Heart ultrasound image segmentation method based on state space model and feature interactive perception

Abstract

A heart ultrasound image segmentation method based on a state space model and feature interactive perception relates to the field of image segmentation, and a dual perception state space enhancement module DPSSEM uses two SSM modules and combines the two SSM modules with SAM and CAM respectively, so that more information can be obtained from two dimensions of space and channel, and the image segmentation efficiency is improved. The dual perception state space enhancement module DPSSEM carries out fine interaction and fusion among different feature maps, the feature identification capability and expression capability are significantly enhanced, the accuracy and robustness of a heart structure in ultrasound image segmentation are effectively improved, and the feature interaction perception decoding module FIPDM carries out multi-feature processing branch and interaction operation, so that the accuracy and robustness of the heart structure in ultrasound image segmentation are effectively improved. And performing multi-path refinement and enhancement on the extracted features in a decoding stage.

Description

Technical Field

[0001] The present invention relates to the field of image segmentation, and in particular to a cardiac ultrasound image segmentation method based on a state space model and feature interactive perception. Background Art

[0002] Cardiac ultrasound image segmentation is crucial for the diagnosis and treatment of heart diseases. However, traditional image segmentation methods face many challenges when processing cardiac ultrasound images. Due to the limitations of ultrasound imaging principles, cardiac ultrasound images often have problems such as noise, artifacts, and low contrast, making it difficult to accurately segment cardiac structures. At the same time, the dynamic characteristics of the heart cause its shape and position to change continuously during different cardiac cycles, further increasing the difficulty of segmentation.

[0003] Mamba's state space model (SSM) brings new opportunities to solve these problems. The state space model (SSM) can effectively capture the complex characteristics of the heart that change over time by accurately modeling and predicting the system state, just like processing a dynamic system. It can selectively propagate or suppress information based on the input data, enhance the ability to process complex inputs, focus on key information in cardiac ultrasound images, and filter out irrelevant interference. Feature interaction perception can deeply explore the relationship between different features in the image and further enhance the recognition of cardiac structure. Combining Mamba's SSM with feature interaction perception is expected to give full play to the advantages of both, improve the accuracy and robustness of cardiac ultrasound image segmentation, and provide solid technical support for the precise diagnosis and treatment of heart diseases. Therefore, it is very important to explore a cardiac ultrasound image segmentation method based on state space model and feature interaction perception to better meet clinical needs and provide a more accurate and reliable basis for the diagnosis and treatment of heart diseases. Summary of the invention

[0004] In order to overcome the deficiencies of the above technologies, the present invention provides an image segmentation method for improving the accuracy and robustness of cardiac ultrasound image segmentation.

[0005] The technical solution adopted by the present invention to overcome the technical problems is:

[0006] A cardiac ultrasound image segmentation method based on a state space model and feature interaction perception, comprising:

[0007] a) Obtain N cardiac ultrasound images, and obtain a cardiac ultrasound image dataset E, where E = {E1, E2, ..., E i ,...,E N}, E i is the i-th cardiac ultrasound image, i∈{1,...,N};

[0008] b) Take the i-th cardiac ultrasound image Ei Preprocessing is performed to obtain the preprocessed cardiac ultrasound image S i All N preprocessed cardiac ultrasound images constitute a preprocessed cardiac ultrasound image dataset S, S = {S1, S2, ..., S i ,...,S N};

[0009] c) dividing the preprocessed cardiac ultrasound image dataset S into a training set, a validation set, and a test set;

[0010] d) establishing a cardiac ultrasound image segmentation network model consisting of an encoder, a skip connection unit, and a decoder;

[0011] e) The i-th preprocessed cardiac ultrasound image S in the training set i Input into the encoder of the cardiac ultrasound image segmentation network model, and output the feature map

[0012] f) The feature map Input into the decoder of the cardiac ultrasound image segmentation network model, extract features through the jump connection unit, and obtain the predicted segmentation image

[0013] g) Calculate the total loss L total , through the total loss L total Training a cardiac ultrasound image segmentation network model to obtain an optimized cardiac ultrasound image segmentation network model;

[0014] h) Take the i-th preprocessed cardiac ultrasound image S in the test set i Input into the optimized cardiac ultrasound image segmentation network model, and output the predicted segmentation image Preferably, in step a), N cardiac ultrasound images are obtained from the CAMUS cardiac segmentation ultrasound image dataset, and the value of N is 500.

[0015] Preferably, the preprocessed cardiac ultrasound image dataset S in step c) is divided into a training set, a validation set, and a test set in a ratio of 7:2:1.

[0016] Further, step b) comprises the following steps:

[0017] b-1) Use the center_crop() function in OpenCV to crop the i-th cardiac ultrasound image E i Perform a center cropping operation with a size of 256×256 to obtain the i-th cardiac ultrasound image E after cropping i ′;

[0018] b-2) Use the random_flip_rotate() function in OpenCV to rotate the i-th cardiac ultrasound image E after cropping i 'Flip along the horizontal or vertical axis with a probability of 0.5, and the flip angle is 90 degrees to obtain the i-th cardiac ultrasound image E after flipping i ″;

[0019] b-3) Flip the i-th cardiac ultrasound image E i ″Perform z-score standardization to obtain the preprocessed cardiac ultrasound image S i .

[0020] Further, step e) comprises the following steps:

[0021] e-1) The encoder of the cardiac ultrasound image segmentation network model is composed of a convolutional layer, a first dual perception state space enhancement module DPSSEM, a first downsampling unit, a second dual perception state space enhancement module DPSSEM, a second downsampling unit, a third dual perception state space enhancement module DPSSEM, a third downsampling unit, a fourth dual perception state space enhancement module DPSSEM, and a fourth downsampling unit;

[0022] e-2) The i-th preprocessed cardiac ultrasound image S in the training set i Input into the convolutional layer of the encoder and output the feature map

[0023] e-3) The first dual perception state space enhancement module DPSSEM of the encoder is composed of a depthwise separable convolutional layer DSConv, a first LayerNorm layer, a first multi-layer perceptron MLP, a first branch, a second branch, a third branch, a second LayerNorm layer, and a second multi-layer perceptron MLP. The feature map The first dual perception state space enhancement module DPSSEM is sequentially input into the deep separable convolution layer DSConv, the first LayerNorm layer, and the first multi-layer perceptron MLP to output the feature map. The feature map Use the torch.chunk() function in pytorch to perform the Split operation and output the feature maps separately Feature Map Feature Map The first branch of the first dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a spatial attention mechanism SAM, a SiLU activation function, and a state space model SSM. Input into the first branch and output the feature map The second branch of the first dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a hole convolutional layer, and a SiLU activation function. Input into the second branch and output the feature map The third branch of the first dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a channel attention mechanism CAM, a SiLU activation function, and a state space model SSM. Input into the third branch and output the feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map It is sequentially input into the second LayerNorm layer of the first dual perception state space enhancement module DPSSEM and the second multi-layer perceptron MLP, and the output is the feature map

[0024] e-4) The first downsampling unit of the encoder is composed of a convolutional layer and a maximum pooling layer in sequence. Input into the first downsampling unit and output the feature map

[0025] e-5) The second dual perception state space enhancement module DPSSEM of the encoder consists of a depthwise separable convolutional layer DSConv, a first LayerNorm layer, a first multi-layer perceptron MLP, a first branch, a second branch, a third branch, a second LayerNorm layer, and a second multi-layer perceptron MLP. The output is the feature map. The feature map Use the torch.chunk() function in pytorch to perform the Split operation and output the feature maps separately Feature Map Feature Map The first branch of the second dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a spatial attention mechanism SAM, a SiLU activation function, and a state space model SSM. Input into the first branch and output the feature map The second branch of the second dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a hole convolutional layer, and a SiLU activation function. Input into the second branch and output the feature map The third branch of the second dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a channel attention mechanism CAM, a SiLU activation function, and a state space model SSM. Input into the third branch and output the feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map It is sequentially input into the second LayerNorm layer of the second dual perception state space enhancement module DPSSEM and the second multi-layer perceptron MLP, and the output is the feature map

[0026] e-6) The second downsampling unit of the encoder is composed of a convolutional layer and a maximum pooling layer in sequence. Input into the second downsampling unit and output the feature map

[0027] e-7) The third dual perception state space enhancement module DPSSEM of the encoder is composed of a depth separable convolution layer DSConv, a first LayerNorm layer, a first multi-layer perceptron MLP, a first branch, a second branch, a third branch, a second LayerNorm layer, and a second multi-layer perceptron MLP. The feature map The output is the deep separable convolutional layer DSConv, the first LayerNorm layer, and the first multi-layer perceptron MLP of the third dual perception state space enhancement module DPSSEM, and the feature map is obtained. The feature map Use the torch.chunk() function in pytorch to perform the Split operation and output the feature maps separately Feature Map Feature Map The first branch of the third dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a spatial attention mechanism SAM, a SiLU activation function, and a state space model SSM. Input into the first branch and output the feature map The second branch of the third dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a hole convolutional layer, and a SiLU activation function. Input into the second branch and output the feature map The third branch of the third dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a channel attention mechanism CAM, a SiLU activation function, and a state space model SSM. Input into the third branch and output the feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map It is sequentially input into the second LayerNorm layer of the third dual perception state space enhancement module DPSSEM and the second multi-layer perceptron MLP, and the output is the feature map

[0028] e-8) The third downsampling unit of the encoder is composed of a convolutional layer and a maximum pooling layer in sequence, Input into the third downsampling unit and output the feature map

[0029] e-9) The fourth dual perception state space enhancement module DPSSEM of the encoder is composed of a depth separable convolution layer DSConv, a first LayerNorm layer, a first multi-layer perceptron MLP, a first branch, a second branch, a third branch, a second LayerNorm layer, and a second multi-layer perceptron MLP. The feature map The output is the deep separable convolutional layer DSConv, the first LayerNorm layer, and the first multi-layer perceptron MLP of the fourth dual perception state space enhancement module DPSSEM, and the feature map is obtained. The feature map Use the torch.chunk() function in pytorch to perform the Split operation and output the feature maps separately Feature Map Feature Map The first branch of the fourth dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a spatial attention mechanism SAM, a SiLU activation function, and a state space model SSM. Input into the first branch and output the feature map The second branch of the fourth dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a hole convolutional layer, and a SiLU activation function. Input into the second branch and output the feature map The third branch of the fourth dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a channel attention mechanism CAM, a SiLU activation function, and a state space model SSM. Input into the third branch and output the feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map It is sequentially input into the second LayerNorm layer of the fourth dual perception state space enhancement module DPSSEM and the second multi-layer perceptron MLP, and the feature map is output.

[0030] e-10) The fourth downsampling unit of the encoder is composed of a convolutional layer and a maximum pooling layer in sequence, Input to the fourth downsampling unit and output the feature map Preferably, in step e-2), the convolution kernel size of the convolution layer of the encoder is 3×3; the convolution kernel size of the depthwise separable convolution layer DSConv of the first dual perception state space enhancement module DPSSEM, the second dual perception state space enhancement module DPSSEM, the third dual perception state space enhancement module DPSSEM, and the fourth dual perception state space enhancement module DPSSEM are all 3×3, the padding is 1, and the step size is 1; the convolution kernel size of the convolution layer of the first branch is 5×5, the padding is 2, and the step size is 1; The convolution kernel size of the convolution layer of the second branch is 3×3, the padding is 1, and the stride is 1. The convolution kernel size of the hole convolution layer of the second branch is 3×3, the padding is 1, the stride is 1, and the expansion rate is 2. The convolution kernel size of the convolution layer of the third branch is 7×7, the padding is 3, and the stride is 1; the convolution kernel size of the convolution layer of the first downsampling unit, the second downsampling unit, the third downsampling unit, and the fourth downsampling unit is 3×3, the padding is 1, and the stride is 1. The pooling kernel of the maximum pooling layer is 2×2, and the stride is 2.

[0031] Further, step f) comprises the following steps:

[0032] f-1) The decoder of the cardiac ultrasound image segmentation network model is composed of a convolutional layer, a first feature interactive perception decoding module FIPDM, a first upsampling unit, a second feature interactive perception decoding module FIPDM, a second upsampling unit, a third feature interactive perception decoding module FIPDM, a third upsampling unit, a fourth feature interactive perception decoding module FIPDM, and a fourth upsampling unit, and the jump connection unit of the cardiac ultrasound image segmentation network model is composed of a first focused feature enhancement transfer module FFETM, a second focused feature enhancement transfer module FFETM, a third focused feature enhancement transfer module FFETM, and a fourth focused feature enhancement transfer module FFETM;

[0033] f-2) The fourth focused feature enhancement transfer module FFETM of the jump connection unit is composed of the first convolution layer, the first branch, the second branch, the maximum pooling layer, the second convolution layer, and the Sigmoid activation function. Input to the first convolutional layer of the fourth focused feature enhancement transfer module FFETM, and output the feature map The feature map Use the torch.chunk() function in pytorch to perform the Split operation and output the feature maps separately Feature Map The first branch of the fourth focused feature enhancement transfer module FFETM is composed of a spatial attention mechanism and a channel attention mechanism, which converts the feature map Input into the first branch and output the feature map The second branch of the fourth focused feature enhancement transfer module FFETM is composed of a convolutional layer, a ReLU activation function, and a BatchNorm layer. Input into the second branch and output the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map The output is sequentially input into the maximum pooling layer, the second convolutional layer, and the Sigmoid activation function of the fourth focused feature enhancement transfer module FFETM, and the feature map is obtained.

[0034] f-3) The feature map With feature map Perform addition operation to obtain feature map

[0035] f-4) The fourth feature interaction perception decoding module FIPDM of the decoder consists of a convolutional layer, a LayerNorm layer, a GeLU activation function, a first branch, a second branch, a third branch, and an MLP layer. It is sequentially input into the convolution layer, LayerNorm layer, and GeLU activation function of the fourth feature interaction perception decoding module FIPDM, and the feature map is output. The feature map Use the torch.chunk() function in pytorch to perform the Split operation and output the feature maps separately Feature Map Feature Map The first branch of the fourth feature interaction perception decoding module FIPDM is composed of a global maximum pooling layer, a convolutional layer, a GeLU activation function, a depth-separable convolutional layer, and a BatchNorm layer. Input into the first branch and output the feature map The second branch of the fourth feature interaction perception decoding module FIPDM is composed of a depth-separable convolutional layer, a BatchNorm layer, and a GeLU activation function. Input into the second branch and output the feature map The third branch of the fourth feature interaction perception decoding module FIPDM is composed of a global average pooling layer and a convolutional layer in sequence. Input into the third branch and output the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map feature map Feature Map Feature Map The Concat operation is performed through the torch.cat() function in pytorch to obtain the feature map The feature map Input to the MLP layer of the fourth feature interaction perception decoding module FIPDM, and output the feature map The feature map With feature map Perform addition operation to obtain feature map

[0036] f-5) The fourth upsampling unit of the decoder is composed of a transposed convolutional layer and a BatchNorm layer in sequence. Input to the fourth upsampling unit and output the feature map

[0037] f-6) The third focused feature enhancement transfer module FFETM of the jump connection unit is composed of the first convolution layer, the first branch, the second branch, the maximum pooling layer, the second convolution layer, and the Sigmoid activation function. Input to the first convolutional layer of the third focused feature enhancement transfer module FFETM, and output the feature map The feature map Use the torch.chunk() function in pytorch to perform the Split operation and output the feature maps separately Feature Map The first branch of the third focused feature enhancement transfer module FFETM is composed of the spatial attention mechanism and the channel attention mechanism in sequence. Input into the first branch and output the feature map The second branch of the third focused feature enhancement transfer module FFETM is composed of a convolutional layer, a ReLU activation function, and a BatchNorm layer. Input into the second branch and output the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map It is sequentially input into the maximum pooling layer, the second convolutional layer, and the Sigmoid activation function of the third focused feature enhancement transfer module FFETM, and the feature map is output.

[0038] f-7) The feature map With feature map Perform addition operation to obtain feature map

[0039] f-8) The third feature interaction perception decoding module FIPDM of the decoder consists of a convolutional layer, a LayerNorm layer, a GeLU activation function, a first branch, a second branch, a third branch, and an MLP layer. It is sequentially input into the convolution layer, LayerNorm layer, and GeLU activation function of the third feature interaction perception decoding module FIPDM, and the feature map is output. The feature map Use the torch.chunk() function in pytorch to perform the Split operation and output the feature maps separately Feature Map Feature Map The first branch of the third feature interaction perception decoding module FIPDM is composed of a global maximum pooling layer, a convolutional layer, a GeLU activation function, a depth-separable convolutional layer, and a BatchNorm layer. Input into the first branch and output the feature map The second branch of the third feature interaction perception decoding module FIPDM is composed of a depth-separable convolutional layer, a BatchNorm layer, and a GeLU activation function. Input into the second branch and output the feature map The third branch of the third feature interaction perception decoding module FIPDM is composed of a global average pooling layer and a convolutional layer in sequence. Input into the third branch and output the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map feature map Feature Map Feature Map The Concat operation is performed through the torch.cat() function in pytorch to obtain the feature map The feature map Input into the MLP layer of the third feature interaction perception decoding module FIPDM, and output the feature map The feature map With feature map Perform addition operation to obtain feature map

[0040] f-9) The third upsampling unit of the decoder is composed of a transposed convolutional layer and a BatchNorm layer in sequence. Input to the third upsampling unit and output the feature map

[0041] f-10) The second focused feature enhancement transfer module FFETM of the jump connection unit is composed of the first convolution layer, the first branch, the second branch, the maximum pooling layer, the second convolution layer, and the Sigmoid activation function. Input to the first convolutional layer of the second focused feature enhancement transfer module FFETM, and output the feature map The feature map Use the torch.chunk() function in pytorch to perform the Split operation and output the feature maps separately Feature Map The first branch of the second focused feature enhancement transfer module FFETM is composed of the spatial attention mechanism and the channel attention mechanism in sequence. Input into the first branch and output the feature map The second branch of the second focused feature enhancement transfer module FFETM is composed of a convolutional layer, a ReLU activation function, and a BatchNorm layer. Input into the second branch and output the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map It is sequentially input into the maximum pooling layer, the second convolutional layer, and the Sigmoid activation function of the second focused feature enhancement transfer module FFETM, and the feature map is output.

[0042] f-11) The feature map With feature map Perform addition operation to obtain feature map

[0043] f-12) The second feature interaction perception decoding module FIPDM of the decoder consists of a convolutional layer, a LayerNorm layer, a GeLU activation function, a first branch, a second branch, a third branch, and an MLP layer. It is sequentially input into the convolutional layer, LayerNorm layer, and GeLU activation function of the second feature interaction perception decoding module FIPDM, and the feature map is output. The feature map Use the torch.chunk() function in pytorch to perform the Split operation and output the feature maps separately Feature Map Feature Map The first branch of the second feature interaction perception decoding module FIPDM is composed of a global maximum pooling layer, a convolutional layer, a GeLU activation function, a depth-separable convolutional layer, and a BatchNorm layer. Input into the first branch and output the feature map The second branch of the second feature interaction perception decoding module FIPDM is composed of a depth-separable convolutional layer, a BatchNorm layer, and a GeLU activation function. Input into the second branch and output the feature map The third branch of the second feature interaction perception decoding module FIPDM is composed of a global average pooling layer and a convolutional layer. Input into the third branch and output the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map feature map Feature Map Feature Map The Concat operation is performed through the torch.cat() function in pytorch to obtain the feature map The feature map Input into the MLP layer of the second feature interaction perception decoding module FIPDM, and output the feature map The feature map With feature map Perform addition operation to obtain feature map

[0044] f-13) The second upsampling unit of the decoder is composed of a transposed convolutional layer and a BatchNorm layer in sequence. Input to the second upsampling unit and output the feature map

[0045] f-14) The first focused feature enhancement transfer module FFETM of the jump connection unit is composed of the first convolution layer, the first branch, the second branch, the maximum pooling layer, the second convolution layer, and the Sigmoid activation function. Input into the first convolutional layer of the first focused feature enhancement transfer module FFETM, and output the feature map The feature map Use the torch.chunk() function in pytorch to perform the Split operation and output the feature maps separately Feature Map The first branch of the first focused feature enhancement transfer module FFETM is composed of the spatial attention mechanism and the channel attention mechanism in sequence. Input into the first branch and output the feature map The second branch of the first focused feature enhancement transfer module FFETM is composed of a convolutional layer, a ReLU activation function, and a BatchNorm layer. Input into the second branch and output the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map It is sequentially input into the maximum pooling layer, the second convolutional layer, and the Sigmoid activation function of the first focused feature enhancement transfer module FFETM, and the feature map is output.

[0046] f-15) The feature map With feature map Perform addition operation to obtain feature map

[0047] f-16) The first feature interaction perception decoding module FIPDM of the decoder consists of a convolutional layer, a LayerNorm layer, a GeLU activation function, a first branch, a second branch, a third branch, and an MLP layer. It is sequentially input into the convolution layer, LayerNorm layer, and GeLU activation function of the first feature interaction perception decoding module FIPDM, and the feature map is output. The feature map Use the torch.chunk() function in pytorch to perform the Split operation and output the feature maps separately Feature Map Feature Map The first branch of the first feature interaction perception decoding module FIPDM is composed of a global maximum pooling layer, a convolutional layer, a GeLU activation function, a depth-separable convolutional layer, and a BatchNorm layer. Input into the first branch and output the feature map The second branch of the first feature interaction perception decoding module FIPDM is composed of a depth-separable convolutional layer, a BatchNorm layer, and a GeLU activation function. Input into the second branch and output the feature map The third branch of the first feature interaction perception decoding module FIPDM is composed of a global average pooling layer and a convolutional layer in sequence. Input into the third branch and output the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map feature map Feature Map Feature Map The Concat operation is performed through the torch.cat() function in pytorch to obtain the feature map The feature map Input into the MLP layer of the first feature interaction perception decoding module FIPDM, and output the feature map The feature map With feature map Perform addition operation to obtain feature map

[0048] f-17) The first upsampling unit of the decoder is composed of a transposed convolutional layer and a BatchNorm layer in sequence. Input into the first upsampling unit and output the feature map

[0049] f-18) The feature map Input into the convolutional layer of the decoder, and output the predicted segmentation image Preferably, the convolution kernel size of the first convolution layer of the first focus feature enhancement transfer module FFETM, the second focus feature enhancement transfer module FFETM, the third focus feature enhancement transfer module FFETM, and the fourth focus feature enhancement transfer module FFETM of the decoder is 3×3, the padding is 1, and the step size is 1; the convolution kernel size of the second convolution layer is 3×3, the padding is 1, and the step size is 1; the pooling kernel of the maximum pooling layer is 2×2, the step size is 1, and the convolution kernel size of the convolution layer of the second branch is 1×1, and the step size is 1; the first feature interaction perception decoding module FIPDM, the second feature interaction perception decoding module FIPDM, the third feature interaction perception decoding module FIPDM, and the fourth feature interaction The convolution kernel size of the convolution layer of the perceptual decoding module FIPDM is 3×3, and the step size is 1. The convolution kernel size of the convolution kernel of the first branch is 3×3, and the step size is 1. The convolution kernel size of the depthwise separable convolution layer of the first branch is 3×3, and the step size is 1, and the padding is 1. The convolution kernel size of the depthwise separable convolution layer of the second branch is 3×3, and the step size is 1, and the padding is 1. The convolution kernel size of the convolution kernel of the convolution layer of the third branch is 1×1, and the step size is 1; the convolution kernel size of the transposed convolution layer of the first upsampling unit, the second upsampling unit, the third upsampling unit, and the fourth upsampling unit of the decoder is 3×3, the padding is 1, and the step size is 2; the convolution kernel size of the convolution layer of the decoder in step f-18) is 1×1.

[0050] Further, step g) comprises the following steps:

[0051] g-1) through the formula L total =αL Dice +βL CE +γL boundary +δL Tversky Calculate the total loss L total , where L Dice is the Dice loss, L CE is the cross entropy loss, L boundary is the boundary loss, L Tversky is Tversky loss, α, β, γ, δ are all hyperparameters;

[0052] g-2) Use the SGD optimizer to take advantage of the total loss L total Train the cardiac ultrasound image segmentation network model to obtain an optimized cardiac ultrasound image segmentation network model.

[0053] Preferably, when training the cardiac ultrasound image segmentation network model, the learning rate is set to 0.001, the batch size is set to 16, the number of iterations is set to 1200, the hyperparameter α is set to 0.3, the hyperparameter β is set to 0.3, the hyperparameter γ is set to 0.2, and the hyperparameter δ is set to 0.2.

[0054] The beneficial effects of the present invention are as follows: the dual perception state space enhancement module DPSSEM can not only obtain more information from the two dimensions of space and channel by using two SSM modules and combining them with SAM and CAM respectively, but also perform deep fusion between different scales and different features. Compared with a single SSM module, the setting of the dual SSM allows each module to focus on the changes in different feature dimensions, improves the model's understanding depth of structural information, and further enhances the diversity of feature expression and detail resolution. Through the design of a multi-branch structure, the dual perception state space enhancement module DPSSEM performs fine interaction and fusion between different feature maps, significantly enhances the recognition and expression capabilities of features, and effectively improves the accuracy and robustness of cardiac structure in ultrasound image segmentation. The feature interaction perception decoding module FIPDM performs multi-path refinement and enhancement of the extracted features in the decoding stage through multiple feature processing branches and interactive operations. The module introduces a variety of feature fusion methods, so that the features retain more detail information while restoring the image spatial resolution, thereby achieving high-precision image reconstruction and improving the network's segmentation accuracy for cardiac ultrasound images. The focused feature enhancement transfer module FFETM can filter and transfer feature information in a focused manner, so that the decoder can better receive the important features transferred by the encoder. By focusing on the features in the jump connection, the module effectively reduces noise interference and significantly improves the adaptability of the segmentation model to cardiac structures of different scales and morphologies. It can better improve the accuracy and robustness of cardiac ultrasound image segmentation, achieve detailed anatomical structure recognition, adapt to complex cardiac anatomical features, and meet clinical needs for high-precision segmentation. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 is a structural diagram of the cardiac ultrasound image segmentation network model of the present invention;

[0056] Figure 2 It is a structural diagram of the dual perception state space enhancement module DPSSEM of the present invention;

[0057] Figure 3 It is a structural diagram of the feature interaction perception decoding module FIPDM of the present invention;

[0058] Figure 4 It is a structural diagram of the focusing feature enhancement transfer module FFETM of the present invention. DETAILED DESCRIPTION

[0059] The following is combined with Figure 1 To Attachment Figure 4 The present invention is further described.

[0060] A cardiac ultrasound image segmentation method based on a state space model and feature interaction perception, comprising:

[0061] a) Obtain N cardiac ultrasound images, and obtain a cardiac ultrasound image dataset E, where E = {E1, E2, ..., E i ,...,E N}, E i is the i-th cardiac ultrasound image, i∈{1,...,N}.

[0062] b) Take the i-th cardiac ultrasound image E i Preprocessing is performed to obtain the preprocessed cardiac ultrasound image S i All N preprocessed cardiac ultrasound images constitute a preprocessed cardiac ultrasound image dataset S, S = {S1, S2, ..., S i ,...,S N}.

[0063] c) Divide the preprocessed cardiac ultrasound image dataset S into a training set, a validation set, and a test set.

[0064] d) Establish a cardiac ultrasound image segmentation network model consisting of an encoder, a skip connection unit, and a decoder.

[0065] e) The i-th preprocessed cardiac ultrasound image S in the training set i Input into the encoder of the cardiac ultrasound image segmentation network model, and output the feature map

[0066] f) The feature map Input into the decoder of the cardiac ultrasound image segmentation network model, extract features through the jump connection unit, and obtain the predicted segmentation image

[0067] g) Calculate the total loss L total , through the total loss L total Train the cardiac ultrasound image segmentation network model to obtain an optimized cardiac ultrasound image segmentation network model.

[0068] h) Take the i-th preprocessed cardiac ultrasound image S in the test set i Input into the optimized cardiac ultrasound image segmentation network model, and output the predicted segmentation image In order to verify the reliability of this patented method, the final predicted segmentation image A systematic evaluation is conducted and compared with existing deep learning segmentation methods for cardiac ultrasound images, as shown in Table 1:

[0069] Table 1 Comparison of different experimental methods

[0070]

[0071] In our experimental method, we compared methods such as U-Net, ACNNS, nn-UNet, and GUDU. Our evaluation indicators used Dice, Jaccard, average surface distance (ASD), and 95% Hausdorff distance (95HD), and achieved performances of 90.44%, 82.71%, 2.05voxel, and 7.36voxel, respectively. From the data in the table, it can be seen that the Dice index of the method of the present invention is 0.57% higher than that of the previous GUDU method, and the Jaccard index is 1.1% higher than that of the GUDU method. In addition, the ASD and 95HD indicators of the method of the present invention also achieved the best results compared with other methods. Overall, such performance improvement shows that our method can reduce boundary errors while accurately segmenting cardiac structures, thereby obtaining better cardiac ultrasound image segmentation effects and providing higher quality image support for clinical diagnosis and treatment.

[0072] In one embodiment of the present invention, in step a), N cardiac ultrasound images are obtained from the CAMUS cardiac segmentation ultrasound image dataset, where N is 500.

[0073] In one embodiment of the present invention, the preprocessed cardiac ultrasound image dataset S in step c) is divided into a training set, a validation set, and a test set in a ratio of 7:2:1.

[0074] In one embodiment of the present invention, step b) comprises the following steps:

[0075] b-1) Use the center_crop() function in OpenCV to crop the i-th cardiac ultrasound image E i Perform a center cropping operation with a size of 256×256 to obtain the i-th cardiac ultrasound image E after cropping i ′.

[0076] b-2) Use the random_flip_rotate() function in OpenCV to rotate the i-th cardiac ultrasound image E after cropping i 'Flip along the horizontal or vertical axis with a probability of 0.5, and the flip angle is 90 degrees to obtain the i-th cardiac ultrasound image E after flipping i ″.

[0077] b-3) Flip the i-th cardiac ultrasound image E i ″Perform z-score standardization to obtain the preprocessed cardiac ultrasound image S i .

[0078] In one embodiment of the present invention, step e) comprises the following steps:

[0079] e-1) The encoder of the cardiac ultrasound image segmentation network model is composed of a convolutional layer, a first dual perception state space enhancement module DPSSEM, a first downsampling unit, a second dual perception state space enhancement module DPSSEM, a second downsampling unit, a third dual perception state space enhancement module DPSSEM, a third downsampling unit, a fourth dual perception state space enhancement module DPSSEM, and a fourth downsampling unit.

[0080] e-2) The i-th preprocessed cardiac ultrasound image S in the training set i Input into the convolutional layer of the encoder and output the feature map

[0081] e-3) The first dual perception state space enhancement module DPSSEM of the encoder is composed of a depthwise separable convolutional layer DSConv, a first LayerNorm layer, a first multi-layer perceptron MLP, a first branch, a second branch, a third branch, a second LayerNorm layer, and a second multi-layer perceptron MLP. The feature map The first dual perception state space enhancement module DPSSEM is sequentially input into the deep separable convolution layer DSConv, the first LayerNorm layer, and the first multi-layer perceptron MLP to output the feature map. The feature map Use the torch.chunk() function in pytorch to perform the Split operation and output the feature maps separately Feature Map Feature Map The first branch of the first dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a spatial attention mechanism SAM, a SiLU activation function, and a state space model SSM. Input into the first branch and output the feature map The second branch of the first dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a hole convolutional layer, and a SiLU activation function. Input into the second branch and output the feature map The third branch of the first dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a channel attention mechanism CAM, a SiLU activation function, and a state space model SSM. Input into the third branch and output the feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map It is sequentially input into the second LayerNorm layer of the first dual perception state space enhancement module DPSSEM and the second multi-layer perceptron MLP, and the output is the feature map

[0082] e-4) The first downsampling unit of the encoder is composed of a convolutional layer and a maximum pooling layer in sequence. Input into the first downsampling unit and output the feature map

[0083] e-5) The second dual perception state space enhancement module DPSSEM of the encoder consists of a depthwise separable convolutional layer DSConv, a first LayerNorm layer, a first multi-layer perceptron MLP, a first branch, a second branch, a third branch, a second LayerNorm layer, and a second multi-layer perceptron MLP. The output is the feature map. The feature map Use the torch.chunk() function in pytorch to perform the Split operation and output the feature maps separately Feature Map Feature Map The first branch of the second dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a spatial attention mechanism SAM, a SiLU activation function, and a state space model SSM. Input into the first branch and output the feature map The second branch of the second dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a hole convolutional layer, and a SiLU activation function. Input into the second branch and output the feature map The third branch of the second dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a channel attention mechanism CAM, a SiLU activation function, and a state space model SSM. Input into the third branch and output the feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map It is sequentially input into the second LayerNorm layer of the second dual perception state space enhancement module DPSSEM and the second multi-layer perceptron MLP, and the output is the feature map

[0084] e-6) The second downsampling unit of the encoder is composed of a convolutional layer and a maximum pooling layer in sequence. Input into the second downsampling unit and output the feature map

[0085] e-7) The third dual perception state space enhancement module DPSSEM of the encoder is composed of a depth separable convolution layer DSConv, a first LayerNorm layer, a first multi-layer perceptron MLP, a first branch, a second branch, a third branch, a second LayerNorm layer, and a second multi-layer perceptron MLP. The feature map The output is the deep separable convolutional layer DSConv, the first LayerNorm layer, and the first multi-layer perceptron MLP of the third dual perception state space enhancement module DPSSEM, and the feature map is obtained. The feature map Use the torch.chunk() function in pytorch to perform the Split operation and output the feature maps separately Feature Map Feature Map The first branch of the third dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a spatial attention mechanism SAM, a SiLU activation function, and a state space model SSM. Input into the first branch and output the feature map The second branch of the third dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a hole convolutional layer, and a SiLU activation function. Input into the second branch and output the feature map The third branch of the third dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a channel attention mechanism CAM, a SiLU activation function, and a state space model SSM. Input into the third branch and output the feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map It is sequentially input into the second LayerNorm layer of the third dual perception state space enhancement module DPSSEM and the second multi-layer perceptron MLP, and the output is the feature map

[0086] e-8) The third downsampling unit of the encoder is composed of a convolutional layer and a maximum pooling layer in sequence, Input into the third downsampling unit and output the feature map

[0087] e-9) The fourth dual perception state space enhancement module DPSSEM of the encoder is composed of a depth separable convolution layer DSConv, a first LayerNorm layer, a first multi-layer perceptron MLP, a first branch, a second branch, a third branch, a second LayerNorm layer, and a second multi-layer perceptron MLP. The feature map The output is the deep separable convolutional layer DSConv, the first LayerNorm layer, and the first multi-layer perceptron MLP of the fourth dual perception state space enhancement module DPSSEM, and the feature map is obtained. The feature map Use the torch.chunk() function in pytorch to perform the Split operation and output the feature maps separately Feature Map Feature Map The first branch of the fourth dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a spatial attention mechanism SAM, a SiLU activation function, and a state space model SSM. Input into the first branch and output the feature map The second branch of the fourth dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a hole convolutional layer, and a SiLU activation function. Input into the second branch and output the feature map The third branch of the fourth dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a channel attention mechanism CAM, a SiLU activation function, and a state space model SSM. Input into the third branch and output the feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map It is sequentially input into the second LayerNorm layer of the fourth dual perception state space enhancement module DPSSEM and the second multi-layer perceptron MLP, and the feature map is output.

[0088] e-10) The fourth downsampling unit of the encoder is composed of a convolutional layer and a maximum pooling layer in sequence, Input to the fourth downsampling unit and output the feature map In this embodiment, the convolution kernel size of the convolution layer of the encoder in step e-2) is 3×3; the convolution kernel size of the depthwise separable convolution layer DSConv of the first dual perception state space enhancement module DPSSEM, the second dual perception state space enhancement module DPSSEM, the third dual perception state space enhancement module DPSSEM, and the fourth dual perception state space enhancement module DPSSEM are all 3×3, the padding is 1, and the step size is 1; the convolution kernel size of the convolution layer of the first branch is 5×5, the padding is 2, and the step size is 1. , the convolution kernel size of the convolution layer of the second branch is 3×3, the padding is 1, the step size is 1, the convolution kernel size of the hole convolution layer of the second branch is 3×3, the padding is 1, the step size is 1, and the expansion rate is 2, the convolution kernel size of the convolution layer of the third branch is 7×7, the padding is 3, and the step size is 1; the convolution kernel size of the convolution layer of the first downsampling unit, the second downsampling unit, the third downsampling unit, and the fourth downsampling unit is 3×3, the padding is 1, and the step size is 1, and the pooling kernel of the maximum pooling layer is 2×2, and the step size is 2. In one embodiment of the present invention, step f) includes the following steps:

[0089] f-1) The decoder of the cardiac ultrasound image segmentation network model is composed of a convolutional layer, a first feature interaction perception decoding module FIPDM, a first upsampling unit, a second feature interaction perception decoding module FIPDM, a second upsampling unit, a third feature interaction perception decoding module FIPDM, a third upsampling unit, a fourth feature interaction perception decoding module FIPDM, and a fourth upsampling unit. The jump connection unit of the cardiac ultrasound image segmentation network model is composed of a first focused feature enhancement transfer module FFETM, a second focused feature enhancement transfer module FFETM, a third focused feature enhancement transfer module FFETM, and a fourth focused feature enhancement transfer module FFETM.

[0090] f-2) The fourth focused feature enhancement transfer module FFETM of the jump connection unit is composed of the first convolution layer, the first branch, the second branch, the maximum pooling layer, the second convolution layer, and the Sigmoid activation function. Input to the first convolutional layer of the fourth focused feature enhancement transfer module FFETM, and output the feature map The feature map Use the torch.chunk() function in pytorch to perform the Split operation and output the feature maps separately Feature Map The first branch of the fourth focused feature enhancement transfer module FFETM is composed of a spatial attention mechanism and a channel attention mechanism, which converts the feature map Input into the first branch and output the feature map The second branch of the fourth focused feature enhancement transfer module FFETM is composed of a convolutional layer, a ReLU activation function, and a BatchNorm layer. Input into the second branch and output the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map The output is sequentially input into the maximum pooling layer, the second convolutional layer, and the Sigmoid activation function of the fourth focused feature enhancement transfer module FFETM, and the feature map is obtained.

[0091] f-3) The feature map With feature map Perform addition operation to obtain feature map

[0092] f-4) The fourth feature interaction perception decoding module FIPDM of the decoder consists of a convolutional layer, a LayerNorm layer, a GeLU activation function, a first branch, a second branch, a third branch, and an MLP layer. It is sequentially input into the convolution layer, LayerNorm layer, and GeLU activation function of the fourth feature interaction perception decoding module FIPDM, and the feature map is output. The feature map Use the torch.chunk() function in pytorch to perform the Split operation and output the feature maps separately Feature Map Feature Map The first branch of the fourth feature interaction perception decoding module FIPDM is composed of a global maximum pooling layer, a convolutional layer, a GeLU activation function, a depth-separable convolutional layer, and a BatchNorm layer. Input into the first branch and output the feature map The second branch of the fourth feature interaction perception decoding module FIPDM is composed of a depth-separable convolutional layer, a BatchNorm layer, and a GeLU activation function. Input into the second branch and output the feature map The third branch of the fourth feature interaction perception decoding module FIPDM is composed of a global average pooling layer and a convolutional layer in sequence. Input into the third branch and output the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map feature map Feature Map Feature Map The Concat operation is performed through the torch.cat() function in pytorch to obtain the feature map The feature map Input to the MLP layer of the fourth feature interaction perception decoding module FIPDM, and output the feature map The feature map With feature map Perform addition operation to obtain feature map

[0093] f-5) The fourth upsampling unit of the decoder is composed of a transposed convolutional layer and a BatchNorm layer in sequence. Input to the fourth upsampling unit and output the feature map

[0094] f-6) The third focused feature enhancement transfer module FFETM of the jump connection unit is composed of the first convolution layer, the first branch, the second branch, the maximum pooling layer, the second convolution layer, and the Sigmoid activation function. Input to the first convolutional layer of the third focused feature enhancement transfer module FFETM, and output the feature map The feature map Use the torch.chunk() function in pytorch to perform the Split operation and output the feature maps separately Feature Map The first branch of the third focused feature enhancement transfer module FFETM is composed of the spatial attention mechanism and the channel attention mechanism in sequence. Input into the first branch and output the feature map The second branch of the third focused feature enhancement transfer module FFETM is composed of a convolutional layer, a ReLU activation function, and a BatchNorm layer. Input into the second branch and output the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map It is sequentially input into the maximum pooling layer, the second convolutional layer, and the Sigmoid activation function of the third focused feature enhancement transfer module FFETM, and the feature map is output.

[0095] f-7) The feature map With feature map Perform addition operation to obtain feature map

[0096] f-8) The third feature interaction perception decoding module FIPDM of the decoder consists of a convolutional layer, a LayerNorm layer, a GeLU activation function, a first branch, a second branch, a third branch, and an MLP layer. It is sequentially input into the convolution layer, LayerNorm layer, and GeLU activation function of the third feature interaction perception decoding module FIPDM, and the feature map is output. The feature map Use the torch.chunk() function in pytorch to perform the Split operation and output the feature maps separately Feature Map Feature Map The first branch of the third feature interaction perception decoding module FIPDM is composed of a global maximum pooling layer, a convolutional layer, a GeLU activation function, a depth-separable convolutional layer, and a BatchNorm layer. Input into the first branch and output the feature map The second branch of the third feature interaction perception decoding module FIPDM is composed of a depth-separable convolutional layer, a BatchNorm layer, and a GeLU activation function. Input into the second branch and output the feature map The third branch of the third feature interaction perception decoding module FIPDM is composed of a global average pooling layer and a convolutional layer in sequence. Input into the third branch and output the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map feature map Feature Map Feature Map The Concat operation is performed through the torch.cat() function in pytorch to obtain the feature map The feature map Input into the MLP layer of the third feature interaction perception decoding module FIPDM, and output the feature map The feature map With feature map Perform addition operation to obtain feature map

[0097] f-9) The third upsampling unit of the decoder is composed of a transposed convolutional layer and a BatchNorm layer in sequence. Input to the third upsampling unit and output the feature map

[0098] f-10) The second focused feature enhancement transfer module FFETM of the jump connection unit is composed of the first convolution layer, the first branch, the second branch, the maximum pooling layer, the second convolution layer, and the Sigmoid activation function. Input to the first convolutional layer of the second focused feature enhancement transfer module FFETM, and output the feature map The feature map Use the torch.chunk() function in pytorch to perform the Split operation and output the feature maps separately Feature Map The first branch of the second focused feature enhancement transfer module FFETM is composed of the spatial attention mechanism and the channel attention mechanism in sequence. Input into the first branch and output the feature map The second branch of the second focused feature enhancement transfer module FFETM is composed of a convolutional layer, a ReLU activation function, and a BatchNorm layer. Input into the second branch and output the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map It is sequentially input into the maximum pooling layer, the second convolutional layer, and the Sigmoid activation function of the second focused feature enhancement transfer module FFETM, and the feature map is output.

[0099] f-11) The feature map With feature map Perform addition operation to obtain feature map

[0100] f-12) The second feature interaction perception decoding module FIPDM of the decoder consists of a convolutional layer, a LayerNorm layer, a GeLU activation function, a first branch, a second branch, a third branch, and an MLP layer. It is sequentially input into the convolutional layer, LayerNorm layer, and GeLU activation function of the second feature interaction perception decoding module FIPDM, and the feature map is output. The feature map Use the torch.chunk() function in pytorch to perform the Split operation and output the feature maps separately Feature Map Feature Map The first branch of the second feature interaction perception decoding module FIPDM is composed of a global maximum pooling layer, a convolutional layer, a GeLU activation function, a depth-separable convolutional layer, and a BatchNorm layer. Input into the first branch and output the feature map The second branch of the second feature interaction perception decoding module FIPDM is composed of a depth-separable convolutional layer, a BatchNorm layer, and a GeLU activation function. Input into the second branch and output the feature map The third branch of the second feature interaction perception decoding module FIPDM is composed of a global average pooling layer and a convolutional layer. Input into the third branch and output the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map feature map Feature Map Feature Map The Concat operation is performed through the torch.cat() function in pytorch to obtain the feature map The feature map Input into the MLP layer of the second feature interaction perception decoding module FIPDM, and output the feature map The feature map With feature map Perform addition operation to obtain feature map

[0101] f-13) The second upsampling unit of the decoder is composed of a transposed convolutional layer and a BatchNorm layer in sequence. Input to the second upsampling unit and output the feature map

[0102] f-14) The first focused feature enhancement transfer module FFETM of the jump connection unit is composed of the first convolution layer, the first branch, the second branch, the maximum pooling layer, the second convolution layer, and the Sigmoid activation function. Input into the first convolutional layer of the first focused feature enhancement transfer module FFETM, and output the feature map The feature map Use the torch.chunk() function in pytorch to perform the Split operation and output the feature maps separately Feature Map The first branch of the first focused feature enhancement transfer module FFETM is composed of the spatial attention mechanism and the channel attention mechanism in sequence. Input into the first branch and output the feature map The second branch of the first focused feature enhancement transfer module FFETM is composed of a convolutional layer, a ReLU activation function, and a BatchNorm layer. Input into the second branch and output the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map It is sequentially input into the maximum pooling layer, the second convolutional layer, and the Sigmoid activation function of the first focused feature enhancement transfer module FFETM, and the feature map is output.

[0103] f-15) The feature map With feature map Perform addition operation to obtain feature map

[0104] f-16) The first feature interaction perception decoding module FIPDM of the decoder consists of a convolutional layer, a LayerNorm layer, a GeLU activation function, a first branch, a second branch, a third branch, and an MLP layer. It is sequentially input into the convolution layer, LayerNorm layer, and GeLU activation function of the first feature interaction perception decoding module FIPDM, and the feature map is output. The feature map Use the torch.chunk() function in pytorch to perform the Split operation and output the feature maps separately Feature Map Feature Map The first branch of the first feature interaction perception decoding module FIPDM is composed of a global maximum pooling layer, a convolutional layer, a GeLU activation function, a depth-separable convolutional layer, and a BatchNorm layer. Input into the first branch and output the feature map The second branch of the first feature interaction perception decoding module FIPDM is composed of a depth-separable convolutional layer, a BatchNorm layer, and a GeLU activation function. Input into the second branch and output the feature map The third branch of the first feature interaction perception decoding module FIPDM is composed of a global average pooling layer and a convolutional layer in sequence. Input into the third branch and output the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map feature map Feature Map Feature Map The Concat operation is performed through the torch.cat() function in pytorch to obtain the feature map The feature map Input into the MLP layer of the first feature interaction perception decoding module FIPDM, and output the feature map The feature map With feature map Perform addition operation to obtain feature map

[0105] f-17) The first upsampling unit of the decoder is composed of a transposed convolutional layer and a BatchNorm layer in sequence. Input into the first upsampling unit and output the feature map

[0106] f-18) The feature map Input into the convolutional layer of the decoder, and output the predicted segmentation image In this embodiment, preferably, the convolution kernel size of the first convolution layer of the first focused feature enhancement transfer module FFETM, the second focused feature enhancement transfer module FFETM, the third focused feature enhancement transfer module FFETM, and the fourth focused feature enhancement transfer module FFETM of the decoder are all 3×3, the padding is 1, and the step size is 1; the convolution kernel size of the second convolution layer is 3×3, the padding is 1, and the step size is 1; the pooling kernel of the maximum pooling layer is 2×2, the step size is 1, and the convolution kernel size of the convolution layer of the second branch is 1×1, and the step size is 1; the first feature interaction perception decoding module FIPDM, the second feature interaction perception decoding module FIPDM, the third feature interaction perception decoding module FIPDM, and the fourth feature interaction perception decoding module FIPDM of the decoder The convolution kernel size of the convolution layer of the feature interaction perception decoding module FIPDM is 3×3, and the step size is 1. The convolution kernel size of the convolution kernel of the first branch is 3×3, and the step size is 1. The convolution kernel size of the depthwise separable convolution layer of the first branch is 3×3, and the step size is 1, and the padding is 1. The convolution kernel size of the depthwise separable convolution layer of the second branch is 3×3, and the step size is 1, and the padding is 1. The convolution kernel size of the convolution layer of the third branch is 1×1, and the step size is 1; the convolution kernel size of the transposed convolution layer of the first upsampling unit, the second upsampling unit, the third upsampling unit, and the fourth upsampling unit of the decoder is 3×3, the padding is 1, and the step size is 2; the convolution kernel size of the convolution layer of the decoder in step f-18) is 1×1.

[0107] In one embodiment of the present invention, step g) comprises the following steps:

[0108] g-1) through the formula L total =αL Dice +βL CE +γL boundary +δL Tversky Calculate the total loss L total , where L Dice is the Dice loss, L CE is the cross entropy loss, L boundary is the boundary loss, L Tversky is Tversky loss, α, β, γ, and δ are all hyperparameters.

[0109] g-2) Use the SGD optimizer to take advantage of the total loss L total Train the cardiac ultrasound image segmentation network model to obtain an optimized cardiac ultrasound image segmentation network model.

[0110] In this embodiment, preferably, when training the cardiac ultrasound image segmentation network model, the learning rate is set to 0.001, the batch size is set to 16, the number of iterations is set to 1200, the hyperparameter α is set to 0.3, the hyperparameter β is set to 0.3, the hyperparameter γ is set to 0.2, and the hyperparameter δ is set to 0.2.

[0111] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A cardiac ultrasound image segmentation method based on state space model and feature interaction perception, characterized in that: include: a) Obtain N cardiac ultrasound images, and obtain a cardiac ultrasound image dataset E, where E = {E1, E2, ..., E i ,...,E N }, E i is the i-th cardiac ultrasound image, i∈{1,...,N}; b) Take the i-th cardiac ultrasound image E i Preprocessing is performed to obtain the preprocessed cardiac ultrasound image S i All N preprocessed cardiac ultrasound images constitute a preprocessed cardiac ultrasound image dataset S, S = {S1, S2, ..., S i ,...,S N }; c) dividing the preprocessed cardiac ultrasound image dataset S into a training set, a validation set, and a test set; d) establishing a cardiac ultrasound image segmentation network model consisting of an encoder, a skip connection unit, and a decoder; e) The i-th preprocessed cardiac ultrasound image S in the training set i Input into the encoder of the cardiac ultrasound image segmentation network model, and output the feature map f) The feature map Input into the decoder of the cardiac ultrasound image segmentation network model, extract features through the jump connection unit, and obtain the predicted segmentation image g) Calculate the total loss L total , through the total loss L total Training a cardiac ultrasound image segmentation network model to obtain an optimized cardiac ultrasound image segmentation network model; h) Take the i-th preprocessed cardiac ultrasound image S in the test set i Input into the optimized cardiac ultrasound image segmentation network model, and output the predicted segmentation image 2. The method for segmenting cardiac ultrasound images based on state space model and feature interactive perception according to claim 1, characterized in that: In step a), N cardiac ultrasound images are obtained from the CAMUS cardiac segmentation ultrasound image dataset, where N is 500.

3. The cardiac ultrasound image segmentation method based on state space model and feature interactive perception according to claim 1, characterized in that: The preprocessed cardiac ultrasound image dataset S in step c) is divided into a training set, a validation set, and a test set in a ratio of 7:2:

1.

4. The cardiac ultrasound image segmentation method based on state space model and feature interactive perception according to claim 1, characterized in that: Step b) comprises the following steps: b-1) Use the center_crop() function in OpenCV to crop the i-th cardiac ultrasound image E i Perform a center cropping operation with a size of 256×256 to obtain the i-th cardiac ultrasound image E after cropping i ′; b-2) Use the random_flip_rotate() function in OpenCV to rotate the cropped i-th cardiac ultrasound image E i 'Flip along the horizontal or vertical axis with a probability of 0.5, and the flip angle is 90 degrees to obtain the i-th cardiac ultrasound image E after flipping i ″; b-3) Flip the i-th cardiac ultrasound image E i ″Perform z-score standardization operation to obtain the preprocessed cardiac ultrasound image S i .

5. The cardiac ultrasound image segmentation method based on state space model and feature interactive perception according to claim 1, characterized in that: Step e) comprises the following steps: e-1) The encoder of the cardiac ultrasound image segmentation network model is composed of a convolutional layer, a first dual perception state space enhancement module DPSSEM, a first downsampling unit, a second dual perception state space enhancement module DPSSEM, a second downsampling unit, a third dual perception state space enhancement module DPSSEM, a third downsampling unit, a fourth dual perception state space enhancement module DPSSEM, and a fourth downsampling unit; e-2) The i-th preprocessed cardiac ultrasound image S in the training set i Input into the convolutional layer of the encoder and output the feature map e-3) The first dual perception state space enhancement module DPSSEM of the encoder is composed of a depthwise separable convolutional layer DSConv, a first LayerNorm layer, a first multi-layer perceptron MLP, a first branch, a second branch, a third branch, a second LayerNorm layer, and a second multi-layer perceptron MLP. The feature map The first dual perception state space enhancement module DPSSEM is sequentially input into the deep separable convolution layer DSConv, the first LayerNorm layer, and the first multi-layer perceptron MLP to output the feature map. The feature map The torch.chunk() function in pytorch is used to perform the Split operation and output the feature maps separately. Feature Map Feature Map The first branch of the first dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a spatial attention mechanism SAM, a SiLU activation function, and a state space model SSM. Input into the first branch and output the feature map The second branch of the first dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a hole convolutional layer, and a SiLU activation function. Input into the second branch and output the feature map The third branch of the first dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a channel attention mechanism CAM, a SiLU activation function, and a state space model SSM. Input into the third branch and output the feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map It is sequentially input into the second LayerNorm layer of the first dual perception state space enhancement module DPSSEM and the second multi-layer perceptron MLP, and the output is the feature map e-4) The first downsampling unit of the encoder is composed of a convolutional layer and a maximum pooling layer in sequence. Input into the first downsampling unit and output the feature map e-5) The second dual perception state space enhancement module DPSSEM of the encoder consists of a depthwise separable convolutional layer DSConv, a first LayerNorm layer, a first multi-layer perceptron MLP, a first branch, a second branch, a third branch, a second LayerNorm layer, and a second multi-layer perceptron MLP. The output is the feature map. The feature map The torch.chunk() function in pytorch is used to perform the Split operation and output the feature maps separately. Feature Map Feature Map The first branch of the second dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a spatial attention mechanism SAM, a SiLU activation function, and a state space model SSM. Input into the first branch and output the feature map The second branch of the second dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a hole convolutional layer, and a SiLU activation function. Input into the second branch and output the feature map The third branch of the second dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a channel attention mechanism CAM, a SiLU activation function, and a state space model SSM. Input into the third branch and output the feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map It is sequentially input into the second LayerNorm layer of the second dual perception state space enhancement module DPSSEM and the second multi-layer perceptron MLP, and the output is the feature map e-6) The second downsampling unit of the encoder is composed of a convolutional layer and a maximum pooling layer in sequence. Input into the second downsampling unit and output the feature map e-7) The third dual perception state space enhancement module DPSSEM of the encoder is composed of a depth separable convolution layer DSConv, a first LayerNorm layer, a first multi-layer perceptron MLP, a first branch, a second branch, a third branch, a second LayerNorm layer, and a second multi-layer perceptron MLP. The feature map The output is the feature map. The feature map The torch.chunk() function in pytorch is used to perform the Split operation and output the feature maps separately. Feature Map Feature Map The first branch of the third dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a spatial attention mechanism SAM, a SiLU activation function, and a state space model SSM. Input into the first branch and output the feature map The second branch of the third dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a hole convolutional layer, and a SiLU activation function. Input into the second branch and output the feature map The third branch of the third dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a channel attention mechanism CAM, a SiLU activation function, and a state space model SSM. Input into the third branch and output the feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map It is sequentially input into the second LayerNorm layer of the third dual perception state space enhancement module DPSSEM and the second multi-layer perceptron MLP, and the output is the feature map e-8) The third downsampling unit of the encoder is composed of a convolutional layer and a maximum pooling layer in sequence, Input into the third downsampling unit and output the feature map e-9) The fourth dual perception state space enhancement module DPSSEM of the encoder is composed of a depth separable convolution layer DSConv, a first LayerNorm layer, a first multi-layer perceptron MLP, a first branch, a second branch, a third branch, a second LayerNorm layer, and a second multi-layer perceptron MLP. The feature map The output is the deep separable convolutional layer DSConv, the first LayerNorm layer, and the first multi-layer perceptron MLP of the fourth dual perception state space enhancement module DPSSEM, and the feature map is obtained. The feature map The torch.chunk() function in pytorch is used to perform the Split operation and output the feature maps separately. Feature Map Feature Map The first branch of the fourth dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a spatial attention mechanism SAM, a SiLU activation function, and a state space model SSM. Input into the first branch and output the feature map The second branch of the fourth dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a hole convolutional layer, and a SiLU activation function. Input into the second branch and output the feature map The third branch of the fourth dual perception state space enhancement module DPSSEM is composed of a linear layer, a convolutional layer, a channel attention mechanism CAM, a SiLU activation function, and a state space model SSM. Input into the third branch and output the feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map It is sequentially input into the second LayerNorm layer of the fourth dual perception state space enhancement module DPSSEM and the second multi-layer perceptron MLP, and the feature map is output. e-10) The fourth downsampling unit of the encoder is composed of a convolutional layer and a maximum pooling layer in sequence, Input to the fourth downsampling unit and output the feature map 6. The cardiac ultrasound image segmentation method based on state space model and feature interactive perception according to claim 5, characterized in that: In step e-2), the convolution kernel size of the convolution layer of the encoder is 3×3; the convolution kernel size of the depthwise separable convolution layer DSConv of the first dual perception state space enhancement module DPSSEM, the second dual perception state space enhancement module DPSSEM, the third dual perception state space enhancement module DPSSEM, and the fourth dual perception state space enhancement module DPSSEM are all 3×3, the padding is 1, and the step size is 1. The convolution kernel size of the convolution layer of the first branch is 5×5, the padding is 2, and the step size is 1. The convolution kernel size of the convolution layer of the first branch is 3×3, the padding is 1, and the stride is 1. The convolution kernel size of the hole convolution layer of the second branch is 3×3, the padding is 1, the stride is 1, and the expansion rate is 2. The convolution kernel size of the convolution layer of the third branch is 7×7, the padding is 3, and the stride is 1; the convolution kernel size of the convolution layer of the first downsampling unit, the second downsampling unit, the third downsampling unit, and the fourth downsampling unit is 3×3, the padding is 1, and the stride is 1. The pooling kernel of the maximum pooling layer is 2×2, and the stride is 2.

7. The cardiac ultrasound image segmentation method based on state space model and feature interactive perception according to claim 1, characterized in that: Step f) comprises the following steps: f-1) The decoder of the cardiac ultrasound image segmentation network model is composed of a convolutional layer, a first feature interactive perception decoding module FIPDM, a first upsampling unit, a second feature interactive perception decoding module FIPDM, a second upsampling unit, a third feature interactive perception decoding module FIPDM, a third upsampling unit, a fourth feature interactive perception decoding module FIPDM, and a fourth upsampling unit. The jump connection unit of the cardiac ultrasound image segmentation network model is composed of a first focused feature enhancement transfer module FFETM, a second focused feature enhancement transfer module FFETM, a third focused feature enhancement transfer module FFETM, and a fourth focused feature enhancement transfer module FFETM; f-2) The fourth focused feature enhancement transfer module FFETM of the jump connection unit is composed of a first convolutional layer, a first branch, a second branch, a maximum pooling layer, a second convolutional layer, and a Sigmoid activation function. The feature map Input to the first convolutional layer of the fourth focused feature enhancement transfer module FFETM, and output the feature map The feature map The torch.chunk() function in pytorch is used to perform the Split operation and output the feature maps separately. Feature Map The first branch of the fourth focused feature enhancement transfer module FFETM is composed of a spatial attention mechanism and a channel attention mechanism, which converts the feature map Input into the first branch and output the feature map The second branch of the fourth focused feature enhancement transfer module FFETM is composed of a convolutional layer, a ReLU activation function, and a BatchNorm layer. Input into the second branch and output the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map The output is sequentially input into the maximum pooling layer, the second convolutional layer, and the Sigmoid activation function of the fourth focused feature enhancement transfer module FFETM, and the feature map is obtained. f-3) The feature map With feature map Perform addition operation to obtain feature map f-4) The fourth feature interaction perception decoding module FIPDM of the decoder consists of a convolutional layer, a LayerNorm layer, a GeLU activation function, a first branch, a second branch, a third branch, and an MLP layer. It is sequentially input into the convolution layer, LayerNorm layer, and GeLU activation function of the fourth feature interaction perception decoding module FIPDM, and the feature map is output. The feature map The torch.chunk() function in pytorch is used to perform the Split operation and output the feature maps separately. Feature Map Feature Map The first branch of the fourth feature interaction perception decoding module FIPDM is composed of a global maximum pooling layer, a convolutional layer, a GeLU activation function, a depth-separable convolutional layer, and a BatchNorm layer. Input into the first branch and output the feature map The second branch of the fourth feature interaction perception decoding module FIPDM is composed of a depth-separable convolutional layer, a BatchNorm layer, and a GeLU activation function. Input into the second branch and output the feature map The third branch of the fourth feature interaction perception decoding module FIPDM is composed of a global average pooling layer and a convolutional layer in sequence. Input into the third branch and output the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map feature map Feature Map Feature Map The Concat operation is performed through the torch.cat() function in pytorch to obtain the feature map The feature map Input to the MLP layer of the fourth feature interaction perception decoding module FIPDM, and output the feature map The feature map With feature map Perform addition operation to obtain feature map f-5) The fourth upsampling unit of the decoder is composed of a transposed convolutional layer and a BatchNorm layer in sequence. Input to the fourth upsampling unit and output the feature map f-6) The third focused feature enhancement transfer module FFETM of the jump connection unit is composed of the first convolution layer, the first branch, the second branch, the maximum pooling layer, the second convolution layer, and the Sigmoid activation function. Input to the first convolutional layer of the third focused feature enhancement transfer module FFETM, and output the feature map The feature map The torch.chunk() function in pytorch is used to perform the Split operation and output the feature maps separately. Feature Map The first branch of the third focused feature enhancement transfer module FFETM is composed of the spatial attention mechanism and the channel attention mechanism in sequence. Input into the first branch and output the feature map The second branch of the third focused feature enhancement transfer module FFETM is composed of a convolutional layer, a ReLU activation function, and a BatchNorm layer. Input into the second branch and output the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map It is sequentially input into the maximum pooling layer, the second convolutional layer, and the Sigmoid activation function of the third focused feature enhancement transfer module FFETM, and the feature map is output. f-7) The feature map With feature map Perform addition operation to obtain feature map f-8) The third feature interaction perception decoding module FIPDM of the decoder consists of a convolutional layer, a LayerNorm layer, a GeLU activation function, a first branch, a second branch, a third branch, and an MLP layer. It is sequentially input into the convolution layer, LayerNorm layer, and GeLU activation function of the third feature interaction perception decoding module FIPDM, and the feature map is output. The feature map The torch.chunk() function in pytorch is used to perform the Split operation and output the feature maps separately. Feature Map Feature Map The first branch of the third feature interaction perception decoding module FIPDM is composed of a global maximum pooling layer, a convolutional layer, a GeLU activation function, a depth-separable convolutional layer, and a BatchNorm layer. Input into the first branch and output the feature map The second branch of the third feature interaction perception decoding module FIPDM is composed of a depth-separable convolutional layer, a BatchNorm layer, and a GeLU activation function. Input into the second branch and output the feature map The third branch of the third feature interaction perception decoding module FIPDM is composed of a global average pooling layer and a convolutional layer in sequence. Input into the third branch and output the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map feature map Feature Map Feature Map The Concat operation is performed through the torch.cat() function in pytorch to obtain the feature map The feature map Input into the MLP layer of the third feature interaction perception decoding module FIPDM, and output the feature map The feature map With feature map Perform addition operation to obtain feature map f-9) The third upsampling unit of the decoder is composed of a transposed convolutional layer and a BatchNorm layer in sequence. Input to the third upsampling unit and output the feature map f-10) The second focused feature enhancement transfer module FFETM of the jump connection unit is composed of the first convolution layer, the first branch, the second branch, the maximum pooling layer, the second convolution layer, and the Sigmoid activation function. Input to the first convolutional layer of the second focused feature enhancement transfer module FFETM, and output the feature map The feature map The torch.chunk() function in pytorch is used to perform the Split operation and output the feature maps separately. Feature Map The first branch of the second focused feature enhancement transfer module FFETM is composed of the spatial attention mechanism and the channel attention mechanism in sequence. Input into the first branch and output the feature map The second branch of the second focused feature enhancement transfer module FFETM is composed of a convolutional layer, a ReLU activation function, and a BatchNorm layer. Input into the second branch and output the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map It is sequentially input into the maximum pooling layer, the second convolutional layer, and the Sigmoid activation function of the second focused feature enhancement transfer module FFETM, and the feature map is output. f-11) The feature map With feature map Perform addition operation to obtain feature map f-12) The second feature interaction perception decoding module FIPDM of the decoder consists of a convolutional layer, a LayerNorm layer, a GeLU activation function, a first branch, a second branch, a third branch, and an MLP layer. It is sequentially input into the convolutional layer, LayerNorm layer, and GeLU activation function of the second feature interaction perception decoding module FIPDM, and the feature map is output. The feature map The torch.chunk() function in pytorch is used to perform the Split operation and output the feature maps separately. Feature Map Feature Map The first branch of the second feature interaction perception decoding module FIPDM is composed of a global maximum pooling layer, a convolutional layer, a GeLU activation function, a depth-separable convolutional layer, and a BatchNorm layer. Input into the first branch and output the feature map The second branch of the second feature interaction perception decoding module FIPDM is composed of a depth-separable convolutional layer, a BatchNorm layer, and a GeLU activation function. Input into the second branch and output the feature map The third branch of the second feature interaction perception decoding module FIPDM is composed of a global average pooling layer and a convolutional layer. Input into the third branch and output the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map feature map Feature Map Feature Map The Concat operation is performed through the torch.cat() function in pytorch to obtain the feature map The feature map Input into the MLP layer of the second feature interaction perception decoding module FIPDM, and output the feature map The feature map With feature map Perform addition operation to obtain feature map f-13) The second upsampling unit of the decoder is composed of a transposed convolutional layer and a BatchNorm layer in sequence. Input to the second upsampling unit and output the feature map f-14) The first focused feature enhancement transfer module FFETM of the jump connection unit is composed of the first convolution layer, the first branch, the second branch, the maximum pooling layer, the second convolution layer, and the Sigmoid activation function. Input into the first convolutional layer of the first focused feature enhancement transfer module FFETM, and output the feature map The feature map The torch.chunk() function in pytorch is used to perform the Split operation and output the feature maps separately. Feature Map The first branch of the first focused feature enhancement transfer module FFETM is composed of the spatial attention mechanism and the channel attention mechanism in sequence. Input into the first branch and output the feature map The second branch of the first focused feature enhancement transfer module FFETM is composed of a convolutional layer, a ReLU activation function, and a BatchNorm layer. Input into the second branch and output the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform multiplication operation to obtain the feature map The feature map It is sequentially input into the maximum pooling layer, the second convolutional layer, and the Sigmoid activation function of the first focused feature enhancement transfer module FFETM, and the feature map is output. f-15) The feature map With feature map Perform addition operation to obtain feature map f-16) The first feature interaction perception decoding module FIPDM of the decoder consists of a convolutional layer, a LayerNorm layer, a GeLU activation function, a first branch, a second branch, a third branch, and an MLP layer. It is sequentially input into the convolution layer, LayerNorm layer, and GeLU activation function of the first feature interaction perception decoding module FIPDM, and the feature map is output. The feature map The torch.chunk() function in pytorch is used to perform the Split operation and output the feature maps separately. Feature Map Feature Map The first branch of the first feature interaction perception decoding module FIPDM is composed of a global maximum pooling layer, a convolutional layer, a GeLU activation function, a depth-separable convolutional layer, and a BatchNorm layer. Input into the first branch and output the feature map The second branch of the first feature interaction perception decoding module FIPDM is composed of a depth-separable convolutional layer, a BatchNorm layer, and a GeLU activation function. Input into the second branch and output the feature map The third branch of the first feature interaction perception decoding module FIPDM is composed of a global average pooling layer and a convolutional layer in sequence. Input into the third branch and output the feature map The feature map With feature map Perform addition operation to obtain feature map The feature map With feature map Perform addition operation to obtain feature map The feature map feature map Feature Map Feature Map The Concat operation is performed through the torch.cat() function in pytorch to obtain the feature map The feature map Input into the MLP layer of the first feature interaction perception decoding module FIPDM, and output the feature map The feature map With feature map Perform addition operation to obtain feature map f-17) The first upsampling unit of the decoder is composed of a transposed convolutional layer and a BatchNorm layer in sequence. Input into the first upsampling unit and output the feature map f-18) The feature map Input into the convolutional layer of the decoder, and output the predicted segmentation image 8. The method for segmenting cardiac ultrasound images based on state space model and feature interactive perception according to claim 7, characterized in that: The convolution kernel size of the first convolution layer of the decoder's first focused feature enhancement transfer module FFETM, the second focused feature enhancement transfer module FFETM, the third focused feature enhancement transfer module FFETM, and the fourth focused feature enhancement transfer module FFETM are all 3×3, the padding is 1, and the step size is 1. The convolution kernel size of the second convolution layer is 3×3, the padding is 1, and the step size is 1. The pooling kernel of the maximum pooling layer is 2×2, and the step size is 1. The convolution kernel size of the convolution layer of the second branch is 1×1, and the step size is 1. The first feature interactive perception decoding module FIPDM, the second feature interactive perception decoding module FIPDM, the third feature interactive perception decoding module FIPDM, and the fourth feature interactive perception The convolution kernel size of the convolution layer of the decoding module FIPDM is 3×3, and the step size is 1. The convolution kernel size of the convolution kernel of the first branch is 3×3, and the step size is 1. The convolution kernel size of the depthwise separable convolution layer of the first branch is 3×3, and the step size is 1, and the padding is 1. The convolution kernel size of the depthwise separable convolution layer of the second branch is 3×3, and the step size is 1, and the padding is 1. The convolution kernel size of the convolution kernel of the convolution layer of the third branch is 1×1, and the step size is 1; the convolution kernel size of the transposed convolution layer of the first upsampling unit, the second upsampling unit, the third upsampling unit, and the fourth upsampling unit of the decoder is 3×3, the padding is 1, and the step size is 2; the convolution kernel size of the convolution layer of the decoder in step f-18) is 1×1.

9. The cardiac ultrasound image segmentation method based on state space model and feature interactive perception according to claim 1, characterized in that: Step g) comprises the following steps: g-1) through the formula L total =αL Dice +βL CE +γL boundary +δL Tversky Calculate the total loss L total , where L Dice is the Dice loss, L CE is the cross entropy loss, L boundary is the boundary loss, L Tversky is Tversky loss, α, β, γ, δ are all hyperparameters; g-2) Use the SGD optimizer to take advantage of the total loss L total Train the cardiac ultrasound image segmentation network model to obtain an optimized cardiac ultrasound image segmentation network model.

10. The cardiac ultrasound image segmentation method based on state space model and feature interactive perception according to claim 9, characterized in that: When training the cardiac ultrasound image segmentation network model, the learning rate was set to 0.001, the batch size was set to 16, the number of iterations was set to 1200, the hyperparameter α was set to 0.3, the hyperparameter β was set to 0.3, the hyperparameter γ was set to 0.2, and the hyperparameter δ was set to 0.2.

Images