An AI-based lung perfusion assessment system

Abstract

The present invention relates to the field of lung perfusion technology, and specifically to an AI-based lung perfusion assessment system. The system includes an image quality screening module, a blood vessel positioning analysis module, a blood flow velocity measurement module, a blood vessel abnormality analysis module, and an abnormal response module. The image quality screening module monitors the pixel density and the amplitude of the color gradient change of the image. The present invention optimizes the image selection process through real-time image clarity evaluation and screening, ensuring that all images used for analysis meet high-quality standards, automatically calculates the grayscale changes and edge contours of blood vessels, accurately locates and marks potential lesion areas, and provides more detailed vascular structure analysis. The system can also quickly analyze the velocity deviation of blood vessel segments, promptly identify blood flow abnormalities, and effectively improve the judgment speed of diseases such as pulmonary embolism or pulmonary hypertension. Automated abnormality detection and recording enhances the efficiency of responding to emergency medical situations, improves the accuracy and speed of judgment, and supports more effective clinical decision-making.

Description

Technical Field

[0001] The present invention relates to the technical field of lung perfusion, and in particular to an AI-based lung perfusion assessment system. Background Art

[0002] Pulmonary perfusion technology mainly involves evaluating and measuring the fluidity of blood in the lungs, which is crucial for the diagnosis and management of various lung diseases, especially in cases such as pulmonary embolism, pulmonary hypertension, chronic obstructive pulmonary disease (COPD), etc. Through pulmonary perfusion assessment, doctors can determine the distribution of blood in the lungs to evaluate abnormalities in pulmonary blood flow. With the advancement of technology, MRI and CT technologies are also widely used to non-invasively evaluate pulmonary perfusion.

[0003] Among them, the pulmonary perfusion assessment system refers to the use of artificial intelligence technology to enhance or optimize the pulmonary perfusion assessment process, and automatically analyze medical imaging data through algorithms to identify pulmonary blood flow problems faster and more accurately. Artificial intelligence can process large amounts of data and recognize patterns, which is particularly important for identifying potential pulmonary blood flow abnormalities in the early stages. The main uses of the system include improving diagnostic accuracy, speeding up the diagnostic process, and providing decision support in clinical practice.

[0004] Existing technologies fail to provide sufficiently rapid responses in terms of quick judgment and immediate decision support, which affects the timeliness of treatment. They rely on complex analytical processes and are easily affected by the operator's experience and judgment criteria, increasing the risk of misdiagnosis or missed diagnosis. Especially in the early stages of disease judgment, when rapid response is required to emergency situations such as acute pulmonary embolism, the response speed of existing technologies is insufficient to provide timely judgments, thereby delaying critical treatment opportunities. Existing technologies also show limitations in adapting to complex cases, making it difficult to meet evolving clinical needs, limiting their application effectiveness in complex medical environments. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and propose an AI-based lung perfusion assessment system.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an AI-based lung perfusion assessment system, the system comprising:

[0007] The image quality screening module monitors the pixel density and color gradient changes of the collected lung continuous image sequence data to determine whether the image meets the image quality standards. If it exceeds the standard, it will be included in the processing sequence and archived. Otherwise, the current frame image will be discarded and the position will be recorded to obtain a qualified image sequence.

[0008] The vascular positioning analysis module analyzes the grayscale distribution and edge contours of the blood vessels in the image based on the qualified image sequence, calculates the grayscale change amplitude and contour closure of each blood vessel in the image, determines the position of the blood vessels and potential lesion areas, and obtains vascular structure positioning data;

[0009] The blood flow velocity measurement module extracts the position identifier of each blood vessel segment based on the blood vessel structure positioning data, calculates the pixel displacement of the blood vessel segment, analyzes the relationship between the displacement data and the time interval, estimates the blood flow velocity, records the velocity deviation of each blood vessel segment, and obtains blood flow rate difference information;

[0010] The vascular abnormality analysis module identifies blood vessels whose velocity changes exceed the normal fluctuation range based on the blood flow velocity difference information, analyzes the diameter changes and velocity differences of the blood vessels, determines the ratio of the blood vessels with abnormal changes to the total number of blood vessels, and obtains abnormal blood vessel indicators.

[0011] The present invention has the following improvements: the qualified image sequence includes image number, storage timestamp, and quality rating; the vascular structure positioning data includes edge recognition index, color grading, and structural complexity; the blood flow rate difference information includes rate extreme value, average rate difference, and rate consistency score; and the abnormal vascular index includes abnormality measurement, variation frequency, and healthy vascular comparison index.

[0012] The present invention is improved in that the image quality screening module includes:

[0013] The pixel density extraction submodule analyzes the width of the grayscale distribution and the change in the center grayscale in each frame based on the collected continuous lung image sequence data. By evaluating the relationship between the grayscale difference and the average value, it identifies the pixel-dense areas and edge areas in the image and generates the pixel density change.

[0014] The color gradient analysis submodule uses the pixel density variation to analyze the grayscale difference between adjacent pixels in each frame of the image, identifies the color gradient in each direction within the image frame, and obtains the color gradient variation amplitude;

[0015] The quality discrimination and screening submodule evaluates the clarity of each frame of image according to the amplitude of the color gradient change, compares it with the clarity benchmark, screens image frames that meet the standards, and records the positions of unqualified image frames to obtain a qualified image sequence.

[0016] The present invention is improved in that the blood vessel positioning analysis module includes:

[0017] The grayscale change extraction submodule extracts consecutive frames from each vascular region of the image based on the qualified image sequence, detects the grayscale difference between the frames and the variation amplitude in the spatial position, analyzes the grayscale variation range between adjacent pixels, and calculates the average grayscale variation amplitude and standard deviation of the pixels in the region to obtain the grayscale variation amplitude mean;

[0018] The contour structure analysis submodule calls the grayscale change amplitude mean value to obtain the spatial coordinate points of the blood vessel edge, calculates the area to perimeter ratio of each closed path, evaluates the density and distribution uniformity of the points in the closed path, and obtains the contour closure index;

[0019] The lesion area recognition submodule selects image segments with drastic grayscale changes and dense edge structures according to the contour closure index, calculates the abnormal grayscale edge density, and then color-codes the area to obtain vascular structure positioning data.

[0020] The present invention is improved in that the blood flow velocity measurement module includes:

[0021] The displacement extraction submodule detects the position identifier of each blood vessel segment based on the blood vessel structure positioning data, compares the position change of the same blood vessel segment in the difference time frame image data, and measures the pixel displacement data;

[0022] The velocity calculation submodule converts the pixel distance into a physical distance based on the pixel displacement data and the inter-frame time difference data, calculates the displacement rate of each blood vessel segment, and obtains an estimated blood flow velocity;

[0023] The velocity deviation comparison submodule calls the blood flow velocity estimation value, compares it with the standard velocity range of the pulmonary microcirculation, identifies the blood vessel segments that deviate from the standard range, and records the velocity deviation of each blood vessel segment to obtain blood flow rate difference information.

[0024] The present invention is improved in that the vascular abnormality analysis module includes:

[0025] The velocity fluctuation detection submodule obtains the velocity time series of each vascular segment based on the blood flow rate difference information, compares the velocity changes at adjacent time points with historical data, defines the normal fluctuation range, identifies and marks the vascular segments where the velocity changes exceed the normal range, and obtains the velocity abnormality recognition rate;

[0026] The diameter change calculation submodule extracts continuous image frames of the corresponding blood vessel segment based on the velocity anomaly recognition rate, analyzes the edge position of the blood vessel in the image, identifies the variation range of the position in the continuous frames, and calculates the blood vessel diameter change rate;

[0027] The synchronous variation judgment submodule extracts the time series of diameter and velocity from the abnormal blood vessel segment based on the blood vessel diameter change rate to obtain the abnormal blood vessel index.

[0028] The present invention is improved in that the system further comprises:

[0029] The abnormal response module evaluates whether the microcirculation warning standard is exceeded based on the abnormal vascular index. If exceeded, it records the number and image location of the abnormal blood vessel, sends an abnormal warning message to medical personnel, and records the response time and blood vessel segment label to obtain an abnormal notification log;

[0030] The abnormality notification log includes warning level, notification time point, and emergency response instructions.

[0031] The present invention is improved in that the abnormal response module includes:

[0032] The index judgment submodule collects blood flow velocity, vessel diameter change and tissue perfusion rate data based on the abnormal vascular index, and checks whether it exceeds the standard according to the microcirculation warning standard. If it exceeds the standard, it is recorded as abnormal, and a record of abnormal index exceeding the standard is obtained;

[0033] The abnormal marking submodule calls the abnormal index exceeding standard record, marks the abnormal blood vessel, records its position in the image, and associates the abnormal blood vessel number with its image coordinates to obtain the abnormal blood vessel positioning index;

[0034] The information notification submodule sends abnormal warning information to medical personnel based on the abnormal blood vessel positioning index, records the response time and blood vessel segment label, and performs sorting processing to obtain an abnormal notification log.

[0035] Compared with the prior art, the advantages and positive effects of the present invention are:

[0036] In the present invention, through real-time image clarity evaluation and screening, the image selection process is optimized to ensure that all images used for analysis meet high quality standards, automatically calculate the grayscale changes and edge contours of blood vessels, accurately locate and mark potential lesion areas, and provide more detailed vascular structure analysis. The system can also quickly analyze the velocity deviation of vascular segments, promptly identify blood flow abnormalities, and effectively improve the judgment speed of diseases such as pulmonary embolism or pulmonary hypertension. Automated abnormality detection and recording enhances the efficiency of responding to emergency medical situations, improves the accuracy and speed of judgment, and supports more effective clinical decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is a system flow chart of the present invention;

[0038] Figure 2 This is a flowchart of the image quality screening module in the present invention;

[0039] Figure 3 This is a flow chart of the blood vessel positioning analysis module in the present invention;

[0040] Figure 4 This is a flow chart of the blood flow velocity measurement module in the present invention;

[0041] Figure 5 This is a flow chart of the vascular abnormality analysis module of the present invention;

[0042] Figure 6 This is a flow chart of the abnormal response module in the present invention. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, 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 intended to limit the present invention.

[0044] 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", etc., indicating directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings and are only for the convenience of describing the present invention and simplifying the description. They do not indicate or imply that the devices or elements referred to must have a specific direction, be constructed and operate in a specific direction, and therefore should not be understood as limiting the present invention. In addition, in the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0045] Example

[0046] See also Figure 1 The present invention provides a technical solution: an AI-based lung perfusion assessment system comprising:

[0047] The image quality screening module monitors the pixel density and color gradient changes of the collected lung continuous image sequence data, and conducts real-time evaluation of the clarity of each frame to determine whether it meets the image quality standard. If it exceeds the standard, it will be included in the processing sequence and filed. Otherwise, the current frame will be discarded and the position will be recorded to obtain a qualified image sequence.

[0048] The vascular positioning analysis module analyzes the grayscale distribution and edge contours of blood vessels in the image based on qualified image sequences, calculates the grayscale change amplitude and contour closure of each blood vessel in the image, determines the location of the blood vessels and potential lesion areas, and color-codes areas with frequent grayscale changes and dense structures to obtain vascular structure positioning data;

[0049] The blood flow velocity measurement module extracts the location identifier of each vascular segment based on the vascular structure positioning data, calculates the pixel displacement of the vascular segment, analyzes the relationship between the displacement data and the time interval, estimates the blood flow velocity, and compares the measured data with the standard velocity range of the pulmonary microcirculation. It records the velocity deviation of each vascular segment and obtains the blood flow rate difference information;

[0050] The vascular anomaly analysis module identifies blood vessels whose velocity changes exceed the normal fluctuation range based on blood flow rate difference information, analyzes the diameter changes and velocity differences of the blood vessels, determines whether there are synchronous variations or structural abnormalities, and determines the ratio of abnormally changing blood vessels to the total number of blood vessels to obtain abnormal blood vessel indicators;

[0051] The abnormal response module evaluates whether the microcirculation warning standard is exceeded based on abnormal vascular indicators. If exceeded, it records the number and image location of the abnormal blood vessel, sends an abnormal warning message to medical personnel, and records the response time and blood vessel segment label to obtain an abnormal notification log.

[0052] Qualified image sequences include image numbers, storage timestamps, and quality ratings; vascular structure positioning data include edge recognition indicators, color grading, and structural complexity; blood flow rate difference information includes rate extremes, average rate differences, and rate consistency scores; abnormal vascular indicators include abnormality metrics, variation frequencies, and healthy vascular control indexes; and abnormal notification logs include warning levels, notification time points, and emergency response instructions.

[0053] Image number acquisition method in qualified image sequence: After each frame of image passes through the image quality screening module and is judged to be qualified, the system automatically generates a unique number (such as according to the acquisition order, timestamp, or hash value);

[0054] Storage timestamp acquisition method: During image acquisition and archiving, the system records the specific time (year / month / day hour:minute:second) when the image frame is saved into the database. Function: Assists in image traceability and sequence analysis, and facilitates synchronization with physiological signals and device parameters.

[0055] Quality rating acquisition method: The image quality screening module automatically assesses the clarity, contrast, noise level, etc. of each frame based on algorithms such as pixel density and color gradient, and assigns a score or rating (such as A / B / C or 1-10 points). Its function is to select the optimal sequence, perform subsequent weighted processing, or mark abnormal data.

[0056] Method for obtaining edge recognition indicators in vascular structure positioning data: The vascular positioning analysis module extracts vascular contours using edge detection algorithms (such as Canny, Sobel, or deep learning-based segmentation networks) and outputs quantitative indicators such as edge continuity and closure. Its function is to evaluate the accuracy and completeness of vascular segmentation and support subsequent structural and functional analysis.

[0057] Color grading acquisition method: Based on the grayscale or color information of the vascular area, the system grades the blood vessels in color scale or brightness (e.g., light gray, medium gray, dark gray). This can be combined with color space conversion (RGB, HSV, Lab, etc.). This helps determine vascular perfusion and analyze the color feature differences between diseased and healthy areas.

[0058] Structural complexity is obtained by analyzing the number of vascular branches, bifurcation angles, length-width ratios, fractal dimensions, etc. to quantify the geometric complexity of the vascular network. Function: It is used to characterize regional vascular distribution and assist in identifying structural abnormalities and pathological changes.

[0059] The blood flow velocity measurement module extracts the maximum and minimum velocity values ​​from the blood flow velocity difference information after analyzing the velocity of all blood vessel segments, marking the extreme blood flow velocity points. This function is to detect local high- or low-velocity blood vessels, indicating possible abnormalities such as embolism and stenosis.

[0060] Mean velocity difference is obtained by calculating the average velocity of all vascular segments, then calculating the difference between each segment and the average velocity, and then giving the standard deviation or variability of the overall velocity distribution. Its purpose is to comprehensively assess the uniformity of pulmonary blood perfusion and the relative location of abnormal areas.

[0061] Rate consistency score is obtained by combining the discreteness and continuity of global and local vascular velocities and scoring using mathematical models (such as consistency index and coefficient of variation). Function: It reflects the stability of the perfusion process and provides a basis for quantitative evaluation.

[0062] See also Figure 2 , the image quality screening module includes:

[0063] The pixel density extraction submodule analyzes the width of the grayscale distribution and the change in the center grayscale in each frame based on the collected continuous lung image sequence data. By evaluating the relationship between the grayscale difference and the average value, it identifies the pixel-dense areas and edge areas in the image and generates the pixel density change.

[0064] Based on the continuous image sequence of the lungs, for each frame of the image, the image data is first read as a two-dimensional grayscale matrix, and the distribution number of pixel grayscales is counted by row and column respectively, and the horizontal and vertical grayscale curves of the image are constructed. The maximum and minimum grayscale values ​​in the curves are then extracted and the difference is calculated to determine the width of the grayscale distribution. Subsequently, the grayscale average value of all pixels in the image is extracted as the center reference, and the pixels with grayscale values ​​greater than the average value in the image rows and columns are scanned. The grayscale median and grayscale dispersion of the concentrated area are counted to determine whether the center grayscale is offset. If the deviation between the average value and the ideal center grayscale exceeds the set tolerance, it is considered a center-shifted image. After that, the grayscale of each pixel in the image is traversed pixel by pixel. The pixel points with grayscale values ​​significantly higher or lower than the average value are marked as high-difference points, and the proportion of high-difference points in each local area of ​​the image is calculated. When the proportion exceeds a specific density threshold, the area is marked as a pixel-dense area, otherwise it is marked as an edge area. In a specific example, if the average grayscale value of an image deviates from the center grayscale value by seven grayscale levels, and there are more than 30 high-difference points in every 100 pixels in a certain area of ​​the image, then the area is directly judged as pixel-dense. After the above extraction and judgment process, all pixel-dense areas and edge areas form a spatial distribution of pixel density in the entire frame image, and then the degree of change of such distribution between frames is calculated to obtain the pixel density change.

[0065] The color gradient analysis submodule uses the pixel density change to analyze the grayscale difference between adjacent pixels in each frame of the image, identify the color gradient in each direction within the image frame, and obtain the color gradient change amplitude;

[0066] The grayscale difference between adjacent pixels in the image frame is calculated sequentially. A directional traversal is performed in the up-down, left-right, and diagonal directions of the image, and the grayscale value differences of adjacent pixels are compared. If the grayscale difference of a pair of pixels exceeds a set threshold, it is recorded as a mutation point. The proportion of mutation points in each direction to the total pixel pairs is then counted as a representation of the degree of color scale mutation in that direction. The one with the largest ratio among the four directions is used as an indicator of the overall color scale gradient change amplitude. The image is further divided into several local windows, and the grayscale difference in each window is counted. The maximum difference and average difference in each local area are obtained, and the ratio between the two is calculated to determine whether the local color scale distribution is uniform. If the ratio of a region exceeds the set ratio threshold, it is marked as a region with drastic color scale change. If the overall maximum directional mutation ratio of the image is large and the local uniformity is poor, the color scale gradient change of the frame is considered to be significant. In actual applications, in a certain frame of image, the proportion of mutation points in the horizontal direction is 25%, and the proportion in other directions is about 15%. The local color scale non-uniform area accounts for 40% of the total image area, which indicates that the color scale gradient change amplitude of the frame is significant.

[0067] The quality judgment and screening submodule evaluates the clarity of each frame of the image based on the amplitude of the color gradient change, compares it with the clarity benchmark, screens the image frames that meet the standards, and records the positions of the unqualified image frames to obtain a qualified image sequence;

[0068] Image clarity is assessed based on the aforementioned color gradient change amplitude, and this value is compared with a benchmark value obtained in advance by statistically analyzing a large number of clear images. The benchmark value is set to a maximum directional mutation ratio of no less than 18%, and a local color gradient change area ratio of no more than 20%. In the discrimination process, the maximum mutation direction ratio of the current image frame is first extracted and compared with the first clarity standard. If it is less than the threshold, the image is directly marked as unqualified. The proportion of the local color gradient area in the image is then evaluated. If this value exceeds the specified upper limit, it is also screened out. In the example application, the maximum mutation direction of a frame is 16%, and the local gradient area ratio is 35%. Both of these values ​​fail to meet the standards, so the image quality is determined to be insufficient. The frame number of the image in the original image sequence is recorded and marked as an unqualified frame for subsequent verification. After the clarity assessment and screening of all images are completed frame by frame, the image frames that meet the conditions are sorted into a qualified image sequence in chronological order for subsequent vascular analysis and processing.

[0069] See also Figure 3 , the vascular positioning analysis module includes:

[0070] The grayscale change extraction submodule extracts consecutive frames from each vascular region of the image based on a qualified image sequence, detects the grayscale difference between frames and its variation in spatial position, analyzes the grayscale variation range between adjacent pixels, and calculates the average grayscale variation amplitude and standard deviation of the pixels in the region to obtain the mean grayscale variation amplitude.

[0071] Identify the position and range of the vascular area in each frame of the image, and continuously extract the same vascular area in adjacent frames. First, align each vascular area in consecutive frames according to the consistency of pixel coordinates to ensure the spatial correspondence of the compared objects. Then read the pixel grayscale values ​​in the vascular area in adjacent frames respectively, calculate the grayscale difference between the two frames for the same pixel coordinate position, record all the differences and count their maximum, minimum and distribution range to determine the range of grayscale changes between frames, and calculate the average value of all grayscale differences as the mean index of the overall grayscale change of the vascular area. Then perform statistical distribution processing on the grayscale differences to obtain The degree of discreteness reflects the stability of the change. If a certain blood vessel area contains 200 pixels in image frames A and B, and the grayscale values ​​of 140 of these pixels in frame B are increased by 15 to 25 grayscale levels compared with frame A, the average grayscale change amplitude is 20. If most of the changes are concentrated in the range of 18 to 22, and the standard deviation is 2.5, it means that the grayscale change in this area is relatively concentrated and stable. After detecting multiple blood vessel areas, the average change amplitude and standard deviation of each area can be unified into a set of grayscale change indicators corresponding to the image frame, and the average grayscale change amplitude of the image frame as a whole can be calculated as the input data of subsequent modules.

[0072] The contour structure analysis submodule uses the mean grayscale change amplitude to obtain the spatial coordinate points of the blood vessel edge, calculates the area-to-perimeter ratio of each closed path, evaluates the density and distribution uniformity of the points in the closed path, and obtains the contour closure index;

[0073] The mean value of grayscale variation is called and used as a reference standard for evaluating edge stability and boundary clarity. The edge detection method is used to extract the coordinates of edge pixels in each vascular area, and a closed path is formed based on the coordinate connection. The area value of the pixel area enclosed by each closed path and the perimeter value represented by the total number of pixels in the path are then calculated. The ratio is taken to reflect the compactness of the contour. The number of edge points on the path is counted and distributed on a two-dimensional coordinate map into multiple sector blocks. The number of points in each block is compared, and the difference between the maximum and minimum values ​​is calculated to determine whether the distribution is uniform. If the difference accounts for more than 2 of the total number of points, the area value is calculated. If the edge points are 0%, it is considered unevenly distributed. For example, if a blood vessel contour path consists of 85 edge points, the area is 360 square pixels, and the perimeter ratio is 4.24, after being divided into 8 sector blocks, the number of edge points in each block is 9, 12, 10, 8, 11, 10, 13, and 12, respectively, the maximum is 13, the minimum is 8, the difference is 5, and the proportion is 5.88%, then the edge points are judged to be evenly distributed. If the difference between the maximum and minimum reaches 18, the proportion is 21%, and the closed path is marked as unevenly distributed. Finally, the compactness and edge point distribution of all blood vessel paths are summarized to form the contour closure index corresponding to each area.

[0074] The lesion area recognition submodule selects image segments with drastic grayscale changes and dense edge structures based on the contour closure index, using the formula:

[0075] ;

[0076] Calculate the density of abnormal grayscale edges , used to evaluate the feature density of potential lesion areas in the image area, and then color-code the area to obtain vascular structure positioning data, where, It is The gray value of a pixel, is the grayscale mean of the pixels in the current area, It is The pixel edge segment length refers to the actual length of the edge segment corresponding to the pixel point on the edge contour of the blood vessel. is the average length of edge segments within the region, It is The gradient intensity of the edge point corresponding to a pixel indicates the intensity of the grayscale change at that point. It is The number of structural points in the local area where the pixel is located. The structural points reflect the complexity of the blood vessels or other tissue structures in the local area. is the total number of pixels in the image area;

[0077] The area containing continuous vascular structures in the image is selected as the analysis object, and the area is divided into several pixel groups. The grayscale value of the pixels in each group is extracted. , edge segment length , edge gradient intensity and the number of local structural points First, based on all Calculate the grayscale mean , then Calculate its average length , in order to construct the difference distribution term , and the gradient strength of each pixel After the joint product is accumulated, the numerator is obtained, and at the same time, all pixels Add together to form the denominator value, and further substitute it into the formula, let the total number of pixels in the area , and their grayscale values ​​are , and the edge lengths are , the edge gradient strength is , the number of structural points is , first calculate the mean:

[0078] ;

[0079] ;

[0080] Then the numerator is calculated as follows:

[0081] Item 1: ;

[0082] Item 2: ;

[0083] Item 3: ;

[0084] The sum of the numerators is ;

[0085] The denominator is calculated as follows:

[0086] Item 1: ;

[0087] Item 2: ;

[0088] Item 3: ;

[0089] The sum of the denominators is ;

[0090] Substituting into the formula we get: ;

[0091] The calculation results show that the abnormal grayscale edge density of the current area is 11.22. If the lesion area recognition benchmark value is set to 10 in the system, the current area is judged as a structure-dense area and is assigned color coding information to generate vascular structure positioning data.

[0092] See also Figure 4 , blood flow velocity measurement module includes:

[0093] The displacement extraction submodule detects the position marker of each vascular segment based on the vascular structure positioning data, compares the position changes of the same vascular segment in the difference time frame image data, and measures the pixel displacement data;

[0094] The spatial position label of each vascular segment when it is first identified in the image sequence is extracted. The position label consists of three items: the center coordinates of the vascular segment's contour, the direction of the vascular direction, and the range of the vascular width. Each item is accurately recorded in pixel units. In the subsequent processing, the area corresponding to the vascular segment is retrieved from the difference time frame, and the center coordinates and boundary information are extracted again from the images at different time points. The change in the position of the vascular segment in the current frame compared to the previous frame is calculated by point-by-point matching. When judging the displacement of the vascular segment, the Euclidean distance of the center point coordinates of each segment is used as the change value. If the time difference between frames is fixed at 40 milliseconds, then the Pixel distance changes can be used as a spatial reference for blood flow movement. The same information needs to be repeatedly extracted for each blood vessel segment between multiple consecutive frames, and multiple displacement change values ​​​​are recorded in chronological order. If the center position of a blood vessel segment in frame 1 is 45 pixels horizontally and 60 pixels vertically, and in frame 2 it is 49 pixels horizontally and 63 pixels vertically, then the pixel displacement is 5 units. Through the image resolution setting, it can be converted into actual physical distance. For example, if the pixel resolution is 0.1 mm, the actual displacement of the blood vessel segment is 0.5 mm. After repeating this operation between multiple frames, each blood vessel segment obtains a corresponding displacement trajectory data set, which serves as the basic result for subsequent blood flow velocity calculations.

[0095] The velocity calculation submodule converts pixel distance into physical distance based on pixel displacement data and combines it with inter-frame time difference data to calculate the displacement rate of each blood vessel segment using the formula:

[0096] ;

[0097] Get blood flow velocity estimates, where Representative The estimated blood flow velocity of a segmented vessel is the velocity of blood flow in the vessel segment, which is calculated based on the position change and time difference. Indicates the The pixel displacement value of a blood vessel segment is the number of pixels corresponding to the position change of the blood vessel segment between consecutive frames. is the conversion factor from pixel to actual distance, used to convert pixel displacement into actual physical distance. For the The inter-frame time difference of the segmented blood vessel segment represents the time interval between adjacent frames in the image sequence;

[0098] Extract the time interval between each two frames based on the metadata of the image frame sequence. If the image acquisition frame rate is frame, the time difference between two adjacent frames is for , and then introduce the conversion factor between pixels and actual physical distance , which is usually calibrated by the image system when it is set, for example, ,Right now , then targeting the vascular segment Call their pixel displacement values ​​respectively, it is known Pixel displacement of the segment , Duan Wei , Duan Wei , substitute the above parameters into the formula:

[0099] for part,

[0100] ;

[0101] for part,

[0102] ;

[0103] for part,

[0104] ;

[0105] This result shows the actual blood flow movement distance of the blood vessel segment per unit time. By combining the pixel displacement between image frames with the physical scale and time interval, it is converted into a physically meaningful flow velocity index, whose unit is millimeters per second. The larger the value, the faster the blood flow speed in the blood vessel segment.

[0106] The velocity deviation comparison submodule calls the blood flow velocity estimate and compares it with the standard velocity range of the pulmonary microcirculation, identifies the vascular segments that deviate from the standard range, and records the velocity deviation of each vascular segment to obtain blood flow rate difference information;

[0107] The estimated blood flow velocity value is called and the estimated velocity data of each blood vessel segment is compared with the standard velocity range of the pulmonary microcirculation. First, the functional classification of different blood vessel segments is performed, such as classifying segment z1 as the venous segment and segment z2 and z3 as the arterial segment. According to medical literature, the standard interval of blood flow velocity in the venous segment is , the arterial segment is During the comparison process, the system calls the estimated velocity value of segment z1, which is 7.5 mm / s, and determines that it is within the standard interval of the venous segment and is determined to be normal flow velocity. The velocity of segment z2 is 12.5 mm / s, which is within the standard interval of the arterial segment and is also determined to be normal. However, the velocity of segment z3 is 18.75 mm / s. Although it does not exceed the standard interval, it is close to the upper limit and is the focus of subsequent monitoring. The recording rule for the deviation is set as follows: if the absolute value of the difference between the blood flow velocity estimate and the upper or lower limit of the interval exceeds 1.0 mm / s, it is considered a segment with significant deviation and the velocity deviation needs to be recorded. For example, the difference between the z3 segment and the upper limit of 20.0 mm / s is 1.25 mm / s, which meets the deviation recording condition. The deviation is recorded as +1.25 mm / s, and the segment number, the correspondence between the velocity estimate and the standard interval are retained as structured data for subsequent analysis. Further, in the output, the numbers of all vascular segments that meet the deviation screening conditions and the corresponding offset value intervals are summarized to obtain blood flow rate difference information.

[0108] See also Figure 5 , the vascular abnormality analysis module includes:

[0109] The velocity fluctuation detection submodule obtains the velocity time series of each vascular segment based on blood flow rate difference information, compares the velocity changes at adjacent time points with historical data, defines the normal fluctuation range, identifies and marks vascular segments with velocity changes outside the normal range, and obtains the velocity anomaly recognition rate;

[0110] Extract the velocity value of each vascular segment in the entire time series, establish a velocity sequence curve arranged by time frame, calculate the difference in velocity values ​​between adjacent time points, and compare the difference with the average velocity change of the vascular segment in the past three to five frames. If the difference between the current change value and the historical average change value exceeds the set fluctuation threshold, it is considered an abnormal fluctuation event. The fluctuation threshold is set according to the average velocity difference of the stable vascular segments in the overall sample and its standard deviation. If the velocity of a vascular segment at time points T1 to T4 is 0.8, 0.83, 0.85, and 0.84, respectively , then its historical average change is 0.02. If the speed rises to 1.1 at T5, the change is 0.26, which is 0.24 different from the historical average. This value is greater than the set fluctuation threshold of 0.1, then the segment is marked as speed abnormal at T5. During the judgment process, all blood vessels need to be checked segment by segment and scanned frame by frame to establish a complete fluctuation behavior record for each blood vessel segment. The number of blood vessel segments marked as abnormal is divided by the total number of detected blood vessel segments to be used as the speed abnormality recognition rate. This ratio is recorded by frame or region classification, and the speed abnormality recognition rate in the entire range is output in the final result.

[0111] The diameter change calculation submodule extracts continuous image frames of the corresponding blood vessel segment based on the velocity anomaly recognition rate, analyzes the edge position of the blood vessel in the image, identifies the amplitude of the position change in the continuous frames, and calculates the blood vessel diameter change rate;

[0112] Based on the recognition rate of velocity anomaly, the vascular segment judged as velocity anomaly is determined, and the image data of the vascular segment in the corresponding time frame and several frames before and after it are extracted from the original image sequence. The edge pixel coordinates of the blood vessel are extracted from each frame, and the edge contour is scanned in sequence in the horizontal or vertical direction. The pixel distance between the left and right edges of the blood vessel in each frame is measured as the current frame diameter value. The difference between the diameter values ​​of each frame is calculated in multiple frames of image, and the maximum change amplitude and the total change trend are obtained. The maximum change amplitude is then divided by the diameter value of the previous frame to calculate the vascular diameter change rate. If the change rate exceeds the set threshold, the segment is recorded as a diameter anomaly. Usually, the threshold setting is based on the diameter stability analysis of the blood vessel segments in normal sample data. Generally, the maximum diameter fluctuation rate does not exceed 10% as the normal range. If the diameter values ​​of a blood vessel segment in consecutive frames are 14, 14.2, 14.5, 15.3, and 14.9 pixels, respectively, and the corresponding change rates are 1.4%, 2.1%, 5.5%, and -2.6%, respectively, then the maximum change rate is 5.5%, which is lower than the set threshold of 10%, so it is not judged as abnormal. If a segment has a sudden change in diameter between frames, for example, from 12 pixels to 14 pixels, the corresponding change rate is 16.7%, and it is directly recorded as an abnormal segment.

[0113] The synchronous variation judgment submodule extracts the time series of diameter and velocity from abnormal blood vessel segments based on the rate of change of blood vessel diameter, using the formula:

[0114] ;

[0115] Obtain abnormal vascular indicators , represents the mean value of the structural change and velocity change of all abnormal vascular segments, where Indicates the Segment blood vessels in time Blood flow rate, data obtained by medical imaging technology, Indicates the Segment blood vessels in time The diameter of the blood vessels is obtained by medical imaging technology. is the total number of vessel segments marked as abnormal, and Respectively indicate time (i.e. the previous time point) blood flow rate and diameter of segmental vessels;

[0116] First, the abnormal segments marked by velocity abnormality recognition rate and vascular diameter change rate are used as indexes, and each abnormal vascular segment is analyzed at two consecutive time points. and The velocity and diameter data are extracted and matched. If the five blood vessels numbered 1 to 5 are marked as abnormal, the blood flow velocity and diameter are measured by ultrasound imaging and edge recognition respectively, and the following values ​​are obtained:

[0117] The first segment of the blood vessels The speed is 15.2mm / s and the diameter is 2.4mm. The moment speed is 16.0 mm / s and the diameter is 2.5 mm;

[0118] The second segment of the blood vessels The speed is 13.8mm / s and the diameter is 2.5mm. The moment speed is 14.5mm / s and the diameter is 2.4mm;

[0119] The third segment of the blood vessels The speed is 14.6mm / s and the diameter is 2.3mm. The moment speed is 15.2mm / s and the diameter is 2.6mm;

[0120] The 4th segment of the blood vessels The speed is 16.1mm / s and the diameter is 2.6mm. The moment speed is 15.8mm / s and the diameter is 2.7mm;

[0121] The 5th segment of the blood vessels The speed is 12.9mm / s and the diameter is 2.2mm. The velocity at that moment is 13.3 mm / s and the diameter is 2.1 mm. Then, for each blood vessel segment, the change in the product of velocity and diameter is calculated according to the formula and the absolute value is taken. The following calculation is performed one by one:

[0122] The first paragraph is ;

[0123] The second paragraph is ;

[0124] Paragraph 3 is ;

[0125] Paragraph 4 is ;

[0126] Paragraph 5 is ;

[0127] calculate :

[0128] ;

[0129] ;

[0130] The results show that It is 2.202, which reflects the intensity and amplitude of the linkage changes between structure (diameter) and dynamics (rate). The larger its absolute value, the stronger the synchronous variation in structure and flow velocity of the corresponding vascular segment, showing more severe instability or nonlinear coupling characteristics.

[0131] See also Figure 6 , the exception response module includes:

[0132] The indicator judgment submodule collects data on blood flow velocity, vessel diameter change, and tissue perfusion rate based on abnormal vascular indicators, and checks whether they exceed the standard according to the microcirculation warning standard. If they exceed the standard, it will be recorded as abnormal, and a record of abnormal indicator exceeding the standard will be obtained;

[0133] The velocity value sequence of each vascular segment marked as abnormal, the corresponding diameter change value in the continuous image, and the tissue perfusion rate data calculated according to the regional density distribution analysis results are extracted in turn. The blood flow velocity is compared with the set velocity fluctuation threshold. If the value exceeds the upper threshold or is lower than the lower threshold, the blood flow in this segment is directly marked as abnormal. The velocity judgment threshold is set to the range of 0.3 to 1.2 mm / s with reference to the physiological standard of pulmonary microcirculation. The diameter value extracted from different image frames of the vascular segment is further read, and the change rate between adjacent frames is calculated. If the change rate is greater than 10% of the diameter change warning value, it is judged as a diameter abnormality. Finally, the group of the area where the blood vessel is located is extracted. The tissue perfusion rate is calculated by the proportion of perfused pixels per unit area in the region. If the value is lower than 20%, it is recorded as insufficient perfusion. The above three types of data are then uniformly compared with the microcirculation warning standard. The standard defines that any two abnormalities are considered to be exceeding the standard blood vessels. If any two of the three items exceed the threshold standard, the system records the number of the blood vessel segment, the exceeded item and its specific value in the current image frame, and marks it as an abnormal indicator exceeding the standard in the final data set. For example, if the speed of a blood vessel segment is 1.5 mm / s, the diameter change rate is 18%, and the perfusion rate is 17%, then two of the three judgment conditions of this segment exceed the corresponding threshold, so it is marked as abnormal and added to the abnormal record queue.

[0134] The abnormal marking submodule calls the abnormal index exceeding standard record, marks the abnormal blood vessels, records their positions in the image, and associates the abnormal blood vessel numbers with their image coordinates to obtain the abnormal blood vessel location index;

[0135] The vascular segment number identified as abnormal and its corresponding image frame index position are retrieved. The retained vascular structure positioning data are then used to extract the pixel area of ​​the corresponding numbered vascular segment in the image. The center position coordinates, contour boundary points, and actual size in image resolution units of the area are further calculated. This spatial position information is then associated with the vascular segment number to form a complete spatial positioning index entry. During the annotation operation, the system overlays and marks the original image, marking all abnormal vascular segments with a uniform color stroke. The edge of the marked area is marked with a four-pixel red border, and the center of the marked area is superimposed with a numbered label. A coordinate comparison list is generated outside the image, recording the center coordinate value and frame number of each numbered vascular segment in the image. If a vascular segment in the image is numbered V017 and the corresponding coordinates are 100 pixels horizontally and 65 pixels vertically, it is marked as "V017@(100, 65)" and added to the positioning index. All index entries are arranged in ascending order of frame order and vascular number to form a vascular abnormality positioning index list.

[0136] The information notification submodule sends abnormal warning information to medical personnel based on the abnormal blood vessel positioning index, records the response time and blood vessel segment label, and performs sorting processing to obtain the abnormal notification log;

[0137] The number, image frame number, and image coordinate information of each abnormal vascular segment are read in sequence and encapsulated as an abnormal event entry. The abnormal warning mechanism is then triggered, and an alarm signal is sent to the bound medical response terminal. The signal content includes the vascular segment number, the index of the image frame where it is located, the center coordinate information, and the type of exceeded item, with a text description indicating the specific numerical range of the abnormal data exceeding the standard. After sending the warning, the system immediately records the response timestamp and binds it to the vascular segment number to form a response record table entry. If the vascular segment number is V023, its image coordinates are (88, 102), the abnormal type is speed exceeding the standard by 1.9 mm / s and diameter changing by 17%, and the sending time is 13:45:02, then the response record is "V023|(88, 102)|13:45:02|speed+diameter". All record entries are sorted in ascending order according to the response time and summarized to generate the abnormal notification log in the image analysis of the day.

[0138] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. An AI-based lung perfusion assessment system, characterized in that: The system comprises: The image quality screening module monitors the pixel density and color gradient changes of the collected lung continuous image sequence data to determine whether the image meets the image quality standards. If it exceeds the standard, it will be included in the processing sequence and archived. Otherwise, the current frame image will be discarded and the position will be recorded to obtain a qualified image sequence. The vascular positioning analysis module analyzes the grayscale distribution and edge contours of the blood vessels in the image based on the qualified image sequence, calculates the grayscale change amplitude and contour closure of each blood vessel in the image, determines the position of the blood vessels and potential lesion areas, and obtains vascular structure positioning data; The blood flow velocity measurement module extracts the position identifier of each blood vessel segment based on the blood vessel structure positioning data, calculates the pixel displacement of the blood vessel segment, analyzes the relationship between the displacement data and the time interval, estimates the blood flow velocity, records the velocity deviation of each blood vessel segment, and obtains blood flow rate difference information; The blood flow velocity measurement module includes: The displacement extraction submodule detects the position identifier of each blood vessel segment based on the blood vessel structure positioning data, compares the position change of the same blood vessel segment in the difference time frame image data, and measures the pixel displacement data; The velocity calculation submodule converts the pixel distance into a physical distance based on the pixel displacement data and the inter-frame time difference data, calculates the displacement rate of each blood vessel segment, and obtains an estimated blood flow velocity; The velocity deviation comparison submodule calls the blood flow velocity estimate, compares it with the standard velocity range of the pulmonary microcirculation, identifies the blood vessel segments that deviate from the standard range, and records the velocity deviation of each blood vessel segment to obtain blood flow rate difference information; The vascular abnormality analysis module identifies blood vessels whose velocity changes exceed the normal fluctuation range based on the blood flow velocity difference information, analyzes the diameter changes and velocity differences of the blood vessels, determines the ratio of the blood vessels with abnormal changes to the total number of blood vessels, and obtains an abnormal blood vessel index; The vascular abnormality analysis module includes: The velocity fluctuation detection submodule obtains the velocity time series of each vascular segment based on the blood flow rate difference information, compares the velocity changes at adjacent time points with historical data, defines the normal fluctuation range, identifies and marks the vascular segments where the velocity changes exceed the normal range, and obtains the velocity abnormality recognition rate; The diameter change calculation submodule extracts continuous image frames of the corresponding blood vessel segment based on the velocity anomaly recognition rate, analyzes the edge position of the blood vessel in the image, identifies the variation range of the position in the continuous frames, and calculates the blood vessel diameter change rate; The synchronous variation judgment submodule extracts the time series of diameter and velocity from the abnormal blood vessel segment based on the blood vessel diameter change rate to obtain the abnormal blood vessel index.

2. The AI-based lung perfusion assessment system according to claim 1, characterized in that: The qualified image sequence includes image number, storage timestamp, and quality rating; the vascular structure positioning data includes edge recognition index, color grading, and structural complexity; the blood flow rate difference information includes rate extremes, average rate differences, and rate consistency scores; and the abnormal vascular indicators include abnormality metrics, variation frequencies, and healthy vascular comparison indexes.

3. The AI-based lung perfusion assessment system according to claim 1, characterized in that: The image quality screening module includes: The pixel density extraction submodule analyzes the width of the grayscale distribution and the change in the center grayscale in each frame based on the collected continuous lung image sequence data. By evaluating the relationship between the grayscale difference and the average value, it identifies the pixel-dense areas and edge areas in the image and generates the pixel density change. The color gradient analysis submodule uses the pixel density variation to analyze the grayscale difference between adjacent pixels in each frame of the image, identifies the color gradient in each direction within the image frame, and obtains the color gradient variation amplitude; The quality discrimination and screening submodule evaluates the clarity of each frame of image according to the amplitude of the color gradient change, compares it with the clarity benchmark, screens image frames that meet the standards, and records the positions of unqualified image frames to obtain a qualified image sequence.

4. The AI-based lung perfusion assessment system according to claim 1, characterized in that: The blood vessel positioning analysis module includes: The grayscale change extraction submodule extracts consecutive frames from each vascular region of the image based on the qualified image sequence, detects the grayscale difference between the frames and the variation amplitude in the spatial position, analyzes the grayscale variation range between adjacent pixels, and calculates the average grayscale variation amplitude and standard deviation of the pixels in the region to obtain the grayscale variation amplitude mean; The contour structure analysis submodule calls the grayscale change amplitude mean value to obtain the spatial coordinate points of the blood vessel edge, calculates the area to perimeter ratio of each closed path, evaluates the density and distribution uniformity of the points in the closed path, and obtains the contour closure index; The lesion area recognition submodule selects image segments with drastic grayscale changes and dense edge structures according to the contour closure index, calculates the abnormal grayscale edge density, and then color-codes the area to obtain vascular structure positioning data.

5. The AI-based lung perfusion assessment system according to claim 1, characterized in that: The system further comprises: The abnormal response module evaluates whether the microcirculation warning standard is exceeded based on the abnormal vascular index. If exceeded, it records the number and image location of the abnormal blood vessel, sends an abnormal warning message to medical personnel, and records the response time and blood vessel segment label to obtain an abnormal notification log; The abnormality notification log includes warning level, notification time point, and emergency response instructions.

6. The AI-based lung perfusion assessment system according to claim 5, characterized in that: The abnormal response module includes: The index judgment submodule collects blood flow velocity, vessel diameter change and tissue perfusion rate data based on the abnormal vascular index, and checks whether it exceeds the standard according to the microcirculation warning standard. If it exceeds the standard, it is recorded as abnormal, and a record of abnormal index exceeding the standard is obtained; The abnormal marking submodule calls the abnormal index exceeding standard record, marks the abnormal blood vessel, records its position in the image, and associates the abnormal blood vessel number with its image coordinates to obtain the abnormal blood vessel positioning index; The information notification submodule sends abnormal warning information to medical personnel based on the abnormal blood vessel positioning index, records the response time and blood vessel segment label, and performs sorting processing to obtain an abnormal notification log.

Images