A Two-Dimensional Geometric Measurement Method and System for Tumors with Imperfect Masks

Abstract

This invention discloses a method and system for two-dimensional geometric measurement of tumors using imperfect masks, comprising: acquiring a medical image slice to be identified and its corresponding tumor mask; constructing a global image branch and a target region branch, and extracting features based on the medical image slice to be identified and its corresponding tumor mask, wherein the global image branch outputs global features and the target region branch outputs local features; introducing a semantic query vector and a quantized query vector, and combining the global and local features to obtain semantic features and geometric features to determine fusion features; and determining the tumor category and geometric measurement value corresponding to the medical image slice to be identified based on the fusion features. Thus, even when the tumor mask has imperfections such as boundary deviation and noise, the tumor category and geometric measurement value can be stably output without relying on the precise boundary of the mask, effectively improving the accuracy of type identification and the stability of geometric measurement.

Description

Technical Field

[0001] This invention relates to the fields of medical imaging and artificial intelligence technology, and in particular to a two-dimensional geometric measurement method and system for tumors oriented towards imperfect masks. Background Technology

[0002] In clinical diagnosis and treatment of tumors, tumor type identification and two-dimensional geometric measurement based on medical imaging are the core basis for efficacy evaluation and follow-up. For example, by measuring geometric indicators such as the longest diameter, vertical short diameter, and lesion area of ​​the tumor, the size change of the lesion can be judged, thereby evaluating the treatment effect and guiding the adjustment of the treatment plan.

[0003] Current mainstream automated tumor geometric measurement methods mainly include cascaded "segmentation-measurement" methods and direct regression methods. Cascaded methods first segment the tumor region in medical images using a segmentation model to obtain a tumor mask, and then calculate geometric measurement parameters based on the mask boundaries. This approach is highly dependent on the precise boundaries of the mask; however, differences in scanning parameters across different medical centers, variations in radiologists' annotation habits and subjective judgments, and prediction errors in edge details by automated segmentation models can all lead to defects in the mask such as boundary deviations, noise, and holes. Direct regression methods, while bypassing the mask, completely deviate from the tumor anatomical semantic information provided by the mask, lacking clear anatomical constraints and failing to identify the tumor subregion corresponding to the measurement indicators, resulting in poor interpretability. Therefore, masks are prone to imperfections such as boundary deviations and noise, and traditional methods rely on the accuracy of mask boundaries, leading to unstable tumor identification and measurement results. Summary of the Invention

[0004] This invention provides a two-dimensional geometric measurement method and system for tumors using imperfect masks, in order to solve the defect in the prior art of "unstable identification and measurement results due to insufficient mask accuracy".

[0005] This invention provides a two-dimensional geometric measurement method for tumors oriented towards imperfect masks, comprising: Obtain the medical image slice to be identified and the corresponding tumor mask; Construct a global image branch and a target region branch, and extract global features and local features based on the medical image slice to be identified and the corresponding tumor mask, respectively. The global image branch outputs global features, and the target region branch outputs local features. A semantic query vector and a quantized query vector are introduced, and semantic features and geometric features are obtained by combining the global features and the local features, so as to determine the fusion features; Based on the fusion features, the tumor category and geometric measurement value corresponding to the medical image slice to be identified are determined.

[0006] According to the present invention, a two-dimensional geometric measurement method for tumors oriented towards imperfect masks is provided, wherein the global features contain tumor context information and the local features contain information about the region of interest.

[0007] According to the present invention, a two-dimensional geometric measurement method for tumors oriented towards imperfect masks, the step of outputting global features specifically includes: The medical image slice to be identified is input into a visual encoder to extract global visual features, and the global visual features are standardized by linear projection. The standardized global visual features are combined with the masking information of the tumor mask to obtain global features.

[0008] According to the present invention, a two-dimensional geometric measurement method for tumors oriented towards imperfect masks, the step of outputting local features specifically includes: The tumor region of interest information is determined based on the medical image slice to be identified and the tumor mask. The tumor region of interest information is input into a visual encoder to extract local geometric features, and the local geometric features are standardized by linear projection. The standardized local geometric features are combined with the positional codes generated based on the tumor mask to obtain local features.

[0009] According to the present invention, a two-dimensional geometric measurement method for tumors oriented towards imperfect masks includes the following steps: introducing a semantic query vector and a quantized query vector, and combining the global features and the local features to obtain semantic features and geometric features. Set up multiple semantic query vectors corresponding to the number of tumor subregion categories, and multiple quantization query vectors corresponding to the number of geometric measurement parameters; The global features, the local features, multiple semantic query vectors, and the quantization query vector are input into the adapter, and interact through a multi-head cross-attention mechanism to output semantic features and quantization features. The semantic features are used for semantic classification tasks, and the quantization features are used for geometric quantization regression tasks.

[0010] According to the present invention, a two-dimensional geometric measurement method for tumors oriented towards imperfect masks is provided, wherein the step of determining the fusion features specifically includes: The semantic features and the quantization features interact through a bidirectional cross-attention mechanism, wherein the first direction uses the semantic features as the query and the quantization features as the key and value, and the second direction uses the quantization features as the query and the semantic features as the key and value; The semantic features obtained after interaction are concatenated and fused with the quantization features to obtain fused features.

[0011] According to the present invention, a two-dimensional geometric measurement method for tumors oriented towards imperfect masks, the step of determining the tumor category and geometric measurement value corresponding to the medical image slice to be identified based on the fusion features specifically includes: The fused features are input into the semantic classification head and the geometric regression head, respectively. The semantic classification head is used to output the tumor category corresponding to the medical image slice to be identified, and the geometric regression head is used to output the geometric measurement value of the tumor region corresponding to the medical image slice to be identified.

[0012] The present invention also provides a two-dimensional geometric measurement system for tumors oriented towards imperfect masks, comprising: The data acquisition module is used to acquire the medical image slices to be identified and the corresponding tumor masks; The dual-branch feature extraction module is used to construct a global image branch and a target region branch, and extract global features and local features based on the medical image slice to be identified and the corresponding tumor mask, respectively. The global image branch outputs global features, and the target region branch outputs local features. The dual-query interaction and fusion module is used to introduce semantic query vectors and quantized query vectors, and combine the global features and the local features to obtain semantic features and geometric features, so as to determine the fusion features; The classification and regression output module is used to determine the tumor category and geometric measurement value corresponding to the medical image slice to be identified based on the fusion features.

[0013] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the two-dimensional geometric measurement method for tumors oriented towards imperfect masks as described above.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the tumor two-dimensional geometric measurement method oriented towards imperfect masks as described above.

[0015] The tumor two-dimensional geometric measurement method provided by this invention for imperfect masks, by constructing global image branches and target region branches, introducing and fusing semantic and quantization dual query vectors, can stably output tumor category and geometric measurement values ​​without relying on the precise boundary of the mask, even when the tumor mask has imperfections such as boundary deviation and noise. This effectively improves the accuracy of type recognition and the stability of geometric measurement, and avoids the problem of unstable recognition and measurement results caused by insufficient mask accuracy in traditional methods. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the two-dimensional geometric measurement method for tumors oriented towards imperfect masks provided by the present invention. Figure 2 This is a framework diagram of the tumor two-dimensional geometric measurement method oriented towards imperfect masks provided by the present invention; Figure 3 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0018] The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and should not be construed as limiting the scope of the invention.

[0019] In the description of this invention, it should be noted that the terms "center," "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0020] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "connected" and "linked" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0021] See Figure 1 This application provides a two-dimensional geometric measurement method for tumors using imperfect masks, comprising the following steps: S110: Obtain the medical image slice to be identified and the corresponding tumor mask; S120: Construct a global image branch and a target region branch, and extract global features and local features based on the medical image slice to be identified and the corresponding tumor mask, wherein the global image branch outputs global features and the target region branch outputs local features; S130: Introduce semantic query vectors and quantized query vectors, and combine global and local features to obtain semantic and geometric features in order to determine the fusion features; S140: Determine the tumor category and geometric measurement value corresponding to the medical image slice to be identified based on the fusion features.

[0022] For two-dimensional geometric measurement of tumors, the "segmentation-measurement" cascade approach is currently the most commonly used measurement paradigm in clinical practice and research. However, this method has the following drawbacks: ① It is highly sensitive to mask accuracy. In actual clinical scenarios, masks may come from different annotators, different medical centers, or different segmentation models, resulting in issues such as boundary deviations, noise, and annotation differences. These minute deviations are directly amplified, leading to drastic fluctuations in measurement results. ② This two-stage process separates the original image information from the mask's prior information, ignoring key structural information such as tumor edges, textures, and scale in the original medical images, further reducing the stability and reliability of the measurement results. ③ The mask post-processing optimization stage introduces additional structural biases, and in scenarios with blurred tumor boundaries, mask optimization itself is uncertain, failing to fundamentally solve the measurement robustness problem.

[0023] Therefore, traditional methods heavily rely on mask accuracy. This application fully considers the objective factors in real-world clinical scenarios, such as equipment differences, annotator subjectivity, and blurred tumor boundaries, which generally result in imperfections in tumor masks, including boundary deviations and noise. Under these conditions, this application first obtains the medical image slice to be identified and its corresponding tumor mask, allowing for imperfections such as boundary deviations and noise common in real-world scenarios. Next, it constructs a global branch to extract global features containing the tumor context and a target region branch to extract local features focusing on local morphology, providing complementary information for subsequent fusion to enhance the stability of feature expression without relying on a perfect mask. Then, it introduces semantic query vectors and quantized query vectors to model semantic and geometric features respectively and fuse them interactively, achieving decoupling and complementarity between semantics and geometry. This maintains accuracy in category recognition and stability in geometric measurement even with imperfect masks. Finally, it outputs both the tumor category and geometric measurement values ​​based on the fused features. Thus, this application solves the technical problems of low tumor type recognition accuracy and large fluctuations in geometric measurement results caused by insufficient mask accuracy in traditional methods, achieving stable output of tumor category and geometric measurement values ​​even under imperfect mask conditions.

[0024] In S110, the medical image slice (MRI slice) is derived from a single tomographic plane in the three-dimensional volume data obtained by magnetic resonance imaging equipment scanning. The tumor mask comes from manual annotation or automatic segmentation models (such as deep learning segmentation networks). Based on clinical reality, this application does not have strict requirements for mask accuracy and allows common imperfections such as boundary noise and annotation differences.

[0025] After obtaining the medical image slices to be identified and the tumor mask, the process also includes standardizing the input medical image slices to be identified and the tumor mask to ensure consistency in subsequent feature extraction.

[0026] Preprocessing specifically includes: Image normalization: normalizing the pixel values ​​of medical image slices to the [0,1] range to eliminate pixel value differences caused by different devices and different scanning parameters.

[0027] Unified resolution: The image and mask are scaled to a fixed resolution (e.g., 256×256) to avoid the impact of resolution differences on feature extraction and measurement results.

[0028] Format conversion: Converts single-channel medical images into a three-channel format to meet the input requirements of the visual encoder.

[0029] Mask preprocessing: The tumor mask is simply binarized to retain only the distinction between the tumor region and the background, without the need for fine boundary optimization, thus simulating the imperfect masking scenario in clinical practice.

[0030] In S120, a dual-branch feature extraction structure is constructed, which includes a global image branch and a target region branch. The global features contain tumor context information, and the local features contain information about the region of interest. Thus, the tumor context information is extracted through the global image branch, and the local morphological information is extracted through the target region branch. The two complement each other, and a stable feature representation is obtained without relying on a perfect mask.

[0031] Global Image Branch: Taking a complete MRI slice as input, it incorporates the mask as a spatial cue and is responsible for capturing the global location of the tumor, surrounding anatomical structures, and contextual semantic features for tumor subregion identification and semantic constraints. Target Region Branch: Taking a mask-weighted region of interest (ROI) as input, it focuses on the local morphology, scale, and edge details of the tumor and is specifically used for geometric feature learning and accurate measurement, reducing the impact of mask boundary deviation.

[0032] For example, see Figure 2This method employs a pre-trained visual encoder (such as the visual backbone network of CLIP) as a shared feature extractor. During training, it learns general visual features on large-scale image datasets, enabling efficient extraction of deep representations such as edges, textures, and shapes from medical images. The visual encoder is used to extract features from the dual-branch input separately. (1) Global Image Branch Input: The preprocessed complete medical image slices are input into the visual encoder to extract global visual features. The global visual features are then standardized by linear projection. Mask information is injected through a mask generator and combined with the standardized global visual features to obtain global features, which are used to capture the anatomical context and global semantic information around the tumor. The mask generator is used to obtain mask information based on the input tumor mask, providing a rough regional location constraint for the tumor and guiding feature extraction to focus on the tumor region without relying on the precise boundary of the mask.

[0033] (2) Target Region Branch Input: The region of interest (ROI) of the tumor is determined by the medical image slice to be identified and the tumor mask. The ROI is then input into the visual encoder to extract local geometric features. These local geometric features are standardized by linear projection. The standardized local geometric features are then combined with the positional encoding to obtain local features, which are used to enhance the perception of tumor morphology, scale, and edge structure. By encoding the positional information into vector form and injecting it into the local geometric features, the absolute or relative spatial position of the tumor in the original image can be provided, so that the spatial context is not lost while focusing on the local morphology.

[0034] It should be noted that the extracted global visual features and local geometric features are mapped to a unified dimension through linear projection to facilitate cross-attention interaction in the subsequent adapter (Q-Former).

[0035] In S130, the steps of introducing semantic query vectors and quantization query vectors, and combining global and local features to obtain semantic and geometric features include: setting multiple semantic query vectors corresponding to the number of tumor sub-region categories, and multiple quantization query vectors corresponding to the number of geometric measurement parameters; inputting global features, local features, multiple semantic query vectors, and quantization query vectors into the adapter, interacting through a multi-head cross-attention mechanism, and outputting semantic and quantization features; wherein, semantic features are used for semantic classification tasks, and quantization features are used for geometric quantization regression tasks.

[0036] In this step, two independent sets of query vectors, semantic query vector and quantization query vector, are introduced. Through Q-Former, cross-modal interaction with global and local features is achieved to decouple semantic features from geometric features, thus balancing classification accuracy and measurement stability.

[0037] The semantic query vector is used for tumor semantic classification tasks, focusing on learning tumor category and subregion information (such as the enhancement area (ET), edema area (ED), and necrosis area (NCR) of gliomas), providing clear anatomical constraints for geometric measurements and preventing measurement results from deviating from the target region. The quantized query vector is used for tumor geometric measurement regression tasks, focusing on learning geometric measurement information such as the longest diameter, shortest diameter, and area of ​​the tumor, concentrating on extracting tumor morphological and scale features to ensure the accuracy of measurement parameters. The Q-Former employs a multi-head cross-attention-based interaction structure to achieve precise information interaction between the query vector and global and local features. Through attention weight allocation, it focuses on extracting task-relevant features, suppressing irrelevant noise interference, and improving feature effectiveness.

[0038] For example, two independent sets of query vectors are initialized. The semantic query vectors are set to N (N corresponds to the number of tumor sub-region categories, such as 3 for glioma, corresponding to the enhancement area, edema area, and necrosis area respectively). The number of quantization query vectors corresponds to the number of geometric parameters to be measured (e.g., longest diameter, shortest diameter, area). Adapter interaction: The standardized global features, local features, and the two sets of query vectors are input into the Q-Former adapter, and interaction is achieved through a multi-head cross-attention mechanism. The attention weights are mainly allocated to features related to the tumor region to suppress background noise interference. After interaction, semantic features and quantization features are obtained, corresponding to semantic classification and geometric measurement tasks respectively.

[0039] Furthermore, the steps for determining the fusion features specifically include: interacting semantic features and quantization features through a bidirectional cross-attention mechanism, wherein the first direction uses semantic features as the query and quantization features as the key and value, and the second direction uses quantization features as the query and semantic features as the key and value; and concatenating and fusing the interacting semantic features and quantization features to obtain the fusion features.

[0040] In this step, a bidirectional cross-attention mechanism is used to achieve complementary enhancement of semantic and quantitative features, establishing a bidirectional information transmission channel between them to further improve measurement robustness and interpretability. Specifically, semantic feature guidance uses semantic features as queries and quantitative features as keys and values, enabling the semantic representation to perceive tumor geometric structure information and ensuring that geometric measurements always revolve around the tumor target region and corresponding subregions, avoiding measurement bias. Quantitative feature assistance uses quantitative features as queries and semantic features as keys and values, introducing tumor category-related constraints into the quantitative representation during the update process, improving measurement stability when mask quality is poor, and simultaneously enhancing the clinical interpretability of the measurement results.

[0041] Specifically, bidirectional information fusion is achieved through a bidirectional cross-attention mechanism: (1) First-direction interaction: Using semantic features as the query and quantitative features as the key and value, attention calculation is used to integrate semantic features into geometric structure information and clarify the differences in geometric features of different sub-regions; (2) Second-direction interaction: Using quantitative features as queries and semantic features as keys and values, attention calculation is used to introduce semantic constraints into the quantitative features to ensure that the measurement results correspond to the correct tumor subregions; (3) Fusion feature output: The semantic features and quantization features after bidirectional interaction are concatenated and fused through a fully connected layer to obtain the final fused feature representation.

[0042] In S140, the step of determining the tumor category and geometric measurement value corresponding to the medical image slice to be identified based on the fusion features specifically includes: inputting the fusion features into the semantic classification head and the geometric regression head respectively; the semantic classification head is used to output the tumor category corresponding to the medical image slice to be identified; and the geometric regression head is used to output the geometric measurement value of the tumor region corresponding to the medical image slice to be identified.

[0043] In this step, based on the fused features, a dual-task output structure is constructed to achieve tumor semantic classification and geometric measurement respectively. Multiple constraints are applied simultaneously to ensure that the output results meet actual clinical needs. (1) Semantic classification head: The fused features are input into the semantic classification head (fully connected layer + Softmax activation function) to output the tumor category and sub-region category. The classification loss is calculated using the cross-entropy loss function for model optimization, providing clear anatomical constraints for geometric measurement and enhancing the interpretability of the measurement results.

[0044] (2) Geometric Regression Head: The fused features are input into the geometric regression head (fully connected layer), which outputs geometric quantification parameters such as the longest diameter of the tumor, the shortest vertical diameter, and the lesion area. The regression loss is calculated using a smoothed L1 loss function for model optimization. At the same time, dual constraints are applied to the regression output: ① Non-negativity constraint, ensuring that the measured values ​​conform to the physical rules of geometric measurement, that is, the quantification parameters output by the geometric regression head are processed by the non-negativity constraint function to satisfy the physical constraint that the geometric measurement values ​​are non-negative; ② Geometric consistency constraint, ensuring that the longest diameter ≥ the shortest diameter, and that the area calculation and diameter parameter maintain logical consistency.

[0045] Within the reasoning framework described above, model training is performed. A training sample set is prepared, with each sample containing a medical image slice, a corresponding tumor mask, and two types of supervision labels: tumor subregion category and ground truth geometric measurement value. During training, the classification loss and regression loss are weighted and summed (the weight ratio can be adjusted according to actual needs, such as 1:1) as the total loss function. The Adam optimizer is used to train the entire model end-to-end until the model converges. After the model training is complete, a new medical image slice to be identified and an imperfect mask are input. After the above process, the final tumor category and geometric measurement parameters are output for clinical follow-up and efficacy evaluation.

[0046] The following describes a tumor two-dimensional geometric measurement system oriented towards imperfect masks provided by the present invention. The tumor two-dimensional geometric measurement system oriented towards imperfect masks described below can be referred to in correspondence with the tumor two-dimensional geometric measurement method oriented towards imperfect masks described above.

[0047] This application also provides a two-dimensional geometric measurement system for tumors oriented towards imperfect masks, including: The data acquisition module is used to acquire the medical image slices to be identified and the corresponding tumor masks; The dual-branch feature extraction module is used to construct a global image branch and a target region branch. Based on the medical image slice to be identified and the corresponding tumor mask, global features and local features are extracted respectively. The global image branch outputs global features, and the target region branch outputs local features. The dual-query interaction and fusion module is used to introduce semantic query vectors and quantized query vectors, and combine global and local features to obtain semantic and geometric features in order to determine the fusion features; The classification and regression output module is used to determine the tumor category and geometric measurement value corresponding to the medical image slice to be identified based on the fusion features.

[0048] Figure 3 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 3 As shown, the electronic device may include a processor 310, a communications interface 320, a memory 330, and a communication bus 340. The processor 310, communications interface 320, and memory 330 communicate with each other via the communication bus 340. The processor 310 can call logical instructions from the memory 330 to execute a two-dimensional geometric measurement method for tumors oriented towards imperfect masks.

[0049] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0050] On the other hand, the present invention also provides a computer program product, the computer program product including a computer program that can be stored on a non-transitory computer-readable storage medium, and when the computer program is executed by a processor, the computer is able to execute the tumor two-dimensional geometric measurement method for imperfect masks provided by the above methods.

[0051] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the tumor two-dimensional geometric measurement method for imperfect masks provided by the methods described above.

[0052] The device embodiments described above are merely illustrative. 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 modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0053] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0054] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention 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; and these 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 the present invention.

Claims

1. A two-dimensional geometric measurement method for tumors oriented towards imperfect masks, characterized in that, include: Obtain the medical image slice to be identified and the corresponding tumor mask; Construct a global image branch and a target region branch, and extract global features and local features based on the medical image slice to be identified and the corresponding tumor mask, respectively. The global image branch outputs global features, and the target region branch outputs local features. A semantic query vector and a quantized query vector are introduced, and semantic features and geometric features are obtained by combining the global features and the local features, so as to determine the fusion features; Based on the fusion features, the tumor category and geometric measurement value corresponding to the medical image slice to be identified are determined.

2. The two-dimensional geometric measurement method for tumors oriented towards imperfect masks according to claim 1, characterized in that, The global features contain tumor context information, and the local features contain information about the region of interest.

3. The two-dimensional geometric measurement method for tumors oriented towards imperfect masks according to claim 1, characterized in that, The step of outputting global features specifically includes: The medical image slice to be identified is input into a visual encoder to extract global visual features, and the global visual features are standardized by linear projection. The standardized global visual features are combined with the masking information of the tumor mask to obtain global features.

4. The two-dimensional geometric measurement method for tumors oriented towards imperfect masks according to claim 1, characterized in that, The step of outputting local features specifically includes: The tumor region of interest information is determined based on the medical image slice to be identified and the tumor mask. The tumor region of interest information is input into a visual encoder to extract local geometric features, and the local geometric features are standardized by linear projection. The standardized local geometric features are combined with the positional codes generated based on the tumor mask to obtain local features.

5. The two-dimensional geometric measurement method for tumors oriented towards imperfect masks according to claim 1, characterized in that, The steps of introducing semantic query vectors and quantized query vectors, and combining the global features and the local features to obtain semantic features and geometric features specifically include: Set up multiple semantic query vectors corresponding to the number of tumor subregion categories, and multiple quantization query vectors corresponding to the number of geometric measurement parameters; The global features, the local features, multiple semantic query vectors, and the quantization query vector are input into the adapter, and interact through a multi-head cross-attention mechanism to output semantic features and quantization features. The semantic features are used for semantic classification tasks, and the quantization features are used for geometric quantization regression tasks.

6. The two-dimensional geometric measurement method for tumors oriented towards imperfect masks according to claim 1, characterized in that, The step of determining the fusion features specifically includes: The semantic features and the quantization features interact through a bidirectional cross-attention mechanism, wherein the first direction uses the semantic features as the query and the quantization features as the key and value, and the second direction uses the quantization features as the query and the semantic features as the key and value; The semantic features obtained after interaction are concatenated and fused with the quantization features to obtain fused features.

7. The two-dimensional geometric measurement method for tumors oriented towards imperfect masks according to claim 1, characterized in that, The step of determining the tumor category and geometric measurement value corresponding to the medical image slice to be identified based on the fusion features specifically includes: The fused features are input into the semantic classification head and the geometric regression head, respectively. The semantic classification head is used to output the tumor category corresponding to the medical image slice to be identified, and the geometric regression head is used to output the geometric measurement value of the tumor region corresponding to the medical image slice to be identified.

8. A two-dimensional geometric measurement system for tumors oriented towards imperfect masks, characterized in that, include The data acquisition module is used to acquire the medical image slices to be identified and the corresponding tumor masks; The dual-branch feature extraction module is used to construct a global image branch and a target region branch, and extract global features and local features based on the medical image slice to be identified and the corresponding tumor mask, respectively. The global image branch outputs global features, and the target region branch outputs local features. The dual-query interaction and fusion module is used to introduce semantic query vectors and quantized query vectors, and combine the global features and the local features to obtain semantic features and geometric features, so as to determine the fusion features; The classification and regression output module is used to determine the tumor category and geometric measurement value corresponding to the medical image slice to be identified based on the fusion features.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the tumor two-dimensional geometric measurement method oriented towards imperfect masks as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the tumor two-dimensional geometric measurement method oriented towards imperfect masks as described in any one of claims 1 to 7.

Images