Breast cancer molecular subtype evaluation system based on unsupervised learning and related equipment

Abstract

The invention relates to the technical field of computer-aided medicine, and provides a breast cancer molecular subtype evaluation system based on unsupervised learning and related equipment.The breast cancer molecular subtype evaluation system comprises an acquisition module used for acquiring and processing ultrasonic image data to obtain processed data; the extraction module is used for performing feature extraction on the preprocessed data to obtain image features; the analysis module is used for analyzing the image features to obtain classification clusters of classification identifiers of breast cancer molecular subtypes; the grouping module is used for grouping the classification identifiers to obtain a breast cancer induction area and a lesion probability; the matching module is used for determining and analyzing a potential breast cancer induction source based on the breast cancer induction area to obtain an onset risk probability; and the evaluation module is used for evaluating the onset risk probability and generating an onset risk report. By means of the system, the problems that a traditional method depends on complex biological sample treatment and a long experimental period, the diagnosis speed is reduced, and the processing efficiency is low when large-scale data are processed are solved.

Description

Technical Field

[0001] The present application relates to the technical field of computer-assisted medicine, and in particular to a breast cancer molecular subtype assessment system based on unsupervised learning and related equipment. Background Art

[0002] As one of the most common cancers in women worldwide, the diagnosis and treatment of breast cancer has become the focus of medical research. Due to the complex pathogenesis of breast cancer and the different pathological characteristics of each patient, how to accurately and quickly identify the molecular subtypes of breast cancer is of great significance for customized treatment plans. Therefore, it is particularly important to develop an efficient breast cancer molecular subtype assessment system.

[0003] Among the relevant technical means, the use of gene expression profiling and immunohistochemistry to evaluate the molecular subtypes of breast cancer has achieved certain results. Through these methods, doctors can classify tumors according to their biological characteristics and then choose the most appropriate treatment strategy. For example, patients with ER-positive and HER2-negative breast cancer may be more suitable for hormone therapy rather than chemotherapy. The application of these technologies has significantly improved the individualization and precision of treatment.

[0004] Regarding the above technical solutions, although the molecular subtypes of breast cancer can be accurately determined through gene expression profiles and immunohistochemistry, in terms of data acquisition and processing efficiency, traditional methods rely on complex biological sample processing and long experimental cycles, which reduces the diagnosis speed and has the problem of low processing efficiency when processing large-scale data. Summary of the invention

[0005] In order to improve the traditional methods that rely on complex biological sample processing and long experimental cycles, reduce the diagnosis speed, and have low processing efficiency when processing large-scale data, the present application provides a breast cancer molecular subtype assessment system and related equipment based on unsupervised learning.

[0006] The present invention provides a breast cancer molecular subtype assessment system based on unsupervised learning, comprising: an acquisition module, used to acquire ultrasound image data of breast cancer patients, preprocess the ultrasound image data, and obtain preprocessed ultrasound image data; an extraction module, used to perform feature extraction on the preprocessed ultrasound image data through a preset unsupervised learning model, and obtain a number of specific image features; an analysis module, used to perform cluster analysis on all the specific image features through a self-organizing mapping network algorithm, and obtain classification clusters of classification identifiers of a number of breast cancer molecular subtypes; a grouping module, used to classify all the breast cancer molecular subtypes The classification identifiers are grouped to obtain a breast cancer molecular subtype group, and the breast tissue status of each classification cluster in the breast cancer molecular subtype group is analyzed by a spatiotemporal neural network model to obtain a breast cancer inducing area and a lesion probability; a matching module is used to match the breast cancer inducing area with a preset clinical database to obtain a matching result, determine a potential breast cancer inducing source according to the matching result, perform a multidimensional feature analysis on the potential breast cancer inducing source and the lesion probability to obtain a disease risk probability; an evaluation module is used to evaluate the disease risk probability to obtain an evaluation result, and generate a disease risk report according to the evaluation result.

[0007] As a preferred solution, the step of acquiring ultrasound image data of breast cancer patients, preprocessing the ultrasound image data, and obtaining preprocessed ultrasound image data includes: acquiring ultrasound image data of breast cancer patients based on a hospital system library, performing noise reduction filtering on the ultrasound image data to obtain noise-reduced ultrasound image data, refining and identifying pixel features of the noise-reduced ultrasound image data to obtain pixel texture information; comparing the pixel texture information with texture features of a reference ultrasound image to obtain valid ultrasound image data; performing image enhancement processing on the valid ultrasound image data to obtain enhanced image data; and segmenting the enhanced image data according to a preset segmentation standard to obtain preprocessed ultrasound image data.

[0008] As a preferred solution, the step of extracting features from the preprocessed ultrasound image data through a preset unsupervised learning model to obtain a number of specific image features includes: extracting features from the breast area in the preprocessed ultrasound image data through a convolutional neural network to obtain a number of initial image features; analyzing the breast tissue in the preprocessed ultrasound image data using the initial image features to obtain a breast analysis result, and extracting feature vectors related to breast hyperplasia and tissue morphological changes in the breast analysis result using principal component analysis and an autoencoder; inputting the feature vectors into an unsupervised learning model for aggregation analysis to obtain a number of target image features; classifying the corresponding breast tissue in the preprocessed ultrasound image data based on the target image features, and generating a corresponding feature matrix according to the classification results; wherein the breast tissue includes healthy tissue, suspicious lesion area, and cancerous area; and performing association analysis using the image features in the feature matrix and preset molecular data to obtain a number of specific image features related to breast cancer molecules; wherein the molecular data includes gene expression data and protein expression data.

[0009] As a preferred embodiment, the step of performing cluster analysis on all the specific image features through a self-organizing map network algorithm to obtain classification identifications of several breast cancer molecular subtypes includes: performing unsupervised clustering processing on all breast areas in the preprocessed ultrasound image data according to the specific image features to obtain initial clustering results; optimizing the initial clustering results using a self-organizing map network algorithm to obtain several breast tissue clusters, and performing position mapping analysis on each of the breast tissue clusters to obtain spatial distribution information; matching the spatial distribution information with a database of known molecular subtypes of breast cancer to obtain feature distributions corresponding to different molecular subtypes, and generating classification identifications of several breast cancer molecular subtypes based on the feature distributions.

[0010] As a preferred embodiment, the steps of grouping the classification identifiers of all the breast cancer molecular subtypes to obtain a breast cancer molecular subtype group, analyzing each breast tissue state of the breast cancer molecular subtype group by a spatiotemporal neural network model, and obtaining a breast cancer inducing area and a lesion probability include: grouping the classification identifiers of the breast cancer molecular subtypes to obtain a grouping result, and screening a breast cancer molecular subtype group with similar lesion characteristics according to the grouping result; inputting the breast cancer molecular subtype group into the spatiotemporal neural network model to perform an evolution analysis of the breast tissue state, and performing a time series analysis on the proliferation rate and lesion characteristics of each breast cancer molecular subtype in the breast cancer molecular subtype group according to the output result of the spatiotemporal neural network model to obtain a proliferation trend of breast tissue; inputting the proliferation trend into a preset standard lesion model to obtain a breast cancer inducing area, and predicting the probability of the breast cancer inducing area developing into cancer according to the proliferation rate and morphological characteristics of the breast cancer inducing area to obtain a lesion probability.

[0011] As a preferred embodiment, the steps of matching the breast cancer inducing area with a preset clinical database to obtain a matching result, determining a potential breast cancer inducing source according to the matching result, and performing a multi-dimensional feature analysis on the potential breast cancer inducing source and the lesion probability to obtain a risk probability of disease, include: performing position and feature matching on the breast cancer inducing area with a known cancerous area in a clinical database to obtain a matching result; using the matching result to verify the clinical manifestation of each breast cancer inducing area to determine a potential breast cancer inducing source; performing a multi-dimensional feature analysis on the potential breast cancer inducing source and the lesion probability to obtain a multi-dimensional analysis result, performing a quantitative analysis on the multi-dimensional result to obtain a quantitative analysis result; and generating a risk probability of disease for each potential breast cancer inducing source according to the quantitative analysis result.

[0012] As a preferred scheme, the step of evaluating the disease risk probability, obtaining an evaluation result, and generating a disease risk report based on the evaluation result includes: inputting the disease risk probability into a preset evaluation model to obtain an evaluation result, analyzing the onset time window and high-risk level of the breast cancer triggering source according to the evaluation result, classifying the onset time window and the high-risk level to obtain a classification result, and generating a disease risk report for the patient based on the classification result.

[0013] The present application also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, the breast cancer molecular subtype assessment system based on unsupervised learning described in any one of the above is implemented.

[0014] The present application also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the processor executes the breast cancer molecular subtype assessment system based on unsupervised learning as described in any one of the above.

[0015] Compared with the prior art, the present application has the following beneficial effects: fast diagnosis and high efficiency. The unsupervised learning model is used for feature extraction and the self-organizing map network algorithm is used for cluster analysis, allowing the system to automatically identify lesion characteristics and subtype classifications without clear labels, thereby improving the automation and accuracy of the analysis; the application of the spatiotemporal neural network model makes the analysis of the breast tissue status in the classification cluster more in-depth and detailed, and can effectively predict the inducing area and lesion probability of breast cancer, and further enhances the reliability and practicality of the diagnosis results through matching with the preset clinical database and multidimensional feature analysis, while improving the diagnosis speed and reducing the need for biochemical experiments that rely on traditional biological specimens, thereby reducing costs and patient waiting time, improving the traditional method that relies on complex biological sample processing and long experimental cycles, reducing the diagnosis speed, and when processing large-scale data, there is a problem of low processing efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0017] The structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with this technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantive technical significance. Any structural modification, change in proportion or adjustment of size, without affecting the effects and purposes that can be achieved by the present invention, should still fall within the scope of the technical contents disclosed by the present invention.

[0018] Figure 1 is a schematic block diagram of the structure of a breast cancer molecular subtype assessment system based on unsupervised learning provided by an embodiment of the present invention; Figure 2 It is a schematic block diagram of the structure of an electronic device provided by an embodiment of the present invention.

[0019] Description of reference numerals: 10. Breast cancer molecular subtype assessment system based on unsupervised learning; 11. Acquisition module; 12. Extraction module; 13. Analysis module; 14. Grouping module; 15. Matching module; 16. Assessment module; 20. Electronic device; 21. Memory; 22. Processor. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0022] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0023] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0024] The technical solution of the present invention is further described below with reference to the accompanying drawings and through specific implementation methods.

[0025] Embodiment 1: like Figure 1 As shown, the breast cancer molecular subtype assessment system 10 based on unsupervised learning provided in the embodiment of the present application includes an acquisition module 11, an extraction module 12, an analysis module 13, a grouping module 14, a matching module 15 and an assessment module 16.

[0026] The extraction module 12 is used to extract features from the preprocessed ultrasound image data through a preset unsupervised learning model to obtain a number of specific image features.

[0027] In this module, deep learning algorithms such as convolutional neural networks (CNNs) are used to automatically identify and extract key image features, such as shape, size, texture, etc. These features are crucial for subsequent classification and analysis; specifically, the CNN model captures subtle features in breast ultrasound images through a combination of a series of convolutional layers, pooling layers, and fully connected layers. The convolutional layer scans the image through filters of different numbers and sizes to detect basic visual elements such as edges, corners, and spots; the pooling layer is used to reduce the size of the sample, highlight important features through maximum pooling or average pooling, compress data dimensions, and reduce the computational burden. In addition, activation functions such as ReLU (rectified linear unit) can be used to introduce nonlinearity to enhance the expressive power of the model. At the same time, in order to adapt to the characteristics of breast ultrasound images, a shallow network architecture with smaller convolution kernels is designed to extract finer texture and shape information.

[0028] For example, the model can be trained to recognize the typical texture differences between benign and malignant tumors. These texture features appear as different grayscale distributions in ultrasound images. During the training process, a labeled tumor dataset is used, including ultrasound images of various benign and malignant tumors. The model can effectively distinguish benign from malignant tumors by learning the differentiated texture patterns in these images, especially the morphological features and texture details (such as the burr-like edges and uneven echoes common in malignant tumors). During the training process, the convolution kernel size, number of layers, and learning rate parameters of the CNN model are optimized through parameter adjustment and validation set testing to improve its sensitivity to specific texture features and recognition accuracy.

[0029] The analysis module 13 is used to perform cluster analysis on all specific image features through a self-organizing map network algorithm to obtain classification clusters of classification identifiers of several breast cancer molecular subtypes.

[0030] In this module, the self-organizing map network algorithm (SOM) is used to perform unsupervised clustering on the extracted specific image features. The SOM network automatically maps the input features to the low-dimensional network topology through a competitive learning mechanism to form feature clusters, thereby automatically separating different breast cancer molecular subtypes in the feature space; specifically, the topology and initial parameters of the SOM algorithm, such as the learning rate, neighborhood function, and grid size, are configured to ensure the accuracy and convergence of clustering in view of the complexity of breast cancer molecular subtypes. During the training process, the network nodes gradually adjust their positions so that similar image features naturally cluster on adjacent nodes, thereby forming obvious classification clusters in the feature space.

[0031] For example, by adjusting the learning rate, the network can adapt to the feature distribution in the data set more quickly, thereby more accurately classifying similar features into the same classification cluster. Specifically, lowering the learning rate can refine the clustering boundaries and increase sensitivity to feature details and subtle differences; while increasing the neighborhood range can enhance the network's grasp of the global feature distribution. By clustering the ultrasound image data of a large number of breast cancer patients, different molecular subtypes will form different classification clusters based on their unique feature distribution. These classification clusters can accurately reflect the differences in the performance of breast cancer on ultrasound images, thereby providing data support for subsequent pathological analysis and the formulation of personalized treatment plans. At the same time, in order to improve the stability and accuracy of clustering, cross-validation can also be used to adjust the parameters of the SOM network so that it exhibits good generalization ability on the test set.

[0032] The grouping module 14 is used to group the classification identifiers of all breast cancer molecular subtypes to obtain breast cancer molecular subtype groups, and analyze the breast tissue status of each classification cluster in the breast cancer molecular subtype group through a spatiotemporal neural network model to obtain breast cancer-inducing areas and lesion probabilities.

[0033] In this module, the spatiotemporal neural network model is applied to further analyze the spatiotemporal characteristics of each breast cancer molecular subtype group, so as to evaluate the breast tissue status and its changing trend of each classification cluster in the specific image features; specifically, by constructing time series data, such as multiple ultrasound image records, the spatiotemporal neural network can learn the dynamic changes of breast tissue at different time points and reveal its proliferative or degenerative changes. This network is specially designed to process time series data, combining the advantages of convolutional neural network (CNN) in spatial feature extraction and the ability of long short-term memory (LSTM) network in time series modeling to cope with the complexity of breast tissue status changing over time.

[0034] For example, by analyzing the time series of ultrasound image data, changes in breast tissue before and after treatment can be monitored, providing immediate feedback on the efficacy and prediction of future disease progression. Specifically, using historical treatment data, a set of ultrasound image sequences of patients before, during, and after treatment can be constructed. Through the learning of spatiotemporal neural networks, the network can identify changes in breast tissue at different time points, such as a trend of tumor volume reduction or imaging changes after chemotherapy. This analysis can not only provide doctors with real-time efficacy feedback, but also help predict future disease progression, such as determining whether the treatment is effective and whether the treatment plan needs to be adjusted.

[0035] The matching module 15 is used to match the breast cancer inducing area with the preset clinical database to obtain a matching result, determine the potential breast cancer inducing source according to the matching result, perform multi-dimensional feature analysis on the potential breast cancer inducing source and the lesion probability, and obtain the risk probability of disease.

[0036] In this module, pattern matching technology is used to identify the similarity between breast cancer-inducing areas and known cases in combination with a preset clinical database; specifically, image processing and machine learning techniques are used to compare and analyze the patient's breast ultrasound image features with the features of breast cancer cases marked in the database, including shape, edge features, echo intensity, etc. At the same time, machine learning models such as support vector machines (SVM) or random forests are used for classification analysis to preliminarily screen out possible similar cases, thereby inferring potential inducing sources.

[0037] For example, by comparing the characteristics of breast cancer-inducing areas of specific patients with the characteristics of known cases, it is possible to discover the correlation between certain subtypes of breast cancer and specific genetic markers or lifestyle habits. Specifically, it may be found that patients with a certain type of tumor are more likely to carry a certain gene mutation, or show a higher probability of developing the disease under the influence of specific environmental factors (such as a high-fat diet and lack of exercise). This multi-dimensional comparative analysis can provide an effective basis for subsequent personalized treatment and early intervention.

[0038] The evaluation module 16 is used to evaluate the probability of disease risk, obtain evaluation results, and generate a disease risk report based on the evaluation results.

[0039] In this module, the analysis results of the previous modules are combined, and statistical models and risk assessment algorithms are used to calculate the probability of disease risk and generate a detailed disease risk report; specifically, Bayesian network, logistic regression or decision tree models are used to make full use of the patient's multidimensional data, including imaging features, clinical data and family history, for comprehensive evaluation. Through probability theory and statistical methods, the model can output a comprehensive risk score that indicates the possibility of a patient developing breast cancer in the future.

[0040] For example, by integrating the analysis of the image features of the breast cancer-inducing area, the matching results and the potential inducing source analysis, a personalized risk report can be generated to provide patients with customized prevention and intervention recommendations. Specifically, through the comprehensive evaluation of the image features of the breast cancer-inducing area, the matching results with the clinical database, and the potential inducing source analysis, a personalized risk report can be generated for each patient. The report will list in detail the risk level of breast cancer, the recommended monitoring frequency and related health intervention measures, providing strong support for the formulation of a long-term follow-up plan and helping to reduce the risk of breast cancer.

[0041] In this embodiment, the ultrasonic image data of the breast cancer patient is obtained by the acquisition module 11, and the ultrasonic image data is preprocessed to obtain the preprocessed ultrasonic image data. Next, the extraction module 12 extracts the features of the preprocessed ultrasonic image data through a preset unsupervised learning model, thereby obtaining several specific image features. These specific image features are then clustered by the analysis module 13 using a self-organizing map network algorithm, and the result is that the specific image features are divided into classification clusters of classification identifiers of several breast cancer molecular subtypes. The grouping module 14 groups the classification identifiers of all breast cancer molecular subtypes to form a breast cancer molecular subtype group, and conducts an in-depth analysis of the breast tissue state in each classification cluster through a spatiotemporal neural network model to evaluate the breast cancer inducing area and the probability of lesions. Finally, the matching module 15 matches the breast cancer inducing area with a preset clinical database, and determines the potential breast cancer inducing source based on the matching result, and then performs multidimensional feature analysis on the potential breast cancer inducing source and the probability of lesions to obtain the probability of morbidity risk. The evaluation module 16 is responsible for evaluating the probability of morbidity risk and generating a morbidity risk report based on the evaluation result.

[0042] The use of unsupervised learning models for feature extraction and self-organizing mapping network algorithms for cluster analysis allows the system to automatically identify lesion characteristics and subtype classifications without explicit labels, which improves the automation and accuracy of the analysis. The application of spatiotemporal neural network models makes the analysis of breast tissue status in classification clusters more in-depth and detailed, and can effectively predict the inducing areas and lesion probabilities of breast cancer. In addition, matching with the preset clinical database and multidimensional feature analysis further enhance the reliability and practicality of the diagnostic results, providing doctors with a scientific basis, which helps to formulate more personalized and accurate treatment plans. The design of the entire system not only improves the speed of diagnosis, but also reduces the need for biochemical experiments that rely on traditional biological specimens, thereby reducing costs and patient waiting time.

[0043] Embodiment 2: In the acquisition module 11, ultrasound image data of breast cancer patients are acquired based on the hospital's system library, noise reduction filtering is performed on the ultrasound image data to obtain noise-reduced ultrasound image data, and pixel features of the noise-reduced ultrasound image data are refined and identified to obtain pixel texture information.

[0044] By applying advanced noise suppression algorithms, such as wavelet transform, random noise in the image can be effectively reduced; specifically, multi-scale wavelet decomposition is applied to ultrasound image data, and threshold processing technology is used to remove small, non-characteristic noise points, thereby clearly retaining and highlighting the essential features of the image.

[0045] For example, using a bidirectional filter to smooth an image can reduce high-frequency noise in the image without blurring the edges, thereby enhancing the contrast between pixels and better showing the details of breast tissue.

[0046] The pixel texture information is compared with the texture features of the reference ultrasound image to obtain valid ultrasound image data.

[0047] Through the image texture analysis tool, the processed image is compared with the samples in the benchmark library, and the texture similarity is evaluated using texture matching algorithms such as the gray-level co-occurrence matrix (GLCM); specifically, the differences between the processed image and the benchmark image in four texture parameters: contrast, correlation, energy, and uniformity are calculated.

[0048] For example, by comparing the texture roughness of a patient's ultrasound image with a library of images of healthy breast tissue, it is possible to effectively distinguish normal tissue from abnormal tissue, thereby accurately identifying breast areas that require further analysis.

[0049] Perform image enhancement processing on the effective ultrasonic image data to obtain enhanced image data.

[0050] By using image enhancement algorithms such as histogram equalization and contrast-limited adaptive histogram equalization (CLAHE), the local contrast of the image is enhanced; specifically, these enhancement techniques are applied to ultrasound image data to improve the visibility of the image, especially in the detailed representation of breast tissue.

[0051] For example, the CLAHE algorithm is used to perform local contrast enhancement on ultrasound images of breast tissue, making tissue boundaries clearer and facilitating more accurate location of lesion areas during subsequent image analysis and feature extraction.

[0052] The enhanced image data is segmented according to a preset segmentation standard to obtain pre-processed ultrasound image data.

[0053] By applying image segmentation techniques, such as region-based growing algorithms or deep learning segmentation models, key regions in breast images are automatically identified and isolated; specifically, these algorithms are used to precisely cut out breast tissue, tumors, and other structures for further detailed analysis.

[0054] For example, the U-Net convolutional network is used for image segmentation. By learning a large amount of training data, the network can accurately identify different structures of breast tissue, such as tumors and cysts, thereby providing accurate input data for subsequent feature extraction and lesion analysis.

[0055] In the extraction module 12, features of the breast region in the preprocessed ultrasound image data are extracted by a convolutional neural network to obtain a number of initial image features.

[0056] By designing a deep convolutional neural network architecture specifically for the characteristics of ultrasound images, key image features such as edges, textures, and shapes are automatically extracted; specifically, the number of network layers, filter size, and activation function are adjusted to optimize the ability to recognize subtle lesion features.

[0057] For example, by training a CNN model to recognize typical texture patterns of breast cancer, such as linear structures and non-uniform echogenicity, these patterns are crucial for diagnosis in the early stages of cancer.

[0058] The initial image features are used to analyze the breast tissue in the preprocessed ultrasound image data to obtain the breast analysis results. Principal component analysis and autoencoder are used to extract feature vectors related to breast hyperplasia and tissue morphological changes in the breast analysis results.

[0059] By combining principal component analysis (PCA) and autoencoders, the most representative and diagnostically valuable features are screened out from a large number of initial image features; specifically, PCA is used to reduce the dimension and highlight key variables, while the autoencoder is used to learn and reconstruct high-level abstract features of breast image data.

[0060] For example, the main components extracted by PCA reveal the common variations in breast tissue, and the autoencoder can reconstruct fine image features closely related to breast hyperplasia from these variations, thereby providing more accurate data support for diagnosis.

[0061] The feature vector is input into the unsupervised learning model for aggregation analysis to obtain several target image features.

[0062] Through unsupervised learning algorithms, such as k-means clustering or Gaussian mixture models, feature vectors are organized into meaningful groups to reveal the intrinsic correlation between different breast lesions; specifically, the centroid of the cluster is optimized through algorithm iteration to ensure that the obtained image feature group can effectively distinguish healthy, benign and malignant breast tissues.

[0063] For example, by using k-means clustering to process feature vectors, it can be found that the image features of breast cancer lesions are significantly different from those of healthy tissues. This distinction helps in subsequent accurate diagnosis and treatment decisions.

[0064] The corresponding breast tissue in the preprocessed ultrasound image data is classified based on the target image features, and a corresponding feature matrix is ​​generated according to the classification results; wherein the breast tissue includes healthy tissue, suspicious lesion area, and cancerous area.

[0065] Through machine learning classifiers, such as support vector machines (SVMs) or decision trees, breast tissue is classified into different categories based on the target image features; specifically, the breast tissue is evaluated through the trained model and a feature matrix is ​​generated, which details the probability and features of each classification.

[0066] For example, inputting the data in the feature matrix into the decision tree model can accurately determine whether breast tissue has become cancerous and the possibility of cancer. This information is crucial for formulating personalized treatment plans in clinical practice.

[0067] The image features in the feature matrix are used to perform association analysis with preset molecular data to obtain several specific image features related to breast cancer molecules; wherein the molecular data includes gene expression data and protein expression data.

[0068] Through an analysis platform that integrates image features and molecular biology data, correlation analysis and machine learning prediction models are used to identify the relationship between images and molecular markers; specifically, the correspondence between image features and gene and protein expression patterns is analyzed to identify reliable biomarkers.

[0069] For example, association analysis can reveal that specific image features, such as texture heterogeneity of breast tissue, are directly associated with overexpression of certain genes or abnormal protein accumulation. These findings can help further molecular-level diagnosis and the development of targeted treatment strategies.

[0070] In the analysis module 13, unsupervised clustering processing is performed on the breast regions in all pre-processed ultrasound image data according to specific image features to obtain an initial clustering result.

[0071] Advanced unsupervised clustering techniques, such as spectral clustering or hierarchical clustering, are used to perform detailed analysis of specific image features to identify the natural distribution of different breast tissues in the image feature space; specifically, these algorithms are used to analyze and optimize clustering results to ensure clustering accuracy and consistency.

[0072] For example, spectral clustering technology can reveal various subtypes of breast cancer in the feature space. These subtypes are difficult to distinguish using traditional methods, but through careful clustering in high-dimensional feature space, the boundaries between different subtypes can be clearly seen, which helps with accurate classification and subsequent treatment.

[0073] The initial clustering results were optimized using a self-organizing map network algorithm to obtain several breast tissue clusters. Position mapping analysis was performed on each breast tissue cluster to obtain spatial distribution information.

[0074] The clustering results are refined and optimized through the self-organizing map network (SOM) algorithm to form a more accurate and distinct breast tissue clustering map; specifically, SOM adaptively learns the topological relationship of specific image features so that clusters of similar features are spatially close in the network, thereby facilitating the identification and analysis of the spatial distribution of breast tissue.

[0075] For example, SOM can effectively map the clustering differences between early and late stages of breast cancer. This spatial difference shows the image characteristics of different cancer stages, thereby assisting doctors in more accurate pathological grading and treatment planning.

[0076] The spatial distribution information is matched with the database of known molecular subtypes of breast cancer to obtain the characteristic distribution corresponding to different molecular subtypes, and several classification identifiers of breast cancer molecular subtypes are generated based on the characteristic distribution.

[0077] Through advanced database matching technology, the spatial distribution information is compared and analyzed with the preset breast cancer molecular subtype database; specifically, pattern recognition and similarity measurement algorithms are used to ensure that the obtained classification identification can reflect the exact association between breast tissue and specific molecular subtypes.

[0078] For example, by analyzing the characteristic distribution of different breast cancer molecular subtypes, specific molecular markers can be accurately identified. These markers are critical for determining the specific type of cancer and treatment methods, providing more personalized treatment options.

[0079] In the grouping module 14, the classification identifiers of the breast cancer molecular subtypes are grouped to obtain a grouping result, and a breast cancer molecular subtype group with similar pathological characteristics is screened out according to the grouping result.

[0080] By analyzing and processing each classification identifier, similar breast cancer molecular subtypes are grouped together; specifically, data aggregation and grouping algorithms, such as hierarchical clustering or network analysis, are used to form breast cancer molecular subtype groups based on shared features.

[0081] For example, the grouping module 14 can identify breast cancer molecular subtypes with similar proliferation rates and pathological characteristics and classify them into the same group, which helps in the formulation of subsequent targeted treatment and monitoring strategies and ensures the targetedness and effectiveness of the treatment plan.

[0082] The breast cancer molecular subtype group was input into the spatiotemporal neural network model to analyze the evolution of breast tissue status. Based on the output results of the spatiotemporal neural network model, a time series analysis was performed on the proliferation rate and lesion characteristics of each breast cancer molecular subtype in the breast cancer molecular subtype group to obtain the proliferation trend of breast tissue.

[0083] Through the spatiotemporal neural network model, the evolution of breast cancer molecular subtypes is dynamically monitored and predicted by comprehensively considering time and space variables; specifically, the model predicts the proliferation rate of breast tissue and the pathology evolution path by analyzing time series data.

[0084] For example, spatiotemporal models can predict the pathological progression of a certain subtype under specific conditions. For example, rapidly proliferating molecular subtypes of breast cancer show earlier signs of deterioration, providing a time window for early intervention.

[0085] The proliferation trend is input into the preset standard lesion model to obtain the breast cancer inducing area. According to the proliferation speed and morphological characteristics of the breast cancer inducing area, the probability of the breast cancer inducing area developing into cancer is predicted to obtain the lesion probability.

[0086] By integrating multi-source data and predictive models, the risk level of breast cancer-inducing areas can be accurately assessed; specifically, statistical and machine learning methods, such as logistic regression or random forests, are used to comprehensively assess the likelihood of lesions based on proliferation trends and morphological features.

[0087] For example, by quantitatively analyzing the characteristic parameters of breast cancer-inducing areas, such as proliferation rate and morphological changes, the time and probability of the area developing into cancer can be predicted, which is crucial for developing prevention strategies and early treatment.

[0088] In the matching module 15, the breast cancer-inducing region is matched with the known cancerous regions in the clinical database in terms of position and features to obtain a matching result.

[0089] Through efficient image matching and data comparison technology, the image features of breast cancer-inducing areas are compared with known cancer cases stored in the clinical database; specifically, image registration and feature comparison algorithms are used to ensure high accuracy and relevance of the matching.

[0090] For example, comparative analysis can reveal common features between breast cancer-inducing areas and similar cases in the database, thereby verifying diagnostic hypotheses and accurately identifying potential cancer risks, providing strong data support for clinical decision-making.

[0091] The matching results were used to verify the clinical manifestations of each breast cancer-inducing area and identify the potential breast cancer-inducing source.

[0092] By comprehensively analyzing the matching results and clinical manifestations, the breast cancer predisposing areas were studied in depth; specifically, the case-control study method was used to compare the clinical manifestations of potential predisposing sources with those of known cases to ensure the accuracy of diagnosis and clinical relevance.

[0093] For example, this method can be used to confirm the specific type and stage of the breast cancer-inducing area, and further analyze the potential pathological mechanism and treatment response of the area based on clinical data, thereby providing patients with more targeted treatment plans.

[0094] A multidimensional feature analysis is performed on potential breast cancer triggers and lesion probabilities to obtain multidimensional analysis results, and the multidimensional analysis results are quantitatively analyzed to obtain quantitative analysis results.

[0095] Through multidimensional data analysis techniques, such as factor analysis and cluster analysis, various clinical and imaging features are comprehensively considered to form detailed multidimensional analysis results; specifically, these techniques help reveal the intrinsic connections between different data dimensions and improve the accuracy of lesion prediction.

[0096] For example, applying cluster analysis to group together breast cancer-inducing areas with similar characteristics can clearly distinguish lesions of different stages and types. This multidimensional analysis not only helps understand the development process of breast cancer, but also provides more targeted data support for treatment.

[0097] The risk probability of each potential breast cancer risk factor is generated based on the quantitative analysis results.

[0098] Through precise data modeling and probability calculation, the risk of disease in each predisposing area is predicted based on the multidimensional analysis results that have been obtained; specifically, statistical models such as logistic regression are used to evaluate and calculate the probability of disease in these areas.

[0099] For example, applying the quantitative analysis results to a probability model can output a disease risk score for each induced area. This score is based on multiple factors such as lesion characteristics, proliferation rate and clinical history, helping doctors to conduct risk assessment and early intervention.

[0100] In the evaluation module 16, the disease risk probability is input into a preset evaluation model to obtain an evaluation result, and the onset time window and high-risk level of the breast cancer triggering source are analyzed based on the evaluation result. The onset time window and high-risk level are classified to obtain a classification result, and a patient's disease risk report is generated based on the classification result.

[0101] Through advanced risk assessment models, such as risk scoring systems or dynamic prediction models, the individual characteristics of patients and the probability of disease risk are comprehensively considered; specifically, the model analyzes and outputs the potential onset time and risk level of breast cancer triggers, enabling doctors to develop targeted monitoring and intervention strategies.

[0102] For example, dynamic prediction models can be used to predict the time window for future onset of disease based on lesion characteristics and clinical data, which helps doctors develop personalized monitoring plans and preventive measures for high-risk patients. At the same time, the generated disease risk reports can be directly used for patients' disease management and treatment decisions.

[0103] In this embodiment, by comprehensively using image processing technology, machine learning, data analysis and risk assessment technology, in-depth analysis and accurate assessment of breast cancer-inducing areas are achieved. This method not only improves the accuracy of diagnosis and prediction, but also greatly enhances the personalization of treatment plans and the timeliness of preventive measures. Through early identification and risk assessment of breast cancer, this embodiment provides doctors with a powerful decision support tool, thereby improving patients' treatment outcomes and quality of life.

[0104] Embodiment 3: like Figure 2 As shown, the present application also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, the breast cancer molecular subtype assessment system based on unsupervised learning in Example 1 is implemented.

[0105] In this embodiment, by integrating software and hardware, the electronic device uses its processor and memory to implement the breast cancer molecular subtype assessment system based on unsupervised learning described in Example 1. The electronic device can automatically process and analyze ultrasound image data from breast cancer patients. The entire process from acquiring image data to outputting risk assessment reports is executed by the processor in the electronic device, ensuring the efficiency and accuracy of the assessment process. In addition, the device not only improves the processing speed by optimizing algorithms and data processing procedures, but also ensures the accuracy and reliability of data analysis, so that doctors can make accurate diagnoses and formulate personalized treatment plans based on the detailed information provided by the system.

[0106] Embodiment 4: The present application also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the processor executes the breast cancer molecular subtype assessment system based on unsupervised learning as in Example 1.

[0107] In this embodiment, a computer-readable storage medium is provided, in which the computer program of the breast cancer molecular subtype assessment system based on unsupervised learning in Example 1 is stored, so that any compatible processor can realize the functions of the system when executing the program. The provision of such a storage medium allows the system to be deployed in a variety of different electronic devices, which can be installed and run from professional equipment in hospitals to portable medical devices, greatly increasing the accessibility and flexibility of the system. In this way, the computer program can be widely used in various medical environments, increasing the popularity of breast cancer diagnosis technology, and helping to improve the efficiency of early detection and treatment of breast cancer in a wider area.

[0108] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A breast cancer molecular subtype assessment system based on unsupervised learning, characterized in that: include: An acquisition module is used to acquire ultrasound image data of a breast cancer patient, and preprocess the ultrasound image data to obtain preprocessed ultrasound image data; An extraction module, used for performing feature extraction on the preprocessed ultrasound image data through a preset unsupervised learning model to obtain a number of specific image features; An analysis module, used for performing cluster analysis on all the specific image features through a self-organizing map network algorithm to obtain classification clusters of classification identifiers of several breast cancer molecular subtypes; A grouping module is used to group the classification identifiers of all the breast cancer molecular subtypes to obtain a breast cancer molecular subtype group, and analyze the breast tissue status of each classification cluster in the breast cancer molecular subtype group by a spatiotemporal neural network model to obtain a breast cancer-inducing area and a lesion probability; A matching module, used to match the breast cancer inducing area with a preset clinical database to obtain a matching result, determine a potential breast cancer inducing source according to the matching result, perform multi-dimensional feature analysis on the potential breast cancer inducing source and the lesion probability, and obtain a disease risk probability; The evaluation module is used to evaluate the probability of the disease risk, obtain an evaluation result, and generate a disease risk report based on the evaluation result.

2. The breast cancer molecular subtype assessment system based on unsupervised learning according to claim 1, characterized in that: The step of acquiring ultrasound image data of a breast cancer patient, preprocessing the ultrasound image data, and obtaining preprocessed ultrasound image data comprises: Acquire ultrasound image data of breast cancer patients based on the hospital's system library, perform noise reduction filtering on the ultrasound image data to obtain noise-reduced ultrasound image data, and refine and identify pixel features of the noise-reduced ultrasound image data to obtain pixel texture information; Comparing the pixel texture information with texture features of a reference ultrasound image to obtain valid ultrasound image data; Performing image enhancement processing on the valid ultrasonic image data to obtain enhanced image data; The enhanced image data is segmented according to a preset segmentation standard to obtain pre-processed ultrasound image data.

3. The breast cancer molecular subtype assessment system based on unsupervised learning according to claim 1, characterized in that: The step of extracting features from the preprocessed ultrasound image data using a preset unsupervised learning model to obtain a number of specific image features includes: Extracting features of the breast region in the preprocessed ultrasound image data by using a convolutional neural network to obtain a number of initial image features; Analyzing breast tissue in the preprocessed ultrasound image data using the initial image features to obtain a breast analysis result, and extracting feature vectors related to breast hyperplasia and tissue morphological changes in the breast analysis result using principal component analysis and an autoencoder; Inputting the feature vector into an unsupervised learning model for aggregation analysis to obtain a number of target image features; Classifying the corresponding breast tissue in the preprocessed ultrasound image data based on the target image features, and generating a corresponding feature matrix according to the classification results; wherein the breast tissue includes healthy tissue, suspicious lesion area, and cancerous area; The image features in the feature matrix are used to perform association analysis with preset molecular data to obtain a number of specific image features related to breast cancer molecules; wherein the molecular data includes gene expression data and protein expression data.

4. The breast cancer molecular subtype assessment system based on unsupervised learning according to claim 1, characterized in that: The step of performing cluster analysis on all the specific image features by using a self-organizing map network algorithm to obtain classification identifications of several breast cancer molecular subtypes includes: performing unsupervised clustering processing on the breast regions in all the preprocessed ultrasound image data according to the specific image features to obtain an initial clustering result; Optimizing the initial clustering result using a self-organizing map network algorithm to obtain a number of breast tissue clusters, and performing position mapping analysis on each of the breast tissue clusters to obtain spatial distribution information; The spatial distribution information is matched with a database of known molecular subtypes of breast cancer to obtain characteristic distributions corresponding to different molecular subtypes, and a number of classification identifiers of breast cancer molecular subtypes are generated based on the characteristic distributions.

5. The breast cancer molecular subtype assessment system based on unsupervised learning according to claim 1, characterized in that: The step of grouping the classification identifiers of all the breast cancer molecular subtypes to obtain a breast cancer molecular subtype group, and analyzing each breast tissue state of the breast cancer molecular subtype group by a spatiotemporal neural network model to obtain a breast cancer-inducing area and a lesion probability comprises: Grouping the classification identifiers of the breast cancer molecular subtypes to obtain a grouping result, and screening out a breast cancer molecular subtype group with similar pathological characteristics according to the grouping result; Inputting the breast cancer molecular subtype group into the spatiotemporal neural network model to perform an evolution analysis of the breast tissue state, and performing a time series analysis on the proliferation speed and pathological characteristics of each breast cancer molecular subtype in the breast cancer molecular subtype group according to the output results of the spatiotemporal neural network model to obtain the proliferation trend of the breast tissue; The proliferation trend is input into a preset standard lesion model to obtain a breast cancer inducing area. According to the proliferation speed and morphological characteristics of the breast cancer inducing area, the probability of the breast cancer inducing area developing into cancer is predicted to obtain the lesion probability.

6. The breast cancer molecular subtype assessment system based on unsupervised learning according to claim 1, characterized in that: The step of matching the breast cancer inducing region with a preset clinical database to obtain a matching result, determining a potential breast cancer inducing source according to the matching result, and performing a multi-dimensional feature analysis on the potential breast cancer inducing source and the lesion probability to obtain a risk probability of disease comprises: Matching the breast cancer-inducing region with known cancerous regions in a clinical database in terms of position and features to obtain a matching result; Using the matching results to verify the clinical manifestations of each breast cancer-inducing area to determine the potential breast cancer-inducing source; Performing multidimensional feature analysis on the potential breast cancer inducing source and the lesion probability to obtain a multidimensional analysis result, and performing quantitative analysis on the multidimensional result to obtain a quantitative analysis result; The incidence risk probability of each potential breast cancer risk factor is generated according to the quantitative analysis results.

7. The breast cancer molecular subtype assessment system based on unsupervised learning according to claim 1, characterized in that: The step of evaluating the probability of disease risk to obtain an evaluation result, and generating a disease risk report according to the evaluation result comprises: The disease risk probability is input into a preset evaluation model to obtain an evaluation result, the onset time window and high-risk level of the breast cancer triggering source are analyzed according to the evaluation result, the onset time window and the high-risk level are classified to obtain a classification result, and a patient's disease risk report is generated based on the classification result.

8. An electronic device, characterized in that: The system comprises a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, the system for evaluating molecular subtypes of breast cancer based on unsupervised learning according to any one of claims 1 to 7 is implemented.

9. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor, the processor is enabled to execute the breast cancer molecular subtype assessment system based on unsupervised learning as claimed in any one of claims 1 to 7.

Images