Medical ultrasonic image segmentation method and system based on adaptive feature fusion

Abstract

The invention relates to the technical field of medical imaging, in particular to a medical ultrasonic image segmentation method and system based on adaptive feature fusion. The method comprises the following steps: acquiring medical ultrasonic image historical data, and performing ultrasonic medical history feature extraction according to the medical ultrasonic image historical data so as to obtain medical ultrasonic image medical history data; constructing an ultrasonic image disease classification model according to the medical history data of the medical ultrasonic image so as to obtain the ultrasonic image disease classification model; a patient medical record ultrasonic image is obtained, zero pixel points are removed according to the patient medical record ultrasonic image, and therefore a patient non-zero pixel point ultrasonic image is obtained; and performing ultrasonic layered tissue analysis according to the non-zero pixel point ultrasonic image of the patient to obtain tissue lesion area data. The medical ultrasonic image segmentation accuracy is improved based on the medical imaging technology.

Description

Technical Field

[0001] The present invention relates to the field of medical imaging technology, and in particular to a medical ultrasound image segmentation method and system based on adaptive feature fusion. Background Art

[0002] Medical ultrasound imaging is an important imaging technology widely used in clinical diagnosis. Its non-invasiveness and real-time imaging capabilities make it an important means of observing tissue structure and function. Current adaptive feature fusion methods are often optimized for specific types of medical ultrasound images and lack sufficient versatility. This results in different application effects on different types of ultrasound images (such as heart, liver or tumors, etc.), and adaptive feature fusion is still subject to certain subjectivity in the feature selection and fusion process. Different feature selection strategies lead to significant differences in fusion effects, which in turn affect the final segmentation effect. Existing algorithms lack sufficient flexibility in dealing with different lesion types or different imaging conditions. For specific types of tumors or lesions, their adaptive mechanisms cannot effectively capture key features, thus affecting the accuracy of segmentation results. Summary of the invention

[0003] Based on this, it is necessary for the present invention to provide a medical ultrasound image segmentation method and system based on adaptive feature fusion to solve at least one of the above technical problems.

[0004] To achieve the above object, a medical ultrasound image segmentation method based on adaptive feature fusion comprises the following steps:

[0005] Step S1: acquiring medical ultrasound image history data, and extracting ultrasound medical history features based on the medical ultrasound image history data, thereby obtaining medical ultrasound image medical history data; constructing an ultrasound image disease classification model based on the medical ultrasound image medical history data, thereby obtaining an ultrasound image disease classification model;

[0006] Step S2: Acquire the patient's medical record ultrasound image, and remove zero pixels according to the patient's medical record ultrasound image, so as to obtain the patient's non-zero pixel ultrasound image; perform ultrasound layered tissue analysis according to the patient's non-zero pixel ultrasound image, so as to obtain tissue lesion area data;

[0007] Step S3: calculating the lesion area ratio according to the tissue lesion area data and the non-zero pixel point ultrasound image of the patient, thereby obtaining lesion area ratio data; classifying the tissue lesion area data into disease types according to the ultrasound image disease classification model, thereby obtaining tissue lesion type data;

[0008] Step S4: Adaptively fusing the non-zero pixel ultrasound image of the patient according to the tissue lesion type data to obtain adaptive multi-feature data of the tissue lesion type; accurately segmenting the tissue lesion area data according to the adaptive multi-feature data of the tissue lesion type to obtain accurate segmentation data of the tissue lesion area;

[0009] Step S5: Perform multimodal feature analysis based on the precise segmentation data of the tissue lesion area and the lesion area ratio data, so as to obtain multimodal feature data of the tissue lesion and upload it to the medical cloud platform.

[0010] The present invention utilizes a large amount of historical ultrasound image data to extract characteristic information related to medical history and uses it to construct a classification model. The beneficial effect is that a disease characteristic model based on historical data is established, so that researchers can be provided with a variety of medical history information sources based on the image characteristics of previous cases. This model can not only be used for further ultrasound image analysis, but also help researchers identify disease evolution trends from historical data, and provide more accurate background information for subsequent image analysis. By eliminating useless zero pixels, image analysis is more focused on effective information areas, thereby improving the efficiency of data processing. Ultrasonic layered tissue analysis helps researchers conduct in-depth analysis of the specific characteristics of the lesion area, and obtain key tissue lesion area data from the level of tissue structure. This process is particularly important for tissue pathology research and helps to further understand the microstructure of the lesion area, thereby providing more detailed pathological basis. The calculation of the lesion area ratio provides researchers with a quantitative evaluation indicator that can accurately measure the proportion of lesions relative to the overall tissue, which is critical for evaluating the spread and potential impact of lesions. Combined with the application of classification models, researchers can associate lesion areas with disease types to better understand the imaging features of different lesion types, which not only helps to classify lesions, but also provides an area-based statistical analysis basis for subsequent research. Through adaptive multi-feature fusion, the system can dynamically adjust and extract features of different lesion types to generate more accurate lesion area data. This fusion process helps to integrate multiple imaging features (such as grayscale, texture, shape, etc.) and dynamically adjust the segmentation strategy according to different types of lesions, greatly improving the segmentation accuracy of lesion areas, thereby providing researchers with high-quality regional divisions and supporting detailed analysis of lesions and subsequent research. Through multimodal feature analysis, researchers can analyze the different features of the lesion area from multiple angles (including shape, texture, dynamic changes, etc.) and establish a more comprehensive lesion feature library. Multimodal data analysis can reveal lesion details that cannot be captured under a single modality, improving researchers' overall understanding of the lesion area. Uploading these data to the medical cloud platform will not only facilitate data sharing and collaborative research, but also provide a reliable data basis for further analysis of large-scale lesion characteristics, which is conducive to supporting interdisciplinary and cross-institutional collaborative research.

[0011] Optionally, step S1 specifically includes:

[0012] Step S11: acquiring medical ultrasound image history data, and performing ultrasound medical history feature extraction based on the medical ultrasound image history data, thereby obtaining medical ultrasound image medical history data;

[0013] Step S12: classifying the disease type according to the medical ultrasound image medical history data, thereby obtaining in vivo symptom disease data and in vitro symptom disease data;

[0014] Step S13: performing symptom-driven feature enhancement according to the in vivo symptom disease data, thereby obtaining in vivo symptom disease-driven feature enhancement data;

[0015] Step S14: performing symptom-driven feature enhancement according to the in vitro symptom disease data, thereby obtaining in vitro symptom disease-driven feature enhancement data;

[0016] Step S15: integrating disease characteristic data according to the in vivo symptom disease driving characteristic enhancement data and the in vitro symptom disease driving characteristic enhancement data, thereby obtaining disease characteristic data;

[0017] Step S16: constructing an ultrasound image disease classification model according to the disease feature data, thereby obtaining an ultrasound image disease classification model.

[0018] By collecting and analyzing historical ultrasound image data, the present invention enables researchers to extract key medical history features from a large number of previous cases, providing a solid data foundation for further research on ultrasound images. The role of medical history data extraction is to reveal the long-term evolution of different types of lesions, which helps to analyze the changing trends of specific diseases in images, thereby providing a basis for subsequent research and model construction. Dividing diseases into two categories, internal symptoms and external symptoms, helps to separate different feature manifestations, which is convenient for refined research. Internal symptoms mainly involve internal organs and structures, while external symptoms involve epidermis or superficial tissues. The beneficial effect is that it provides researchers with a clearer analysis dimension, making subsequent feature extraction and analysis more targeted, thereby improving the accuracy of data analysis. Through symptom-driven feature enhancement, researchers can focus on enhancing the relevant features of in vivo disease symptoms to capture more detailed and important features. This enhancement not only helps to improve researchers' understanding of in vivo lesions, but also helps to mine complex image features, providing more representative data for in vivo disease research and model development. Feature enhancement for in vitro symptoms (such as skin, superficial tissue, etc.) can improve the accuracy of researchers' recognition of in vitro lesions. Unlike in vivo symptoms, in vitro symptoms are more superficial in ultrasound images. Strengthening these features can capture more microscopic features related to in vitro lesions, providing the necessary analytical basis for studying the external manifestations of the disease. By integrating the feature data of in vivo and in vitro symptoms, researchers can conduct a more comprehensive analysis of the disease from a holistic perspective. In vivo and in vitro symptoms complement each other. Through integration, the overall picture of the disease can be better analyzed, providing important support for studying complex and multi-level lesions. This data integration lays a solid foundation for the subsequent construction of classification models and is conducive to revealing the common and differential characteristics of the disease. Based on the disease feature data after in vivo and in vitro symptoms are enhanced, a comprehensive classification model is constructed, allowing researchers to automatically and accurately classify different types of diseases. The model can be used to analyze different types of disease features, providing researchers with a powerful tool for automatically analyzing, comparing and classifying different ultrasound image features. Its core role is to improve the efficiency and accuracy of the image analysis process, especially for pattern recognition and lesion tracking in medical imaging research.

[0019] Optionally, step S13 is specifically:

[0020] Step S131: performing pain degree statistics according to the in vivo symptom data, thereby obtaining high-pain in vivo symptom data;

[0021] Step S132: performing grayscale texture conversion according to the high-pain in vivo symptom data, thereby obtaining high-pain in vivo symptom grayscale texture data;

[0022] Step S133: performing inhomogeneity analysis based on the high pain in vivo symptom grayscale texture data, thereby obtaining inhomogeneity grayscale texture data;

[0023] Step S134: performing edge analysis based on the high pain in vivo symptom grayscale texture data, thereby obtaining high pain in vivo symptom fuzzy edge data;

[0024] Step S135: Integrate the in vivo symptom disease characteristics according to the uneven grayscale texture data and the high-pain in vivo symptom fuzzy edge data, so as to obtain the in vivo symptom disease driving feature enhancement data.

[0025] The present invention can provide data related to the severity of the lesion for research by statistically analyzing the pain degree of symptoms in the body. The high pain body symptom data reflects the characteristics of the pain distribution in the lesion area, and provides a characteristic data based on the patient's pain response for subsequent research. This data can be used to analyze the relationship between pain and image features, and help researchers better understand the characteristic expression of pain in the lesion area. Through grayscale texture conversion, the symptom data of the high pain area is presented in a more delicate image form, which is conducive to studying the imaging manifestations of the high pain area. The grayscale texture can capture the subtle changes in the lesion area and provide researchers with more intuitive data for analyzing the texture characteristics of the lesion area. This grayscale conversion makes the image easier to extract features and subsequent analysis, which is suitable for the study of tissue microstructure. Inhomogeneity analysis helps to reveal the detailed changes and complexity of the texture of the lesion area. By performing inhomogeneity analysis on the texture of the high pain area, researchers can further explore the heterogeneity of the internal structure of the lesion tissue, which is of great significance for understanding the complexity of different lesion tissues and the microscopic variation characteristics of the tissue, and helps to improve the ability to distinguish different lesion areas in ultrasound images. Edge analysis can help researchers identify the degree of boundary fuzziness of the lesion area, which is crucial to exploring whether the lesion area is clearly defined. The fuzzy edge data provides important clues for the study of lesion spread and invasiveness. High-pain areas are often accompanied by fuzzy edge performance. The analysis of fuzzy edge data provides support for the study of diffusion characteristics and boundary complexity of lesion areas. By integrating non-uniform grayscale texture data and fuzzy edge data, the lesion characteristics of high-pain areas can be comprehensively enhanced and comprehensively studied. This feature integration provides multi-angle information for analyzing complex in vivo lesions, and is particularly suitable for studying lesion spread, invasiveness, and intra-tissue heterogeneity. By integrating the enhancement of driving features, researchers can deeply understand the characteristic performance of complex lesion areas, thereby better advancing research related to lesion characteristics.

[0026] Optionally, step S14 is specifically:

[0027] Step S141: performing body temperature statistics according to the extracorporeal symptom disease data, thereby obtaining high temperature extracorporeal symptom disease data;

[0028] Step S142: dividing the high temperature extracorporeal symptom disease data into swelling regions, thereby obtaining high temperature swelling region data;

[0029] Step S143: measuring the skin thickness according to the high temperature swelling area data, thereby obtaining the skin thickness data of the high temperature swelling area;

[0030] Step S144: analyzing the skin color change according to the high temperature swelling area data, thereby obtaining the skin color data of the high temperature swelling area;

[0031] Step S145: Integrate the in vitro symptom disease characteristics according to the skin thickness data and the skin color data of the high temperature swelling area, so as to obtain the in vitro symptom disease driving characteristic enhancement data.

[0032] The present invention can capture the temperature change characteristics related to the lesion by statistically analyzing the body temperature of the external symptoms, which is of great value for studying the distribution of external symptoms such as fever and inflammation. The high-temperature external symptom data provides a basis for studying the temperature fluctuations in the area of ​​body surface lesions such as inflammation and infection, and helps to analyze the influence of lesions on the external temperature and its association with pathological changes. By dividing the swelling area of ​​the high-temperature external symptom data, the distribution of the swelling area caused by inflammation or infection on the body surface can be identified. This division helps to study the changes in body surface morphology caused by lesions, and provides quantitative data of symptom manifestations such as tissue expansion or fluid accumulation. The swelling area division data provides an important basis for subsequent analysis of the scope and spread of body surface lesions, which is a key step in understanding the development of external symptoms. The measurement of skin thickness can reveal the degree of tissue thickening in the swelling area, helping researchers to evaluate the impact of lesions on body surface tissues. The skin thickness data of the high-temperature swelling area provides important parameters for studying tissue changes caused by inflammation or infection, especially for understanding the changes in skin structure caused by lesions. This data helps to evaluate the differences in external manifestations of different lesions, thereby promoting in-depth research on the external characteristics of lesions. Skin color changes reflect the blood circulation, oxygenation status and inflammatory response of the lesion area. By analyzing the skin color changes in the high temperature and swollen area, we can understand the visual manifestation characteristics of the lesion on the body surface, especially the symptoms of inflammation, infection or redness and purpleness of the lesion area. This data provides a scientific basis for studying the color characteristic changes caused by surface diseases, and helps to understand the activity of the lesion area or the intensity of the inflammatory response. Integrating skin thickness and color change data helps to establish a multidimensional set of in vitro symptom disease characteristics. This step provides a more comprehensive understanding of in vitro lesions by integrating multiple surface symptom characteristics, which is conducive to studying the relationship between external symptom characteristics. For example, the correlation analysis between skin thickness and color changes can reveal the pattern and intensity of inflammatory spread. This feature integration provides richer data support for the study of lesion extension and in vitro feature manifestations.

[0033] Optionally, step S133 is specifically:

[0034] Gray-level co-occurrence matrix data are calculated based on the gray-level texture data of the high-pain body symptoms, thereby obtaining gray-level co-occurrence matrix data;

[0035] The local area standard deviation is calculated according to the gray level co-occurrence matrix data, thereby obtaining the local area standard deviation data;

[0036] Calculate the coefficient of variation based on the local area standard deviation and the local area standard deviation data to obtain the coefficient of variation data;

[0037] Entropy calculation is performed according to the gray-level co-occurrence matrix data to obtain entropy data;

[0038] The uneven grayscale texture is evaluated based on the coefficient of variation data and the entropy data, thereby obtaining the uneven grayscale texture data.

[0039] The present invention can quantify the spatial relationship between the grayscale values ​​of pixels in the ultrasound image by calculating the grayscale co-occurrence matrix (GLCM), thereby revealing the texture features in the image. This process provides a basis for studying the internal structural changes of tissues, especially in lesions related to high pain symptoms, and can capture the inherent complexity and heterogeneity of tissues. The grayscale co-occurrence matrix data can be used to analyze the texture change characteristics of the lesion area, which helps to reveal the law of lesion expansion and potential tissue heterogeneity. By calculating the local area standard deviation, researchers can quantify the degree of texture change in the image. The increase in standard deviation often indicates an increase in the complexity of the tissue structure, which is of great significance for analyzing tissue heterogeneity, fibrosis or other morphological changes in the lesion area. The local area standard deviation data provides quantitative support for studying the local heterogeneity of the lesion tissue, which helps to evaluate the characteristics of the lesion tissue and its changes over time. The coefficient of variation is the ratio of the standard deviation to the mean, which can reflect the relative degree of change in the local area. By calculating the coefficient of variation, the texture changes in the lesion area can be further quantified, providing a relative, dimensionless measurement method. The coefficient of variation data can reveal the dynamic characteristics of the internal heterogeneity of the tissue, which helps to deeply understand the tissue response and pathological progression in the high pain area. Entropy is an indicator to measure the degree of uncertainty and disorder in a system. By calculating the entropy of the gray-level co-occurrence matrix, the texture complexity and randomness in the image can be evaluated. Entropy data can help researchers quantitatively evaluate the structural complexity and heterogeneity in the lesion area, especially when analyzing high-pain lesions with complex tissue structures, which can effectively reveal the irregularity and disorder of the tissue. By combining the coefficient of variation and entropy data, the grayscale texture of the lesion area can be evaluated for inhomogeneity. This evaluation method can comprehensively reflect the structural complexity, texture irregularity and local heterogeneity in the lesion area. Inhomogeneous grayscale texture data provides researchers with important information about the irregularity of tissue structure and the complexity of pathology, which helps to analyze and study the potential development trend of the lesion area and the changing pattern of tissue response.

[0040] Optionally, step S134 is specifically:

[0041] Performing gradient calculation according to the high pain in vivo symptom grayscale texture data, thereby obtaining the high pain in vivo symptom grayscale texture gradient data;

[0042] Threshold data is obtained by setting a threshold value according to grayscale texture data of high pain in vivo symptoms;

[0043] According to the threshold data, the grayscale texture gradient data of the high pain body symptom is edge segmented to obtain weak edge data;

[0044] Perform edge refinement according to the weak edge data, thereby obtaining weak edge refinement data;

[0045] According to the weak edge data, the direction change amount is counted to obtain the uneven direction change data of the weak edge;

[0046] The fuzziness of the high pain in vivo symptom grayscale texture data is calculated according to the weak edge direction uneven change data and the weak edge refinement data, so as to obtain the high pain in vivo symptom fuzzy edge data.

[0047] The present invention can detect the areas in the image where the pixel grayscale changes most obviously, usually the edges and transition areas, through gradient calculation. This process helps reveal the changes in the boundaries and internal structures of the symptom areas in the body with high pain, laying the foundation for subsequent edge identification and accurate analysis of the lesion area. The grayscale texture gradient data enables researchers to better locate the boundaries of the pain area, and then understand the internal texture characteristics of these areas. Threshold setting is used to distinguish important texture change areas from background noise in the image, optimize the results of gradient calculation, and by setting an adaptive threshold for a specific lesion area, it can effectively filter out irrelevant information and highlight meaningful tissue features, thereby improving the accuracy of subsequent edge segmentation. Threshold data helps to ensure the accuracy of the edge segmentation process, thereby improving the texture analysis effect of the lesion area. By thresholding the gradient data, weak edge data is generated. This step effectively identifies the edge area where the grayscale changes slowly. The weak edge data can provide the research with weak transition areas corresponding to the lesion tissue. These areas usually contain useful tissue change information, which helps to further study the performance of edge features in the pain area. Edge refinement further improves the edge clarity of the lesion area by refining the boundaries of the weak edge area. Edge refinement data can effectively reduce noise interference and ensure accurate capture of tiny tissue edges. This step provides higher quality data support for subsequent edge direction change and fuzziness calculation, ensuring more accurate edge details in the process of studying high pain symptom tissues. By statistically analyzing the direction change of weak edges, the uneven change characteristics of edges in different directions can be captured. The uneven direction change data of weak edges helps to reveal the complexity of the lesion area, especially in tissues related to high pain symptoms, and can accurately quantify the characteristics of edge direction changes, so as to deeply study the trend and direction of changes in tissue morphology. Fuzziness calculation can measure the clarity of the edge. By combining the uneven direction change data and edge refinement data, the fuzzy edge data can reveal the changes in the degree of edge blur in the lesion area, which is of great research value for analyzing abnormal tissue proliferation or lesion spread. The quantitative evaluation of fuzziness can provide researchers with an objective measure of edge clarity, especially for complex or highly uneven lesion areas, which can better understand the fuzziness and uncertainty of the edge.

[0048] Optionally, step S4 is specifically:

[0049] Step S41: calculating the area of ​​the lesion region according to the non-zero pixel ultrasound image of the patient, thereby obtaining the area data of the lesion region;

[0050] Step S42: calculating the boundary line of the lesion area according to the non-zero pixel point ultrasound image of the patient, thereby obtaining the boundary line data of the lesion area;

[0051] Step S43: evaluating the lesion compactness according to the lesion area data and the lesion area boundary line data, thereby obtaining lesion compactness data;

[0052] Step S44: measuring the roundness of the lesion region and evaluating the complexity of the lesion region contour according to the lesion region area data and the lesion region boundary line data, thereby obtaining the lesion region roundness data and the lesion region contour complexity data;

[0053] Step S45: performing lesion invasiveness analysis according to the lesion area roundness data and the lesion area contour complexity data, thereby obtaining lesion invasiveness data;

[0054] Step S46: performing tissue lesion feature weight allocation on the lesion invasiveness data and the lesion compactness data according to the tissue lesion type data, thereby obtaining tissue lesion type adaptive multi-feature data;

[0055] Step S47: accurately segmenting the tissue lesion region data according to the tissue lesion type adaptive multi-feature data, thereby obtaining accurate segmentation data of the tissue lesion region.

[0056] The present invention can accurately measure the size of the lesion area by calculating the area of ​​non-zero pixels in the lesion area in the ultrasound image. The lesion area data provides an important basis for studying the development trend of the lesion, the diffusion speed of the lesion, and the difference in area between different lesions. The data provides effective support for subsequent morphological feature evaluation and invasiveness analysis. By calculating the boundary line, the outline of the lesion area can be accurately defined to help analyze the shape characteristics of the lesion. The boundary line data of the lesion area lays the foundation for further studying the geometric characteristics of the lesion area (such as contour complexity, roundness, etc.). These geometric characteristics can be used to analyze the diffusion morphology and tissue invasiveness of the lesion. The lesion compactness assessment is used to quantitatively analyze the compactness of the lesion area, that is, the degree of filling of the lesion area within its boundary line. The compactness data can reveal the uniformity of the lesion tissue. Areas with higher compactness show more concentrated lesion characteristics, while areas with lower compactness suggest a scattered distribution of lesions. This data is of great significance to the study of lesion morphology. Roundness measurement helps to evaluate whether the shape of the lesion area is close to a circle, while contour complexity assessment is used to measure the complexity of the edge of the lesion area. Roundness data can reflect the symmetry and regularity of the lesion, while contour complexity data reveals whether the lesion area presents complex and diverse boundary shapes. These geometric features provide an important basis for the study of the invasiveness and expansion behavior of lesions. Lesion invasiveness analysis is based on roundness and contour complexity data, which can reveal whether the lesion area presents invasive growth characteristics. Lesions with strong invasiveness usually show complex contours and low roundness. Invasiveness data helps to analyze the potential path and pattern of lesion expansion, and thus provides data support for in-depth research on the spread mechanism of lesions. By combining lesion type, invasiveness data and compactness data, adaptive weighted allocation of features can be performed, and more accurate feature weights can be set for different lesion types. Adaptive multi-feature data provides a flexible research method for the accurate characterization of different lesion types, which helps to comprehensively analyze the multidimensional features of different lesion types, thereby improving the understanding of lesion characteristics and the ability to refine classification. Based on the adaptive multi-feature data, accurate segmentation of lesion areas can be performed, and different types of lesions can be accurately divided into regions in ultrasound images. This process improves the adaptability of the segmentation algorithm to different lesion types, ensures more accurate segmentation analysis of complex tissue structures and irregular lesion areas, and facilitates the study of the detailed characteristics of lesion tissues and their development trends.

[0057] Optionally, step S45 is specifically:

[0058] Step S451: Calculating the roundness score according to the lesion area roundness data, thereby obtaining the lesion area roundness score data;

[0059] Step S452: Calculating the contour complexity score according to the contour complexity data of the lesion area, thereby obtaining the contour complexity score data of the lesion area;

[0060] Step S453: obtaining medical clinical database data, and extracting lesion aggressiveness weight features according to the medical clinical database data, thereby obtaining lesion aggressiveness weight data;

[0061] Step S454: Perform an invasiveness score according to the lesion invasiveness weight data, the lesion area roundness score data, and the lesion area contour complexity score data, so as to obtain lesion invasiveness data.

[0062] The present invention can quantitatively evaluate the shape regularity of the lesion area by converting the roundness data of the lesion area into a score. The roundness score reflects the symmetry and geometric regularity of the lesion area. The higher the roundness, the closer the lesion area is to a regular circle, indicating that the lesion is less invasive, and provides accurate numerical support for further integrating geometric features for invasive analysis. The contour complexity score is used to quantitatively evaluate the edge morphological complexity of the lesion area. The more complex the contour, the higher the score, indicating that the lesion boundary presents a more irregular shape, indicating lesion characteristics with stronger invasiveness, providing important geometric indicators for quantitative analysis of the invasiveness of the lesion, and laying the foundation for subsequent invasiveness scoring and feature extraction. Extracting weight data related to the invasiveness of the lesion from the medical clinical database can be combined with the statistical results of a large number of clinical cases to objectively quantify the weights of various features of the invasiveness of the lesion. Through the large sample data in the database, researchers can extract more universal invasive feature weights, which helps to improve the accuracy and scientificity of invasive analysis, ensure that the analysis model has better generalization ability, and can be applied to a variety of different types of lesion research. By combining multiple feature scores (such as roundness score, contour complexity score) and invasive weight data in the clinical database, a comprehensive invasiveness score can be performed, which can more accurately evaluate the invasiveness of the lesion area. This comprehensive scoring mechanism helps to reveal the potential development trends and invasive characteristics of different lesion areas, and provides reliable data support for further research on lesion spread behavior and feature classification.

[0063] Optionally, step S47 is specifically:

[0064] Step S471: extracting texture features based on the adaptive multi-feature data of tissue lesion types, thereby obtaining texture data;

[0065] Step S472: selecting texture seed point pixels for tissue lesion area data according to the texture data, thereby obtaining texture seed point pixel data;

[0066] Step S473: defining a growth criterion according to the texture seed point pixel data, thereby obtaining growth criterion data;

[0067] Step S474: performing region growth on the tissue lesion region data according to the texture seed point pixel data and the growth criterion data, thereby obtaining tissue lesion region growth data;

[0068] Step S475: performing precise regional segmentation on the tissue lesion region data according to the tissue lesion region growth data, thereby obtaining precise tissue lesion region segmentation data.

[0069] The present invention can accurately extract the texture features of tissues by utilizing the adaptive multi-feature data of tissue lesion types. The texture features can reflect the subtle changes and complexity of tissue structures, and are particularly important in the morphological study of lesion areas, and are helpful in studying the texture differences of different lesion areas, providing a solid foundation for subsequent regional segmentation and morphological analysis. By selecting representative seed point pixels from the texture feature data, the core features of the lesion area are determined. This process helps to define the central features of the area, provides a starting point for subsequent regional growth and precise segmentation, reduces the risk of error accumulation, and enhances the accuracy of the segmentation process. Defining the growth criterion is the key to ensuring the accurate application of the region growing algorithm. By combining the texture seed point pixel data, the growth criterion can be tailored according to the specific texture features of the lesion area to ensure the accuracy and flexibility of the regional growth. This step ensures that too much noise or erroneous information will not be introduced during the regional expansion process, and adapts to different types of lesion tissues. The region growing algorithm can gradually expand the initial lesion area to the complete lesion boundary based on the seed point and the growth criterion. This process can effectively capture the overall morphology of the lesion area, help to improve the integrity and accuracy of the segmentation, and provide basic data for the comprehensive analysis of the lesion area, making further regional segmentation more effective and accurate. Accurate segmentation is one of the ultimate goals of ultrasound image analysis. By taking region growth data as input, the boundaries of the lesion area can be divided more accurately, which not only improves the accuracy of segmentation, but also can effectively adapt to different types of tissue lesions, making the segmentation results more robust and reliable in various situations.

[0070] Optionally, the present specification also provides a medical ultrasound image segmentation method and system based on adaptive feature fusion, which are used to perform the medical ultrasound image segmentation method based on adaptive feature fusion as described above. The medical ultrasound image segmentation system includes:

[0071] Disease classification model building module: used to obtain medical ultrasound image history data, and perform ultrasound medical history feature extraction based on the medical ultrasound image history data, thereby obtaining medical ultrasound image medical history data; and to construct an ultrasound image disease classification model based on the medical ultrasound image medical history data, thereby obtaining an ultrasound image disease classification model;

[0072] Ultrasonic layered tissue analysis module: used to obtain the patient's medical record ultrasound image, and remove zero pixels according to the patient's medical record ultrasound image, so as to obtain the patient's non-zero pixel ultrasound image; perform ultrasonic layered tissue analysis according to the patient's non-zero pixel ultrasound image, so as to obtain tissue lesion area data;

[0073] Tissue lesion classification module: used to calculate the lesion area ratio based on the tissue lesion area data and the patient's non-zero pixel ultrasound image, so as to obtain lesion area ratio data; classify the tissue lesion area data into disease types based on the ultrasound image disease classification model, so as to obtain tissue lesion type data;

[0074] Tissue lesion area segmentation module: used to perform adaptive multi-feature fusion on the patient's non-zero pixel ultrasound image according to the tissue lesion type data, so as to obtain adaptive multi-feature data of the tissue lesion type; accurately segment the tissue lesion area data according to the adaptive multi-feature data of the tissue lesion type, so as to obtain accurate segmentation data of the tissue lesion area;

[0075] Tissue lesion multimodal feature analysis module: used to perform multimodal feature analysis based on the precise segmentation data of tissue lesion areas and lesion area ratio data, so as to obtain multimodal feature data of tissue lesions and upload it to the medical cloud platform.

[0076] The present invention provides a medical ultrasound image segmentation method and system based on adaptive feature fusion. The system can implement any medical ultrasound image segmentation method based on adaptive feature fusion of the present invention, and is used to combine the operation and signal transmission medium between various modules to complete the medical ultrasound image segmentation method based on adaptive feature fusion. The internal modules of the system cooperate with each other to improve the accuracy of medical ultrasound image segmentation. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments thereof made with reference to the following drawings:

[0078] Figure 1 It is a schematic diagram of the steps of the medical ultrasound image segmentation method based on adaptive feature fusion of the present invention;

[0079] Figure 2 Detailed step flow diagram of step S1 in the present invention;

[0080] Figure 3 Detailed step flow diagram of step S13 in the present invention;

[0081] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0082] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. 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 technicians in this field without creative work are within the scope of protection of the present invention.

[0083] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0084] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0085] To achieve this, please refer to Figures 1 to 3 The present invention provides a medical ultrasound image segmentation method based on adaptive feature fusion, the method comprising the following steps:

[0086] Step S1: acquiring medical ultrasound image history data, and extracting ultrasound medical history features based on the medical ultrasound image history data, thereby obtaining medical ultrasound image medical history data; constructing an ultrasound image disease classification model based on the medical ultrasound image medical history data, thereby obtaining an ultrasound image disease classification model;

[0087] In this embodiment, the medical ultrasound image history data is obtained by connecting the data with the hospital's information system. This process includes extracting the stored ultrasound image data and using a database query language (such as SQL) to extract relevant images according to specific screening conditions (such as date range, patient medical record number or image type). The extracted ultrasound image is subjected to an image processing algorithm to extract ultrasound medical history features, and the techniques used include gray-level co-occurrence matrix (GLCM) analysis, texture feature extraction and edge detection. Specifically, GLCM is used to calculate features such as image contrast, homogeneity and entropy, and these features constitute medical ultrasound image medical history data. Subsequently, an ultrasound image disease classification model is constructed based on the extracted medical history feature data. The model uses machine learning technology, especially random forests or support vector machines, to generate a classifier model by training feature data with known disease categories, and can automatically classify disease types on new images.

[0088] Step S2: Acquire the patient's medical record ultrasound image, and remove zero pixels according to the patient's medical record ultrasound image, so as to obtain the patient's non-zero pixel ultrasound image; perform ultrasound layered tissue analysis according to the patient's non-zero pixel ultrasound image, so as to obtain tissue lesion area data;

[0089] In this embodiment, the data for obtaining the ultrasound image of the patient's medical record comes from the ultrasound image storage system of the hospital. The image is obtained by identifying the patient ID and the relevant imaging examination record, and the image format is DICOM. For the processing of the medical record image, the zero pixel point processing is first performed. Using the threshold method, the pixel value below the preset threshold (for example, the threshold is set to 0.1) is regarded as a zero pixel, and the image is binarized to remove these zero-value pixels and retain valid non-zero pixel data. Next, the ultrasound layered tissue analysis of the patient's non-zero pixel ultrasound image is performed by image processing technology, and the technology based on the regional growing algorithm is used to identify different tissue types and lesion areas from the ultrasound image. By constructing the growth seed point and gradually expanding it in combination with the grayscale and spatial information, the tissue lesion area data is finally obtained. These data contain tissue features at different levels, laying the foundation for subsequent analysis.

[0090] Step S3: calculating the lesion area ratio according to the tissue lesion area data and the non-zero pixel point ultrasound image of the patient, thereby obtaining lesion area ratio data; classifying the tissue lesion area data into disease types according to the ultrasound image disease classification model, thereby obtaining tissue lesion type data;

[0091] In this embodiment, the lesion area ratio is calculated based on the tissue lesion area data and the non-zero pixel ultrasound image of the patient. The lesion area ratio is obtained by dividing the total number of pixels in the lesion area by the total number of non-zero pixels, and the formula is: lesion area ratio = (number of pixels in the lesion area) / (number of non-zero pixels). This ratio data provides a quantitative basis for studying the severity of the lesion. At the same time, the ultrasound image disease classification model is used to classify the extracted tissue lesion area data into disease types. This step evaluates the features of the lesion area through pixel-by-pixel feature analysis combined with the feature weights of the model to determine its similarity with known disease types. The classification result is tissue lesion type data, which contains the probability distribution of each lesion and can reflect the potential classification information of the lesion.

[0092] Step S4: Adaptively fusing the non-zero pixel ultrasound image of the patient according to the tissue lesion type data to obtain adaptive multi-feature data of the tissue lesion type; accurately segmenting the tissue lesion area data according to the adaptive multi-feature data of the tissue lesion type to obtain accurate segmentation data of the tissue lesion area;

[0093] In this embodiment, an adaptive multi-feature fusion operation is performed on the non-zero pixel ultrasound image of the patient, and this process determines the fusion strategy according to the tissue lesion type data. Image processing algorithms, such as wavelet transform, principal component analysis, etc., are used to extract different types of features, including texture, shape and statistical features. The weight of each feature is dynamically adjusted according to the feature importance score, thereby forming adaptive multi-feature data of tissue lesion types. This data combines the information of multiple image features to enhance the expression ability of lesion features. Subsequently, based on the adaptive multi-feature data, accurate segmentation of the tissue lesion area is performed, and the boundary is refined by methods such as the watershed algorithm or the active contour model, and the gradient information and regional similarity are combined to ensure accurate segmentation of the lesion area. This segmentation result reflects the true boundary of the lesion area and provides a basis for subsequent feature analysis.

[0094] Step S5: Perform multimodal feature analysis based on the precise segmentation data of the tissue lesion area and the lesion area ratio data, so as to obtain multimodal feature data of the tissue lesion and upload it to the medical cloud platform.

[0095] In this embodiment, the precise segmentation data of the tissue lesion area is combined with the lesion area ratio data for multimodal feature analysis. This process uses techniques including morphological analysis, grayscale histogram analysis, and texture feature statistics to conduct in-depth research on the diversity and complexity of lesion areas. The analysis process involves extracting the features of each lesion area, including specific parameters such as grayscale mean, standard deviation, shape features (such as roundness, aspect ratio), and texture features (such as energy, contrast, homogeneity). Multimodal feature data provides a basis for quantitative analysis of the comprehensive characteristics of tissue lesions, and is ultimately uploaded to the medical cloud platform. During the upload process, data is transmitted through a secure API interface to ensure data integrity and security. This series of data processing and uploading provides a rich source of information for subsequent research, which facilitates further statistical analysis and model training. Optionally, step S1 is specifically as follows:

[0096] Step S11: acquiring medical ultrasound image history data, and performing ultrasound medical history feature extraction based on the medical ultrasound image history data, thereby obtaining medical ultrasound image medical history data;

[0097] In this embodiment, the acquisition of medical ultrasound image history data is completed by accessing the hospital's electronic health record system, and the data source includes the patient's ultrasound examination record. This process uses a data query language (such as SQL) to extract relevant image information from the database, and screening conditions such as time range and image type need to be set. The extracted ultrasound image data is preprocessed, including denoising and normalization. Next, image processing technology is used to extract ultrasound medical history features. The specific steps include using image filters (such as high-pass filters) to enhance edge information and calculate image texture features (such as contrast, homogeneity, and energy). These features are extracted through a gray-level co-occurrence matrix (GLCM). The final medical ultrasound image history data will be used for subsequent disease type classification.

[0098] Step S12: classifying the disease type according to the medical ultrasound image medical history data, thereby obtaining in vivo symptom disease data and in vitro symptom disease data;

[0099] In this embodiment, after obtaining the medical ultrasound image medical history data, the process of classifying the disease type is achieved by constructing a classification algorithm. Using a machine learning algorithm, such as random forest, support vector machine (SVM) or K nearest neighbor (KNN), the extracted medical history feature data is first trained with known disease labels. In the specific implementation, 80% of the data is selected for training and 20% of the data is used to verify the model performance. During the model training process, the optimal parameter settings, such as maximum depth, minimum sample segmentation, etc., are determined by cross-validation. After training, the model can output two types of results based on the input feature data: in vivo symptom disease data and in vitro symptom disease data, representing different types of disease symptom characteristics.

[0100] Step S13: performing symptom-driven feature enhancement according to the in vivo symptom disease data, thereby obtaining in vivo symptom disease-driven feature enhancement data;

[0101] In this embodiment, symptom-driven feature enhancement is performed based on the in vivo symptom disease data. First, the main features related to the in vivo symptoms are extracted, such as the shape, size, and texture features of the lesion area. This process is achieved through image analysis technology, using edge detection algorithms (such as the Canny algorithm) to identify the boundaries of the lesions, and then calculating shape features, such as roundness, aspect ratio, etc. The extracted features are enhanced and normalized using a standardized method to ensure that different features are at the same scale. The in vivo symptom features are combined with the original features using a weighted average method to obtain in vivo symptom disease-driven feature enhancement data, providing a more sensitive feature description for subsequent analysis.

[0102] Step S14: performing symptom-driven feature enhancement according to the in vitro symptom disease data, thereby obtaining in vitro symptom disease-driven feature enhancement data;

[0103] In this embodiment, the symptom-driven feature enhancement of in vitro symptom disease data focuses on the spatial information analysis of the image. Using the region growing algorithm, the seed point positions are first determined, and these seed points are manually selected by clinical experts based on image features. Subsequently, the surrounding area is expanded according to growth criteria (such as grayscale differences and neighborhood similarities) to extract the complete lesion area. Next, morphological operations (such as opening and closing operations) are applied to process the extracted area to remove noise and smooth the boundaries. Through these processing steps, the characteristics of the in vitro symptoms are further enhanced, and ultimately the in vitro symptom disease-driven feature enhancement data is formed.

[0104] Step S15: integrating disease characteristic data according to the in vivo symptom disease driving characteristic enhancement data and the in vitro symptom disease driving characteristic enhancement data, thereby obtaining disease characteristic data;

[0105] In this embodiment, the in vivo symptom disease-driven feature enhancement data and the in vitro symptom disease-driven feature enhancement data are integrated, and the integrated feature space needs to be determined first. This process involves feature selection and dimensionality reduction of the two sets of data, and the principal component analysis (PCA) is used to extract the main components to ensure that the integrated features can retain the original information. Through the weighted synthesis method, the features are weighted according to their respective importance to generate disease feature data. The integrated data not only reflects the internal and external characteristics of the lesion, but also provides a comprehensive feature description for subsequent analysis.

[0106] Step S16: constructing an ultrasound image disease classification model according to the disease feature data, thereby obtaining an ultrasound image disease classification model.

[0107] In this embodiment, an ultrasound image disease classification model is constructed based on the integrated disease feature data, and a machine learning algorithm (such as random forest or deep learning neural network) is used for model training. First, the integrated feature data set is divided into a training set and a test set to ensure the generalization ability of the model. During the training process, the grid search method is used to adjust hyperparameters, such as learning rate, batch size, etc. The model performance is evaluated using cross-validation, and the optimal model is selected to achieve disease classification of new ultrasound images. The ultrasound image disease classification model finally generated has a high classification accuracy, which provides a powerful tool for subsequent research.

[0108] Optionally, step S13 is specifically:

[0109] Step S131: performing pain degree statistics according to the in vivo symptom data, thereby obtaining high-pain in vivo symptom data;

[0110] In this embodiment, when performing statistics on high pain in vivo symptom data, in vivo symptom data related to the patient are obtained, which include the patient's subjective pain score and related clinical data. Statistical analysis software (such as R or Python's Pandas library) is used to aggregate the pain score, set a threshold standard (for example, a pain score greater than 7 is considered high pain), and filter out symptom data that meets the standard through simple conditional statements. The final statistical results will form a high pain in vivo symptom data set, including all records with pain scores exceeding the set threshold, for subsequent processing and analysis.

[0111] Step S132: performing grayscale texture conversion according to the high-pain in vivo symptom data, thereby obtaining high-pain in vivo symptom grayscale texture data;

[0112] In this embodiment, the process of grayscale texture conversion of high pain in vivo symptom data requires converting the high pain symptom data into a form suitable for image processing. The symptom image is processed using a grayscale conversion formula (such as converting the RGB color model into a weighted average of grayscale values). In the implementation, image processing libraries such as OpenCV are used to read high pain in vivo symptom images, and the cv2.cvtColor function is used for color space conversion to obtain high pain in vivo symptom grayscale texture data. This data reflects the distribution of pain areas at different grayscale levels, providing a basis for subsequent texture analysis.

[0113] Step S133: performing inhomogeneity analysis based on the high pain in vivo symptom grayscale texture data, thereby obtaining inhomogeneity grayscale texture data;

[0114] In this embodiment, when performing inhomogeneity analysis, the local contrast of the image is calculated using the high pain in vivo symptom grayscale texture data. The image is divided into multiple small areas (such as 3x3 or 5x5 windows) through the sliding window technology, and the grayscale variance in each window is calculated. The larger the variance value, the more significant the grayscale change in the area, reflecting the inhomogeneity of the texture. In the specific implementation, a custom function is used to traverse each window and calculate its variance, and finally an inhomogeneous grayscale texture data set is generated to reflect the grayscale inhomogeneity of the high pain in vivo symptom area.

[0115] Step S134: performing edge analysis based on the high pain in vivo symptom grayscale texture data, thereby obtaining high pain in vivo symptom fuzzy edge data;

[0116] In this embodiment, when performing edge analysis on the grayscale texture data of high pain in vivo symptoms, an edge detection algorithm (such as the Canny edge detection algorithm) is used to extract edge information. First, the grayscale image is Gaussian blurred to remove noise, using a Gaussian kernel with a standard deviation of 1.0. Subsequently, the Canny algorithm is applied to set a low threshold and a high threshold (for example, a low threshold of 50 and a high threshold of 150) for edge detection. Through the cv2.Canny function, high pain in vivo symptom fuzzy edge data is generated to show the boundary of the pain area and its degree of blur, providing important edge information for subsequent feature integration.

[0117] Step S135: Integrate the in vivo symptom disease characteristics according to the uneven grayscale texture data and the high-pain in vivo symptom fuzzy edge data, so as to obtain the in vivo symptom disease driving feature enhancement data.

[0118] In this embodiment, according to the process of integrating the in vivo symptom disease characteristics based on the uneven grayscale texture data and the fuzzy edge data of high-pain in vivo symptoms, the inhomogeneity data and the fuzzy edge data are feature-weighted using the weighted synthesis method to combine the information of the two. The corresponding values ​​of the inhomogeneity features and the fuzzy edge features are standardized to ensure that they are in the same dimension. Afterwards, the two sets of features are weighted averaged by defining weight parameters (such as 0.6 and 0.4), and finally the in vivo symptom disease driving feature enhancement data is obtained, which comprehensively reflects the texture features and edge features of high-pain in vivo symptoms, providing rich information for subsequent analysis.

[0119] Optionally, step S14 is specifically:

[0120] Step S141: performing body temperature statistics according to the extracorporeal symptom disease data, thereby obtaining high temperature extracorporeal symptom disease data;

[0121] In this embodiment, when statistics of high temperature extracorporeal symptom disease data are performed, the patient's body temperature data is collected, which is usually obtained through medical equipment such as electronic thermometers. The temperature record of each patient includes measurements at multiple time points. Through data processing software (such as Python's Pandas library), these body temperature data are statistically analyzed and threshold standards are set. For example, records with a body temperature exceeding 38.5°C are regarded as high temperature data. By conditional screening, all high temperature records are extracted to form a high temperature extracorporeal symptom disease data set, which will provide a basis for the analysis of subsequent steps.

[0122] Step S142: dividing the high temperature extracorporeal symptom disease data into swelling regions, thereby obtaining high temperature swelling region data;

[0123] In this embodiment, when dividing the high-temperature swelling area, it is necessary to obtain image data related to the high-temperature in vitro symptoms, usually shot using a high-resolution digital camera or a mobile phone camera. Using image processing software (such as OpenCV), the image is converted into a black and white binary image through an image segmentation algorithm (such as a threshold segmentation method). An appropriate threshold (such as the lowest temperature based on the high-temperature data) is set to separate the high-temperature area from the normal area. Subsequently, the contour detection technology is used to identify the boundary of the swelling area. Finally, the high-temperature swelling area data is generated to identify the swelling site of the in vitro symptoms, providing a clear regional basis for subsequent measurement and analysis.

[0124] Step S143: measuring the skin thickness according to the high temperature swelling area data, thereby obtaining the skin thickness data of the high temperature swelling area;

[0125] In this embodiment, when measuring the skin thickness of the high temperature swelling area, a suitable measuring tool is selected, such as an ultrasonic thickness meter or a high frequency ultrasonic instrument. These devices can provide accurate thickness measurement. The measurement point is determined using the high temperature swelling area data to ensure that the measured part is the center of the swelling area. The thickness value of the skin is recorded at each measurement point, usually in millimeters (mm). By performing statistical analysis on the thickness data of multiple measurement points, the average value and standard deviation are calculated to generate the skin thickness data of the high temperature swelling area. These data provide key physical parameters for subsequent feature integration.

[0126] Step S144: analyzing the skin color change according to the high temperature swelling area data, thereby obtaining the skin color data of the high temperature swelling area;

[0127] In this embodiment, when analyzing the skin color change of the high temperature swelling area, it is necessary to obtain relevant image data to ensure that the image is taken under good lighting conditions to ensure the accuracy of the color. Use image processing software (such as OpenCV) to convert the acquired image into a color space from RGB space to HSV or Lab space to better analyze the color features. During the analysis, the color histogram of the swelling area is extracted, and the average hue, saturation and brightness values ​​are calculated to identify color changes. By setting a standard (for example, the amplitude of the change exceeds a certain value, such as ΔE>2), the skin color data of the high temperature swelling area is finally generated to provide a quantitative analysis of skin changes.

[0128] Step S145: Integrate the in vitro symptom disease characteristics according to the skin thickness data and the skin color data of the high temperature swelling area, so as to obtain the in vitro symptom disease driving characteristic enhancement data.

[0129] In this embodiment, in the process of integrating in vitro symptom disease characteristics, it is necessary to integrate the skin thickness data and skin color data from the high temperature swelling area. To achieve integration, the two sets of data are first standardized to ensure that their dimensions are consistent, and the Z-score standardization method is usually used. Next, the weighting parameters are defined, such as assigning a 60% weight to skin thickness and a 40% weight to skin color. Using the weighted average formula, the two sets of feature data are integrated to generate in vitro symptom disease-driven feature enhancement data. This data set reflects the physical characteristics and color changes of the high temperature swelling area, providing rich comprehensive information for subsequent analysis.

[0130] Optionally, step S133 is specifically:

[0131] Gray-level co-occurrence matrix data are calculated based on the gray-level texture data of the high-pain body symptoms, thereby obtaining gray-level co-occurrence matrix data;

[0132] In this embodiment, when calculating the grayscale co-occurrence matrix, it is first necessary to extract an image from the grayscale texture data of high-pain in vivo symptoms. Use image processing software (such as MATLAB or OpenCV library in Python) to convert the image into a grayscale image for subsequent analysis. Define a relatively small window (such as 5x5 or 7x7) and count the frequency of occurrence of grayscale value pairs in each window. The calculation process of the grayscale co-occurrence matrix is ​​achieved by traversing the entire image and recording the frequency of different grayscale value combinations between adjacent pixels. This method will provide an important basis for subsequent texture feature analysis and form grayscale co-occurrence matrix data. The specific grayscale value range can be set to 0 to 255 to ensure that all grayscale levels are covered.

[0133] The local area standard deviation is calculated according to the gray level co-occurrence matrix data, thereby obtaining the local area standard deviation data;

[0134] In this embodiment, when calculating the standard deviation of a local area, the gray level co-occurrence matrix data obtained previously is used. An appropriate area (such as a 5x5 or 7x7 window) is selected, and the standard deviation of the gray value in each window is calculated. The formula is:

[0135]

[0136] Where μ is the average grayscale value in the window, and N is the number of pixels in the window. This calculation needs to be performed on each local area of ​​the entire image to obtain the local area standard deviation data. This data is used to evaluate the degree of texture change and reflect the texture features in the image.

[0137] Calculate the coefficient of variation based on the local area standard deviation and the local area standard deviation data to obtain the coefficient of variation data;

[0138] In this embodiment, when calculating the coefficient of variation, it is first necessary to obtain the standard deviation data and the corresponding mean data of each local area. The calculation formula of the coefficient of variation is:

[0139] CV=μσ

[0140] Among them, σ is the standard deviation of the local area, and μ is the average gray value of the area. Using programming tools (such as Python), traverse all local areas and calculate the coefficient of variation one by one. The final coefficient of variation data will provide an important basis for the evaluation of inhomogeneity for subsequent analysis, reflect the relative degree of grayscale change in the image, and provide a reference for the quantitative analysis of texture features.

[0141] Entropy calculation is performed according to the gray-level co-occurrence matrix data to obtain entropy data;

[0142] In this embodiment, when performing entropy calculation, the gray level co-occurrence matrix data is first used to calculate the probability distribution of each gray level. The entropy calculation formula is:

[0143]

[0144] Among them, P(i) is the probability of gray level i appearing. By traversing the entire gray level co-occurrence matrix, the number of occurrences of each gray level is counted and normalized into a probability form. The calculation result will generate entropy data, reflecting the complexity and uncertainty of the image. The level of entropy will provide strong support for subsequent texture analysis and reveal the texture characteristics of the image.

[0145] The uneven grayscale texture is evaluated based on the coefficient of variation data and the entropy data, thereby obtaining the uneven grayscale texture data.

[0146] In this embodiment, when evaluating the uneven grayscale texture, the unevenness evaluation formula must first be defined by combining the coefficient of variation data and the entropy data. By setting the weights (for example, the coefficient of variation accounts for 70% and the entropy accounts for 30%), the following formula is used for calculation:

[0147] U=w CV CV+w H ·H

[0148] Among them, w CV With w H is the corresponding weight coefficient. Combining the calculated coefficient of variation data and entropy data, a comprehensive evaluation is performed through programming tools to obtain the inhomogeneity grayscale texture data. This data reflects the texture distribution characteristics of the image, provides support for subsequent medical research, and allows researchers to analyze the texture differences of different tissues or lesion areas.

[0149] Optionally, step S134 is specifically:

[0150] Performing gradient calculation according to the high pain in vivo symptom grayscale texture data, thereby obtaining the high pain in vivo symptom grayscale texture gradient data;

[0151] In this embodiment, in the processing of high pain in vivo symptom grayscale texture data, gradient calculation is first performed to extract important features. By selecting a suitable image processing tool, such as OpenCV or MATLAB, a grayscale image of high pain in vivo symptoms is imported to ensure that the image is an 8-bit grayscale image. The Sobel operator is used for gradient calculation, and horizontal and vertical convolution kernels are defined. The calculation process includes applying the Sobel convolution kernel to each pixel of the image to obtain the gradient value of each pixel in the x and y directions. By calculating the gradient amplitude in these two directions, the high pain in vivo symptom grayscale texture gradient data is formed. The key to this step is to select a suitable convolution kernel. Parameters such as the convolution kernel size and scale are set to 3x3 in actual applications to ensure the effectiveness of the gradient calculation.

[0152] Threshold data is obtained by setting a threshold value according to grayscale texture data of high pain in vivo symptoms;

[0153] In this embodiment, the threshold is set according to the grayscale texture data of the high pain body symptoms to ensure the accuracy of the subsequent edge detection. The Otsu method is used to determine the optimal threshold. The process starts with calculating the grayscale histogram of the image, counting the number of pixels at each grayscale level, and calculating the total number of pixels in the image. All thresholds are traversed, and the threshold that maximizes the variance between classes is selected. In the specific implementation, the optimal threshold is calculated to be 120, which is used as the standard for subsequent processing. The threshold data provides a clear reference basis for edge segmentation, ensuring the effectiveness and accuracy of the edge detection process.

[0154] According to the threshold data, the grayscale texture gradient data of the high pain body symptom is edge segmented to obtain weak edge data;

[0155] In this embodiment, after obtaining the threshold data, the threshold is used to perform edge segmentation on the grayscale texture gradient data of the high pain body symptom. Binarization is performed by comparing each pixel value with the set threshold. If the gradient value is greater than or equal to the threshold 120, the pixel is marked as an edge point and the value is set to 255; if it is less than the threshold, it is set to 0. The weak edge data formed by this operation can intuitively reflect the edge characteristics of the high pain body symptom, and provide basic data for subsequent refinement and fuzziness calculation.

[0156] Perform edge refinement according to the weak edge data, thereby obtaining weak edge refinement data;

[0157] In this embodiment, the weak edge data is thinned to improve the clarity and accuracy of the edge. The edge thinning is performed using a non-maximum suppression method, and the process includes traversing each weak edge pixel and comparing its intensity value with that of adjacent pixels. For each edge pixel, if the pixel has the maximum intensity value in its gradient direction, the pixel is retained; otherwise, its value is set to 0. This thinning process can effectively remove unnecessary noise and ensure the clarity of the edge, providing more accurate edge data for subsequent directional change statistics.

[0158] According to the weak edge data, the direction change amount is counted to obtain the uneven direction change data of the weak edge;

[0159] In this embodiment, based on the weak edge data, the direction change amount is counted to evaluate the complexity of the edge. For each edge point, the direction change between it and the adjacent edge point is calculated. First, a small range is set to identify adjacent edge pixels, usually a range of 1 pixel. For each edge point, its direction relative to the neighboring pixels is calculated, and the changes in these directions are counted. By collecting and analyzing this direction change information, the uneven data of weak edge direction changes are obtained, which lays an important foundation for the fuzziness calculation of high pain body symptoms.

[0160] The fuzziness of the high pain in vivo symptom grayscale texture data is calculated according to the weak edge direction uneven change data and the weak edge refinement data, so as to obtain the high pain in vivo symptom fuzzy edge data.

[0161] In this embodiment, fuzziness calculation is performed based on the weak edge direction change uneven data and weak edge refinement data. First, the standard for fuzziness calculation is set, usually considering the standard deviation of the direction change uneven data and the strength of the weak edge refinement data. By analyzing the edge changes and the refinement results, the fuzziness value of each edge point is calculated. This calculation can generate a fuzziness index for each edge point, reflecting the edge characteristics of high pain in vivo symptoms, thereby providing support for subsequent research. The high pain in vivo symptom fuzzy edge data finally obtained provides valuable information for further analysis and research.

[0162] Optionally, step S4 is specifically:

[0163] Step S41: calculating the area of ​​the lesion region according to the non-zero pixel ultrasound image of the patient, thereby obtaining the area data of the lesion region;

[0164] In this embodiment, in the process of calculating the area of ​​the lesion region according to the non-zero pixel ultrasound image of the patient, the processed ultrasound image data is imported. By converting the ultrasound image into a grayscale image and setting a threshold (for example, the threshold is set to 50), the non-zero pixels in the image are extracted to form a binary image. In this binary image, the non-zero pixels represent the lesion area, and the zero pixels represent the background. Next, the lesion area is identified and extracted by performing region marking processing on the binary image. The area of ​​the marked area is calculated using an image processing tool (such as OpenCV), and the area is expressed in the number of pixels. The lesion area data finally obtained provides important information for subsequent analysis.

[0165] Step S42: calculating the boundary line of the lesion area according to the non-zero pixel point ultrasound image of the patient, thereby obtaining the boundary line data of the lesion area;

[0166] In this embodiment, in the process of calculating the boundary line of the lesion area based on the non-zero pixel ultrasound image of the patient, the binary image obtained in the previous step continues to be used. An edge detection algorithm, such as Canny edge detection, is applied to detect the boundary of the lesion area. The Canny algorithm performs edge detection on the image by setting high and low thresholds (for example, the high threshold is 100 and the low threshold is 50) to identify the boundary line of the lesion area. In this process, the image needs to be smoothed to reduce the influence of noise, and Gaussian blur processing is usually used. The lesion area boundary line data finally obtained will provide the shape and distribution characteristics of the lesion, which is convenient for subsequent analysis.

[0167] Step S43: evaluating the lesion compactness according to the lesion area data and the lesion area boundary line data, thereby obtaining lesion compactness data;

[0168] In this embodiment, after obtaining the area data of the lesion area and the boundary line data of the lesion area, the lesion compactness is evaluated. The compactness can be calculated by the area and perimeter of the lesion area. The specific method is to first calculate the perimeter of the lesion area, usually using the boundary line data, and calculate the perimeter by accumulating the coordinates of the boundary points. Then, the area of ​​the lesion area is compared with the square of the perimeter to evaluate the compactness. The larger the compactness value, the more compact the lesion area. The evaluation results provide important indicators of lesion characteristics, which are helpful for further analyzing the nature of the lesion.

[0169] Step S44: measuring the roundness of the lesion region and evaluating the complexity of the lesion region contour according to the lesion region area data and the lesion region boundary line data, thereby obtaining the lesion region roundness data and the lesion region contour complexity data;

[0170] In this embodiment, in the process of measuring the roundness of the lesion area and evaluating the complexity of the lesion area contour, the lesion area area data and the lesion area boundary line data continue to be used. The roundness evaluation is calculated by the formula, and the formula is (4×π×area) / (perimeter 2). The roundness data is obtained based on the previously calculated area and perimeter of the lesion area. The evaluation of contour complexity is completed by calculating the ratio of the perimeter to the area of ​​the lesion area. A higher complexity value indicates that the lesion contour is more complex, and vice versa. The results of this step provide strong support for studying the characteristics of the lesion.

[0171] Step S45: performing lesion invasiveness analysis according to the lesion area roundness data and the lesion area contour complexity data, thereby obtaining lesion invasiveness data;

[0172] In this embodiment, the lesion invasiveness analysis is performed based on the roundness data of the lesion area and the complexity data of the lesion area contour. The invasiveness analysis combines the comprehensive evaluation of the above two indicators and uses a weighted algorithm to set the weights (for example, the roundness weight is set to 0.6 and the complexity weight is set to 0.4). The roundness and complexity are linearly combined by weight to obtain a comprehensive invasiveness score. This score can effectively reflect the potential invasive characteristics of the lesion and provide data support for subsequent tissue lesion analysis.

[0173] Step S46: performing tissue lesion feature weight allocation on the lesion invasiveness data and the lesion compactness data according to the tissue lesion type data, thereby obtaining tissue lesion type adaptive multi-feature data;

[0174] In this embodiment, after obtaining the lesion invasiveness data and lesion compactness data, tissue lesion feature weights are allocated according to the tissue lesion type data. According to literature or clinical data, a specific weight allocation rule is set (for example, for adenomatous lesions, the compactness weight is 0.7 and the invasiveness weight is 0.3), and adjusted according to the different tissue lesion types. The data of each feature is applied to the weighting formula through parameter transfer to ensure that the feature weights of different types of lesions reflect their actual characteristics. Finally, adaptive multi-feature data of tissue lesion types is obtained, laying the foundation for subsequent precise segmentation.

[0175] Step S47: accurately segmenting the tissue lesion region data according to the tissue lesion type adaptive multi-feature data, thereby obtaining accurate segmentation data of the tissue lesion region.

[0176] In this embodiment, the tissue lesion area data is accurately segmented according to the adaptive multi-feature data of the tissue lesion type. In this step, an ultrasound image segmentation algorithm (such as region growing or graph-based segmentation algorithm) is used to accurately segment the lesion area. According to the feature data and weights in step S46, the parameters of the segmentation algorithm are adjusted (for example, the similarity threshold in region growing is set to 0.8) to ensure that the algorithm can fully consider the influence of each feature during the segmentation process. The final output of the tissue lesion area accurate segmentation data provides accurate and reliable basic data for further research.

[0177] Optionally, step S45 is specifically:

[0178] Step S451: Calculating the roundness score according to the lesion area roundness data, thereby obtaining the lesion area roundness score data;

[0179] In this embodiment, in the calculation of the roundness score of the lesion area, the roundness data of the lesion area obtained in step S44 is first processed. The range of roundness values ​​is usually between 0 and 1, where 1 represents a perfect circle and 0 represents an irregular shape. In order to obtain the roundness score, the roundness data is first normalized to ensure that it is evaluated under a unified scoring standard. Specifically, the minimum-maximum normalization method is used to map the roundness value to a scoring range of 0 to 100. This process requires setting a minimum and maximum value, which corresponds to the scores when the roundness is 0 and 1, respectively. The score data is stored in a specific data structure for subsequent analysis.

[0180] Step S452: Calculating the contour complexity score according to the contour complexity data of the lesion area, thereby obtaining the contour complexity score data of the lesion area;

[0181] In this embodiment, the calculation of the contour complexity score is based on the contour complexity data of the lesion area obtained from step S44. First, the evaluation criteria for contour complexity are defined, and the complexity is usually expressed by the ratio of contour length to area. In specific implementation, the contour length and area are calculated, and the complexity score is derived using the ratio. For ease of scoring, the value is mapped to a range of 0 to 100. The same minimum-maximum normalization method as the roundness score is used so that the contour complexity score can be consistent with other feature scores, ensuring that all scores can be compared under a unified standard.

[0182] Step S453: obtaining medical clinical database data, and extracting lesion aggressiveness weight features according to the medical clinical database data, thereby obtaining lesion aggressiveness weight data;

[0183] In this embodiment, in the process of extracting weighted features of lesion aggressiveness, relevant data is first extracted from a medical clinical database. The database contains a large amount of information about lesion types, clinical manifestations, and their aggressiveness. When extracting features, the main focus is on variables related to lesion aggressiveness, such as the size of the lesion, the rate of development, and the patient's physiological parameters. These data are analyzed using data mining techniques to determine the importance of each feature in the assessment of aggressiveness. The results of feature extraction are presented in the form of weighted data, where features with high weights indicate that they have a greater influence in the assessment of aggressiveness. The weighted data will serve as the basis for subsequent scoring.

[0184] Step S454: Perform an invasiveness score according to the lesion invasiveness weight data, the lesion area roundness score data, and the lesion area contour complexity score data, so as to obtain lesion invasiveness data.

[0185] In this embodiment, the calculation of the invasiveness score combines the lesion area roundness score, the contour complexity score and the lesion invasiveness weight data. In the specific operation, the three data sources are first integrated together to form a comprehensive evaluation model. Each feature score is weighted according to its corresponding weight to ensure that the contribution of each feature to the final score reflects its actual importance. The final invasiveness score is calculated by weighted summation. The value of this score can be used to judge the potential invasiveness of the lesion and provide important reference data for subsequent research. Finally, the invasiveness score results are stored in a preset database for further analysis and comparison.

[0186] Optionally, step S47 is specifically:

[0187] Step S471: extracting texture features based on the adaptive multi-feature data of tissue lesion types, thereby obtaining texture data;

[0188] In this embodiment, during the process of texture feature extraction, texture features are first extracted from the adaptive multi-feature data of tissue lesion types obtained in step S46. Texture feature extraction uses grayscale co-occurrence matrix (GLCM) technology, which can effectively capture the spatial relationship between pixels in the image. In specific implementation, the ultrasound image of the tissue lesion area is first grayed to obtain a grayscale image. Then, a grayscale co-occurrence matrix is ​​constructed, and a variety of texture features are calculated, such as contrast, similarity, entropy and energy. These features will reflect the microstructure and texture characteristics of the tissue.

[0189] Step S472: selecting texture seed point pixels for tissue lesion area data according to the texture data, thereby obtaining texture seed point pixel data;

[0190] In this embodiment, in the process of selecting texture seed point pixels, operations are performed based on the texture data obtained in step S471. Threshold processing is performed on the extracted texture features to determine the significant features of the texture. By calculating the texture feature value of each pixel, pixels with feature values ​​higher than the set threshold are selected as seed points. This threshold can be dynamically adjusted according to the type of tissue lesions and the distribution of texture features. The 95% quantile is usually used as a standard to ensure that the selected seed points are representative and relevant. The selected texture seed point pixels will be recorded and stored as texture seed point pixel data for the subsequent definition of growth criteria.

[0191] Step S473: defining a growth criterion according to the texture seed point pixel data, thereby obtaining growth criterion data;

[0192] In this embodiment, in the process of defining the growth criteria, the texture seed point pixel data is used to formulate the criteria for regional growth. The growth criteria mainly include the similarity of adjacent pixels and the growth restriction conditions, and the specific implementation adopts the similarity threshold and the maximum growth area restriction. First, the similarity threshold is defined, which is set according to the standard deviation of the texture feature, and is generally set to the range of ±10% of the mean of the texture feature. The maximum limit of the growth area is set, that is, during the regional growth process, the growth area is limited to no more than four times the area of ​​the initial seed point. The above growth criteria will be encoded as a data structure for use in subsequent regional growth steps.

[0193] Step S474: performing region growth on the tissue lesion region data according to the texture seed point pixel data and the growth criterion data, thereby obtaining tissue lesion region growth data;

[0194] In this embodiment, the operation of regional growth is performed based on texture seed point pixel data and defined growth criteria. Starting from each selected seed point, an 8-neighborhood connection method is used to gradually check whether the surrounding pixels meet the growth criteria. For each neighborhood pixel, the texture similarity with the seed point is calculated, and it is determined whether it meets the similarity threshold. If it meets, the pixel is included in the current region, and its neighborhood pixels are further checked to continue expanding the region. This process is repeated until no new pixels can be added to form complete tissue lesion region growth data. The final growth data will be stored in a data structure for use in subsequent precise segmentation steps.

[0195] Step S475: performing precise regional segmentation on the tissue lesion region data according to the tissue lesion region growth data, thereby obtaining precise tissue lesion region segmentation data.

[0196] In this embodiment, in the process of accurate regional segmentation, processing is performed based on the growth data of the tissue lesion area. The growth area is post-processed to improve the accuracy of the segmentation. According to the boundaries of each area in the growth data, the edge detection algorithm (such as the Canny operator) is used to extract the edges of the area to ensure clear boundaries. Then, the region is merged and split, and the adjacent areas are judged using thresholds to ensure that the texture features of the adjacent areas are significantly different. Connectivity analysis is applied to ensure that all independent areas are reasonably segmented. After this series of processing, the obtained accurate segmentation data of the tissue lesion area will be stored for subsequent analysis and research.

[0197] Optionally, the present specification also provides a medical ultrasound image segmentation method and system based on adaptive feature fusion, which are used to perform the medical ultrasound image segmentation method based on adaptive feature fusion as described above. The medical ultrasound image segmentation system includes:

[0198] Disease classification model building module: used to obtain medical ultrasound image history data, and perform ultrasound medical history feature extraction based on the medical ultrasound image history data, thereby obtaining medical ultrasound image medical history data; and to construct an ultrasound image disease classification model based on the medical ultrasound image medical history data, thereby obtaining an ultrasound image disease classification model;

[0199] Ultrasonic layered tissue analysis module: used to obtain the patient's medical record ultrasound image, and remove zero pixels according to the patient's medical record ultrasound image, so as to obtain the patient's non-zero pixel ultrasound image; perform ultrasonic layered tissue analysis according to the patient's non-zero pixel ultrasound image, so as to obtain tissue lesion area data;

[0200] Tissue lesion classification module: used to calculate the lesion area ratio based on the tissue lesion area data and the patient's non-zero pixel ultrasound image, so as to obtain lesion area ratio data; classify the tissue lesion area data into disease types based on the ultrasound image disease classification model, so as to obtain tissue lesion type data;

[0201] Tissue lesion area segmentation module: used to perform adaptive multi-feature fusion on the patient's non-zero pixel ultrasound image according to the tissue lesion type data, so as to obtain adaptive multi-feature data of the tissue lesion type; accurately segment the tissue lesion area data according to the adaptive multi-feature data of the tissue lesion type, so as to obtain accurate segmentation data of the tissue lesion area;

[0202] Tissue lesion multimodal feature analysis module: used to perform multimodal feature analysis based on the precise segmentation data of tissue lesion areas and lesion area ratio data, so as to obtain multimodal feature data of tissue lesions and upload it to the medical cloud platform.

[0203] The present invention provides a medical ultrasound image segmentation method and system based on adaptive feature fusion. The system can implement any medical ultrasound image segmentation method based on adaptive feature fusion of the present invention, and is used to combine the operation and signal transmission medium between various modules to complete the medical ultrasound image segmentation method based on adaptive feature fusion. The internal modules of the system cooperate with each other to improve the accuracy of medical ultrasound image segmentation.

[0204] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0205] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A medical ultrasound image segmentation method based on adaptive feature fusion, characterized in that: The following steps are involved: Step S1: acquiring medical ultrasound image history data, and extracting ultrasound medical history features based on the medical ultrasound image history data, thereby obtaining medical ultrasound image medical history data; constructing an ultrasound image disease classification model based on the medical ultrasound image medical history data, thereby obtaining an ultrasound image disease classification model; Step S2: Acquire the patient's medical record ultrasound image, and remove zero pixels according to the patient's medical record ultrasound image, so as to obtain the patient's non-zero pixel ultrasound image; perform ultrasound layered tissue analysis according to the patient's non-zero pixel ultrasound image, so as to obtain tissue lesion area data; Step S3: calculating the lesion area ratio according to the tissue lesion area data and the non-zero pixel point ultrasound image of the patient, thereby obtaining lesion area ratio data; classifying the tissue lesion area data into disease types according to the ultrasound image disease classification model, thereby obtaining tissue lesion type data; Step S4: Adaptively fusing the non-zero pixel ultrasound image of the patient according to the tissue lesion type data to obtain adaptive multi-feature data of the tissue lesion type; accurately segmenting the tissue lesion area data according to the adaptive multi-feature data of the tissue lesion type to obtain accurate segmentation data of the tissue lesion area; Step S5: Perform multimodal feature analysis based on the precise segmentation data of the tissue lesion area and the lesion area ratio data, so as to obtain multimodal feature data of the tissue lesion and upload it to the medical cloud platform.

2. The medical ultrasound image segmentation method based on adaptive feature fusion according to claim 1, characterized in that: Step S1 is specifically as follows: Step S11: acquiring medical ultrasound image history data, and performing ultrasound medical history feature extraction based on the medical ultrasound image history data, thereby obtaining medical ultrasound image medical history data; Step S12: classifying the disease type according to the medical ultrasound image medical history data, thereby obtaining in vivo symptom disease data and in vitro symptom disease data; Step S13: performing symptom-driven feature enhancement according to the in vivo symptom disease data, thereby obtaining in vivo symptom disease-driven feature enhancement data; Step S14: performing symptom-driven feature enhancement according to the in vitro symptom disease data, thereby obtaining in vitro symptom disease-driven feature enhancement data; Step S15: integrating disease characteristic data according to the in vivo symptom disease driving characteristic enhancement data and the in vitro symptom disease driving characteristic enhancement data, thereby obtaining disease characteristic data; Step S16: constructing an ultrasound image disease classification model according to the disease feature data, thereby obtaining an ultrasound image disease classification model.

3. The medical ultrasound image segmentation method based on adaptive feature fusion according to claim 2, characterized in that: Step S13 is specifically as follows: Step S131: performing pain degree statistics according to the in vivo symptom data, thereby obtaining high-pain in vivo symptom data; Step S132: performing grayscale texture conversion according to the high-pain in vivo symptom data, thereby obtaining high-pain in vivo symptom grayscale texture data; Step S133: performing inhomogeneity analysis based on the high pain in vivo symptom grayscale texture data, thereby obtaining inhomogeneity grayscale texture data; Step S134: performing edge analysis based on the high pain in vivo symptom grayscale texture data, thereby obtaining high pain in vivo symptom fuzzy edge data; Step S135: Integrate the in vivo symptom disease characteristics according to the uneven grayscale texture data and the high-pain in vivo symptom fuzzy edge data, so as to obtain the in vivo symptom disease driving feature enhancement data.

4. The medical ultrasound image segmentation method based on adaptive feature fusion according to claim 2, characterized in that: Step S14 is specifically as follows: Step S141: performing body temperature statistics according to the extracorporeal symptom disease data, thereby obtaining high temperature extracorporeal symptom disease data; Step S142: dividing the high temperature extracorporeal symptom disease data into swelling regions, thereby obtaining high temperature swelling region data; Step S143: measuring the skin thickness according to the high temperature swelling area data, thereby obtaining the skin thickness data of the high temperature swelling area; Step S144: analyzing the skin color change according to the high temperature swelling area data, thereby obtaining the skin color data of the high temperature swelling area; Step S145: Integrate the in vitro symptom disease characteristics according to the skin thickness data and the skin color data of the high temperature swelling area, so as to obtain the in vitro symptom disease driving characteristic enhancement data.

5. The medical ultrasound image segmentation method based on adaptive feature fusion according to claim 3, characterized in that: Step S133 is specifically as follows: Gray-level co-occurrence matrix data are calculated based on the gray-level texture data of the high-pain body symptoms, thereby obtaining gray-level co-occurrence matrix data; The local area standard deviation is calculated according to the gray level co-occurrence matrix data, thereby obtaining the local area standard deviation data; Calculate the coefficient of variation based on the local area standard deviation and the local area standard deviation data to obtain the coefficient of variation data; Entropy calculation is performed according to the gray-level co-occurrence matrix data to obtain entropy data; The uneven grayscale texture is evaluated based on the coefficient of variation data and the entropy data, thereby obtaining the uneven grayscale texture data.

6. The medical ultrasound image segmentation method based on adaptive feature fusion according to claim 3, characterized in that: Step S134 is specifically as follows: Performing gradient calculation according to the high pain in vivo symptom grayscale texture data, thereby obtaining the high pain in vivo symptom grayscale texture gradient data; Threshold data is obtained by setting a threshold value according to grayscale texture data of high pain in vivo symptoms; According to the threshold data, the grayscale texture gradient data of the high pain body symptom is edge segmented to obtain weak edge data; Perform edge refinement according to the weak edge data, thereby obtaining weak edge refinement data; According to the weak edge data, the direction change amount is counted to obtain the uneven direction change data of the weak edge; The fuzziness of the high pain in vivo symptom grayscale texture data is calculated according to the weak edge direction uneven change data and the weak edge refinement data, so as to obtain the high pain in vivo symptom fuzzy edge data.

7. The medical ultrasound image segmentation method based on adaptive feature fusion according to claim 1, characterized in that: Step S4 is specifically as follows: Step S41: calculating the area of ​​the lesion region according to the non-zero pixel ultrasound image of the patient, thereby obtaining the area data of the lesion region; Step S42: calculating the boundary line of the lesion area according to the non-zero pixel point ultrasound image of the patient, thereby obtaining the boundary line data of the lesion area; Step S43: evaluating the lesion compactness according to the lesion area data and the lesion area boundary line data, thereby obtaining lesion compactness data; Step S44: measuring the roundness of the lesion region and evaluating the complexity of the lesion region contour according to the lesion region area data and the lesion region boundary line data, thereby obtaining the lesion region roundness data and the lesion region contour complexity data; Step S45: performing lesion invasiveness analysis according to the lesion area roundness data and the lesion area contour complexity data, thereby obtaining lesion invasiveness data; Step S46: performing tissue lesion feature weight allocation on the lesion invasiveness data and the lesion compactness data according to the tissue lesion type data, thereby obtaining tissue lesion type adaptive multi-feature data; Step S47: accurately segmenting the tissue lesion region data according to the tissue lesion type adaptive multi-feature data, thereby obtaining accurate segmentation data of the tissue lesion region.

8. The medical ultrasound image segmentation method based on adaptive feature fusion according to claim 1, characterized in that: Step S45 is specifically as follows: Step S451: Calculating the roundness score according to the lesion area roundness data, thereby obtaining the lesion area roundness score data; Step S452: Calculating the contour complexity score according to the contour complexity data of the lesion area, thereby obtaining the contour complexity score data of the lesion area; Step S453: obtaining medical clinical database data, and extracting lesion aggressiveness weight features according to the medical clinical database data, thereby obtaining lesion aggressiveness weight data; Step S454: Perform an invasiveness score according to the lesion invasiveness weight data, the lesion area roundness score data, and the lesion area contour complexity score data, so as to obtain lesion invasiveness data.

9. The medical ultrasound image segmentation method based on adaptive feature fusion according to claim 8, characterized in that: Step S47 is specifically as follows: Step S471: extracting texture features based on the adaptive multi-feature data of tissue lesion types, thereby obtaining texture data; Step S472: selecting texture seed point pixels for tissue lesion area data according to the texture data, thereby obtaining texture seed point pixel data; Step S473: defining a growth criterion according to the texture seed point pixel data, thereby obtaining growth criterion data; Step S474: performing region growth on the tissue lesion region data according to the texture seed point pixel data and the growth criterion data, thereby obtaining tissue lesion region growth data; Step S475: performing precise regional segmentation on the tissue lesion region data according to the tissue lesion region growth data, thereby obtaining precise tissue lesion region segmentation data.

10. A medical ultrasound image segmentation method and system based on adaptive feature fusion, characterized in that: Used to perform the medical ultrasound image segmentation method based on adaptive feature fusion as claimed in claim 1, the medical ultrasound image segmentation system comprises: Disease classification model building module: used to obtain medical ultrasound image history data, and perform ultrasound medical history feature extraction based on the medical ultrasound image history data, thereby obtaining medical ultrasound image medical history data; and to construct an ultrasound image disease classification model based on the medical ultrasound image medical history data, thereby obtaining an ultrasound image disease classification model; Ultrasonic layered tissue analysis module: used to obtain the patient's medical record ultrasound image, and remove zero pixels according to the patient's medical record ultrasound image, so as to obtain the patient's non-zero pixel ultrasound image; perform ultrasonic layered tissue analysis according to the patient's non-zero pixel ultrasound image, so as to obtain tissue lesion area data; Tissue lesion classification module: used to calculate the lesion area ratio based on the tissue lesion area data and the patient's non-zero pixel ultrasound image, so as to obtain lesion area ratio data; classify the tissue lesion area data into disease types based on the ultrasound image disease classification model, so as to obtain tissue lesion type data; Tissue lesion area segmentation module: used to perform adaptive multi-feature fusion on the patient's non-zero pixel ultrasound image according to the tissue lesion type data, so as to obtain adaptive multi-feature data of the tissue lesion type; accurately segment the tissue lesion area data according to the adaptive multi-feature data of the tissue lesion type, so as to obtain accurate segmentation data of the tissue lesion area; Tissue lesion multimodal feature analysis module: used to perform multimodal feature analysis based on the precise segmentation data of tissue lesion areas and lesion area ratio data, so as to obtain multimodal feature data of tissue lesions and upload it to the medical cloud platform.

Images