The tumor image enhancement module includes a tumor image degradation simulation unit; the attenuation effect of the tumor image during the imaging process and the noise characteristics of the imaging device are introduced; the tumor image degradation simulation is expressed as: ; ; Wherein, T is the transmission diagram, which is used to describe the attenuation degree of the tumor image during the imaging process; A is the background light; is the noise of the imaging device; and is the weighting coefficient of noise; is Gaussian noise; It’s salt and pepper noise; and is the noise probability; I and J are low-quality tumor images and high-quality tumor images, respectively.

2. The intelligent tumor image processing system according to claim 1, characterized in that: The tumor image enhancement module is based on a generative adversarial network, inputs a low-quality tumor image I, and outputs a high-quality tumor image J as a target image; specifically, it includes the following contents: Tumor image degradation simulation unit; Tumor image enhancement inversion unit; accurately estimate the background light according to the noise distribution of the tumor image; introduce low-frequency background noise by estimating the transmission map, and the background light estimation is expressed as: ; The transmission graph estimation is expressed as: ;in, is the estimated background light during the enhancement process; GB(·) is the Gaussian blur; is the estimated transmission map during the enhancement process; is the Sigmoid activation function; and They are 7×7 convolution and 3×3 convolution respectively; and is the adjustment parameter; It is low-frequency background noise; Tumor image degradation estimation unit; the transmission map and background light estimation during the degradation process are expressed as: ; ;in, and are the estimated transmission map and background light in the degradation process, respectively; Max(·) is the maximum pooling operation; are features extracted from high-quality tumor images J; It is residual learning; Loss function design unit; construct adversarial loss, expressed as: ; ; Construct cycle consistency loss, expressed as: ; Construct the structural loss, expressed as: ; ; Construct tumor boundary loss , expressed as: Final loss It is expressed as: ; Where L1 and L2 are the adversarial losses of I and J respectively; D(·) is the discriminator; is a low-quality image enhanced by the generator; It is a high-quality image after being processed by the degradation network; and is the image restored by enhancing the degradation process; is the L1 norm; and The loss is obtained by weighted averaging all images in the low-quality tumor image and high-quality tumor image datasets; and are the estimated transmission map and the loss of background light, respectively; and They are the estimated transmission diagrams during the enhancement process and the degradation process, respectively; and are the estimated background light during the enhancement process and the degradation process, respectively; , , and is the loss weight; r is the pixel index; is the gradient difference between I and J at pixel r.

3. The intelligent tumor image processing system according to claim 2, characterized in that: The loss function design unit is to construct adversarial loss, which is expressed as: ; ; Construct cycle consistency loss, expressed as: ; Construct the structural loss, expressed as: ; ; Construct tumor boundary loss , expressed as: Final loss It is expressed as: ; Where L1 and L2 are the adversarial losses of I and J respectively; D(·) is the discriminator; is a low-quality image enhanced by the generator; It is a high-quality image after being processed by the degradation network; and is the image restored by enhancing the degradation process; is the L1 norm; and The loss is obtained by weighted averaging all images in the low-quality tumor image and high-quality tumor image datasets; and are the estimated transmission map and the loss of background light, respectively; and They are the estimated transmission diagrams during the enhancement process and the degradation process, respectively; and are the estimated background light during the enhancement process and the degradation process, respectively; , , and is the loss weight; r is the pixel index; is the gradient difference between I and J at pixel r.

4. The intelligent tumor image processing system according to claim 3, characterized in that: The loss function design unit specifically includes the following contents: The tumor image segmentation module processes the enhanced tumor image, specifically including the following contents: Progressive enhancement unit; by alternately enhancing the global features of tumor location and morphology and the local features of tumor boundaries, and switching the processing of global and local features by adjusting the input order of feature maps; Progressive enhancement is expressed as: ; ; ;in, and are the global feature enhancement results and local feature enhancement results generated in the iterative process, k is the number of iterations; α and β are parameters for adjusting the fusion of global and local features; s(·) is the fused feature; F(·) is the fusion operation; R(·) is the transformation operation; gd and ad represent the global feature and local feature, respectively; Feature aggregation unit; combine feature aggregation with context information; introduce multi-scale feature enhancement and bilateral filtering; feature aggregation is expressed as: ; ; ; ;in, It is the result of aggregating the current feature and the feature of the previous layer through adjacent features; and is the weight coefficient of feature map fusion; is the feature of the current layer; is another adjacent feature of the current layer; It is the feature map obtained in the previous layer; It is the output feature map after the pooling operation; It is a feature fusion operation; It is the feature map after upsampling; is an upsampling operation; Concat(·) is concatenation and fusion; EP(·) is bilateral filtering; , and is the adjustment coefficient; U and Q are the total number of scales of convolution and pooling, respectively, and u and q are the corresponding indices; and They are the convolution and pooling operations of corresponding scales respectively; and is the scale weighting coefficient; Degenerate feature selection unit; repair tumor image and maintain phase information; remove interference through Fourier transform; restore structural information through phase spectrum; expressed as: ; The frequency domain characteristics are expressed as: ; Amplitude spectrum and phase spectrum representation: ; ; Introducing wavelet transform, the amplitude reconstruction is expressed as: ; Amplitude error It is expressed as: ; Wherein, T(·) is the image after Fourier transformation, u and v are the frequency domain coordinates; f(·) is the pixel value of the tumor image in the spatial domain, x and y are the spatial coordinates; M and N are the spatial dimensions; R(·) and I(·) are the real and imaginary parts of the frequency domain, respectively; A(·) is the amplitude spectrum; P(·) is the phase spectrum; is the reconstructed amplitude spectrum; is the convolution kernel; j is the imaginary unit; is the wavelet reconstruction enhancement coefficient; WT(·) is the wavelet transform operation; Segmentation loss function building block.

5. The intelligent tumor image processing system according to claim 4, characterized in that: The segmentation loss function construction unit is in the segmentation task of the tumor image, and each iteration includes: gradually optimizing the global and local features of the tumor image through progressive enhancement; refining the features of the tumor area through feature aggregation; repairing the noise and image degradation in the tumor image based on the degradation feature selection, and restoring the structural information; converting the loss after the iteration into With each iteration loss Weighted segmentation loss function , and completed the training based on historical tumor image data; the segmentation loss function is expressed as: ; ; ; Where S is the total number of iterations and s is the iteration index; is the iteration weight; is the iteration adjustment factor; is the current average iteration loss; y is the true tumor region label, i.e., whether it is a tumor region; p is the predicted tumor region label; A is the segmented region; and B is the true tumor region.

6. The intelligent tumor image processing system according to claim 1, characterized in that: The tumor image processing module acquires tumor image data in real time, enhances and segments the tumor image, and realizes tumor image processing after the training of the tumor image enhancement module and the tumor image segmentation module is completed.

7. The intelligent tumor image processing system according to claim 1, characterized in that: In the image acquisition module, the historical tumor image data includes low-quality tumor images and high-quality tumor images; and the tumor area is delineated for the historical tumor image data.

Images