A complex model clinical image semantic segmentation method based on frequency phase guidance

Abstract

The application discloses a kind of complex model clinical image semantic segmentation methods based on frequency phase guide, comprising: the original clinical image is preprocessed;Data input Gabor wavelet transform module is converted into complex value domain image;Complex value domain image is input into coding network, and the perception of phase component is selectively adjusted and the preference to different frequency components by complex value's mixed attention module, and complex value domain feature is obtained;Complex value domain feature is input into frequency self-adapting separation module, and complex value domain feature is decomposed into high-frequency feature and low-frequency feature, and important feature is screened out by frequency channel attention module by merging low-frequency feature and high-frequency feature;Important feature is input into next level encoder, and high-frequency feature is merged with decoder each level feature, and the segmentation result of clinical image is obtained.The application effectively utilizes phase information and frequency information, can better distinguish target area in clinical image from similar background, and improve segmentation effect.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of image segmentation, relates to computer vision technology, and particularly relates to a complex model clinical image semantic segmentation method based on frequency phase guidance. BACKGROUND

[0002] In numerous clinical tasks, accurate clinical image segmentation is of great significance. However, the process of manually drawing the contours of tumors or organs involves professional knowledge in multiple fields and requires analyzing multiple examination sequences. This process is subjective and time-consuming. The introduction of an automated segmentation method can significantly reduce the workload of doctors and improve the consistency and accuracy of analysis. With the help of automated segmentation technology, doctors can not only speed up the diagnosis process, but also more quickly develop treatment plans and monitor changes in the disease.

[0003] Recently, data-driven deep learning research has significantly advanced the automatic segmentation of clinical images. Deep learning methods, such as convolutional neural networks (CNN) and Transformers, have been widely used in clinical image segmentation tasks, bringing many advantages to medical image analysis. Deep learning methods can automatically learn features and rules, eliminating the need for manual feature extractor design, thereby reducing the need for human intervention and improving automation. However, when faced with the challenge of significant similarity and distribution proximity between clinical diseases, many current deep learning architectures still have some limitations. SUMMARY

[0004] The application aims to overcome the deficiencies in the prior art and provide a complex model clinical image semantic segmentation method based on frequency phase guidance. The designed segmentation model can effectively utilize phase information and frequency information, better distinguish target regions from similar backgrounds in clinical images, and assist in better screening of early lesions.

[0005] Technical solution: To achieve the above-mentioned purpose, the application provides a complex model clinical image semantic segmentation method based on frequency phase guidance, comprising the following steps:

[0006] S1: Preprocessing the original clinical image;

[0007] S2: Inputting the preprocessed image data obtained in step S1 into a Gabor wavelet transform module to convert it into a complex-valued domain image;

[0008] S3: Inputting the complex-valued domain image obtained in step S2 into an encoding network, selectively adjusting the perception of phase components and the preference for different frequency components through a complex-valued hybrid attention module to obtain a complex-valued domain feature;

[0009] S4: Input the complex-valued domain features obtained in step S3 into the constructed frequency adaptive separation module, decompose the complex-valued domain features into high-frequency features and low-frequency features, and merge the low-frequency features with the high-frequency features to filter out important features through the frequency channel attention module;

[0010] S5: The important features obtained in step S4 are input into the next level encoder, and the high-frequency features are merged with the features of each level of the decoder. The real and imaginary features are integrated at the last layer of the decoder to obtain the segmentation results of the clinical image.

[0011] Furthermore, the preprocessing method in step S1 is: first normalize to reduce the range of pixel values, and then use data enhancement strategy to expand the data set, specifically:

[0012] A1: Change the image size to the same size through interpolation so that it can be uniformly input into the network training later;

[0013] A2: Normalize the pixel value range of the image to 0-1 to reduce the interference caused by uneven lighting in clinical images;

[0014] A3: Use three data augmentation methods, namely flipping, rotation, and skewing, to expand the data volume.

[0015] Furthermore, the Gabor wavelet transform module in step S2 is composed of a filter bank with 4 scales and 5 directions, and the clinical image is converted into a complex domain by constructing a set of Gabor filters, specifically including:

[0016] B1: For the filter parameters: scale λ and direction θ, take 4 and 5 values ​​uniformly within a certain range respectively to construct a set of Gabor filters. The specific formula is as follows:

[0017]

[0018] Where x′ = xcos(θ) + ysin(θ), y′ = -xsin(θ) + ycos(θ), θ represents the tilt angle of the Gabor kernel function; λ and ψ represent the filter wavelength and phase offset, respectively; the standard deviation σ and the spatial aspect ratio δ determine the ellipticity of the Gabor kernel function;

[0019] B2: Filter the input image using the constructed Gabor filter set to obtain an image in the complex-valued domain.

[0020] Furthermore, in step S3, the encoding network adopts the UNet network, and the UNet network is modified into a complex form, including expressing the convolution kernel, activation function and batch normalization in complex form, as follows:

[0021] The complex convolution is expressed as a convolution operation between complex feature map h and convolution filter W, where h=a+ib, W=A+iB; according to the definition of complex operation, the calculation process of complex convolution is expressed as:

[0022] W*h=(A*a-B*b)+i(B*a+A*b);

[0023] The complex activation ReLU satisfies the Cauchy-Riemann equation, as follows:

[0024]

[0025] The batch normalization calculation of the complex input z is as follows:

[0026]

[0027] where, is the normalization of z, V is the covariance matrix, and E[·] is the mean of the input z.

[0028] Further, the complex-valued mixing attention module in step S3 includes a cross-phase fusion module and an amplitude deviation interference module, which are sequentially passed through to obtain the complex-valued domain feature.

[0029] Further, the cross-phase fusion module in step S3 is operated in the following manner:

[0030] C1: 3x3 and 1x1 convolution operations are respectively applied to obtain features F 3 and F 1 , respectively, and then these features are fused together, 1x1 convolution and S-shaped activation function σ(·) are performed to generate a attention map β representing the importance of local features, as follows:

[0031] β=σ(Conv 1×1 (F 3 +F 1 ));

[0032] C2: 1-β is expressed as β' at the same time, and the two soft attention maps are used to automatically select more representative features from each receptive field, as follows:

[0033] F c =cat(β′×F 3 ,β×F 1 ).

[0034] Further, the amplitude deviation interference module in step S3 is operated in the following manner:

[0035] D1: The complex-valued features outputted by step C2 and the input of the cross-phase fusion module are added pixel by pixel and inputted into the 3x3 as amplitude bias interference module;

[0036] D2: The output of step D1 is subjected to a 1x1 convolution operation to reduce the channel dimension, and is divided into two independent processing paths;

[0037] D3: The upper branch uses deep separable convolution to extract global context information;

[0038] D4: The lower branch feature is separated into amplitude and phase components, and then the amplitude information F s A slight perturbation is added, and the perturbed amplitude information and phase information are recombined, which is represented as follows:

[0039] Amplitude = Conv 1×1 · (Conv 1×1 · |F s |),

[0040] F complex = Amplitude x cos (arctan (|F s |)) + j x Amplitude x sin (arctan (|F s |));

[0041] D5: F complex is multiplied as a weight with the upper branch, an interference feature vector is obtained, and the original channel dimension is restored through a 3x3 convolution operation.

[0042] The complex-valued network in step S3 is proved to be a data-driven nonlinear wavelet packet, and the image after Gabor wavelet transform can be easily approximated by the complex-valued network. In addition, the complex-valued network can preserve and process complex-valued features, so they can preserve and restore the phase information, and thus the complex-valued network is more sensitive to the phase structure. The hybrid attention module first combines the multi-scale convolution through the cross-phase fusion module to adaptively adjust the receptive field to select important features. Secondly, the amplitude bias interference module solves the problem of over-reliance on specific frequencies by the network.

[0043] Further, the running process of the frequency adaptive separation module in step S4 is as follows:

[0044] E1: One-level wavelet decomposition is performed on the feature map along the channel dimension to obtain low-frequency coefficients and high-frequency coefficients

[0045] E2: The low-frequency coefficients are replaced with a zero matrix and the high-frequency coefficients are subjected to inverse wavelet transform to obtain high-frequency features F H ∈RC×H×W ;

[0046] E3: Inverse wavelet transform using zero value replacement of high frequency coefficients to obtain low frequency feature F L ∈R C×H×W , and high frequency feature F H is merged into F L ; cat Input frequency channel attention module;

[0047] E4: F cat is divided into n parts by channel, denoted as Then DCT operation is performed on F i and spliced together to obtain the overall compressed feature as follows:

[0048] F dct = cat([DCT(F0), DCT(F1), …, DCT(F n-1 )]);

[0049] E5: Apply full connection operation and activation operation to F dct , then multiply F cat to obtain F hybrid ∈R 2C×H×W .

[0050] The frequency adaptive separation module in step S4 is designed based on the theory that deep learning-based models tend to learn specific frequencies according to specific tasks. In the segmentation task, the encoder first learns lower frequencies, which contain most of the basic semantic information related to the overall representation. In contrast, the decoder is required to retain the required high frequency information containing fine-grained details, thereby emphasizing the extraction of subtle changes and contours. The frequency adaptive separation module first automatically separates the high frequency and low frequency features using wavelet decomposition, so that specific high frequency information can be selectively transmitted to the decoder through skip connection. In addition, the frequency adaptive separation module is equipped with a channel attention module to refine the most representative frequency features to the next layer of encoder.

[0051] Further, the step S5 is specifically:

[0052] F1: The feature F hybrid obtained in step E5 is changed to C channels through 1x1 convolution as the input of the next level encoder;

[0053] F2: The high frequency feature F H obtained in step E2 is transmitted to the corresponding decoder stage through skip connection to supplement fine-grained information;

[0054] F3: the real part feature and the imaginary part feature of the network output are integrated to obtain a final segmentation feature map, and the specific formula is as follows:

[0055]

[0056] Wherein F final , F real and F imag represent the final segmentation feature, the real part feature and the imaginary part feature respectively.

[0057] Further, the segmentation result is obtained by the activation function Softmax in the step F3, and the calculation formula of Softmax is as follows:

[0058]

[0059] The method explores the properties of the complex value model under the guidance of the phase and frequency perspective. The complex value based neural network can directly process the phase component, providing a better potential solution for the perception structure of clinical image segmentation. The application designs a complex hybrid attention module to selectively adjust the perception of the phase component and the bias of the model. In addition, the application designs a frequency adaptive separation module, which uses a module combining wavelet decomposition and frequency channel attention to emphasize the frequency features that the encoder and decoder prefer to process.

[0060] The complex model clinical image semantic segmentation method based on frequency phase guidance provided by the application provides a new idea for clinical image segmentation. It is an end-to-end architecture and fully utilizes the phase information to capture the structural features of the image, further improving the effectiveness of the method.

[0061] Advantages: compared with the prior art, the application has the following advantages:

[0062] 1. The method can realize automatic segmentation of clinical images based on deep learning, thereby avoiding manual heuristic learning and improving the generalization ability of the model.

[0063] 2. The method can fully utilize the phase information of the clinical image, explore the properties of the complex value model under the guidance of the phase and frequency perspective, and directly process the phase component based on the complex value based neural network. The method is suitable for intelligent analysis of clinical images, and successfully solves the problems of blurred boundaries and difficult differentiation between target regions and similar backgrounds in clinical image segmentation by utilizing the structural information of the phase and the preference of the network for specific frequency components, thereby improving the semantic segmentation effect of the clinical image. BRIEF DESCRIPTION OF DRAWINGS

[0064] Figure 1 is a flowchart of the method of the application;

[0065] Figure 2 is a structural framework diagram of the method of the present application applied to automatic segmentation of a target in a clinical image;

[0066] Figure 3 is a result diagram of a comparison between a preprocessed image and an original image taken as an example of a breast ultrasound image;

[0067] Figure 4 is a result diagram of a comparison between a complex-valued image obtained after step S2 and a real-valued image obtained in step S1 taken as an example of a breast ultrasound image;

[0068] Figure 5 is an architectural diagram of the frequency adaptive module in step S4;

[0069] Figure 6 is a segmentation result diagram of an input image by the frequency-phase guided complex model taken as an example of a breast ultrasound image. DETAILED DESCRIPTION

[0070] The present application will be further clarified by the following examples and figures, which should be understood as merely illustrating the present application and not limiting the scope thereof, and after reading the present application, those skilled in the art will make various modifications to the present application, which fall within the scope defined by the claims attached hereto.

[0071] As shown in Figure 1 and Figure 2 , the present application provides a frequency-phase guided complex model clinical image semantic segmentation method, comprising the following steps:

[0072] S1: preprocessing the original clinical image;

[0073] The preprocessing method in step S1 is to first normalize the pixel value range, and then use a data augmentation strategy to expand the data set, specifically:

[0074] A1: change the image size to the same size by interpolation, so as to unify the subsequent network training;

[0075] A2: normalize the pixel value range of the image to 0-1, to reduce the interference caused by uneven light in the clinical image;

[0076] A3: use three data augmentation methods, namely flipping, rotating and oblique cutting, to expand the data volume.

[0077] In this embodiment, the breast ultrasound image is taken as an example, and the comparison between the preprocessed image and the original image is specifically shown in Figure 3 .

[0078] S2: input the pre-processed image data obtained in step S1 into a Gabor wavelet transform module to convert into a complex value domain image;

[0079] The Gabor wavelet transform module is composed of a filter bank selected from 4 scales and 5 directions, which can cover the Gabor wavelet transform family and well describe the local structure information corresponding to the scale and direction selectivity;

[0080] The clinical image is converted to a complex value domain by constructing a set of Gabor filters, the phase information of the image is recovered, and good localization characteristics are maintained in both time domain and frequency domain, which specifically includes:

[0081] B1: the parameters of the filter, scale λ and direction θ, are respectively taken as 4 values and 5 values within a certain range to construct a set of Gabor filters, and the specific formula is as follows:

[0082]

[0083] Wherein, x' = xcos(θ) + ysin(θ), y' = -xsin(θ) + ycos(θ), θ represents the tilt angle of the Gabor kernel function; λ and ψ represent the filter wavelength and phase shift, respectively; the standard deviation σ and the space aspect ratio δ determine the ellipticity of the Gabor kernel function;

[0084] B2: filter the input image by constructing a set of Gabor filters to obtain a complex value domain image.

[0085] This embodiment takes breast ultrasound images as an example, and the comparison results of the complex value domain image obtained after step S2 and the real value image obtained in step S1 are specifically shown in Figure 4 .

[0086] S3: input the complex value domain image obtained in step S2 into an encoding network, and selectively adjust the perception of the phase component and the preference for different frequency components through a complex value hybrid attention module to obtain a complex value domain feature;

[0087] The encoding network adopts a UNet network, and the UNet network is modified into a complex number form, including representing the convolution kernel, the activation function and the batch normalization in a complex number form, which is specifically as follows:

[0088] The complex number convolution is represented as the convolution operation between the complex number feature map h and the convolution filter W, wherein h = a + ib, W = A + iB; according to the definition of complex number operation, the calculation process of complex number convolution is represented as:

[0089] W*h = (A*a-B*b) + i(B*a+A*b);

[0090] Complex activation ReLU, satisfying the Cauchy-Riemann equation, is as follows:

[0091]

[0092] Batch normalization calculation on complex input z is as follows:

[0093]

[0094] wherein, is the normalization of z, V is the covariance matrix, and E[·] is the mean of input z.

[0095] The mixed attention module of complex values includes a cross-phase fusion module and an amplitude deviation interference module, and the cross-phase fusion module is first passed through, and then the amplitude deviation interference module is passed through to obtain a complex value domain feature.

[0096] Referring to Figure 2 , the operation mode of the cross-phase fusion module is as follows:

[0097] C1: 3x3 and 1x1 convolution operations are respectively applied to obtain features F 3 and F 1 , and then the features are fused together, 1x1 convolution and S-shaped activation function σ(·) are performed to generate an attention map β representing the importance of local features, as follows:

[0098] β=σ(Conv 1×1 (F 3 +F 1 ));

[0099] C2: 1-β is expressed as β' at the same time, and the two soft attention maps are used to automatically select more representative features from each receptive field, as follows:

[0100] F c =cat(β′×F 3 ,β×F 1 )。

[0101] Referring to Figure 2 , the operation mode of the amplitude deviation interference module is as follows:

[0102] D1: The complex value features output by step C2 and the input of the cross-phase fusion module are added pixel by pixel and passed through 3x3 as the input of the amplitude deviation interference module;

[0103] D2: 1x1 convolution operation is performed on the output of step D1 to reduce the channel dimension, and the output is divided into two independent processing paths;

[0104] D3: The upper branch uses deep separable convolution to extract global context information;

[0105] D4: The lower branch feature is separated into amplitude and phase components, and then the amplitude information F s A subtle perturbation is added, and the perturbed amplitude information is recombined with the phase information, as follows:

[0106] Amplitude = Conv 1×1 ·(Conv 1×1 ·|F s |),

[0107] F complex =Amplitude×cos(arctan(|F s |))+j·Amplitude×sin(arctan(|F s |));

[0108] D5: F complex is multiplied as a weight with the upper branch, and the interference feature vector is obtained and the original channel dimension is restored through a 3x3 convolution operation.

[0109] S4: The complex-valued feature obtained in step S3 is input into the frequency adaptive separation module constructed, and the complex-valued feature is decomposed into high-frequency feature and low-frequency feature, and the low-frequency feature and the high-frequency feature are combined to filter out important features through the frequency channel attention module;

[0110] Referring to Figure 5 , the operation process of the frequency adaptive separation module is as follows:

[0111] E1: One-level wavelet decomposition is performed on the feature map along the channel dimension to obtain low-frequency coefficients and high-frequency coefficients

[0112] E2: The low-frequency coefficients are replaced with a zero matrix and the high-frequency coefficients are subjected to inverse wavelet transform to obtain high-frequency features F H ∈R C×H×W ;

[0113] E3: The high-frequency coefficients are replaced with zero values to obtain low-frequency features F L ∈R C×H×W , and the high-frequency features F H and the low-frequency features F L are combined into F cat and input into the frequency channel attention module;

[0114] E4: F cat is divided into n parts along the channel, denoted as Then, Fi The DCT operation is performed and spliced together to obtain the overall compressed features, as follows:

[0115] F dct = cat([DCT(F0), DCT(F1), …, DCT(F n-1 )]);

[0116] E5: Apply the full connection operation and the activation operation to F dct , and then multiply F cat to obtain F hybrid ∈R 2C×H×W .

[0117] S5: Input the important features obtained in step S4 into the next level encoder, and combine the high-frequency features with the features of each level of the decoder, and integrate the real part and the imaginary part features at the last layer of the decoder to obtain the segmentation result of the clinical image;

[0118] Step S5 is specifically:

[0119] F1: Change the channel number of the features F hybrid obtained in step E5 to C through 1x1 convolution, and input it into the next level encoder;

[0120] F2: Transmit the high-frequency features F H obtained in step E2 to the corresponding decoder stage through the skip connection to supplement the fine-grained information;

[0121] F3: Integrate the real part features and the imaginary part features output by the network to obtain the final segmentation feature map, and the specific formula is as follows:

[0122]

[0123] Where F final , F real and F imag represent the final segmentation feature, the real part feature and the imaginary part feature, respectively.

[0124] The segmentation result is obtained by the activation function Softmax in step F3, and the calculation formula of Softmax is as follows:

[0125]

[0126] In this embodiment, the breast ultrasound image is taken as an example, and the segmentation result of the input image by the frequency-phase guided complex model is specifically shown as Figure 6 , and the Dice and Jaccard thereof reach 0.914 and 0.842, respectively, which shows that the segmentation effect is very good.

[0127] In summary, the application is suitable for automatic segmentation of clinical images. This innovative deep learning method is based on a frequency phase guided complex model, providing a new idea for clinical image semantic segmentation, fully utilizing the structural information in the image, realizing fast automatic segmentation of clinical images, and improving the segmentation effect.

[0128] The embodiment also provides a frequency phase guided complex model based clinical image semantic segmentation system, which comprises a network interface, a memory and a processor; wherein the network interface is used for receiving and sending signals in the process of transmitting information with other external network elements; the memory is used for storing computer program instructions capable of running on the processor; and the processor is used for executing the steps of the consensus method when running the computer program instructions.

[0129] The embodiment also provides a computer storage medium storing a computer program, which can realize the above-described method when the processor executes the computer program. The computer readable medium can be considered tangible and non-transitory. Non-limiting examples of non-transitory tangible computer readable media include non-volatile memory circuits (such as flash memory circuits, erasable programmable read-only memory circuits, or mask read-only memory circuits), volatile memory circuits (such as static random access memory circuits or dynamic random access memory circuits), magnetic storage media (such as analog or digital magnetic tapes or hard disk drives), and optical storage media (such as CDs, DVDs, or Blu-ray discs), etc. The computer program includes processor executable instructions stored on at least one non-transitory tangible computer readable medium. The computer program can also include or depend on stored data. The computer program can include a basic input / output system (BIOS) that interacts with the hardware of a special-purpose computer, a device driver that interacts with a specific device of a special-purpose computer, one or more operating systems, user applications, background services, background applications, etc.

[0130] Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0131] The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other processing device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other processing device to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions specified in the flowchart block or blocks. Figure 1 The flowchart and / or block diagram in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments. In this regard, each flowchart block and / or combination of flowchart blocks can be implemented by electronic hardware, computer software, or combinations of both. The Figure 1 The flowchart and / or block diagram in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments. In this regard, each flowchart block and / or combination of flowchart blocks can be implemented by electronic hardware, computer software, or combinations of both. The

Claims

1. A method for semantic segmentation of clinical images based on a complex model guided by frequency and phase, characterized in that: The steps include: S1: Preprocessing of original clinical images; S2: The pre-processed image data obtained in step S1 is input into the Gabor wavelet transform module to be converted into a complex-valued domain image; S3: The complex-valued image obtained in step S2 is input into the encoding network, and the complex-valued mixed attention module selectively adjusts the perception of the phase component and the preference for different frequency components to obtain complex-valued features; S4: Input the complex-valued domain features obtained in step S3 into the constructed frequency adaptive separation module, decompose the complex-valued domain features into high-frequency features and low-frequency features, and merge the low-frequency features with the high-frequency features to filter out important features through the frequency channel attention module; S5: The important features obtained in step S4 are input into the next level encoder, and the high-frequency features are merged with the features of each level of the decoder. The real and imaginary features are integrated at the last layer of the decoder to obtain the segmentation result of the clinical image; In step S3, the encoding network adopts the UNet network, and the UNet network is modified to a complex form, including the convolution kernel, activation function and batch normalization are all expressed in complex form, as follows: Complex convolution is represented as the convolution operation between the complex feature map h and the convolution filter W, where h = a + ib, W = A + iB; according to the definition of complex operations, the calculation process of complex convolution is expressed as: W*h=(A*aB*b)+i(B*a+A*b); The complex activation ReLU satisfies the Cauchy-Riemann equation as shown below: Batch normalization for complex input z is calculated as follows: in, is the normalization of z, V is the covariance matrix and E[·] is the mean of the input z.

2. The method for clinical image semantic segmentation based on a complex model guided by frequency and phase according to claim 1, characterized in that: The preprocessing method in step S1 is: first normalize to reduce the range of pixel values, and then use data enhancement strategy to expand the data set, specifically: A1: Change the image size to the same size through interpolation so that it can be uniformly input into the network training later; A2: Normalize the pixel value range of the image to 0-1 to reduce the interference caused by uneven lighting in clinical images; A3: Use three data augmentation methods, namely flipping, rotation, and skewing, to expand the data volume.

3. The method for clinical image semantic segmentation based on a complex model guided by frequency and phase according to claim 1, characterized in that: The Gabor wavelet transform module in step S2 is composed of a filter group with 4 scales and 5 directions. The clinical image is converted into a complex domain by constructing a set of Gabor filters, which specifically includes: B1: For the filter parameters: scale λ and direction θ, take 4 values ​​and 5 values ​​respectively to construct a set of Gabor filters. The specific formula is as follows: Where x' = xcos(θ) + ysin(θ), y' = -xsin(θ) + ycos(θ), θ represents the tilt angle of the Gabor kernel function; λ and ψ represent the filter wavelength and phase offset, respectively; the standard deviation σ and the spatial aspect ratio δ determine the ellipticity of the Gabor kernel function; B2: Filter the input image using the constructed Gabor filter set to obtain an image in the complex-valued domain.

4. The method for clinical image semantic segmentation based on a complex model guided by frequency and phase according to claim 1, characterized in that: The complex-valued hybrid attention module in step S3 includes a cross-phase fusion module and an amplitude deviation interference module, which first passes through the cross-phase fusion module and then passes through the amplitude deviation interference module to obtain the complex-valued domain features.

5. The method for clinical image semantic segmentation based on a complex model guided by frequency and phase according to claim 4, characterized in that: The operation mode of the cross-phase fusion module in step S3 is: C1: Apply 3×3 and 1×1 convolution operations to obtain features F respectively. 3 and F 1 , and then these features are fused together, subjected to 1×1 convolution and S-type activation function σ(·), to generate an attention map β representing the importance of local features, as shown below: β=σ(Conv 1×1 (F 3 +F 1 )); C2: 1-β is expressed as β' at the same time, and these two soft attention maps are used to automatically select more representative features from each receptive field, as shown below: F c =cat(β'×F 3 ,β×F 1 )。 6. The method for clinical image semantic segmentation based on a frequency-phase guided complex model according to claim 5, characterized in that: The operation mode of the amplitude deviation interference module in step S3 is: D1: The complex-valued features output from step C2 and the input of the cross-phase fusion module are added pixel by pixel and then multiplied by 3×3 as the input of the amplitude deviation interference module; D2: Apply a 1×1 convolution operation to the output of step D1 to reduce the channel dimension and split it into two independent processing paths; D3: The upper branch uses deep separable volumes to extract global context information; D4: The lower branch features are separated into amplitude and phase components, and then the amplitude information F is processed by applying convolution with controlled noise. s A slight perturbation is added, and the perturbed amplitude information is recombined with the phase information as follows: Amplitude=Conv 1×1 ·(Conv 1×1 ·|F s |), F complex =Amplitude×cos(arctan(|F s |))+j·Amplitude×sin(arctan(|F s |)); D5:F complex It is multiplied with the upper branch as a weight to obtain the interference feature vector and restore the original channel dimension through a 3×3 convolution operation.

7. The method for clinical image semantic segmentation based on a complex model guided by frequency and phase according to claim 1, characterized in that: The operation process of the frequency adaptive separation module in step S4 is as follows: E1: Perform a first-level wavelet decomposition on the feature map along the channel dimension to obtain low-frequency coefficients and high frequency coefficients E2: Replace the low-frequency coefficients with zero matrix and perform inverse wavelet transform together with the high-frequency coefficients to obtain the high-frequency feature F H ∈R C ×H×W ; E3: Use zero values ​​to replace high-frequency coefficients and perform inverse wavelet transform to obtain low-frequency features F L ∈R C×H×W , and the high-frequency feature F H and low-frequency features F L Merge into F cat Input frequency channel attention module; E4: F cat Divide the channel into n parts, denoted as Then to F i Perform a DCT operation and concatenate together to get the overall compressed features as follows: F dct =cat([DCT(F0),DCT(F1),…,DCT(F n-1 )]); E5: For F dct Apply the fully connected operation and activation operation, and then with F cat Multiply to get F hybrid ∈R 2C×H×W .

8. The method for clinical image semantic segmentation based on a complex model guided by frequency and phase according to claim 7, characterized in that: The step S5 is specifically as follows: F1: Feature F obtained in step E5 hybrid The number of channels is changed to C through 1×1 convolution, which is used as the input of the next encoder. F2: high-frequency feature F obtained in step E2 H Transmitted to the corresponding decoder stage via skip connections to supplement fine-grained information; F3: Integrate the real and imaginary features of the network output to obtain the final segmentation feature map. The specific formula is as follows: Among them F final , F real and F imag Represent the final segmentation features, real features and imaginary features respectively.

9. The method for clinical image semantic segmentation based on a complex model guided by frequency and phase according to claim 8, characterized in that: In step F3, the segmentation result is obtained by activating the Softmax function. The calculation formula of Softmax is:

Images