Composite image-based tumor cell assisted positioning system and method

Abstract

The application provides a tumor cell auxiliary positioning system and method based on a composite image, relates to the technical field of medical imaging, and comprises the following steps: a data acquisition module is used for synchronously collecting radio frequency echo signals, ultrasonic images and computed tomography data of a target biological tissue, and generating original fusion data sets; a feature processing module is used for spatial registration and feature extraction on the original fusion data sets, so as to obtain a multi-dimensional feature set; and an identification imaging module is used for extracting multi-frequency band radio frequency echo signal intensity level values of a target region based on the multi-dimensional feature set, generating a tissue radio frequency characteristic diagram through logarithmic compression, and performing pixel-level fusion with a synchronous ultrasonic image to generate a composite image. Through multi-modal data fusion and accurate calibration, the application determines a tumor boundary three-dimensional coordinate in combination with anatomical structure constraints, and improves the accuracy of tumor cell positioning.

Description

Technical Field

[0001] This invention relates to the field of medical imaging technology, and in particular to a tumor cell-assisted localization system and method based on composite images. Background Technology

[0002] In early breast cancer screening, traditional techniques for locating tumor cells have some limitations. The imaging clarity is not well adapted to the tissue depth. For example, when ultrasound technology is used alone for breast cancer screening, the shape and boundary of superficial breast tumors can usually be clearly shown. However, when the tumor is located in the deep layers of the breast (such as near the chest wall), the ultrasound waves will be attenuated when passing through thick breast tissue or chest wall muscles, which may result in blurred tumor images and unclear edges, increasing the difficulty of accurately locating the size and extent of tumor invasion.

[0003] Furthermore, its ability to integrate multi-dimensional information is relatively weak. For example, while magnetic resonance imaging (MRI) can present the anatomical details of breast soft tissue well when used alone for breast cancer tumor localization, it is not direct enough in reflecting functional information such as tumor blood perfusion. When encountering some early tumors with atypical blood supply characteristics, it may be difficult to accurately distinguish the boundary between the tumor and the surrounding benign proliferative tissue based solely on anatomical images, thus affecting the accuracy of localization. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a tumor cell-assisted localization system and method based on composite images, which can accurately fuse multi-source image information and improve the accuracy and efficiency of tumor cell localization.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0006] Firstly, a tumor cell-assisted localization system based on composite images includes:

[0007] The data acquisition module is used to simultaneously acquire radio frequency echo signals, ultrasound images, and computed tomography data of the target biological tissue to generate a raw fusion dataset.

[0008] The feature processing module is used to perform spatial registration and feature extraction on the original fused dataset to obtain a multi-dimensional feature set;

[0009] The recognition imaging module is used to extract the intensity level of multi-band radio frequency echo signals in the target area based on a multi-dimensional feature set, generate a tissue radio frequency characteristic map through logarithmic compression, and perform pixel-level fusion with synchronous ultrasound images to generate a composite image.

[0010] The localization analysis module is used to perform subcellular distribution analysis on composite images to obtain the set of coordinates of nuclear localization positive cells;

[0011] The positioning calibration module is used to select a set of preset biomarkers from the set of coordinates of positive cells with nuclear localization, construct spatial topological relationships, quantify the microenvironmental heterogeneity of spatial topological relationships, and generate positioning calibration parameters based on the regional heterogeneity distribution characteristics.

[0012] The positioning output module is used to determine the three-dimensional coordinates of the tumor boundary based on positioning calibration parameters, through spatial gradient change analysis and radio frequency signal characteristic attenuation correlation rules, so as to obtain the tumor cell positioning results.

[0013] Furthermore, spatial registration and feature extraction are performed on the original fused dataset to obtain a multi-dimensional feature set, including:

[0014] Based on the original fused dataset, rigid transformation and affine transformation are performed sequentially on radio frequency echo signals, ultrasound images and computed tomography data to achieve precise spatial alignment and generate a spatially registered fused dataset.

[0015] For the spatially registered fusion dataset, the texture features of ultrasound images and the density distribution features of computed tomography scans are extracted, and the frequency domain energy features of the radio frequency echo signal are calculated by fast Fourier transform.

[0016] The texture features of ultrasound images, density distribution features of computed tomography scans, and frequency domain energy features are integrated to obtain a multi-dimensional feature set.

[0017] Furthermore, based on the multi-band radio frequency echo signal intensity level values ​​extracted from the target region using a multi-dimensional feature set, a tissue radio frequency characteristic map is generated through logarithmic compression and then pixel-level fused with the synchronous ultrasound image to generate a composite image, including:

[0018] From the multi-dimensional feature set, the radio frequency echo signal intensity level values ​​of multiple frequency bands corresponding to the target area are extracted and separated;

[0019] The dynamic range logarithmic compression processing is performed on the multi-band radio frequency echo signal intensity level values ​​of the target area to generate a radio frequency signal intensity distribution matrix characterizing the radio frequency signal intensity distribution of the target area;

[0020] The radio frequency signal intensity distribution matrix is ​​mapped onto a two-dimensional plane corresponding to the spatial location of the ultrasound image to generate a tissue radio frequency characteristic map that reflects the radio frequency characteristics of the target area.

[0021] The radio frequency characteristic map of the tissue is combined with the ultrasound image of the corresponding target area acquired simultaneously according to a predetermined rule to achieve pixel-level superposition and fusion, generating a composite image.

[0022] Furthermore, subcellular distribution analysis was performed on the composite image to obtain the set of coordinates of nuclear-localized positive cells, including:

[0023] For the cell nuclear marker signal and the target protein marker signal in the composite image, the intensity correlation of the two signals is calculated pixel by pixel to obtain a three-dimensional data set that characterizes the degree of colocalization of each three-dimensional spatial location point;

[0024] Based on a three-dimensional dataset representing the degree of colocalization, by setting a threshold for determining nuclear localization positivity, all three-dimensional spatial regions in the three-dimensional dataset with values ​​greater than the threshold are identified as candidate nuclear localization positivity cell regions.

[0025] Based on the candidate nuclei, positive cell regions are located, and three-dimensional spatial connectivity analysis is applied to segment interconnected spatial regions into independent three-dimensional structures of individual candidate cells.

[0026] Based on the independent three-dimensional structure of a single candidate cell, the centroid coordinates of the cell in three-dimensional space are obtained by calculating the geometric center.

[0027] The centroid coordinates of all independent candidate cells in three-dimensional space are integrated to obtain the set of coordinates of nuclear-localized positive cells.

[0028] Furthermore, a set of pre-defined biomarkers is selected from the set of coordinates of nuclear-localized positive cells to construct spatial topological relationships. Microenvironmental heterogeneity is quantified within these spatial topological relationships, and localization calibration parameters are generated based on regional heterogeneity distribution characteristics, including:

[0029] Based on the set of coordinates of nuclear-positive cells, a set of three-dimensional coordinates of a subset of cells with preset biomarker identifiers is generated by filtering cell coordinates with preset biomarker identifiers.

[0030] Based on the three-dimensional coordinate set of the target biomarker cell subset, a three-dimensional spatial adjacency network is constructed with cell coordinates as nodes and spatial adjacency relationships as edges by calculating the three-dimensional Euclidean distance between each coordinate point and applying the adjacency distance threshold to determine connectivity.

[0031] Based on a three-dimensional spatial adjacency network, a comprehensive heterogeneity quantification dataset is generated by calculating the set of local cell density gradient values, the set of node clustering coefficient values, and the global spatial entropy value of all nodes in the network.

[0032] Based on the comprehensive heterogeneity quantification dataset, the heterogeneity spatial distribution characteristics of the metabolic target region are extracted by mapping it to a three-dimensional spatial grid of the metabolic target region and analyzing the numerical distribution within the grid cells.

[0033] Based on the description of heterogeneous spatial distribution characteristics, the boundary correction vector parameter set is derived by establishing the spatial mapping relationship between the features and the original boundary of the metabolic target area, and the positioning calibration parameters are generated.

[0034] Furthermore, based on the positioning calibration parameters, the three-dimensional coordinates of the tumor boundary are determined collaboratively through spatial gradient change analysis and radio frequency signal characteristic attenuation correlation rules to obtain tumor cell localization results, including:

[0035] Based on the positioning calibration parameters and lesion classification results, the initial spatial location data of the calibrated candidate boundary is generated by superimposing the spatial mapping information of the positioning calibration parameters onto the three-dimensional spatial coordinate frame of the original fusion dataset.

[0036] Based on the calibrated initial spatial location data of the candidate boundaries, the radio frequency signal attenuation feature value of the region near the candidate location is extracted; at the same time, the spatial gradient intensity value of the region near the same location is calculated in the ultrasound image component data, and the matching verification is performed based on the preset spatial collaborative association rules to generate a candidate boundary spatial location dataset.

[0037] Based on the candidate boundary spatial location dataset, a dynamic programming algorithm is applied to adjust the spatial continuity of boundary points and generate a set of boundary points that conform to anatomical structure constraints.

[0038] Based on the boundary point set that conforms to anatomical structure constraints, three-dimensional spatial coordinate information is extracted to obtain the three-dimensional coordinate point set of the tumor boundary as the result of tumor cell localization.

[0039] Furthermore, based on the candidate boundary spatial location dataset, a dynamic programming algorithm is applied to adjust the spatial continuity of the boundary points, generating a set of boundary points that conforms to anatomical structure constraints, including:

[0040] The three-dimensional coordinate set of the tumor boundary is used as a candidate boundary spatial location dataset and input into a framework based on a dynamic programming algorithm;

[0041] For the candidate boundary spatial location dataset, based on the anatomical structure constraints of the target organ, a path evaluation rule is generated to measure the smoothness of the connection path between any two candidate boundary points.

[0042] Based on the path evaluation rules, a dynamic programming algorithm is used to calculate and find the final boundary point spatial connection sequence of global smoothness in the candidate boundary spatial location dataset;

[0043] Based on the spatial connection sequence of boundary points, a final set of tumor boundary points that conforms to the smoothness constraints of anatomical structure is generated.

[0044] Secondly, tumor cell-assisted localization methods based on composite images include:

[0045] Step 1: Simultaneously acquire radio frequency echo signals, ultrasound images, and computed tomography data of the target biological tissue to generate the original fusion dataset;

[0046] Step 2: Spatial registration and feature extraction are performed on the original fused dataset to obtain a multi-dimensional feature set;

[0047] Step 3: Extract the multi-band radio frequency echo signal intensity level values ​​of the target area based on the multi-dimensional feature set, generate a tissue radio frequency characteristic map through logarithmic compression, and perform pixel-level fusion with the synchronous ultrasound image to generate a composite image;

[0048] Step 4: Perform subcellular distribution analysis on the composite image to obtain the set of coordinates of nuclear-localized positive cells;

[0049] Step 5: Select a set of preset biomarkers from the set of nuclear localization positive cell coordinates, construct spatial topological relationships, quantify the microenvironmental heterogeneity of spatial topological relationships, and generate localization calibration parameters based on regional heterogeneity distribution characteristics.

[0050] Step 6: Based on the positioning calibration parameters, the three-dimensional coordinates of the tumor boundary are determined by combining spatial gradient change analysis with the correlation rules of radio frequency signal characteristic attenuation to obtain the tumor cell localization results.

[0051] Thirdly, a computing device includes:

[0052] One or more processors;

[0053] A storage device for storing one or more programs that, when executed by one or more processors, enable the one or more processors to implement the system.

[0054] Fourthly, a computer-readable storage medium storing a program that, when executed by a processor, implements the system.

[0055] The above-described solution of the present invention has at least the following beneficial effects:

[0056] In terms of data fusion and utilization, radiofrequency echo signals, ultrasound images, and computed tomography data are collected simultaneously. Spatial registration is used to achieve precise alignment of multimodal data, and multidimensional features are extracted and integrated to provide a comprehensive and rich information foundation for tumor localization. This reduces localization bias caused by data partiality. In terms of localization accuracy and depth, a progressive localization path from macro to micro is constructed. The coordinates of nuclear localization positive cells are obtained through subcellular distribution analysis, which improves localization accuracy to the cellular level and enhances localization accuracy.

[0057] In terms of handling complex environments and calibration capabilities, by selecting preset biomarkers, constructing spatial topological relationships and quantifying microenvironmental heterogeneity, and generating positioning calibration parameters, the system fully considers the impact of the complexity of the tumor microenvironment on positioning, effectively corrects deviations in the positioning process, and makes the positioning results more consistent with the actual distribution of tumor cells, thus enhancing the reliability of positioning in complex tumor microenvironments. In terms of clinical application value, based on the positioning calibration parameters, the system coordinates the three-dimensional coordinates of the tumor boundary with spatial gradient change analysis and radiofrequency signal characteristic attenuation association rules, and ensures that the boundary conforms to anatomical constraints through dynamic programming algorithms. This provides a reliable basis for accurate tumor diagnosis, surgical planning, and efficacy evaluation, and reduces damage to normal tissues during surgery. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of a tumor cell-assisted localization system based on composite images provided in an embodiment of the present invention.

[0059] Figure 2 This is a schematic flowchart of a tumor cell-assisted localization method based on composite images provided in an embodiment of the present invention. Detailed Implementation

[0060] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0061] like Figure 1 As shown, embodiments of the present invention propose a tumor cell-assisted localization system based on composite images, comprising:

[0062] The data acquisition module is used to simultaneously acquire radio frequency echo signals, ultrasound images, and computed tomography data of the target biological tissue to generate a raw fusion dataset.

[0063] The feature processing module is used to perform spatial registration and feature extraction on the original fused dataset to obtain a multi-dimensional feature set;

[0064] The recognition imaging module is used to extract the intensity level of multi-band radio frequency echo signals in the target area based on a multi-dimensional feature set, generate a tissue radio frequency characteristic map through logarithmic compression, and perform pixel-level fusion with synchronous ultrasound images to generate a composite image.

[0065] The localization analysis module is used to perform subcellular distribution analysis on composite images to obtain the set of coordinates of nuclear localization positive cells;

[0066] The positioning calibration module is used to select a set of preset biomarkers from the set of coordinates of positive cells with nuclear localization, construct spatial topological relationships, quantify the microenvironmental heterogeneity of spatial topological relationships, and generate positioning calibration parameters based on the regional heterogeneity distribution characteristics.

[0067] The positioning output module is used to determine the three-dimensional coordinates of the tumor boundary based on positioning calibration parameters, through spatial gradient change analysis and radio frequency signal characteristic attenuation correlation rules, so as to obtain the tumor cell positioning results.

[0068] In this embodiment of the invention, simultaneously acquiring multiple data from the target biological tissue and generating a raw fusion dataset allows for more comprehensive information acquisition, covering biological tissue characteristics across different dimensions. This avoids the information limitations associated with single data types. Spatial registration and feature extraction of the raw fusion dataset ensure precise spatial correspondence between data from different sources, while extracting key multi-dimensional features. This allows the analysis to focus more on valuable information, improving the effectiveness and relevance of data processing. Based on the multi-dimensional feature set, high-metabolic target areas can be located precisely, pinpointing regions with high lesion activity. The three-dimensional images clearly present the spatial structure and distribution of the target area, providing an intuitive and concrete basis for a deeper understanding of lesion characteristics. Subcellular distribution analysis of the composite images yields the final determination. The nuclear localization positive cell coordinate set allows for in-depth analysis at the cellular level, accurately capturing the location of cells with specific characteristics. This makes the localization more closely match the actual state of biological tissues. By selecting preset biomarkers from the nuclear localization positive cell coordinate set to construct spatial topological relationships and generate localization calibration parameters, the differences in the microenvironment of biological tissues can be fully considered. Through quantitative analysis and parameter calibration, localization deviations caused by microenvironmental heterogeneity are reduced, improving the reliability of localization results. Based on the localization calibration parameters, the three-dimensional coordinates of the tumor boundary are determined, realizing the synergistic application of multiple aspects of information such as spatial gradient changes and radio frequency signal characteristic attenuation. This enables more precise delineation of tumor boundaries and accurate tumor cell localization results, providing reliable spatial location information for tumor diagnosis and treatment.

[0069] In a preferred embodiment of the present invention, spatial registration and feature extraction are performed on the original fused dataset to obtain a multi-dimensional feature set, which may include:

[0070] Based on the original fused dataset, rigid transformation and affine transformation are performed sequentially on radio frequency echo signals, ultrasound images and computed tomography data to achieve precise spatial alignment and generate a spatially registered fused dataset.

[0071] For the spatially registered fusion dataset, the texture features of ultrasound images and the density distribution features of computed tomography scans are extracted, and the frequency domain energy features of the radio frequency echo signal are calculated by fast Fourier transform.

[0072] The texture features of ultrasound images, density distribution features of computed tomography scans, and frequency domain energy features are integrated to obtain a multi-dimensional feature set.

[0073] In this embodiment of the invention, the spatial coordinate systems of radio frequency echo signals, ultrasound images, and computed tomography data are extracted from the original fusion dataset. Stable anatomical structures in biological tissues (such as bone edges and blood vessel bifurcation points) are selected as reference markers. The three-dimensional coordinates of these markers in the three types of data are recorded respectively. The deviation between the coordinates of the reference markers and the preset standard coordinates in each type of data is calculated. The origin of the coordinates is adjusted by translation operation to reduce the positional deviation of the reference markers in the x, y, and z axes to a preset range. Then, the coordinate axis direction is adjusted by rotation operation to keep the spatial orientation of the reference markers consistent. This process is repeated until the coordinate error of all reference markers in the three types of data is lower than a set threshold, thus completing the rigid transformation.

[0074] Based on the rigid transformation, more subtle anatomical feature points (such as tissue texture boundaries and density abrupt change points) are added as registration markers. The actual coordinates of these markers in the three types of data are measured respectively, and a coordinate correspondence table is established. According to the correspondence, the scaling factor that maximizes the overlap of the marker points in different data is calculated. The data scale is adjusted according to the coefficient in the x, y, and z axes to match the size of the marker points. At the same time, the shearing parameter is calculated to adjust the tilt angle of the data in the plane and space to eliminate the deformation caused by the difference in imaging perspective. Through multiple iterations to optimize the scaling and shearing parameters, the average coordinate deviation of all marker points is minimized, and a spatially registered fusion dataset is generated.

[0075] In the spatially registered ultrasound image, several non-overlapping square regions are divided. For all pixels within each region, the specific grayscale values ​​are counted. The sum of all grayscale values ​​is calculated and divided by the number of pixels to obtain the average grayscale value. The squared difference between each grayscale value and the average value is calculated, summed, and divided by the number of pixels to obtain the grayscale variance. Adjacent pixels within the region (horizontal, vertical, and diagonal directions) are selected, and the sum of their grayscale value products is calculated to obtain the correlation index. The sum of the absolute values ​​of the grayscale value differences between adjacent pixels is calculated to obtain the contrast index. The frequency of different grayscale values ​​within the region is counted, and the proportion of each grayscale value's frequency to the total number of pixels is calculated. The squared proportion is summed to obtain the energy index. The negative values ​​of the product of the proportion and the natural logarithm are summed to obtain the entropy index. These indices from all regions are summarized to form the texture features of the ultrasound image.

[0076] In the spatially registered computed tomography (CT) scan data, the three-dimensional range of the target tissue is defined, and the CT values ​​of voxels within this range are extracted one by one. The sum of all CT values ​​is calculated and divided by the number of voxels to obtain the average density value. The maximum and minimum values ​​of the CT values ​​are selected to determine the density variation range. The squared difference between each CT value and the average density value is calculated, summed, and divided by the number of voxels to obtain the density standard deviation. The density range is divided into several continuous sub-ranges, and the number of voxels in each sub-range is counted. The proportion of the number of voxels in each sub-range to the total number of voxels is calculated to form the density distribution histogram. The CT value (mode) corresponding to the peak value in the histogram, as well as the skewness and kurtosis of the histogram, are recorded. These data are combined to obtain the density distribution characteristics of the computed tomography scan.

[0077] Radio frequency echo signals of continuous time periods are extracted from the spatially registered fused dataset. The signal amplitude value at each time point is recorded, and the amplitude values ​​are arranged in chronological order to form a time-domain signal sequence. When applying the Fast Fourier Transform, the time-domain signal sequence is first divided into equal-length subsequences. The transform operation is performed on each subsequence to obtain the corresponding frequency components and the amplitude of each frequency. The energy (amplitude squared) of each frequency is calculated based on the amplitude. The total energy of all frequencies in the preset low-frequency, mid-frequency, and high-frequency bands is calculated to obtain the total energy of each frequency band. The proportion of energy of each frequency band to the total energy is calculated, and the frequency point with the highest energy (peak frequency) and the frequency range (bandwidth) from the peak frequency to when the energy drops to half are recorded. These data are summarized to form the frequency domain energy characteristics of the radio frequency echo signal.

[0078] We collect all specific values ​​of ultrasound image texture features, computed tomography density distribution features, and radio frequency echo signal frequency domain energy features. For each feature, we find its maximum and minimum values, calculate the difference between each feature value and the minimum value, and then divide by the difference between the maximum and minimum values ​​to standardize the feature values ​​to the 0-1 range. We arrange the standardized feature values ​​in the order of "texture features - density features - frequency domain features" to form a feature matrix containing multiple types of indicators. We check whether there are missing values ​​in the matrix and fill the missing data with the average value of the same type of feature. Finally, we obtain a complete multi-dimensional feature set.

[0079] The step-by-step operation of rigid transformation and affine transformation enables high-precision spatial matching of radiofrequency echo signals, ultrasound images, and computed tomography (CT) data. This allows for accurate correspondence between the same tissue regions described by different types of data. Extracting texture features from ultrasound images allows for detailed capture of the uniformity of grayscale distribution within the tissue and the correlation between pixels. These features effectively reflect structural differences in tissues, such as the difference in texture complexity between tumor tissue and normal tissue, providing rich structural evidence for distinguishing lesion areas. Extraction of density distribution features from CT scans can quantitatively reflect the density attributes and distribution patterns of tissues, revealing the differences in tissue density across different pathological states. These differences can serve as important criteria for identifying abnormal tissues, improving the accuracy of tissue property assessment. The frequency domain energy characteristics of radio frequency echo signals obtained through fast Fourier transform can convert frequency characteristics that are difficult to observe directly in the time domain into quantifiable indicators, reflecting the differences in tissue reflection characteristics to radio frequency signals. This provides a new dimension for distinguishing different tissue types, enriching the diversity of features. The integration of multi-dimensional feature sets combines features such as structure, density, and frequency, integrating the advantages of different imaging methods, avoiding the limitations of single features, and allowing the feature set to more comprehensively reflect the physiological and pathological characteristics of biological tissues, thus improving the reliability of overall analysis.

[0080] In a preferred embodiment of the present invention, the extraction of multi-band radio frequency echo signal intensity levels of the target region based on a multi-dimensional feature set, the generation of a tissue radio frequency characteristic map through logarithmic compression, and the pixel-level fusion with a synchronous ultrasound image to generate a composite image may include:

[0081] From the multi-dimensional feature set, the radio frequency echo signal intensity level values ​​of multiple frequency bands corresponding to the target area are extracted and separated;

[0082] The dynamic range logarithmic compression processing is performed on the multi-band radio frequency echo signal intensity level values ​​of the target area to generate a radio frequency signal intensity distribution matrix characterizing the radio frequency signal intensity distribution of the target area;

[0083] The radio frequency signal intensity distribution matrix is ​​mapped onto a two-dimensional plane corresponding to the spatial location of the ultrasound image to generate a tissue radio frequency characteristic map that reflects the radio frequency characteristics of the target area.

[0084] The radio frequency characteristic map of the tissue is combined with the ultrasound image of the corresponding target area acquired simultaneously according to a predetermined rule to achieve pixel-level superposition and fusion, generating a composite image.

[0085] In this embodiment of the invention, the data source range of the multi-dimensional feature set is determined. The feature set includes all radio frequency echo signal related data of the target area, such as a specific tissue area of ​​the human body and its surrounding area, including features such as signal acquisition time, signal propagation path, and original signal amplitude. Then, the feature data corresponding to the target area is screened. First, the spatial coordinate range of the target area is set, such as the XYZ axis coordinate interval determined by the previous positioning. The spatial position information of each signal data is checked one by one from the multi-dimensional feature set. All signal data whose spatial coordinates fall within the target area are retained, and surrounding signal data whose coordinates exceed the target area are removed.

[0086] Determine the multiple frequency bands to be separated, such as preset low-frequency, mid-frequency, and high-frequency bands. Each band corresponds to a fixed frequency range, such as 1-3MHz for low-frequency, 3-5MHz for mid-frequency, and 5-8MHz for high-frequency. Establish frequency filtering thresholds for each band. Perform frequency detection on the signal data of the filtered target area, calculating the actual frequency value of each signal data point one by one. Calculate the frequency using the "1 / period" method by detecting the signal period. Compare the calculated actual frequency value with the frequency threshold of each band. If the actual frequency of a signal falls within the low-frequency band threshold range, it is classified as a low-frequency radio frequency echo signal. Similarly, classify mid-frequency and high-frequency band signals respectively to complete the separation of signals in different frequency bands. Calculate the intensity level of all signals within each separated band. For each signal in each band, first measure its original amplitude data, then convert it according to the preset level value conversion standard, such as comparing the amplitude data with the reference voltage value, according to "level value = 20 × lg". The logic of "(signal amplitude / reference amplitude)" converts the amplitude data into the corresponding intensity level value by successively calculating the logarithm and multiplying by 20, ensuring that each signal in each frequency band has a unique corresponding intensity level value.

[0087] Determine the dynamic range of the multi-band RF echo signal strength levels in the target area. Collect the calculated strength levels in all frequency bands, compare each level value, and record the maximum and minimum values. The difference between the two values ​​is the current dynamic range of the signal strength. Set the target dynamic range for logarithmic compression. For example, based on the subsequent image display requirements, the preset target dynamic range is 0-60dB. Calculate the compression ratio by dividing the current dynamic range by the target dynamic range to obtain the compression coefficient. For example, if the current dynamic range is 0-120dB and the target is 0-60dB, the compression coefficient is 2. Perform point-by-point logarithmic compression on each strength level value in each frequency band. First, take each original strength level value. For example, if a signal level value is 80dB, calculate the difference between it and the minimum value of the original dynamic range. Then, divide this difference by the compression coefficient and finally add the minimum value of the target dynamic range to obtain the compressed level value. Repeat this operation to complete the logarithmic compression of all signal strength levels, ensuring that all compressed level values ​​fall within the target dynamic range.

[0088] To generate a radio frequency signal intensity distribution matrix, first determine the number of rows and columns of the matrix, ensuring this number matches the pixel row and column number of the subsequent ultrasound image. For example, if the ultrasound image is 512×512 pixels, the matrix is ​​set to 512 rows and 512 columns. Then, fill the compressed signal intensity level values ​​of each frequency band into the matrix according to the spatial location correspondence principle. Based on the original spatial coordinates (X and Y axes) of each signal, find the corresponding row number in the matrix (e.g., X coordinate corresponds to row number and Y coordinate corresponds to column number). Fill the compressed signal intensity level value into the corresponding matrix cell. If there is no direct signal data at a certain matrix cell, take the average of the signal intensity levels of its four adjacent positions, add the four adjacent values ​​together, and divide by 4 to obtain the intensity level value of that cell. Finally, a complete two-dimensional matrix representing the radio frequency signal intensity distribution of the target area is formed.

[0089] Acquire the spatial parameters of the synchronous ultrasound image, including the pixel size of the ultrasound image (e.g., each pixel corresponds to 0.1mm × 0.1mm in actual space), and the spatial coordinate system of the image (e.g., the upper left corner as the origin, the positive X-axis to the right, and the positive Y-axis downwards). Establish the correspondence between the radio frequency signal intensity distribution matrix and the spatial position of the ultrasound image, associating the row and column numbers of the matrix with the Y-axis and X-axis coordinates of the ultrasound image, respectively. For example, the cell in the first row and first column of the matrix corresponds to the pixel position of the upper left corner of the ultrasound image (X=0, Y=0), and the cell in the i-th row and j-th column of the matrix corresponds to the pixel position of the ultrasound image (X=j × 0.1mm, Y=i × 0.1mm). Ensure that the spatial position of each cell in the matrix corresponds to the pixel position of the ultrasound image. The spatial location is perfectly matched. The mapping rules between the level value and the image grayscale / color are set. For example, the compressed level value of 0-20dB is mapped to black, 20-40dB to gray, and 40-60dB to white, or different level values ​​are corresponding to rainbow color levels. The tissue radio frequency characteristic map is generated point by point. According to the level value of each cell in the radio frequency signal intensity distribution matrix, the grayscale value or color value corresponding to the cell is determined by referring to the mapping rules. For example, when the level value is 30dB, it corresponds to the medium grayscale value 128. Then, the grayscale / color value is assigned to the pixel point of the corresponding spatial location in the ultrasound image. All cells are processed one by one in the order of matrix rows and columns, and finally a tissue radio frequency characteristic map that is completely consistent with the spatial size of the ultrasound image is formed.

[0090] Acquire synchronously acquired ultrasound images of the target area, determine the pixel format of the ultrasound images (e.g., RGB or grayscale) and the numerical range of each pixel (e.g., grayscale image pixel values ​​are 0-255), set predetermined rules for pixel-level fusion (e.g., weighted overlay rules), preset the weight of the radiofrequency characteristic map to 0.4 and the weight of the ultrasound image to 0.6, verify the pixel position alignment of the two images, check the number of pixels in the tissue radiofrequency characteristic map and the ultrasound image one by one (e.g., both are 512×512 pixels), and randomly select 10 feature points (e.g., obvious contour points at the edge of the target area), check whether the pixel coordinates of the feature points in the radiofrequency characteristic map are completely consistent with the pixel coordinates of the corresponding feature points in the ultrasound image. If there is a deviation (e.g., a feature point is (100, 100) in the radiofrequency image and (101, 100) in the ultrasound image), then shift the radiofrequency characteristic map pixels to the left by 1 pixel to ensure that the coordinates of all feature points are aligned.

[0091] For each pixel location in the image, the pixel value of that location is first read from the tissue radiofrequency characteristic map, and then the pixel value of that location is read from the ultrasound image. The fused pixel value is calculated according to the weighting rules. If it is in RGB format, the fused value is calculated for the pixel values ​​of the R, G, and B channels respectively according to the same weighting rules. The fused pixel values ​​of all pixel locations are arranged in the row and column order of the original image to form a complete composite image, ensuring that each pixel contains both radiofrequency characteristic information and ultrasound image information.

[0092] By first screening the target area signal, then separating the frequency bands by frequency, and finally calculating the level value, the process ensures that the extracted signal comes only from the target area, avoiding the influence of surrounding noise. At the same time, separating the signal by frequency band and calculating the level value separately can preserve the unique information of radio frequency signals in different frequency bands, compressing the excessive intensity differences in the original signal to a range suitable for image display, and avoiding the loss of details caused by strong signals being too bright and weak signals being too dark in the original signal. Meanwhile, logarithmic compression can preserve the relative differences of weak signals, ensuring that subtle changes in radio frequency signals within the target area are not masked, and improving the image's ability to present subtle tissue features. By matching the spatial parameters of the ultrasound image and establishing the matrix-pixel correspondence, the process can ensure that the spatial size and coordinates of the tissue radio frequency characteristic map and the ultrasound image are completely consistent, guaranteeing the accurate spatial correspondence of the two image information. At the same time, mapping the level value to grayscale / color can transform abstract radio frequency signal data into intuitive image information, reducing the difficulty of interpreting signal data.

[0093] In a preferred embodiment of the present invention, performing subcellular distribution analysis on the composite image to obtain a set of coordinates of nuclear-localized positive cells may include:

[0094] For the cell nuclear marker signal and the target protein marker signal in the composite image, the intensity correlation of the two signals is calculated pixel by pixel to obtain a three-dimensional data set that characterizes the degree of colocalization of each three-dimensional spatial location point;

[0095] Based on a three-dimensional dataset representing the degree of colocalization, by setting a threshold for determining nuclear localization positivity, all three-dimensional spatial regions in the three-dimensional dataset with values ​​greater than the threshold are identified as candidate nuclear localization positivity cell regions.

[0096] Based on the candidate nuclei, positive cell regions are located, and three-dimensional spatial connectivity analysis is applied to segment interconnected spatial regions into independent three-dimensional structures of individual candidate cells.

[0097] Based on the independent three-dimensional structure of a single candidate cell, the centroid coordinates of the cell in three-dimensional space are obtained by calculating the geometric center.

[0098] The centroid coordinates of all independent candidate cells in three-dimensional space are integrated to obtain the set of coordinates of nuclear-localized positive cells.

[0099] In this embodiment of the invention, cell nuclear marker signals and target protein marker signals are extracted from the composite image. These two signals are processed pixel by pixel. Each pixel corresponds to a location point in three-dimensional space. For each location point, the intensity values ​​of the cell nuclear signal and the target protein signal are read respectively, and the correlation between these two intensity values ​​is calculated. Specifically, the trend of the intensity changes of the two signals at multiple adjacent pixels is statistically analyzed. If the trends are consistent, the correlation is high; otherwise, it is low. This correlation is represented by a specific numerical value, ranging from 0 to 1, where 1 represents complete correlation and 0 represents complete non-correlation. The correlation values ​​of all location points are arranged according to their three-dimensional spatial positions to form a three-dimensional data set representing the degree of co-location of each three-dimensional spatial location point.

[0100] A threshold for determining nuclear localization positivity is established. This threshold is set by analyzing the colocalization values ​​of known nuclear localization positive and negative samples. For example, a threshold of 0.6 is set, meaning that locations with a colocalization value greater than 0.6 are considered to be nuclear localization positive regions. The three-dimensional dataset representing colocalization is traversed, and the value of each location is checked one by one. If the value is greater than 0.6, the location is marked as a suspected nuclear localization positive point. The three-dimensional spatial regions where all marked suspected nuclear localization positive points are located are integrated to obtain all three-dimensional spatial regions in the three-dimensional dataset with values ​​greater than the determination threshold. These regions are the candidate nuclear localization positive cell regions.

[0101] For candidate nuclear localization positive cell regions, a three-dimensional connected component analysis algorithm is applied. First, an unlabeled location point is randomly selected from the candidate region as the starting point. Then, it is checked whether the adjacent locations of this starting point in three-dimensional space (front, back, left, right, up, and down) also belong to the candidate nuclear localization positive cell region. If they do, they are included in the current connected component. Then, using these newly included locations as the starting point, the adjacent locations are checked again, and the connected component range is continuously expanded until no new locations can be included. In this way, an independent connected component is obtained, representing a single candidate cell three-dimensional structure. The above process is repeated to process all unlabeled locations, dividing the entire candidate nuclear localization positive cell region into multiple independent single candidate cell three-dimensional structures.

[0102] For each independent candidate cell 3D structure, the coordinate information of all its 3D spatial locations is collected. The coordinates of each location point consist of three values: x, y, and z. The average value of the x-coordinates of these locations is calculated to obtain the center position of the cell 3D structure in the x-axis direction. Similarly, the average values ​​of the y-coordinates and z-coordinates are calculated to obtain the center positions in the y-axis and z-axis directions, respectively. The coordinates formed by these three average values ​​(x-average, y-average, z-average) are the geometric center of the single candidate cell 3D structure, that is, the centroid coordinates of the cell in 3D space.

[0103] Collect the centroid coordinates of all independent candidate cells in three-dimensional space, arrange them in a certain order (such as in ascending order of x-coordinate), check whether there are duplicates or errors in these coordinates, and correct or remove them if there are. The final set of all centroid coordinates is the set of coordinates of nuclear localization positive cells.

[0104] The intensity correlation of two signals is calculated pixel-by-pixel to obtain a three-dimensional data set characterizing the degree of co-localization. This method can accurately capture the spatial association between the cell nucleus and the target protein at the subcellular level. The correlation value at each location point can objectively reflect the degree of co-localization at that point, which helps improve the accuracy of nuclear localization analysis. Candidate nuclear localization positive cell regions are identified based on a judgment threshold. By reasonably setting the threshold, regions that may have nuclear localization positivity can be quickly screened from a large amount of three-dimensional data, reducing interference from irrelevant regions and making the analysis more focused on the target region. This improves the efficiency and specificity of nuclear localization positive cell identification. A three-dimensional connected component analysis algorithm is applied to segment the three-dimensional structure of independent individual candidate cells. This method can accurately divide complex, interconnected regions into individual cellular structures, avoiding confusion between multiple cellular structures and allowing each candidate cell to be analyzed separately, thus ensuring the accuracy of single-cell analysis. By calculating the geometric center, the centroid coordinates of the cell are obtained, enabling a concise coordinate value to represent the position of a single cell in three-dimensional space. This accurately reflects the spatial distribution characteristics of the cell, making the description of cell position more intuitive. By integrating all centroid coordinates into a set of coordinates for nuclear-localized positive cells, the spatial location information of all nuclear-localized positive cells is summarized, forming a systematic and complete coordinate data, which is helpful for in-depth research on the spatial distribution and functional relationship of nuclear-localized positive cells.

[0105] In a preferred embodiment of the present invention, a set of preset biomarkers is selected from the set of coordinates of nuclear localization positive cells to construct spatial topological relationships. Microenvironmental heterogeneity is quantified for these spatial topological relationships, and localization calibration parameters are generated based on regional heterogeneity distribution characteristics. This may include:

[0106] Based on the set of coordinates of nuclear-positive cells, a set of three-dimensional coordinates of a subset of cells with preset biomarker identifiers is generated by filtering cell coordinates with preset biomarker identifiers.

[0107] Based on the three-dimensional coordinate set of the target biomarker cell subset, a three-dimensional spatial adjacency network is constructed with cell coordinates as nodes and spatial adjacency relationships as edges by calculating the three-dimensional Euclidean distance between each coordinate point and applying the adjacency distance threshold to determine connectivity.

[0108] Based on a three-dimensional spatial adjacency network, a comprehensive heterogeneity quantification dataset is generated by calculating the set of local cell density gradient values, the set of node clustering coefficient values, and the global spatial entropy value of all nodes in the network.

[0109] Based on the comprehensive heterogeneity quantification dataset, the heterogeneity spatial distribution characteristics of the metabolic target region are extracted by mapping it to a three-dimensional spatial grid of the metabolic target region and analyzing the numerical distribution within the grid cells.

[0110] Based on the description of heterogeneous spatial distribution characteristics, the boundary correction vector parameter set is derived by establishing the spatial mapping relationship between the features and the original boundary of the metabolic target area, and the positioning calibration parameters are generated.

[0111] In this embodiment of the invention, the coordinate information of each cell is extracted one by one from the set of nuclear-localized positive cell coordinates. At the same time, it is checked whether each cell carries the identifier of a preset biomarker. The cell coordinates with the preset biomarker identifier are filtered out and organized into a new set. For example, if the preset biomarker is "CD44+", then all cell coordinates marked "CD44+" are selected from the nuclear-localized positive cell coordinates and arranged in order of their position in three-dimensional space to form a three-dimensional coordinate set of the target biomarker cell subset. Each coordinate contains specific values ​​of the three dimensions x, y, and z.

[0112] For each coordinate point in the three-dimensional coordinate set of the target biomarker cell subset, the three-dimensional Euclidean distance between it and all other coordinate points in the set is calculated. During the calculation, the difference between two coordinate points on the x, y, and z axes is calculated separately, each difference is squared and summed, and the square root is taken to obtain the straight-line distance between the two points. An adjacency distance threshold (e.g., 50 micrometers) is set. If the three-dimensional Euclidean distance between two coordinate points is less than or equal to the threshold, the two points are determined to be connected. Each cell coordinate is used as a node in the network, and the nodes with connectivity are connected by edges to form a three-dimensional spatial adjacency network with spatial adjacency relationships as edges, which intuitively presents the spatial connection between cells.

[0113] In a three-dimensional adjacency network, for each node, the number of cells within a sphere with a radius of 100 micrometers centered on that node is counted, and then divided by the volume of the sphere to obtain the local cell density of that node. By comparing the local cell densities of adjacent nodes, the rate of change of density, i.e., the local cell density gradient value, is calculated. The gradient values ​​of all nodes are summarized into a set of local cell density gradient values. The clustering coefficient of each node is calculated by first finding all adjacent nodes of that node, counting the actual number of edges between these adjacent nodes, and then dividing by the maximum possible number of edges (i.e., the number of any two combinations of adjacent nodes) to obtain the clustering coefficient of that node. The clustering coefficients of all nodes form a set of node clustering coefficient values. When calculating the global spatial entropy value of the network, the three-dimensional space is divided into several equally sized cubic grids, and the number of nodes in each grid is counted. Based on the distribution probability of the number of nodes in each grid, the global spatial entropy value is obtained according to the entropy calculation method. The set of local cell density gradient values, the set of node clustering coefficient values, and the global spatial entropy value of the network are integrated together to form a comprehensive heterogeneity quantification dataset.

[0114] The values ​​in the comprehensive heterogeneity quantification dataset are mapped onto a three-dimensional spatial grid of the metabolic target region, ensuring that each grid cell corresponds to a specific heterogeneity quantification value. The distribution of values ​​within each grid cell is analyzed, such as the average, maximum, and minimum values, as well as the proportion of values ​​in different intervals. Based on these statistical results, the spatial distribution characteristics of heterogeneity within the metabolic target region are described, such as which regions have high heterogeneity, which regions have low heterogeneity, and whether the trend of heterogeneity changes from the center to the edge is increasing or decreasing, thus forming a description of the spatial distribution characteristics of heterogeneity.

[0115] Based on the description of heterogeneous spatial distribution characteristics, a spatial mapping relationship between these characteristics and the original boundary of the metabolic target area is established, clarifying the original boundary position corresponding to different heterogeneous characteristics. According to the deviation between the heterogeneous distribution and the original boundary, a set of boundary correction vector parameters is derived. Each correction vector contains the distance and direction that need to be adjusted in the x, y, and z directions. For example, if the heterogeneous characteristics of a certain region show that the original boundary is biased inward, the correction vector is positive in that direction, indicating that the boundary is adjusted outward by a certain distance. These boundary correction vector parameter sets are integrated to generate positioning calibration parameters for precise calibration of the tumor boundary.

[0116] By selecting cell coordinates with preset biomarkers to generate a 3D coordinate set, it is possible to accurately focus on specific cell populations related to the research target, eliminate interference from irrelevant cells, and lay the foundation for constructing accurate spatial relationships. Calculating 3D Euclidean distance and combining it with an adjacency distance threshold to construct a 3D spatial adjacency network objectively reflects the actual spatial connectivity between cells, transforming abstract cell coordinates into an intuitive network structure. This facilitates clear observation of cell spatial distribution patterns and interactions, providing an effective tool for analyzing cell population organization. Calculating local cell density gradients, clustering coefficients, and global spatial entropy generates a comprehensive heterogeneity quantification dataset, quantifying the heterogeneity of cell distribution from local to global dimensions. This method captures the spatial distribution of cells, including density variations, aggregation levels, and overall disorder. These features are mapped onto a three-dimensional spatial grid, and heterogeneous spatial distribution characteristics are extracted. By closely integrating quantified heterogeneity data with the spatial location of metabolic target areas, the specific distribution patterns of heterogeneity within the target area are clearly revealed. This provides detailed spatial information for understanding the complexity of the microenvironment of metabolic target areas, helping to discover potential regional differences. Based on heterogeneity characteristics, boundary correction vectors are derived to generate positioning calibration parameters. This allows for precise correction of the original boundary according to the actual heterogeneity of the microenvironment, making the positioning results more consistent with the true state of biological tissues. This reduces positioning deviations caused by ignoring microenvironmental differences and improves the accuracy and reliability of tumor boundary positioning.

[0117] In a preferred embodiment of the present invention, based on positioning calibration parameters, the three-dimensional coordinates of the tumor boundary are determined collaboratively through spatial gradient change analysis and radio frequency signal characteristic attenuation correlation rules to obtain tumor cell localization results, which may include:

[0118] Based on the positioning calibration parameters and lesion classification results, the initial spatial location data of the calibrated candidate boundary is generated by superimposing the spatial mapping information of the positioning calibration parameters onto the three-dimensional spatial coordinate frame of the original fusion dataset.

[0119] Based on the calibrated initial spatial location data of the candidate boundaries, the radio frequency signal attenuation feature value of the region near the candidate location is extracted; at the same time, the spatial gradient intensity value of the region near the same location is calculated in the ultrasound image component data, and the matching verification is performed based on the preset spatial collaborative association rules to generate a candidate boundary spatial location dataset.

[0120] Based on a candidate boundary spatial location dataset, a dynamic programming algorithm is applied to adjust the spatial continuity of boundary points and generate a set of boundary points that conform to anatomical structural constraints. Specifically, this involves: inputting the 3D coordinate set of the tumor boundary as the candidate boundary spatial location dataset into a framework based on the dynamic programming algorithm; generating path evaluation rules for the candidate boundary spatial location dataset, based on the anatomical structural constraints of the target organ, to measure the smoothness of the connection path between any two candidate boundary points; calculating the final spatial connection sequence of boundary points with global smoothness in the candidate boundary spatial location dataset using the dynamic programming algorithm based on the path evaluation rules; and generating the final tumor boundary point set that conforms to the anatomical structural smoothness constraints based on the boundary point spatial connection sequence.

[0121] Based on the boundary point set that conforms to anatomical structure constraints, three-dimensional spatial coordinate information is extracted to obtain the three-dimensional coordinate point set of the tumor boundary as the result of tumor cell localization.

[0122] In this embodiment of the invention, positioning calibration parameters and lesion classification results are obtained. The positioning calibration parameters include data such as equipment error compensation and patient position offset correction, such as the spatial coordinate deviation values ​​between different imaging devices (e.g., ultrasound and radiofrequency devices), and the position correction amount caused by the patient's breathing and limb movement. The lesion classification results clarify the approximate type of the tumor, the possible growth range, and other information, such as whether it is a benign or malignant tumor, as well as the common boundary morphological characteristics of this type of tumor based on past cases. Next, the three-dimensional spatial coordinate frame of the original fusion dataset is processed. The original fusion dataset is a collection formed by integrating multiple modal data such as ultrasound image data and radiofrequency signal data. Its three-dimensional spatial coordinate frame is established according to the scanning range and accuracy of the imaging device. Each data point has corresponding x, y, and z axis coordinate values, which are used to identify the specific position of the point in the three-dimensional space of the human body.

[0123] Then, the spatial mapping information is superimposed. The spatial mapping information in the positioning calibration parameters, that is, the coordinate transformation rules after error correction, is applied to the three-dimensional spatial coordinate frame of the original fusion dataset. For example, if the positioning calibration parameters show that the x-axis of the ultrasound device has a 2mm offset, the x-coordinate of all data points in the original frame is increased by 2mm to complete the calibration and generate the initial spatial position data of the candidate boundary after calibration. These data points initially outline the possible boundary range of the tumor.

[0124] Based on the generated calibrated initial spatial location data of the candidate boundaries, each candidate location point is determined. Then, a specific nearby region is delineated centered on each candidate location point, such as a sphere with a radius of 5 mm. Within this nearby region, the attenuation characteristic value of the radio frequency signal is extracted. Specifically, the intensity data of all radio frequency signals in the region are collected, and the attenuation of the signal from transmission to reception is calculated. For example, if the transmitted signal strength is 100 units and the received signal strength is 30 units, then the attenuation at this point is 70 units. The attenuation values ​​of all points in the region are then combined, and the average or median value is taken as the radio frequency signal attenuation characteristic value of the region.

[0125] Simultaneously, a spatial location range with the same location as the candidate location is found in the ultrasound image component data. Within this range, the image spatial gradient intensity value is calculated. The gray values ​​of different pixels in the ultrasound image are different, and the gradient intensity reflects the rate of change of gray values. During the calculation, the gray value of each pixel in the region is first obtained, and then the gray value difference between adjacent pixels is compared. For example, for two adjacent pixels in the horizontal direction, the gray values ​​are 80 and 120 respectively, and their difference is 40. Then, the average gradient value is calculated as the image spatial gradient intensity value by combining the gray value differences of all adjacent pixels in the region.

[0126] Next, the preset spatial collaborative association rules are invoked, which stipulate the matching relationship that should be satisfied between the radiofrequency signal attenuation feature value and the ultrasound image spatial gradient intensity value at the tumor boundary. For example, when the radiofrequency signal attenuation feature value is 60-80 units, the ultrasound image spatial gradient intensity value should be in the range of 30-50 units. The two feature values ​​of the region calculated earlier are compared with the rules. If the matching conditions are met, the candidate location point is included in the candidate boundary spatial location dataset; if not, the point is excluded, and finally the candidate boundary spatial location dataset is generated.

[0127] Dynamic programming is used to adjust the spatial continuity of boundary points. The dynamic programming algorithm processes each candidate boundary point in a certain order (e.g., from the top to the bottom of the tumor, or from left to right). For two adjacent boundary points, the spatial distance and angle between them are calculated. If the distance is too large or the angle change is too abrupt, it indicates that the continuity between the two points is poor and there may be an error. The algorithm will adjust according to the preset continuity standard. For example, the maximum allowable distance between adjacent points is set to 10mm and the maximum allowable angle change is 30 degrees. If the distance between two adjacent points is 15mm, which exceeds the standard, a new point will be inserted between the two points to make the distance between adjacent points meet the requirements. If the angle change is too large, the coordinates of one of the points will be fine-tuned to make the angle within a reasonable range.

[0128] During the adjustment process, anatomical constraints are introduced. These constraints are determined based on human anatomy. For example, the tumor boundary cannot cross specific anatomical structures such as bones and blood vessels, and should conform to the morphological characteristics of normal tissues in that area. For instance, the boundary of a tumor near the lung cannot enter the normal air-filled area of ​​the lung; the boundary of a tumor near the liver should conform to the contour of the liver. The dynamic programming algorithm checks whether each adjusted boundary point conforms to these anatomical constraints. If it does not, the position of the boundary point is further adjusted until the curve or surface formed by all boundary points is spatially continuous and conforms to the anatomical constraints, ultimately generating a set of boundary points that conforms to the anatomical constraints.

[0129] Based on the generated set of boundary points that conform to anatomical constraints, the three-dimensional spatial coordinate information of each boundary point is extracted one by one, that is, the specific values ​​of the x, y, and z axes corresponding to each point. These three-dimensional coordinate points are arranged in order to form a complete set, which is the three-dimensional coordinate point set of the tumor boundary. This set of points can accurately depict the shape and position of the tumor boundary in three-dimensional space, thereby obtaining the tumor cell localization results and providing accurate location information for diagnosis and treatment.

[0130] The original data is calibrated by positioning calibration parameters to make the initial spatial location data more accurate. Then, by combining the synergistic matching and verification of radiofrequency signal attenuation characteristics and ultrasound image spatial gradient intensity, points that conform to the characteristics of tumor boundaries are further screened, improving the accuracy of tumor boundary localization and laying the foundation for accurate determination of tumor extent. Dynamic programming algorithm is applied to adjust the spatial continuity of boundary points to ensure a smooth transition of the tumor boundary in space and avoid abrupt turns or breaks. At the same time, anatomical structure constraints are introduced to make the tumor boundary conform to the normal anatomical structure characteristics of the human body and avoid the boundary crossing unreasonable tissue areas, thereby enhancing the rationality and reliability of the localization results and reducing the possibility of misdiagnosis and missed diagnosis. The obtained three-dimensional coordinate point set of the tumor boundary can clearly and accurately show the specific location and morphology of the tumor in the body, and more precise treatment plans can be formulated based on this information.

[0131] like Figure 2 As shown, embodiments of the present invention also provide a tumor cell-assisted localization method based on composite images, including:

[0132] Step 1: Simultaneously acquire radio frequency echo signals, ultrasound images, and computed tomography data of the target biological tissue to generate the original fusion dataset;

[0133] Step 2: Spatial registration and feature extraction are performed on the original fused dataset to obtain a multi-dimensional feature set;

[0134] Step 3: Extract the multi-band radio frequency echo signal intensity level values ​​of the target area based on the multi-dimensional feature set, generate a tissue radio frequency characteristic map through logarithmic compression, and perform pixel-level fusion with the synchronous ultrasound image to generate a composite image;

[0135] Step 4: Perform subcellular distribution analysis on the composite image to obtain the set of coordinates of nuclear-localized positive cells;

[0136] Step 5: Select a set of preset biomarkers from the set of nuclear localization positive cell coordinates, construct spatial topological relationships, quantify the microenvironmental heterogeneity of spatial topological relationships, and generate localization calibration parameters based on regional heterogeneity distribution characteristics.

[0137] Step 6: Based on the positioning calibration parameters, the three-dimensional coordinates of the tumor boundary are determined by combining spatial gradient change analysis with the correlation rules of radio frequency signal characteristic attenuation to obtain the tumor cell localization results.

[0138] It should be noted that this method is the same as the method described above for the system. All implementation methods in the above system embodiments are applicable to this embodiment and can achieve the same technical effect.

[0139] Embodiments of the present invention also provide a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0140] Embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0141] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A tumor cell-assisted localization system based on composite images, characterized in that, include: The data acquisition module is used to simultaneously acquire radio frequency echo signals, ultrasound images, and computed tomography data of the target biological tissue to generate a raw fusion dataset. The feature processing module is used to perform spatial registration and feature extraction on the original fused dataset to obtain a multi-dimensional feature set; The recognition imaging module is used to extract the intensity level of multi-band radio frequency echo signals in the target area based on a multi-dimensional feature set, generate a tissue radio frequency characteristic map through logarithmic compression, and perform pixel-level fusion with synchronous ultrasound images to generate a composite image. The localization analysis module is used to perform subcellular distribution analysis on composite images to obtain a set of coordinates for nuclear-localized positive cells. Specifically, it includes: calculating the intensity correlation of nuclear marker signals and target protein marker signals in the composite image pixel-by-pixel to obtain a three-dimensional data set characterizing the degree of co-localization of each three-dimensional spatial location point; based on this three-dimensional data set characterizing the degree of co-localization, identifying all three-dimensional spatial regions in the data set whose values ​​are greater than the threshold for nuclear localization positivity by setting a threshold, as candidate nuclear-localized positive cell regions; applying three-dimensional spatial connectivity analysis to segment interconnected spatial regions into independent individual candidate cell three-dimensional structures based on the candidate nuclear-localized positive cell three-dimensional structures; obtaining the centroid coordinates of the cell in three-dimensional space by calculating the geometric center based on the independent individual candidate cell three-dimensional structures; and integrating the centroid coordinates of all independent candidate cells in three-dimensional space to obtain the set of coordinates for nuclear-localized positive cells. The positioning and calibration module is used to select a set of preset biomarkers from the set of nuclear-positioned positive cell coordinates, construct spatial topological relationships, quantify the microenvironmental heterogeneity of the spatial topological relationships, and generate positioning and calibration parameters based on the regional heterogeneity distribution characteristics. Specifically, it includes: based on the set of nuclear-positioned positive cell coordinates, by filtering cell coordinates with preset biomarker identifiers, generating a three-dimensional coordinate set of a subset of target biomarker cells; and based on the three-dimensional coordinate set of the target biomarker cell subset, by calculating the three-dimensional Euclidean distance between each coordinate point and applying an adjacency distance threshold to determine connectivity, constructing a system with cell coordinates as nodes. A three-dimensional spatial adjacency network with spatial adjacency relationships as edges is constructed. Based on the three-dimensional spatial adjacency network, a comprehensive heterogeneity quantification dataset is generated by calculating the set of local cell density gradient values, the set of node clustering coefficient values, and the global spatial entropy value of all nodes in the network. According to the comprehensive heterogeneity quantification dataset, the heterogeneity spatial distribution feature description in the metabolic target area is extracted by mapping it to a three-dimensional spatial grid of the metabolic target area and analyzing the numerical distribution within the grid cells. Based on the heterogeneity spatial distribution feature description, the boundary correction vector parameter set is derived by establishing the spatial mapping relationship between the features and the original boundary of the metabolic target area, and the positioning calibration parameters are generated. The positioning output module is used to determine the three-dimensional coordinates of the tumor boundary based on positioning calibration parameters, through spatial gradient change analysis and radio frequency signal characteristic attenuation correlation rules, so as to obtain the tumor cell positioning results.

2. The tumor cell-assisted localization system based on composite images according to claim 1, characterized in that, Spatial registration and feature extraction are performed on the original fused dataset to obtain a multi-dimensional feature set, including: Based on the original fused dataset, rigid transformation and affine transformation are performed sequentially on radio frequency echo signals, ultrasound images and computed tomography data to achieve precise spatial alignment and generate a spatially registered fused dataset. For the spatially registered fusion dataset, the texture features of ultrasound images and the density distribution features of computed tomography scans are extracted, and the frequency domain energy features of the radio frequency echo signal are calculated by fast Fourier transform. The texture features of ultrasound images, density distribution features of computed tomography scans, and frequency domain energy features are integrated to obtain a multi-dimensional feature set.

3. The tumor cell-assisted localization system based on composite images according to claim 2, characterized in that, Multi-band radio frequency echo signal intensity levels of the target region are extracted based on a multi-dimensional feature set. These levels are then logarithmically compressed to generate a tissue radio frequency characteristic map, which is then pixel-level fused with a synchronous ultrasound image to generate a composite image, including: From the multi-dimensional feature set, the radio frequency echo signal intensity level values ​​of multiple frequency bands corresponding to the target area are extracted and separated; The dynamic range logarithmic compression processing is performed on the multi-band radio frequency echo signal intensity level values ​​of the target area to generate a radio frequency signal intensity distribution matrix characterizing the radio frequency signal intensity distribution of the target area; The radio frequency signal intensity distribution matrix is ​​mapped onto a two-dimensional plane corresponding to the spatial location of the ultrasound image to generate a tissue radio frequency characteristic map that reflects the radio frequency characteristics of the target area. The radio frequency characteristic map of the tissue is combined with the ultrasound image of the corresponding target area acquired simultaneously according to a predetermined rule to achieve pixel-level superposition and fusion, generating a composite image.

4. The tumor cell-assisted localization system based on composite images according to claim 3, characterized in that, Based on the positioning calibration parameters, the three-dimensional coordinates of the tumor boundary are determined collaboratively through spatial gradient change analysis and radio frequency signal characteristic attenuation correlation rules to obtain tumor cell localization results, including: Based on the positioning calibration parameters and lesion classification results, the initial spatial location data of the calibrated candidate boundary is generated by superimposing the spatial mapping information of the positioning calibration parameters onto the three-dimensional spatial coordinate frame of the original fusion dataset. Based on the calibrated initial spatial location data of the candidate boundaries, the radio frequency signal attenuation feature value of the region near the candidate location is extracted; at the same time, the spatial gradient intensity value of the region near the same location is calculated in the ultrasound image component data, and the matching verification is performed based on the preset spatial collaborative association rules to generate a candidate boundary spatial location dataset. Based on the candidate boundary spatial location dataset, a dynamic programming algorithm is applied to adjust the spatial continuity of boundary points and generate a set of boundary points that conform to anatomical structure constraints. Based on the boundary point set that conforms to anatomical structure constraints, three-dimensional spatial coordinate information is extracted to obtain the three-dimensional coordinate point set of the tumor boundary as the result of tumor cell localization.

5. The tumor cell-assisted localization system based on composite images according to claim 4, characterized in that, Based on the candidate boundary spatial location dataset, a dynamic programming algorithm is applied to adjust the spatial continuity of boundary points, generating a set of boundary points that conforms to anatomical structure constraints, including: The three-dimensional coordinate set of the tumor boundary is used as a candidate boundary spatial location dataset and input into a framework based on a dynamic programming algorithm; For the candidate boundary spatial location dataset, based on the anatomical structure constraints of the target organ, a path evaluation rule is generated to measure the smoothness of the connection path between any two candidate boundary points. Based on the path evaluation rules, a dynamic programming algorithm is used to calculate and find the final boundary point spatial connection sequence of global smoothness in the candidate boundary spatial location dataset; Based on the spatial connection sequence of boundary points, a final set of tumor boundary points that conforms to the smoothness constraints of anatomical structure is generated.

6. A method for tumor cell-assisted localization based on composite images, wherein the method implements the system as described in any one of claims 1 to 5, characterized in that, include: Step 1: Simultaneously acquire radio frequency echo signals, ultrasound images, and computed tomography data of the target biological tissue to generate the original fusion dataset; Step 2: Spatial registration and feature extraction are performed on the original fused dataset to obtain a multi-dimensional feature set; Step 3: Extract the multi-band radio frequency echo signal intensity level values ​​of the target area based on the multi-dimensional feature set, generate a tissue radio frequency characteristic map through logarithmic compression, and perform pixel-level fusion with the synchronous ultrasound image to generate a composite image; Step 4: Perform subcellular distribution analysis on the composite image to obtain the set of coordinates of nuclear-localized positive cells; Step 5: Select a set of preset biomarkers from the set of nuclear localization positive cell coordinates, construct spatial topological relationships, quantify the microenvironmental heterogeneity of spatial topological relationships, and generate localization calibration parameters based on regional heterogeneity distribution characteristics. Step 6: Based on the positioning calibration parameters, the three-dimensional coordinates of the tumor boundary are determined by combining spatial gradient change analysis with the correlation rules of radio frequency signal characteristic attenuation to obtain the tumor cell localization results.

7. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the system as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the system as described in any one of claims 1 to 5.

Images