Brain tumor image segmentation system based on Unet variant network

Abstract

The present invention discloses a brain tumor image segmentation system based on a Unet variant network, comprising: an acquisition module configured to acquire a brain tumor image to be segmented; a preprocessing module configured to preprocess the brain tumor image to be segmented; and an image segmentation module configured to segment the preprocessed brain tumor image using a trained Unet variant network to obtain segmented brain tumor regions. The system provides auxiliary diagnosis technology for doctors to predict, evaluate and treat brain tumors in patients.

Description

Technical Field

[0001] The invention relates to the technical field of image processing, and in particular to a brain tumor image segmentation system based on a Unet variant network. Background Art

[0002] The statements in this section merely mention background art related to the present invention and do not necessarily constitute prior art.

[0003] Gliomas are the most common primary malignant brain tumors, and their segmentation is a very useful application of lesion segmentation. However, manual brain tumor segmentation is a highly technical and time-consuming task. Therefore, automated brain tumor segmentation will be of great help in the development of this field. However, accurate segmentation of brain tumors remains a challenging task due to their high variations in size, shape, location, and inconsistent intensity and contrast in various image modalities. Summary of the invention

[0004] In order to address the deficiencies of the prior art, the present invention provides a brain tumor image segmentation system based on a Unet variant network, and provides auxiliary diagnosis technology for doctors to predict, evaluate and treat patients' brain tumors.

[0005] In a first aspect, the present invention provides a brain tumor image segmentation system based on a Unet variant network;

[0006] The brain tumor image segmentation system based on the Unet variant network includes:

[0007] An acquisition module is configured to: acquire a brain tumor image to be segmented;

[0008] A preprocessing module is configured to: preprocess the brain tumor image to be segmented;

[0009] The image segmentation module is configured to segment the preprocessed brain tumor image using a trained Unet variant network to obtain a segmented brain tumor area.

[0010] In a second aspect, the present invention further provides an electronic device, comprising:

[0011] a memory for non-transitory storage of computer-readable instructions; and

[0012] a processor for executing the computer readable instructions,

[0013] When the computer readable instructions are executed by the processor, the following steps are performed:

[0014] Acquire a brain tumor image to be segmented;

[0015] Preprocessing the brain tumor image to be segmented;

[0016] The preprocessed brain tumor image is segmented using the trained Unet variant network to obtain the segmented brain tumor area.

[0017] In a third aspect, the present invention further provides a storage medium that non-temporarily stores computer-readable instructions, wherein when the non-temporary computer-readable instructions are executed by a computer, the following steps are performed:

[0018] Acquire a brain tumor image to be segmented;

[0019] Preprocessing the brain tumor image to be segmented;

[0020] The preprocessed brain tumor image is segmented using the trained Unet variant network to obtain the segmented brain tumor area.

[0021] In a fourth aspect, the present invention further provides a computer program product, comprising a computer program, wherein the computer program is used to perform the following steps when running on one or more processors:

[0022] Acquire a brain tumor image to be segmented;

[0023] Preprocessing the brain tumor image to be segmented;

[0024] The preprocessed brain tumor image is segmented using the trained Unet variant network to obtain the segmented brain tumor area.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] The present invention proposes a simple to train two-dimensional Unet variant network, using bottleneck residual blocks to form a deeply stacked encoder and adding dropout to achieve good performance in lesion segmentation in brain MRIs. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0028] Figure 1 This is a functional module diagram of Embodiment 1;

[0029] Figure 2 This is a segmentation example diagram of the first embodiment;

[0030] Figure 3 This is a schematic diagram of the network structure of Embodiment 1;

[0031] FIG4(a) and FIG4(b) are bottleneck residual blocks and typical residual blocks of Embodiment 1;

[0032] Figure 5 This is a schematic diagram of the network structure of Embodiment 1;

[0033] Figure 6 This is a schematic diagram of the image segmentation result of Example 1. DETAILED DESCRIPTION

[0034] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0035] It should be noted that the terms used herein are only for describing specific embodiments, and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that the terms "include" and "have" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0036] In addition, in order to clearly describe the technical solutions of the embodiments of the present invention, in the embodiments of the present invention, the words "first", "second", etc. are used to distinguish the same items or similar items with substantially the same functions and effects. Those skilled in the art can understand that the words "first" and "second" do not limit the quantity and execution order, and the words "first", "second", etc. do not necessarily limit the difference.

[0037] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0038] In this embodiment, all data is obtained in compliance with laws and regulations and based on the user's consent, and is used legally.

[0039] Embodiment 1

[0040] This embodiment provides a brain tumor image segmentation system based on a Unet variant network;

[0041] like Figure 1 As shown, the brain tumor image segmentation system based on the Unet variant network includes:

[0042] An acquisition module is configured to: acquire a brain tumor image to be segmented;

[0043] A preprocessing module is configured to: preprocess the brain tumor image to be segmented;

[0044] The image segmentation module is configured to segment the preprocessed brain tumor image using a trained Unet variant network to obtain a segmented brain tumor area.

[0045] Furthermore, the preprocessing module comprises:

[0046] An image standardization unit is configured to: perform image standardization processing on the brain tumor image to be segmented;

[0047] The image conversion unit is configured to convert the image after the standardization process into a two-dimensional image.

[0048] Furthermore, the Unet variant network is an improvement on the original Unet network. It is assumed that the original Unet network includes an encoder, a bridge and a decoder. The Unet variant network replaces the convolutional layers of the encoder and the bridge with bottleneck residual modules, and replaces the convolutional layers of the decoder with typical residual modules.

[0049] Furthermore, the Unet variant network comprises: an encoder, a bridge and a decoder connected in sequence;

[0050] The encoder comprises: a first convolution unit a1, a first downsampling layer s1, a second convolution unit a2, a second downsampling layer s2, a third convolution unit a3, a third downsampling layer s3, a fourth convolution unit a4, a fourth downsampling layer s4, a fifth convolution unit a5 and a fifth downsampling layer s5 connected in sequence;

[0051] The bridge comprises: a sixth convolution unit a6 and a first upsampling layer x1 connected in sequence; wherein the convolution unit a6 is connected to the fifth downsampling layer s5, and the first upsampling layer x1 is connected to the first splicer p1;

[0052] The decoder includes: a first splicer p1, a first convolution unit f1, a second upsampling layer x2, a second splicer p2, a second convolution unit f2, a third upsampling layer x3, a third splicer p3, a third convolution unit f3, a fourth upsampling layer x4, a fourth splicer p4, a fourth convolution unit f4, a fifth upsampling layer x5, a fifth splicer p5 and a fifth convolution unit f5 connected in sequence; the fifth convolution unit f5 is connected to the output layer.

[0053] Further, the output end of the first convolution unit a1 is connected to the input end of the fifth splicer p5;

[0054] The output end of the second convolution unit a2 is connected to the input end of the fourth splicer p4;

[0055] The output end of the third convolution unit a3 is connected to the input end of the third splicer p3;

[0056] The output end of the fourth convolution unit a4 is connected to the input end of the second splicer p2;

[0057] The output end of the fifth convolution unit a5 is connected to the input end of the first splicer p1.

[0058] Furthermore, the first convolution unit a1 includes: two bottleneck residual modules connected in sequence, each bottleneck residual module includes a normalization layer, an activation function layer, a 1*1 convolution layer c1, a normalization layer, an activation function layer, a 3*3 convolution layer c2, a normalization layer, an activation function layer and a 1*1 convolution layer c3 connected in sequence; wherein, the convolution layer c3 of the second residual module is connected to the adder j1, the input end of the first convolution unit a1 is also connected to the adder j1, and the output end of the adder j1 is the output end of the first convolution unit a1.

[0059] Furthermore, the second convolution unit a2 includes: three bottleneck residual modules connected in sequence, each bottleneck residual module includes a normalization layer, an activation function layer, a 1*1 convolution layer c4, a normalization layer, an activation function layer, a 3*3 convolution layer c5, a normalization layer, an activation function layer, and a 1*1 convolution layer c6 connected in sequence; wherein, the convolution layer c6 of the third residual module is connected to the adder j2, the input end of the second convolution unit a2 is also connected to the adder j2, and the output end of the adder j2 is the output end of the second convolution unit a2.

[0060] Furthermore, the third convolution unit a3 includes: three bottleneck residual modules connected in sequence, each bottleneck residual module includes a normalization layer, an activation function layer, a 1*1 convolution layer c7, a normalization layer, an activation function layer, a 3*3 convolution layer c8, a normalization layer, an activation function layer, and a 1*1 convolution layer c9 connected in sequence; wherein, the convolution layer c9 of the third residual module is connected to the adder j3, the input end of the third convolution unit a3 is also connected to the adder j3, and the output end of the adder j3 is the output end of the third convolution unit a3.

[0061] Furthermore, the fourth convolution unit a4 includes: five bottleneck residual modules connected in sequence, each bottleneck residual module includes a normalization layer, an activation function layer, a 1*1 convolution layer c10, a normalization layer, an activation function layer, a 3*3 convolution layer c11, a normalization layer, an activation function layer, and a 1*1 convolution layer c12 connected in sequence; wherein, the convolution layer c12 of the fifth residual module is connected to the adder j4, the input end of the fourth convolution unit a4 is also connected to the adder j4, and the output end of the adder j4 is the output end of the fourth convolution unit a4.

[0062] Furthermore, the fifth convolution unit a5 includes: fourteen bottleneck residual modules connected in sequence, each bottleneck residual module includes a normalization layer, an activation function layer, a 1*1 convolution layer c13, a normalization layer, an activation function layer, a 3*3 convolution layer c14, a normalization layer, an activation function layer, and a 1*1 convolution layer c15 connected in sequence; wherein, the convolution layer c15 of the fourteenth residual module is connected to the adder j5, the input end of the fifth convolution unit a5 is also connected to the adder j5, and the output end of the adder j5 is the output end of the fifth convolution unit a5.

[0063] Furthermore, the sixth convolution unit a6 includes: four bottleneck residual modules connected in sequence, each bottleneck residual module includes a normalization layer, an activation function layer, a 1*1 convolution layer c16, a normalization layer, an activation function layer, a 3*3 convolution layer c17, a normalization layer, an activation function layer, and a 1*1 convolution layer c18 connected in sequence; wherein, the convolution layer c18 of the fourth residual module is connected to the adder j6, the input end of the sixth convolution unit a6 is also connected to the adder j6, and the output end of the adder j6 is the output end of the sixth convolution unit a6.

[0064] Furthermore, the first convolution unit f1 is implemented by a typical residual module; the typical residual module includes a normalization layer, an activation function layer, a 3*3 convolution layer h1, a normalization layer, an activation function layer, a 3*3 convolution layer h2, and an adder j7 connected in sequence; the input end of the adder j7 is also connected to the input end of the first convolution unit f1; the output end of the adder j7 serves as the output end of the first convolution unit f1.

[0065] Furthermore, the second convolution unit f2 is implemented by a typical residual module; the typical residual module includes a normalization layer, an activation function layer, a 3*3 convolution layer h3, a normalization layer, an activation function layer, a 3*3 convolution layer h4, and an adder j8 connected in sequence; the input end of the adder j8 is also connected to the input end of the second convolution unit f2; the output end of the adder j8 serves as the output end of the second convolution unit f2.

[0066] Furthermore, the third convolution unit f3 is implemented by a typical residual module; the typical residual module includes a normalization layer, an activation function layer, a 3*3 convolution layer h5, a normalization layer, an activation function layer, a 3*3 convolution layer h6, and an adder j9 connected in sequence; the input end of the adder j9 is also connected to the input end of the third convolution unit f3; the output end of the adder j9 serves as the output end of the third convolution unit f3.

[0067] Furthermore, the fourth convolution unit f4 is implemented by a typical residual module; the typical residual module includes a normalization layer, an activation function layer, a 3*3 convolution layer h7, a normalization layer, an activation function layer, a 3*3 convolution layer h8, and an adder j10 connected in sequence; the input end of the adder j10 is also connected to the input end of the fourth convolution unit f4; the output end of the adder j10 serves as the output end of the fourth convolution unit f4.

[0068] Furthermore, the fifth convolution unit f5 is implemented by a typical residual module; the typical residual module includes a normalization layer, an activation function layer, a 3*3 convolution layer h9, a normalization layer, an activation function layer, a 3*3 convolution layer h10, and an adder j11 connected in sequence; the input end of the adder j11 is also connected to the input end of the fifth convolution unit f5; the output end of the adder j11 serves as the output end of the fifth convolution unit f5.

[0069] Exemplarily, the network of the present invention is a variant network of Unet, and the basic architecture is as follows Figure 3 , as shown in Figure 4(a) and Figure 4(b). The detailed architecture description is as follows Figure 5 The network of the present invention is a typical U-shaped network consisting of three parts: encoder, decoder and bridge. In the encoding and decoding areas connected by the bridge part, five stacked residual blocks with convolutional layers and identify mapping are used, namely, bottleneck residual blocks and typical residual blocks, as shown in FIG. Figure 3 As shown in the figure, the number of bottleneck residual blocks is also marked. The feature representation is encoded in the encoding area and the pixel level is restored in the decoding area.

[0070] Because deep architectures have many layers, we need to consider the problems of overfitting and dead neurons in activation when training the network. The network of the present invention uses a small dropout rate (egdropout=0.5) for regularization after each layer. To help feature mapping, the input of each layer of the decoder is added with a cascade connection from the encoder output after upsampling. The network of the present invention downsamples with a step size of 2 in the first convolution of each layer. Upsampling is performed using a 2D transposed convolution with a step size of 2x2. The last layer uses a 1x1 convolution and generates classification scores for pixels to represent the three tumor classes and the background. The category of the pixel is determined by the maximum output channel (argmax), and softmax is used for training. The input of the network of the present invention is a two-dimensional image slice, in which the four channels represent each MRI modality.

[0071] Figures 4(a) and 4(b) show in detail the outlines of the Bottleneck Residual Block and ResidualBlock in the encoding and decoding parts. The network of the present invention adopts a Bottleneck design in the encoding and bridging parts, including a 1x1 convolution layer for dimensionality reduction, a 3x3 convolution layer, and a 1x1 convolution layer for dimensionality recovery. In the decoding part, a typical residual block is used, which consists of two stacked 3x3 convolutions. And batch normalization (BN) and rectified linear unit (ReLU) activation are used in all residual blocks.

[0072] Furthermore, the training process of the trained Unet variant network includes:

[0073] Constructing a training set; the training set is a brain tumor image with known brain tumor segmentation results;

[0074] The training set is input into the Unet variant network, and the Unet variant network is trained. When the loss function value no longer decreases, the training is stopped to obtain the trained Unet variant network.

[0075] Exemplarily, the training set uses the BraTS 2020 dataset, which is a dataset of the MICCAI brain tumor segmentation competition. The training set (TrainingData) has 369 cases and the validation set (ValidationData) has 125 cases. Each case contains 4 modalities (multimodal). The 4 modalities are t1, t2, flair, and t1ce. Multi-modal image information can complement each other and effectively improve the accuracy of segmentation; the use of different magnetic resonance imaging modes also relatively increases the difficulty of the brain tumor segmentation task. And all BraTS multimodal scans are provided in the form of NIfTI files (.nii.gz).

[0076] MR image attribute information: (240x240x155) An MR sequence has 155 images, and the size of each image is 240x240.

[0077] Each case in the BraTS 2020 dataset contains three segmented parts (labels), namely background, necrosis (NET, non-enhancing tumor), edema (ED, peritumoral edema), and enhancing tumor (ET, enhancing tumor). Figure 3 As shown in the figure, and the data set stipulates that the background (label 0), ① gangrene (label 1), ② edema area (label 2), ③ enhanced tumor area (label 4). These four labels can jointly reflect whole tumor (WT), enhanced tumor (ET), and tumor core (TC).

[0078] Right now

[0079] 1. WT=ED+ET+NET

[0080] 2.TC=ET+NET

[0081] 3.ET

[0082] When the Brats compete, they compete on the division of these three parts.

[0083] The dataset BraTS 2020 cannot be used directly. We need to perform some (more format) preprocessing on this dataset before training the model. With the help of the code, MRI images in nifti (.nii) format are normalized and converted into individual 2D png slices. The present invention uses data normalization to preprocess the images based on the entire 3D MRI modality, and normalizes them using equation (1), where v is a given voxel, and the mean (m) and standard deviation (SD) are for images with voxels>0. Therefore, the images have been rescaled before conversion to 2D images.

[0084]

[0085] Each slice of the image was saved as an independent png image with each channel representing one MRI modality. The only image augmentation applied was to randomly flip the image in left-right and front-back directions with a probability of 50%.

[0086] Furthermore, the preprocessed brain tumor image is segmented using the trained Unet variant network to obtain the segmented brain tumor area; the specific working principle includes:

[0087] The preprocessed brain tumor image is input into the trained Unet variant network. The feature representation is encoded in the encoder. The input of each layer of the decoder is added with a cascade connection from the encoder output after upsampling. The pixel level is decoded and restored in the decoder, and the brain tumor segmentation area is output.

[0088] The network of the present invention is trained using the ADAM optimizer with a learning rate of 1e-4 and He initialization. And batchsizes=10, epochs=50. The images of all subjects will be converted into a single RGBA image per slice and input into the model in a random order. Training and evaluation will be performed with different dropout rates. The network of the present invention is implemented in Kears using the tensorflow backend and executed using an NVIDIA GTX 2080Ti graphics card with 12GB of memory.

[0089] The network of the invention was evaluated on 125 cases of the validation set and preprocessed in the same way as the training set. The generated two-dimensional mask was reconverted into nifti (.nii) format with the help of the code and evaluated on the IPP website, which calculates and evaluates the lesion mask and outputs the indicators DSC, sensitivity (Sensitivity), specificity (Specificity) and hausdorff95 (HD95).

[0090] The evaluation criteria in image segmentation mainly use the Dice Similariy Coefficient (DSC). The Dice coefficient is a set similarity measurement indicator, which is usually used to calculate the similarity of two samples. The value range is 0-1. The best segmentation result is 1, and the worst is 0.

[0091]

[0092]

[0093]

[0094] Among them, TP is true positive, TN is true negative, FP is false positive, and FN is false negative.

[0095] Dice is more sensitive to the internal filling of the mask, while Hausdorff distance is more sensitive to the segmented boundaries.

[0096]

[0097] Table 1 Segmentation 2D masks examples and evaluation results

[0098] DSC Sensitivity Specificity ET 0.65818 0.70128 0.99954 WT 0.87378 0.90789 0.99865 TC 0.67305 0.65075 0.99971

[0099] Figure 2 This is a segmentation example diagram of the first embodiment; Figure 6 This is a schematic diagram of the segmentation result of Example 1.

[0100] Embodiment 2

[0101] This embodiment further provides an electronic device, including: one or more processors, one or more memories, and one or more computer programs; wherein the processor is connected to the memory, the one or more computer programs are stored in the memory, and when the electronic device is running, the processor executes the one or more computer programs stored in the memory to complete the following steps:

[0102] Acquire a brain tumor image to be segmented;

[0103] Preprocessing the brain tumor image to be segmented;

[0104] The preprocessed brain tumor image is segmented using the trained Unet variant network to obtain the segmented brain tumor area.

[0105] The specific step details of each step in the second embodiment are consistent with the step details of the first embodiment.

[0106] Embodiment 3

[0107] This embodiment further provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the following steps are performed:

[0108] Acquire a brain tumor image to be segmented;

[0109] Preprocessing the brain tumor image to be segmented;

[0110] The preprocessed brain tumor image is segmented using the trained Unet variant network to obtain the segmented brain tumor area.

[0111] The specific step details of each step in the third embodiment are consistent with the step details of the first embodiment.

[0112] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A brain tumor image segmentation system based on a Unet variant network, characterized by: include: An acquisition module is configured to: acquire a brain tumor image to be segmented; A preprocessing module is configured to: preprocess the brain tumor image to be segmented; An image segmentation module is configured to: segment the preprocessed brain tumor image using a trained Unet variant network to obtain a segmented brain tumor region; The Unet variant network is an improvement on the original Unet network, which includes an encoder, a bridge and a decoder. The Unet variant network replaces the convolutional layers of the encoder and the bridge with bottleneck residual modules, and replaces the convolutional layers of the decoder with typical residual modules. The Unet variant network comprises: an encoder, a bridge and a decoder connected in sequence; The encoder comprises: a first convolution unit a1, a first downsampling layer s1, a second convolution unit a2, a second downsampling layer s2, a third convolution unit a3, a third downsampling layer s3, a fourth convolution unit a4, a fourth downsampling layer s4, a fifth convolution unit a5 and a fifth downsampling layer s5 connected in sequence; The bridge comprises: a sixth convolution unit a6 and a first upsampling layer x1 connected in sequence; wherein the convolution unit a6 is connected to the fifth downsampling layer s5, and the first upsampling layer x1 is connected to the first splicer p1; The decoder comprises: a first splicer p1, a first convolution unit f1, a second upsampling layer x2, a second splicer p2, a second convolution unit f2, a third upsampling layer x3, a third splicer p3, a third convolution unit f3, a fourth upsampling layer x4, a fourth splicer p4, a fourth convolution unit f4, a fifth upsampling layer x5, a fifth splicer p5 and a fifth convolution unit f5 connected in sequence; the fifth convolution unit f5 is connected to the output layer; The output end of the first convolution unit a1 is connected to the input end of the fifth splicer p5; the output end of the second convolution unit a2 is connected to the input end of the fourth splicer p4; the output end of the third convolution unit a3 is connected to the input end of the third splicer p3; the output end of the fourth convolution unit a4 is connected to the input end of the second splicer p2; the output end of the fifth convolution unit a5 is connected to the input end of the first splicer p1; The first convolution unit a1 includes: two bottleneck residual modules connected in sequence, each bottleneck residual module includes a normalization layer, an activation function layer, a 1*1 convolution layer c1, a normalization layer, an activation function layer, a 3*3 convolution layer c2, a normalization layer, an activation function layer and a 1*1 convolution layer c3 connected in sequence; wherein the convolution layer c3 of the second residual module is connected to the adder j1, the input end of the first convolution unit a1 is also connected to the adder j1, and the output end of the adder j1 is the output end of the first convolution unit a1; The first convolution unit f1 is implemented by a typical residual module; the typical residual module includes a normalization layer, an activation function layer, a 3*3 convolution layer h1, a normalization layer, an activation function layer, a 3*3 convolution layer h2, and an adder j7 connected in sequence; the input end of the adder j7 is also connected to the input end of the first convolution unit f1; the output end of the adder j7 serves as the output end of the first convolution unit f1.

2. The brain tumor image segmentation system based on the Unet variant network as claimed in claim 1, characterized in that: The preprocessing module comprises: An image standardization unit is configured to: perform image standardization processing on the brain tumor image to be segmented; The image conversion unit is configured to convert the image after the standardization process into a two-dimensional image.

3. The brain tumor image segmentation system based on Unet variant network as claimed in claim 1, characterized in that: The preprocessed brain tumor image is segmented using a trained Unet variant network to obtain a segmented brain tumor region; The specific working principles include: The preprocessed brain tumor image is input into the trained Unet variant network. The feature representation is encoded in the encoder. The input of each layer of the decoder is added with a cascade connection from the encoder output after upsampling. The pixel level is decoded and restored in the decoder, and the brain tumor segmentation area is output.

4. An electronic device, comprising: a memory for non-transitory storage of computer readable instructions; as well as a processor for executing the computer readable instructions, When the computer readable instructions are executed by the processor, the following steps are performed: Acquire a brain tumor image to be segmented; Preprocessing the brain tumor image to be segmented; The preprocessed brain tumor image is segmented using the trained Unet variant network to obtain the segmented brain tumor area; The Unet variant network is an improvement on the original Unet network, which includes an encoder, a bridge and a decoder. The Unet variant network replaces the convolutional layers of the encoder and the bridge with bottleneck residual modules, and replaces the convolutional layers of the decoder with typical residual modules. The Unet variant network comprises: an encoder, a bridge and a decoder connected in sequence; The encoder comprises: a first convolution unit a1, a first downsampling layer s1, a second convolution unit a2, a second downsampling layer s2, a third convolution unit a3, a third downsampling layer s3, a fourth convolution unit a4, a fourth downsampling layer s4, a fifth convolution unit a5 and a fifth downsampling layer s5 connected in sequence; The bridge comprises: a sixth convolution unit a6 and a first upsampling layer x1 connected in sequence; wherein the convolution unit a6 is connected to the fifth downsampling layer s5, and the first upsampling layer x1 is connected to the first splicer p1; The decoder comprises: a first splicer p1, a first convolution unit f1, a second upsampling layer x2, a second splicer p2, a second convolution unit f2, a third upsampling layer x3, a third splicer p3, a third convolution unit f3, a fourth upsampling layer x4, a fourth splicer p4, a fourth convolution unit f4, a fifth upsampling layer x5, a fifth splicer p5 and a fifth convolution unit f5 connected in sequence; the fifth convolution unit f5 is connected to the output layer; The output end of the first convolution unit a1 is connected to the input end of the fifth splicer p5; the output end of the second convolution unit a2 is connected to the input end of the fourth splicer p4; the output end of the third convolution unit a3 is connected to the input end of the third splicer p3; the output end of the fourth convolution unit a4 is connected to the input end of the second splicer p2; the output end of the fifth convolution unit a5 is connected to the input end of the first splicer p1; The first convolution unit a1 includes: two bottleneck residual modules connected in sequence, each bottleneck residual module includes a normalization layer, an activation function layer, a 1*1 convolution layer c1, a normalization layer, an activation function layer, a 3*3 convolution layer c2, a normalization layer, an activation function layer and a 1*1 convolution layer c3 connected in sequence; wherein the convolution layer c3 of the second residual module is connected to the adder j1, the input end of the first convolution unit a1 is also connected to the adder j1, and the output end of the adder j1 is the output end of the first convolution unit a1; The first convolution unit f1 is implemented by a typical residual module; the typical residual module includes a normalization layer, an activation function layer, a 3*3 convolution layer h1, a normalization layer, an activation function layer, a 3*3 convolution layer h2, and an adder j7 connected in sequence; the input end of the adder j7 is also connected to the input end of the first convolution unit f1; the output end of the adder j7 serves as the output end of the first convolution unit f1.

5. A storage medium characterized by being non-transitory wherein when the non-transitory stored computer readable instructions are executed by a computer, the following steps are performed: Acquire a brain tumor image to be segmented; Preprocessing the brain tumor image to be segmented; The preprocessed brain tumor image is segmented using the trained Unet variant network to obtain the segmented brain tumor area; The Unet variant network is an improvement on the original Unet network, which includes an encoder, a bridge and a decoder. The Unet variant network replaces the convolutional layers of the encoder and the bridge with bottleneck residual modules, and replaces the convolutional layers of the decoder with typical residual modules. The Unet variant network comprises: an encoder, a bridge and a decoder connected in sequence; The encoder comprises: a first convolution unit a1, a first downsampling layer s1, a second convolution unit a2, a second downsampling layer s2, a third convolution unit a3, a third downsampling layer s3, a fourth convolution unit a4, a fourth downsampling layer s4, a fifth convolution unit a5 and a fifth downsampling layer s5 connected in sequence; The bridge comprises: a sixth convolution unit a6 and a first upsampling layer x1 connected in sequence; wherein the convolution unit a6 is connected to the fifth downsampling layer s5, and the first upsampling layer x1 is connected to the first splicer p1; The decoder comprises: a first splicer p1, a first convolution unit f1, a second upsampling layer x2, a second splicer p2, a second convolution unit f2, a third upsampling layer x3, a third splicer p3, a third convolution unit f3, a fourth upsampling layer x4, a fourth splicer p4, a fourth convolution unit f4, a fifth upsampling layer x5, a fifth splicer p5 and a fifth convolution unit f5 connected in sequence; the fifth convolution unit f5 is connected to the output layer; The output end of the first convolution unit a1 is connected to the input end of the fifth splicer p5; the output end of the second convolution unit a2 is connected to the input end of the fourth splicer p4; the output end of the third convolution unit a3 is connected to the input end of the third splicer p3; the output end of the fourth convolution unit a4 is connected to the input end of the second splicer p2; the output end of the fifth convolution unit a5 is connected to the input end of the first splicer p1; The first convolution unit a1 includes: two bottleneck residual modules connected in sequence, each bottleneck residual module includes a normalization layer, an activation function layer, a 1*1 convolution layer c1, a normalization layer, an activation function layer, a 3*3 convolution layer c2, a normalization layer, an activation function layer and a 1*1 convolution layer c3 connected in sequence; wherein the convolution layer c3 of the second residual module is connected to the adder j1, the input end of the first convolution unit a1 is also connected to the adder j1, and the output end of the adder j1 is the output end of the first convolution unit a1; The first convolution unit f1 is implemented by a typical residual module; the typical residual module includes a normalization layer, an activation function layer, a 3*3 convolution layer h1, a normalization layer, an activation function layer, a 3*3 convolution layer h2, and an adder j7 connected in sequence; the input end of the adder j7 is also connected to the input end of the first convolution unit f1; the output end of the adder j7 serves as the output end of the first convolution unit f1.

6. A computer program product, characterized in that The invention comprises a computer program which, when run on one or more processors, is used to perform the following steps: Acquire a brain tumor image to be segmented; Preprocessing the brain tumor image to be segmented; The preprocessed brain tumor image is segmented using the trained Unet variant network to obtain the segmented brain tumor area; The Unet variant network is an improvement on the original Unet network, which includes an encoder, a bridge and a decoder. The Unet variant network replaces the convolutional layers of the encoder and the bridge with bottleneck residual modules, and replaces the convolutional layers of the decoder with typical residual modules. The Unet variant network comprises: an encoder, a bridge and a decoder connected in sequence; The encoder comprises: a first convolution unit a1, a first downsampling layer s1, a second convolution unit a2, a second downsampling layer s2, a third convolution unit a3, a third downsampling layer s3, a fourth convolution unit a4, a fourth downsampling layer s4, a fifth convolution unit a5 and a fifth downsampling layer s5 connected in sequence; The bridge comprises: a sixth convolution unit a6 and a first upsampling layer x1 connected in sequence; wherein the convolution unit a6 is connected to the fifth downsampling layer s5, and the first upsampling layer x1 is connected to the first splicer p1; The decoder comprises: a first splicer p1, a first convolution unit f1, a second upsampling layer x2, a second splicer p2, a second convolution unit f2, a third upsampling layer x3, a third splicer p3, a third convolution unit f3, a fourth upsampling layer x4, a fourth splicer p4, a fourth convolution unit f4, a fifth upsampling layer x5, a fifth splicer p5 and a fifth convolution unit f5 connected in sequence; the fifth convolution unit f5 is connected to the output layer; The output end of the first convolution unit a1 is connected to the input end of the fifth splicer p5; the output end of the second convolution unit a2 is connected to the input end of the fourth splicer p4; the output end of the third convolution unit a3 is connected to the input end of the third splicer p3; the output end of the fourth convolution unit a4 is connected to the input end of the second splicer p2; the output end of the fifth convolution unit a5 is connected to the input end of the first splicer p1; The first convolution unit a1 includes: two bottleneck residual modules connected in sequence, each bottleneck residual module includes a normalization layer, an activation function layer, a 1*1 convolution layer c1, a normalization layer, an activation function layer, a 3*3 convolution layer c2, a normalization layer, an activation function layer and a 1*1 convolution layer c3 connected in sequence; wherein the convolution layer c3 of the second residual module is connected to the adder j1, the input end of the first convolution unit a1 is also connected to the adder j1, and the output end of the adder j1 is the output end of the first convolution unit a1; The first convolution unit f1 is implemented by a typical residual module; the typical residual module includes a normalization layer, an activation function layer, a 3*3 convolution layer h1, a normalization layer, an activation function layer, a 3*3 convolution layer h2, and an adder j7 connected in sequence; the input end of the adder j7 is also connected to the input end of the first convolution unit f1; the output end of the adder j7 serves as the output end of the first convolution unit f1.

Images