Method for processing prostate cancer in multi-parametric magnetic resonance images and related device

Abstract

A kind of processing method of prostate cancer in multi-parameter magnetic resonance image and related equipment.Compared with prior art, the processing method of prostate cancer in multi-parameter magnetic resonance image and related equipment provided by the present application can crop, intensity normalize and register the images of different image sequences in multi-parameter magnetic resonance image sequence;To solve the problem of feature extraction and fusion of different image sequences in multi-parameter magnetic resonance image sequence, a convolution attention module is used to extract several different level feature maps of different sequence images, and an attention mechanism is used to fuse the extracted feature maps and segmentation mask;Attention mechanism is used to realize feature extraction and feature fusion, and the pre-set feature extraction network model can automatically learn suitable feature extraction and feature fusion parameters during training process;Using Retina U-Net architecture, the tasks of detection, segmentation and classification can be completed simultaneously.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of medical image segmentation, and in particular to a processing method for prostate cancer in multi-parameter magnetic resonance images and related equipment. BACKGROUND

[0002] Prostate cancer (PCa) is a malignant tumor that occurs in the prostate of men. MRI technology can provide images of different contrasts (i.e. modalities), which is a non-invasive and well-performing soft tissue contrast imaging method. MRI images can provide information about the shape, size, and location of organs and lesions, and play a key role in disease analysis and diagnosis. The Prostate Imaging-Reporting and Data System (PI-RADS) is a structured reporting scheme for multi-parameter prostate MRI to assess suspected prostate cancer in untreated prostates.

[0003] Compared with single-modality medical images, multi-modality medical images can provide more information about the lesion area and its surrounding area from multiple perspectives, and display the characteristics of the lesion from different angles, which is an important means for patients to diagnose tumors in recent years. However, the detection, segmentation, and diagnosis of lesions in multi-modality medical images require high technical skills from doctors and are time-consuming and labor-intensive. Using deep learning technology to automatically detect, segment, and diagnose lesions in multi-modality medical images can reduce the workload of doctors and speed up the diagnosis, which is a research hotspot in the field of medical image segmentation in recent years. However, there are difficulties in feature extraction and fusion of multi-modality medical images in the prior art, and there is a lack of models that can simultaneously perform detection, segmentation, and diagnosis.

[0004] Therefore, there is an urgent need for a new medical image processing method to automatically detect, segment, and classify lesions, and to provide assistance to doctors. SUMMARY

[0005] The present application aims to provide a processing method for prostate cancer in multi-parameter magnetic resonance images and related equipment that can efficiently extract and fuse features of multi-modality medical images, and simultaneously perform automatic detection, segmentation, and classification.

[0006] In a first aspect, the present application provides a processing method for prostate cancer in multi-parameter magnetic resonance images, comprising the following steps:

[0007] obtaining a multi-parameter magnetic resonance image sequence containing a prostate region, the multi-parameter magnetic resonance image sequence including an apparent diffusion coefficient image sequence, a diffusion weighted image sequence, and a T2 weighted image sequence;

[0008] preprocessing the multi-parameter magnetic resonance image sequence, the preprocessing comprising cropping images of different image sequences in the multi-parameter magnetic resonance image sequence into the same size and intensity normalization, and then registering the apparent diffusion coefficient image sequence and the diffusion weighted image sequence to the T2 weighted image sequence;

[0009] extracting different level feature maps in the apparent diffusion coefficient image sequence, the diffusion weighted image sequence and the T2 weighted image sequence based on a preset feature extraction network model, respectively, to obtain apparent diffusion coefficient feature maps, diffusion weighted feature maps and T2 weighted feature maps;

[0010] taking the T2 weighted image sequence as an input of the preset feature extraction network model, and obtaining prostate segmentation masks, central zone gland segmentation masks and peripheral zone segmentation masks through processing; concatenating the apparent diffusion coefficient feature maps, the diffusion weighted feature maps, the T2 weighted feature maps, the prostate segmentation masks, the central zone gland segmentation masks and the peripheral zone segmentation masks to form concatenated feature maps;

[0011] processing the concatenated feature maps through an attention mechanism to obtain fusion feature maps;

[0012] performing evaluation, classification, lesion detection and lesion segmentation on the fusion feature maps through a preset detection network architecture to obtain final detection results.

[0013] Preferably, the preset feature extraction network model comprises a convolutional attention module, a first semantic segmentation network and a second semantic segmentation network, and the convolutional attention module comprises a channel attention module and a spatial attention module.

[0014] Preferably, the convolutional attention module satisfies the following relationship:

[0015]

[0016] wherein, represents element-wise multiplication, F" represents the final refined output, M C represents channel attention, M S represents spatial attention, F represents the input of channel attention, and F' represents the output of channel attention.

[0017] Preferably, the channel attention module satisfies the following relationship:

[0018]

[0019] wherein, σ represents a sigmoid activation function, represents W0 is a C / r x C matrix, wherein W1 is a CxC / r matrix, AvgPool denotes an average pooling layer, MaxPool denotes a max pooling layer, and MLP denotes a multi-layer perceptron, denotes the calculation result of AvgPool(F), denotes the calculation result of MaxPool(F).

[0020] Preferably, the spatial attention module satisfies the following relationship:

[0021]

[0022] wherein f 7×7 denotes a convolution with a kernel size of 7x7.

[0023] Preferably, the T2-weighted image sequence is taken as an input of the preset feature extraction network model, and a processing procedure for obtaining the prostate segmentation mask, the central zone gland segmentation mask, and the peripheral zone segmentation mask includes:

[0024] taking the T2-weighted image sequence as an input of the first semantic segmentation network in the preset feature extraction network model to obtain the prostate segmentation mask;

[0025] taking the T2-weighted image sequence and the prostate segmentation mask as inputs of the second semantic segmentation network in the preset feature extraction network model to obtain the central zone gland segmentation mask;

[0026] subtracting the central zone gland segmentation mask from the prostate segmentation mask to obtain the peripheral zone segmentation mask.

[0027] Preferably, the attention mechanism adopts ECANet.

[0028] Preferably, the preset detection network architecture is Retina U-Net.

[0029] In a second aspect, the present application further provides a computer device, comprising a memory, a processor, and a multi-parameter magnetic resonance image prostate cancer processing program stored in the memory and executable on the processor, wherein the processor implements the steps of the multi-parameter magnetic resonance image prostate cancer processing method as described in any one of the above embodiments when executing the multi-parameter magnetic resonance image prostate cancer processing program.

[0030] In a third aspect, the present application also provides a computer readable storage medium, which stores a processing program for prostate cancer in multi-parameter magnetic resonance images, and the processing program for prostate cancer in multi-parameter magnetic resonance images, when executed by a processor, implements the steps in the processing method for prostate cancer in multi-parameter magnetic resonance images.

[0031] Compared with the prior art, the processing method for prostate cancer in multi-parameter magnetic resonance images and the related device provided by the present application can crop, intensity normalize and register images in different image sequences in a multi-parameter magnetic resonance image sequence; in view of the difficulty in feature extraction and fusion of different image sequences in the multi-parameter magnetic resonance image sequence, a convolution attention module is used to extract a plurality of different level feature maps of different sequence images, and an ECANet attention mechanism is used to fuse the extracted feature maps and segmentation masks; the prostate region segmentation mask, the central zone gland segmentation mask and the peripheral zone segmentation mask are integrated into the feature fusion module to improve the subsequent classification ability, detection ability and segmentation ability of prostate cancer; the attention mechanism is used to realize feature extraction and feature fusion, and the preset feature extraction network model can automatically learn suitable feature extraction and feature fusion parameters in the training process; the Retina U-Net architecture is used to complete the tasks of full-automatic detection, segmentation and classification at the same time, improve the inference speed, and multiple tasks share one architecture, the related tasks complement each other through shared information, and the apparent of each other is improved. BRIEF DESCRIPTION OF DRAWINGS

[0032] The above and other aspects of the present application will become more apparent and more readily appreciated from the following detailed description, taken in conjunction with the accompanying drawings, in which:

[0033] Figure 1 is a flowchart of a processing method for prostate cancer in multi-parameter magnetic resonance images provided by an embodiment of the present application;

[0034] Figure 2 is a technical solution flowchart of a processing method for prostate cancer in multi-parameter magnetic resonance images provided by an embodiment of the present application;

[0035] Figure 3 is a convolution attention module structure diagram of a processing method for prostate cancer in multi-parameter magnetic resonance images provided by an embodiment of the present application;

[0036] Figure 4 is a channel attention module structure diagram of a processing method for prostate cancer in multi-parameter magnetic resonance images provided by an embodiment of the present application;

[0037] Figure 5is a spatial attention module structure diagram of a prostate cancer processing method in a multi-parameter magnetic resonance image provided by an embodiment of the application.

[0038] Figure 6 is an ECANet structure diagram of a prostate cancer processing method in a multi-parameter magnetic resonance image provided by an embodiment of the application.

[0039] Figure 7 is a Retina U-Net structure diagram of a prostate cancer processing method in a multi-parameter magnetic resonance image provided by an embodiment of the application.

[0040] Figure 8 is a schematic diagram of a computer device for processing a prostate cancer in a multi-parameter magnetic resonance image provided by an embodiment of the application. DETAILED DESCRIPTION

[0041] The specific embodiments / embodiments described herein are specific embodiments of the application, used to illustrate the concept of the application, and are explanatory and exemplary, and should not be interpreted as limiting the embodiments of the application and the scope of the application. In addition to the embodiments described herein, those skilled in the art can also employ other technical solutions based on the content disclosed in the claims and the specification of the present application, which include technical solutions that make any obvious substitutions and modifications to the embodiments described herein, and these technical solutions are within the protection scope of the application.

[0042] The specific embodiments of the application will be described in detail below with reference to the accompanying drawings.

[0043] (Embodiment One)

[0044] Please refer to Figures 1-7 The application provides a prostate cancer processing method in a multi-parameter magnetic resonance image, which comprises the following steps:

[0045] S101, a multi-parameter magnetic resonance image sequence containing a prostate region is acquired, and the multi-parameter magnetic resonance image sequence comprises an apparent diffusion coefficient image sequence, a diffusion weighted image sequence and a T2 weighted image sequence.

[0046] In the embodiment of the present application, the multiparametric magnetic resonance image mpMRI sequence includes an apparent diffusion coefficient image ADC sequence, a diffusion weighted image DWI sequence, and a T2 weighted image T2W sequence. Specifically, the apparent diffusion coefficient image ADC sequence is used to describe the speed and range of molecular diffusion motion in different directions in the diffusion weighted image sequence; the diffusion weighted image DWI sequence can reflect the diffusion motion and the degree of restriction of water molecules in tissues and lesions; and the T2 weighted image T2W sequence can clearly show the position and size of the lesion.

[0047] S102, pre-processing the multiparametric magnetic resonance image sequence, the pre-processing including cutting images of different image sequences in the multiparametric magnetic resonance image sequence into the same size and performing intensity normalization, and then registering the apparent diffusion coefficient image sequence and the diffusion weighted image sequence to the T2 weighted image sequence.

[0048] In the embodiment of the present application, all images in the multiparametric magnetic resonance image are cropped into a prostate peripheral region with a size of 160x160x24 voxels and a spacing of (0.5, 0.5, 3) mm, wherein third-order B-spline interpolation is used for all image interpolation; the intensity of each channel of the cropped image is normalized; and then the apparent diffusion coefficient image sequence and the diffusion weighted image sequence are registered to the T2 weighted image sequence. Specifically, non-rigid registration (based on B-spline transformation) is performed between the spatial gradient of the T2 weighted image sequence and the apparent diffusion coefficient image sequence using the Python library SimpleTK, Mattes Mutual Information is used as the loss function, and gradient descent is used as the optimization of B-spline parameters.

[0049] S103, extracting different level feature maps in the apparent diffusion coefficient image sequence, the diffusion weighted image sequence, and the T2 weighted image sequence based on a preset feature extraction network model, respectively, to obtain an apparent diffusion coefficient feature map, a diffusion weighted feature map, and a T2 weighted feature map.

[0050] In the embodiments of the present application, specifically, the preset feature extraction network model adopts CBAM (Convolutional Block Attention Module), which is a lightweight convolutional attention module. The convolutional attention module CBAM includes two sub-modules, a channel attention module CAM (Channel Attention Module) and a spatial attention module SAM (Spartial Attention Module), which respectively perform attention mechanisms on channels and spaces. The input features will first pass through a channel attention module to obtain a weighted result, and then pass through a spatial attention module, and finally obtain a result after weighting. The overall attention mechanism can be summarized as:

[0051]

[0052] wherein, represents element-wise multiplication, F" represents the final refined output, M C represents channel attention, M S represents spatial attention, F represents the input of channel attention, and F' represents the output of channel attention.

[0053] The channel attention module focuses on meaningful information in the input features. The above feature map is input through two parallel maximum pooling layers and average pooling layers, so that the size of the above feature map is changed from CxHxW to Cx1x1, and then a Share MLP module is passed. In this module, it first compresses the channel number to 1 / r (Reduction, reduction rate) times of the original, and then expands to the original channel number, and obtains two activated results through the ReLU activation function. Element-wise addition is performed on the two output results, and then a sigmoid activation function is used to obtain the output result of the channel attention module. The output result is multiplied by the original image to change the size back to CxHxW. The channel attention module satisfies the following relationship:

[0054]

[0055] wherein, σ represents a sigmoid activation function, W0 is a C / rx C matrix, W1 is a CxC / r matrix, AvgPool represents an average pooling layer, MaxPool represents a maximum pooling layer, and MLP represents a multi-layer perceptron, represents the calculation result of AvgPool (F), represents the calculation result of MaxPool (F).

[0056] The spatial attention module focuses on the position information of the target. The output result of the channel attention module is obtained through maximum pooling and average pooling to obtain two 1xHxW feature maps, then the two feature maps are spliced through a Concat operation, and are converted into a 1-channel feature map through a 7x7 convolution, and then a sigmoid function is used to obtain a spatial attention feature map. Finally, the output result is multiplied by the original image to return to the size of CxHxW to obtain the apparent diffusion coefficient feature map, the diffusion weighted feature map and the weighted feature map. The spatial attention module satisfies the following relationship:

[0057]

[0058] Wherein, f 7×7 represents a convolution with a kernel size of 7x7.

[0059] S104, the T2 weighted image sequence is taken as the input of the preset feature extraction network model, and the prostate segmentation mask, the central zone gland segmentation mask and the peripheral zone segmentation mask are obtained through processing; the apparent diffusion coefficient feature map, the diffusion weighted feature map, the T2 weighted image sequence, the prostate segmentation mask, the central zone gland segmentation mask and the peripheral zone segmentation mask are connected in series to form a series feature map;

[0060] In the embodiment of the application, while extracting different level feature maps in the image sequence in step S103, based on the first semantic segmentation network U-Net pre-trained in the preset feature extraction network model, the T2 weighted image sequence is taken as the input of the first semantic segmentation network U-Net to obtain the prostate segmentation mask; then based on the second semantic segmentation network U-Net pre-trained in the preset feature extraction network model, the T2 weighted image sequence and the prostate segmentation mask are taken as the input of the second semantic segmentation network U-Net to obtain the central zone gland segmentation mask (CG segmentation mask); then the central zone gland segmentation mask is subtracted from the prostate segmentation mask to obtain the peripheral zone segmentation mask (PZ segmentation mask). Finally, the series feature map is formed by connecting the apparent diffusion coefficient feature map, the diffusion weighted feature map, the T2 weighted image sequence, the prostate segmentation mask, the central zone gland segmentation mask and the peripheral zone segmentation mask. The prostate region segmentation mask, the central zone gland segmentation mask and the peripheral zone segmentation mask are connected in series, which improves the subsequent step of the prostate cancer classification ability, detection ability and segmentation ability.

[0061] S105, the series feature map is processed through an attention mechanism to obtain a fusion feature map;

[0062] In the embodiment of the present application, the attention mechanism adopts ECANet (efficient channel attention). The ECANet is a channel attention mechanism. After global average pooling is performed on the input feature map, the size of the feature map is changed from CxHxW to Cx1x1, and then an adaptive one-bit convolution kernel size is calculated and applied to one-dimensional convolution to obtain the weight of each channel of the feature map. Then, the normalized weight and the original input feature map are multiplied channel by channel to generate a weighted fusion feature map.

[0063] In S106, the fusion feature map is evaluated, classified, lesion detected and lesion segmented by a preset detection network architecture to obtain a final detection result.

[0064] In the embodiment of the present application, the evaluation classification adopts the Prostate Imaging-Reporting and Data System (PI-RADS), which is a structured reporting scheme for evaluating suspected prostate cancer in untreated prostate. The preset detection network architecture is Retina U-Net, which combines the Retina Net detector with the U-Net segmentation network. Among them, the Retina Net is a simple one-stage detection network based on FPN. As shown in Figure 7 P3-P6. The pyramid level Pj represents the feature map of the jth decoder level, where j increases with the decrease of the resolution. Since the existence of small objects in medical images is taken into account, the pyramid level of the subnetwork operation in the Retina U-Net architecture is shifted to P2-P5. In addition, two high-resolution pyramid levels are added to the FPN in the Retina U-Net, thereby generating a final segmentation layer, making the expanded FPN architecture very similar to the U-Net. Therefore, the segmentation of the lesion is independent of the detection, which greatly simplifies the structure.

[0065] Compared with the prior art, the processing method of prostate cancer in multi-parameter magnetic resonance images and the related equipment provided by the application can crop, intensity normalize and register images of different image sequences in a multi-parameter magnetic resonance image sequence; in view of the difficulty in feature extraction and fusion of different image sequences in the multi-parameter magnetic resonance image sequence, a convolution attention module is used to extract a plurality of different level feature maps of different sequence images, and an ECANet attention mechanism is used to fuse the extracted feature maps and a segmentation mask; the prostate region segmentation mask, the central zone gland segmentation mask and the peripheral zone segmentation mask are integrated into the feature fusion module to improve the subsequent classification ability, detection ability and segmentation ability of prostate cancer; the attention mechanism is used to realize feature extraction and feature fusion, and the preset feature extraction network model can automatically learn suitable feature extraction and feature fusion parameters in the training process; the Retina U-Net architecture is used to complete the tasks of full-automatic detection, segmentation and classification at the same time, improve the inference speed, and multiple tasks share one architecture, the related tasks complement each other by sharing information, and improve the performance of each other.

[0066] (Embodiment Two)

[0067] The embodiment of the application further provides a computer device, please refer to Figure 8 , Figure 8 is a structural schematic diagram of the computer device provided by the embodiment of the application, the computer device 200 comprises a memory 202, a processor 201 and a computer program stored on the memory 202 and capable of running on the processor 201.

[0068] The processor 201 invokes the computer program stored in the memory 202 to execute the steps in the processing method of prostate cancer in multi-parameter magnetic resonance images provided by the embodiment of the application, please combine Figure 1 , specifically comprising the following steps:

[0069] S101, a multi-parameter magnetic resonance image sequence containing a prostate region is acquired, and the multi-parameter magnetic resonance image sequence comprises an apparent diffusion coefficient image sequence, a diffusion weighted image sequence and a T2 weighted image sequence;

[0070] S102, the multi-parameter magnetic resonance image sequence is preprocessed, and the preprocessing comprises cropping images in different image sequences in the multi-parameter magnetic resonance image sequence into the same size and performing intensity normalization, and then registering the apparent diffusion coefficient image sequence and the diffusion weighted image sequence to the T2 weighted image sequence;

[0071] S103, different hierarchical feature maps in the apparent diffusion coefficient image sequence, the diffusion weighted image sequence and the T2 weighted image sequence are extracted based on a preset feature extraction network model, to obtain an apparent diffusion coefficient feature map, a diffusion weighted feature map and a T2 weighted feature map respectively;

[0072] S104, the T2 weighted image sequence is taken as an input of the preset feature extraction network model, and processed to obtain a prostate segmentation mask, a central zone gland segmentation mask and a peripheral zone segmentation mask, and the apparent diffusion coefficient feature map, the diffusion weighted feature map, the T2 weighted feature map, the prostate segmentation mask, the central zone gland segmentation mask and the peripheral zone segmentation mask are connected in series to form a series connection feature map;

[0073] S105, the series connection feature map is processed through an attention mechanism to obtain a fusion feature map;

[0074] S106, the fusion feature map is evaluated, classified, lesion detected and lesion segmented through a preset detection network architecture to obtain a final detection result.

[0075] The computer device 200 provided by the embodiment of the present application can realize the steps in the processing method of prostate cancer in multi-parameter magnetic resonance images as in the above embodiment, and can realize the same technical effects, and the description in the above embodiment is referred to, and will not be repeated here.

[0076] (Embodiment three)

[0077] The embodiment of the present application further provides a computer readable storage medium, and the computer readable storage medium stores a processing program of prostate cancer in multi-parameter magnetic resonance images, the processing program of prostate cancer in multi-parameter magnetic resonance images is executed by a processor to realize each process and step in the processing method of prostate cancer in multi-parameter magnetic resonance images provided by the embodiment of the present application, and can realize the same technical effects, and to avoid repetition, which will not be repeated here.

[0078] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by a computer program instructing related hardware, and the program can be stored in a computer readable storage medium, and the program can include the processes of the above-mentioned embodiment when executed. The storage medium can be a magnetic disc, an optical disc, a read-only memory (ROM) or a random access memory (RAM).

[0079] It should be noted that, in this document, the terms "comprises", "comprising", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a" does not, without more constraints, exclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0080] Those skilled in the art can clearly understand the above-mentioned embodiment method can be realized by means of software and necessary general hardware platform, of course, also can be realized by hardware, but in many cases, the former is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of software product, and the computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), including a plurality of instructions to make a terminal (may be mobile phone, computer, server, air conditioner, or network equipment, etc.) execute the method described in various embodiments of the present application.

[0081] The embodiments of the present application are described above in conjunction with the accompanying drawings, the disclosed is only the preferred embodiment of the present application, but the present application is not limited to the above specific embodiments, the above specific embodiments are only illustrative, but not restrictive, those skilled in the art can make many equivalent changes without departing from the purpose of the present application and the scope of the claims, all of which are within the protection of the present application.

Claims

1. A method for processing prostate cancer in multi-parameter magnetic resonance images, characterized in that: The following steps are involved: Acquiring a multi-parameter magnetic resonance image sequence including a prostate region, wherein the multi-parameter magnetic resonance image sequence includes an apparent diffusion coefficient image sequence, a diffusion-weighted image sequence, and a T2-weighted image sequence; Preprocessing the multi-parametric magnetic resonance image sequence, the preprocessing comprising cropping images of different image sequences in the multi-parametric magnetic resonance image sequence into the same size and performing intensity normalization, and then registering the apparent diffusion coefficient image sequence and the diffusion-weighted image sequence to the T2-weighted image sequence; Extracting feature maps of different levels from the apparent diffusion coefficient image sequence, the diffusion-weighted image sequence, and the T2-weighted image sequence based on a preset feature extraction network model, respectively, to obtain an apparent diffusion coefficient feature map, a diffusion-weighted feature map, and a T2-weighted feature map; The T2-weighted image sequence is used as input of the preset feature extraction network model to obtain a prostate segmentation mask, a central zone gland segmentation mask, and a peripheral zone segmentation mask after processing; the apparent diffusion coefficient feature map, the diffusion weighted feature map, the T2-weighted feature map, the prostate segmentation mask, the central zone gland segmentation mask, and the peripheral zone segmentation mask are serially connected to form a serial feature map; Processing the concatenated feature maps through an attention mechanism to obtain a fused feature map; The fused feature map is simultaneously evaluated and classified, lesion detected, and lesion segmented through a preset detection network architecture to obtain the final detection result.

2. The method for processing prostate cancer in multi-parameter magnetic resonance images according to claim 1, wherein: The preset feature extraction network model includes a convolutional attention module, a first semantic segmentation network and a second semantic segmentation network, and the convolutional attention module includes a channel attention module and a spatial attention module.

3. The method for processing prostate cancer in multi-parameter magnetic resonance images according to claim 2, wherein: The convolutional attention module satisfies the following relationship: in, represents element-by-element multiplication, F″ represents the final refined output, and M c represents channel attention, M s represents spatial attention, F represents the input of channel attention, and F' represents the output of channel attention.

4. The method for processing prostate cancer in multi-parameter magnetic resonance images according to claim 3, wherein: The channel attention module satisfies the following relationship: Among them, σ represents the sigmoid activation function, Indicates that W0 is a C / r×C matrix, Indicates that W1 is a C×C / r matrix, AvgPool represents the average pooling layer, MaxPool represents the maximum pooling layer, and MLP represents the multi-layer perceptron. represents the calculation result of AvgPool(F), Indicates the calculation result of MaxPool(F).

5. The method for processing prostate cancer in multi-parameter magnetic resonance images according to claim 4, wherein: The spatial attention module satisfies the following relationship: Among them, f 7×7 Represents a convolution with a kernel size of 7×7.

6. The method for processing prostate cancer in multi-parameter magnetic resonance images according to claim 2, wherein: The T2-weighted image sequence is used as the input of the preset feature extraction network model, and the processing process of obtaining the prostate segmentation mask, the central zone gland segmentation mask, and the peripheral zone segmentation mask includes: Using the T2-weighted image sequence as input to the first semantic segmentation network in the preset feature extraction network model to obtain the prostate segmentation mask; Using the T2-weighted image sequence and the prostate segmentation mask as inputs to the second semantic segmentation network in the preset feature extraction network model to obtain a central zone gland segmentation mask; The peripheral band segmentation mask is calculated by subtracting the central band gland segmentation mask from the prostate segmentation mask.

7. The method for processing prostate cancer in multi-parameter magnetic resonance images according to claim 1, wherein: The attention mechanism adopts ECANet.

8. The method for processing prostate cancer in multi-parameter magnetic resonance images according to claim 1, wherein: The preset detection network architecture is Retina U-Net.

9. A computer device, characterized in that: include: A memory, a processor, and a processing program for prostate cancer in multi-parameter magnetic resonance images stored in the memory and executable on the processor, wherein when the processor executes the processing program for prostate cancer in multi-parameter magnetic resonance images, the steps of the method for processing prostate cancer in multi-parameter magnetic resonance images as claimed in any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a processing program for prostate cancer in multi-parameter magnetic resonance images, and when the processing program for prostate cancer in multi-parameter magnetic resonance images is executed by a processor, the steps of the processing method for prostate cancer in multi-parameter magnetic resonance images as described in any one of claims 1 to 8 are implemented.

Images