Method and electronic device for segmenting myoma-like biological tissue

Abstract

The present disclosure provides a myoma biological tissue segmentation method and an electronic device. The method comprises: performing spatial conversion and channel feature extraction on each first feature map obtained by performing multiple feature extractions on a medical image of a myoma biological tissue based on a spatial channel attention mechanism to obtain each second feature map; performing channel compression and feature extraction at least once on each up-sampling feature map obtained by up-sampling each second feature map to obtain a third feature map, wherein the same up-sampling feature map has the same number of times of channel compression and feature extraction, different up-sampling feature maps have different numbers of times of channel compression and feature extraction, any up-sampling feature map first performs channel compression once and then performs feature extraction once, and the image of the latter channel compression is the image obtained by fusing the feature map obtained by the former feature extraction and the up-sampling feature map; and obtaining the myoma biological tissue by using the third feature map and a fourth feature map obtained based on the second feature map. The accuracy of segmenting the myoma biological tissue is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to a method for segmenting fibroid biological tissue and an electronic device. Background Art

[0002] Fibroids are the most common tumors of the reproductive organs. An example is uterine fibroids. There are various ways to examine fibroids. Ultrasound is the most commonly used diagnostic method in clinical practice. It can reveal the number, location, and size of fibroids, as well as the presence of calcifications, cystic changes, and other changes within them.

[0003] Currently, ultrasound-based segmentation methods for myoma-like biological tissues are mainly concentrated in the field of traditional image processing. That is, based on the pixel distribution in the image of myoma-like biological tissue, a specific threshold is set, and areas above or below the threshold are determined as selected areas and segmented. However, this method requires a preset threshold, and the threshold cannot be flexibly changed. This method is only suitable for situations where the recognition scene is single and the foreground and background in the image are highly distinguishable. However, due to the high noise content of ultrasound images and the different morphology and grayscale of myoma-like biological tissues in the image, the use of the existing methods will result in a low accuracy rate in segmenting myoma-like biological tissues. Summary of the Invention

[0004] In an exemplary embodiment of the present disclosure, a method and electronic device for segmenting tumor-like biological tissue are provided to improve the accuracy of segmenting tumor-like biological tissue.

[0005] A first aspect of the present disclosure provides a method for segmenting tumor-like biological tissue, the method comprising:

[0006] Continuously performing multiple feature extractions on the medical image of the myoma-like biological tissue to obtain first feature maps, wherein the number of the first feature maps is the same as the number of feature extractions;

[0007] For any first feature map, perform spatial transformation and channel feature extraction on the first feature map based on the spatial channel attention mechanism to obtain a second feature map;

[0008] Upsampling each second feature map to obtain each upsampled feature map, and performing channel compression and feature extraction on each upsampled feature map at least once to obtain a third feature map, wherein the number of channel compressions for the same upsampled feature map is the same as the number of feature extractions thereof, and the number of channel compressions and feature extractions for different upsampled feature maps is different, and any upsampled feature map is first subjected to channel compression once and then to feature extraction once each time, and the image used for the latter channel compression is an image obtained by fusing the feature map obtained by the previous feature extraction and the upsampled feature map;

[0009] The structure of the myoma-like biological tissue is obtained by fusing the third feature map and a fourth feature map obtained based on the second feature maps.

[0010] In the embodiment, multiple first feature maps are obtained by continuously performing feature extraction on the medical image of the myoma-like biological tissue, and a second feature map is obtained by performing spatial conversion and channel feature extraction on the first feature map. Each second feature map is up-sampled, and each up-sampled second feature map is subjected to at least one channel compression and at least one feature extraction to obtain a third feature map. Finally, the third feature map and a fourth feature map obtained by decoding the second feature map are fused to obtain the structure of the myoma-like biological tissue. Thus, the myoma-like biological tissue is automatically segmented by using a neural network, and a preset threshold is not required, thereby improving the accuracy of segmenting the myoma-like biological tissue.

[0011] In one embodiment, the continuously performing feature extraction on the medical image of the myoma-like biological tissue to obtain the first feature maps comprises:

[0012] The multiple encoders are sequentially used to continuously perform feature extraction on the medical image to obtain the first feature maps. The number of the first feature maps is the same as the number of the encoders, and the image input to a subsequent encoder is the first feature map output by a previous encoder.

[0013] In the embodiment, the multiple encoders are used to continuously perform feature extraction on the medical image to obtain the multiple first feature maps, so that more dimensional feature information can be extracted, and the accuracy of segmentation is further improved.

[0014] In one embodiment, the sequentially using the multiple encoders to continuously perform feature extraction on the medical image to obtain the first feature maps comprises:

[0015] For any encoder, the following steps are performed:

[0016] Feature extraction is performed on a first target image by using a dilated convolution layer in the encoder to obtain a first intermediate feature map, wherein the first target image is the medical image or the first intermediate feature map output by a previous encoder; and

[0017] A non-linear transformation is performed on the first intermediate feature map by using a Leaky Relu function to obtain a second intermediate feature map;

[0018] Feature extraction is performed on the second intermediate feature map by using a dilated convolution layer to obtain a third intermediate feature map;

[0019] Fusing the third intermediate feature map with the first target image to obtain a fourth intermediate feature map;

[0020] After performing a nonlinear transformation on the fourth intermediate feature map using the Leaky Relu function, a maximum pooling layer is used to perform image compression on the fourth intermediate feature map after the nonlinear transformation to obtain the first feature map.

[0021] In this embodiment, the conventional convolution layer in the encoder is replaced by a dilated convolution layer. Since the receptive field of the dilated convolution layer is larger than that of the conventional convolution layer under the same size of the convolution kernel, more feature information can be extracted using dilated convolution. In this embodiment, the third intermediate feature map obtained by the dilated convolution layer is fused with the first target image, so that the final first feature map can retain more detail information, further improving the accuracy of segmenting tumor-like biological tissues.

[0022] In one embodiment, if feature extraction is performed on the medical image for the first time, the first target image is the medical image; and

[0023] If this is not the first time that feature extraction is performed on the medical image, the first target image is the first intermediate feature map output by the previous encoder.

[0024] In this embodiment, the first target images input by encoders at different positions are different, thereby extracting features of more dimensions and improving the accuracy of segmenting fibroid-like biological tissues.

[0025] In one embodiment, performing spatial transformation and channel feature extraction on the first feature map based on the spatial channel attention mechanism to obtain the second feature map includes:

[0026] Performing spatial transformation on the first feature map using the spatial attention mechanism in the spatial channel attention mechanism to obtain a fifth intermediate feature map; and

[0027] Performing channel feature extraction on the first feature map through the channel attention mechanism in the spatial channel attention mechanism to obtain a sixth intermediate feature map;

[0028] The fifth intermediate feature map and the sixth intermediate feature map are fused to obtain a seventh intermediate feature map, and feature extraction is performed on the seventh intermediate feature map to obtain the second feature map.

[0029] In this embodiment, by performing spatial transformation and channel feature extraction on the first feature map, a second feature map is generated in which elements in important spaces and important channels are enlarged, thereby weakening unimportant spaces and channels and further improving the accuracy of segmentation.

[0030] In one embodiment, the utilizing the spatial attention mechanism in the spatial channel attention mechanism to perform spatial transformation on the first feature map to obtain a fifth intermediate feature map includes:

[0031] sequentially performing multiple feature extractions on the first feature map using multiple convolutional layers to obtain a first sub-feature map;

[0032] Performing a fully connected operation on the first sub-feature map using a combination of a fully connected layer and an activation function to generate a target weight for each position of the first sub-feature map;

[0033] The first sub-feature map after the nonlinear transformation is updated based on the target weights of the respective positions to obtain the fifth intermediate feature map.

[0034] This embodiment performs spatial transformation on the first feature map through a spatial attention mechanism, thereby expanding important spatial information and weakening unimportant spatial information, thereby further improving the accuracy of segmentation.

[0035] In one embodiment, performing channel feature extraction on the first feature map by using the channel attention mechanism in the spatial channel attention mechanism to obtain a sixth intermediate feature map includes:

[0036] sequentially performing multiple feature extractions on the first feature map using multiple convolutional layers to obtain a second sub-feature map;

[0037] Performing channel feature extraction on the second sub-feature map to obtain a third sub-feature map;

[0038] The third sub-feature map is fused with the second sub-feature map to obtain the sixth intermediate feature map.

[0039] This embodiment performs channel feature extraction processing on the first feature map through a channel attention mechanism, thereby expanding the information on important channels and weakening the information on unimportant channels, further improving the accuracy of segmentation.

[0040] In one embodiment, the fourth feature map is obtained by:

[0041] Multiple decoders are used in sequence to continuously decode the second target image multiple times to obtain the fourth feature map, wherein the second target image is an image obtained by fusing the second feature map with the first feature map or an image obtained by fusing the second feature map with the fourth feature map output by the previous decoder, and the number of the decoders is the same as the number of encoders.

[0042] In this embodiment, the second target image is decoded by the same number of decoders as the number of encoders, so that the determined fourth feature map is more accurate.

[0043] In one embodiment, the sequentially using multiple decoders to continuously decode the second target image multiple times to obtain the fourth feature map includes:

[0044] For any decoder, perform the following steps:

[0045] Upsampling the second target image to obtain a global feature map of the fibroid biological tissue;

[0046] Normalizing the global feature map using a Sigmoid function to obtain a normalized global feature map;

[0047] Using a dilated convolutional layer to extract features from the normalized global feature map to obtain a local feature map;

[0048] The local feature map is normalized using a Sigmoid function to obtain the fourth feature map.

[0049] In one embodiment, determining the structure of the myoma-like biological tissue using the third characteristic map and a fourth characteristic map obtained based on each of the second characteristic maps includes:

[0050] fusing the third feature map and the fourth feature map to obtain a mask image of the fibroid-like biological tissue, wherein the mask image is a binary image including the outline of the fibroid-like biological tissue;

[0051] The mask image is subjected to contour detection using a preset edge detection algorithm, and the structure of the myoma-like biological tissue is determined in the medical image.

[0052] This embodiment determines the structure of the myoma-like biological tissue by fusing the third feature map with the fourth feature map, so that the structure of the myoma-like biological tissue is determined to have more detailed information, making the determined structure of the myoma-like biological tissue more accurate.

[0053] A second aspect of the present disclosure provides an electronic device, including a storage unit and a processor, wherein:

[0054] The storage unit is configured to store medical images of myoma-like biological tissues;

[0055] The processor is configured to:

[0056] Continuously performing multiple feature extractions on the medical image of the myoma-like biological tissue to obtain first feature maps, wherein the number of the first feature maps is the same as the number of feature extractions;

[0057] For any first feature map, perform spatial transformation and channel feature extraction on the first feature map based on the spatial channel attention mechanism to obtain a second feature map;

[0058] Upsampling each second feature map to obtain each upsampled feature map, and performing channel compression and feature extraction on each upsampled feature map at least once to obtain a third feature map, wherein the number of channel compressions for the same upsampled feature map is the same as the number of feature extractions thereof, and the number of channel compressions and feature extractions for different upsampled feature maps is different, and any upsampled feature map is first subjected to channel compression once and then to feature extraction once each time, and the image used for the latter channel compression is an image obtained by fusing the feature map obtained by the previous feature extraction and the upsampled feature map;

[0059] The structure of the myoma-like biological tissue is obtained by using the third characteristic map and a fourth characteristic map obtained based on the second characteristic maps.

[0060] In one embodiment, the processor performs the feature extraction on the medical image of the myoma-like biological tissue multiple times continuously to obtain each first feature map, and is specifically configured as follows:

[0061] Multiple encoders are used in sequence to perform multiple feature extractions on the medical image to obtain the first feature maps, wherein the number of the first feature maps is the same as the number of the multiple encoders, and the image input to the latter encoder is the first feature map output by the previous encoder.

[0062] In one embodiment, the processor executes the step of sequentially using multiple encoders to continuously extract features from the medical image multiple times to obtain the first feature maps, and is specifically configured as follows:

[0063] For any encoder, perform the following steps:

[0064] Performing feature extraction on a first target image using a dilated convolutional layer in the encoder to obtain a first intermediate feature map, wherein the first target image is the medical image or the first intermediate feature map output by a previous encoder; and

[0065] Performing a nonlinear transformation on the first intermediate feature map using a leaky rectified linear unit (Leaky Relu) function to obtain a second intermediate feature map;

[0066] Performing feature extraction on the second intermediate feature map using a dilated convolutional layer to obtain a third intermediate feature map;

[0067] Fusing the third intermediate feature map with the first target image to obtain a fourth intermediate feature map;

[0068] After performing a nonlinear transformation on the fourth intermediate feature map using the Leaky Relu function, a maximum pooling layer is used to perform image compression on the fourth intermediate feature map after the nonlinear transformation to obtain the first feature map.

[0069] In one embodiment, if feature extraction is performed on the medical image for the first time, the first target image is the medical image; and

[0070] If this is not the first time that feature extraction is performed on the medical image, the first target image is the first intermediate feature map output by the previous encoder.

[0071] In one embodiment, performing spatial transformation and channel feature extraction on the first feature map based on the spatial channel attention mechanism to obtain the second feature map includes:

[0072] Performing spatial transformation on the first feature map using the spatial attention mechanism in the spatial channel attention mechanism to obtain a fifth intermediate feature map; and

[0073] Performing channel feature extraction on the first feature map through the channel attention mechanism in the spatial channel attention mechanism to obtain a sixth intermediate feature map;

[0074] The fifth intermediate feature map and the sixth intermediate feature map are fused to obtain a seventh intermediate feature map, and feature extraction is performed on the seventh intermediate feature map to obtain the second feature map.

[0075] In one embodiment, the processor uses the spatial attention mechanism in the spatial channel attention mechanism to perform spatial transformation on the first feature map to obtain the fifth intermediate feature map, and is specifically configured to:

[0076] sequentially performing multiple feature extractions on the first feature map using multiple convolutional layers to obtain a first sub-feature map;

[0077] Performing a fully connected operation on the first sub-feature map using a combination of a fully connected layer and an activation function to generate a target weight for each position of the first sub-feature map;

[0078] The first sub-feature map after the nonlinear transformation is updated based on the target weights of the respective positions to obtain the fifth intermediate feature map.

[0079] In one embodiment, the processor performs the channel feature extraction process on the first feature map through the channel attention mechanism in the spatial channel attention mechanism to obtain a sixth intermediate feature map, and is specifically configured as follows:

[0080] sequentially performing multiple feature extractions on the first feature map using multiple convolutional layers to obtain a second sub-feature map;

[0081] Performing channel feature extraction on the second sub-feature map to obtain a third sub-feature map;

[0082] The third sub-feature map is fused with the second sub-feature map to obtain the sixth intermediate feature map.

[0083] In one embodiment, the processor is further configured to:

[0084] The fourth characteristic map is obtained by:

[0085] Multiple decoders are used in sequence to continuously decode the second target image multiple times to obtain the fourth feature map, wherein the second target image is an image obtained by fusing the second feature map with the first feature map or an image obtained by fusing the second feature map with the fourth feature map output by the previous decoder, and the number of the decoders is the same as the number of encoders.

[0086] In one embodiment, the processor executes the sequentially using multiple decoders to continuously decode the second target image multiple times to obtain the fourth feature map, and is specifically configured as follows:

[0087] For any decoder, perform the following steps:

[0088] Upsampling the second target image to obtain a global feature map of the fibroid biological tissue;

[0089] Normalizing the global feature map using a Sigmoid function to obtain a normalized global feature map;

[0090] Using a dilated convolutional layer to extract features from the normalized global feature map to obtain a local feature map;

[0091] The local feature map is normalized using a Sigmoid function to obtain the fourth feature map.

[0092] In one embodiment, the processor executes the step of determining the structure of the fibroid-like biological tissue by using the third feature map and the fourth feature map obtained based on the second feature maps, and is specifically configured as follows:

[0093] fusing the third feature map and the fourth feature map to obtain a mask image of the fibroid-like biological tissue, wherein the mask image is a binary image including the outline of the fibroid-like biological tissue;

[0094] The mask image is subjected to contour detection using a preset edge detection algorithm, and the structure of the myoma-like biological tissue is determined in the medical image.

[0095] According to a third aspect of an embodiment of the present disclosure, there is provided an electronic device, including:

[0096] At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executed by the at least one processor; the instructions are executed by the at least one processor to enable the at least one processor to execute the method as described in the first aspect.

[0097] According to a fourth aspect provided by an embodiment of the present disclosure, a computer storage medium is provided, wherein the computer storage medium stores a computer program, and the computer program is used to execute the method as described in the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0098] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0099] Figure 1 A schematic diagram of an applicable scenario according to an embodiment of the present disclosure;

[0100] Figure 2 FIG1 is a flow chart of a method for segmenting fibroid-like biological tissue according to an embodiment of the present disclosure;

[0101] Figure 3 1 is a connection diagram of an encoder according to an embodiment of the present disclosure;

[0102] Figure 4 is a structural diagram of an encoder according to an embodiment of the present disclosure;

[0103] Figure 5A Schematic diagram of the structure of the spatial channel attention mechanism according to one embodiment of the present disclosure;

[0104] Figure 5B Schematic diagram of the structure of the spatial attention mechanism according to one embodiment of the present disclosure;

[0105] Figure 5CSchematic diagram of the structure of the channel attention mechanism according to one embodiment of the present disclosure;

[0106] Figure 6 1 is a schematic structural diagram of a shape flow module according to an embodiment of the present disclosure;

[0107] Figure 7A is a schematic diagram of a mask image according to one embodiment of the present disclosure;

[0108] Figure 7B is a schematic diagram of a myoma-like biological tissue structure in a medical image according to one embodiment of the present disclosure;

[0109] Figure 8 Schematic diagram of the structure of a neural network model for segmenting myoma-like biological tissue according to one embodiment of the present disclosure;

[0110] Figure 9 1 is a schematic structural diagram of a decoder according to an embodiment of the present disclosure;

[0111] Figure 10 A schematic diagram of a segmentation device based on myoma-like biological tissue;

[0112] Figure 11 FIG. 4 is a schematic structural diagram of an electronic device according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0113] To make the objectives, technical solutions, and advantages of the embodiments of the present disclosure more clear, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present disclosure, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present disclosure without making any creative efforts shall fall within the scope of protection of the present disclosure.

[0114] In the embodiments of the present disclosure, the term "and / or" describes the association relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone. The character " / " generally indicates that the associated objects are in an "or" relationship.

[0115] The application scenarios described in the embodiments of the present disclosure are intended to more clearly illustrate the technical solutions of the embodiments of the present disclosure and do not constitute a limitation on the technical solutions provided by the embodiments of the present disclosure. Persons skilled in the art will appreciate that, as new application scenarios emerge, the technical solutions provided by the embodiments of the present disclosure will also be applicable to similar technical problems. In the description of the present disclosure, unless otherwise specified, "multiple" means two or more.

[0116] In the prior art, ultrasound-based segmentation methods for myoma-like biological tissues are mainly concentrated in the field of traditional image processing. That is, a specific threshold is set based on the pixel distribution in the image of myoma-like biological tissue, and the area above or below the threshold is determined as the selected area and segmented. However, this method requires a preset threshold, and the threshold cannot be flexibly changed. This method is only suitable for situations where the recognition scene is single and the foreground and background in the image are highly distinguishable. However, due to the high noise content of ultrasound images and the different morphologies and grayscales of myoma-like biological tissues in the image, the use of the methods in the prior art will result in a low accuracy rate in segmenting myoma-like biological tissues.

[0117] Therefore, the present disclosure provides a method for segmenting fibroid-like biological tissues, which obtains multiple first feature maps by continuously performing multiple feature extractions on medical images of fibroid-like biological tissues, and performs spatial conversion and channel feature extraction on the first feature maps to obtain second feature maps, and upsamples each of the second feature maps, and performs at least one channel compression and at least one feature extraction on each upsampled second feature map to obtain a third feature map, and fuses the third feature map with the fourth feature map obtained by decoding the second feature map. Thus, the present invention automatically segments fibroid-like biological tissues by means of a neural network, and does not require a preset threshold, thereby improving the accuracy of segmenting fibroid-like biological tissues. Below, the scheme of the present disclosure is described in detail in conjunction with the accompanying drawings.

[0118] like Figure 1 As shown, an application scenario of a method for segmenting tumor-like biological tissue includes an ultrasound device 110 (scanning unit 111 and display unit 112) and a server 120. Server 120 can be implemented as a single server or multiple servers. Server 120 can be implemented as a physical server or a virtual server.

[0119] In a possible application scenario, a medical image of a myoma-like biological tissue is acquired by a scanning unit 111 of an ultrasound device 110, and the medical image is sent to a server 120. After receiving the medical image, the server 120 continuously performs feature extraction on the medical image of the myoma-like biological tissue multiple times to obtain multiple first feature maps, where the number of the first feature maps is the same as the number of times of feature extraction. Then, for any one first feature map, the server 120 performs spatial conversion and channel feature extraction on the first feature map based on a spatial channel attention mechanism to obtain a second feature map. After that, the server 120 performs upsampling on each second feature map to obtain each upsampled feature map, and performs channel compression and feature extraction on each upsampled feature map at least once to obtain a third feature map, where the number of times of channel compression of a same upsampled feature map is the same as the number of times of feature extraction of the upsampled feature map, the number of times of channel compression and feature extraction of different upsampled feature maps is different, and any upsampled feature map performs channel compression and feature extraction at a time, the image used by the latter channel compression is an image obtained by fusing the feature map obtained by the former feature extraction and the upsampled feature map. Finally, the server 120 obtains the structure of the myoma-like biological tissue by using the third feature map and a fourth feature map obtained based on the second feature maps. The server 120 sends a medical image in which the structure of the myoma-like biological tissue is determined to the ultrasound device 110, and the ultrasound device 110 displays the medical image by using a display unit 112.

[0120] As shown in FIG. 1, a flowchart of a method for segmenting a myoma-like biological tissue according to the present disclosure can include the following steps: Figure 2

[0121] Step 201: Continuously performing feature extraction on a medical image of a myoma-like biological tissue multiple times to obtain first feature maps, where the number of the first feature maps is the same as the number of times of feature extraction.

[0122] In this embodiment, the medical image is an ultrasound image, and in order to ensure the accuracy of segmentation, the medical image needs to be preprocessed before step 201 is performed, including the following two steps:

[0123] 1. Anisotropic diffusion filtering:

[0124] The medical image is filtered by using a speckle-suppression anisotropic diffusion filter, so that the local texture detail changes caused by speckle noise are suppressed while the main myoma-like biological tissue morphological information is maintained, and the mixed noise in the ultrasound image is eliminated while the organ contour is retained.

[0125] ​2. CLAHE algorithm (Contrast Limited Adaptive Histogram Equalization, limited contrast adaptive histogram equalization algorithm):

[0126] The CLAHE algorithm increases the contrast of medical images. Histogram imbalance in medical images primarily stems from adjustments to ultrasound imaging parameters and different acquisition techniques. From an image perspective, this imbalance manifests as varying brightness levels. When images of varying brightness levels are fed into a network model, the neural network learns these features. This brightness level slows network convergence and can lead to errors. Therefore, increasing the contrast of medical images is crucial to reducing errors in the segmentation of myoma-like tissue.

[0127] In one embodiment, the first feature maps are obtained by:

[0128] Multiple encoders are used in sequence to perform multiple feature extractions on the medical image to obtain the first feature maps, wherein the number of the first feature maps is the same as the number of the multiple encoders, and the image input to the latter encoder is the first feature map output by the previous encoder.

[0129] like Figure 3 As shown in FIG, a connection diagram of multiple encoders in this embodiment is shown, which includes 4 encoders. Among them, the first encoder inputs a medical image, and the inputs of the subsequent encoders are all outputs of the previous encoder, but each encoder will obtain the corresponding first feature map.

[0130] It should be noted that the neural network in this embodiment uses four encoders for feature extraction. The number of encoders in this embodiment is only used for illustration and does not limit the number of encoders in the neural network. The number of encoders can be limited according to actual conditions.

[0131] Next, the method of obtaining each first feature map in this implementation is described in detail in conjunction with the structure of the encoder:

[0132] like Figure 4 As shown in FIG, it is a structural diagram of any encoder, which includes a dilated convolution layer 410, a leaky Relu function 420, a dilated convolution layer 430, a leaky Relu function 440 and a maximum pooling layer 450. The following is an introduction to the process of obtaining the first feature map using the encoder:

[0133] The atrous convolution layer 410 in the encoder is used to extract features from a first target image to obtain a first intermediate feature map, wherein the first target image is the medical image or the first intermediate feature map output by the previous encoder; the first intermediate feature map is nonlinearly transformed using a rectified linear unit Leaky Relu function 420 to obtain a second intermediate feature map; the second intermediate feature map is extracted using a atrous convolution layer 430 to obtain a third intermediate feature map; the third intermediate feature map is fused with the first target image to obtain a fourth intermediate feature map; the fourth intermediate feature map is nonlinearly transformed using a Leaky Relu function 440, and then the fourth intermediate feature map after the nonlinear transformation is compressed using a maximum pooling layer 450 to obtain the first feature map.

[0134] In one embodiment, if feature extraction is performed on the medical image for the first time, the first target image is the medical image; if feature extraction is not performed on the medical image for the first time, the first target image is the first intermediate feature map output by the previous encoder.

[0135] For example, Figure 3 If the first target image input by the first encoder is a medical image, then the first target image input by the second encoder and each encoder after the second encoder is the first intermediate feature map output by the previous encoder.

[0136] Step 202: For any first feature map, perform spatial transformation and channel feature extraction on the first feature map based on the spatial channel attention mechanism to obtain a second feature map;

[0137] In one embodiment, the second feature map is obtained by:

[0138] The first feature map is spatially transformed by using the spatial attention mechanism in the spatial channel attention mechanism to obtain a fifth intermediate feature map; and the first feature map is subjected to channel feature extraction by using the channel attention mechanism in the spatial channel attention mechanism to obtain a sixth intermediate feature map; the fifth intermediate feature map and the sixth intermediate feature map are fused to obtain a seventh intermediate feature map, and feature extraction is performed on the seventh intermediate feature map to obtain the second feature map.

[0139] For example, Figure 5AAs shown, the spatial channel attention mechanism includes a spatial attention mechanism 510, a channel attention mechanism 520 and a convolution layer 530, wherein the spatial attention mechanism 510 is used to perform spatial transformation on the first feature map to obtain a fifth intermediate feature map; the channel attention mechanism 520 is used to perform channel feature extraction on the first feature map to obtain a sixth intermediate feature map, the fifth intermediate feature map and the sixth intermediate feature map are fused to obtain a seventh intermediate feature map, and the convolution layer 530 is used to perform feature extraction on the seventh intermediate feature map to obtain the second feature map.

[0140] Next, the structures and workflows of the spatial attention mechanism 510 and the channel attention mechanism 520 are introduced respectively:

[0141] 1. Spatial Attention Mechanism 510:

[0142] like Figure 5B As shown, the spatial attention mechanism 510 includes two convolutional layers 5B21, a combination structure of two fully connected layers + activation functions 5B22, and a ReLU function 5B23. The following describes the specific process of obtaining the fifth intermediate feature map by spatially transforming the first feature map based on the spatial attention mechanism:

[0143] Two convolutional layers 5B21 are used in sequence to perform multiple feature extractions on the first feature map to obtain a first sub-feature map. A full connection operation is performed on the first sub-feature map using a combined structure 5B22 of two fully connected layers and an activation function to generate target weights for each position of the first sub-feature map. A nonlinear transformation is performed on the first sub-feature map using a Relu function 5B23 to obtain a first sub-feature map after the nonlinear transformation. Finally, the first sub-feature map after the nonlinear transformation is updated based on the target weights for each position to obtain the fifth intermediate feature map.

[0144] It should be noted that the combination structure 5B22 of the fully connected layer + activation function in this embodiment includes a fully connected layer + ReLU function and a fully connected layer + Sigmoid function. The specific combination structure of the fully connected layer + activation function can be set according to actual conditions and is not limited in this embodiment.

[0145] (2) Channel Attention Mechanism 520:

[0146] like Figure 5C As shown, the channel attention mechanism 520 includes two convolutional layers 5C31 and a channel feature extraction layer 5C32. The specific process of performing channel feature extraction processing on the first feature map through the channel attention mechanism to obtain the sixth intermediate feature map is described in detail below:

[0147] Two convolutional layers 5C31 are used in sequence to perform multiple feature extractions on the first feature map to obtain a second sub-feature map. Channel feature extraction layer 5C32 is used to perform channel feature extraction on the second sub-feature map to obtain a third sub-feature map. The third sub-feature map is fused with the second sub-feature map to obtain the sixth intermediate feature map.

[0148] It should be noted that the channel feature extraction layer in this embodiment is implemented using a 1*1 convolutional layer.

[0149] Step 203: upsampling each second feature map to obtain each upsampled feature map, and performing channel compression and feature extraction on each upsampled feature map at least once to obtain a third feature map, wherein the number of channel compressions for the same upsampled feature map is the same as the number of feature extractions, and the number of channel compressions and feature extractions for different upsampled feature maps is different, and any upsampled feature map is first subjected to channel compression once and then to feature extraction once each time, and the image used for the latter channel compression is an image obtained by fusing the feature map obtained by the previous feature extraction and the upsampled feature map;

[0150] In this embodiment, the third feature map is obtained by the shape flow module, such as Figure 6 , which is a schematic diagram of the structure of the shape flow module, includes four transposed convolutional layers 610, four channel compression layers 620, and four convolutional layers 630. The following describes the specific process of obtaining the third feature map based on the shape flow module:

[0151] The second feature map is upsampled using four transposed convolutional layers 610 to obtain upsampled feature maps, which are then input into corresponding channel compression layers 620. After each channel compression layer 620 receives the upsampled feature map, the channel compression layers at different positions have different processing flows: the first channel compression layer 620 performs channel compression on the received upsampled feature map, and then inputs the channel-compressed upsampled feature map into the first convolutional layer 630 for feature extraction. The first convolutional layer 630 then inputs the feature-extracted upsampled feature map into the second channel compression layer. After receiving the feature-extracted upsampled feature map, the second feature map is fused with the feature-extracted upsampled feature map to obtain a fused image. The fused image is then subjected to channel compression and feature extraction. The processing flows of the channel compression modules after the second channel compression module are the same as those of the second channel compression module and will not be described in detail here.

[0152] Step 204: Obtain the structure of the myoma-like biological tissue using the third characteristic map and a fourth characteristic map obtained based on the second characteristic maps.

[0153] In one embodiment, step 204 can be implemented as follows: fusing the third feature map with the fourth feature map to obtain a mask image of the tumor-like biological tissue, wherein the mask image is a binary image containing the contour of the tumor-like biological tissue; performing contour detection on the mask image using a preset edge detection algorithm to determine the structure of the tumor-like biological tissue in the medical image.

[0154] like Figure 7A As shown, it is a mask image of a myoma-like biological tissue, which is a binary image containing the outline of the myoma-like biological tissue. Figure 7A The white area in the figure is the outline of the tumor-like biological tissue. The edge detection algorithm is used to Figure 7A By performing contour detection in the image, the structure of the myoma-like biological tissue can be determined in the medical image, such as Figure 7B As shown, the circled area is the structure of the myoma-like biological tissue, and the long diameter and short diameter of the myoma-like biological tissue structure can be displayed, that is, Figure 7B Cross lines in .

[0155] In one embodiment, the fourth feature map is obtained by:

[0156] Multiple decoders are used in sequence to continuously decode the second target image multiple times to obtain the fourth feature map, wherein the second target image is an image obtained by fusing the second feature map with the first feature map or an image obtained by fusing the second feature map with the fourth feature map output by the previous decoder, and the number of the encoders is the same as the number of decoders.

[0157] It should be noted that: according to the order in which the decoders work, for any decoder, if the decoder is determined to be the first decoder to work, the second target image is the image obtained by fusing the second feature map with the first feature map; if the decoder is not the first decoder to work, the second target image is the image obtained by fusing the second feature map with the fourth feature map output by the previous decoder.

[0158] For example, Figure 8 , which is a structural diagram of a neural network model for segmenting tumor-like biological tissue, includes an encoder 810 , a spatial channel attention mechanism 820 , a decoder 830 and a shape flow module 840 .

[0159] like Figure 8As can be seen, in this embodiment, four encoders 810 are first used to continuously perform multiple feature extractions on the medical image to obtain the four first feature maps. Then, for any of the first feature maps, the spatial channel attention mechanism 820 performs spatial transformation and channel feature extraction on the first feature map to obtain a second intermediate feature map. The spatial channel attention mechanism 820 sends each of the obtained second intermediate feature maps to the shape flow module 840 and the decoder 830, respectively. The shape flow module 840 obtains a third feature map, and the decoder 830 obtains a fourth feature map. Finally, the third feature map and the fourth feature map are fused to obtain a mask image of the myoma-like biological tissue. Based on the mask image of the myoma-like biological tissue, the structure of the myoma-like biological tissue is determined in the medical image.

[0160] The following describes the decoding process for any decoder. Figure 9 , which is a schematic diagram of the structure of any decoder, including a transposed convolution layer 910, a Sigmoid function 920, a hole convolution layer 930 and a Sigmoid function 940.

[0161] First, the second target image is upsampled using a transposed convolution layer 910 to obtain a global feature map of the tumor-like biological tissue. Then, the global feature map is normalized using a Sigmoid function 920 to obtain a normalized global feature map. A hole convolution layer 930 is used to extract features from the normalized global feature map to obtain a local feature map. Finally, the local feature map is normalized using a Sigmoid function 940 to obtain the fourth feature map.

[0162] Based on the same disclosed concept, the above-mentioned method for segmenting fibroid tissue can also be implemented by a device for segmenting fibroid tissue. The effect of the device for segmenting fibroid tissue is similar to that of the above-mentioned method, and will not be described in detail here.

[0163] Figure 10 Schematic diagram of the structure of a device for segmenting tumor-like biological tissue according to an embodiment of the present disclosure.

[0164] like Figure 10 As shown, the fibroid biological tissue segmentation device 1000 disclosed herein may include a first feature map determination module 1010 , a second feature map determination module 1020 , a third feature map determination module 1030 and a fibroid biological tissue structure determination module 1040 .

[0165] A first feature map determining module 1010 is configured to continuously perform multiple feature extractions on the medical image of the myoma-like biological tissue to obtain first feature maps, wherein the number of first feature maps is the same as the number of feature extractions;

[0166] A second feature map determining module 1020 is configured to perform spatial transformation and channel feature extraction on any first feature map based on a spatial channel attention mechanism to obtain a second feature map;

[0167] A third feature map determining module 1030 is configured to upsample each second feature map to obtain an upsampled feature map, and perform channel compression and feature extraction on each upsampled feature map at least once to obtain a third feature map, wherein the number of channel compressions performed on the same upsampled feature map is the same as the number of feature extractions performed on it, and the number of channel compressions and feature extractions performed on different upsampled feature maps is different, and each upsampled feature map is first subjected to channel compression once and then to feature extraction once, and the image used for the latter channel compression is an image obtained by fusing the feature map obtained by the previous feature extraction and the upsampled feature map;

[0168] The myoma-like biological tissue structure determination module 1040 is configured to obtain the structure of the myoma-like biological tissue using the third feature map and a fourth feature map obtained based on the second feature maps.

[0169] In one embodiment, the first feature map determining module 1010 is specifically configured to:

[0170] Multiple encoders are used in sequence to perform multiple feature extractions on the medical image to obtain the first feature maps, wherein the number of the first feature maps is the same as the number of the multiple encoders, and the image input to the latter encoder is the first feature map output by the previous encoder.

[0171] In one embodiment, the first feature map determining module 1010 is specifically configured to:

[0172] For any encoder, perform the following steps:

[0173] Performing feature extraction on a first target image using a dilated convolutional layer in the encoder to obtain a first intermediate feature map, wherein the first target image is the medical image or the first intermediate feature map output by a previous encoder; and

[0174] Performing a nonlinear transformation on the first intermediate feature map using a leaky rectified linear unit (Leaky Relu) function to obtain a second intermediate feature map;

[0175] Performing feature extraction on the second intermediate feature map using a dilated convolutional layer to obtain a third intermediate feature map;

[0176] Fusing the third intermediate feature map with the first target image to obtain a fourth intermediate feature map;

[0177] After performing a nonlinear transformation on the fourth intermediate feature map using the Leaky Relu function, a maximum pooling layer is used to perform image compression on the fourth intermediate feature map after the nonlinear transformation to obtain the first feature map.

[0178] In one embodiment, if feature extraction is performed on the medical image for the first time, the first target image is the medical image; and

[0179] If this is not the first time that feature extraction is performed on the medical image, the first target image is the first intermediate feature map output by the previous encoder.

[0180] In one embodiment, the second feature map determining module 1020 is specifically configured to:

[0181] Performing spatial transformation on the first feature map using the spatial attention mechanism in the spatial channel attention mechanism to obtain a fifth intermediate feature map; and

[0182] Performing channel feature extraction on the first feature map through the channel attention mechanism in the spatial channel attention mechanism to obtain a sixth intermediate feature map;

[0183] The fifth intermediate feature map and the sixth intermediate feature map are fused to obtain a seventh intermediate feature map, and feature extraction is performed on the seventh intermediate feature map to obtain the second feature map.

[0184] In one embodiment, the second feature map determining module 1020 is further configured to:

[0185] sequentially performing multiple feature extractions on the first feature map using multiple convolutional layers to obtain a first sub-feature map;

[0186] Performing a fully connected operation on the first sub-feature map using a combination of a fully connected layer and an activation function to generate a target weight for each position of the first sub-feature map;

[0187] The first sub-feature map after the nonlinear transformation is updated based on the target weights of the respective positions to obtain the fifth intermediate feature map.

[0188] In one embodiment, the second feature map determining module 1020 is further configured to:

[0189] sequentially performing multiple feature extractions on the first feature map using multiple convolutional layers to obtain a second sub-feature map;

[0190] Performing channel feature extraction on the second sub-feature map to obtain a third sub-feature map;

[0191] The third sub-feature map is fused with the second sub-feature map to obtain the sixth intermediate feature map.

[0192] In one embodiment, the apparatus further comprises:

[0193] The fourth feature map determining module 1050 is configured to obtain the fourth feature map by:

[0194] Multiple decoders are used in sequence to continuously decode the second target image multiple times to obtain the fourth feature map, wherein the second target image is an image obtained by fusing the second feature map with the first feature map or an image obtained by fusing the second feature map with the fourth feature map output by the previous decoder, and the number of encoders is the same as the number of decoders.

[0195] In one embodiment, the fourth feature map determining module 1050 is specifically configured to:

[0196] For any decoder, perform the following steps:

[0197] Upsampling the second target image to obtain a global feature map of the fibroid biological tissue;

[0198] Normalizing the global feature map using a Sigmoid function to obtain a normalized global feature map;

[0199] Using a dilated convolutional layer to extract features from the normalized global feature map to obtain a local feature map;

[0200] The local feature map is normalized using a Sigmoid function to obtain the fourth feature map.

[0201] In one embodiment, the myoma-like biological tissue structure determination module 1040 is specifically configured to:

[0202] fusing the third feature map and the fourth feature map to obtain a mask image of the fibroid-like biological tissue, wherein the mask image is a binary image including the outline of the fibroid-like biological tissue;

[0203] The mask image is subjected to contour detection using a preset edge detection algorithm, and the structure of the myoma-like biological tissue is determined in the medical image.

[0204] After introducing a method and apparatus for segmenting tumor-like biological tissue according to an exemplary embodiment of the present disclosure, an electronic device according to another exemplary embodiment of the present disclosure is introduced next.

[0205] Those skilled in the art will appreciate that various aspects of the present disclosure may be implemented as systems, methods, or program products. Therefore, various aspects of the present disclosure may be implemented in the following forms: a complete hardware implementation, a complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software implementations, which may be collectively referred to herein as "circuits," "modules," or "systems."

[0206] In some possible implementations, the electronic device according to the present disclosure may include at least one processor and at least one computer storage medium. The computer storage medium stores program code, and when the program code is executed by the processor, the processor executes the steps of the method for segmenting fibroid tissue according to various exemplary embodiments of the present disclosure described above. For example, the processor may execute the following steps: Figure 2 Steps 201-204 shown in .

[0207] Refer to the following Figure 11 1100 according to this embodiment of the present disclosure will be described. Figure 11 The electronic device 1100 shown is merely an example and should not limit the functions and scope of use of the embodiments of the present disclosure.

[0208] like Figure 11 As shown, electronic device 1100 is a general electronic device. Components of electronic device 1100 may include, but are not limited to, at least one processor 1101, at least one computer storage medium 1102, and a bus 1103 connecting different system components (including computer storage medium 1102 and processor 1101).

[0209] Bus 1103 represents one or more of several types of bus structures, including a computer storage media bus or computer storage media controller, a peripheral bus, a processor, or a local bus using any of a variety of bus architectures.

[0210] Computer storage media 1102 may include readable media in the form of volatile computer storage media, such as random access computer storage media (RAM) 1121 and / or cache storage media 1122 , and may further include read-only computer storage media (ROM) 1123 .

[0211] The computer storage medium 1102 may also include a program / utility 1125 having a set (at least one) of program modules 1124, such program modules 1124 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0212] The electronic device 1100 may also communicate with one or more external devices 1104 (e.g., keyboards, pointing devices, etc.), one or more devices that enable a user to interact with the electronic device 1100, and / or any device that enables the electronic device 1100 to communicate with one or more other electronic devices (e.g., routers, modems, etc.). Such communication may be performed through an input / output (I / O) interface 1105. Furthermore, the electronic device 1100 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) through a network adapter 1106. Figure 11 As shown, the network adapter 1106 communicates with other modules for the electronic device 1100 via the bus 1103. It should be understood that, although not shown in the figure, other hardware and / or software modules can be used in conjunction with the electronic device 1100, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0213] In some possible embodiments, various aspects of the method for segmenting tumor-like biological tissue provided by the present disclosure can also be implemented in the form of a program product, which includes program code. When the program product is run on a computer device, the program code is used to enable the computer device to execute the steps of the method for segmenting tumor-like biological tissue according to various exemplary embodiments of the present disclosure described above in this specification.

[0214] The program product may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or component, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: an electrical connection having one or more wires, a portable disk, a hard disk, a random access computer storage medium (RAM), a read-only computer storage medium (ROM), an erasable programmable read-only computer storage medium (EPROM or flash memory), an optical fiber, a portable compact disk read-only computer storage medium (CD-ROM), an optical computer storage medium, a magnetic computer storage medium, or any suitable combination thereof.

[0215] The program product for segmenting fibroid-like biological tissue according to an embodiment of the present disclosure can be implemented as a portable compact disk read-only computer storage medium (CD-ROM) and include program code, and can be run on an electronic device. However, the program product of the present disclosure is not limited thereto. In this document, a readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0216] A readable signal medium may include a data signal transmitted in baseband or as part of a carrier wave, which carries readable program code. Such a transmitted data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium that can transmit, propagate, or transfer a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0217] The program code embodied on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0218] The program code for performing the operations of the present disclosure may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, and the like, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user electronic device, partially on the user device, as a stand-alone software package, partially on the user electronic device and partially on a remote electronic device, or entirely on a remote electronic device or server. In cases involving a remote electronic device, the remote electronic device may be connected to the user electronic device via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external electronic device (e.g., via the Internet using an Internet service provider).

[0219] It should be noted that although several modules of the device are mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules described above can be embodied in one module. Conversely, the features and functions of one module described above can be further divided and embodied by multiple modules.

[0220] Furthermore, although the operations of the disclosed method are described in a particular order in the accompanying drawings, this does not require or imply that the operations must be performed in this particular order, or that all illustrated operations must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps.

[0221] Those skilled in the art will appreciate that the embodiments of the present disclosure may be provided as methods, systems, or computer program products. Therefore, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present disclosure may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk computer storage media, CD-ROM, optical computer storage media, etc.) containing computer-usable program code.

[0222] The present disclosure is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the present disclosure. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0223] These computer program instructions may also be stored in a computer-readable computer storage medium that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable computer storage medium produce an article of manufacture including an instruction device that implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0224] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0225] Obviously, those skilled in the art may make various changes and modifications to the present disclosure without departing from the spirit and scope of the present disclosure. Thus, if these modifications and variations of the present disclosure fall within the scope of the claims of the present disclosure and their equivalents, the present disclosure is intended to include these modifications and variations.

Claims

1. A method for segmenting fibroid tissue, characterized in that: The method comprises: Multiple encoders are used in sequence to perform multiple feature extractions on the medical image to obtain each first feature map, wherein for any encoder, the following steps are performed: a hole convolution layer in the encoder is used to extract features from the first target image to obtain a first intermediate feature map, wherein the first target image is the medical image or the first intermediate feature map output by the previous encoder; the first intermediate feature map is nonlinearly transformed using a leaky rectified linear unit (Leaky Relu) function to obtain a second intermediate feature map; a hole convolution layer is used to extract features from the second intermediate feature map to obtain a third intermediate feature map; the third intermediate feature map is fused with the first target image to obtain a fourth intermediate feature map; after nonlinearly transforming the fourth intermediate feature map using the Leaky Relu function, a maximum pooling layer is used to compress the fourth intermediate feature map after nonlinear transformation to obtain the first feature map; the number of the first feature maps is the same as the number of the multiple encoders, and the image input to the latter encoder is the first feature map output by the previous encoder, and the number of first feature maps is the same as the number of feature extractions; For any first feature map, perform spatial transformation and channel feature extraction on the first feature map based on the spatial channel attention mechanism to obtain a second feature map; Upsampling each second feature map to obtain each upsampled feature map, and performing channel compression and feature extraction on each upsampled feature map at least once to obtain a third feature map, wherein the number of channel compressions for the same upsampled feature map is the same as the number of feature extractions thereof, and the number of channel compressions and feature extractions for different upsampled feature maps is different, and any upsampled feature map is first subjected to channel compression once and then to feature extraction once each time, and the image used for the latter channel compression is an image obtained by fusing the feature map obtained by the previous feature extraction and the upsampled feature map; The structure of the myoma-like biological tissue is obtained by using the third characteristic map and a fourth characteristic map obtained based on the second characteristic maps.

2. The method according to claim 1, characterized in that If feature extraction is performed on the medical image for the first time, the first target image is the medical image; and If this is not the first time that feature extraction is performed on the medical image, the first target image is the first intermediate feature map output by the previous encoder.

3. The method according to claim 1, characterized in that The performing spatial transformation and channel feature extraction on the first feature map based on the spatial channel attention mechanism to obtain a second feature map includes: Performing spatial transformation on the first feature map using the spatial attention mechanism in the spatial channel attention mechanism to obtain a fifth intermediate feature map; and Performing channel feature extraction on the first feature map through the channel attention mechanism in the spatial channel attention mechanism to obtain a sixth intermediate feature map; The fifth intermediate feature map and the sixth intermediate feature map are fused to obtain a seventh intermediate feature map, and feature extraction is performed on the seventh intermediate feature map to obtain the second feature map.

4. The method according to claim 3, characterized in that The utilizing the spatial attention mechanism in the spatial channel attention mechanism to perform spatial transformation on the first feature map to obtain a fifth intermediate feature map includes: sequentially performing multiple feature extractions on the first feature map using multiple convolutional layers to obtain a first sub-feature map; Performing a fully connected operation on the first sub-feature map using a combination of a fully connected layer and an activation function to generate a target weight for each position of the first sub-feature map; The first sub-feature map after the nonlinear transformation is updated based on the target weights of the respective positions to obtain the fifth intermediate feature map.

5. The method according to claim 3, characterized in that The performing channel feature extraction on the first feature map by the channel attention mechanism in the spatial channel attention mechanism to obtain a sixth intermediate feature map includes: sequentially performing multiple feature extractions on the first feature map using multiple convolutional layers to obtain a second sub-feature map; Performing channel feature extraction on the second sub-feature map to obtain a third sub-feature map; The third sub-feature map is fused with the second sub-feature map to obtain the sixth intermediate feature map.

6. The method according to claim 1, characterized in that The fourth characteristic map is obtained by: Multiple decoders are used in sequence to continuously decode the second target image multiple times to obtain the fourth feature map, wherein the second target image is an image obtained by fusing the second feature map with the first feature map or an image obtained by fusing the second feature map with the fourth feature map output by the previous decoder, and the number of the decoders is the same as the number of encoders.

7. The method according to claim 6, characterized in that The sequentially using multiple decoders to continuously decode the second target image multiple times to obtain the fourth feature map includes: For any decoder, perform the following steps: Upsampling the second target image to obtain a global feature map of the fibroid biological tissue; Normalizing the global feature map using a Sigmoid function to obtain a normalized global feature map; Using a dilated convolutional layer to extract features from the normalized global feature map to obtain a local feature map; The local feature map is normalized using a Sigmoid function to obtain the fourth feature map.

8. An electronic device, characterized in that: comprising a storage unit and a processor, wherein: The storage unit is configured to store medical images of myoma-like biological tissues; The processor is configured to: Multiple encoders are used in sequence to perform multiple feature extractions on the medical image to obtain first feature maps, wherein for any encoder, the following steps are performed: a hole convolution layer in the encoder is used to extract features from a first target image to obtain a first intermediate feature map, wherein the first target image is the medical image or the first intermediate feature map output by the previous encoder; a leaky rectified linear unit (Leaky Relu) function is used to perform nonlinear transformation on the first intermediate feature map to obtain a second intermediate feature map; a hole convolution layer is used to perform feature extraction on the second intermediate feature map to obtain a third intermediate feature map; the third intermediate feature map is fused with the first target image to obtain a fourth intermediate feature map; a Leaky Relu function is used to perform nonlinear transformation on the fourth intermediate feature map, and a maximum pooling layer is used to perform image compression on the nonlinearly transformed fourth intermediate feature map to obtain the first feature map; the number of the first feature maps is the same as the number of the multiple encoders, and the image input to the next encoder is the first feature map output by the previous encoder, and the number of first feature maps is the same as the number of feature extractions; For any first feature map, perform spatial transformation and channel feature extraction on the first feature map based on the spatial channel attention mechanism to obtain a second feature map; Upsampling each second feature map to obtain each upsampled feature map, and performing channel compression and feature extraction on each upsampled feature map at least once to obtain a third feature map, wherein the number of channel compressions for the same upsampled feature map is the same as the number of feature extractions thereof, and the number of channel compressions and feature extractions for different upsampled feature maps is different, and any upsampled feature map is first subjected to channel compression once and then to feature extraction once each time, and the image used for the latter channel compression is an image obtained by fusing the feature map obtained by the previous feature extraction and the upsampled feature map; The structure of the myoma-like biological tissue is obtained by using the third characteristic map and a fourth characteristic map obtained based on the second characteristic maps.

Images