Adrenal metastatic tumor prediction analysis method and system based on image recognition

Abstract

The invention relates to the field of image recognition, in particular to an adrenal metastatic tumor prediction analysis method and system based on image recognition. The method comprises the following steps: acquiring PET-CT monitoring images of the adrenal gland of a patient at a plurality of time points, performing metastatic tumor visual detection at one time point by one time point, and extracting metastatic tumor bounding boxes at the plurality of time points; performing surrounding tissue microenvironment dynamic fluctuation analysis and boundary immune influence degree evaluation according to the metastatic tumor bounding box, and constructing a multi-cycle immune reaction intensity diagram; performing metastatic tumor change trend evolution on the metastatic tumor bounding box to obtain metastatic tumor image change trend charts at different time points; and performing peripheral tissue texture feature difference identification according to the metastatic tumor bounding box so as to extract potential tumor metastatic cells. According to the method, the future change of the metastatic tumor is accurately predicted through dynamic prediction of the development trend of the adrenal metastatic tumor, and more intuitive metastatic tumor change conditions at different time points are provided.

Description

Technical Field

[0001] The present invention relates to the field of image recognition, and particularly to a method and system for predicting and analyzing adrenal metastases based on image recognition. Background Art

[0002] With the continuous development of medical imaging technology, image recognition technology, as a new artificial intelligence application, has made remarkable breakthroughs in the field of medical diagnosis. Especially in the early detection and accurate diagnosis of tumors, image recognition technology has shown great potential. As a common type of malignant tumor, adrenal metastases often pose great challenges due to their inconspicuous symptoms and difficulty in early diagnosis. With the progress of computer vision and deep learning algorithms, the prediction and analysis of adrenal metastases based on image recognition have gradually become an important tool for early tumor diagnosis, greatly improving the accuracy and efficiency of tumor detection.

[0003] The diagnosis of adrenal metastases usually relies on medical imaging data such as CT and MRI. However, traditional imaging analysis methods often rely on manual interpretation, which has certain subjectivity and errors. Since the manifestations of adrenal metastases are similar to those of other kidney diseases or tumors, some subtle lesion features may be missed during image analysis, resulting in missed diagnoses or misdiagnoses. At the same time, the diagnosis of adrenal metastases usually requires multiple imaging examinations, increasing the patient's visit burden and the consumption of medical resources. Therefore, there is an urgent need for a more efficient and accurate diagnostic method to help improve the early detection of adrenal metastases. Summary of the Invention

[0005] To solve the above technical problems, the present invention proposes a method and system for predicting and analyzing adrenal metastases based on image recognition to solve at least one of the above technical problems.

[0006] To achieve the above object, the present invention provides a method for predicting and analyzing adrenal metastases based on image recognition, including the following steps: Step S1: Collect the adrenal PET-CT monitoring images of a patient at multiple time points, and perform visual detection of metastases at each time point one by one to extract the bounding boxes of metastases at multiple time points; Step S2: Perform dynamic fluctuation analysis of the surrounding tissue microenvironment and evaluate the degree of immune influence on the boundary according to the bounding boxes of metastases, and construct a multi-cycle immune response intensity map; Step S3: Evolve the change trend of metastases for the bounding boxes of metastases to obtain the image change trend map of metastases at different time points; Step S4: Identify the differences in texture features of the surrounding tissues according to the bounding boxes of metastases, so as to extract potential tumor metastatic cells; Step S5: performing dynamic metastasis trajectory tracking according to the adrenal gland PET-CT monitoring image and the metastasis tumor image change trend diagram, and performing full-process metastasis trajectory fitting to construct a time-series metastasis trajectory diagram; Step S6: Perform rolling prediction of metastasis status based on the multi-cycle immune response intensity map and the potential tumor metastasis cell time series metastasis trajectory map to construct a metastasis prediction timeline.

[0007] The present invention can obtain the image data of the patient's adrenal metastasis at different time points through image monitoring at multiple time points, and provide more comprehensive dynamic data for subsequent analysis. Through visual detection of metastases at each time point, the bounding box of the metastasis at each time point can be accurately extracted, thereby providing basic data for subsequent analysis. The bounding box of the metastasis can provide effective spatial coordinates for the analysis and prediction of the tumor, making the subsequent prediction more reliable and operable. By analyzing the microenvironment of the tissue surrounding the metastasis, it can help reveal the dynamic changes of immune cells around the tumor, and then provide deep insights into the evolution of the tumor microenvironment. By evaluating the intensity of the immune response, the changes in the immune state in the tumor microenvironment can be identified, helping to evaluate the potential response of metastases to immunotherapy. By analyzing the change trend of the bounding box of the metastasis, the evolution process of the tumor at different time points can be clarified, revealing the trends of tumor growth, shrinkage or stability. The tumor image change trend map can help identify whether the tumor has the potential to metastasize at an early stage, and the change trend map can provide support for the early warning of metastases, help identify potential metastasis risks, and then take preventive measures. By analyzing the differences in the texture characteristics of the surrounding tissues, the potential existence of tumor metastatic cells can be identified, providing a more accurate metastasis warning. The differential analysis of texture features makes the image recognition process more refined, helps to distinguish normal tissue from metastatic tumor cells, and improves the detection accuracy of tumor metastasis. By analyzing the texture changes of surrounding tissues, we can gain a deep understanding of the changes in the tumor microenvironment and the interaction between tumor cells and surrounding tissues, providing support for subsequent clinical decision-making. Through the dynamic tracking of the metastasis trajectory, the path and speed of tumor metastasis can be effectively monitored to provide a more intuitive development trend of the disease. The full-process trajectory fitting can help track the evolution of tumors at different time points and reveal the full picture of tumor metastasis. The metastasis trajectory map can provide the key nodes on the metastasis path. Through the comprehensive analysis of the immune response intensity map and the metastasis trajectory map, the dynamic prediction of tumor development trends can be achieved, and the future changes of metastatic tumors can be predicted.

[0008] In this specification, a system for predicting and analyzing adrenal metastases based on image recognition is provided, which is used to execute the method for predicting and analyzing adrenal metastases based on image recognition as described above, comprising: An image processing module, configured to collect adrenal PET-CT monitoring images of a patient at multiple time points, perform visual detection of metastatic tumors at each time point one by one, and extract the bounding boxes of metastatic tumors at multiple time points; An immune response evaluation module, configured to perform dynamic fluctuation analysis of the surrounding tissue microenvironment and evaluate the degree of immune influence on the boundary according to the bounding boxes of the metastatic tumors, and construct a multi-cycle immune response intensity map; An image change module, configured to evolve the change trend of metastatic tumors for the bounding boxes of metastatic tumors, and obtain an image change trend map of metastatic tumors at different time points; A difference recognition module, configured to identify the differences in the texture features of the surrounding tissues according to the bounding boxes of metastatic tumors, so as to extract potential tumor metastatic cells; A metastasis trajectory tracking module, configured to perform dynamic metastasis trajectory tracking according to the adrenal PET-CT monitoring images and the image change trend map of metastatic tumors, and perform full-process metastasis trajectory fitting to construct a time-series metastasis trajectory map; A situation prediction module, configured to perform rolling prediction of the metastasis situation of the time-series metastasis trajectory map according to the multi-cycle immune response intensity map and potential tumor metastatic cells, and construct a metastasis prediction timeline.

[0009] The present invention provides comprehensive dynamic data for subsequent analysis by collecting adrenal PET-CT images at multiple time points, and can effectively track the changes of tumors. The metastatic tumors and their boundaries are accurately identified through image visual detection, avoiding subjective differences in manual judgment, and improving the detection efficiency and accuracy. The bounding boxes of metastatic tumors at multiple time points extracted provide core data support for subsequent analysis and are the basis for the analysis of other modules. By analyzing the immune response around the tumor, the change trend of the tumor microenvironment can be evaluated, and then the sensitivity of the tumor to the immune response can be inferred. Through the image change trend map, the changes of metastatic tumors at different time points can be tracked in real time, revealing the trends of tumor growth, shrinkage or stability. The change trend map can help identify the development trend of metastatic tumors, provide early warnings, and take intervention measures in advance. The identification of texture feature differences can reveal the subtle differences between tumor metastatic cells and surrounding normal tissues, improving the detection accuracy of metastatic cells. Through texture difference analysis, potential diffusion regions of metastatic tumors can be identified early, enhancing the early diagnosis ability of metastatic tumors. It can accurately track the path and speed of tumor metastasis, provide a clearer tumor development pattern, and help evaluate the risk and severity of metastasis. Through the fitting of the metastasis trajectory map, from the initial metastasis to possible diffusion regions. By analyzing the immune response and tumor metastatic cells and combining the metastasis trajectory map, the future trend prediction of tumor metastasis can be realized. Description of the Drawings

[0011] Figure 1Schematic diagram of the step process of a method for predicting and analyzing adrenal metastases based on image recognition according to the present invention; Figure 2 Schematic diagram of the detailed implementation steps of step S1; Figure 3 Schematic diagram of the detailed implementation steps of step S2; Figure 4 Schematic diagram of the detailed implementation steps of step S3. Detailed implementation manners

[0013] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0014] The embodiments of the present application provide a method and system for predicting and analyzing adrenal metastases based on image recognition. The execution subjects of the method and system for predicting and analyzing adrenal metastases based on image recognition include, but are not limited to, mechanical equipment, data processing platforms, cloud server nodes, network upload devices, etc. that carry the system, which can be regarded as general computing nodes of the present application. The data processing platform includes, but is not limited to, at least one of an audio and image management system, an information management system, and a cloud data management system.

[0015] Please refer to Figures 1 to 4 , the present invention provides a method for predicting and analyzing adrenal metastases based on image recognition. The method for predicting and analyzing adrenal metastases based on image recognition includes the following steps: Step S1: Collect adrenal PET-CT monitoring images of a patient at multiple time points, and perform visual detection of metastases at each time point one by one, and extract the bounding boxes of metastases at multiple time points; Step S2: Perform dynamic fluctuation analysis of the surrounding tissue microenvironment and evaluate the degree of immune influence on the boundary according to the bounding boxes of the metastases, and construct a multi-cycle immune response intensity map; Step S3: Evolve the change trend of the metastases for the bounding boxes of the metastases to obtain an image change trend map of the metastases at different time points; Step S4: Identify the differences in texture features of the surrounding tissues according to the bounding boxes of the metastases, so as to extract potential tumor metastatic cells; Step S5: Perform dynamic metastasis trajectory tracking according to the adrenal PET-CT monitoring images and the image change trend map of the metastases, and perform full-process metastasis trajectory fitting to construct a time-series metastasis trajectory map; Step S6: Perform rolling prediction of the metastasis situation on the time-series metastasis trajectory map according to the multi-cycle immune response intensity map and potential tumor metastatic cells, and construct a metastasis prediction timeline.

[0016] Through image monitoring at multiple time points, the present invention can obtain the imaging data of a patient's adrenal metastases at different time points, providing more comprehensive dynamic data for subsequent analysis. Through visual detection of metastases at each time point, the bounding boxes of metastases at each time point can be accurately extracted, providing basic data for subsequent analysis. The bounding boxes of metastases can provide effective spatial coordinates for tumor analysis and prediction, making subsequent predictions more reliable and operable. By analyzing the microenvironment of tissues surrounding the metastases, it is possible to help reveal the dynamic changes of immune cells around the tumor, and thus provide profound insights into the evolution of the tumor microenvironment. By evaluating the intensity of the immune response, changes in the immune state in the tumor microenvironment can be identified, helping to evaluate the potential response of metastases to immunotherapy. By analyzing the changing trend of the bounding boxes of metastases, the evolution process of the tumor at different time points can be clarified, revealing trends such as tumor growth, shrinkage, or stability. The tumor imaging change trend graph can help identify the potential for tumor metastasis at an early stage. The change trend graph can provide support for early warning of metastases, helping to identify potential metastatic risks, and then taking preventive measures. By analyzing the differences in texture features of surrounding tissues, the potential presence of tumor metastatic cells can be identified, providing a more accurate metastasis warning. The differential analysis of texture features makes the image recognition process more refined, helps to distinguish normal tissues from metastatic tumor cells, and improves the detection accuracy of tumor metastasis. By analyzing the texture changes of surrounding tissues, the changes in the tumor microenvironment and the interaction between tumor cells and surrounding tissues can be deeply understood, providing support for subsequent clinical decisions. By dynamically tracking the metastasis trajectory, the path and speed of tumor metastasis can be effectively monitored, providing a more intuitive trend of the disease development. The full-process trajectory fitting can help track the evolution of the tumor at different time points, revealing the whole picture of tumor metastasis. The metastasis trajectory graph can provide key nodes on the metastasis path. Through the comprehensive analysis of the immune response intensity graph and the metastasis trajectory graph, the dynamic prediction of the tumor development trend can be achieved, predicting the future changes of metastases.

[0017] In an embodiment of the present invention, refer to Figure 1 , which is a schematic diagram of the step flow of a method for predicting and analyzing adrenal metastases based on image recognition according to the present invention. In this example, the steps of the method for predicting and analyzing adrenal metastases based on image recognition include: Step S1: Collect the adrenal PET-CT monitoring images of the patient at multiple time points, and perform visual detection of metastases at each time point to extract the bounding boxes of metastases at multiple time points; In this embodiment, after obtaining the authorization of the patient, multiple PET-CT scans are performed, usually before treatment, after treatment, and during follow-up. The interval between each time point is several weeks to several months, and the specific time is determined according to clinical requirements. After each scan, the image data is saved and stored in the DICOM standard format to facilitate subsequent analysis and processing. Before visual detection, the collected PET-CT images are preprocessed, including denoising, normalization, and image enhancement. Gaussian filtering and other methods are usually used to remove noise and improve the image quality. The images are normalized to ensure that images at different time points can be compared. This can be achieved by histogram equalization of the images. Appropriate visual detection methods are selected, usually combining the experience of medical imaging experts and computer-aided diagnosis (CAD) systems to ensure the accuracy of the identification of metastatic tumors. During visual detection, the suspicious metastatic tumor areas are marked one by one. For each identified metastatic tumor, a bounding box is drawn to ensure accurate enclosure of the tumor area. The bounding box should tightly enclose the metastatic tumor for subsequent analysis. The coordinates, size, and shape features of each bounding box are recorded for subsequent data analysis and comparison. This information is usually stored in the form of CSV or a database for subsequent processing. The metastatic tumor bounding box data at each time point is integrated into a database to ensure that it can be accessed and analyzed at any time. This database should include the basic information of the patient, the image data at each time point, and the characteristics of the metastatic tumors. Ensure the integrity of the data and back up the database regularly to prevent data loss. Quality control is performed on the collected images and the marked bounding boxes to ensure the accuracy and reliability of the data. The marked results can be reviewed by random sampling, and internal team discussions and training are carried out regularly to ensure that all participants are proficient in image analysis methods and techniques. Preliminary analysis is performed on the extracted metastatic tumor bounding box data to evaluate the changes in metastatic tumors at different time points, including volume changes, morphological changes, etc. This will provide important basic data for subsequent metastatic tumor prediction analysis. If the volume change of the metastatic tumor at different time points shows a significant increase, it indicates that the tumor may be progressing.

[0018] Step S2: Perform dynamic fluctuation analysis of the surrounding tissue microenvironment and evaluate the degree of immune influence on the boundary according to the metastatic tumor bounding box, and construct a multi-cycle immune response intensity map; In this embodiment, first, it is necessary to extract the tissue region around the metastatic tumor from the PET-CT images at each time point. The coordinate information of the bounding box is used to crop a specific region, usually set to a range of 5-10 mm outside the bounding box, to capture the microenvironmental changes of the surrounding tissues. Record the specific location, size, shape, and corresponding image data of each extracted region for subsequent analysis. Image processing is performed on the extracted tissue regions, including denoising, enhancement, and normalization, to improve the accuracy of analysis. Commonly used image processing methods include Gaussian filtering, histogram equalization, etc. Microenvironmental features such as tissue density, blood flow, and cell density are extracted from the processed images. These features are usually implemented through image analysis software (such as ImageJ or MATLAB), and techniques such as threshold segmentation and morphological operations are used for analysis. Using the extracted microenvironmental feature data, calculate the feature values at each time point and perform time series analysis on them to evaluate the dynamic changes in the surrounding tissue microenvironment. The moving average or difference method can be used to analyze the fluctuations of the feature values. If the cell density of the surrounding tissue at a certain time point is 1000 cells / mm² and the next time point is 1200 cells / mm², record this change and calculate the change rate. According to the characteristics of the surrounding tissue and the boundary information of the metastatic tumor, construct an immune response intensity map. This map reflects the impact of the metastatic tumor on the surrounding immune environment, and immunohistochemical staining or immunofluorescence techniques are usually used to obtain the immune cell distribution information. By staining the tissue region (such as markers like CD3, CD8, CD68, etc.) and taking images with a microscope, evaluate the immune cell infiltration at different time points. Image analysis is performed on the stained tissue region to count the number and distribution of different immune cells (such as T cells, macrophages, etc.). Image analysis software can be used to automatically count the number of cells and calculate their proportion in the surrounding tissue region. If at a certain time point, the number of CD8+ T cells is 200 and the number of macrophages is 150, the proportion of immune cells and its change trend can be calculated. According to the distribution and number of immune cells, evaluate the impact of the metastatic tumor on the surrounding tissue immune environment. At each time point, calculate the density and activity of immune cells and perform comprehensive analysis in combination with the microenvironmental characteristics of the surrounding tissue. If significant CD8+ T cell infiltration is found near the boundary of the metastatic tumor and less infiltration in the region far from the boundary, it indicates that the tumor has an obvious impact on the immune environment. Integrate the immune response intensity data at each time point to construct a multi-cycle immune response intensity map. The image at each time point should indicate the corresponding immune cell characteristics, microenvironmental characteristics, and their changes. Generate a heat map containing multiple time points, with the intensity of the immune response represented by the depth of color for intuitive analysis. Record the immune response intensity map for each time window and generate charts or images to display the immune environment changes at different time points. Visualization tools such as GraphPad Prism or Tableau can be used for data display.The generated heatmap can display the dynamic changes in the intensity of the immune response over the timeline, providing support for subsequent analysis and clinical decision-making. Analyze the generated multi-cycle immune response intensity maps to explore their relationship with the progression of metastatic tumors. Analyze how the dynamic changes in the immune environment affect tumor growth and metastasis. If it is found that the intensity of the immune response significantly decreases during the period when the tumor grows rapidly, it indicates that the immune escape mechanism may be at work.

[0019] Step S3: Evolve the change trend of the metastatic tumor for the metastatic tumor bounding box to obtain the image change trend map of the metastatic tumor at different time points; In this embodiment, before analyzing the changing trend of metastatic tumors, it is first necessary to determine key imaging indicators. These indicators usually include the volume, area, shape factors (such as perimeter, roundness, etc.) of metastatic tumors, as well as image gray values, etc. Selecting these indicators can help comprehensively evaluate the changes of tumors. The volume can be estimated by using a geometric model based on the coordinate information of the bounding box and the resolution of the image. At each time point, for the bounding box of each metastatic tumor, record the values of the above indicators. These data will serve as an important basis for subsequent analysis. At time point T1, if the volume of the metastatic tumor is 15 cm³ and the perimeter is 25 cm, record this information in the database. For the image data at different time points, image registration is required to ensure comparison in the same coordinate system. This usually adopts feature point-based registration methods such as SIFT (Scale-Invariant Feature Transform) or SURF (Speeded-Up Robust Features). Through registration, the image deviation caused by the change of the patient's position or equipment differences can be eliminated, thereby improving the accuracy of change analysis. Use automated image analysis tools (such as ImageJ, MATLAB, or professional medical image analysis software) to analyze the metastatic tumors at each time point, extract the boundaries of the metastatic tumors, and calculate the key indicators. Use the threshold segmentation method to extract the contours of the metastatic tumors and calculate their volume and area. For the metastatic tumor indicators at each time point, calculate the change rate (such as percentage change) between adjacent time points. This can help identify the rate of tumor growth or shrinkage. If the volume at time point T1 is 15 cm³ and the volume at time point T2 is 20 cm³, the change rate is [(20 - 15) / 15] × 100% = 33.33%. Integrate the change data of the metastatic tumor indicators at different time points to generate a changing trend chart. The chart can be in the form of a line chart or a bar chart, etc., to visually display the changes of the tumor. Draw a line chart with the time point on the X-axis and the volume of the metastatic tumor on the Y-axis, and mark the volume value and change rate at each time point. Analyze the generated changing trend chart to observe the growth pattern and rate of the tumor. By comparing the changes between different time points, evaluate the progress of the tumor. If the growth rate of the tumor volume significantly accelerates within a certain period, it may indicate a change in the biological behavior of the tumor. Based on the changing trend chart, combined with clinical information, discuss and evaluate the prognosis of the metastatic tumor. Analyze the relationship between tumor changes and treatment effects or patient survival. If it is found that the tumor volume significantly shrinks after treatment, it can be inferred that the treatment is effective, while if the tumor continues to grow, the treatment strategy may need to be adjusted.

[0020] Step S4: Identify the differences in the texture features of the surrounding tissues according to the bounding box of the metastatic tumor, so as to extract potential tumor metastatic cells; In this embodiment, based on the metastatic tumor bounding box determined in step S1, the bounding box is extended to define the surrounding tissue region. The extension range is usually 5 - 10 mm to ensure capturing sufficient peripheral tissue features. Record the coordinates and sizes of these extended regions for subsequent analysis. Preprocess the extracted peripheral tissue images, including denoising, normalization, and enhancement. Common denoising methods such as Gaussian filtering, mean filtering, etc. can improve the image quality and reduce interference. Apply a Gaussian filter with a standard deviation of 1 to smooth the image and remove small noise points. Use the gray-level co-occurrence matrix (GLCM) method to extract the texture features of the peripheral tissue. For each extracted region, calculate the GLCM in multiple directions (such as 0°, 45°, 90°, 135°) and extract eigenvalue, such as contrast, correlation, energy, and entropy, etc. If the GLCM energy value calculated in a certain direction is 0.5, then the texture uniformity in this direction can be recorded and compared with the features in other directions. Compare the texture features of the peripheral tissue of the metastatic tumor with those of the normal tissue region, and use statistical analysis methods (such as t-test or ANOVA) to evaluate the significant differences in features. If the average energy of the peripheral tissue of the metastatic tumor is 0.45 and the average energy of the normal tissue is 0.65, conduct a t-test analysis to determine the statistical significance of the difference. Set the significance level (usually p < 0.05) to judge whether there are significant differences in the extracted texture features between the peripheral tissue of the metastatic tumor and the normal tissue. If the p-value is less than the set threshold, it is considered that there are significant differences in the texture features between the two. If a p-value of 0.03 is found in the statistical analysis, it can be considered that there are significant differences in this feature between the tissue around the metastatic tumor and the normal tissue. Potential metastatic cell identification: Identify potential tumor metastatic cells based on the differences in texture features. Usually, the presence of tumor metastatic cells will cause significant changes in the texture features of the surrounding tissue (such as increased cell density, disordered tissue structure, etc.). If a significant increase in the entropy value of the texture features is found in a certain region compared to the normal tissue, this may indicate the presence of potential tumor metastatic cells in this region. Select appropriate cell identification methods, usually including image segmentation techniques (such as threshold segmentation, region growing, or deep learning-based segmentation methods). Combine the information on the differences in texture features to implement the identification and extraction of cells. Adopt a threshold-based segmentation method and set an appropriate gray-level threshold to separate the tumor metastatic cells from the surrounding tissue. Count the identified potential tumor metastatic cells and record their positions and morphological features (such as size, shape, etc.). These data will provide a basis for subsequent biological analysis. If 50 potential metastatic cells are counted in a certain region, record the cell density and distribution characteristics of this region. Integrate the data of the extracted potential tumor metastatic cells with the analysis results of the peripheral tissue texture features to form a complete analysis report. This report should include the cell quantity, position, texture features, and their clinical significance.The generated report may include that the cell density of the tissue around the metastatic tumor is 200 cells / mm², and the texture features of this area are significantly different from those of normal tissues, thus indicating the risk of potential metastasis.

[0021] Step S5: Perform dynamic metastasis trajectory tracking based on the adrenal PET-CT monitoring image and the image change trend chart of the metastatic tumor, and perform full-process metastasis trajectory fitting to construct a time-sequential metastasis trajectory chart; In this embodiment, adrenal PET-CT monitoring images and images showing the changing trends of metastatic tumors at multiple time points are collected. These images should have been preprocessed and analyzed to ensure reliable data quality. Integrate information such as the location, volume, and bounding box of metastatic tumors at each time point for subsequent tracking. Usually, a database is used to record this information to ensure data traceability. Review the bounding boxes of metastatic tumors extracted in step S1 and confirm whether the locations of metastatic tumors at each time point are consistent. This can be achieved through image registration techniques to ensure that images at different time points are aligned in the same coordinate system. The method of feature point matching (such as SIFT or SURF algorithm) is used for image registration to ensure accurate identification of the dynamic changes of metastatic tumors. Track the location of metastatic tumors at each time point and record the center coordinates of the bounding box at each time point. By calculating the coordinate changes between consecutive time points, the movement trajectory of the tumor can be understood. If the center coordinates of the metastatic tumor at time point T1 are (30, 40) and at time point T2 are (32, 42), then record the displacement between these two time points. Organize the coordinate data at each time point into a sequence to form the dynamic trajectory data of the metastatic tumor. These data can include information such as the coordinates, volume, and change rate at each time point. Organize the coordinate data at each time point into a list for subsequent trajectory analysis. Use data visualization tools (such as Matplotlib or Tableau) to plot the dynamic trajectory of the metastatic tumor as a graph. The X-axis usually represents time, and the Y-axis represents the change in the location or volume of the metastatic tumor. Plot the movement trajectory of the tumor at different time points, mark the location and volume changes at each time point, and form an intuitive dynamic trajectory graph. Select an appropriate fitting model to analyze the dynamic metastasis trajectory. Commonly used models include linear regression, polynomial regression, or more complex machine learning models (such as support vector regression SVR). Determine the parameters and fitting methods of the model so that it can accurately describe the growth and migration trends of the tumor. Fit the organized trajectory data, calculate the fitting curve. Optimize the model parameters by methods such as the least squares method to obtain the best fitting effect. If a polynomial regression model is used, it may be necessary to select an appropriate polynomial order (such as quadratic or cubic polynomial) to ensure the accuracy of the fitting. Evaluate the fitting results and use indicators such as R² (coefficient of determination) to judge the goodness of fit. An R² value close to 1 indicates a good fitting effect of the model. If the R² value of the model is 0.95, it means that the fitting effect is good and can accurately reflect the dynamic changes of metastatic tumors. Use the fitting results to construct a time-series metastasis trajectory graph. This graph should show the changes of metastatic tumors at each time point, including information such as location, volume, and growth rate. Generate a time-series graph containing the fitting curve to intuitively show the change trend of the tumor during the entire observation period. Record the constructed time-series metastasis trajectory graph and related data in the database to ensure data integrity and traceability. These records will provide a basis for subsequent research and clinical decisions.Generate a report to record the location, volume change of metastatic tumors, and fitting model parameters at each time point. Based on the analysis results of the temporal metastasis trajectory diagram, discuss its application value in clinical practice. Analyze how the dynamic changes of metastatic tumors affect treatment strategies and prognosis judgment.

[0022] Step S6: Perform a rolling prediction of the metastatic tumor situation on the temporal metastasis trajectory diagram according to the multi-cycle immune response intensity diagram and potential tumor metastatic cells, and construct a metastatic tumor prediction timeline.

[0023] In this embodiment, relevant data of multi-cycle immune response intensity maps and potential tumor metastatic cells are collected. The immune response intensity maps should be able to reflect the immune cell infiltration status and its changes at each time point, while the data of potential tumor metastatic cells should include the identified cell number, location, and characteristics. Record the immune response intensity values and the counts of potential metastatic cells at different time points for subsequent analysis. Review the previously generated temporal metastasis trajectory map to ensure the accuracy of the dynamic change data (such as volume, location, etc.) of the metastatic tumors. This graph will serve as the basic data for the prediction model. Confirm the volume and location information of the metastatic tumors at each time point and ensure consistency with the data in the immune response intensity maps. Integrate the dynamic change data of the metastatic tumors with the data of immune response intensity and potential metastatic cells to form a comprehensive dataset. The data at each time point should include information such as the volume of the metastatic tumor, immune response intensity, and the number of potential metastatic cells. Construct a table containing time points, metastatic tumor volume, immune intensity, and metastatic cell counts for subsequent analysis. Select an appropriate prediction model for the rolling prediction of the metastatic tumor situation. Commonly used models include time series analysis models (such as ARIMA) and machine learning models (such as regression models, random forests, etc.). Model selection should consider the characteristics of the data and the prediction purpose. If the changes of the metastatic tumors show obvious seasonality or periodicity, a seasonal ARIMA model can be considered for prediction. Use the integrated dataset to train the selected prediction model. Divide the historical data into a training set and a test set to ensure that the model can accurately capture the relationships among the volume of the metastatic tumor, immune response, and potential metastatic cells. Use the data of the past 6 months for model training and evaluate the performance of the model on the validation set to ensure good prediction ability. Conduct rolling predictions for future time windows to generate the expected volume, immune response intensity, and changes in potential metastatic cells of the metastatic tumors at future time points. The prediction results for each time window will provide a basis for subsequent analysis. Predict the volume changes and immune response intensity of the metastatic tumors for each month within the next 3 months to identify potential tumor progression risks. According to the results of the rolling predictions, construct a metastatic tumor prediction timeline. The timeline should show the changes in the volume of the metastatic tumor, immune response intensity, and potential metastatic cells for each time window to visually display the dynamic change trend of the tumor. Generate a timeline containing different time points, mark the predicted volume and immune response intensity at each time point, and facilitate the observation of the tumor development trend. Record the generated prediction timeline and relevant data in the database to ensure data integrity and traceability. At the same time, use visualization tools (such as Matplotlib, Tableau, etc.) to display the prediction results in a graphical form to enhance readability. Create charts to show the volume changes and immune intensity of the metastatic tumors at different time points.

[0024] In this embodiment, refer to Figure 2, which is a schematic diagram of the detailed implementation steps of step S1. In this embodiment, the detailed implementation steps of step S1 include: Collect adrenal PET-CT monitoring images of the patient at multiple time points; Sharpen and enhance the PET-CT monitoring images and perform histogram equalization to construct a globally optimized monitoring image; Perform adaptive filtering and noise reduction on the globally optimized monitoring image to obtain a filtered and noise-reduced optimized image; Perform visual detection of metastatic tumors at each time point on the filtered and noise-reduced optimized image and mark the positions of metastatic tumors in the image; Perform bounding box segmentation based on the positions of metastatic tumors in the image and extract the bounding boxes of metastatic tumors at multiple time points.

[0025] In this embodiment, after obtaining the patient's authorization, when performing a PET-CT scan, ensure that the equipment is calibrated and in good condition, select an appropriate radioactive tracer (such as [^18F]FDG), and calculate the appropriate dose according to the patient's weight. Usually, PET-CT scans need to be performed in multiple follow-ups, for example, once every three months, and image data at multiple time points are recorded to monitor the changes of adrenal metastases. A standardized scanning protocol is adopted to ensure consistent image quality for each scan. After each scan, professional software (such as the PACS system) is used to store and manage the image data to ensure its traceability. Record the scan parameters at each time point, such as the scan time (60 minutes), image resolution (512x512 pixels), etc., for subsequent processing and analysis. Before image processing, first read the PET-CT monitoring images at each time point to ensure that all images are processed under the same reference framework. Conduct a preliminary inspection of the images to confirm that there are no artifacts or obvious scanning errors. Check the signal-to-noise ratio (SNR) and contrast of the images to ensure that they are suitable for subsequent sharpening and equalization processing. Use image sharpening techniques (such as Laplacian filtering or high-pass filtering) to sharpen the images at each time point to enhance the boundaries of the adrenal gland and metastases in the images. This step helps to improve the accuracy of subsequent detection. By applying the Laplacian operator, calculate the second derivative of the image to enhance the edge information and record the changes in the sharpened images. Perform histogram equalization on the processed images to improve the contrast of the images, making the metastases more obvious. This step improves the visibility of the images by expanding the gray level range of the images. Use the CLAHE (Contrast Limited Adaptive Histogram Equalization) method to ensure equalization in local regions, thus avoiding over-enhancing noise. Synthesize the sharpened and equalized images to form a globally optimized monitoring image. This image will serve as the basis for subsequent analysis and detection. Record the optimized image parameters at each time point, such as the mean gray value, contrast enhancement factor, etc., for subsequent analysis and comparison. Select a suitable adaptive filtering method, such as Wiener filtering or adaptive median filtering, with the goal of reducing random noise in the images while maintaining important edge information. Determine the filtering parameters, such as the window size and filter type, for subsequent processing. Apply the adaptive filtering algorithm to the globally optimized monitoring images at each time point. For each pixel point, dynamically adjust the filtering intensity according to the statistical characteristics of its neighboring pixels. Using Wiener filtering, analyze the local regions of the images, calculate the local mean and variance, and adaptively adjust the filtering according to the noise level to ensure that the detailed parts of the images are retained. Record the image parameters after filtering, such as the peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM), to evaluate the filtering effect. Ensure that the recognizability of the adrenal gland and metastases is improved during the noise reduction process. Evaluate the change in the PSNR value before and after filtering. If the PSNR value increases from 20 to 30, it indicates a significant noise reduction effect.Collect filtered and noise-reduced optimized images at multiple time points, and ensure that the images are visually detected in the same coordinate system. Prepare detection tools and marking software for subsequent work on marking metastatic tumors. Develop a standardized visual detection process to ensure that each inspector has consistent definitions and markings for metastatic tumors. Conduct a visual inspection of each image one by one to identify the location and size of metastatic tumors. Use professional software tools for marking to ensure the accuracy of the markings. Record the coordinate positions (x, y) and diameters of each metastatic tumor to ensure that the characteristic information of the metastatic tumors is completely recorded. Mark the positions of metastatic tumors in the images at each time point to form an image library with markings for subsequent analysis and comparison. Ensure that each marking has corresponding image number and time point information. Record the marking information, including the location, size, shape, etc. of the metastatic tumors in the images, and generate a marking report for subsequent data analysis. Select a suitable bounding box segmentation algorithm, such as threshold-based segmentation, contour detection (such as Canny edge detection), or region growing method, with the goal of extracting bounding boxes from the marked metastatic tumors. Determine the segmentation parameters, such as the threshold range and minimum contour area, to ensure the accuracy and integrity of the bounding boxes. Apply the selected bounding box segmentation algorithm to the marked images at each time point to extract the boundary information of the metastatic tumors. Generate corresponding bounding boxes according to the output results of the algorithm. Through Canny edge detection, identify the edges of the metastatic tumors and generate bounding boxes using the minimum bounding rectangle method. Record the bounding box information of the metastatic tumors at each time point, including the coordinates, width, and height of the bounding boxes. Form a dataset for subsequent comparison and analysis.

[0026] In this embodiment, the specific steps for sharpening and enhancing the PET-CT monitoring image and constructing a globally optimized monitoring image are as follows: Perform Gaussian blur filtering on the PET-CT monitoring image to obtain a blurred filtered image; Calculate the pixel difference between the blurred filtered image and the PET-CT monitoring image to obtain a sharpening mask; Perform sharpening enhancement processing on the PET-CT monitoring image according to the sharpening mask to construct a sharpened enhanced monitoring image; Calculate the gray histogram of the sharpened enhanced monitoring image; Count the occurrence frequencies of each gray level in the gray histogram; Calculate the pixel cumulative frequency according to the occurrence frequencies of each gray level to obtain a cumulative distribution function; Perform normalization processing on the cumulative distribution function to obtain a gray level mapping range parameter; Perform global pixel equalization transformation on the sharpened enhanced monitoring image through the gray level mapping range parameter, thereby constructing a globally optimized monitoring image.

[0027] In this embodiment, Gaussian blur is a commonly used image processing technique aimed at reducing image noise and details. This process smooths the image by convolving the image with a Gaussian kernel. Determine the size and standard deviation (σ) of the Gaussian kernel. Usually, a 3x3 or 5x5 kernel is selected, and the σ value can be set to 1.0 or 1.5, with the specific selection depending on the noise level of the image. Apply the selected Gaussian kernel to each pixel point, calculate the weighted average of its surrounding pixels, and generate a blurred filtered image. If a 3x3 Gaussian kernel is used, when calculating the value of the central pixel, consider the weighted values of the 8 surrounding pixels, and the resulting image will appear smoother with some details reduced. Record the image parameters after blurred filtering, including the peak signal-to-noise ratio (PSNR) and the structural similarity index (SSIM), to evaluate the effect of the blur process. If the PSNR value of the image after blur processing is 30, it indicates that the noise reduction effect is significant and is suitable for subsequent sharpening and enhancement processing. Compare the blurred filtered image with the original PET-CT monitoring image pixel by pixel and calculate the difference between them. This process aims to identify the details hidden in the original image by the blur process. For each pixel, the difference formula is calculated as: D(x,y)=∣I(x,y)−F(x,y)∣, where I is the original image, F is the blurred image, and D is the difference image. Generate a sharpening mask based on the calculated difference. Usually, set the difference threshold to a certain critical value (such as 10) to filter out significant edge information. If the difference of a pixel is greater than 10, it is marked as 1 (significant edge) on the sharpening mask, otherwise it is marked as 0 (non-significant edge). Record the statistical characteristics of the sharpening mask, including the number and distribution of edge pixels, for reference in subsequent sharpening and enhancement processing. If the significant edge pixels in the sharpening mask account for 5% of the total pixels, it indicates that there is certain detail information in the image that needs to be enhanced. Select a suitable sharpening and enhancement method, usually using weighted average filtering or Laplacian filtering. Enhance the edges and details by combining the sharpening mask with the original image. Determine the parameters of the sharpening and enhancement, such as the enhancement factor (such as 1.5), to control the intensity of the sharpening and avoid artifacts caused by oversharpening. Apply the sharpening mask to the original PET-CT monitoring image. The specific steps are: apply the formula E(x,y)=I(x,y)+α⋅D(x,y) to each pixel, where E is the sharpened and enhanced image, and α is the enhancement factor. If the original value of a pixel is 100 and the difference of the sharpening mask is 15, then the sharpened value is 100 + 1.5×15 = 122.5, and this pixel becomes more prominent in the enhanced image.Perform a grayscale histogram calculation on the sharpened enhanced image and record the pixel occurrence frequencies at each grayscale level (usually from 0 to 255). This process helps to evaluate the brightness distribution of the image. The specific steps are as follows: Traverse each pixel of the sharpened enhanced image, count the number of occurrences of each grayscale level, and record it in the histogram. Visualize the calculated grayscale histogram to facilitate the analysis of the brightness and contrast information of the image. The generated histogram shows that most pixels are concentrated in the middle grayscale level, indicating the uniformity of the image brightness. Perform statistics on the generated grayscale histogram, record the occurrence frequency of each grayscale level, and form a frequency distribution table. This step provides the basic data for the subsequent calculation of the cumulative distribution function. If the occurrence frequency of grayscale level 255 is 50 times, it is recorded as part of the frequency data. Record the occurrence frequency of each grayscale level and analyze the characteristics of the frequency distribution, such as whether there are obvious deviations or peaks. If the frequency of grayscale level 128 is much higher than other grayscale levels, it indicates that there are more details in the image at this grayscale level. Calculate the cumulative frequency of each grayscale level according to the occurrence frequency of each grayscale level. Cumulative frequency:. , where f(j) is the frequency of grayscale level j. If the frequency of grayscale level 0 is 10 and the frequency of grayscale level 1 is 20, then the cumulative frequency of grayscale level 2 is 30. Normalize the cumulative distribution function and adjust its value range to between 0 and 1 for subsequent grayscale level mapping. Record the normalized grayscale level mapping parameters for subsequent image processing and conversion. According to the normalized grayscale level mapping range parameters, perform a global pixel equalization transformation on the sharpened enhanced monitoring image. The specific steps are as follows: Map each pixel value using the mapping formula M(x,y)=N(I(x,y)), where M is the pixel value after equalization transformation and I is the sharpened enhanced image. If a pixel value is 120, it may become 200 after mapping, ensuring that the brightness and contrast of the image are improved. Record the main parameters of the globally optimized monitoring image, such as contrast, brightness, and image quality, to evaluate the effect of the equalization transformation.

[0028] In this embodiment, refer to Figure 3 , which is a schematic diagram of the detailed implementation steps of step S2. In this embodiment, the detailed implementation steps of the said step S2 include: Perform a quantitative analysis of the change in blood vessel density of the surrounding tissues according to the metastatic tumor bounding box, and extract the characteristics of the change in blood vessel density; Identify the infiltration condition at the boundary of the metastatic tumor bounding box; Calculate the change in the metabolic level of the boundary tissues of the metastatic tumor bounding box to generate data on the change in metabolic level; Perform a dynamic fluctuation analysis of the surrounding tissue microenvironment based on the characteristics of the change in blood vessel density, the infiltration condition at the boundary, and the data on the change in metabolic level, so as to generate the law of change in the surrounding tissue microenvironment; Evaluate the degree of boundary immune influence according to the variation law of the surrounding tissue microenvironment to generate an evaluation value of the degree of boundary immune influence; Evolve the temporal immune interaction intensity of the metastatic tumor bounding box according to the evaluation value of the degree of boundary immune influence, and construct a multi-cycle immune response intensity map.

[0029] In this embodiment, a suitable method for calculating vascular density is selected. Usually, image processing techniques such as threshold segmentation and morphological operations are used to extract the vascular structure in the image. The objective of the analysis is to quantitatively analyze the vascular density in the tissues inside and outside the bounding box of the metastatic tumor. Determine the threshold method, such as using the Otsu algorithm, to automatically determine the optimal segmentation threshold and enhance the visualization of the vascular structure. Extract the vascular structure separately inside and outside the bounding box of the metastatic tumor, calculate the ratio of the total area of the blood vessels to the total area of the tissue in each region, and obtain the vascular density (expressed as a percentage of the vascular area). If the vascular area inside the bounding box of the metastatic tumor is 500 square millimeters and the total tissue area is 2000 square millimeters, the vascular density is 25%. Record the characteristics of the vascular density changes, including the vascular density values inside and outside the bounding box, and compare the vascular density changes at different time points. If the vascular densities of the tissues around the metastatic tumor are 25%, 30%, and 35% at different time points, it indicates that the vascular density has an upward trend over time. Select a suitable method for identifying the infiltration situation. Usually, image analysis techniques (such as edge detection and region growing) are used to judge the tissue infiltration situation at the boundary of the metastatic tumor. Confirm the degree of infiltration by observing the degree of edge blurring and changes in tissue structure. Determine the edge detection algorithm, such as using the Canny edge detection or Sobel operator, to enhance the boundary features. Analyze the edge of the bounding box of the metastatic tumor to identify whether there is obvious tissue infiltration. By comparing the gray level and texture features of normal tissue and infiltrated tissue, judge the severity of the infiltration. If the tissue gray level at the edge is significantly reduced and the texture becomes blurred, it can be judged that infiltration exists. Record the evaluation results of the infiltration situation, including the scope, depth of infiltration and its impact on the surrounding tissues, for subsequent analysis. If the infiltration depth at the boundary of the metastatic tumor is 2 millimeters and it affects the surrounding normal tissues, special attention should be paid to the pathological changes in this area. Select a suitable method for calculating the metabolic level. Usually, image processing and analysis techniques are used, combined with the metabolic information of PET-CT images, to evaluate the metabolic activity of the metastatic tumor and its surrounding tissues. Determine the evaluation index of the metabolic level, such as the standardized uptake value (SUV), as a quantitative analysis of the metabolic activity. Calculate the metabolic level separately inside and outside the bounding box of the metastatic tumor, and record the SUV value of each region. This process includes extracting the distribution information of the radioactive tracer in the image and calculating its average value in the metastatic tumor and the surrounding tissues. If the SUV value in the metastatic tumor is 5.0 and the SUV value in the surrounding tissues is 2.0, it indicates that the metabolic activity of the metastatic tumor is significantly higher than that of normal tissues. Record the data of the metabolic level changes, analyze the metabolic activity changes at different time points, and identify the biological behavior of the metastatic tumor. If the SUV value of the metastatic tumor gradually increases during the follow-up, it may indicate that the tumor is progressing. Combine the data of vascular density, infiltration situation and metabolic level, and select a suitable statistical analysis method, such as regression analysis and variance analysis, to explore the relationships between these factors.Determine the key parameters for analysis, such as correlation coefficients and P-values, and evaluate the impact of different factors on the surrounding tissue microenvironment. Comprehensively analyze the changes in vascular density, infiltration, and metabolic levels to identify the dynamic fluctuation patterns of the surrounding tissue microenvironment. By comparing the data at different time points, evaluate the changing trend of the microenvironment. If it is found that the increase in vascular density is positively correlated with the increase in metabolic level, it indicates that there may be a connection between angiogenesis and tumor metabolism. Record the changing patterns of the surrounding tissue microenvironment, including the change amplitude, trend, and its clinical significance, and generate charts for subsequent analysis. Generate a curve graph of the microenvironment feature changes to show the changes in vascular density, infiltration degree, and metabolic level at different time points. Select a suitable method for evaluating the immune impact, usually by combining clinical data, histological examinations, and image analysis, to evaluate the infiltration and distribution of immune cells in the boundary area of metastatic tumors. Determine the evaluation indicators, such as the density and distribution characteristics of tumor-infiltrating lymphocytes (TILs). According to the changing patterns of the microenvironment, evaluate the immune impact degree at the boundary of metastatic tumors. By observing the distribution and activity of immune cells, judge their inhibitory or promoting effects on tumor growth. If a large number of TILs are observed at the boundary of metastatic tumors, it indicates that the immune system has a certain response to the tumor. Record the immune impact evaluation values, including the density, distribution of immune cells, and their correlation with tumor growth, to support subsequent analysis. If the evaluation values show a high density of infiltrating lymphocytes, it may indicate the immune escape mechanism of the tumor. Select a suitable method for evaluating the immune interaction intensity, usually using time series analysis techniques, to evaluate the interaction between immune cells and tumor cells. Determine the evaluation parameters, such as the immune cell activity index, cytokine levels, etc., to quantify the immune response. Conduct an evolutionary analysis of the temporal immune interaction intensity for the metastatic tumor boundary box, and analyze the intensity changes of the immune response based on the immune evaluation values at different time points. By plotting the curve of the immune response intensity over time, observe the relationship between immune cell activity and tumor growth. Record the changing data of the multi-cycle immune response intensity and generate a multi-cycle immune response intensity graph to facilitate the visual display of the dynamic evolution of the immune response. The generated intensity graph shows the immune response changes at different time points, providing a basis for further clinical decision-making.

[0030] In this embodiment, refer to Figure 4 , which is a schematic diagram of the detailed implementation steps of step S3. In this embodiment, the detailed implementation steps of step S3 include: Conduct a three-dimensional morphological analysis of the metastatic tumor for each boundary box of the metastatic tumor boundary box to generate the three-dimensional morphological characteristics of the metastatic tumor; Identify the morphological change characteristics of the three-dimensional morphological characteristics of the metastatic tumor at multiple time points; Calculate the density of the metastatic tumor at multiple time points based on the metastatic tumor boundary box to generate the density of the metastatic tumor at multiple time points; Mine the spatial density distribution of the density of metastatic tumors at multiple time points to generate the spatial density distribution characteristics of metastatic tumors; Evolve the change trend of metastatic tumors based on the morphological change characteristics and the spatial density distribution characteristics of metastatic tumors to obtain the image change trend diagram of metastatic tumors at different time points.

[0031] In this embodiment, a suitable three-dimensional morphological analysis tool is selected, usually a medical image processing software (such as 3DSlicer or ITK-SNAP), to ensure that three-dimensional reconstruction can be performed on the image data within the bounding box of the metastatic tumor. Determine the analysis objectives and parameters, such as the volume, surface area, and shape characteristics (such as the ratio of the long axis to the short axis) of the metastatic tumor. For each metastatic tumor bounding box, extract the three-dimensional data therein and perform image segmentation and reconstruction. Utilize threshold segmentation and region growing techniques to ensure accurate extraction of the three-dimensional morphology of the metastatic tumor. If the size of the bounding box of a metastatic tumor is 30mm x 20mm x 15mm, its three-dimensional model can be obtained through reconstruction, and its volume and surface area are recorded. Record the three-dimensional morphological feature data of each metastatic tumor, including volume, surface area, shape index, etc., and conduct comparative analysis to identify morphological changes at different time points. If the volume of a metastatic tumor changes to 5.0 cm³, 6.5 cm³, and 8.0 cm³ at different time points, it indicates that the tumor is growing. Select a suitable morphological change analysis method, usually combining statistical analysis and visualization techniques, to compare the three-dimensional morphological characteristics at different time points and identify their change trends. Determine key parameters, such as the growth rate, morphological change rate, etc., to quantify the morphological changes. Compare the three-dimensional characteristics of the metastatic tumor at each time point and calculate the volume change rate and shape change rate. Growth rate = (Vt−V0) / V0×100% where Vt is the current volume and V0 is the initial volume. If the metastatic tumor increases from 5.0 cm³ to 6.5 cm³, the growth rate is 30%, and record this change for subsequent analysis. For each metastatic tumor bounding box at each time point, calculate the volume of the metastatic tumor and the volume of the surrounding tissue, and then calculate the density of the metastatic tumor. If the volume of a metastatic tumor is 6.5 cm³ and the volume of the surrounding tissue is 100 cm³, the density of the metastatic tumor is 0.065 cm³ / cm³. Record the densities of the metastatic tumor at multiple time points and compare the changes at different time points to identify the density change trend of the metastatic tumor. If the densities of the metastatic tumor are 0.065, 0.070, and 0.075 at different time points, it indicates that the density of the metastatic tumor is gradually increasing. Select a suitable spatial density analysis method, usually using heat map generation or spatial interpolation methods (such as Kriging interpolation), to visualize the spatial distribution of the metastatic tumor density. Determine the spatial range and resolution of the analysis to ensure that the generated density distribution map has high precision. Perform spatial interpolation on the metastatic tumor density data at multiple time points to generate the corresponding spatial density distribution map. Evaluate the metastatic tumor density in different regions through Kriging interpolation. The generated heat map shows the distribution of tumor density in different parts, facilitating the observation of high-density regions. Record the generated spatial density distribution feature data, including the central position, distribution range, and high-density regions of the density distribution, for subsequent analysis. If the density of the metastatic tumor in a certain region is significantly higher than that in other regions, special attention should be paid to the pathological changes in this region.Combined with morphological change characteristics and spatial density distribution characteristics, select an appropriate trend analysis method, such as linear regression analysis or non-linear fitting, to evaluate the change trend of metastatic tumors. Determine the key parameters for analysis, such as the change rate, trend direction, etc. Use statistical analysis methods to comprehensively analyze the morphological change and density distribution data at different time points to identify the change trend of metastatic tumors. If the analysis results show that both the volume and density of metastatic tumors are on the rise, it indicates that the tumor is progressing.

[0032] In this embodiment, step S4 includes the following steps: Perform multi-scale decomposition processing on the metastatic tumor bounding box to obtain metastatic tumor sub-images with different frequency components; Traverse the pixel gray levels of the metastatic tumor sub-images with different frequency components to extract the gray level co-occurrence matrix for each frequency; Perform texture analysis for each frequency based on the metastatic tumor sub-images with different frequency components to generate the metastatic tumor texture features for each frequency; Perform deep convolutional learning on the gray level co-occurrence matrix for each frequency and the metastatic tumor texture features for each frequency to extract the deep metastatic tumor texture features; Based on the deep metastatic tumor texture features, identify the texture feature differences of the surrounding tissues of the metastatic tumor bounding box, and extract the texture difference features between the metastatic tumor and the surrounding normal tissues; Based on the texture difference features, perform micro-detail recognition of the surrounding components to extract potential tumor metastatic cells.

[0033] In this embodiment, a suitable multi-scale decomposition technique is selected. Usually, the wavelet transform or the Laplacian pyramid method is used to decompose the image in the metastatic tumor bounding box into different frequency components. This process helps to extract image features at different scales. Determine the wavelet basis function (such as the Haar wavelet or the Daubechies wavelet), and set the number of decomposition levels, usually choosing 3 to 5 levels to ensure that sufficient detail and contour information are captured. Perform wavelet transform on the original image within the metastatic tumor bounding box to generate sub-images of multiple frequency components. The first-level wavelet decomposition will produce low-frequency and high-frequency sub-images, representing the overall structure and details of the image respectively. If the size of the image in the metastatic tumor bounding box is 256x256 pixels, through 3-level wavelet decomposition, a low-frequency component and multiple high-frequency components (such as horizontal, vertical, and diagonal high-frequency components) are obtained. Record the image features of each frequency component, including resolution, composition, and information content, for subsequent texture analysis. Analyzing the image of the low-frequency component shows the overall structure, while the high-frequency components show the detailed features, providing a basis for subsequent texture analysis. Select a suitable method for calculating the gray-level co-occurrence matrix. Usually, calculate its gray-level co-occurrence matrix according to the different frequency components of the image. The GLCM is used to describe the spatial relationship between pixel gray levels in the image. Determine the direction and distance parameters. Commonly used directions include 0° (horizontal), 90° (vertical), and 45° (diagonal), and the distance parameter is usually set to 1. Traverse the pixel gray levels of the metastatic tumor sub-images of each frequency component to construct the corresponding GLCM. For each pixel, view the gray values of its neighboring pixels and update the co-occurrence matrix. If, in the horizontal direction, the gray value of a pixel is 100 and the gray value of its neighboring pixel is 120, then update the corresponding value in the GLCM. Record the GLCM features of each frequency component, such as the symmetry, energy, entropy, etc. of the matrix, for subsequent texture feature extraction. Calculate the energy and entropy values of the GLCM. If the entropy value of the high-frequency component is 1.2 while that of the low-frequency component is 0.5, it indicates that the detailed information is more abundant in the high-frequency component. Select a suitable texture feature extraction method. Usually, use the statistical features (such as contrast, correlation, energy, and entropy) of the gray-level co-occurrence matrix to characterize the texture features of different frequency components. Determine the calculation formula for texture features to facilitate extracting effective information from the GLCM. Perform statistical analysis on the GLCM of each frequency component to extract the texture features of each frequency. Calculate the contrast and entropy values of the GLCM to describe the texture complexity of the image. If the contrast of the high-frequency image is 0.8 and that of the low-frequency image is 0.3, it indicates that the texture of the high-frequency image is more complex. Record the texture feature data of each frequency to provide a basis for subsequent deep convolutional learning. Summarize the statistical features of each frequency to form a feature table for subsequent analysis and modeling.Select a suitable Convolutional Neural Network (CNN) model. Usually, pre-trained models (such as VGG16, ResNet) can be used for transfer learning to facilitate the extraction of deep texture features of metastatic tumors. Determine the shape of the input layer and the parameters of the convolutional layer to adapt to the features of different frequency components. Use the GLCM and texture features of each frequency component as input and train through a deep convolutional network to extract deep texture features. Input the feature maps of high-frequency and low-frequency components, and after multiple layers of convolution and pooling, finally obtain the deep features of each frequency. Record the extracted deep texture features and evaluate the effect of convolutional learning, for example, by classification accuracy or feature reconstruction error. Select a suitable method for difference identification. Usually, statistical tests (such as t-tests or ANOVA) or machine learning classifiers (such as Support Vector Machine SVM) are used to evaluate the texture feature differences between metastatic tumors and surrounding normal tissues. Determine the key features for identification, such as the deep texture features of metastatic tumors and the eigenvalue of surrounding tissues. Compare the deep texture features of metastatic tumors and surrounding normal tissues to identify significantly different features. Through statistical analysis, judge which features are significantly higher or lower in tumor tissues than in normal tissues. If the texture feature entropy value of a metastatic tumor is 1.5 while that of normal tissue is 0.8, it indicates a significant difference in texture complexity between the two. Record the identification results of texture difference features, including the significance level and the degree of difference, for subsequent analysis. If the p-value is less than 0.05, it indicates a significant difference between metastatic tumors and normal tissues, and further analysis of its clinical significance is required. Select a suitable method for micro-detail identification. Usually, image segmentation techniques such as threshold segmentation, region growing, or machine learning segmentation algorithms are used to identify the fine structures in the tissues surrounding metastatic tumors. Determine the parameters and thresholds of the segmentation algorithm to ensure accurate identification of fine structures. For the identified texture difference features, extract the fine structures from the tissues surrounding metastatic tumors to identify potential tumor metastatic cells. If cell clusters smaller than 5 μm are found at the boundary of a metastatic tumor, they can be determined as potential metastatic cells. Record the results of fine structure identification, including the number, location, and distribution characteristics of potential tumor metastatic cells, to support subsequent clinical decisions. If 10 potential metastatic cells are identified at the boundary of a metastatic tumor, further pathological analysis and evaluation are required.

[0034] In this embodiment, the bounding box of the metastatic tumor is processed by multi-scale decomposition to obtain sub-images of the metastatic tumor with different frequency components; Traverse the pixel gray levels of the sub-images of the metastatic tumor with different frequency components to extract the gray-level co-occurrence matrix of each frequency; Perform texture analysis for each frequency based on the sub-images of the metastatic tumor with different frequency components to generate the texture features of the metastatic tumor for each frequency; Perform deep convolutional learning on the gray-level co-occurrence matrix of each frequency and the texture features of the metastatic tumor for each frequency to extract deep metastatic tumor texture features; Identify the texture feature differences of the surrounding tissues of the metastatic tumor boundary box according to the texture features of the deep metastatic tumor, and extract the texture difference features between the metastatic tumor and the surrounding normal tissues; Based on the texture difference features, perform micro-detail recognition of the surrounding components to extract potential tumor metastatic cells.

[0035] In this embodiment, the gray-level co-occurrence matrix (GLCM) is a tool for describing the spatial relationship between pixel grayscales in an image. By calculating pairs of gray values of adjacent pixels, the texture information of the image can be obtained. The features of the GLCM are used to quantify the texture features of the image, such as contrast, correlation, energy, and entropy. Select metastatic tumor sub-images of different frequency components. By traversing each pixel, record its gray value and the gray value of adjacent pixels to construct the corresponding GLCM. First, determine the direction (such as horizontal, vertical, and diagonal) and distance (usually set to 1 pixel) for calculating the GLCM. For a high-frequency component sub-image, traverse each pixel and check the gray value of its adjacent pixels. If the gray value of the current pixel is 100 and the gray value of the adjacent pixel is 120, then increment the count at the (100, 120) position in the GLCM. In this way, the GLCM of each frequency component is generated. Record the GLCM of each frequency component and calculate its statistical features, such as symmetry, energy, and entropy. These data will provide the necessary information for subsequent texture feature analysis. If the GLCM energy value of the high-frequency component is 0.75 and that of the low-frequency component is 0.30, it indicates that the texture features of the high-frequency component are more abundant, suggesting that there may be a more complex tissue structure. Use the statistical features of the GLCM (such as contrast, correlation, energy, entropy, etc.) to characterize the texture features of different frequency components. These features effectively describe the texture complexity and consistency of the image. Calculate the GLCM of each frequency component and extract its texture features. The calculation formulas usually include: Contrast: Measures the sum of squares of gray differences. Correlation: Describes the linear relationship between gray values. Energy: The sum of squares of all elements in the GLCM, reflecting the uniformity of the image. Entropy: Reflects the complexity of image information. For the GLCM of the high-frequency component, the calculated contrast is 0.85 and the correlation is 0.6, indicating a relatively high texture complexity of this component. Select a suitable convolutional neural network (CNN) model. Usually, use a pre-trained model (such as VGG16, ResNet) for transfer learning to extract deep texture features from different frequency components. Take the GLCM and texture features of each frequency component as inputs, and train them through the convolutional neural network to extract deep texture features. Through multi-layer convolution and pooling operations, gradually abstract higher-level texture feature representations. For the feature maps of high-frequency and low-frequency components, input them into the CNN. Through the connections between layers, finally obtain the deep features of each frequency. Adopt statistical tests (such as t-tests, ANOVA) or machine learning classifiers (such as support vector machines, SVMs) to evaluate the texture feature differences between metastatic tumors and surrounding normal tissues. Compare the deep texture features of metastatic tumors and surrounding normal tissues to identify significantly different features. Use statistical analysis to determine which features are significantly higher or lower in tumor tissues than in normal tissues. If the entropy value of the deep texture feature of the metastatic tumor is 1.8, while that of the normal tissue is 0.7, it indicates a significant difference in texture complexity between the two.Record the recognition results of texture difference features, including the significance level and the degree of difference, for subsequent analysis. If the p-value is less than 0.05, it indicates that there is a significant difference between the metastatic tumor and the normal tissue, and further analysis of its clinical significance is required. Use image segmentation techniques, such as threshold segmentation, region growing, or machine learning segmentation algorithms, to identify the fine structures in the tissue surrounding the metastatic tumor. More accurate segmentation can be achieved by combining texture difference features. Apply the selected segmentation algorithm to the identified texture difference features to extract the fine structures in the tissue surrounding the metastatic tumor and identify potential tumor metastatic cells. Use the region growing algorithm to segment according to the texture difference features and identify cell clusters smaller than 5 μm, and determine them as potential metastatic cells. Record the results of the fine structure recognition, including the number, location, and distribution characteristics of potential tumor metastatic cells, to support subsequent clinical decisions. If 10 potential metastatic cells are identified at the boundary of the metastatic tumor, further pathological analysis and evaluation are required.

[0036] In this embodiment, the specific steps of step S6 are as follows: Distinguish the metastatic hot spots based on the multi-cycle immune response intensity map and potential tumor metastatic cells, and mark multiple metastatic hot spots; Perform a rolling prediction of the metastatic tumor situation on multiple metastatic hot spots and the temporal metastasis trajectory map to obtain the predicted situation of the metastatic tumor in multiple time windows; Analyze the metastasis time point and the diffusion trend based on the predicted situation of the metastatic tumor in multiple time windows, and construct a predicted time axis for the metastatic tumor.

[0037] In this embodiment, the metastatic hot spot region refers to the region with a relatively high risk of tumor metastasis, which is usually closely related to the intensity of the immune response in the body and the distribution of metastatic cells. At this stage, the goal is to identify and label these regions for subsequent analysis and prediction. Collect multi-cycle immune response intensity maps and distribution data of potential tumor metastatic cells. The immune response intensity maps are usually obtained through PET scans or immunohistochemistry methods, while the distribution of metastatic cells is extracted through previous image analysis techniques. Using image processing techniques, such as threshold segmentation and region growing algorithms, by analyzing the immune response intensity maps, regions with intensities higher than a set threshold (e.g., 0.6) are identified. Combining the positions of potential tumor metastatic cells, overlapping analysis is performed to screen out the metastatic hot spot regions. Setting the intensity threshold to 0.6, 3 hot spot regions are identified, namely A, B, and C. For each region, its coordinates, area, and immune response intensity are recorded. The labeled metastatic hot spot regions and their characteristic data are recorded in the database, providing a basis for subsequent prediction of the metastatic tumor situation. If the immune response intensity of region A is 0.75 and the area is 20 mm², then this information will be used for subsequent analysis to help judge its metastasis risk. Collect metastatic tumor imaging data at multiple time points to construct a metastatic trajectory map. This image shows the changes of the tumor over the time axis, and can clearly reflect the growth and diffusion trends of the tumor. Using image registration techniques to align the images at different time points to ensure comparison in the same coordinate system. Select a suitable time series prediction model, such as ARIMA (Autoregressive Integrated Moving Average Model) or LSTM (Long Short-Term Memory Network), for rolling prediction of the changes in the metastatic hot spot regions. These models need to be trained by combining historical data. Determine the parameters of the model, such as the size of the time window (e.g., 3 months) and the prediction step size (e.g., 1 month), for effective prediction. Use the trained model to perform rolling prediction on the identified metastatic hot spot regions, and record the situation of the metastatic tumor in each time window. If the predicted results of a certain hot spot region show volume changes of 10%, 20%, and 25% respectively in the next 3 months, then record these changes. The predicted volume changes of region A are 10 cm³, 12 cm³, and 15 cm³, showing the growth trend of the tumor. Record the predicted situations of the metastatic tumor in multiple time windows, including volume, diffusion range, and potential metastatic paths. This information will provide a basis for subsequent analysis of the metastatic time point and diffusion trend. If in the first time window, the volume of the metastatic tumor increases from 10 cm³ to 12 cm³, it indicates that the tumor shows a growth trend during this period. Based on the prediction results of multiple time windows, potential metastatic time points are identified. These time points are usually the nodes where the tumor volume or diffusion speed changes significantly. Set a volume change threshold (e.g., 20%) to identify key metastatic time points. If in a certain time window, the tumor volume increases from 10 cm³ to 15 cm³ (a 50% increase), then mark this time point as an important metastatic node.Analyze the identified metastasis time points to evaluate the spread trend of the tumor. Adopt trend analysis methods, such as linear regression or polynomial fitting, to analyze the rate and direction of the change in tumor volume. Through trend analysis, if it is found that the tumor spreads significantly faster in a certain direction than in other directions, record this direction as the main spread path. According to the metastasis time points and spread trend, construct a predicted time axis for metastatic tumors, marking the key time nodes and spread paths. If the predicted metastasis time points are T1, T2, and T3, and the spread paths are A, B, then clearly mark this information on the time axis. Record the constructed predicted time axis for metastatic tumors, including time points, volume changes, spread paths, and their clinical significance. Ensure the integrity and traceability of the data.

[0038] In this embodiment, a prediction and analysis system for adrenal metastatic tumors based on image recognition is provided, which is used to execute the prediction and analysis method for adrenal metastatic tumors based on image recognition as described above, including: An image processing module, which is used to collect the adrenal PET-CT monitoring images of the patient at multiple time points, and perform visual detection of metastatic tumors at each time point one by one, and extract the bounding boxes of metastatic tumors at multiple time points; An immune response evaluation module, which is used to analyze the dynamic fluctuations of the surrounding tissue microenvironment and evaluate the degree of immune influence on the boundary according to the bounding boxes of metastatic tumors, and construct a multi-cycle immune response intensity map; An imaging change module, which is used to evolve the change trend of metastatic tumors for the bounding boxes of metastatic tumors to obtain the imaging change trend map of metastatic tumors at different time points; A difference recognition module, which is used to identify the differences in the texture features of the surrounding tissues according to the bounding boxes of metastatic tumors, so as to extract potential tumor metastatic cells; A metastasis trajectory tracking module, which is used to perform dynamic metastasis trajectory tracking according to the adrenal PET-CT monitoring images and the imaging change trend map of metastatic tumors, and perform full-process metastasis trajectory fitting to construct a time-series metastasis trajectory map; A situation prediction module, which is used to perform rolling prediction of the metastasis situation of metastatic tumors on the time-series metastasis trajectory map according to the multi-cycle immune response intensity map and potential tumor metastatic cells, and construct a predicted time axis for metastatic tumors.

[0039] The present invention provides comprehensive dynamic data for subsequent analysis by collecting adrenal PET-CT images at multiple time points, enabling effective tracking of tumor changes. It accurately identifies metastatic tumors and their boundaries through image visual detection, avoiding subjective differences in manual judgment and improving detection efficiency and accuracy. The bounding boxes of metastatic tumors at multiple time points extracted provide core data support for subsequent analysis and are the basis for the analysis of other modules. By analyzing the immune response around the tumor, the changing trend of the tumor microenvironment can be evaluated, and then the sensitivity of the tumor to the immune response can be inferred. Through the image change trend graph, the changes of metastatic tumors at different time points can be tracked in real time, revealing the trends of tumor growth, shrinkage, or stability. The change trend graph can help identify the development trend of metastatic tumors, provide early warnings, and take intervention measures in advance. Texture feature difference recognition can reveal the subtle differences between tumor metastatic cells and surrounding normal tissues, improving the detection accuracy of metastatic cells. Through texture difference analysis, potential diffusion regions of metastatic tumors can be identified at an early stage, enhancing the early diagnosis ability of metastatic tumors. It can accurately track the path and speed of tumor metastasis, provide a clearer tumor development pattern, and help evaluate the risk and severity of metastasis. Through the fitting of the metastasis trajectory graph, from the initial metastasis to possible diffusion regions. By analyzing the immune response and tumor metastatic cells and combining with the metastasis trajectory graph, the future trend prediction of tumor metastasis can be achieved.

[0040] Therefore, from any perspective, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Thus, all changes falling within the meaning and scope of the equivalent elements of the application document are intended to be encompassed within the present invention.

[0041] As described above, these are only specific embodiments of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can 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 these embodiments shown herein, but rather will conform to the broadest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting and analyzing adrenal metastases based on image recognition, characterized in that, It includes the following steps: Step S1: Collect the adrenal PET-CT monitoring images of the patient at multiple time points, perform visual detection of metastatic tumors at each time point one by one, and extract the bounding boxes of metastatic tumors at multiple time points; Step S2: Perform dynamic fluctuation analysis of the surrounding tissue microenvironment and evaluate the degree of immune influence on the boundary according to the bounding boxes of metastatic tumors, and construct a multi-cycle immune response intensity map; Step S3: Evolve the change trend of metastatic tumors for the bounding boxes of metastatic tumors to obtain the image change trend map of metastatic tumors at different time points; Step S4: Identify the differences in texture features of the surrounding tissues according to the bounding boxes of metastatic tumors, so as to extract potential tumor metastatic cells; Step S5: Perform dynamic metastasis trajectory tracking according to the adrenal PET-CT monitoring images and the image change trend map of metastatic tumors, and perform full-process metastasis trajectory fitting to construct a time-series metastasis trajectory map; Step S6: Perform rolling prediction of the metastatic tumor situation on the time-series metastasis trajectory map according to the multi-cycle immune response intensity map and potential tumor metastatic cells, and construct a metastatic tumor prediction timeline.

2. The adrenal metastasis prediction and analysis method based on image recognition according to claim 1, wherein, The specific steps of Step S1 are as follows: Collect the adrenal PET-CT monitoring images of the patient at multiple time points; Perform sharpening enhancement and histogram equalization on the PET-CT monitoring images to construct a globally optimized monitoring image; Perform adaptive filtering and noise reduction on the globally optimized monitoring image to obtain a filtered and noise-reduced optimized image; Perform visual detection of metastatic tumors at each time point on the filtered and noise-reduced optimized image, and mark the positions of metastatic tumors in the image; Perform bounding box segmentation according to the positions of metastatic tumors in the image to extract the bounding boxes of metastatic tumors at multiple time points.

3. The method for predicting and analyzing adrenal metastases based on image recognition according to claim 2, wherein The specific steps of performing sharpening enhancement and histogram equalization on the PET-CT monitoring images to construct a globally optimized monitoring image are as follows: Perform Gaussian blur filtering on the PET-CT monitoring images to obtain a blurred filtering image; Calculate the pixel difference between the blurred filtering image and the PET-CT monitoring images to obtain a sharpening mask; Perform sharpening enhancement processing on the PET-CT monitoring images according to the sharpening mask to construct a sharpened enhancement monitoring image; Calculate the gray histogram of the sharpened enhancement monitoring image; Count the occurrence frequencies of each gray level in the gray histogram; Calculate the pixel cumulative frequency according to the occurrence frequencies of each gray level to obtain a cumulative distribution function; Perform normalization processing on the cumulative distribution function to obtain a gray level mapping range parameter; Perform global pixel equalization conversion on the sharpened enhancement monitoring image through the gray level mapping range parameter, so as to construct a globally optimized monitoring image.

4. The adrenal metastasis prediction and analysis method based on image recognition according to claim 1, wherein The specific steps of Step S2 are as follows: Perform quantitative analysis of the change in blood vessel density of the surrounding tissues according to the bounding boxes of metastatic tumors, and extract the blood vessel density change features; Identify the infiltration situation at the boundary of the bounding boxes of metastatic tumors; Calculate the change in the metabolic level of the boundary tissues for the bounding boxes of metastatic tumors to generate metabolic level change data; Perform dynamic fluctuation analysis of the surrounding tissue microenvironment according to the blood vessel density change features, the infiltration situation at the boundary, and the metabolic level change data, so as to generate the change law of the surrounding tissue microenvironment; Evaluate the degree of boundary immune influence according to the changing law of the surrounding tissue microenvironment to generate an evaluation value of the degree of boundary immune influence; Evolve the temporal immune interaction intensity of the metastatic tumor bounding box according to the evaluation value of the degree of boundary immune influence, and construct a multi-cycle immune response intensity map.

5. The method for predicting and analyzing adrenal metastases based on image recognition according to claim 1, wherein The specific steps of step S3 are as follows: Perform three-dimensional morphological analysis of the metastatic tumor bounding box one by one to generate three-dimensional morphological features of the metastatic tumor; Identify the morphological change features of the three-dimensional morphological features of the metastatic tumor at multiple time points; Calculate the metastatic tumor density at multiple time points according to the metastatic tumor bounding box to generate the metastatic tumor density at multiple time points; Mine the spatial density distribution of the metastatic tumor density at multiple time points to generate the spatial density distribution features of the metastatic tumor; Evolve the change trend of the metastatic tumor based on the morphological change features and the spatial density distribution features of the metastatic tumor to obtain the metastatic tumor image change trend map at different time points.

6. The method for predicting and analyzing adrenal metastases based on image recognition according to claim 1, wherein The specific steps of step S4 are as follows: Perform multi-scale decomposition processing on the metastatic tumor bounding box to obtain metastatic tumor sub-images with different frequency components; Traverse the pixel gray levels of the metastatic tumor sub-images with different frequency components, and extract the gray level co-occurrence matrix of each frequency; Perform texture analysis on the metastatic tumor sub-images with different frequency components one by one to generate the metastatic tumor texture features of each frequency; Perform deep convolutional learning on the gray level co-occurrence matrix of each frequency and the metastatic tumor texture features of each frequency to extract the deep metastatic tumor texture features; Identify the texture feature differences between the metastatic tumor and the surrounding normal tissues based on the deep metastatic tumor texture features of the metastatic tumor bounding box, and extract the texture difference features between the metastatic tumor and the surrounding normal tissues; Perform micro-detail recognition of the surrounding components based on the texture difference features to extract potential tumor metastatic cells.

7. The adrenal metastasis prediction and analysis method based on image recognition according to claim 1, characterized in that, The specific steps of step S5 are as follows: Mark the metastatic tumor position points at different time points according to the adrenal PET-CT monitoring image; Track the dynamic metastasis trajectory based on the metastatic tumor image change trend map and the metastatic tumor position points at different time points, and extract the dynamic metastasis trajectory; Calculate the metastasis speed and metastasis direction of the dynamic metastasis trajectory; Perform full-process metastasis trajectory fitting according to the metastasis speed and metastasis direction to construct a temporal metastasis trajectory map.

8. The method for predicting and analyzing adrenal metastases based on image recognition according to claim 1, wherein The specific steps of step S6 are as follows: Distinguish the metastatic hot spots according to the multi-cycle immune response intensity map and potential tumor metastatic cells, and mark multiple metastatic hot spots; Perform rolling prediction of the metastatic tumor situation on multiple metastatic hot spots and the temporal metastasis trajectory map to obtain the metastatic tumor prediction situation of multiple time windows; Analyze the metastasis time point and diffusion trend according to the metastatic tumor prediction situation of multiple time windows, and construct a metastatic tumor prediction time axis.

9. An adrenal metastasis prediction and analysis system based on image recognition, characterized in that, For executing the adrenal metastasis tumor prediction and analysis method based on image recognition as described in claim 1, including: An image processing module, configured to collect adrenal PET-CT monitoring images of a patient at multiple time points, perform visual detection of metastatic tumors at each time point one by one, and extract metastatic tumor bounding boxes at multiple time points; An immune response assessment module, which is used to analyze the dynamic fluctuations of the surrounding tissue microenvironment and evaluate the degree of immune influence on the boundary according to the metastatic tumor bounding box, and construct a multi-cycle immune response intensity map; An imaging change module, which is used to evolve the change trend of the metastatic tumor for the metastatic tumor bounding box to obtain an imaging change trend map of the metastatic tumor at different time points; A difference recognition module, which is used to recognize the differences in the texture features of the surrounding tissues according to the metastatic tumor bounding box, so as to extract potential tumor metastatic cells; A metastasis trajectory tracking module, which is used to perform dynamic metastasis trajectory tracking according to the adrenal PET-CT monitoring image and the imaging change trend map of the metastatic tumor, and perform full-process metastasis trajectory fitting to construct a time-series metastasis trajectory map; A situation prediction module, which is used to perform rolling prediction of the metastatic tumor situation on the time-series metastasis trajectory map according to the multi-cycle immune response intensity map and potential tumor metastatic cells, and construct a metastatic tumor prediction timeline.

Images