System and method for predicting radiation pneumonitis after radiotherapy of chest tumor

Abstract

The invention relates to the technical field of medical image analysis, in particular to a radiation pneumonitis prediction system and method after chest tumor radiotherapy, and the system comprises a structure hot area extraction module, a density breakpoint recognition module, a disturbance trajectory analysis module, a path clustering matching module and a risk trend output module. According to the method, through continuous arrangement of voxel gray scale ratios on a time axis and coincidence extraction of the voxel gray scale ratios and a dose region, precise positioning of a functional latent active region and sensitive identification of an evolution trend are realized, time sequence variation analysis is enhanced, and in combination with gray scale direction reversal point extraction, quantitative definition of a texture mutation region is improved; unifying ventilation and perfusion path time axes, dynamically mastering the continuity and mutability of functional disturbance of the pulmonary segment, completing behavior clustering based on trajectory consistency and inflection point alignment rate, enhancing the classification and evaluation ability of evolution trend, extracting a disturbance trend synchronization interval, and enhancing prospective identification of key early warning events. And the risk expression has higher hierarchy and timeliness.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image analysis, and particularly to a system and method for predicting radiation pneumonitis after radiotherapy for chest tumors. Background Art

[0002] The technical field of medical image analysis includes multiple links such as the acquisition, processing, recognition, and analysis of medical images. The core content of this technical field is to obtain the structural and change information of internal tissues or organs of the human body through images, and further extract and analyze features such as their morphology, texture, and density distribution. This technical field covers steps such as image preprocessing, image segmentation, feature extraction, and pattern recognition. Its overall goal is to provide information support for related applications through in-depth understanding and quantitative description of image data. Common data sources include CT images, MRI images, and X-ray images, etc. Its processing methods combine methods such as image enhancement, filtering, edge detection, and structural analysis to gradually achieve the structural recognition and description of the target area.

[0003] Among them, the system for predicting radiation pneumonitis after radiotherapy for chest tumors refers to, for the abnormal changes in the lungs that occur after radiotherapy, based on chest CT images, data modeling and analysis are carried out to predict the risk of radiation pneumonitis occurring in specific parts. By obtaining chest CT image data at different time points before and after radiotherapy, an image comparison structure for analyzing the density and texture changes in the lung area is constructed, and means such as image gray histogram, change trend of lung texture distribution, and regional statistical features are used to identify the abnormal evolution process of lung tissue. Through image registration at a set fixed time interval, the lung area is aligned, quantitative features are extracted in the image space in combination with the lung segment division rules, and a risk level division model is constructed according to the feature distribution pattern, so as to realize the conversion process from image data to prediction conclusions.

[0004] In the prediction of radiation pneumonia, existing technologies rely on a single grayscale distribution or texture statistics to extract regional features, and fail to form a dynamic evolution analysis path based on multi-time point sequence data, resulting in a lack of continuous understanding of the changing trend and difficulty in accurately capturing the key time points in the functional mutation process. Image registration at fixed time intervals has the problem of insensitivity to capturing the details of the changes, which is prone to errors when the functional area fluctuates violently, reducing the effectiveness of risk warning. Existing feature extraction is mostly based on static images, lacking the ability to comprehensively evaluate ventilation and perfusion pathways in time series, resulting in rough disturbance identification results and difficulty in forming a classifiable behavior pattern structure. In terms of risk assessment, the overall regional density or texture change mean is often used as the basis for risk judgment, ignoring the aggregation characteristics of disturbance behavior in the time series and morphological dimensions, making it difficult to effectively warn of the evolution trend at the lung segment level. For example, if the burst rate of some areas in a group of lung segments changes dramatically but does not reach the overall mean change threshold, the existing model will find it difficult to identify such detailed changes, resulting in the risk of missing early abnormalities. The lack of multi-dimensional data integration and sequence behavior analysis mechanism is an important shortcoming of existing technologies in terms of risk prediction accuracy and sensitivity. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a system and method for predicting radiation pneumonia after radiotherapy for chest tumors.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: A radiation pneumonia prediction system after chest tumor radiotherapy includes: The structural hot zone extraction module is based on the lung texture boundary map and lung perfusion difference map of individuals after chest tumor radiotherapy. It arranges the voxel grayscale ratios in chronological order, analyzes the grayscale change trend, screens the voxel areas whose continuous difference exceeds the lung function evolution benchmark threshold, locates the active functional latent change area within the radiotherapy target area, and generates a concentrated area map of lung function latent change. The density breakpoint identification module calls the lung function latent change concentrated area map, compares the grayscale changes at the previous and next time points, identifies the key grid points where the grayscale direction is reversed and the amplitude exceeds the texture change threshold, records the position, and generates a texture mutation grid coordinate set; The disturbance trajectory analysis module extracts the pulmonary ventilation and perfusion paths at multiple time points according to the pulmonary segment numbers corresponding to the texture mutation grid coordinate set, locates the rate mutation points, and collects the key disturbance information in the paths to generate a sequence of pulmonary segment function disturbance inflection points; The path cluster matching module calls the inflection point sequence of the lung segment function disturbance, extracts the high-frequency disturbance segments and time interval clusters, analyzes the inflection point timing and distribution morphology of the cluster center trajectory, compares the target path and the inflection point alignment rate and trajectory shape consistency, and obtains a list of disturbance behavior patterns.

[0007] As a further solution of the present invention, the lung function latent change concentrated area map includes functional abnormality clustering areas, grayscale fluctuation abnormal areas, and dose overlap high-frequency areas. The texture mutation grid coordinate set includes grayscale reversal points, local strong mutation points, and abnormal contrast edge points. The lung segment function disturbance inflection point sequence includes rate mutation time points, disturbance intensity landmark points, and ventilation and perfusion inflection point paths. The disturbance behavior pattern list includes inflection point clustering sequences, trajectory morphology categories, and disturbance segment frequency characteristics.

[0008] As a further solution of the present invention, the structural hot zone extraction module includes: The lung texture ratio calculation submodule is based on the lung texture boundary map and lung perfusion difference map of individuals after chest tumor radiotherapy. It arranges the voxel grayscale values ​​in time series, identifies the adjacent time series grayscale ratios and forms a ratio sequence, compares the grayscale difference with the lung function evolution benchmark threshold item by item, locates the difference voxel position, and generates the grayscale change trend value. The function shift area screening submodule screens the voxel set whose difference exceeds the reference threshold of lung function evolution based on the grayscale change trend value, extracts the spatial coordinates of the voxel set and matches them with the perfusion difference map, extracts the spatial overlap area and defines the connectivity range, and obtains the function shift area interval; The latent change area positioning submodule spatially overlaps the functional offset area interval with the radiotherapy dose distribution map, selects the area with the top dose value and density in the overlapping voxels, calibrates the corresponding position at the edge of the radiotherapy target area, and generates a concentrated area map of lung function latent changes.

[0009] As a further solution of the present invention, the density breakpoint identification module includes: The local grayscale extraction submodule calls the lung function latent change concentrated area map, divides the grid area, extracts the local grayscale extreme value and average value, records the grayscale distribution at each time point, and generates a grayscale distribution data set; The grayscale direction judgment submodule compares the grayscale mean values ​​of the grids before and after the grayscale distribution data set, identifies the reversal trend of the grayscale direction, records the corresponding grid index, and generates a grayscale direction change type set; The texture change screening submodule extracts the grayscale amplitude variation value and neighborhood grayscale variance of the direction reversal grid according to the grayscale direction change type set, calculates the texture variation amplitude value by combining the direction angle difference and the grayscale mean difference, screens the grid position whose amplitude exceeds the texture change threshold, identifies the corresponding coordinates in the area map, and generates a texture mutation grid coordinate set.

[0010] As a further solution of the present invention, the disturbance trajectory analysis module includes: The lung segment number extraction submodule matches the lung segment area corresponding to the lung image atlas according to the texture mutation grid coordinate set, extracts and integrates the lung segment numbers, performs grid and lung segment mapping, and obtains the lung segment identification code value; The ventilation perfusion path calculation sub-module calls the lung segment identification coding value, collects the ventilation and perfusion values of the lung segments at the collection time point, analyzes the time series and calculates the rate per unit time, identifies the ventilation and perfusion rate sequences, and performs fusion to obtain the ventilation perfusion rate sequence; The perturbation inflection point positioning sub-module, according to the ventilation perfusion rate sequence, identifies the absolute peak value of the difference between the ventilation and perfusion rates of the lung segments, extracts the perturbation sensitivity array, analyzes the mutation interval, calculates the rate perturbation coefficient within the monitoring period, selects the time points corresponding to the extreme values and corresponds them to the lung segments to generate the lung segment function perturbation inflection point sequence.

[0011] As a further solution of the present invention, the path clustering and matching module includes: The perturbation section extraction sub-module calls the lung segment function perturbation inflection point sequence, screens the sections with perturbation amplitudes exceeding the perturbation amplitude reference value, identifies the perturbation frequency and time interval, and clusters the sections with time intervals lower than the time interval threshold to obtain the high-frequency perturbation section set; The inflection point trajectory analysis sub-module, based on the high-frequency perturbation section set, extracts the inflection points of the clustering center trajectory, collects the inflection point time series and spatial coordinates, identifies the time series variance and distribution density, calculates the inflection point trajectory dispersion value, determines the distribution characteristics of the trajectory inflection points according to the dispersion, and obtains the inflection point trajectory distribution characteristic set; The path alignment consistency comparison sub-module, according to the inflection point trajectory distribution characteristic set, extracts the target path inflection point data, analyzes the matching inflection point quantity ratio and the trajectory shape consistency ratio, and screens the paths meeting the alignment rate and consistency requirements to obtain the perturbation behavior pattern list.

[0012] As a further solution of the present invention, the system further includes a risk trend output module: The risk trend output module, according to the matching degree between the paths in the perturbation behavior pattern list and the risk center trajectory, extracts the lung segment numbers and time periods where they are located, marks the perturbation trend synchronization intervals, classifies and locates the key warning events in the evolution process of the lung segments to obtain the radioactive pneumonia evolution risk signal set; The radioactive pneumonia evolution risk signal set includes risk lung segment numbers, potential risk time periods, and trend synchronization feature identifiers.

[0013] As a further solution of the present invention, the risk trend output module includes: The path screening sub-module, according to the spatio-temporal proximity between the paths in the perturbation behavior pattern list and the post-radiotherapy lung risk center trajectory of the patient, screens the paths meeting the perturbation proximity threshold, extracts the lung segment numbers and corresponding time periods where the paths are located, and generates the post-radiotherapy risk path interval value set; The synchronization marker sub-module calls the lung segment numbers and time periods in the post-radiotherapy risk path interval value set, and marks the time periods of disturbance synchronization according to the consistency of the change trend of lung segment disturbance in the path time sequence arrangement, so as to obtain the lung segment disturbance trend synchronization interval segment value; The key event positioning sub-module extracts the ratio of the disturbance amplitude change rate to the duration according to the lung segment numbers and time periods in the lung segment disturbance trend synchronization interval segment value, and screens the lung segment sequences exceeding the evolution mutation threshold, locates the corresponding time points and lung segment numbers, and obtains the radioactive pneumonia evolution risk signal set.

[0014] The method for predicting radioactive pneumonia after chest tumor radiotherapy is executed based on the above-mentioned system for predicting radioactive pneumonia after chest tumor radiotherapy, and includes the following steps: S1: Based on the lung texture boundary map and the lung region perfusion difference map of an individual after chest tumor radiotherapy, arrange the voxel gray ratio in chronological order, extract the continuous gray difference value, screen the voxels with the difference value greater than the lung function evolution threshold, and superimpose the radiotherapy dose distribution map for regional screening to obtain the functional creep intersection region map; S2: Based on the functional creep intersection region map, identify the local gray range in the grid and the gray difference between the front and back time points, screen the grid coordinates with the gray direction reversed and the change amplitude exceeding the texture change threshold, and obtain the texture mutation marker set; S3: Based on the lung segment numbers corresponding to the texture mutation marker set, extract the ventilation path and perfusion path at multiple time points, perform sequence alignment on the path rate change values, locate the rate fluctuation mutation points, and establish a lung segment ventilation and perfusion disturbance sequence map; S4: Based on the lung segment ventilation and perfusion disturbance sequence map, extract the high-frequency disturbance section and the corresponding disturbance time period, cluster the disturbance trajectory nodes, analyze the order and distribution consistency of the trajectory turning, screen the concentrated area of disturbance behavior, and obtain the radioactive change behavior clustering map; S5: Based on the comparison result between the disturbance trajectory in the radioactive change behavior clustering map and the risk evolution trajectory library, screen the lung segments and time periods with overlapping trajectories, mark the disturbance and risk synchronization intervals, and obtain the radioactive pneumonia evolution risk signal set.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In the present invention, by continuously arranging voxel gray-scale ratios on the time axis and extracting the region coinciding with the radiation dose distribution, the positioning of the active region of functional creep can be focused on, highly sensitive identification of the evolution trend of lung function can be achieved, the temporal analysis ability of functional variation detection can be enhanced, and by comparing the local gray-scale range and extracting the direction reversal points, a fine texture mutation map is constructed at the image level, making the identification of abnormal regions have stronger quantitative accuracy and structural recognition clarity. By comparing the rates of ventilation and perfusion paths under the unified time axis, the continuity and mutability of lung segment function disturbances can be grasped from the dynamic dimension, the key turning states in the evolution process can be captured, and a stable foundation can be laid for subsequent path analysis. By comparing path behaviors through the inflection point alignment rate and trajectory consistency index, the laws of functional disturbances are clustered and integrated, the classification ability of the logical evolution of lung segment regions is improved, and the potential evolution trends have the characteristics of being traceable, classifiable, and evaluable. Based on the comparison of paths and risk trajectories, the synchronous interval and evolution period of disturbance trends are extracted, the forward-looking identification ability of key warning events is effectively improved, and a logical support system is established for the accurate output of high-risk evolution signals. Through fine quantitative analysis from multiple dimensions such as multi-dimensional structure, density, dynamic path, and temporal behavior, the formed prediction basis has the ability of full-process closed-loop tracking and lung segment level positioning, making the expression of risk trends more accurate, distinguishable, and timely. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is the system flow chart of the present invention; Figure 2 is the flow chart of the structural hot zone extraction module in the present invention; Figure 3 is the flow chart of the density breakpoint identification module in the present invention; Figure 4 is the flow chart of the disturbance trajectory analysis module in the present invention; Figure 5 is the flow chart of the path clustering and matching module in the present invention; Figure 6 is the flow chart of the risk trend output module in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] In order to make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. 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.

[0018] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0019] See also Figure 1 The present invention provides a technical solution: a radiation pneumonia prediction system after chest tumor radiotherapy includes: The structural hot zone extraction module is based on the lung texture boundary map and lung perfusion difference map of individuals after chest tumor radiotherapy. It arranges the voxel grayscale ratios in chronological order, analyzes the grayscale change trend, screens the voxel areas whose continuous difference exceeds the lung function evolution benchmark threshold, extracts the areas that overlap with the radiotherapy dose distribution, locates the functional latent active areas within the radiotherapy target area, and generates a concentrated area map of lung function latent changes; The density breakpoint recognition module calls the lung function latent change concentrated area map, extracts the local grayscale range and contrast of the grid area, compares the grayscale changes at the previous and next time points, identifies the key grid points where the grayscale direction is reversed and the amplitude exceeds the texture change threshold, records the position, and generates a texture mutation grid coordinate set; The disturbance trajectory analysis module extracts the pulmonary ventilation and perfusion paths at multiple time points according to the lung segment numbers corresponding to the texture mutation grid coordinate set, compares the continuous rate changes after unifying the time axis, locates the rate mutation points, and collects the key disturbance information in the path to generate a sequence of lung segment function disturbance inflection points; The path cluster matching module calls the inflection point sequence of lung segment function disturbance, extracts high-frequency disturbance segments and time interval clusters, analyzes the inflection point timing and distribution morphology of the cluster center trajectory, compares the alignment rate of the target path and the inflection point and the trajectory shape consistency, and obtains a list of disturbance behavior patterns; The risk trend output module extracts the lung segment number and time period according to the matching degree between the path in the disturbance behavior pattern list and the risk center trajectory, marks the disturbance trend synchronization interval, classifies and locates the key warning events in the lung segment evolution process, and obtains the radiation pneumonia evolution risk signal set.

[0020] The pulmonary function latent change concentrated area map includes functional abnormality clustering areas, grayscale fluctuation abnormal areas, and dose overlap high-frequency areas. The texture mutation grid coordinate set includes grayscale reversal points, local strong mutation points, and abnormal contrast edge points. The pulmonary segment function disturbance inflection point sequence includes rate mutation time points, disturbance intensity landmark points, and ventilation and perfusion inflection point paths. The disturbance behavior pattern list includes inflection point clustering sequences, trajectory morphology categories, and disturbance segment frequency characteristics. The radiation pneumonia evolution risk signal set includes risk lung segment numbers, potential risk periods, and trend synchronization feature identifiers. Lung function latent change concentrated area map: By analyzing the lung texture boundary map and lung perfusion difference map of individuals after chest tumor radiotherapy, the voxel grayscale ratio is arranged in chronological order, and the voxel areas with continuous differences exceeding the baseline threshold of lung function evolution are screened, and finally the active functional latent change area within the radiotherapy target area is located. By analyzing the changing trend of the voxel grayscale ratio and combining the radiotherapy dose distribution map, the location of the functional latent change area is accurately extracted to generate a lung function latent change concentrated area map, which helps to accurately locate the potential change area of ​​the lung, especially after radiotherapy, to determine which areas may have functional changes; Texture mutation grid coordinate set: By comparing the grayscale changes of the lung function latent change concentrated area map at different time points, the grid points where the grayscale direction is reversed and the amplitude exceeds the texture change threshold are identified, and the coordinate positions of these changes are recorded. The process is achieved by analyzing the grayscale change trend of the lung texture, and the key areas with significant texture changes are marked. The grid coordinate set provides precise spatial coordinate support for subsequent lung segment function disturbance analysis, ensuring that the texture mutation area can be correctly identified and located; Lung segment function perturbation inflection point sequence: Based on the texture mutation grid coordinate set, the system extracts the ventilation and perfusion paths of the corresponding lung segments, and locates the rate mutation points according to the rate change values ​​of these paths. By collecting the key perturbation information appearing in the ventilation and perfusion paths, a sequence of inflection points of lung segment function perturbation is generated. The sequence marks the key turning points in the changes in lung segment function, which helps to accurately identify the drastic functional changes that may occur after radiotherapy and provides important clinical reference data; Disturbance behavior pattern list: By clustering and analyzing the inflection point sequence of pulmonary segment function disturbance, the high-frequency disturbance segment and its corresponding time interval are extracted, and the pattern list of disturbance behavior is obtained by comparing the consistency of the target path and the inflection point trajectory morphology. The analysis compares the trajectory distribution characteristics of different disturbance paths, identifies similar behavior patterns, and then classifies and evaluates the potential risk of radiation pneumonia. The pattern list provides a basis for subsequent risk assessment and treatment decisions, and can better guide clinical intervention measures.

[0021] See also Figure 2 , the structural heat zone extraction module includes: The lung texture ratio calculation submodule is based on the lung texture boundary map and lung perfusion difference map of individuals after chest tumor radiotherapy. It arranges the voxel grayscale values ​​in time series, identifies the adjacent time series grayscale ratios and forms a ratio sequence, compares the grayscale difference with the lung function evolution benchmark threshold item by item, locates the difference voxel position, and generates the grayscale change trend value. Each voxel in the lung texture boundary map is processed to identify the lung area in the image. The grayscale ratio of each voxel in the area to its adjacent time series is calculated. By comparing the grayscale ratio sequence, the grayscale difference is compared item by item with the lung function evolution benchmark threshold to determine whether it exceeds the set threshold, and the position of the voxel that meets the conditions is calibrated. For example, assuming that at a specific moment, the grayscale value of a voxel is 150, and the grayscale value of the adjacent voxel in the time series is 160, the calculated grayscale ratio is 150 / 160=0.9375, and then compared with the set threshold. If the threshold is set to 0.95, the voxel will be marked as abnormal. Through this series of calculations, a grayscale change trend value is finally generated, which reflects the trend of lung function changes in a specific area and further locates the functionally impaired area after radiotherapy.

[0022] The functional deviation area screening submodule screens the voxel set whose difference exceeds the baseline threshold of lung function evolution based on the grayscale change trend value, extracts the spatial coordinates of the voxel set and matches them with the perfusion difference map, extracts the spatial overlap area and defines the connectivity range to obtain the functional deviation area interval; When screening the functional deviation area, based on the grayscale change trend value, the voxel set whose grayscale change difference exceeds the baseline threshold of lung function evolution is screened out, and the coordinate value of the voxel in three-dimensional space is further extracted, and spatial matching is performed with the lung area perfusion difference map to find the overlapping area at the same position. For example, assuming that at a certain moment, the grayscale change amplitude of a voxel is 0.98, which exceeds the preset 0.95 baseline threshold, then the voxel will be included in the screening set, and the spatial coordinates of the voxel will be compared with the perfusion difference map. If the perfusion difference value of a certain area is 20%, and there is significant overlap in the voxel position within the voxel set, it can be determined as a functional deviation area interval. In this process, a spatial coordinate matching algorithm is used to screen out areas with potential functional impairment by calculating the relative positions between voxels.

[0023] The latent change area positioning submodule spatially overlaps the functional deviation area interval with the radiotherapy dose distribution map, selects the area with the highest dose value and density in the overlapping voxels, calibrates the corresponding position at the edge of the radiotherapy target area, and generates a concentrated area map of lung function latent changes; During the creep zone positioning process, the functional offset zone interval is spatially overlapped with the radiotherapy dose distribution map. By calculating the dose values and densities within the overlapping region, the voxel regions with the leading dose values and densities are selected, and the regions will be marked as creep zones. Specifically, assuming that the radiotherapy dose of a certain region is 40 Gy and the density of this region is 90%, then in the ranking of dose values and densities, the voxels are in the leading positions. Through spatial overlap analysis, regions with higher doses and densities are selected, further marked as creep regions, and coincide with the edge regions of the radiotherapy target area to ensure that potential lung function impairment regions can be identified, generating a map of the concentrated location of lung function creep.

[0024] Please refer to Figure 3 , the density breakpoint identification module includes: The local gray-scale extraction sub-module calls the map of the concentrated location of lung function creep, divides the grid regions, extracts the local gray-scale extreme values and average values, records the gray-scale distribution at each time point, and generates a gray-scale distribution data set; In the clinical application of radiation pneumonitis prediction, dividing each grid region and extracting the extreme values and average values of the local gray scale are key steps. The process of dividing the grid regions involves partitioning the CT image into squares of 5 mm×5 mm, and the gray-scale values of each region are calculated to obtain its statistical data, such as the maximum and minimum gray-scale values, which helps to identify the radiation changes in specific regions in the follow-up. For example, in an actual patient examination, if the average gray-scale value of a certain grid is significantly higher than that of the surrounding grids, it indicates the occurrence of inflammation in this region. This process not only helps doctors quickly identify the lesion regions, but also provides data support for subsequent detailed analysis. Such refinement makes the gray-scale distribution data set more accurate. Recording the gray-scale distribution at each time point provides a comparison of the changes before and after, and accurately recording the data is the basis for realizing dynamic monitoring. The synthesized gray-scale distribution data set is an indispensable part of quantitative analysis, enabling radiologists to observe the progression of the disease or the effect of treatment by comparing the data at different time points, generating a gray-scale distribution data set.

[0025] The gray-scale direction judgment sub-module, based on the gray-scale distribution data set, compares the gray-scale means of the grids at different time points before and after, identifies the reverse trend of the gray-scale direction, records the corresponding grid indices, and generates a set of gray-scale direction change types; The process of comparing the grayscale mean values of each grid at different time points is a commonly used method for analyzing radiographic images, which is used to detect the evolution of lung diseases. It automatically calculates the difference in grayscale mean values of each grid before and after treatment, and determines whether there are significant changes by setting a grayscale difference threshold. If the grayscale mean value of a grid shows a change from increasing to decreasing from the previous time point to the next time point, it indicates that the degree of inflammation in the lung lesion area has decreased. The set of this type of change (the set of grayscale direction change types) provides doctors with an intuitive way to observe the trend of the disease. The set of data reflects the preliminary basis for judging the treatment effect. The results of data analysis are very crucial and directly affect the doctor's adjustment of the next treatment plan for the patient. Record the corresponding grid index and generate the set of grayscale direction change types.

[0026] According to the set of grayscale direction change types, the texture change screening sub-module extracts the grayscale change amplitude value and the neighborhood grayscale variance of the grids with reversed directions, combines the direction angle difference and the grayscale mean difference, and uses the formula: ; Calculate the texture change amplitude measurement value, screen the grid positions with change amplitudes exceeding the texture change threshold, identify the corresponding coordinates in the location map, and generate the set of texture mutation grid coordinates; Among them, represents the texture change amplitude measurement value, is the grayscale change amplitude value of the th grid, is the neighborhood grayscale variance of the th grid, represents the grayscale direction angle difference of the th grid, is a non-zero offset constant, is the absolute value of the grayscale mean difference of the th grid, is the total number of grids with reversed directions; Extract all the grid areas where grayscale direction reversal occurs, and calculate the grayscale change amplitude value, neighborhood grayscale variance, grayscale direction angle difference, and grayscale mean difference for each grid. The grayscale change amplitude value is obtained by extracting the difference between the maximum and minimum grayscale values of the grid at the two time points before and after. In a specific example, if the grayscale of a certain grid is 185 at the first time point and 135 at the second time point, then , and the neighborhood grayscale variance is achieved by calculating the sample variance after extracting the grayscale values of the 8 adjacent grids around the grid. For example, if the surrounding grayscale values are 170, 172, 175, 168, 165, 174, 169, 171, the sample variance is , and the grayscale direction angle difference Measured by the change in the direction of the gray - scale gradient vector, assuming it is 22 degrees (converted to the radian unit 0.384), a non - zero offset constant is introduced to prevent the denominator from being zero, the difference in gray - scale means is the difference in gray - scale means at the previous and current time points. For example, if they are 160 and 145 respectively, then , select 5 grids with reversals to participate in the calculation, so ; Let the grid parameters be (unit: dimensionless gray - scale value, angle unit is radian): Grid 1: ; Grid 2: ; Grid 3: ; Grid 4: ; Grid 5: ; Substitute into the calculation : Grid 1: ; Grid 2: ; Grid 3: ; Grid 4: ; Grid 5: ; Sum and take the average: ; Through the comprehensive measure that fuses the gray - scale amplitude change, neighborhood variability, and angle variability, joint control of complex image features is carried out when screening for abnormal texture changes, effectively filtering out misjudgments caused by gray - scale jitter. The result shows that the texture amplitude measurement value in the current direction reversal region is 382.37, which has significantly exceeded the empirical texture mutation threshold (set as 200), indicating that the grid is a key position for gray - scale mutation and is identified as the texture mutation grid coordinate set in the follow - up.

[0027] Please refer to Figure 4 , the perturbation trajectory analysis module includes: The lung segment number extraction sub - module, according to the texture mutation grid coordinate set, matches the corresponding lung segment regions in the lung image atlas, extracts and integrates the lung segment numbers, performs grid - to - lung segment mapping, and obtains the lung segment identification code value; In medical image analysis, texture mutation information is extracted from a patient's CT scan data to identify different lung segments in the lungs. This process involves using advanced image processing techniques to accurately identify the lung segment positions of each coordinate point. For example, when planning lung cancer surgery, accurate lung segment identification can help doctors formulate more precise surgical strategies. Medical technicians load the patient's CT images into a dedicated software, and the software automatically performs image segmentation, marking different lung segments. The lung segment numbers of each coordinate point are extracted and integrated and de-duplicated through algorithms to ensure the uniqueness of each number. For example, when multiple coordinate points correspond to the same lung segment, the data points are merged through algorithms to generate a unique set of numbers containing all the necessary lung segment information, obtaining the lung segment identification coding value.

[0028] The ventilation perfusion path calculation sub-module calls the lung segment identification coding value, collects the ventilation and perfusion values of the lung segments at the acquisition time points, analyzes the time series and calculates the rate per unit time, identifies the ventilation and perfusion rate sequences, and performs fusion to obtain the ventilation perfusion rate sequence; In pulmonary function tests, the dynamic changes in pulmonary ventilation and blood perfusion are evaluated. Taking pneumonia patients as an example, the ventilation and perfusion conditions of the lungs are monitored through sequence imaging techniques to understand the specific location of the inflammatory area and its impact on pulmonary function. During implementation, through a ventilator and a blood flow dynamic monitoring device, the ventilation and perfusion data of the patient at each time point are continuously recorded. This data is automatically matched with the previously extracted lung segment identification coding value by medical monitoring software to ensure that each data point accurately corresponds to a specific lung segment. Through the data, the medical team can observe the changes in pulmonary ventilation and perfusion at different time points, helping to evaluate the treatment effect or disease progression, obtaining the ventilation perfusion rate sequence.

[0029] The perturbation inflection point positioning sub-module, based on the ventilation perfusion rate sequence, identifies the absolute peak value of the difference between the ventilation and perfusion rates of the lung segments, extracts the perturbation sensitivity array, analyzes the mutation interval, and uses the formula: ; Calculates the rate perturbation coefficient within the monitoring period, selects the time points and lung segments corresponding to the extreme values, and generates the lung segment function perturbation inflection point sequence; Among them, represents the rate perturbation coefficient within the monitoring period, represents the absolute value of the rate change at time point , is the sum of the ventilation value and the perfusion value at time point , is the rate change amount between adjacent time points , , are respectively the maximum and minimum values of the rate difference at time point , is the median value of ventilation perfusion difference, is the total number of time points; Calculate the absolute value of the first-order difference of the rate and take the peak value of the change at the lung segment level. First, obtain the ventilation rate of the lung segment at different times and perfusion rate , calculate their combined rate , then obtain the change amount between its consecutive time points , and take the absolute value for mutation identification. In practice, the unit of ventilation rate is mL / s, and the unit of perfusion rate is mL / min. To unify the dimension, all data need to be unit-converted and normalized. The ventilation rate remains mL / s, and the perfusion rate is converted to mL / s and then linearly normalized. For example, assume the maximum ventilation rate is 800 mL / s and the minimum is 300 mL / s. Map this interval to 0 to 1 for normalization operation. The normalization formula is: , similarly, the perfusion rate is also normalized. After normalization, reconstruct the ventilation perfusion rate sequence, and then obtain the absolute value of the change in the combined rate at each time point. In the set time range to , calculate the following participating items for each lung segment respectively: Assume the normalized rate , , , , ; Get , ; For , that is, the sum of ventilation and perfusion. Assume the normalized values of ventilation and perfusion are as follows: , , , ; For , that is, the square of the adjacent rate change amount: , , , ; For , , take the extreme value in the rate change at the current time point: ; ; Set the median value ; Substitute all values into the original formula: ; The result shows that there is an obvious fluctuation with a rate perturbation coefficient of 1.567 in this lung segment during the monitoring period, indicating a non-stationary mutation in its ventilation and perfusion dynamics. Since the perturbation coefficient has exceeded the discriminant benchmark value of 1.3 set by the system, it is determined as a functionally perturbed sensitive area. By introducing the combined calculation of three quantification factors, namely the absolute value of the rate change, the combined ventilation and perfusion volume, and the square of the change amount, the perturbation coefficient can not only reflect the numerical mutation, but also capture the temporal trend and the differences between lung segments, forming a sensitive identification basis for the mutated areas of lung function, so as to more precisely locate the inflection point sequence of the lung segment functional perturbation.

[0030] Please refer to Figure 5 , and the path clustering and matching module includes: The perturbed section extraction sub-module calls the inflection point sequence of the lung segment functional perturbation, screens the sections with a perturbation amplitude exceeding the perturbation amplitude benchmark value, identifies the perturbation frequency and time interval, and clusters the sections with a time interval lower than the time interval threshold to obtain a set of high-frequency perturbed sections; Starting from the acquisition of the inflection point sequence of the lung segment functional perturbation, through the process of screening and clustering to generate a set of high-frequency perturbed sections, this process is applied in lung function tests, especially suitable for monitoring the changes in lung function of patients with chronic obstructive pulmonary disease (COPD). In daily monitoring, the lung perturbation data collected by the device is first parsed into a sequence, which records the fluctuation points of each breath. By comparing whether the amplitude of the fluctuation points exceeds a preset threshold, abnormal perturbed sections can be preliminarily screened. This threshold is set as twice the standard deviation of normal breathing fluctuations. Calculate the frequency of the high-perturbed section and the time interval between adjacent perturbation points. The time interval analysis helps to identify the irregularities in the breathing pattern, such as short-term apnea or abnormally rapid breathing. Cluster analysis further classifies the time intervals to determine whether there are common pathological breathing patterns, generating a set of high-frequency perturbed sections, providing a detailed report on the patient's condition changes for doctors. Such monitoring and analysis work is crucial for early diagnosis and treatment plan formulation. Among them, the setting of the threshold is based on the statistical analysis of the original patient data, and through an example demonstration, if the average perturbation amplitude of COPD patients in the original data is 5%, the threshold is set to 10%.

[0031] Based on the set of high-frequency perturbed sections, the inflection point trajectory analysis sub-module extracts the inflection points of the clustering center trajectory, collects the time series and spatial coordinates of the inflection points, identifies the variance and distribution density of the time series, and uses the formula: ; Calculate the inflection point trajectory dispersion value, determine the distribution characteristics of the trajectory inflection points according to the dispersion, and obtain a set of inflection point trajectory distribution characteristics; Among them, represents the inflection point trajectory dispersion value, represents the time series value of the th inflection point, represents the mean of the inflection point time series, represents the distance from the th inflection point to the central trajectory, represents the number of local neighbors of the th inflection point; Extract the central trajectory of the cluster from each section, call all the perturbed inflection points in the trajectory and record their time series values and spatial coordinate information. To achieve the quantitative analysis of the inflection point trajectory distribution, first calculate the time distribution characteristics of the inflection points. By squaring and summing the difference between the timestamp of each inflection point and the average time of all inflection points , the variance value of the time distribution is obtained. Subsequently, it is necessary to evaluate the spatial distribution density of the inflection points and the cluster center. Therefore, record the Euclidean distance from each inflection point to the cluster center, and count the number of its neighbors in the local area. All units need to be unified into the standard format, that is, the time unit is unified to seconds, and the spatial unit is unified to millimeters. To eliminate the influence caused by different dimensions of the participating items, all participating values need to be normalized. Among them, the time-type parameters (such as , ) are normalized using the min-max normalization method: ; The distance-type parameters ( ) are uniformly divided by the maximum distance between inflection points in this section of the trajectory. The neighbor number parameter has a small value range (not exceeding 10), and is normalized by direct proportion. Set the actual scenario data as follows: Suppose there are 5 inflection points, and their normalized time series values are , , , , ; Calculate the average of the time series as: ; Then assume that the normalized distances of each inflection point from the center are , , , , , and the corresponding neighbor numbers are , , , , ; Substitute into the formula: ; Thus, the inflection point trajectory dispersion value is obtained as 0.1997. This value is a quantitative description index of the clustering center trajectory of this segment and is used for subsequent trajectory distribution stability classification and comparison. The lower this value, the more concentrated the inflection points of the trajectory are distributed in time and space, which can be used as a reference for the relatively stable pulmonary function after radiotherapy. This result is a measure of the concentration of the inflection point trajectory distribution characteristics; By introducing both the time series dispersion term and the space density term, the distribution characteristics of the trajectory no longer depend on a single dimension, and the number of neighbors parameter is introduced simultaneously Adjust the spatial proportion of each inflection point. Under the condition of avoiding the influence of spatial density, the accuracy of trajectory pattern recognition is improved. This result shows that the trajectory dispersion value is 0.1997, which is lower than the set reference threshold of 0.25 for stable trajectories. Therefore, it can be judged that the inflection point distribution of this trajectory is stable and can be classified as a post-intervention recovery type trajectory, and should be preferentially included in the reference trajectory sample in subsequent behavior pattern comparison.

[0032] The path alignment consistency comparison sub-module extracts the inflection point data of the target path according to the inflection point trajectory distribution characteristic set, analyzes the ratio of the number of matching inflection points and the trajectory shape consistency ratio, and screens the paths that meet the alignment rate and consistency requirements to obtain the list of perturbation behavior patterns; From the application of the inflection point trajectory distribution characteristic set to the generation of the perturbation behavior pattern list, this process can be widely applied to the diagnosis of respiratory-related diseases and the monitoring of disease progression. In clinical practice, by comparing the patient's breathing path with the known disease characteristic path for alignment, doctors can judge the severity and change trend of the condition, call the patient's real-time breathing data from the monitoring device, and compare it with the disease characteristic path in the database to calculate the inflection point matching ratio and the consistency of the trajectory shape. The calculation not only examines the similarity of the paths, but also evaluates the consistency of time and shape, providing an important basis for doctors to analyze the condition. Through example calculation, if the ratio of the number of matching inflection points of the patient reaches more than 70%, and the shape consistency is also higher than 80%, it is considered that the patient's condition is highly consistent with a specific disease model and urgent intervention is required. The setting of this ratio and consistency benchmark is obtained through the cumulative analysis of long-term clinical data, and the list of perturbation behavior patterns is obtained.

[0033] Please refer to Figure 6 , the risk trend output module includes: The path screening sub-module screens the paths that meet the perturbation proximity threshold according to the spatio-temporal proximity between the paths in the perturbation behavior pattern list and the post-radiotherapy lung risk center trajectory of the patient, extracts the lung segment numbers and corresponding time periods where the paths are located, and generates the post-radiotherapy risk path interval value set; By analyzing the spatio-temporal data of the paths, the paths that meet the perturbation proximity threshold are screened out. These paths will be used to extract representative lung segment numbers and corresponding time periods. Combining with the lung CT image data, the screened paths are verified to ensure that the path selection is consistent with the actual situation of lung radiation changes. For example, if a radiologist finds an increase in lung density in a specific area during the follow-up after treatment, the information will be verified corresponding to the path data to confirm the accuracy of the path selection. This process not only depends on clinical data but also needs to combine pathological reports to determine the nature and scope of lung changes. Through the analysis of comprehensive information, a set of post-radiotherapy risk path interval values is finally generated. The set reflects the specific intervals of radioactive changes in the lungs, providing key data support for subsequent treatment and monitoring. This method precisely calls the parameters in the lung images and calculates the matching degree with the risk trajectory, specifically including the length, width of the path, and the distance from key lung segments, etc. The data is input into the analysis model to obtain a comprehensive score, which reflects the proximity of each path to the risk center trajectory. The score is calculated by comparing the path scores of similar cases in the original data and the current data to ensure its accuracy and reliability. If this score is higher than the set risk threshold, the path is considered a path with potential risk. The threshold is set based on the analysis of past cases and expert opinions. The specific threshold is set to 0.8, which means that any path with a score higher than this value will be regarded as a high-risk path and requires further analysis and monitoring.

[0034] The synchronous marking sub-module calls the lung segment numbers and time periods in the post-radiotherapy risk path interval value set, and marks the time segments with synchronous perturbations based on the consistency of the lung segment perturbation trend changes in the path time sequence arrangement, and obtains the lung segment perturbation trend synchronous interval segment value; Synchronously mark the time segments of lung segment perturbations. The time overlap and spatial coincidence degrees between each path will be calculated to determine the time segments with synchronous perturbations. For example, if two paths appear in the same lung segment within the same or close time periods, this will be marked as a synchronous interval. The marking process is automated and depends on the time and space matching criteria set in the algorithm. The criteria include that the time coincidence degree is at least 70% and the spatial coincidence degree is at least 80% to ensure the accuracy and consistency of the marking. Through this method, the time and area of radioactive pneumonia complication in the patient's lungs after radiotherapy can be effectively identified, providing support for clinical decision-making. The marked data will be used to generate the lung segment perturbation trend synchronous interval segment value. The value set is obtained by summarizing and analyzing all marked synchronous intervals, reflecting the radioactive change trend of the lungs within a specific time period. The identification of the trend is crucial for predicting the progression of the patient's radioactive pneumonia. The marked synchronous intervals will also be used for subsequent risk assessment and treatment plan adjustment.

[0035] The key event positioning sub-module synchronizes the lung segment numbers and time periods in the interval segment values according to the lung segment perturbation trend, extracts the ratio of the perturbation amplitude change rate to the duration, and screens the lung segment sequences that exceed the evolution mutation threshold, locates the corresponding time points and lung segment numbers, and obtains the radioactive pneumonia evolution risk signal set; Analyze the perturbation data of each lung segment to ensure that the selected lung segments are the key areas of radioactive pneumonia evolution risk. For example, if a lung segment shows a faster density increase than other areas in consecutive time periods, the change will be regarded as a key early warning of evolution. This determination is based on the calculation of the ratio of the perturbation amplitude change rate to time, which is obtained by dividing the actually measured density change rate by the time when the change occurs. If the ratio exceeds the preset mutation threshold, the lung segment is recorded and further analyzed. The threshold is set based on the original data and statistical analysis, and the specific value is 1.5, indicating the situation where the lung segment density increases rapidly in a short time. This process not only provides a quantitative way to evaluate the risk of radioactive pneumonia, but also enables doctors to better monitor the patient's condition changes during the treatment process. Finally, a radioactive pneumonia evolution risk signal set is generated. The signal set contains all the lung segments marked as high-risk and their related information, providing a targeted early warning for clinical practice, helping doctors adjust treatment strategies and preventive measures to reduce or avoid the consequences of severe radioactive pneumonia.

[0036] The prediction method of radioactive pneumonia after chest tumor radiotherapy is performed based on the above-mentioned prediction system for radioactive pneumonia after chest tumor radiotherapy, and includes the following steps: S1: Based on the lung texture boundary map and lung region perfusion difference map of an individual after chest tumor radiotherapy, arrange the voxel gray ratio in chronological order, extract the continuous gray difference, screen the voxels with a difference greater than the lung function evolution threshold, and superimpose the radiotherapy dose distribution map for regional screening to obtain the functional creep intersection region map; S2: Based on the functional creep intersection region map, identify the local gray range within the grid and the gray difference between the previous and subsequent time points, screen the grid coordinates with a reversed gray direction and an amplitude change exceeding the texture change threshold, and obtain the texture mutation marker set; S3: Based on the lung segment numbers corresponding to the texture mutation marker set, extract the ventilation path and perfusion path at multiple time points, perform sequence alignment on the path rate change values, locate the rate fluctuation mutation points, and establish a lung segment ventilation-perfusion perturbation sequence map; S4: Based on the lung segment ventilation-perfusion perturbation sequence map, extract the high-frequency perturbation sections and the corresponding perturbation time periods, cluster the perturbation trajectory nodes, analyze the trajectory turning order and distribution consistency, screen the concentrated areas of perturbation behavior, and obtain the radioactive change behavior clustering map; S5: Based on the comparison results between the perturbation trajectories in the radioactive change behavior clustering map and the risk evolution trajectory library, screen the lung segments and time periods with overlapping trajectories, mark the perturbation and risk synchronization intervals, and obtain the radioactive pneumonia evolution risk signal set.

[0037] The above are only the preferred embodiments of the present invention, and do not limit the present invention in other forms. Any person skilled in the art may use the technical content disclosed above to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A prediction system for radiation pneumonitis after radiotherapy of chest tumors, characterized in that, The system comprises: The structural hot zone extraction module is based on the lung texture boundary map and lung perfusion difference map of individuals after chest tumor radiotherapy. It arranges the voxel grayscale ratios in chronological order, analyzes the grayscale change trend, screens the voxel areas whose continuous difference exceeds the lung function evolution benchmark threshold, locates the active functional latent change area within the radiotherapy target area, and generates a concentrated area map of lung function latent change. The density breakpoint identification module calls the lung function latent change concentrated area map, compares the grayscale changes at the previous and next time points, identifies the key grid points where the grayscale direction is reversed and the amplitude exceeds the texture change threshold, records the position, and generates a texture mutation grid coordinate set; The disturbance trajectory analysis module extracts the pulmonary ventilation and perfusion paths at multiple time points according to the pulmonary segment numbers corresponding to the texture mutation grid coordinate set, locates the rate mutation points, and collects the key disturbance information in the paths to generate a sequence of pulmonary segment function disturbance inflection points; The path cluster matching module calls the inflection point sequence of the lung segment function disturbance, extracts the high-frequency disturbance segments and time interval clusters, analyzes the inflection point timing and distribution morphology of the cluster center trajectory, compares the target path and the inflection point alignment rate and trajectory shape consistency, and obtains a list of disturbance behavior patterns.

2. The radioactive pneumonia prediction system after chest tumor radiotherapy according to claim 1, wherein The lung function latent variation concentrated area map includes functional abnormality clustering areas, grayscale fluctuation abnormal areas, and dose overlap high-frequency areas. The texture mutation grid coordinate set includes grayscale reversal points, local strong mutation points, and abnormal contrast edge points. The lung segment function disturbance inflection point sequence includes rate mutation time points, disturbance intensity landmark points, and ventilation and perfusion inflection point paths. The disturbance behavior pattern list includes inflection point clustering sequences, trajectory morphology categories, and disturbance segment frequency characteristics.

3. The radioactive pneumonia prediction system after chest tumor radiotherapy according to claim 1, wherein The structural heat zone extraction module includes: The lung texture ratio calculation submodule is based on the lung texture boundary map and lung perfusion difference map of individuals after chest tumor radiotherapy. It arranges the voxel grayscale values ​​in time series, identifies the adjacent time series grayscale ratios and forms a ratio sequence, compares the grayscale difference with the lung function evolution benchmark threshold item by item, locates the difference voxel position, and generates the grayscale change trend value. The function shift area screening submodule screens the voxel set whose difference exceeds the reference threshold of lung function evolution based on the grayscale change trend value, extracts the spatial coordinates of the voxel set and matches them with the perfusion difference map, extracts the spatial overlap area and defines the connectivity range, and obtains the function shift area interval; The latent change area positioning submodule spatially overlaps the functional offset area interval with the radiotherapy dose distribution map, selects the area with the top dose value and density in the overlapping voxels, calibrates the corresponding position at the edge of the radiotherapy target area, and generates a concentrated area map of lung function latent changes.

4. The radioactive pneumonia prediction system after chest tumor radiotherapy according to claim 3, wherein The density breakpoint identification module comprises: The local grayscale extraction submodule calls the lung function latent change concentrated area map, divides the grid area, extracts the local grayscale extreme value and average value, records the grayscale distribution at each time point, and generates a grayscale distribution data set; The grayscale direction judgment submodule compares the grayscale mean values ​​of the grids before and after the grayscale distribution data set, identifies the reversal trend of the grayscale direction, records the corresponding grid index, and generates a grayscale direction change type set; The texture change screening sub-module extracts the gray value change amplitude of the direction reversal grid and the neighborhood gray variance according to the gray direction change type set, combines the direction angle difference and the gray mean difference, calculates the texture change measurement value, screens the grid positions with the change amplitude exceeding the texture change threshold, identifies the corresponding coordinates in the location map, and generates a set of texture mutation grid coordinates.

5. The radioactive pneumonia prediction system after chest tumor radiotherapy according to claim 4, wherein The disturbance trajectory analysis module includes: The lung segment number extraction sub-module matches the corresponding lung segment regions in the lung image atlas according to the set of texture mutation grid coordinates, extracts and integrates the lung segment numbers, performs grid and lung segment mapping, and obtains the lung segment identification code value; The ventilation and perfusion path calculation sub-module calls the lung segment identification code value, collects the ventilation and perfusion values of the lung segments at the time points, analyzes the time series and calculates the unit time rate, identifies the ventilation and perfusion rate sequences, and performs fusion to obtain the ventilation and perfusion rate sequence; The disturbance inflection point positioning sub-module identifies the absolute peak value of the difference between the ventilation and perfusion rates of the lung segments according to the ventilation and perfusion rate sequence, extracts the disturbance sensitivity array, analyzes the mutation interval, calculates the rate disturbance coefficient within the monitoring period, selects the time points corresponding to the extreme values and the lung segments, and generates a sequence of lung segment function disturbance inflection points.

6. The radioactive pneumonia prediction system after chest tumor radiotherapy according to claim 5, characterized in that, The path clustering and matching module includes: The disturbance section extraction sub-module calls the sequence of lung segment function disturbance inflection points, screens the sections with the disturbance amplitude exceeding the disturbance amplitude reference value, identifies the disturbance frequency and time interval, and clusters the sections with the time interval lower than the time interval threshold to obtain a set of high-frequency disturbance sections; The inflection point trajectory analysis sub-module extracts the clustering center trajectory inflection points based on the set of high-frequency disturbance sections, collects the inflection point time series and spatial coordinates, identifies the time series variance and distribution density, calculates the inflection point trajectory dispersion value, determines the distribution characteristics of the trajectory inflection points according to the dispersion, and obtains a set of inflection point trajectory distribution characteristics; The path alignment consistency comparison sub-module extracts the target path inflection point data according to the set of inflection point trajectory distribution characteristics, analyzes the matching inflection point number ratio and the trajectory shape consistency ratio, screens the paths meeting the alignment rate and consistency requirements, and obtains a list of disturbance behavior patterns.

7. The radioactive pneumonia prediction system after radiotherapy for chest tumors according to claim 1, characterized in that, The system further includes a risk trend output module: The risk trend output module extracts the lung segment numbers and time periods where they are located according to the matching degree between the paths in the list of disturbance behavior patterns and the risk center trajectory, marks the disturbance trend synchronization intervals, classifies and locates the key warning events in the evolution process of the lung segments, and obtains a set of radiation pneumonia evolution risk signals; The set of radiation pneumonia evolution risk signals includes risk lung segment numbers, potential risk time periods, and trend synchronization feature identifiers.

8. The radioactive pneumonia prediction system after chest tumor radiotherapy according to claim 7, characterized in that The risk trend output module includes: The path screening sub-module screens the paths that meet the disturbance proximity threshold according to the spatio-temporal proximity between the paths in the list of disturbance behavior patterns and the post-radiotherapy lung risk center trajectory of the patient, extracts the lung segment numbers and corresponding time periods where the paths are located, and generates a set of post-radiotherapy risk path interval values; The synchronization marking sub-module calls the lung segment numbers and time periods in the set of post-radiotherapy risk path interval values, and marks the time sections with synchronized disturbances according to the consistency of the change in the disturbance trend of the lung segments in the path time series arrangement, and obtains the lung segment disturbance trend synchronization interval section values; The key event positioning sub-module synchronizes the lung segment numbers and time periods in the interval segment values according to the lung segment perturbation trend, extracts the ratio of the perturbation amplitude change rate to the duration, and screens the lung segment sequences exceeding the evolution mutation threshold, locates the corresponding time points and lung segment numbers, and obtains the radioactive pneumonia evolution risk signal set.

9. A method for predicting radiation pneumonitis after radiotherapy for chest tumors, characterized in that, The method is used to implement the execution of the radioactive pneumonia prediction system according to any one of claims 1-8, and includes the following steps: S1: Based on the lung texture boundary map and the lung region perfusion difference map of an individual after chest tumor radiotherapy, arrange the voxel gray ratio in chronological order, extract the continuous gray difference, screen the voxels with the difference greater than the lung function evolution threshold, and superimpose the radiotherapy dose distribution map for regional screening to obtain the functional creep intersection region map; S2: Based on the functional creep intersection region map, identify the local gray range in the grid and the gray difference between the previous and subsequent time points, screen the grid coordinates with the gray direction reversed and the amplitude change exceeding the texture change threshold, and obtain the texture mutation marker set; S3: Based on the lung segment numbers corresponding to the texture mutation marker set, extract the ventilation path and perfusion path at multiple time points, perform sequence alignment on the path rate change values, locate the rate fluctuation mutation points, and establish the lung segment ventilation-perfusion perturbation sequence map; S4: Based on the lung segment ventilation-perfusion perturbation sequence map, extract the high-frequency perturbation sections and the corresponding perturbation time periods, cluster the perturbation trajectory nodes, analyze the trajectory turning order and distribution consistency, screen the concentrated areas of perturbation behavior, and obtain the radioactive change behavior clustering map; S5: Based on the comparison results between the perturbation trajectories in the radioactive change behavior clustering map and the risk evolution trajectory library, screen the lung segments and time periods with overlapping trajectories, mark the perturbation and risk synchronization intervals, and obtain the radioactive pneumonia evolution risk signal set.

Images