Method, device, equipment and storage medium for segmenting uterine fibroids in medical images

Abstract

The present invention belongs to the technical field of image segmentation and discloses a uterine fibroid segmentation method for medical images. The method comprises the following steps: collecting a plurality of medical images for preprocessing to obtain a plurality of sample images to form a sample data set; constructing a U-shaped network model; training the U-shaped network model using the sample data set to obtain a uterine fibroid image segmentation model with optimal model parameters; inputting a medical image to be tested into the uterine fibroid image segmentation model to obtain a tumor area corresponding to the uterine fibroid; wherein the U-shaped network model uses Unet as a backbone network, combines a dilated convolution and an attention mechanism, suppresses unimportant features by applying a channel attention mechanism between each layer of an encoder, and combines a dilated spatial convolution pooling pyramid with a convolution attention module at the bottom of the encoder, thereby improving the model's ability to recognize small-volume tumors, thereby improving segmentation accuracy and reliability.

Description

Technical Field

[0001] The present invention belongs to the technical field of image segmentation, and in particular relates to a method, device, equipment and storage medium for segmenting uterine fibroids in medical images. Background Art

[0002] Traditional treatment for uterine fibroids is mostly surgical resection, but surgical resection is very traumatic.

[0003] Nowadays, high-intensity focused ultrasound (HIFU) technology is used to treat uterine fibroids. HIFU ablation technology uses a certain focusing technology to focus low-energy ultrasound outside the body to a specific area in the body. With the help of the biological effects of ultrasound, the tissue produces irreversible coagulative necrosis without damaging the surrounding normal tissue, thereby achieving the purpose of ablating the tumor.

[0004] Compared with traditional surgical methods, HIFU ablation technology has the advantages of rapid recovery time, repeatable treatment, and uterine preservation, and is currently widely used worldwide. Achieving more accurate HIFU efficacy prediction before surgery is crucial for doctors to select treatment plans and improve treatment success rates. Magnetic resonance imaging (MRI), with its advantages of high soft tissue resolution and multidimensional imaging, is a key technology for diagnosing uterine fibroids and an effective means of predicting and evaluating the effectiveness of HIFU surgical treatment. MRI image segmentation of uterine fibroids and the outer contour of the uterus aids in preoperative planning, intraoperative navigation, and postoperative efficacy assessment, and is a necessary step in HIFU focused ultrasound therapy.

[0005] However, manually segmenting uterine fibroids from MRI images is a time-consuming, subjective task for doctors. Furthermore, the contrast between uterine fibroids and other tissues in MRI images is low, making it difficult to distinguish the boundaries between them. Furthermore, the number, morphology, and size of uterine fibroids vary greatly between patients. Therefore, accurate automatic segmentation methods for uterine fibroids and delineation of the outer contour of the uterus are crucial.

[0006] More and more deep learning models are being applied to medical image research. The most widely used technology is based on convolutional neural networks (CNNs). However, in practice, it is found that due to the indirect transmission of feature information, the features after transmission increase in accuracy, which may lead to inaccurate position information and inaccurate segmentation of lesion edges. There are often problems with identifying small tumors and poor boundary segmentation, resulting in low reliability of segmentation prediction results. Summary of the Invention

[0007] The purpose of the present invention is to provide a method and apparatus, device, and storage medium for uterine fibroid segmentation in medical images, which can improve the ability to recognize small tumors and thereby enhance the segmentation accuracy and reliability.

[0008] A first aspect of the present invention discloses a method for segmenting uterine fibroids in a medical image, comprising:

[0009] S1. Collect several medical images and perform preprocessing to obtain several sample images to form a sample data set;

[0010] S2. Construct a U-shaped network model; wherein the U-shaped network model uses Unet as the backbone network, including an encoder and a decoder. The network structures of the encoder and the decoder are respectively divided into five layers from top to bottom. The first four layers of the encoder are jump-connected to the corresponding layers of the decoder. The fifth layer of the encoder is connected to the fifth layer of the decoder through a series of void space convolution pooling pyramids and convolution attention modules. Each layer of the encoder includes a series of residual blocks and a first channel attention module. The residual block includes three series of convolution blocks. After each convolution block, a batch normalization layer and activation function, the three convolution blocks are the first convolution block, the second convolution block and the third convolution block, the first convolution block is used to add a specified number of channels, the second convolution block is used for downsampling, and the third convolution block is used to extract features without changing the number of channels. The first channel attention module is used to assign weights to the feature maps of multiple channels according to the weight information of different channels; the void spatial convolution pooling pyramid includes four stacked void convolution layers with expansion rates of 1, 6, 12, and 18 respectively, and the convolution attention module includes a second channel attention module and a spatial attention module connected in series;

[0011] S3. Using the sample data set to train the U-shaped network model, a uterine fibroid image segmentation model with optimal model parameters is obtained;

[0012] S4. Input the medical image to be tested into the uterine fibroid image segmentation model to obtain the tumor area corresponding to the uterine fibroid.

[0013] In some embodiments, the medical image to be tested is input into a uterine fibroid image segmentation model to obtain a tumor region corresponding to the uterine fibroid, including:

[0014] The medical image to be tested is input into the uterine fibroid image segmentation model, and the channels of the medical image to be tested are increased from 1 to 64, 128, 256, 512, and 1024 in sequence through the residual block of each layer of the encoder. The first channel attention module of each layer of the encoder assigns weights to the feature maps of multiple channels according to the weight information of different channels to obtain the encoded feature map;

[0015] When the number of channels reaches 1024, the encoded feature map is sampled at different sampling rates through the four dilated convolutional layers to obtain multi-scale features, and the multi-scale features are fused to obtain the final feature extraction result;

[0016] Inferring an attention map in the channel and space respectively through the second channel attention module and the spatial attention module, and multiplying the attention map with the feature extraction result after multi-scale feature fusion to obtain an adaptive feature map;

[0017] The adaptive feature map is decoded layer by layer by the decoder until the first layer of the decoder reduces the number of channels to 64. The result is mapped to probability through a 1×1 convolution operator followed by a sigmoid activation function layer to obtain the final segmentation result, which is marked as the tumor area corresponding to the uterine fibroids in the medical image to be tested.

[0018] In some embodiments, each layer of the decoder is provided with a dual convolution module; and the decoding of the adaptive feature map layer by layer by the decoder includes:

[0019] The input adaptive feature map is fused with the features of the corresponding layer of the jump-connected encoder through the double convolution module of each layer of the encoder, and then an upsampling operation is performed to restore the image size.

[0020] In some embodiments, the first convolution block performs a 1×1 convolution operation with a specified number of convolution kernels to increase a specified number of channels; the second convolution block performs a 3×3 convolution operation with a stride of 2 to downsample the feature map resolution; and the third convolution block performs a 1×1 convolution operation to perform feature extraction.

[0021] In some embodiments, a plurality of medical images are collected and pre-processed, including:

[0022] Perform a unified resizing operation on several medical images, adjust them to 256×256 pixels, and resave them as sample images in PNG format.

[0023] In some embodiments, the method further includes: dividing the sample dataset into a training set, a test set, and a validation set according to a specified ratio.

[0024] In some embodiments, the specified ratio is 8:1:1.

[0025] A second aspect of the present invention discloses a uterine fibroid segmentation device for medical images, comprising:

[0026] A preprocessing unit is used to collect a number of medical images for preprocessing to obtain a number of sample images to form a sample data set;

[0027] A construction unit for constructing a U-shaped network model; wherein the U-shaped network model uses Unet as the backbone network, including an encoder and a decoder, and the network structures of the encoder and the decoder are respectively divided into five layers from top to bottom. The first four layers of the encoder are jump-connected to the corresponding layers of the decoder, and the fifth layer of the encoder is connected to the fifth layer of the decoder through a series of hollow spatial convolution pooling pyramids and convolution attention modules; each layer of the encoder includes a series of residual blocks and a first channel attention module, and the residual block includes three series of convolution blocks, and a batch normalization is provided after each convolution block. Layer and activation function, the three convolution blocks are the first convolution block, the second convolution block and the third convolution block, the first convolution block is used to increase the specified number of channels, the second convolution block is used for downsampling, and the third convolution block is used to extract features without changing the number of channels. The first channel attention module is used to assign weights to the feature maps of multiple channels according to the weight information of different channels; the void spatial convolution pooling pyramid includes four stacked void convolution layers with expansion rates of 1, 6, 12, and 18 respectively, and the convolution attention module includes a second channel attention module and a spatial attention module connected in series;

[0028] A training unit, used for training a U-shaped network model using a sample data set to obtain a uterine fibroid image segmentation model with optimal model parameters;

[0029] The prediction unit is used to input the medical image to be tested into the uterine fibroid image segmentation model to obtain the tumor area corresponding to the uterine fibroid.

[0030] The third aspect of the present invention discloses an electronic device, comprising a memory storing executable program code and a processor coupled to the memory; the processor calls the executable program code stored in the memory to execute the uterine fibroid segmentation method of medical images disclosed in the first aspect.

[0031] A fourth aspect of the present invention discloses a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the uterine fibroid segmentation method for medical images disclosed in the first aspect.

[0032] The beneficial effect of the present invention lies in proposing a U-shaped network that combines void convolution and attention mechanism. By applying the channel attention mechanism between each layer of the encoder to suppress unimportant features, and combining the void spatial convolution pooling pyramid with the convolution attention module at the bottom of the encoder, the model's ability to recognize small-volume tumors can be improved, thereby improving the segmentation accuracy and reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The accompanying drawings herein illustrate specific examples of the technical solutions described in the present invention, and together with the specific implementation methods constitute a part of the specification, and are used to explain the technical solutions, principles and effects of the present invention.

[0034] Unless otherwise specified or defined, the same reference numerals in different drawings represent the same or similar technical features, and the same or similar technical features may also be represented by different reference numerals.

[0035] Figure 1 This is a flow chart of a method for segmenting uterine fibroids in a medical image disclosed in an embodiment of the present invention;

[0036] Figure 2 Schematic diagram of the network structure of the U-shaped network model disclosed in the embodiment of the present invention;

[0037] Figure 3 It is a schematic diagram of a partial structure of an encoder of a U-shaped network model disclosed in an embodiment of the present invention;

[0038] Figure 4 It is a schematic diagram of a partial structure of an encoder of a U-shaped network model disclosed in an embodiment of the present invention;

[0039] Figure 5 Schematic diagram of the structure of the dilated spatial convolutional pooling pyramid and the convolutional attention module disclosed in an embodiment of the present invention;

[0040] Figure 6 1 is a schematic structural diagram of a uterine fibroid segmentation device for medical images disclosed in an embodiment of the present invention;

[0041] Figure 7 It is a structural diagram of an electronic device disclosed in an embodiment of the present invention.

[0042] Description of reference numerals:

[0043] 601, pre-processing unit; 602, construction unit; 603, training unit; 604, prediction unit; 701, memory; 702, processor. DETAILED DESCRIPTION

[0044] To facilitate understanding of the present invention, specific embodiments of the present invention will be described in more detail below with reference to the accompanying drawings.

[0045] Unless otherwise specified or defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art. In the context of combining the technical solution of the present invention with realistic scenarios, all technical and scientific terms used herein may also have meanings corresponding to the purpose of implementing the technical solution of the present invention. "First, second..." used herein is merely used to distinguish names and does not represent a specific quantity or order. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0046] It should be noted that when an element is considered to be "fixed to" another element, it can be directly fixed to the other element or there can be an intermediate element; when an element is considered to be "connected to" another element, it can be directly connected to the other element or there can be an intermediate element; when an element is considered to be "mounted on" another element, it can be directly mounted on the other element or there can be an intermediate element. When an element is considered to be "located on" another element, it can be directly located on the other element or there can be an intermediate element.

[0047] Unless otherwise specified or defined, the “said” and “the” used in this document refer to the technical features or technical contents mentioned or described before the corresponding position, and the technical features or technical contents may be the same as or similar to the technical features or technical contents mentioned therein. In addition, the terms “including” and “having” and any variations thereof used in this document are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or devices.

[0048] The present invention discloses a method for uterine fibroid segmentation in medical images, which can be implemented through computer programming. The method can be performed by an electronic device such as a computer, laptop, or tablet computer, or by a uterine fibroid segmentation device embedded in the electronic device, although this invention is not limited thereto.

[0049] like Figure 1 As shown, the uterine fibroid segmentation method of medical images includes the following steps S1 to S4:

[0050] S1. Collect several medical images and perform preprocessing to obtain several sample images to form a sample data set.

[0051] In an embodiment of the present invention, MRI images of 70 patients with uterine fibroids are collected, wherein the MRI images are specifically T2-weighted or T2-weighted medical images. For example, 1582 T2-weighted MRI images are obtained and marked as medical images to be preprocessed.

[0052] The preprocessing process includes: performing a uniform resizing operation on the medical images, adjusting them to 256×256 pixels, and then resaving them as sample images in PNG format. At the same time, in order to protect the patient's sensitive information, the re-saved sample images are anonymized. The sample data set is divided into a training set, a test set, and a validation set according to a specified ratio. For example, if the specified ratio is set to 8:1:1, 8 / 10 of the sample images are determined as the training set, 1 / 10 of the sample images are determined as the test set, and 1 / 10 of the sample images are determined as the validation set. Furthermore, in order to expand the sample set, the sample images in the training set and the validation set can be randomly transformed through operations such as rotation, flipping, cropping, contrast adjustment, and saturation adjustment to derive more sample images.

[0053] S2. Construct a U-shaped network model.

[0054] like Figure 2 As shown in the figure, the U-shaped network model uses Unet as the backbone network, including an encoder and a decoder. The encoder acts as a downsampling path to expand the receptive field while reducing computational cost, while the decoder acts as an upsampling path to restore the resolution lost in the downsampling path. The entire network structure is divided into five layers from top to bottom. That is, the network structures of the encoder and decoder are divided into five layers from top to bottom. Each of the first four layers of the encoder has a jump connection with the corresponding layer of the decoder. The fifth layer of the encoder and the fifth layer of the decoder are connected via a series of atrous spatial convolutional pyramid pooling (ASPP) and convolutional block attention module (CBAM).

[0055] In the present invention, in addition to the traditional 2×2 maximum pooling between the layers of the encoder, residual connections combined with the channel attention mechanism are also used to ensure feature extraction capabilities. Figure 3As shown in the figure, each layer of the encoder consists of a series of residual blocks and a first channel attention (Squeeze-and-Excitation, SE) module. The residual block consists of three series of convolutional blocks, each followed by a batch normalization (BN) layer and a Relu activation function. The three convolutional blocks are the first, second, and third convolutional blocks. The first convolutional block is used to increase the specified number of channels by performing a 1×1 convolution operation with a specified number of convolution kernels to obtain feature maps of multiple channels. For example, the first convolutional block of the residual block in the first layer of the encoder is used to change the number of channels from 1 to 64, and therefore includes 64 convolution kernels. Similarly, the first convolutional blocks of the residual blocks in other layers use convolution operations to convert the channels to 128, 256, 512, and finally to 1024. The second convolutional block is used for downsampling, specifically performing a 3×3 convolution operation with a stride of 2 to downsample the feature map resolution to preserve information and reduce computational cost. The third convolution block is used to extract features without changing the number of channels. Specifically, a 1×1 convolution operation is performed for feature extraction.

[0056] The first-channel attention module is used to assign weights to multi-channel feature maps based on the weight information of different channels, thereby obtaining more accurate encoded feature maps. Introducing the residual module into the encoder improves the model's feature extraction capabilities. Adding the first-channel attention module after each layer's residual block assigns weights to different channels, thereby suppressing less reliable features.

[0057] like Figure 4 As shown, the atrous spatial convolution pooling pyramid ASPP includes four stacked atrous convolution layers with expansion rates (also known as sampling rates) of 1, 6, 12, and 18, respectively. The encoded feature map input by the first channel attention module of the last layer of the encoder is sampled at different sampling rates through the four atrous convolution layers with different sampling rates. The sampling results are then added together to expand the number of channels, that is, the obtained multi-scale features are fused to obtain the final feature extraction result. The convolutional attention module CBAM is placed after ASPP. CBAM includes a second channel attention module and a spatial attention module connected in series. The fused multi-channel features are input into CBAM. The attention map is inferred in the channel and space respectively through the second channel attention module and the spatial attention module. The attention map is then multiplied by the input multi-scale features to perform adaptive feature optimization to obtain an adaptive feature map, which enables the model to more accurately locate the lesion position and improve the model segmentation accuracy.

[0058] The decoder then performs upsampling on the adaptive feature maps with the smallest size, performing an upsampling operation for decoding. Features in the downsampling path are connected to those in the upsampling path via skip connections to provide added information without abstracting the downsampling information. Each decoder layer incorporates a dual convolutional module, which first fuses the skip-connected encoder features, then uses upsampling to restore the image size. Finally, the number of channels is reduced to 64, and the results are mapped to probabilities using a 1×1 convolution operator followed by a sigmoid activation layer to produce the final segmentation result.

[0059] As an optional implementation, given an intermediate feature map F∈R C×H×W As input, where R is a matrix, C is the number of channels, H and W represent the height and width of the image. First, average pooling and maximum pooling operations are used to aggregate the spatial information of the feature map to generate two different spatial context descriptors: F c avg and F c max In order to generate the channel attention map M c (f) will pass through a shared network consisting of a multi-layer perceptron (MLP) and a hidden layer, where the feature map will be compressed with a compression rate of r. The attention calculation formulas of the first channel attention module and the second channel attention module are shown in the following formula (1):

[0060] M c (f)=σ(MLP(AvgPool / (f))+MLP(MaxPool / (f)))

[0061] =σ(W1(W0(F c avg ))+W1(W0(F c max ))) (1)

[0062] Where σ is the sigmoid activation function, W0∈R C / r × C , W1∈R C × C / r ; W0 and W1 are the weights of the multi-layer perceptron, sharing the input and RELU activation function of W0.

[0063] As an optional implementation, two pooling operations are used to aggregate the channel information of the function map to generate two 2D maps: F s avg ∈R 1×HxW and F s max ∈R 1×HxWrepresents average pooling and maximum pooling. Then the attention calculation formula of the spatial attention module is as follows (2):

[0064] M s (f)=σ(f 7×7 (AvgPool / (f);MaxPool / (F)])))

[0065] =σ(f 7×7 (F s avg ; F s max ])) (2)

[0066] Where σ is the sigmoid activation function, f 7×7 It is a 7×7 convolution.

[0067] S3. Use the sample data set to train the U-shaped network model to obtain a uterine fibroid image segmentation model with optimal model parameters.

[0068] The loss function is used to evaluate the degree of inconsistency between the model's predicted value and the true value. The lower the loss function during the training process of the network model, the closer the predicted value is to the true value, and the better the robustness of the network model. In the training process of the network model of the embodiment of the present invention, the cross entropy loss function (BinaryCross Entropy) is used. This loss function uses a binary classification model with an output dimension of 1 to calculate the loss value, that is, the loss value. BCEWithLogitsLoss is defined as:

[0069] BCE_Loss={l1,…,l N},l n =-[y n ·logσ(x n )+(1-y n )·log(1-σ(x n ))] (3)

[0071] Among them, σ is the sigmoid activation function, which can map x to the interval (0,1):

[0072]

[0073] The optimizer used in this embodiment of the present invention is a stochastic gradient descent (SGD) optimizer with an initial momentum of 0.9. Early stopping is used during network model training to prevent overfitting. The parameters of the trained uterine tumor image segmentation model that achieve the highest accuracy on the validation set are saved as the optimal model parameters.

[0074] Different quantitative metrics are used to comprehensively evaluate the segmentation performance compared with other existing segmentation methods. These metrics include Dice similarity coefficient, Jaccard coefficient (JC), Hausdorff distance (HD), average surface distance (ASD), and recall.

[0075] Among them, the Dice coefficient is one of the most commonly used medical segmentation evaluation indicators, as shown in the following formula (5):

[0076]

[0077] Where A and B represent the gold standard and predicted image respectively.

[0078] Another way to express the Dice coefficient is as shown in the following formula (6):

[0079]

[0080] Where TP represents the true value when the uterine fibroid pixel in the gold standard is correctly classified in the predicted image, FN misclassifies the uterine fibroid pixel into the non-uterine fibroid pixel in the predicted image, and FP means that the non-uterine fibroid pixel in the gold standard is incorrectly labeled as the uterine fibroid pixel in the predicted image.

[0081] The Jaccard coefficient (JC) is used to compare the differences and similarities between sample sets. The higher the Jaccard coefficient, the more similar the two samples are. The Jaccard coefficient is calculated as shown in the following formula (7):

[0082]

[0083] Where HD is the distance between two subsets in the metric space, which is calculated as follows:

[0084] H(A,B)=max{h(A,B),h(B,A)} (8)

[0086] ASD represents the average surface distance between the segmentation result and the gold standard, and is calculated as follows:

[0087]

[0088] Recall describes how many true positive examples in the test set are picked up by the binary classifier from the perspective of real results, that is, how many true positive examples are recalled by the binary classifier. Recall is calculated as shown in the following formula (10):

[0089]

[0090] The saved model parameters are used to automatically predict the tumor area on the uterine tumor test set, and the segmentation results of other existing models on the test set are as follows: Figure 5 The quantitative results are shown in Table 1 below.

[0091] Table 1 Quantitative comparison of segmentation results of the test set

[0092]

[0093] S4. Input the medical image to be tested into the uterine fibroid image segmentation model to obtain the tumor area corresponding to the uterine fibroid.

[0094] In practical applications, the unknown medical image to be tested is input into the uterine fibroid image segmentation model to predict the tumor area corresponding to the uterine fibroid. Specifically, step S4 may include the following steps S41 to S44 (not shown):

[0095] S41, inputting the medical image to be tested into the uterine fibroid image segmentation model, sequentially increasing the channels of the medical image to be tested from 1 to 64, 128, 256, 512, and 1024 through the residual block of each layer of the encoder, and assigning weights to the feature maps of the multiple channels according to the weight information of different channels through the first channel attention module of each layer of the encoder to obtain an encoded feature map;

[0096] S42. When the number of channels reaches 1024, the encoded feature map is sampled at different sampling rates through the four dilated convolutional layers to obtain multi-scale features, and the multi-scale features are fused to obtain the final feature extraction result;

[0097] S43, inferring an attention map in the channel and space respectively through the second channel attention module and the spatial attention module, multiplying the attention map with the feature extraction result after multi-scale feature fusion to obtain an adaptive feature map;

[0098] S44. The adaptive feature map is decoded layer by layer by the decoder until the first layer of the decoder reduces the number of channels to 64, and the result is mapped to probability through a 1×1 convolution operator followed by a sigmoid activation function layer to obtain a final segmentation result, which is marked as the tumor area corresponding to the uterine fibroids in the medical image to be tested.

[0099] Furthermore, decoding the adaptive feature map layer by layer by the decoder may include:

[0100] The input adaptive feature map is fused with the features of the corresponding layer of the jump-connected encoder through the double convolution module of each layer of the encoder, and then an upsampling operation is performed to restore the image size.

[0101] In summary, in view of the characteristics of individual uterine fibroids, such as variable shapes, large differences in size, unknown number, and low contrast between adjacent organs, the embodiments of the present invention not only use residual connections in the encoder to improve the model's ability to extract features, but also combine the spatial attention mechanism with the channel attention mechanism, and connect CBAM with ASPP at the bottom layer of the encoder, thereby improving the model's ability to extract features, thereby improving the ability to recognize small tumors, and further improving the segmentation accuracy and reliability.

[0102] like Figure 6 As shown, the embodiment of the present invention discloses a uterine fibroid segmentation device for medical images, comprising a pre-processing unit 601, a construction unit 602, a training unit 603, and a prediction unit 604, wherein:

[0103] A preprocessing unit 601 is used to collect a number of medical images for preprocessing to obtain a number of sample images to form a sample data set;

[0104] A construction unit 602 is used to construct a U-shaped network model; wherein the U-shaped network model uses Unet as a backbone network, including an encoder and a decoder, and the network structures of the encoder and the decoder are respectively divided into five layers from top to bottom. The first four layers of the encoder are jump-connected to the corresponding layers of the decoder, and the fifth layer of the encoder is connected to the fifth layer of the decoder through a series of hollow spatial convolution pooling pyramids and convolution attention modules; each layer of the encoder includes a series of residual blocks and a first channel attention module, and the residual block includes three series of convolution blocks, and a batch normalization module is provided after each convolution block. convolution layer and activation function, the three convolution blocks are the first convolution block, the second convolution block and the third convolution block, the first convolution block is used to increase the specified number of channels, the second convolution block is used for downsampling, and the third convolution block is used to extract features without changing the number of channels. The first channel attention module is used to assign weights to the feature maps of multiple channels according to the weight information of different channels; the void spatial convolution pooling pyramid includes four stacked void convolution layers with expansion rates of 1, 6, 12, and 18 respectively, and the convolution attention module includes a second channel attention module and a spatial attention module connected in series;

[0105] A training unit 603 is configured to train a U-shaped network model using a sample data set to obtain a uterine fibroid image segmentation model having optimal model parameters;

[0106] The prediction unit 604 is configured to input the medical image to be tested into the uterine fibroid image segmentation model to obtain a tumor region corresponding to the uterine fibroid.

[0107] like Figure 7 As shown, an embodiment of the present invention discloses an electronic device, including a memory 701 storing executable program code and a processor 702 coupled to the memory 701;

[0108] The processor 702 calls the executable program code stored in the memory 701 to execute the uterine fibroid segmentation method of the medical image described in the above embodiments.

[0109] An embodiment of the present invention further discloses a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the uterine fibroid segmentation method of a medical image described in the above embodiments.

[0110] The purpose of the above embodiments is to exemplify and deduce the technical solution of the present invention, and to fully describe the technical solution, purpose and effect of the present invention. Its purpose is to enable the public to have a more thorough and comprehensive understanding of the disclosed content of the present invention, and it does not limit the scope of protection of the present invention.

[0111] The above embodiments are not exhaustive and may include many other embodiments not listed above. Any replacements and improvements made without violating the concept of the present invention are within the scope of protection of the present invention.

Claims

1. A method for segmenting uterine fibroids in medical images, characterized in that: include: S1. Collect several medical images and perform preprocessing to obtain several sample images to form a sample data set; S2. Construct a U-shaped network model; wherein the U-shaped network model uses Unet as the backbone network, including an encoder and a decoder. The network structures of the encoder and the decoder are respectively divided into five layers from top to bottom. The first four layers of the encoder are jump-connected to the corresponding layers of the decoder. The fifth layer of the encoder is connected to the fifth layer of the decoder through a series of void space convolution pooling pyramids and convolution attention modules. Each layer of the encoder includes a series of residual blocks and a first channel attention module. The residual block includes three series of convolution blocks. After each convolution block, a batch normalization layer and activation function, the three convolution blocks are the first convolution block, the second convolution block and the third convolution block, the first convolution block is used to add a specified number of channels, the second convolution block is used for downsampling, and the third convolution block is used to extract features without changing the number of channels. The first channel attention module is used to assign weights to the feature maps of multiple channels according to the weight information of different channels; the void spatial convolution pooling pyramid includes four stacked void convolution layers with expansion rates of 1, 6, 12, and 18 respectively, and the convolution attention module includes a second channel attention module and a spatial attention module connected in series; S3. Using the sample data set to train the U-shaped network model, a uterine fibroid image segmentation model with optimal model parameters is obtained; S4. Input the medical image to be tested into the uterine fibroid image segmentation model to obtain the tumor area corresponding to the uterine fibroid.

2. The method for segmenting uterine fibroids in medical images according to claim 1, wherein: The medical image to be tested is input into the uterine fibroid image segmentation model to obtain the tumor area corresponding to the uterine fibroid, including: The medical image to be tested is input into the uterine fibroid image segmentation model, and the channels of the medical image to be tested are increased from 1 to 64, 128, 256, 512, and 1024 in sequence through the residual block of each layer of the encoder. The first channel attention module of each layer of the encoder assigns weights to the feature maps of multiple channels according to the weight information of different channels to obtain the encoded feature map; When the number of channels reaches 1024, the encoded feature map is sampled at different sampling rates through the four dilated convolutional layers to obtain multi-scale features, and the multi-scale features are fused to obtain the final feature extraction result; Inferring an attention map in the channel and space respectively through the second channel attention module and the spatial attention module, and multiplying the attention map with the feature extraction result after multi-scale feature fusion to obtain an adaptive feature map; The adaptive feature map is decoded layer by layer by the decoder until the first layer of the decoder reduces the number of channels to 64. The result is mapped to probability through a 1×1 convolution operator followed by a sigmoid activation function layer to obtain the final segmentation result, which is marked as the tumor area corresponding to the uterine fibroids in the medical image to be tested.

3. The method for segmenting uterine fibroids in medical images according to claim 2, wherein: Each layer of the decoder is provided with a double convolution module; the adaptive feature map is decoded layer by layer by the decoder, comprising: The input adaptive feature map is fused with the features of the corresponding layer of the jump-connected encoder through the double convolution module of each layer of the encoder, and then an upsampling operation is performed to restore the image size.

4. The method for segmenting uterine fibroids in medical images according to claim 1, wherein: The first convolution block performs a 1×1 convolution operation with a specified number of convolution kernels to add a specified number of channels; The second convolution block downsamples the feature map resolution by performing a 3×3 convolution operation with a stride of 2; The third convolution block performs a 1×1 convolution operation to perform feature extraction.

5. The method for segmenting uterine fibroids in medical images according to any one of claims 1 to 4, wherein: Collect several medical images for preprocessing, including: Perform a unified resizing operation on several medical images, adjust them to 256×256 pixels, and resave them as sample images in PNG format.

6. The method for segmenting uterine fibroids in medical images according to any one of claims 1 to 4, wherein: The method further comprises: The sample data set is divided into a training set, a test set, and a validation set according to a specified ratio.

7. The method for segmenting uterine fibroids in medical images according to claim 6, wherein: The specified ratio is 8:1:

1.

8. A uterine fibroid segmentation device for medical images, characterized in that: include: A preprocessing unit is used to collect a number of medical images for preprocessing to obtain a number of sample images to form a sample data set; A construction unit for constructing a U-shaped network model; wherein the U-shaped network model uses Unet as the backbone network, including an encoder and a decoder, and the network structures of the encoder and the decoder are respectively divided into five layers from top to bottom. The first four layers of the encoder are jump-connected to the corresponding layers of the decoder, and the fifth layer of the encoder is connected to the fifth layer of the decoder through a series of hollow spatial convolution pooling pyramids and convolution attention modules; each layer of the encoder includes a series of residual blocks and a first channel attention module, and the residual block includes three series of convolution blocks, and a batch normalization is provided after each convolution block. Layer and activation function, the three convolution blocks are the first convolution block, the second convolution block and the third convolution block, the first convolution block is used to increase the specified number of channels, the second convolution block is used for downsampling, and the third convolution block is used to extract features without changing the number of channels. The first channel attention module is used to assign weights to the feature maps of multiple channels according to the weight information of different channels; the void spatial convolution pooling pyramid includes four stacked void convolution layers with expansion rates of 1, 6, 12, and 18 respectively, and the convolution attention module includes a second channel attention module and a spatial attention module connected in series; A training unit, used for training a U-shaped network model using a sample data set to obtain a uterine fibroid image segmentation model with optimal model parameters; The prediction unit is used to input the medical image to be tested into the uterine fibroid image segmentation model to obtain the tumor area corresponding to the uterine fibroid.

9. An electronic device, characterized in that It comprises a memory storing executable program code and a processor coupled to the memory; the processor calls the executable program code stored in the memory to execute the uterine fibroid segmentation method of medical images according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, wherein the computer program enables a computer to execute the method for segmenting uterine fibroids in a medical image according to any one of claims 1 to 7.

Images