A double-sequence-based method, device, medium and product for predicting muscle layer invasion of bladder cancer
By extracting and fusing features from dual-sequence MRI images, the problems of misdiagnosis and uncertainty in the prediction of bladder cancer muscle layer invasion were solved, achieving higher prediction accuracy and generalization ability, while reducing data acquisition and safety risks.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHENZHEN UNIV
- Filing Date
- 2026-03-03
- Publication Date
- 2026-07-14
AI Technical Summary
Existing methods for predicting bladder cancer muscle invasion have a history of misdiagnosis, especially when the VI-RADS score is 2 or 3, where the uncertainty is relatively high. Furthermore, traditional methods involve large amounts of data collection, high annotation complexity, and safety risks.
Feature extraction was performed using dual-sequence MRI images (T2WI and DWI sequences). First and second image features were obtained through a feature extractor, and feature fusion was performed using a dual-sequence attention module. Combining anatomical structure and functional micro-features, the muscle layer invasion category of bladder cancer was predicted.
It improved the accuracy of predicting bladder cancer muscle layer invasion, reduced the complexity of data collection and annotation, avoided the safety risks associated with gadolinium contrast agents, and enhanced the model's generalization ability.
Smart Images
Figure CN122391061A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of biomedical engineering technology, and in particular to a method, device, medium and product for predicting bladder cancer muscle layer invasion based on dual sequences. Background Technology
[0002] Currently, the common method for clinical staging of bladder cancer is to obtain tumor and muscle layer tissue and combine it with intraoperative transurethral endoscopy for evaluation to complete pathological staging. However, research results show that misdiagnosis frequently occurs based on specimens from transurethral resection of bladder tumor (TURBT), misclassifying muscle-invasive bladder cancer (MIBC) as non-muscle-invasive bladder cancer (NMIBC), leading to an underestimation of bladder cancer staging. Furthermore, TURBT carries the potential risk of serious complications such as bladder perforation.
[0003] In recent years, MRI image sequences have developed rapidly, helping radiologists reduce misdiagnosis of bladder cancer during transurethral resection of bladder tumor (TURBT) and improve the accuracy of predicting bladder cancer muscle invasion. However, the existing VI-RADS scoring method for bladder cancer assessment based on MRI image sequences shows uncertainty in its prediction of subsets of 2 and 3 points.
[0004] Therefore, existing technologies still need to be improved and enhanced. Summary of the Invention
[0005] The technical problem to be solved by this application is to provide a method, device, medium and product for predicting bladder cancer muscle layer invasion based on dual sequences, in order to address the shortcomings of the existing technology.
[0006] To address the aforementioned technical problems, the first aspect of this application provides... A second aspect of this application provides an electronic device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the dual-sequence-based bladder cancer muscle layer invasion prediction method as described above.
[0007] A third aspect of this application provides a storage medium, which is a computer-readable storage medium, on which a computer program is stored, which, when executed by a processor, implements the steps of the double-sequence-based bladder cancer muscle layer invasion prediction method as described above.
[0008] A fourth aspect of this application provides a computer program product comprising a computer program that, when executed by a processor, implements the steps of the dual-sequence-based bladder cancer myometrial invasion prediction method as described above.
[0009] Beneficial Effects: Compared with existing technologies, this application provides a method, device, medium, and product for predicting bladder cancer muscle layer invasion based on dual-sequence MRI images. The method includes acquiring dual-sequence MRI images; extracting features from the dual-sequence MRI images using a feature extractor to obtain first image features and second image features; fusing the first image features and second image features using a dual-sequence attention module to obtain fused features; and predicting the bladder cancer muscle layer invasion category based on the fused features using a prediction module. After acquiring the first image features and second image features, this application utilizes a dual-sequence attention module to fuse the first image features and second image features. This allows for dynamic learning of the weights of the dual-sequence MRI images, effectively combining anatomical structural features with functional microscopic features, reducing the complexity of data acquisition and annotation while improving the accuracy of bladder cancer muscle layer invasion prediction. Attached Figure Description
[0010] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0011] Figure 1 This is a structural diagram of a bladder cancer muscle layer invasion prediction model.
[0012] Figure 2 A flowchart of a method for predicting bladder cancer muscle layer invasion based on dual sequences, provided in an embodiment of this application.
[0013] Figure 3 This is a network structure diagram of a specific example of a bladder cancer muscle layer invasion prediction model.
[0014] Figure 4 This is an example of a heatmap.
[0015] Figure 5 A schematic block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation
[0016] This application provides a method, device, medium, and product for predicting bladder cancer muscle layer invasion based on dual sequences. To make the objectives, technical solutions, and effects of this application clearer, the following will provide a more detailed description of this application in conjunction with the accompanying drawings and examples. It should be noted that the specific embodiments described herein are only for explaining this application and are not intended to limit it.
[0017] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further clarified that the term “comprising” as used in this specification indicates the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or combinations thereof. It should be understood that when an element is referred to as being “connected” or “coupled” to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, the term “connected” or “coupled” as used herein encompasses wireless connection and wireless coupling. The term “and / or” as used herein includes all, any, and all combinations of one or more of the associated listed items.
[0018] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to be consistent with their meaning in the prior art context and should not be interpreted in an idealized or overly formal sense unless specifically defined herein.
[0019] It should be understood that the sequence number and size of each step in this embodiment do not represent the order of execution. The execution order of each process is determined by its function and internal logic, and should not constitute any limitation on the implementation process of this application embodiment.
[0020] Research has shown that the current common method for clinical staging of bladder cancer is to obtain tumor and muscle layer tissue and combine it with intraoperative transurethral endoscopy for evaluation to complete pathological staging. However, research results indicate that misdiagnosis frequently occurs based on specimens obtained from transurethral resection of bladder tumor (TURBT), misclassifying muscle-invasive bladder cancer (MIBC) as non-muscle-invasive bladder cancer (NMIBC), leading to an underestimation of bladder cancer staging. Furthermore, TURBT carries the potential risk of serious complications such as bladder perforation.
[0021] In recent years, MRI image sequences have developed rapidly, enabling radiologists to overcome misdiagnosis issues in bladder cancer assessment during transurethral resection of bladder tumor (TURBT) and improve the accuracy of predicting bladder cancer muscle invasion. However, existing VI-RADS scoring methods for bladder cancer assessment based on MRI image sequences show uncertainty in predicting subsets of 2 and 3 scores.
[0022] To address the uncertainty in VI-RADS scores of 2 and 3, existing research has applied deep learning to predict bladder cancer muscle invasion. For example, a deep learning-based diagnostic model for bladder cancer muscle invasion, after experimental validation, showed superior predictive efficacy compared to traditional VI-RADS scores in subsets of cases with VI-RADS scores of 2 and 3. However, current research largely relies on T2WI single-sequence modeling, leading to insufficient generalization ability of bladder cancer muscle invasion diagnostic models during external validation. While some methods utilize three-parameter MRI sequences to construct predictive models for muscle-invasive bladder cancer (MIBC), indicating that multi-sequence prediction models outperform single-sequence models, the use of three-parameter MRI sequences presents challenges due to the large data acquisition volume and high annotation complexity. Furthermore, the use of gadolinium contrast agents in DCE sequences is associated with the risk of renal systemic fibrosis (NFS), potentially posing safety risks to humans. Therefore, finding a method that can reduce data acquisition and annotation complexity while improving predictive accuracy and generalization ability is of significant clinical importance.
[0023] Based on this, in this embodiment, dual-sequence MRI images are acquired; features are extracted from the dual-sequence MRI images using the feature extractor to obtain first image features and second image features; the first image features and second image features are fused using the dual-sequence attention module to obtain fused features; and the prediction module predicts the bladder cancer muscle layer invasion category based on the fused features. After acquiring the first image features and second image features, this application uses the dual-sequence attention module to fuse the first image features and second image features. This allows for dynamic learning of the weights of the dual-sequence MRI images, effectively combining the anatomical structural features of the T2WI sequence with the functional microscopic features of the DWI sequence, thereby improving the accuracy of bladder cancer muscle layer invasion prediction. Simultaneously, this application uses dual-sequence MRI images of T2WI and DWI sequences as input data, which reduces the amount of data acquisition and annotation complexity, and avoids the safety risks associated with the use of gadolinium contrast agents in DCE sequences.
[0024] The application content will be further explained below with reference to the accompanying drawings and the description of the embodiments.
[0025] This embodiment provides a two-sequence method for predicting bladder cancer muscle layer invasion, applying a trained bladder cancer muscle layer invasion prediction model. For example... Figure 1As shown, the bladder cancer muscle layer invasion prediction model includes a feature extractor, a dual-sequence attention module, and a prediction module. The feature extractor is connected to both the dual-sequence attention module and the prediction module, and the dual-sequence attention module is connected to the prediction module. Specifically, the feature extractor extracts features from dual-sequence MRI images, the dual-sequence attention module fuses the dual-sequence image features extracted by the feature extractor, and the prediction module predicts bladder cancer muscle layer invasion based on the dual-sequence image features extracted by the feature extractor and the fused features obtained by the dual-sequence attention module, thereby determining the bladder cancer muscle layer invasion category.
[0026] like Figure 2 As shown, the dual-sequence method for predicting bladder cancer muscle invasion provided in this application specifically includes: S10. Acquire dual-sequence MRI images.
[0027] Specifically, dual-sequence MRI images are obtained from dual-parameter MRI scans of the same patient, including T2WI and DWI sequences. T2WI sequences provide high-contrast macroscopic anatomical features, clearly showing the layered structure of the bladder wall and the morphology, size, and spatial boundaries of the tumor, serving as the morphological basis for assessing whether the tumor has disrupted the continuity of the muscle layer. DWI sequences reflect the Brownian motion of water molecules (i.e., the degree of diffusion restriction), providing functional features of high cell density. DWI sequences compensate for the limitations of T2WI in cases of edema or inflammation, enhancing the ability to differentiate muscle layer invasion (MIBC).
[0028] It is important to note that in practical applications, consistency of scanning parameters must be ensured when acquiring dual-sequence MRI images to guarantee high-quality results. Furthermore, after obtaining the dual-sequence MRI images, preprocessing can be performed, such as normalization, cropping, and resampling, to enhance image contrast and clarity, reduce noise and artifacts, and make the images more suitable for subsequent feature extraction and analysis.
[0029] S20. The feature extractor is used to extract features from the dual-sequence MRI images to obtain the first image features and the second image features.
[0030] Specifically, the feature extractor is used to extract features from dual-sequence MRI images. The feature extractor may include a feature extraction branch, through which features are extracted sequentially from two single-sequence MRI images in the dual-sequence MRI images to obtain a first image feature and a second image feature; or, it may include two parallel feature extraction branches, through which features are extracted in parallel from two single-sequence MRI images in the dual-sequence MRI images to obtain a first image feature and a second image feature.
[0031] In one specific embodiment, such as Figure 3 As shown, the feature extractor includes two parallel feature extraction branches, both of which use the same network structure. For example, both parallel feature extraction branches use a ResNet mesh structure, which includes cascaded convolutional units and four residual stages. The convolutional units include convolutional layers (Conv), normalization layers (BN), ReLU activation function layers, and max pooling layers. The first residual stage includes one residual block (Bottleneck block) and two basic blocks. The second residual stage includes one residual block (Bottleneck block) and three basic blocks. The third residual stage includes one residual block (Bottleneck block) and four basic blocks. The fourth residual stage includes one residual block (Bottleneck block) and two basic blocks.
[0032] The Bottleneck residual block comprises a first residual branch, a second residual branch, and a ReLU activation function layer. The input terms of the first and second residual branches are identical, and their output terms are fused by an adder before being input into the ReLU activation function layer. The first residual branch includes two first convolutional blocks and one second convolutional block, while the second residual branch includes one second convolutional block. The first convolutional block includes a convolutional layer (Conv), a normalization layer (BN), and a ReLU activation function layer, while the second convolutional block includes a convolutional layer (Conv) and a normalization layer (BN).
[0033] The basic block comprises a first basic branch, a second basic branch, and a ReLU activation function layer. The input terms of the first and second basic branches are identical, and their output terms are fused together by an adder before being input into the ReLU activation function layer. The first basic branch includes a first convolutional block and a second convolutional block, and the second basic branch includes a second convolutional block.
[0034] When extracting features from dual-sequence MRI images, the T2WI and DWI MRI sequences can be input into the feature extractor. The rich anatomical information in the T2WI sequence is processed through convolution, normalization, and activation operations using a ResNet mesh structure to gradually extract features related to bladder anatomy, yielding the first image features. Similarly, the intra-tissue water molecule diffusion information contained in the DWI sequence is processed through convolution, normalization, and activation operations using a ResNet mesh structure to gradually extract features reflecting tumor cell density and metabolic activity, yielding the second image features. The first and second image features are used only to distinguish between features extracted from T2WI and DWI sequences; therefore, features extracted from DWI sequences can be designated as the first image feature, and features extracted from T2WI sequences as the second image feature.
[0035] This application embodiment uses a ResNet mesh structure as a feature extractor. The residual structure of ResNet enables the network to learn more complex features, making the extracted first and second image features more representative and distinguishable. This not only effectively avoids the gradient vanishing and gradient exploding problems, but also improves the efficiency and accuracy of feature extraction.
[0036] S30. The first image features and the second image features are fused using the dual-sequence attention module to obtain fused features.
[0037] Specifically, the dual-sequence attention module is used to fuse the first image features and the second image features. That is, the fused features in this application are not simply obtained by splicing the first image features and the second image features, but rather by adaptively learning the importance weights of the first image features and the second image features in different regions and channels through the dual-sequence attention module. This is to obtain representative and discriminative key features from the first image features and the second image features and suppress redundant information, so as to effectively combine the anatomical structural features of the T2WI sequence with the functional microscopic features of the DWI sequence, so that the fused features can more comprehensively and accurately reflect the feature information of bladder cancer.
[0038] In one embodiment, the dual-sequence attention module includes a feature dimensionality reduction unit, an attention unit, and a fusion unit. The feature dimensionality reduction unit performs feature dimensionality reduction on the first image features and the second image features. The attention unit performs channel attention learning on the first image features and the second image features. The fusion unit performs feature fusion. Therefore, the step of fusing the first image features and the second image features through the dual-sequence attention module to obtain fused features specifically includes: The first image feature and the second image feature are reduced in dimensionality by the feature dimensionality reduction unit to obtain the first dimensionality reduction feature corresponding to the first image feature and the second dimensionality reduction feature corresponding to the second image feature. Channel attention learning is performed on the first image feature and the second image feature by an attention unit to obtain the first attention weight corresponding to the first image feature and the second attention weight corresponding to the second image feature. The first dimensionality reduction feature and the second dimensionality reduction feature are fused by the fusion unit based on the first attention weight and the second attention weight to obtain the fused feature.
[0039] Specifically, the feature dimensionality reduction unit is used to perform feature dimensionality reduction. The feature dimensionality reduction unit includes a global average pooling layer AvgPool, which compresses the first image features and the second image features into a one-dimensional vector through the average pooling layer. That is, the high-dimensional first image features are compressed into a one-dimensional first dimensionality reduction feature through the global average pooling layer, and the high-dimensional second image features are compressed into a one-dimensional second dimensionality reduction feature.
[0040] An attention unit is used to perform channel attention learning on the first image features and the second image features. A dual-sequence attention module may include one attention unit, through which the first and second image features sequentially undergo channel attention learning; alternatively, the dual-sequence attention module may include two parallel attention units, one for performing channel attention learning on the first image features and the other for performing channel attention learning on the second image features.
[0041] like Figure 3 As shown, the attention unit includes a globally average pooling layer (AvgPool), a fully connected layer (FC), a first activation function, another fully connected layer (FC), and a second activation function, all cascaded together. The input to the multiplier includes the input to the attention unit and the output of the second activation function. The globally average pooling layer compresses the input into a one-dimensional vector to obtain global channel information. The fully connected layer, the first activation function, the fully connected layer, and the second activation function work together to dynamically learn the channel weights. The first activation function can be a ReLU activation function, and the second activation function can be a tanh activation function.
[0042] In this embodiment, the attention unit compresses high-dimensional features into a one-dimensional vector using a global average pooling layer, acquiring global channel information and reducing computational cost. Then, it dynamically learns channel weights through the combined action of a fully connected layer, a first activation function, another fully connected layer, and a second activation function to strengthen key features and suppress redundant information. This highlights the crucial information in the first and second image features that plays a significant role in predicting bladder cancer muscle layer invasion. After channel attention learning, the first attention weight corresponding to the first image feature and the second attention weight corresponding to the second image feature are obtained. These two weights reflect the importance of each channel in different image features.
[0043] In one embodiment, the dual-sequence attention module further includes a fully connected layer, the input of which is the output of the feature dimensionality reduction unit, and the output of which is the input of the fusion unit. Based on this, the fusion of the first dimensionality-reduced feature and the second dimensionality-reduced feature by the fusion unit based on a first attention weight and a second attention weight to obtain the fused feature specifically includes: The first dimensionality reduction feature and the second dimensionality reduction feature are processed by a fully connected layer to obtain the first fully connected feature corresponding to the first dimensionality reduction feature and the second fully connected feature corresponding to the second dimensionality reduction feature. The fusion unit fuses the first fully connected features and the second fully connected features based on the first attention weight and the second attention weight to obtain fused features.
[0044] Specifically, the dual-sequence attention module can include one or two fully connected layers in parallel, which perform fully connected processing on the first dimensionality-reduced feature and the second dimensionality-reduced feature. When the dual-sequence attention module includes one fully connected layer, the first and second dimensionality-reduced features are processed sequentially through this layer; when the dual-sequence attention module includes two fully connected layers, the first and second dimensionality-reduced features are processed in parallel through the two fully connected layers.
[0045] After obtaining the first fully connected feature and the second fully connected feature, the first and second fully connected features are input into the fusion unit. The fusion unit multiplies the first attention weight and the first fully connected feature to obtain the first attention feature, and multiplies the second attention weight and the second fully connected feature to obtain the second attention feature. Finally, the first attention feature and the second attention feature are fused to obtain the fused feature. For example, the first attention feature and the second attention feature can be concatenated.
[0046] S40. The prediction module predicts the bladder cancer muscle layer invasion category based on the fusion features.
[0047] Specifically, after obtaining the fusion features, the prediction module uses these features to predict the bladder cancer muscle layer invasion category. The bladder cancer muscle layer invasion category is categorized as either bladder cancer muscle layer invasion or no bladder cancer muscle layer invasion.
[0048] Furthermore, in practical applications, to better demonstrate the basis for tumor classification judgments made by the bladder cancer muscle layer invasion prediction model, feature visualization is performed simultaneously with obtaining the bladder cancer muscle layer invasion category. Therefore, the bladder cancer muscle layer invasion prediction method based on dual sequences further includes: Read the output features of the last convolutional layer in the bladder cancer muscle invasion prediction model; The gradient information of the output features is obtained, and a heatmap is generated based on the gradient information. The heatmap is used to reflect the degree of attention paid to different regions of the input image during prediction.
[0049] Specifically, the output features of the last convolutional layer are the output features of the last convolutional layer in the feature extractor of the bladder cancer muscle layer invasion prediction model. These output features include those corresponding to T2WI MRI images and DWI MRI images. After reading the output features, the gradient information of the output features is obtained, and a heatmap is generated based on this gradient information. This heatmap reflects the degree of attention paid to different regions of the input image during prediction. The heatmap can be generated using gradient-weighted class activation mapping (Grad-CAM) technology to visualize the output features. For example, the heatmap generated based on the gradient information is as follows: Figure 4 As shown in the heatmap, the regions that the bladder cancer muscle invasion prediction model highly focuses on and regions that receive less attention when making predictions are displayed.
[0050] In one embodiment, an imbalance in the number of different bladder cancer types may occur during training. To overcome this problem, this application employs a balanced focus loss function as the loss function for the bladder cancer muscle invasion prediction model during training. The balanced focus loss function dynamically adjusts the loss weight of each category based on the inherent imbalance characteristics of different categories, thereby resolving the category imbalance problem. The formula for calculating the balanced focus loss function is as follows: , in, Represents the equilibrium focus loss function. This represents the predicted probability for the target category. Indicates the focus factor. α t This is the category balance factor.
[0051] Furthermore, during training, the bladder cancer muscle invasion prediction model was trained ten times using the training dataset. Each completed training iteration of the bladder cancer muscle invasion prediction model was then validated on the validation set. The number of training epochs was fixed for each iteration, but the environment was changed with random seeds to reduce random errors generated by the network. Finally, the prediction probabilities of the ten models were averaged to obtain the final result.
[0052] The training and validation datasets consist of dual-sequence MRI images obtained from the institution, with a radiologist using ITK-SNAP software to perform 3D annotation of bladder tumors. Specifically, rectangles were drawn at the beginning and end of the lesion's cross-section to determine the lesion's spatial extent along the z-axis; simultaneously, a rectangle precisely enclosing the lesion was drawn at the most prominent level to define the x and y axis boundaries, thus constructing a 3D bounding box containing the complete tumor, which served as the basis for image cropping. All regions of interest (VOIs) underwent rigorous verification to ensure no bladder cancer lesions were missed. The specific image preprocessing workflow is as follows: The 3D voxels of interest (VOIs) within the region of interest were segmented into several image patches, and multiple 128×128×8 image patches were extracted from these voxels. The sliding step size in the z-direction was set to 5, meaning that one 128×128×8 image patch was obtained for every 5 slices of the MRI image volume data. If the number of voxels of interest along the z-direction is less than 8, a zero-padding strategy was used. Simultaneously, the intensity of the input image was normalized to the [0,1] interval. Image augmentation was performed using the following methods: horizontal and vertical flipping, image cropping, affine transformations including scaling and translation; image sharpening, contrast adjustment, and addition of Gaussian noise. Image augmentation was performed randomly on each patch using these methods, with an augmentation probability of 100%. The purpose of image augmentation is to better characterize the intrinsic features of the data and effectively alleviate model overfitting, thereby increasing the model's generalization ability.
[0053] Before inputting the data into the bladder cancer muscle layer invasion prediction model, the dual-sequence MRI images were preprocessed using the LBP (Local Binary Patterns) operator. LBP is a powerful texture descriptor that generates a binary code by comparing the gray values of a pixel with those of its neighboring pixels, and then converts it into a hash value to capture the local texture features of the image. Ultimately, it can effectively extract the local features and potential micro-texture changes of the image.
[0054] Furthermore, to further illustrate the performance of the bladder cancer muscle invasion prediction model provided in this application, MedCalc Software was used for data analysis. In the comparison of clinical variables between groups, the chi-square test was used for categorical variables, and the independent samples t-test was used for continuous variables. Cohen's kappa coefficient was used to assess the consistency of the VI-RADS scores between the two radiologists. The diagnostic performance of the deep learning (DL) method and VI-RADS was evaluated using the area under the curve (AUC) and its 95% confidence interval (CI), and the optimal cutoff value was determined based on the principle of maximizing the receiver operating characteristic (ROC) curve exponent, thereby calculating accuracy (ACC), sensitivity (SEN), and specificity (SPE). The Delong test was used to compare the AUC differences between the bladder cancer muscle invasion prediction model, the T2WI model, and the VI-RADS scores of the two doctors. A two-sided p-value <0.05 was considered statistically significant.
[0055] In the internal validation set of Institution 1, for tumors scored 2-3 by Radiologists 1 and 2, the diagnostic accuracy of the bladder cancer muscle invasion prediction model provided in this application was 0.893, significantly higher than the 0.679 of the T2WI model (p=0.031); the AUC values of the two models were 0.962 and 0.846, respectively. In the external validation set of Institution 2, for tumors assessed as 2-3 by Radiologists 1 and 2, there was no significant difference in AUC and accuracy between the bladder cancer muscle invasion prediction model provided in this application and the T2WI model. In the external validation set of Institution 3, in the tumor cohort with a score of 2-3 from Radiology 1, the AUC of the bladder cancer muscle invasion prediction model provided in this application was 0.900, which was significantly higher than that of the T2WI model (0.763), and the difference in accuracy (0.884 vs. 0.721) was also significant. In the tumors with the same score assessed by Radiology 2, the AUC (0.901) and accuracy (0.883) of the bladder cancer muscle invasion prediction model provided in this application were also significantly better than those of the T2WI model (0.778) and accuracy (0.698).
[0056] The above data show that the bladder cancer muscle invasion prediction model provided in this application is comparable to the VI-RADS score and the T2WI model (single sequence model) in terms of accuracy; however, its accuracy is significantly better than the T2WI model in cases with VI-RADS scores of 2 and 3, and it exhibits better generalization ability on the external test set. Furthermore, in terms of performance, the bladder cancer muscle invasion prediction model provided in this application shows excellent AUC values on all validation sets. This indicates that the bladder cancer muscle invasion prediction model provided in this application has good reliability in practical applications.
[0057] This application provides an electronic device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the dual-sequence-based bladder cancer muscle layer invasion prediction method described in Embodiment 1 above.
[0058] The following is for reference. Figure 5 The diagram illustrates a structural schematic of an electronic device suitable for implementing embodiments of this application. The electronic devices in these embodiments may include, but are not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 5 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.
[0059] like Figure 5 As shown, the electronic device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the electronic device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. While electronic devices with various systems are shown in the figures, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.
[0060] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.
[0061] The electronic device provided in this application employs the dual-sequence-based bladder cancer muscle layer invasion prediction method described in the above embodiments, thereby solving the technical problem of inaccurate bladder cancer assessment. Compared with the prior art, the beneficial effects of the electronic device provided in this application are the same as those of the dual-sequence-based bladder cancer muscle layer invasion prediction method provided in the above embodiments, and other technical features of this electronic device are the same as those disclosed in the previous embodiment method, and will not be repeated here.
[0062] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.
[0063] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
[0064] This application provides a medium, which is a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the dual-sequence-based bladder cancer muscle layer invasion prediction method in the above embodiments.
[0065] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM). ROM: CD Read-only memory, optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system or device. The program code contained on the computer-readable storage medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (Radio Frequency), etc., or any suitable combination thereof.
[0066] The aforementioned computer-readable storage medium may be included in an electronic device or may exist independently without being assembled into an electronic device.
[0067] The aforementioned computer-readable storage medium carries one or more programs, which, when executed by an electronic device, cause the electronic device to: Acquire dual-sequence MRI images; extract features from the dual-sequence MRI images using the feature extractor to obtain first image features and second image features; fuse the first image features and second image features using the dual-sequence attention module to obtain fused features; and predict the bladder cancer muscle layer invasion category based on the fused features, the first image features, and the second image features using the prediction module.
[0068] Computer program code for performing the operations of this application can be written in one or more programming languages or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, and C++, and conventional procedural programming languages such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).
[0069] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.
[0070] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.
[0071] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described method for predicting bladder cancer muscle layer invasion based on dual sequences. This solves the technical problem of inability to share the desktop when running X11 applications in the OpenHarmony desktop operating system. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the method for predicting bladder cancer muscle layer invasion based on dual sequences provided in the above embodiments, and will not be repeated here.
[0072] This application also provides a product, which is a computer program product, including a computer program that, when executed by a processor, implements the steps of the double-sequence-based bladder cancer muscle layer invasion prediction method described above.
[0073] The computer program product provided in this application can solve the technical problem of inaccurate bladder cancer assessment. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the bladder cancer muscle layer invasion prediction method based on dual sequences provided in the above embodiments, and will not be repeated here.
[0074] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.
Claims
1. A method for predicting bladder cancer muscle layer invasion based on dual sequences, characterized in that, A trained bladder cancer muscle layer invasion prediction model is applied. This model includes a feature extractor, a dual-sequence attention module, and a prediction module. Specifically, the dual-sequence-based bladder cancer muscle layer invasion prediction method includes: Acquire dual-sequence MRI images; The feature extractor is used to extract features from the dual-sequence MRI images to obtain first image features and second image features. The dual-sequence attention module performs feature fusion on the first image features and the second image features to obtain fused features; The prediction module predicts the bladder cancer muscle layer invasion category based on the fusion features.
2. The method for predicting bladder cancer muscle layer invasion based on dual sequences according to claim 1, characterized in that, The dual-sequence attention module includes a feature reduction unit, an attention unit, and a fusion unit; the dual-sequence attention module performs feature fusion on the first image features and the second image features. The specific features obtained from the fusion include: The first image feature and the second image feature are reduced in dimensionality by the feature dimensionality reduction unit to obtain the first dimensionality reduction feature corresponding to the first image feature and the second dimensionality reduction feature corresponding to the second image feature. Channel attention learning is performed on the first image feature and the second image feature by an attention unit to obtain the first attention weight corresponding to the first image feature and the second attention weight corresponding to the second image feature. The first dimensionality reduction feature and the second dimensionality reduction feature are fused by the fusion unit based on the first attention weight and the second attention weight to obtain the fused feature.
3. The method for predicting bladder cancer muscle layer invasion based on dual sequences according to claim 2, characterized in that, The attention unit includes a global average pooling, a fully connected layer, a first activation function, a fully connected layer, and a second activation function, which are cascaded in sequence. The input of the multiplier includes the input of the attention unit and the output of the second activation function. The global average pooling is used to compress the input into a one-dimensional vector to obtain global channel information. The fully connected layer, the first activation function, the fully connected layer, and the second activation function work together to dynamically learn the channel weights.
4. The method for predicting bladder cancer muscle layer invasion based on dual sequences according to claim 2, characterized in that, The dual-sequence attention module further includes a fully connected layer; the fusion of the first dimensionality-reduced feature and the second dimensionality-reduced feature by the fusion unit based on the first attention weight and the second attention weight to obtain the fused feature specifically includes: The first dimensionality reduction feature and the second dimensionality reduction feature are processed by a fully connected layer to obtain the first fully connected feature corresponding to the first dimensionality reduction feature and the second fully connected feature corresponding to the second dimensionality reduction feature. The fusion unit fuses the first fully connected features and the second fully connected features based on the first attention weight and the second attention weight to obtain fused features.
5. The method for predicting bladder cancer muscle layer invasion based on dual sequences according to claim 1, characterized in that, The dual-sequence MRI images include T2WI MRI images and DWI MRI images.
6. The method for predicting bladder cancer muscle layer invasion based on dual sequences according to claim 1, characterized in that, In the training process of the bladder cancer muscle invasion prediction model, the balanced focus loss function is used.
7. The method for predicting bladder cancer muscle layer invasion based on dual sequences according to claim 1, characterized in that, The method further includes: Read the output features of the last convolutional layer in the bladder cancer muscle invasion prediction model; The gradient information of the output features is obtained, and a heatmap is generated based on the gradient information. The heatmap is used to reflect the degree of attention paid to different regions of the input image during prediction.
8. An electronic device, characterized in that, The electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the method for predicting bladder cancer muscle invasion based on dual sequences as described in any one of claims 1 to 7.
9. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the method for predicting bladder cancer muscle layer invasion based on dual sequences as described in any one of claims 1 to 7.
10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the method for predicting bladder cancer muscle layer invasion based on dual sequences as described in any one of claims 1 to 7.