A Text Processing-Based Method for Generating Breast Cancer Segmentation Architecture

Abstract

This invention discloses a method for generating a breast cancer segmentation architecture based on text processing, belonging to the field of data processing technology. The method includes the following steps: S1, collecting several medical image articles and processing each article to generate a vector database; S2, processing the initial query statement using the various knowledge bases of the vector database to generate several latest results; S3, using a neural network to select the optimal solution from the latest results to obtain the final neural network architecture, and using the final neural network architecture for breast cancer data processing. This invention constructs a neural network architecture suitable for breast cancer tumor segmentation, greatly reducing the professional knowledge requirements for users and improving the accessibility and application scope of the technology.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, specifically relating to a method for generating a breast cancer segmentation architecture based on text processing. Background Technology

[0002] Breast cancer is one of the most common malignant tumors in women. Regular breast cancer screening is crucial for developing treatment plans and improving survival rates. Currently, ultrasound imaging is widely used in the clinical diagnosis of breast lesions due to its advantages such as high flexibility, non-invasiveness, and low cost. However, the complex ultrasound patterns, similar intensity distributions, variable tumor morphologies, and ambiguous boundaries pose significant challenges to breast tumor segmentation. To address these issues, CNNs (Convolutional Neural Networks), with their powerful nonlinear learning capabilities, have been widely applied to breast cancer tumor segmentation. To adapt to the characteristics of ambiguous boundaries and variable size and shape of target regions in breast ultrasound images, many researchers have designed and proposed various modules through manual neural network design, achieving certain improvements over existing neural network architectures. However, manual neural network architecture design has the following problems: 1. It requires deep professional knowledge. Effectively designing neural network architectures requires a thorough understanding of deep learning theory, different types of network structures (such as convolutional neural networks, recurrent neural networks, graph neural networks, etc.), and various optimization techniques. Designers lacking relevant experience and knowledge may find it difficult to design high-performance networks. 2. Difficulty in finding optimal structures: Human intuition and experience may be insufficient to find optimal or near-optimal network structures in some cases. Automated methods (such as neural architecture search) can explore a wider design space and discover innovative architectures that human designers may overlook. 3. High resource consumption: Manually designing network architectures requires significant human and computational resources, especially when trying multiple different architectures to find the best solution. This is particularly pronounced when resources are limited, restricting the application scope of manual design methods.

[0003] To address the difficulties of manually designing neural network architectures, automated neural architecture search (NAS) has been proposed. While NAS reduces human intervention, it still suffers from the following problems: 1. High computational cost: NAS consumes enormous resources, typically requiring significant computational resources, especially when the search space is large. Training and evaluating a large number of candidate architectures requires extremely high time and computational power. This makes NAS difficult to promote in resource-constrained environments. 2. High time cost: A complete NAS process can take days or even weeks, which is impractical for rapid iteration and real-time applications. 3. Complex search space design: Defining a reasonable search space remains difficult, even for medical professionals unfamiliar with deep learning. An excessively large search space leads to a more complex and time-consuming search process, while an excessively small search space may limit the possibility of finding the optimal architecture. Designing a balanced search space that encompasses sufficient diversity still requires deep expertise and experience.

[0004] In recent years, significant breakthroughs have been achieved in the field of artificial intelligence, particularly in large language models (LLMs). LLMs possess powerful knowledge encoding and storage capabilities, text and code understanding and generation abilities, and reasoning capabilities for complex tasks. Their robust knowledge encoding and storage capabilities, along with their reasoning abilities for complex tasks, make them promising candidates for neural network architecture design. However, general-purpose LLMs lack relevant professional knowledge accumulation. Without corresponding hints and knowledge, they can only generate simple existing architectures based on the foundational knowledge gained during training, such as U-net. When more complex functions are required, LLMs struggle to break free from the existing framework to meet the demands. Summary of the Invention

[0005] To address the problem that LLM struggles to break free from its existing framework when text processing requires more complex functionalities, this invention proposes a method for generating breast cancer segmentation architecture based on text processing.

[0006] The technical solution of this invention is: a method for generating breast cancer segmentation architecture based on text processing, comprising the following steps:

[0007] S1. Collect several medical imaging articles, process each medical imaging article, and generate a vector database;

[0008] S2. Process the initial query statement using the various knowledge bases of the vector database to generate several latest results;

[0009] S3. Use a neural network to select the optimal solution from the latest results to obtain the final neural network architecture, and use the final neural network architecture to process breast cancer data.

[0010] Furthermore, S1 includes the following sub-steps:

[0011] S11. Collect several medical imaging articles and generate a set of knowledge entries;

[0012] S12. Clean the knowledge item set and use the embedding model to vectorize the cleaned knowledge item set.

[0013] S13. Store the set of knowledge items after vectorization into the vector database.

[0014] Furthermore, in S11, the expression for the knowledge item set P is:

[0015] ;

[0016] ;

[0017] In the formula, p1, p2, p3, p i and p I S represents the knowledge entries of medical imaging articles 1, 2, 3, i, and i, respectively. i M represents the summary content of the i-th medical imaging article. i This represents the method content of the i-th medical imaging article. Connect(·) means connecting the summary content with the method content.

[0018] Furthermore, in S13, the expression for the vector database D is:

[0019] ;

[0020] ;

[0021] In the formula, d1, d2, d3, d k and d j V represents the 1st, 2nd, 3rd, kth, and jth knowledge bases of the vector database, respectively. j1 v j2 v j3 and They represent the 1st, 2nd, 3rd and Nth items in the j-th knowledge base, respectively. j N knowledge entries, k represents the total number of knowledge entries in the knowledge base, N j This indicates the total number of knowledge entries contained in the knowledge base.

[0022] Furthermore, S2 includes the following sub-steps:

[0023] S21. Use the various knowledge bases of the vector database to perform semantic matching on the initial query statement and generate semantic matching results;

[0024] S22. Based on the semantic matching results, generate several initial results as a candidate set;

[0025] S23. Optimize each initial result in the candidate set to generate several latest results.

[0026] Furthermore, in S21, the semantic matching results The expression is:

[0027] ;

[0028] ;

[0029] In the formula, r1 represents the first semantic matching result in the knowledge base, r2 represents the second semantic matching result in the knowledge base, and r G This represents the G-th result in the semantic matching database, Q represents the initial query statement, and OptimizePrompt(·) combines the initial query statement with a pre-prepared prompt. This represents the result obtained after feature engineering optimization.

[0030] Furthermore, in S22, the h-th initial result S h The expression is:

[0031] ;

[0032] In the formula, LLM h R represents the large language model corresponding to the h-th initial result. h This represents the matching result between the large language model and the knowledge base corresponding to the h-th initial result, where H represents the number of matching results between the large language model and the knowledge base, and GenerateInitialScheme(·) represents the initial scheme generated by the language model corresponding to the h-th initial result based on the matching result of the corresponding knowledge base.

[0033] Furthermore, in S23, the latest result corresponds to the h-th initial result. The expression is:

[0034] ;

[0035] In the formula, SelectAndBuildScheme(·) means that the latest result corresponding to the h-th initial result corresponds to the large language model selecting a suitable module in the set M and rebuilding a new scheme, where M represents the candidate set composed of the initial scheme set, and H represents the number of sets of matching results between the large language model and the knowledge base.

[0036] Furthermore, the optimal solution S * The expression is:

[0037] ;

[0038] In the formula, S * Let S represent the optimal solution, Score(·) represent the score of the latest result corresponding to the h-th initial result, argmax represent the choice corresponding to the solution that maximizes the score, and S represent the specific solution that maximizes the score.

[0039] The beneficial effects of this invention are:

[0040] (1) The present invention builds a neural network architecture suitable for breast cancer tumor segmentation, which greatly reduces the requirements for users' professional knowledge, improves the accessibility and application scope of the technology, and enables large language models to have sufficient prior knowledge, which is conducive to large language models to perform better in the vertical field of neural network architecture generation. The establishment of decentralized sub-knowledge bases can ensure that the modules in the knowledge base can be fully utilized.

[0041] (2) This invention can improve the accuracy and reliability of architecture generation, and can more comprehensively retrieve and utilize modules in the knowledge base. It avoids the problem of modules being ignored due to the large size of the knowledge base or the limitation of tokens in traditional methods. This invention can also effectively reduce the "illusion" phenomenon that may be generated by a single model, thereby improving the accuracy and reliability of the generated neural network architecture. Attached Figure Description

[0042] Figure 1 A flowchart for generating a text-processing-based breast cancer segmentation architecture. Detailed Implementation

[0043] The embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0044] like Figure 1 As shown, this invention provides a method for generating breast cancer segmentation architecture based on text processing, including the following steps:

[0045] S1. Collect several medical imaging articles, process each medical imaging article, and generate a vector database;

[0046] S2. Process the initial query statement using the various knowledge bases of the vector database to generate several latest results;

[0047] S3. Use a neural network to select the optimal solution from the latest results to obtain the final neural network architecture, and use the final neural network architecture to process breast cancer data.

[0048] In this embodiment of the invention, S1 includes the following sub-steps:

[0049] S11. Collect several medical imaging articles and generate a set of knowledge entries;

[0050] S12. Clean the knowledge item set and use the embedding model to vectorize the cleaned knowledge item set.

[0051] S13. Store the set of knowledge items after vectorization into the vector database.

[0052] In this embodiment of the invention, relevant papers on a specified medical image dataset are collected and downloaded, and stored locally in a processable file format such as PDF. The collected papers undergo data preprocessing. Specifically, a large language model (neural network) first performs coarse extraction on each paper, summarizing the title, abstract, methods, and conclusions to obtain a Summary(S). The methods section of the original paper is then extracted separately to obtain Methods(M), which includes the specific implementation steps of the modules in the paper. Methods(M) are combined with Summary(S) as auxiliary data to obtain the processed set of papers.

[0053] In this embodiment of the invention, in S11, the expression for the knowledge item set P is:

[0054] ;

[0055] ;

[0056] In the formula, p1, p2, p3, p i and p I S represents the knowledge entries of medical imaging articles 1, 2, 3, i, and i, respectively. i M represents the summary content of the i-th medical imaging article. i This represents the method content of the i-th medical imaging article. Connect(·) means connecting the summary content with the method content.

[0057] In this embodiment of the invention, in S13, the expression for the vector database D is:

[0058] ;

[0059] ;

[0060] In the formula, d1, d2, d3, d k and d j V represents the 1st, 2nd, 3rd, kth, and jth knowledge bases of the vector database, respectively. j1 v j2 v j3 and They represent the 1st, 2nd, 3rd and Nth items in the j-th knowledge base, respectively. j N knowledge entries, k represents the total number of knowledge entries in the knowledge base, Nj This indicates the total number of knowledge entries contained in the knowledge base.

[0061] In this embodiment of the invention, S2 includes the following sub-steps:

[0062] S21. Use the various knowledge bases of the vector database to perform semantic matching on the initial query statement and generate semantic matching results;

[0063] S22. Based on the semantic matching results, generate several initial results as a candidate set;

[0064] S23. Optimize each initial result in the candidate set to generate several latest results.

[0065] In this embodiment of the invention, in S21, the semantic matching result... The expression is:

[0066] ;

[0067] ;

[0068] In the formula, r1 represents the first semantic matching result in the knowledge base, r2 represents the second semantic matching result in the knowledge base, and r G This represents the G-th result in the semantic matching database, Q represents the initial query statement, and OptimizePrompt(·) combines the initial query statement with a pre-prepared prompt. This represents the result obtained after feature engineering optimization.

[0069] In this embodiment of the invention, in S22, the h-th initial result S h The expression is:

[0070] ;

[0071] In the formula, LLM h R represents the large language model corresponding to the h-th initial result. h This represents the matching result between the large language model and the knowledge base corresponding to the h-th initial result, where H represents the number of matching results between the large language model and the knowledge base, and GenerateInitialScheme(·) represents the initial scheme generated by the language model corresponding to the h-th initial result based on the matching result of the corresponding knowledge base.

[0072] In this embodiment of the invention, in S23, the latest result corresponds to the h-th initial result. The expression is:

[0073] ;

[0074] In the formula, SelectAndBuildScheme(·) means that the latest result corresponding to the h-th initial result corresponds to the large language model selecting a suitable module in the set M and rebuilding a new scheme, where M represents the candidate set composed of the initial scheme set, and H represents the number of sets of matching results between the large language model and the knowledge base.

[0075] In this embodiment of the invention, the optimal solution S * The expression is:

[0076] ;

[0077] In the formula, S * Let S represent the optimal solution, Score(·) represent the score of the latest result corresponding to the h-th initial result, argmax represent the choice corresponding to the solution that maximizes the score, and S represent the specific solution that maximizes the score.

[0078] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.

Claims

1. A method for generating breast cancer segmentation architecture based on text processing, characterized in that, Includes the following steps: S1. Collect several medical imaging articles, process each medical imaging article, and generate a vector database; S2. Process the initial query statement using the various knowledge bases of the vector database to generate several latest results; S3. Use a neural network to select the optimal solution from the latest results to obtain the final neural network architecture, and use the final neural network architecture to process breast cancer data. S2 includes the following sub-steps: S21. Use the various knowledge bases of the vector database to perform semantic matching on the initial query statement and generate semantic matching results; S22. Based on the semantic matching results, generate several initial results as a candidate set; S23. Optimize each initial result in the candidate set to generate several latest results; In S21, the semantic matching result The expression is: ； ； In the formula, r1 represents the first semantic matching result in the knowledge base, r2 represents the second semantic matching result in the knowledge base, and r G This represents the G-th result in the semantic matching database, Q represents the initial query statement, and OptimizePrompt(·) combines the initial query statement with a pre-prepared prompt. This represents the result obtained after feature engineering optimization; In S22, the h-th initial result S h The expression is: ； In the formula, LLM h R represents the large language model corresponding to the h-th initial result. h This represents the matching result between the large language model and the knowledge base corresponding to the h-th initial result, where H represents the number of matching results between the large language model and the knowledge base, and GenerateInitialScheme(·) represents the initial scheme generated by the language model corresponding to the h-th initial result based on the matching result of the corresponding knowledge base. In S23, the latest result corresponds to the h-th initial result. The expression is: ； In the formula, SelectAndBuildScheme(·) means that the latest result corresponding to the h-th initial result corresponds to the large language model selecting a suitable module in the set M and rebuilding a new scheme, M represents the candidate set composed of the initial scheme set, and H represents the number of sets of matching results between the large language model and the knowledge base; The expression for the optimal solution is: ； In the formula, S * Let S represent the optimal solution, Score(·) represent the score of the latest result corresponding to the h-th initial result, argmax represent the choice corresponding to the solution that maximizes the score, and S represent the specific solution that maximizes the score.

2. The method for generating breast cancer segmentation architecture based on text processing according to claim 1, characterized in that, S1 includes the following sub-steps: S11. Collect several medical imaging articles and generate a set of knowledge entries; S12. Clean the knowledge item set and use the embedding model to vectorize the cleaned knowledge item set. S13. Store the set of knowledge items after vectorization into the vector database.

3. The method for generating breast cancer segmentation architecture based on text processing according to claim 2, characterized in that, In S11, the expression for the knowledge item set P is: ； ； In the formula, p1, p2, p3, p i and p I S represents the knowledge entries of medical imaging articles 1, 2, 3, i, and i, respectively. i M represents the summary content of the i-th medical imaging article. i This represents the method content of the i-th medical imaging article. Connect(·) means connecting the summary content with the method content.

4. The method for generating breast cancer segmentation architecture based on text processing according to claim 2, characterized in that, In S13, the expression for the vector database D is: ； ； In the formula, d1, d2, d3, d k and d j V represents the 1st, 2nd, 3rd, kth, and jth knowledge bases of the vector database, respectively. j1 v j2、 v j3 and These represent the 1st, 2nd, 3rd, and Njth knowledge entries in the j-th knowledge base, respectively, where k represents the total number of knowledge bases, and N represents the total number of knowledge entries. j This indicates the total number of knowledge entries contained in the knowledge base.

Images