A bladder tumor image segmentation method, device, medium, and program product

Abstract

The application belongs to the field of intelligent medical treatment, and particularly relates to a bladder tumor image segmentation method, device, medium and program product. The method comprises the following steps: S101, acquiring an image of a bladder tumor patient; S102, inputting the image into a 2D U-Net for tumor segmentation to obtain a 2D segmentation result; S103, based on the 2D segmentation result, traversing the shape of the tumor, if the current tumor shape is regular, inputting the image into a 3D U-Net for tumor segmentation to obtain a 3D segmentation result, and taking the segmentation result of the current tumor in the 3D segmentation result; if the tumor shape is irregular, taking the segmentation result of the current tumor in the 2D segmentation result, and after the traversal is completed, merging and outputting the segmentation results of multiple tumors. The application can optimize the segmentation performance by using the complementary advantages of the 2D-3D method, and obtain better segmentation results.

Description

Technical Field

[0001] This invention relates to the field of intelligent healthcare, and more specifically, to an image segmentation method, device, medium, and program product for bladder tumors. Background Technology

[0002] Bladder cancer (BCa) is one of the most common malignant tumors of the urinary system, significantly impacting morbidity and quality of life. According to global cancer statistics, BCa ranks ninth globally in both new cases and deaths, with approximately 614,000 new cases and 220,000 deaths in 2022. Histologically, BCa is divided into non-muscle-invasive bladder cancer (NMIBC) and muscle-invasive bladder cancer (MIBC), with 75% of new cases classified as NMIBC. Despite advances in surgical resection, immunotherapy, and chemotherapy, the 5-year survival rate for advanced BCa, especially MIBC, remains low, highlighting the need for better diagnostic and treatment tools. Accurate delineation of tumor boundaries and bladder wall layers allows clinicians to assess tumor size, location, and invasiveness, crucial for the diagnosis, staging, and treatment planning of BCa. For example, NMIBC and MIBC require different treatment strategies, with the latter often necessitating radical cystectomy. Furthermore, accurate segmentation supports monitoring treatment response and recurrence, which is essential for improving patient prognosis. However, the hollow structure of the bladder causes its position, shape, and volume to change dynamically, and the complex noise and artifacts in medical images complicate segmentation. Furthermore, manual segmentation by radiologists is time-consuming, subjective, and susceptible to inter-observer variability, highlighting the need for automated and reliable segmentation methods. Deep learning (DL), with its ability to automatically extract relevant features without manual selection, has become a powerful method for medical image segmentation. Unlike traditional machine learning (ML) algorithms, DL can efficiently process large-scale medical images, utilizing significant computational resources to improve accuracy and robustness. It can analyze complex patterns, making it particularly suitable for medical imaging tasks such as bladder segmentation.

[0003] In deep learning-based segmentation architectures, U-Net is one of the most effective methods for image segmentation within a fully convolutional architecture, laying the foundation for numerous advanced medical image analysis techniques. Several studies have explored the application of U-Net in bladder segmentation. Dolz et al. proposed an expanded U-Net model with gradually increasing expansion rates to enhance background information for bladder wall and tumor segmentation. This model was trained on T2-weighted MRI images of 60 patients with BCa, achieving Dice similarity coefficients (DSC) scores of 0.98, 0.84, and 0.69 for the bladder inner wall, outer wall, and tumor, respectively. Liu et al. proposed an improved U-Net model (PiPNet), evaluated on T2-weighted MRI images of 47 BCa patients, achieving DSC values ​​of 0.89 and 0.95 for the bladder outer wall and tumor, respectively. Lee et al. proposed RDAU-Net to segment the bladder and lesions in 2136 CT images, achieving accuracies of 96% and 93%, respectively, with a 44% reduction in training time. Jang et al. proposed an improved U-Net model for urinary system segmentation using unenhanced CT images. After training with 814 CT scans, the model achieved a DSC value of 0.903 for the bladder organ. These studies highlight the effectiveness of U-Net-based models in bladder segmentation. However, they are often limited by small datasets, potential overfitting, and a lack of generality, which may restrict their clinical applicability. The original U-Net architecture was introduced as a 2D model and has been widely used in biomedical imaging, particularly in medical segmentation tasks. However, 2DU-Net processes single slices, which can lead to inconsistent volumetric structures. To address this limitation, Çiçek et al. extended the architecture to 3DU-Net, incorporating spatial relationships of depth, height, and width, making it more suitable for volumetric segmentation. Both 2D and 3DU-Net have been widely used in medical imaging, particularly for organ and tumor segmentation. However, no systematic studies have compared their performance in bladder segmentation. This study aims to systematically evaluate the segmentation performance of 2D and 3DU-Net in the simultaneous segmentation of the bladder wall and tumors. In addition, subgroup analysis based on tumor clinical characteristics was conducted to determine the optimal model selection under different clinical conditions, providing a reference for its potential clinical application. Summary of the Invention

[0004] In view of the above problems, the present invention provides an image segmentation method for bladder tumors, which selects a more suitable segmentation model from 2D U-Net and 3D U-Net according to different clinical scenarios, and determines the optimal model selection under different clinical conditions.

[0005] This application (first aspect) discloses an image segmentation method for bladder tumors, comprising:

[0006] S101: Acquire images of patients with bladder cancer;

[0007] S102: Input the image into 2D U-Net to perform tumor segmentation and obtain 2D segmentation results;

[0008] S103: Based on the 2D segmentation results, traverse the shape of the tumor. If the current tumor shape is regular, input the image into 3D U-Net for tumor segmentation to obtain 3D segmentation results, and accept the segmentation results of the current tumor in the 3D segmentation results; if the tumor shape is irregular, accept the segmentation results of the current tumor in the 2D segmentation results. After the traversal is completed, merge and output the segmentation results of multiple tumors.

[0009] This application (second aspect) discloses an image segmentation method for bladder tumors, the method comprising:

[0010] S301: Acquire images of patients with bladder tumors;

[0011] S302: Input the image into 3D U-Net to perform tumor segmentation and obtain 3D segmentation results;

[0012] S303: Based on the 3D segmentation results, traverse the shape of the tumor. During the traversal, if the current tumor shape is regular, accept the segmentation result of the current tumor in the 3D segmentation results; if the tumor shape is irregular, input the image into 2D U-Net for tumor segmentation to obtain 2D segmentation results, accept the segmentation result of the current tumor in the 2D segmentation results, and after the traversal is completed, merge and output the segmentation results of multiple tumors.

[0013] Furthermore, the method also includes determining whether the number of tumors is greater than 1 based on the 2D segmentation result; if it is greater than 1, the image is input into a 3D U-Net for tumor segmentation to obtain a 3D segmentation result.

[0014] Based on the 3D segmentation results, the shape of the tumor is traversed. During the traversal, if the current tumor shape is regular, the segmentation result of the current tumor in the 3D segmentation results is accepted; if the tumor shape is irregular, the segmentation result of the current tumor in the 2D segmentation results is accepted. After the traversal is completed, the segmentation results of multiple tumors are merged and output.

[0015] Furthermore, the method also includes: simultaneously acquiring the time limit of the segmentation task; if the time limit is lower than a first time threshold, inputting the image into 3D U-Net for segmentation and outputting the segmentation result; otherwise, inputting the image into 2D U-Net or 3D U-Net for tumor segmentation and traversing the shape of the tumor.

[0016] Furthermore, the method also includes: determining the sphericity of the tumor shape; if the sphericity of the tumor shape is higher than the sphericity threshold, it is determined to be a regular shape; otherwise, it is determined to be an irregular shape.

[0017] Optionally, the sphericity threshold is constructed based on the sphericity of the training set;

[0018] Optionally, a shape discriminant classifier can be used to determine whether the tumor shape is regular. The shape discriminant classifier is constructed by inputting tumor images from the training set and labels indicating whether the shape is regular into the classifier for training.

[0019] Furthermore, the method also includes: inputting the image into a 2D U-Net to perform bladder wall segmentation to obtain bladder wall segmentation results, and performing tumor segmentation based on the image and the bladder wall segmentation results.

[0020] Furthermore, the method also includes: the 2D U-Net model architecture consists of 7 downsampling and 7 upsampling stages, with a basic feature count of 32;

[0021] Optionally, the 3D U-Net model architecture consists of 5 downsampling stages and 5 upsampling stages, with a basic feature count of 32.

[0022] Optionally, the training process for 2D U-Net and 3D U-Net employs test-time augmentation techniques;

[0023] Optionally, the image is an enhanced CT image.

[0024] A third aspect of this application discloses an image segmentation system for bladder tumors, comprising:

[0025] Acquisition module 201: Used to acquire images of patients with bladder tumors;

[0026] Initial segmentation module 202: Used to input images into 2D U-Net for tumor segmentation to obtain 2D segmentation results;

[0027] Traversal Adjustment Output Module 203: Based on the 2D segmentation result, traverse the shape of the tumor. During traversal, if the current tumor shape is regular, input the image into 3D U-Net for tumor segmentation to obtain 3D segmentation results, and accept the segmentation result of the current tumor in the 3D segmentation results; if the tumor shape is irregular, accept the segmentation result of the current tumor in the 2D segmentation results, and after traversal, merge and output the segmentation results of multiple tumors.

[0028] This application (fourth aspect) discloses an image segmentation system for bladder tumors, comprising:

[0029] Second acquisition module 301: used to acquire images of bladder tumor patients;

[0030] Second initial segmentation module 302: Used to input images into 3D U-Net for tumor segmentation to obtain 3D segmentation results;

[0031] The second traversal adjustment output module 303 is used to traverse the shape of the tumor based on the 3D segmentation result. During traversal, if the current tumor shape is regular, the segmentation result of the current tumor in the 3D segmentation result is accepted; if the tumor shape is irregular, the image is input into 2D U-Net for tumor segmentation to obtain 2D segmentation result, and the segmentation result of the current tumor in the 2D segmentation result is accepted. After the traversal is completed, the segmentation results of multiple tumors are merged and output.

[0032] The fifth aspect of this application discloses a computer device, the device comprising: a memory and a processor; the memory being used to store program instructions; the processor being used to invoke the program instructions, which, when executed, are used to perform the steps of the method described above.

[0033] The sixth aspect of this application discloses a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.

[0034] The seventh aspect of this application discloses a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method.

[0035] This application has the following beneficial effects:

[0036] (1) The present application uses a hybrid 2D-3D method, which can take advantage of the complementary advantages of the two to optimize the segmentation performance and obtain better segmentation results;

[0037] (2) This application has determined through research how to select the segmentation model to achieve the best results in different scenarios, so that a more suitable model can be selected from the widely used models in clinical applications, thereby improving the overall effect in bladder cancer tumor segmentation. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention, 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 the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is a schematic diagram of the method flow provided in the first aspect of the present invention;

[0040] Figure 2 This is a schematic diagram of a program product provided in the third aspect of the present invention;

[0041] Figure 3 This is a schematic diagram of a computer device provided in an embodiment of the present invention;

[0042] Figure 4 This is a schematic diagram of the architecture of an exemplary computing device provided in an embodiment of the present invention;

[0043] Figure 5 This is a schematic diagram of the storage medium provided in an embodiment of the present invention;

[0044] Figure 6 This is a schematic diagram illustrating the process of tumor segmentation using 3D U-net and 2D U-net networks with test time enhancement technology, provided in an embodiment of the present invention.

[0045] Figure 7 This invention provides a comparison of the bladder tumor segmentation performance of 2D U-Net and 3D U-Net in internal (A–C) and external (D–F) test cohorts. Box plots illustrate the Dice similarity coefficient (DSC) (A, D), average surface distance (ASD) (B, E), and 95% Hausdorff distance (95% HD) (C, F). Higher DSC indicates better segmentation accuracy, while lower ASD and 95% HD indicate higher boundary accuracy. In both cohorts, 2D U-Net generally achieved a higher DSC but exhibited greater fluctuations in ASD and 95% HD.

[0046] Figure 8 This invention provides an embodiment of the correlation between Dice similarity coefficients (DSC) of 2D U-Net (blue) and 3D U-Net (orange) and radiomics shape features. The scatter plot illustrates the relationship between DSC and tumor shape features, including elongation (A), flatness (B), minimum axial length (C), surface area to volume ratio (D), maximum 2D diameter (E), grid volume (F), sphericity (G), and voxel volume (H). 2D U-Net shows stronger correlations in elongation, flatness, and sphericity, while 3D U-Net is more affected by irregular morphology. The surface area to volume ratio is negatively correlated with DSC, indicating that segmentation in highly irregular tumors is more challenging.

[0047] Figure 9 This is a flowchart of a study subject enrollment screening process provided by an embodiment of the present invention;

[0048] Figure 10 This is a schematic diagram of a 2D-3D selection process provided by an embodiment of the present invention;

[0049] Figure 11 This is a schematic diagram of a 2D-3D selection process provided by an embodiment of the present invention. Detailed Implementation

[0050] To enable those skilled in the art to better understand the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0051] In some of the processes described in the specification, claims, and accompanying drawings of this invention, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as S101, S102, etc., are merely used to distinguish different operations and do not represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types.

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] Figure 1 This is a schematic flowchart of an image segmentation method for evaluating bladder tumors provided by an embodiment of the present invention. Specifically, the method includes the following steps:

[0054] S101: Acquire images of patients with bladder cancer;

[0055] S102: Input the image into 2D U-Net to perform tumor segmentation and obtain 2D segmentation results;

[0056] S103: Based on the 2D segmentation results, traverse the shape of the tumor. If the current tumor shape is regular, input the image into 3D U-Net for tumor segmentation to obtain 3D segmentation results, and accept the segmentation results of the current tumor in the 3D segmentation results; if the tumor shape is irregular, accept the segmentation results of the current tumor in the 2D segmentation results. After the traversal is completed, merge and output the segmentation results of multiple tumors.

[0057] The method described in this application is based on the following research.

[0058] Segmentation of bladder cancer (BCa) is crucial for accurate diagnosis, staging, and treatment planning. This study aimed to systematically compare the performance of 2D U-Net and 3D U-Net in bladder wall and tumor segmentation, focusing on different tumor subgroups and radiomic features. Methods: A retrospective study was conducted on enhanced CT scans of 435 BCa patients from two centers, divided into a training group (n=325), an internal testing group (n=72), and an external testing group (n=38). Both models were trained using the nnU-Net framework and evaluated using the Dice similarity coefficient (DSC), mean surface distance (ASD), and 95th percentile Hausdorff distance (95% HD). Subgroup analyses were performed based on tumor characteristics, including size, multifocality, muscular invasiveness, and shape-based radiomic features. Results: For bladder wall segmentation, 3D U-Net slightly outperformed 2D U-Net in the internal testing cohort but showed lower generalizability in the external cohort. In tumor segmentation, 2D U-Net consistently achieved higher DSC scores, especially for small tumors (< 0.001). For large tumors (> 3 cm), both models performed well, but 2D U-Net maintained higher accuracy (P = 0.028). 3D U-Net showed slightly better performance in multi-tumor segmentation, with lower ASD and significantly faster inference time (P < 0.001). Radiographic feature analysis showed that 2D U-Net performed better on irregular tumors, while 3D U-Net was more effective for spherical lesions. Conclusion: 2D U-Net demonstrates superior segmentation accuracy, especially for small, single, and irregular tumors, making it the preferred choice for early-stage BCa and NMIBC. 3D U-Net, with its faster inference time and volume continuity, may be beneficial for cases with high tumor burden or intraoperative applications. These findings highlight the importance of selecting segmentation models based on tumor characteristics, and future research should explore hybrid architectures to optimize performance.

[0059] Specifically as follows:

[0060] In an approved two-center retrospective study, the ethics committee waived the requirement for written informed consent due to its retrospective design (I-22PJ887). All procedures were conducted in accordance with the Declaration of Helsinki (2013 edition). The dataset included patients pathologically diagnosed with BCa between October 2021 and August 2023. All patients underwent enhanced abdominal and pelvic CT scans within two weeks prior to surgery. Exclusion criteria were as follows: (1) poor image quality, including severe motion artifacts, high noise levels, or inadequate bladder filling that hindered assessment; (2) only thick-slice enhanced CT images were available; and (3) no visible tumors were observed on the CT scan. A total of 435 patients were included, with 325 in the training set, 72 in the internal test set, and 38 in the external test set. The flowchart is shown below. Figure 9 As shown. The imaging protocol CT images were acquired using scanners from different institutions. The dataset includes portal venous CT images with slice thicknesses of 0.625 mm and 1 mm. Patients at our hospital underwent scanning using GE Discovery CT (GE Healthcare) or Siemens SomatomDefinition Flash CT (Siemens Healthineers) with the following parameters: tube voltage = 120 kVp, rotation time = 0.5 s, collimation = 128 × 0.6 mm, pitch = 0.9 mm, slice thickness = 1 mm. The contrast agent injection protocol was 100 mL of iopamidol at 370 mg / mL, with an injection rate of 4–4.5 mL / s. Images were acquired during the arterial phase (30 s), portal venous phase (50 s), and delayed phase (240 s). The external examination center used a Siemens SOMATOMPerspective 64-slice scanner with a detector array of 0.625 mm × 64, tube voltage of 120 kV, tube current of 300 mAs, and a spacing of 0.984 mm. The contrast-enhanced scanning parameters were consistent with those of our hospital. Ground-based live modeling was created using 3DSlicer (https: / / www.slicer.org) to manually delineate the tumor and bladder wall by a radiologist with four years of experience in urogenital imaging. A senior radiologist with 14 years of experience then reviewed and refined the delineation. The final segmentation results served as the baseline for model training. Model architecture and training process: Both 2DU-Net and 3DU-Net models were trained using a self-configured nnU-Net framework. The training methodology was standardized across the models, with the main differences being input patch size, batch size, and model architecture.

[0061] 2DU-Net: The input image is resampled and cropped to 512×512 pixels, then z-score intensity normalized. The model architecture consists of 7 downsampling and 7 upsampling stages, with 32 basic features. The batch size used during training is 12.

[0062] 3DU-Net: Input images are resampled and cropped to 128×128×128 with a batch size of 2. The architecture has 5 downsampling and 5 upsampling stages, with 32 basic features. Both models are trained using 5x cross-validation for 1000 epochs with an initial learning rate of 0.01. Test-time augmentation techniques are applied during inference to enhance prediction robustness. Figure 6 ).

[0063] Model comparisons based on tumor characteristics and prediction time used DSC, mean surface distance (ASD), and 95th percentile Hausdorff distance (95%HD) to evaluate the segmentation performance of 2D and 3DUNet models. Two board-certified experienced radiologists reviewed clinical, imaging, and pathology records. In addition to visually identifiable classification parameters, radiomics features were extracted from manually segmented tumors as a reference standard. Clinical data included sex and age. Imaging parameters included tumor location (high-risk: trigone and ureteral orifice), number (single vs. multiple), and maximum diameter of a single tumor. Tumors were categorized into small tumors (3 cm) and other tumors to assess the impact of size on segmentation performance. Pathological data included the presence of muscle invasion (MIBC or NMIBC).

[0064] Radiomics Features: Thirteen morphoradiomics features were extracted using Python for additional model comparisons. Inference times for both models were measured. Our aim was to evaluate which model achieved superior segmentation performance across various clinical scenarios, thereby identifying the most suitable approach for different diagnostic and therapeutic settings. Statistical Analysis: Statistical analysis was performed using SPSS software (version 26.0, IBM Corp., Armonk, NY, USA) and R software (version 4.4.2, R Foundation for Statistical Computing, Vienna, Austria). Quantitative variables are expressed as mean ± standard deviation (SD), while categorical variables are expressed as frequencies and percentages to describe patient demographics and variable distribution.

[0065] (1) Within-group comparisons of 2D U-Net and 3D U-Net were performed using the nonparametric paired Wilcoxon signed-rank test to compare the segmentation performance of each model (2D U-Net and 3D U-Net) under different tumor features (DSC, ASD, and 95% HD). The comparisons assessed whether there were significant differences in segmentation accuracy measured by DSC between subgroups within the same model (e.g., single tumor vs. multiple tumors or different tumor sizes). A p-value <0.05 was considered statistically significant.

[0066] (2) Comparison between 2DU-Net and 3DU-Net models within each tumor subgroup: The Wilcoxon signed-rank test was used to compare the segmentation performance of the 2DU-Net and 3DU-Net models for each tumor subgroup. Evaluation metrics included DSC, ASD, and 95%HD to assess segmentation accuracy and boundary precision. A p-value <0.05 was considered statistically significant, indicating a significant difference in segmentation performance between the two models for specific tumor features.

[0067] (3) The association between segmentation performance and radiomics features was assessed by linear regression analysis to evaluate the relationship between DSC and morphoradiological features in the 2D U-Net and 3DU-Net models. Regression lines for each model were plotted to visually represent the trends. P-values ​​and confidence intervals (CI) were reported to assess statistical significance, with P < 0.05 considered statistically significant.

[0068] result:

[0069] Patient Characteristics and Overall Segmentation Performance: This study included 435 patients, with 325 in the training cohort, 72 in the internal testing cohort, and 38 in the external testing cohort. The mean age of patients in the training cohort was 64 ± 12 years, the mean age of patients in the internal testing cohort was 66 ± 11 years, and the mean age of patients in the external testing cohort was 65 ± 9 years (Table 1). The segmentation performance of 2D U-Net and 3D U-Net on bladder wall and tumor was evaluated. For bladder wall segmentation, 3D U-Net performed slightly better on the internal testing set, while 2D U-Net performed better on the external testing set. However, for tumor segmentation, 2D U-Net achieved higher Dice scores on both the internal (0.877 ± 0.073 vs. 0.828 ± 0.146) and external (0.87 ± 0.149 vs. 0.81 ± 0.157) testing sets. Furthermore, 2D U-Net exhibits lower error values ​​for ASD and 95% HD, particularly in tumor segmentation, demonstrating better boundary delineation and robustness. (Table 1) Figure 7 ).

[0070] Table 1. Patient demographics and segmentation performance of 2D U-Net and 3D U-Net models

[0071]

[0072] Table Notes: DSC, Dice similarity coefficient; ASD, mean surface distance; 95% HD, 95th percentile Hausdorff distance.

[0073] In-model comparison of segmentation performance based on tumor subtype:

[0074] (1) Tumor location: The segmentation performance of high-risk tumor sites (the triangle and ureteral orifice) was compared with other sites. No statistically significant differences were observed between the two models (P=0.888 for 2DU-Net and P=0.12 for 3DU-Net). This indicates that tumor location has little impact on the segmentation accuracy of the two models (Tables 2-4).

[0075] (2) Tumor Multiples: For 2DU-Net, the Dice score for a single tumor was 0.885±0.080, which was higher than the Dice score for multiple tumors (0.844±0.157), although the difference was not statistically significant. Similarly, for 3DU-Net, the Dice score for a single tumor (0.833±0.152) was higher than the Dice score for multiple tumors (0.796±0.139), but the difference was almost negligible. That is, when the same segmentation technique is used for both single-tumor and multi-tumor segmentation, the single-tumor segmentation task performs better (Table 3).

[0076] (3) Muscle invasiveness: The segmentation accuracy between MIBC and NMIBC was analyzed. The two models performed slightly better for NMIBC tumors, but no significant difference was found, indicating that tumor invasiveness has a negligible impact on segmentation performance.

[0077] (4) Tumor Size: There were significant differences in segmentation performance between the two models for different tumor sizes. For small tumors (3cm), the segmentation performance of both models improved. In 2DU-Net, the Dice score of tumors >3cm (0.915±0.041) was significantly higher than that of other tumors (0.861±0.116, P=0.034). Similarly, in 3DU-Net, the Dice score of larger tumors (0.883±0.061) was significantly higher than that of other tumors (0.804±0.165, P=0.023).

[0078] Table 2. Comparison of intra-group DSC (Dice similarity coefficient) for 2D U-Net and 3D U-Net.

[0079]

[0080] Table Notes: DSC, Dice similarity coefficient; MIBC, muscle-invasive bladder tumor; NMIBC, non-muscle-invasive bladder tumor.

[0081] Table 3. Comparison of intra-group mean surface distances for 2D U-Net and 3D U-Net

[0082]

[0083] Table 4. Comparison of 95% Hausdorff distances within groups for 2D U-Net and 3D U-Net respectively

[0084]

[0085] These findings suggest that both models achieve better segmentation performance for larger tumors, while smaller tumors present a greater challenge, especially for 3DU-Net. Therefore, using 2D UNet for segmentation in smaller tumors yields relatively better results.

[0086] Comparison of segmentation performance models for different tumor subtypes:

[0087] For a single tumor, 2DU-Net significantly outperformed 3DU-Net in terms of Dice (P<0.001), ASD (P=0.002), and 95%HD (P=0.034).

[0088] Similarly, for NMIBC tumors, 2DU-Net showed significantly higher Dice scores (0.879±0.090 vs. 0.817±0.162, P<0.001) and significantly lower ASD (P=0.006).

[0089] Regarding tumor size, for small tumors (3cm), 2DU-Net outperformed 3DU-Net in DSC (P=0.028), while ASD (P=0.008) and 95%HD (P=0.029) were significantly lower, confirming higher segmentation accuracy for larger tumors.

[0090] For multiple tumors, there was a significant difference in Dice scores between 2D U-Net (0.796±0.139) and 3D U-Net (0.844±0.157) (P=0.049), and 3D U-Net had a lower ASD (although not significant), which suggests that 3D U-Net has a potential advantage in treating multiple lesions.

[0091] For high-risk tumor locations and MIBC, there were no significant differences between the two models, indicating comparable performance in these subgroups (Table 5). Comparison based on radiomorphological features showed significant differences among the eight morphological features (Table 5). Figure 8 ). Figure 8The relationship between tumor shape-based radiomics features and segmentation performance (DSC) of 2DU-Net and 3DU-Net was explained. Both models showed better segmentation accuracy for tumors with higher elongation, flatness, minimum axial length, and sphericity, with 2DU-Net showing a stronger correlation in most cases. Surface area to volume ratio was negatively correlated with Dice score (P < 0.001 for both models), with 3DU-Net being more significantly affected. Figure 8 D).

[0092] Table 5. Inter-model comparison of three metrics between 2D U-Net and 3D U-Net in different tumor subgroups.

[0093]

[0094] 2DU-Net showed a significant positive correlation with maximum 2D diameter, mesh volume, and voxel volume, while 3DU-Net did not. 3DU-Net relied more heavily on sphericity (linear regression: P < 0.001), performing best on spherical tumors but struggling with irregular shapes. These results indicate that tumor shape significantly affects segmentation accuracy; 2DU-Net demonstrates stronger adaptability to different morphologies, while 3D U-Net performs better on more regular tumor shapes.

[0095] Prediction Time: On the internal test set, the prediction time of the 2D U-Net model was 630.818 ± 181.072 seconds, while that of the 3D U-Net model was 595.505 ± 179.131 seconds. There was a significant difference between the two models (p < 0.05). Both models achieved high segmentation accuracy, but 2D U-Net still outperformed 3D U-Net. This suggests that 3D U-Net may be suitable for scenarios with higher segmentation speed requirements, such as intraoperative imaging in late-stage MIBC cases or real-time treatment planning.

[0096] The first time threshold is determined based on the performance of the test set if 3D U-Net segmentation is used during intraoperative imaging or real-time treatment.

[0097] In some embodiments, the first time threshold is 10 minutes.

[0098] For image segmentation in NMIBC, 2D U-Net is the preferred choice.

[0099] For a single tumor, 2DU-Net significantly outperformed 3DU-Net in DSC, ASD, and 95% HD, consolidating its advantage in accurately delineating tumor boundaries.

[0100] Conversely, 3DU-Net performs slightly better in multi-tumor segmentation and has a faster inference time, which may make it the preferred choice for intraoperative imaging or real-time treatment planning in advanced MIBC cases, where speed and handling complex geometries are priorities.

[0101] In summary, the superior performance of 2D U-Net in segmenting small tumors and single tumors suggests it has the potential to become a more reliable tool for early BCa assessment, while the efficiency of 3D U-Net in handling multiple lesions may support its use in high tumor burden cases. Radiomics characterization provided additional insights into the segmentation performance of the two models. 2D U-Net showed a stronger correlation with shape-based features such as elongation, flatness, and minimum axial length, indicating its effective segmentation of elongated or irregularly shaped tumors. In contrast, 3D U-Net showed a higher dependence on sphericity, which may explain its performance degradation in non-spherical tumors. The negative correlation between Dice and surface area to volume ratio for both models suggests that highly irregular tumors pose segmentation challenges, but 3D U-Net is more affected, possibly because it relies on volumetric continuity rather than fine spatial details. These results indicate that 2D U-Net is more adaptable to diverse morphologies, while 3D U-Net is best suited for spherical tumors. In conclusion, our findings suggest that different segmentation models may be preferable depending on tumor characteristics and clinical needs. The superior accuracy of 2DU-Net in small tumors (<1cm), non-MIBC, and irregularly shaped lesions suggests its potential for early BCa monitoring and accurate tumor delineation. In contrast, the comparable performance and faster inference time of 3DU-Net across multiple tumors suggest its potential benefit for high tumor burden cases or real-time applications, such as intraoperative guidance. Furthermore, tumor morphology influences segmentation accuracy; 2D U-Net performs well in irregular tumors, while 3DU-Net is better suited for round lesions. Given their complementary advantages, a hybrid 2D-3D approach can optimize segmentation performance by leveraging the detailed feature extraction of 2D models and the volumetric continuity of 3D models. Future research should explore hybrid architectures and integrate multimodal imaging (MRI, PET-CT) and deep learning-based classification models to further enhance clinical decision-making for BCa.

[0102] In summary, this study provides a comprehensive comparison of the application of 2DU-Net and 3DU-Net in BCa segmentation, highlighting the superior accuracy of 2DU-Net in small, single, and irregular tumors, while emphasizing the potential advantages of 3DU-Net in segmenting multiple tumors and its significantly shorter inference time. These findings provide valuable insights into the optimal application of DL-based segmentation models in different clinical scenarios and underscore the necessity of tailoring model selection to specific tumor characteristics.

[0103] In some embodiments, the process of the bladder tumor image segmentation method is as follows: Figure 10 As shown.

[0104] In some embodiments, the portal venous phase image is segmented using the method proposed in this invention.

[0105] In some embodiments, images of the portal venous phase are used Figure 11 The method shown is used for segmentation.

[0106] Figure 3 This is a schematic diagram of a computer device provided in an embodiment of the present invention, such as... Figure 3 As shown, the device 2000 may include: one or more processors 2010 and one or more memories 2020; wherein the memories store computer-readable code that, when run by the one or more processors, can perform the methods described above.

[0107] The processor in this embodiment can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, operations, and logic block diagrams disclosed in this embodiment. The general-purpose processor can be a microprocessor or any conventional processor, and can be based on an x86 or ARM architecture.

[0108] In general, the various exemplary embodiments of this disclosure can be implemented in hardware or dedicated circuitry, software, firmware, logic, or any combination thereof. Some aspects can be implemented in hardware, while others can be implemented in firmware or software that can be executed by a controller, microprocessor, or other computing device. When aspects of embodiments of this disclosure are illustrated or described as block diagrams, flowcharts, or using some other graphical representation, it will be understood that the blocks, apparatuses, systems, techniques, or methods described herein can be implemented as non-limiting examples in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0109] For example, the method or apparatus according to embodiments of this disclosure can also be used by means of Figure 4 The architecture of the computing device 3000 shown is used for implementation. For example... Figure 4As shown, the computing device 3000 may include a bus 3010, one or more CPUs 3020, a read-only memory (ROM) 3030, a random access memory (RAM) 3040, a communication port 3050 connected to a network, an input / output component 3060, a hard disk 3070, etc. The storage devices in the computing device 3000, such as the ROM 3030 or the hard disk 3070, may store various data or files used for processing and / or communication of the methods provided in this disclosure, as well as program instructions executed by the CPU. The computing device 3000 may also include a user interface 3080. Of course, Figure 4 The architecture shown is merely exemplary and can be omitted as needed when implementing different devices. Figure 4 One or more components in the computing device shown.

[0110] This invention also includes a computer-readable storage medium, such as... Figure 5 The diagram illustrates a storage medium 4000 provided in an embodiment of the present invention. The computer storage medium 4020 stores computer-readable instructions 4010. When the computer-readable instructions 4010 are executed by a processor, the method described above according to embodiments of the present disclosure can be performed. The computer-readable storage medium in the embodiments of the present disclosure may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. Non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory may be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic Random Access Memory (SDRAM), Double Data Rate Synchronous Dynamic Random Access Memory (DDRSDRAM), Enhanced Synchronous Dynamic Random Access Memory (ESDRAM), Synchronous Link Dynamic Random Access Memory (SLDRAM), and Direct Memory Bus Random Access Memory (DR RAM). It should be noted that the memory used in the methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0111] This disclosure also provides a computer program product or computer program that, when executed by a processor, implements the steps of the above-described method, such as... Figure 2 As shown, the computer program product or computer program includes:

[0112] Acquisition module 201: Used to acquire images of patients with bladder tumors;

[0113] Initial segmentation module 202: Used to input images into 2D U-Net for tumor segmentation to obtain 2D segmentation results;

[0114] Traversal Adjustment Output Module 203: Based on the 2D segmentation result, traverse the shape of the tumor. During traversal, if the current tumor shape is regular, input the image into 3D U-Net for tumor segmentation to obtain 3D segmentation results, and accept the segmentation result of the current tumor in the 3D segmentation results; if the tumor shape is irregular, accept the segmentation result of the current tumor in the 2D segmentation results, and after traversal, merge and output the segmentation results of multiple tumors.

[0115] It should be noted that 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 disclosure. 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.

[0116] In general, the various exemplary embodiments of this disclosure can be implemented in hardware or dedicated circuitry, software, firmware, logic, or any combination thereof. Some aspects can be implemented in hardware, while others can be implemented in firmware or software that can be executed by a controller, microprocessor, or other computing device. When aspects of embodiments of this disclosure are illustrated or described as block diagrams, flowcharts, or using some other graphical representation, it will be understood that the blocks, apparatuses, systems, techniques, or methods described herein can be implemented as non-limiting examples in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0117] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0118] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0119] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0120] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0121] The exemplary embodiments of this disclosure described in detail above are merely illustrative and not restrictive. Those skilled in the art will understand that various modifications and combinations can be made to these embodiments or their features without departing from the principles and spirit of this disclosure, and such modifications should fall within the scope of this disclosure.

Claims

1. A method for image segmentation of bladder tumors, characterized in that, The method includes: S101: Acquire images of patients with bladder cancer; S102: Input the image into 2D U-Net to perform tumor segmentation and obtain 2D segmentation results; S103: Based on the 2D segmentation results, traverse the shape of the tumor. If the current tumor shape is regular, input the image into 3D U-Net for tumor segmentation to obtain 3D segmentation results, and accept the segmentation results of the current tumor in the 3D segmentation results; if the tumor shape is irregular, accept the segmentation results of the current tumor in the 2D segmentation results. After the traversal is completed, merge and output the segmentation results of multiple tumors. The method further includes determining whether the number of tumors is greater than 1 based on the 2D segmentation result. If it is greater than 1, the image is input into a 3D U-Net for tumor segmentation to obtain a 3D segmentation result. Based on the 3D segmentation result, the shape of the tumor is traversed. During the traversal, if the current tumor shape is regular, the segmentation result of the current tumor in the 3D segmentation result is accepted. If the tumor shape is irregular, the segmentation result of the current tumor in the 2D segmentation result is accepted. After the traversal is completed, the segmentation results of multiple tumors are merged and output.

2. The image segmentation method for bladder tumors according to claim 1, characterized in that, The method further includes: simultaneously acquiring the time limit of the segmentation task; if the time limit is lower than a first time threshold, inputting the image into 3D U-Net for segmentation and outputting the segmentation result; otherwise, inputting the image into 2D U-Net or 3D U-Net for tumor segmentation and traversing the shape of the tumor.

3. The image segmentation method for bladder tumors according to claim 1, characterized in that, The method further includes: determining the sphericity of the tumor shape; if the sphericity of the tumor shape is higher than the sphericity threshold, it is determined to be a regular shape; otherwise, it is determined to be an irregular shape.

4. The image segmentation method for bladder tumors according to claim 3, characterized in that, The sphericity threshold is constructed based on the sphericity of the training set.

5. The image segmentation method for bladder tumors according to claim 3, characterized in that, The shape discrimination classifier determines whether the tumor shape is regular. The shape discrimination classifier is constructed by inputting tumor images and whether the shape is regular into the training set for training.

6. The image segmentation method for bladder tumors according to claim 1, characterized in that, The method further includes: inputting the image into a 2D U-Net to perform bladder wall segmentation to obtain bladder wall segmentation results, and performing tumor segmentation based on the image and bladder wall segmentation results.

7. The image segmentation method for bladder tumors according to claim 1, characterized in that, The method further includes: the 2D U-Net model architecture consists of 7 downsampling and 7 upsampling stages, with a basic feature count of 32.

8. The image segmentation method for bladder tumors according to claim 1, characterized in that, The 3D U-Net model architecture consists of 5 downsampling and 5 upsampling stages, with a basic feature count of 32.

9. The image segmentation method for bladder tumors according to claim 1, characterized in that, The training process for both the 2D U-Net and the 3D U-Net employs test-time augmentation techniques.

10. The image segmentation method for bladder tumors according to claim 1, characterized in that, The images are enhanced CT images.

11. A computer device, characterized in that, The device includes: a memory and a processor; the memory is used to store a computer program; the processor executes the computer program to implement the steps of the method according to any one of claims 1-10.

12. A computer-readable storage medium, characterized in that, It stores a computer program thereon, which, when executed by a processor, implements the steps of the method as described in any one of claims 1-10.

13. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the method described in any one of claims 1-10.

Images