Bone density phantom leakage failure detection, early warning method and system

Abstract

The application discloses a bone density phantom leakage failure detection and early warning method and system, relates to the technical field of early warning systems, and specifically comprises the following steps: generating a bubble distribution map by CT image analysis and identification, calculating the distance from a voxel to a container wall and forming an edge aggregation index, generating an aggregation space map based on cluster algorithm analysis of bubble density, generating a core map by superimposed analysis and screening of abnormal aggregation clusters, calculating a leakage index by combining a risk evaluation index method, marking a failure area when the leakage index exceeds a threshold value, and generating an early warning information table, thereby achieving accurate identification and risk early warning of bone density phantom leakage. In the application, bubbles are distinguished and a spatial distribution map is generated by depth analysis of a scanning image, a distance is calculated to form an edge aggregation index to reveal potential hidden dangers, high-risk aggregation clusters are distinguished by combining cluster analysis, the risk of a leakage channel is highlighted by superimposed identification of quantity and continuity, a risk index is combined for quantitative evaluation and early warning information is output, and the improvement from signal collection to spatial cognition is realized.

Description

Technical Field

[0001] This invention relates to the field of early warning system technology, and in particular to a method and system for detecting and warning of bone density phantom leakage failure. Background Technology

[0002] The field of early warning systems encompasses the monitoring and anomaly identification of the operational status of medical testing equipment. Its core lies in establishing monitoring methods for the stability parameters of the tested object and its operating environment, forming a technical path capable of sensing the status of critical components of the equipment, thereby identifying and alerting to potential failure risks. It covers aspects such as the acquisition and analysis of detection signals, the setting and identification of monitoring parameters, and the generation and transmission of fault warning information, constructing a systematic technical framework centered on detection and early warning.

[0003] The bone density testing phantom leakage failure detection and early warning method and system addresses potential leakage problems that may occur during the use of bone density testing phantoms. It involves designing monitoring methods for leakage characteristic parameters, combining these with methods for testing the sealing performance of the phantom structure, establishing criteria for judging leakage status, and outputting information through an early warning mechanism. The system monitors leakage paths on the phantom surface and internally, identifies abnormal detection signals caused by leakage, and detects and judges critical leakage states. The methods employed include sensor-based signal acquisition, comparison with preset thresholds for anomaly identification, and triggering early warning prompts based on data judgment results.

[0004] Current technologies primarily rely on comparing monitoring signals with preset thresholds to identify leakage problems, resulting in single-point judgments of abnormal states. This threshold-based approach struggles to comprehensively cover complex phantom structures, especially when bubbles are in a critical state or their aggregation distribution has not reached a significant abnormality level. This often leads to inaccurate identification and the failure to promptly expose early potential risks. Lacking consideration of the spatial relationships and distribution trends of bubbles, current technologies are prone to misjudgment or omission when distinguishing between occasional minor anomalies and potential failure signs, affecting the reliability of early warnings. Furthermore, their ability to explain the causes of anomalies is limited; they can only indicate signal fluctuations or deviations but cannot present the spatial characteristics of risk distribution, making it difficult for users to judge the severity and development trend of potential hazards. Long-term reliance on this method may result in some leakage hazards being overlooked in the early stages, only being discovered when structural damage becomes apparent, delaying optimal intervention and increasing the scrap rate of detection phantoms and equipment maintenance costs. Summary of the Invention

[0005] To address the technical problems existing in the prior art, embodiments of the present invention provide a method and system for detecting and warning of bone mineral density phantom leakage failure, the specific technical solution of which is as follows: On the one hand, a method for detecting and warning of leakage failure in a bone mineral density phantom is provided, including the following steps: S1: All pixels are obtained from the CT scan images of the bone density phantom and analyzed by 3D U-Net. The relationship between the pixels is calculated by combining the attention gating mechanism and bubble regions are identified to generate a bubble spatial distribution map. S2: Based on the bubble spatial distribution map, obtain the coordinates of bubble voxels and container walls, calculate the distance from each bubble voxel to the container wall, filter out the number of bubble voxels that have not reached the distance threshold and calculate the proportion of them to the total number of bubble voxels, as the edge aggregation index; S3: Based on the edge aggregation index and bubble voxel coordinates, calculate the number density of bubble voxels in the cluster radius using the DBSCAN clustering algorithm, filter bubble voxel clusters and mark them, and generate a bubble aggregation space map. S4: Overlay the bubble aggregation spatial map with the bubble spatial distribution map for analysis, filter out bubble voxel clusters with abnormal bubble number and continuous voxel bubble length, and generate a voxel bubble aggregation core map. S5: Based on the core diagram of the voxel bubble aggregation, calculate the leakage index of each bubble voxel aggregation cluster. If it exceeds the risk judgment threshold, mark the leakage failure area of ​​the phantom and generate a bone density detection phantom leakage failure early warning information table.

[0006] As a further embodiment of the present invention, the bubble spatial distribution map includes the bubble region range, voxel coordinates, and three-dimensional boundary relationships. The edge aggregation index specifically includes the proportion of bubble voxels, the voxel values ​​near the container wall, and the spatial distribution degree. The bubble aggregation spatial map includes the cluster number, voxel density, and cluster boundary coordinates. The voxel bubble aggregation core map specifically includes the abnormal cluster number, the number of bubbles, and the length of continuous voxels. The bone density detection phantom leakage failure early warning information table includes the failure zone number, the failure zone spatial range, and the leakage index.

[0007] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: Acquire all pixel data of the CT scan image of the bone density phantom, record the gray value and spatial coordinates of each pixel, and sort them according to the spatial location in three-dimensional space to generate pixel spatial data; S102: The pixel spatial data is input into the 3D U-Net network structure. The local features of adjacent pixels are extracted through the convolutional layer, the spatial texture information is summarized by the downsampling layer, and then the spatial relationship of the region is restored through the upsampling layer. The probability distribution of each pixel is calculated to obtain the spatial texture probability. S103: Based on the spatial texture probability, an attention gating mechanism is introduced to perform weighted calculations on the interrelationships between multiple pixels, aggregate the probability values ​​of the associated regions, mark the spatial range of the bubble area and map it in three-dimensional coordinates to generate a bubble spatial distribution map.

[0008] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Based on the bubble spatial distribution map, obtain the bubble voxel coordinates, record the position point of each bubble voxel in three-dimensional space, and store them in index order to obtain the bubble voxel coordinate set; S202: Call the bubble voxel coordinate set and obtain the container wall boundary coordinates. Calculate the straight-line distance from each bubble voxel to the container wall using the shortest path and summarize the results to obtain the bubble wall distance value. S203: Based on the bubble wall distance value, filter out the number of bubble voxels that have not reached the distance threshold, calculate the ratio of the filtered voxel number to the total number of voxels in the bubble voxel coordinate set, and generate the edge aggregation index.

[0009] As a further aspect of the present invention, the distance threshold is set by obtaining all the shortest distance values ​​from the bubble voxel to the container wall, then statistically analyzing the distribution range and central tendency of the values, and finally combining them with the overall scale range of the container wall.

[0010] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Based on the edge aggregation index and bubble voxel coordinates, the bubble voxels are spatially clustered using the DBSCAN clustering algorithm. The cluster radius is set, and the number density of bubble voxels in each cluster region is calculated to obtain the bubble voxel density distribution. S302: Based on the bubble voxel density distribution, filter bubble voxel clusters whose density exceeds a set density threshold, record and mark the bubble voxel coordinates within the clusters, and obtain bubble cluster markers. S303: Based on the bubble cluster marker, identify the aggregation region of bubble voxels in three-dimensional space, map the spatial distribution information of the clusters with the coordinates of the bubble voxels, and obtain a bubble aggregation spatial map.

[0011] As a further aspect of the present invention, the density threshold is determined by calculating the number of neighboring voxels of each bubble voxel within the cluster radius and statistically analyzing the number density, then performing distribution statistics on the number density of all voxels and extracting the concentration interval, and setting the threshold based on the differences between multiple density intervals.

[0012] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: Based on the bubble aggregation spatial map and the bubble spatial distribution map, the two types of spatial data are superimposed and analyzed in a three-dimensional coordinate system. The voxel distribution under the same spatial coordinates is compared and the difference in the number of voxels is statistically analyzed to obtain the difference value of the number of bubble voxels. S402: Based on the difference in the number of bubble voxels, calculate the continuous voxel length of the bubble in the coordinate axis direction for the arrangement of adjacent voxels in each cluster, and compare it with the set continuous length threshold to filter out clusters with abnormal length and obtain the continuous voxel length. S403: Call the difference value between the continuous voxel length and the number of bubble voxels to identify the selected abnormal clusters, map the spatial distribution range to three-dimensional coordinates, and establish a core map of voxel bubble aggregation. The continuous length threshold is set by calculating the length distribution of consecutive adjacent voxels within the cluster, statistically analyzing the maximum and average continuous spacing between voxels, and taking into account the overall structural characteristics of the cluster.

[0013] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: Obtain the core map of the voxel bubble aggregation, accumulate the index values ​​of each cluster using the risk assessment index method and normalize them, calculate the risk score of multiple clusters, and generate the leakage index. S502: Call the leakage index, compare the values ​​of multiple clusters with the set risk judgment threshold one by one, mark the clusters that exceed the risk judgment threshold, and obtain the leakage failure area of ​​the phantom. S503: Based on the amount of leakage failure area of ​​the phantom, and combined with the three-dimensional spatial coordinate information, establish a corresponding early warning table structure, and record the cluster number and coordinate range of the failure area to establish a bone density detection phantom leakage failure early warning information table. The risk assessment threshold is set based on the leakage index distribution characteristics of the bubble voxel clusters.

[0014] On the other hand, a bone mineral density phantom leakage failure detection and early warning system is provided, including: The image recognition module acquires all pixels based on the CT scan image of the bone density phantom and analyzes them using 3D U-Net. It also calculates the interrelationships between pixels and identifies bubble regions by combining an attention gating mechanism, generating a bubble spatial distribution map and transmitting it to the edge aggregation module. The edge clustering module obtains the coordinates of bubble voxels and container walls based on the bubble spatial distribution map, calculates the distance from each bubble voxel to the container wall, filters out the number of bubble voxels that do not reach the distance threshold and calculates the proportion of the total number of bubble voxels, which is used as the edge clustering index and passed to the clustering analysis module. The clustering analysis module, based on the edge clustering index and bubble voxel coordinates, calculates the number density of bubble voxels in the cluster radius using the DBSCAN clustering algorithm, filters and marks bubble voxel clusters, generates a bubble clustering spatial map, and transmits it to the anomaly screening module. The anomaly screening module overlays and analyzes the bubble aggregation spatial map and the bubble spatial distribution map, filters out bubble voxel clusters with abnormal bubble number and continuous voxel bubble length, generates a voxel bubble aggregation core map and transmits it to the risk warning module. The risk warning module calculates the leakage index of each bubble voxel cluster based on the core image of the voxel bubble aggregation using the risk assessment index method. If the leakage index exceeds the risk judgment threshold, the leakage failure area of ​​the phantom is marked and an early warning is issued, generating a bone density detection phantom leakage failure early warning information table.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, by performing depth analysis on phantom scan images, the subtle features of air bubbles in the images can be distinguished and spatial distribution maps generated. This allows the detection of microstructures to move beyond single-point parameters and be based on a holistic spatial perspective. Furthermore, by calculating the distance from the bubbles to the container wall and forming an edge aggregation index, the proximity of bubbles to critical structural regions can be effectively revealed, enabling the quantification of potential leakage hazards before obvious anomalies appear. Further analysis of bubble distribution density using spatial clustering techniques distinguishes dispersed bubbles from clusters that truly pose a structural threat, avoiding misjudgments and redundant warnings. During cluster screening, the superimposed identification of bubble quantity and continuity highlights high-risk areas that could potentially form leakage channels, resulting in a risk distribution that more closely reflects actual failure mechanisms. Finally, by combining the calculation of cluster areas and risk indices, a quantitative assessment of the leakage critical state is performed, and warning information is output. This ensures that the warning results not only remain at the level of anomaly identification but also provide a reference for distribution characteristics and risk levels. The overall logic has been improved from single signal acquisition to multi-dimensional spatial structure cognition, resulting in significant gains in identification accuracy, risk location precision, and forward-looking grasp of leakage trends, providing more reliable support for the long-term safe operation of equipment. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a detailed schematic diagram of S1 of the present invention; Figure 3 This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention; Figure 7 This is a system module diagram of the present invention. Detailed Implementation

[0018] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0019] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0020] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0021] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0022] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0023] Please see Figure 1 This invention provides a method for detecting and warning of bone mineral density phantom leakage failure, comprising the following steps: S1: All pixels are obtained from the CT scan images of the bone density phantom and analyzed by 3D U-Net. The relationship between the pixels is calculated by combining the attention gating mechanism and bubble regions are identified to generate a bubble spatial distribution map. S2: Based on the bubble spatial distribution map, obtain the coordinates of bubble voxels and container walls, calculate the distance from each bubble voxel to the container wall, filter out the number of bubble voxels that have not reached the distance threshold and calculate the proportion of them to the total number of bubble voxels, which is used as the edge aggregation index. S3: Based on the edge aggregation index and bubble voxel coordinates, the number density of bubble voxels in the cluster radius is calculated using the DBSCAN clustering algorithm, bubble voxel clusters are selected and marked, and a bubble aggregation spatial map is generated. S4: Overlay the bubble aggregation spatial map with the bubble spatial distribution map for analysis, screen bubble voxel clusters with abnormal bubble number and continuous voxel bubble length, and generate a voxel bubble aggregation core map. S5: Based on the core map of voxel bubble aggregation, calculate the leakage index of each bubble voxel aggregation cluster. If it exceeds the risk judgment threshold, mark the leakage failure area of ​​the phantom and generate a bone density detection phantom leakage failure early warning information table.

[0024] The bubble spatial distribution map includes the bubble region range, voxel coordinates, and three-dimensional boundary relationships. The edge aggregation index specifically includes the proportion of bubble voxels, the voxel values ​​near the container wall, and the spatial distribution degree. The bubble aggregation spatial map includes the cluster number, voxel density, and cluster boundary coordinates. The voxel bubble aggregation core map specifically refers to the abnormal cluster number, bubble quantity, and continuous voxel length. The bone density detection phantom leakage failure early warning information table includes the failure area number, the spatial range of the failure area, and the leakage index.

[0025] Please see Figure 2 The specific steps of S1 are as follows: S101: Acquire all pixel data of the CT scan image of the bone density phantom, record the gray value and spatial coordinates of each pixel, and sort them according to the spatial location in three-dimensional space to generate pixel spatial data; Based on the acquisition of all pixel data from the CT scan image of the bone density phantom, a DICOM format file was exported from the CT scanner. The image matrix data was read, with an image size of 512 pixels wide, 512 pixels high, 100 slices, a pixel pitch of 0.5 mm, and a slice thickness of 1.0 mm. The grayscale value of each pixel was extracted, ranging from 0 to 4095. For example, the grayscale value of pixels in a typical bone region is approximately 1500, while the grayscale value of pixels in an air region is approximately 0. The spatial coordinates of each pixel were calculated: the x-coordinate equals the pixel column index multiplied by the pixel pitch of 0.5, the y-coordinate equals the pixel row index multiplied by the pixel pitch of 0.5, and the z-coordinate equals the slice index multiplied by the slice thickness of 1.0. The coordinates and grayscale value of each pixel were recorded into a list. For example, a pixel with column index 100, row index 200, slice index 10, grayscale value 1500, coordinates x=50.0 mm, y=100.0 mm, z=10.0 mm, and a pixel with column index 101, row index 200, and slice index 101 ...000, coordinates x=50.0 mm, y=100.0 mm, z=10.0 mm, and a pixel with column index 101, row index 200, and a slice index 101, with a grayscale value of 1500, coordinates x=50.0 mm, y=100.0 mm, z=10.0 mm, and a slice index 10 0. Slice index 10, grayscale value 1520, coordinates x=50.5 mm, y=100.0 mm, z=10.0 mm, pixel column index 100, row index 201. Slice index 10, grayscale value 1480, coordinates x=50.0 mm, y=100.5 mm, z=10.0 mm. Record all pixels similarly, then sort the pixels in the list. The sorting key first compares the z-coordinates; if the z-coordinates are the same, compare the y-coordinates; if the y-coordinates are similar... Comparing x-coordinates, all are sorted in ascending order using a comparison function. For point A, z=10.0, y=100.0, x=50.0; for point B, z=10.0, y=100.0, x=50.5. Point A has a smaller x-coordinate, so it is placed before point B. For point C, z=10.0, y=100.5, x=50.0. Point C has a larger y-coordinate, so it is placed after points A and B. The sorted list is in the order of point A, point B, point C, generating pixel space data.

[0026] Table 1: Example Table of Pixel Data

[0027] As shown in Table 1, the coordinates and grayscale values ​​of some pixels are sorted by first comparing the z coordinates, then comparing the y coordinates if the z coordinates are the same, and then comparing the x coordinates if the y coordinates are the same. All pixels are sorted in ascending order, and the pixel spatial data is organized in three-dimensional spatial order after sorting.

[0028] S102: Input pixel spatial data into the 3D U-Net network structure, extract local features of adjacent pixels through convolutional layers, summarize spatial texture information through downsampling layers, restore the spatial relationship of the region through upsampling layers, calculate the probability distribution of each pixel, and obtain the spatial texture probability. Based on the input of pixel space data into the network structure, sorted pixel space data points A, B, and C are loaded. The data dimensions are 128x128x64x1. A 3D convolution operation is applied with a kernel size of 3x3x3, weight parameters ranging from -0.1 to 0.1, and a bias value of 0.01. The output feature map is calculated, with an output size of 128x128x64x32. For example, the input region is a 3x3x3 grayscale matrix with values ​​ranging from 1500 to 1520. The weighted product plus bias is used to calculate the output feature value, which is the sum of the input value multiplied by the corresponding weight and then the bias is added. For example, the input point is the area around (50, 100, 10), the gray value matrix is ​​1500, 1505, 1510, 1502, 1500, 1508, 1501, 1503, 1500, and the kernel weight matrix has random values ​​of 0.05, 0.03, 0.01, 0.02, 0.04, 0.06, 0.07, 0.09, 0.08. ; The dot product is approximately 676.12, resulting in an output feature value of 676.12. Downsampling is applied using max pooling with a pooling size of 2x2x2 and a stride of 2, resulting in an output size of 64x64x32x32. The maximum grayscale value of each 2x2x2 region is taken; for example, if the region's grayscale values ​​are 1500, 1520, 1480, 1510, 1490, 1530, 1470, and 1505, the maximum value is 1530, resulting in an output of 1530. Upsampling is then applied using trilinear interpolation, upsampling the size from 64x64x32 to 128x128x64. Interpolation weights are calculated based on the distance to neighboring pixels; for example, the point (50, 100, 10) is interpolated from its surrounding points (50, 100, 10). (0), (50, 100, 11), etc., weight calculation is based on the inverse of distance, output value, calculate probability distribution, apply softmax function, input feature value, output probability value range from 0 to 1, for example feature value 676.12, through exponential function normalization, the probability value is calculated as exp(676.12) divided by the sum of the exp of all feature values. Assuming that after performing exponential operation on all output feature values ​​of the layer where the feature value is located and summing them, the denominator value is S, then the probability value of the pixel is exp(676.12) / S, the probability is about exp(676.12) / 1000000. Since exp(676.12) is very large, the probability is close to 1.0, and the spatial texture probability is obtained.

[0029] S103: An attention gating mechanism is introduced based on spatial texture probability to perform weighted calculations on the interrelationships between multiple pixels, aggregate the probability values ​​of related regions, mark the spatial range of the bubble area and map it in three-dimensional coordinates to generate a bubble spatial distribution map. Based on an attention gating mechanism introduced according to spatial texture probability, spatial texture probability data is loaded, with probability values ​​ranging from 0 to 1. Attention weights are calculated using a Gaussian kernel function, where the weights are inversely proportional to the distance between pixels. Distance calculation is based on coordinate differences. For example, pixel A with coordinates (50.0, 100.0, 10.0) has a probability of 0.8, while pixel B with coordinates (50.5, 100.0, 10.0) has a probability of 0.6. `exp` represents the exponential function, and `sqrt` represents the square root function. The distance is calculated as `sqrt(50.5-50.0)^2+(100.0-100.0)^2+(10.0-1)^2`. 0.0)^2 = 0.5 mm, Sigma value set to 1.0 mm, based on empirical values ​​for typical bubble sizes, the weight is approximately exp(-0.25 / 2) = exp(-0.125) ≈ 0.882. The weights are normalized to make the sum equal to 1. Point A has a weight of 0.7, and point B has a weight of 0.3. The weighted calculation is performed on the relationships between multiple pixels, and the weighted probability value is calculated as sum, over, pixels, of, probability multiplied by weight. For example, point A has a probability of 0.8 and a weight of 0.7, and point B has a probability of 0.6 and a weight of 0.3, so the weighted sum = 0.56 + 0.6. 0.3 = 0.56 + 0.18 = 0.74. Aggregate the probability values ​​of the associated regions to obtain an aggregation probability of 0.74. Mark the spatial range of the bubble area. Set a threshold of 0.5. Based on the bubble probability distribution in historical data, pixels with a probability greater than 0.5 are marked as bubbles. For example, if the aggregation probability is greater than 0.74, it is marked as 1; otherwise, it is marked as 0. Map the pixels in three-dimensional coordinates and output the coordinates of the marked pixels. For example, point A with coordinates (50.0, 100.0, 10.0) is marked as 1, and point B with coordinates (50.5, 100.0, 10.0) is marked as 0. Generate a bubble spatial distribution map.

[0030] Please see Figure 3 The specific steps of S2 are as follows: S201: Obtain bubble voxel coordinates based on the bubble spatial distribution map, record the position point of each bubble voxel in three-dimensional space, and store them in index order to obtain the bubble voxel coordinate set; Based on the bubble spatial distribution map, a 3D array of data is read from the map. Point A with coordinates (50.0, 100.0, 10.0) is marked as 1, and point B with coordinates (50.5, 100.0, 10.0) is marked as 0. The array size is 128 pixels wide, 128 pixels high, and 64 slice depths. Each element has a value of 1 to represent a bubble voxel and 0 to represent a non-bubble voxel. All elements in the array are traversed, and nested loops are used to check each voxel value. When the value is equal to 1, the index position of that voxel is obtained: column index i, row index j, and slice index k. The spatial coordinates are calculated, and the x-coordinate is equal to... The column index i is multiplied by the pixel spacing of 0.5 mm, the y-coordinate is equal to the row index j multiplied by the pixel spacing of 0.5 mm, and the z-coordinate is equal to the slice index k multiplied by the layer thickness of 1.0 mm. For example, if the voxel index (10, 20, 5) has a value of 1, the coordinates are x=5.0 mm, y=10.0 mm, z=5.0 mm; if the voxel index (11, 20, 5) has a value of 1, the coordinates are x=5.5 mm, y=10.0 mm, z=5.0 mm; and if the voxel index (10, 21, 5) has a value of 1, the coordinates are x=5.0 mm, y=10.5 mm, z=5.0 mm. Each record... The coordinates of the bubble voxels are stored in a temporary list containing coordinate points such as (5.0, 10.0, 5.0), (5.5, 10.0, 5.0), and (5.0, 10.5, 5.0). Then, the coordinate points in the temporary list are sorted. The sorting key first compares the z-coordinates; if the z-coordinates are the same, the y-coordinates are compared; if the y-coordinates are the same, the x-coordinates are compared. All points are sorted in ascending order using a comparison function. For example, point A has z=5.0, y=10.0, x=5.0, point B has z=5.0, y=10.0, x=5.5, and point A has the smaller x-coordinate, so point A is ranked higher. Before point B, point C has z=5.0, y=10.5, x=5.0. Point C has a larger y-coordinate, so it is placed after points A and B. The sorted list is in the order of points A, B, and C. The sorted list is stored in a bubble voxel coordinate set, which is an array or list in which each element is a (x, y, z) tuple, such as (5.0, 10.0, 5.0), (5.5, 10.0, 5.0), (5.0, 10.5, 5.0), (6.0, 11.0, 5.0), (7.0, 12.0, 5.0). This gives us the bubble voxel coordinate set.

[0031] S202: Call the bubble voxel coordinate set and obtain the container wall boundary coordinates. Calculate the straight-line distance from each bubble voxel to the container wall using the shortest path and summarize the results to obtain the bubble wall distance value. Based on the bubble voxel coordinate set, which contains multiple coordinate points such as (5.0, 10.0, 5.0), (5.5, 10.0, 5.0), (5.0, 10.5, 5.0), (6.0, 11.0, 5.0), (7.0, 12.0, 5.0), the container wall boundary coordinates are obtained. The container wall is defined as the image boundary, with a minimum x-boundary of 0 mm and a maximum x-boundary of 64.0 mm (based on a 128-pixel width multiplied by a 0.5 mm spacing), a minimum y-boundary of 0 mm and a maximum y-boundary of 64.0 mm (based on a 128-pixel height multiplied by a 0.5 mm spacing), and a minimum z-boundary of... The maximum z-boundary is 64.0 mm (based on a slice depth of 64 mm multiplied by a layer thickness of 1.0 mm). Wall coordinates include (0, y, z), (64.0, y, z), (x, 0, z), (x, 64.0, z), (x, y, 0), and (x, y, 64.0). However, for distance calculations, only the wall position needs to be considered. The straight-line distance from each bubble voxel to the container wall is calculated using the Euclidean distance formula, but the minimum value among all wall distances is taken because it is the shortest path straight-line distance. For example, for bubble voxel coordinates (5.0, 10.0, 5.0), the wall distance at x=0 is calculated as sqrt. (5.0-0)^2+(10.0-10.0)^2+(5.0-5.0)^2=5.0 mm, calculated to x=64.0, wall distance sqrt(5.0-64.0)^2+(10.0-10.0)^2+(5.0-5.0)^2=59.0 mm, calculated to y=0, wall distance sqrt(5.0-5.0)^2+(10.0-0)^2+(5.0-5.0)^2=10.0 mm, calculated to y=64.0, wall distance sqrt(5.0-5.0)^2+(10.0-64.0)^2+(5.0-5.0)^2+(5.0-5.0)^2=59.0 mm, calculated to y=64.0, wall distance sqrt(5.0-5.0)^2+(10.0-64.0)^2+(5.0-5.0)^2+(5.0-5.0)^2=59.0 mm. )^2=54.0 mm, calculate the wall distance up to z=0 as sqrt(5.0-5.0)^2+(10.0-10.0)^2+(5.0-0)^2=5.0 mm, calculate the wall distance up to z=64.0 as sqrt(5.0-5.0)^2+(10.0-10.0)^2+(5.0-64.0)^2=59.0 mm, take the minimum value of 5.0 mm, similarly calculate the minimum distance for all bubble voxels, summarize these distance values ​​into a list, for example, distance list [5.0, 4.5, 6.0, 3.0, 7.0, ...], to obtain the bubble wall distance values.

[0032] Table 2: Example Table of Distance from Bubble Voxel to Wall

[0033] As shown in Table 2, the coordinates of some bubble voxels and the minimum straight-line distance to the wall are calculated based on the Euclidean formula to obtain the minimum value of the distance to all walls.

[0034] S203: Based on the bubble wall distance value, filter the number of bubble voxels that have not reached the distance threshold, calculate the ratio of the filtered voxel number to the total number of voxels in the bubble voxel coordinate set, and generate the edge aggregation index. Based on the bubble wall distance values, the list contains the minimum distance value for each bubble voxel, for example, distance values ​​[5.0, 4.5, 6.0, 3.0, 7.0, ...]. A distance threshold of 5.0 mm is set. The threshold is based on typical bubble size and the influence range of the container wall. The bubble diameter is approximately 2-4 mm, so a distance less than 5 mm is considered close to the wall. The number of bubble voxels that do not reach the distance threshold is filtered out. Not reaching the threshold means the distance is less than the threshold, so each distance value is checked to see if it is less than 5.0 mm. For example, a distance of 5.0 mm is not less than 5.0, so voxels with a distance < 5.0 mm are counted. For example, a distance value of 5.0 is not counted, but for the example, a threshold of 5.0 mm is set. Voxels with a distance less than 5.0 mm are screened. For example, if the distance is 4.5 mm less than 5.0 mm, count 1; if the distance is 3.0 mm less than 5.0 mm, count 1; if the distance is 5.0 mm or more than 5.0 mm, do not count; if the distance is 6.0 mm or more than 5.0 mm, do not count. Iterate through all distance values ​​and count the number of voxels with a distance < 5.0 mm. Set the total number of voxels from S201 to N = 1000, and the number of voxels to be screened to M = 200 (for example). Calculate the ratio of the number of voxels to be screened to the total number of voxels N. The ratio is equal to M divided by N. For example, if M = 200 and N = 1000, the ratio = 200 / 1000 = 0.2. Generate the edge aggregation index with an index value of 0.2.

[0035] Please see Figure 4 The specific steps of S3 are as follows: S301: Based on the edge aggregation index and bubble voxel coordinates, the bubble voxels are spatially clustered using the DBSCAN clustering algorithm. The cluster radius is set, and the number density of bubble voxels in each cluster region is calculated to obtain the bubble voxel density distribution. Based on the edge clustering index and bubble voxel coordinates, with an index value of 0.2 indicating that 20% of the bubbles are close to the wall, the bubble voxel coordinate set is obtained from S201. For example, the coordinate point list includes (5.0, 10.0, 5.0), (5.5, 10.0, 5.0), (5.0, 10.5, 5.0), (6.0, 11.0, 5.0), (7.0, 12.0, 5.0), etc. The clustering radius eps is set to 2.0 mm, referring to the typical bubble diameter of 2-4 mm, to ensure that adjacent bubbles can be captured. The minimum number of points minPts is set to 5, based on empirical values ​​to ensure the effectiveness of clustering. Spatial clustering is performed on the bubble voxels, and the Euclidean distance of each point to other points is calculated. sqrt represents the standard abbreviation for square root. The function name is used for example. The distance between point A (5.0, 10.0, 5.0) and point B (5.5, 10.0, 5.0) is sqrt(5.5-5.0)^2 + (10.0-10.0)^2 + (5.0-5.0)^2 = 0.5 mm. The distance between point A and point C (5.0, 10.5, 5.0) is sqrt(5.0-5.0)^2 + (10.5-10.0)^2 + (5.0-5.0)^2 = 0.5 mm. The distance between point A and point D (6.0, 11.0, 5.0) is sqrt(6.0-5.0)^2 + (11.0-10.0)^2 + (5.0-5.0)^2 ≈ 1.414 mm. The distance between point A and point E (7.0, 12.0, 5.0) is... sqrt(7.0-5.0)^2+(12.0-10.0)^2+(5.0-5.0)^2≈2.828 mm. For point A, check the number of points within a radius of 2.0 mm, including points B, C, and D (distance 1.414 < 2.0), a total of 3 points, which is less than 5, so point A is not a core point. Similar to checking all points, set point F (8.0, 13.0, 5.0) to have 5 points within a 2.0 mm radius, mark it as a core point, and form a cluster. For example, cluster 1 includes points B, C, D, E, F, etc. Calculate the number density of bubble voxels in each cluster region. The density is equal to the number of bubble voxels in the cluster divided by the cluster volume. The cluster volume is calculated by the bounding box volume of the cluster point set. For example, the x-coordinate of point in cluster 1 is min5. .0max8.0, y coordinate min10.0max13.0, z coordinate min5.0max5.0, volume (8.0-5.0)(13.0-10.0)(5.0-5.0)=3.03.00.0=0.0mm³, but z is the same, the volume is 0, which is unreasonable. Therefore, an approximate volume is used to set the distribution area of ​​cluster points. If min-z=max-z, then the layer thickness is 1.0mm, so the volume (8.0-5.0)(13.0-10.0)1.0=3.03.01.0=9.0mm³. The number of bubble voxels is set to 10, and the density is 10 / 9.0≈1.111bubbles / mm³. The density of all clusters is calculated similarly, and the density value is [1].[111, 1.333, 1.000, 0.800, 0.600, ...], yielding the bubble voxel density distribution.

[0036] Table 3: Example Table of Cluster Density

[0037] As shown in Table 3, the number, volume, and calculated density values ​​of bubble voxels in some clusters are presented. The density is calculated by dividing the number by the volume.

[0038] S302: Based on the bubble voxel density distribution, filter bubble voxel clusters whose density exceeds a set density threshold, record and mark the bubble voxel coordinates within the clusters, and obtain bubble cluster markers. Based on the bubble voxel density distribution, for example, density values ​​[1.111, 1.333, 1.000, 0.800, 0.600, ...], a density threshold of 1.0 bubbles / mm³ is set. Referring to average density or empirical values, typical bubble clusters have higher densities. Bubble voxel clusters with densities exceeding the set density threshold are screened. Each cluster's density is checked to see if it is greater than 1.0. For example, if density 1.111 > 1.0, cluster 1 is screened; if density 1.333 > 1.0, cluster 2 is screened; if density 1.000 equals 1.0, then... (The text abruptly ends here, likely due to an incomplete sentence or missing information.) "Exceeding" may mean strictly greater than, so the density must be greater than 1.0. Therefore, a density of 1.000 is not filtered, and a density of 0.800 < 1.0 is not filtered. Iterate through all density values ​​and filter out clusters with a density > 1.0. Record the coordinates of the bubble voxels within the clusters. For example, the list of coordinate points for cluster 1 [(5.5, 10.0, 5.0), (5.0, 10.5, 5.0), (6.0, 11.0, 5.0), ...], the list of coordinate points for cluster 2, etc., and label these clusters. Add labels such as high-density cluster to obtain bubble cluster labels.

[0039] S303: Based on the bubble cluster markers, identify the aggregation regions of bubble voxels in three-dimensional space, map the spatial distribution information of the clusters with the coordinates of the bubble voxels, and obtain a bubble aggregation spatial map; Based on the bubble cluster markers, the coordinate point list of cluster 1 [(5.5, 10.0, 5.0), (5.0, 10.5, 5.0), (6.0, 11.0, 5.0), ...], the coordinate point list of cluster 2, etc., identify the aggregation regions of bubble voxels in three-dimensional space. Based on the coordinates of the marked clusters, the spatial range of each cluster is calculated. For example, for cluster 1, the x-coordinates are min5.0, max8.0, the y-coordinates are min10.0, max13.0, the z-coordinates are min5.0, max5.0, and the region center point is (6.5, 11.5, 5.0). Similar to all clusters, the spatial distribution information of the clusters, such as the center point and range, is mapped to the bubble voxel coordinates to generate a three-dimensional array or image, where the region value of the marked clusters is 1, and the others are 0, thus obtaining the bubble aggregation spatial map.

[0040] Please see Figure 5 The specific steps of S4 are as follows: S401: Based on the bubble aggregation spatial map and the bubble spatial distribution map, the two types of spatial data are overlaid and analyzed in a three-dimensional coordinate system. The voxel distribution under the same spatial coordinates is compared and the difference in the number of voxels is statistically analyzed to obtain the difference value of the number of bubble voxels. Based on the bubble aggregation spatial map and bubble spatial distribution map, the image is a three-dimensional array with dimensions of 128x128x64. A value of 1 represents a bubble cluster region, and a value of 0 represents a non-clustered region. The bubble spatial distribution map is obtained from S103, also with the same dimensions, where a value of 1 represents a bubble voxel and a value of 0 represents a non-bubble voxel. Both types of spatial data are loaded into memory, and a three-dimensional coordinate system is created. The coordinate system ranges from x to 64.0 mm (128 pixels x 0.5 mm spacing), y to 64.0 mm, and z to 0 to 64.0 mm (64 slices x 1.0 mm layer thickness). Each spatial coordinate point is traversed, for example, coordinates (10.0, 20.0, 30.0). Read the voxel value corresponding to the bubble cluster spatial map at the specified coordinates. Convert the coordinates to pixel indices: x-index = x / 0.5 = 20, y-index = y / 0.5 = 40, z-index = z / 1.0 = 30. Set the cluster map value to 1, indicating that the point belongs to a cluster. Read the voxel value at the same coordinates in the bubble spatial distribution map, using the same index, and set the value to 0, indicating no bubble voxels. Compare the voxel distribution at the same spatial coordinates, checking if the voxel values ​​of the two maps are consistent. For example, if the cluster map value is 1 and the distribution map value is 0, it is marked as a difference point. Statistically analyze the differences for all coordinate points, including differences where the cluster map has voxels but the distribution map does not (value 1). (vs 0), or the distribution map has it but the cluster map does not (value 0 vs 1). In this embodiment, the difference statistics focus on the difference of bubble voxels in the cluster area. Therefore, the main statistics are the number of distribution map values ​​of 0 at the coordinate points of cluster map value 1, that is, the number of missing bubble voxels in the cluster. The number of points of cluster map value 1 is set to 10,000. Among these points, the number of distribution map values ​​of 0 is 500, so the number of differential voxels is 500, and the difference value of bubble voxel number is 500.

[0041] Table 4: Example Table of Voxel Value Comparison for Coordinate Points

[0042] As shown in Table 4, the voxel values ​​of some coordinate points are compared. The difference is defined as the case where the cluster plot value is 1 and the distribution plot value is 0. The difference value is obtained by counting the number of such points.

[0043] S402: Based on the difference in the number of bubble voxels, calculate the continuous voxel length of the bubble in the coordinate axis direction for the arrangement of adjacent voxels in each cluster, and compare it with the set continuous length threshold to filter out clusters with abnormal length and obtain the continuous voxel length. Based on the difference in the number of bubble voxels, a difference value of 500 indicates that there are 500 points within the cluster that were expected to contain bubbles but actually did not. For the arrangement of adjacent voxels within each cluster, the coordinate point list of each cluster is extracted from the bubble cluster space map. For example, the coordinate points of cluster 1 are [(5.5, 10.0, 5.0), (5.0, 10.5, 5.0), (6.0, 11.0, 5.0), ...]. The length of continuous voxels along the coordinate axis is calculated. First, the coordinate axis direction is selected, such as the x-axis. Voxels within the cluster are sorted by x-coordinate. Continuous voxels are checked; continuity is defined as the distance between adjacent voxels equal to the pixel distance of 0.5 mm. For example, point A x=5.0, point B x=5.5, distance 0.5 mm, continuous; point C x=6.0, distance 0.5 mm from point B, continuous. Therefore, the length of continuous voxels from x=5.0 to x=6.0 is (6.0-5.0) / 0.5. +1 = 3 voxels, which translates to 3 millimeters in length. 0.5 = 1.5 mm. Similar to calculating the y and z directions, but usually the maximum continuous length is used. For example, the continuous length in the x direction is 1.5 mm, and there may be multiple continuous lengths in the y direction. A threshold for the continuous length is set to a preset fixed value, such as 2.0 mm. For example, for cluster 1, the lengths of all continuous segments are calculated, and a length list of [1.5, 2.0, 1.0] mm is set. The maximum continuous spacing is 2.0 mm, and the average spacing is (1.5 + 2.0 + 1.0) / 3 = 1.5 mm. Based on experience, the threshold is set to the average spacing plus 0.5 mm, i.e., 1.5 + 0.5 = 2.0 mm. The maximum continuous length of each cluster is set to 2.0 mm, or a maximum spacing of 2.0 mm is used as a reference. However, to avoid over-filtering, a threshold of 2.0 mm is set. The maximum continuous length of each cluster is then compared with the set continuous length threshold. For example, the maximum continuous length of cluster 1 is 2.0 mm, which is equal to the threshold of 2.0 mm. Only clusters with a maximum continuous length of 2.5 mm are considered abnormal, so they are not filtered. However, if a cluster has a maximum continuous length of 2.5 mm, which is greater than 2.0 mm, it is filtered as abnormal. Clusters with abnormal lengths are filtered to obtain the continuous voxel lengths. For example, the continuous length list of abnormal clusters is [2.5, 3.0] mm.

[0044] S403: Call the difference value between the length of continuous voxels and the number of bubble voxels to identify the selected abnormal clusters, map the spatial distribution range to three-dimensional coordinates, and establish a core map of voxel bubble aggregation. The continuous length threshold is set by calculating the length distribution of consecutive adjacent voxels within a cluster, statistically analyzing the maximum and average continuous spacing between voxels, and combining the overall structural characteristics of the cluster. The difference between the length of the continuous voxel and the number of bubble voxels is called. The length of the continuous voxel is set from S402 to the list of abnormal cluster lengths [2.5, 3.0] mm. The difference value of 500 is obtained from S401. The selected abnormal clusters are identified, for example, abnormal clusters numbered 1 and 2, and a label such as "abnormal cluster" is added. The spatial distribution range is mapped to the three-dimensional coordinates. Based on the coordinate points of the abnormal clusters, the spatial boundary of each cluster is calculated. For example, for cluster 1, x min=5.0, max=8.0 mm, y min=10.0, max=13.0 mm, z min=5.0, max=5.0 mm. A three-dimensional array is generated, in which the value of the abnormal cluster area is 1 and the others are 0. The core map of voxel bubble aggregation is established.

[0045] Please see Figure 6 The specific steps of S5 are as follows: S501: Obtain the core map of voxel bubble aggregation, accumulate the index values ​​of each cluster using the risk assessment index method and normalize them, calculate the risk score of multiple clusters, and generate the leakage index. Based on the core map of voxel bubble aggregation, with x min=5.0, max=8.0 mm, y min=10.0, max=13.0 mm, z min=5.0, max=5.0 mm, this map is a three-dimensional array of data, with dimensions of 128 pixels wide, 128 pixels high, and 64 slice depths. Each pixel value represents the leakage risk level, ranging from 0 to 1, with higher values ​​indicating greater risk. Using a risk assessment index method, information on all bubble clusters is first obtained, including three indicators: bubble density, continuous voxel length, and edge aggregation index. Bubble density is derived from previous clustering analysis results; for example, cluster 1 has a density of 1.2 bubbles per cubic millimeter. Continuous voxel length is derived from continuity analysis; for example, cluster 1 has a continuous length of 2.5 mm. Edge aggregation index is derived from edge analysis; for example, cluster 1 has an index of 0.3. Each indicator is normalized, with bubble density ranging from 0 to... 2. For each cubic millimeter of bubble, the normalized value is the density value divided by 2. For example, the normalized density of cluster 1 is 0.6. The reference range for the length of the continuous voxel is 0 to 5 millimeters. The normalized value is the length value divided by 5. For example, the normalized length of cluster 1 is 0.5. The edge aggregation index itself is between 0 and 1 and does not need to be normalized. Accumulate the three normalized index values ​​of each cluster and sum them to obtain the original risk score. For example, the original score of cluster 1 is 0.6 plus 0.5 plus 0.3, which equals 1.4. Since the maximum possible score is 3, divide the original score by 3 to obtain the normalized leakage index. For example, the leakage index of cluster 1 is 1.4 divided by 3, which is approximately equal to 0.467. The leakage index of cluster 2 is 0.733. Repeat this process for all clusters to obtain a list of leakage indices for each cluster.

[0046] S502: Call the leakage index, compare the values ​​of multiple clusters with the set risk assessment threshold one by one, mark the clusters that exceed the risk assessment threshold, and obtain the leakage failure area of ​​the phantom. The leakage index list calculated above is used. For example, the leakage index of cluster 1 is 1.4 divided by 3, which is approximately 0.467, and the leakage index of cluster 2 is 0.733. The risk judgment threshold is set according to the distribution characteristics of all leakage indices. The average and standard deviation of these indices are calculated. For example, the average of the three indices 0.467, 0.733, and 0.2 is approximately 0.467, and the standard deviation is approximately 0.266. The threshold is set to the average plus twice the standard deviation, approximately 0.467 + 0.266 = 0.733. The leakage index of each cluster is compared with the threshold of 0.733 one by one. Clusters with an index greater than or equal to 0.733 are marked as high risk. For example, cluster 2 with an index of 0.733 is marked because it exceeds or equals the threshold. Clusters with an index less than or equal to 0.733 are not marked. For example, cluster 1 with an index of 0.467 and cluster 3 with an index of 0.2 do not exceed the threshold. The number of marked clusters is counted to obtain the number of phantom leakage failure areas. For example, one cluster is considered to have failed.

[0047] S503: Based on the amount of leakage failure area of ​​the phantom, and combined with the three-dimensional spatial coordinate information, establish a corresponding early warning table structure, and record the cluster number and coordinate range of the failure area to establish a bone density detection phantom leakage failure early warning information table. The risk assessment threshold is set based on the leakage index distribution characteristics of the bubble voxel clusters; Based on the amount of leakage failure area in the phantom, for example, one failure cluster, and combining the three-dimensional spatial coordinate information of cluster 1, its spatial range is obtained. For example, the x-coordinate range of cluster 2 is 15.0 to 16.5 mm, the y-coordinate range is 25.0 to 26.0 mm, and the z-coordinate range is 10.0 mm. An early warning table structure is established, which includes fields such as cluster number, coordinate range, and leakage index. Detailed information of the failure cluster is recorded, such as cluster number 2, x-range 15.0-16.5 mm, y-range 25.0-26.0 mm, z-range 10.0 mm, and leakage index 0.733. This forms a bone density testing phantom leakage failure early warning information table, which is stored in a structured form for easy subsequent querying and analysis.

[0048] Please see Figure 7 This invention also provides a bone mineral density phantom leakage failure detection and early warning system, comprising: The image recognition module acquires all pixels based on the CT scan image of the bone density phantom and analyzes them using 3D U-Net. It also calculates the interrelationships between pixels and identifies bubble regions by combining an attention gating mechanism, generating a bubble spatial distribution map and transmitting it to the edge aggregation module. The edge clustering module obtains the coordinates of bubble voxels and container walls based on the bubble spatial distribution map, calculates the distance from each bubble voxel to the container wall, filters out the number of bubble voxels that do not reach the distance threshold and calculates the proportion of them to the total number of bubble voxels, which is used as the edge clustering index and passed to the clustering analysis module. The clustering analysis module, based on the edge clustering index and bubble voxel coordinates, calculates the number density of bubble voxels in the cluster radius using the DBSCAN clustering algorithm, filters and marks bubble voxel clusters, generates a bubble clustering spatial map, and transmits it to the anomaly screening module. The anomaly screening module overlays and analyzes the bubble aggregation spatial map and the bubble spatial distribution map, filters out bubble voxel clusters with abnormal bubble number and continuous voxel bubble length, generates a voxel bubble aggregation core map and transmits it to the risk warning module. The risk warning module, based on the core image of voxel bubble aggregation, calculates the leakage index of each bubble voxel aggregation cluster using the risk assessment index method. If the leakage index exceeds the risk judgment threshold, the leakage failure area of ​​the phantom is marked and an early warning is issued, generating a bone density detection phantom leakage failure early warning information table.

[0049] For ease of explanation, Figure 7 Only the main components of the system are shown. This embodiment can be used to execute... Figure 1 The technical solutions of the method embodiments shown are similar in principle and in effect, and will not be described again here.

[0050] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for detecting and warning of leakage failure in a bone mineral density phantom, characterized in that, Includes the following steps: S1: All pixels are obtained from the CT scan images of the bone density phantom and analyzed by 3D U-Net. The relationship between the pixels is calculated by combining the attention gating mechanism and bubble regions are identified to generate a bubble spatial distribution map. S2: Based on the bubble spatial distribution map, obtain the coordinates of bubble voxels and container walls, calculate the distance from each bubble voxel to the container wall, filter out the number of bubble voxels that have not reached the distance threshold and calculate the proportion of them to the total number of bubble voxels, as the edge aggregation index; S3: Based on the edge aggregation index and bubble voxel coordinates, calculate the number density of bubble voxels in the cluster radius using the DBSCAN clustering algorithm, filter bubble voxel clusters and mark them, and generate a bubble aggregation space map. S4: Overlay the bubble aggregation spatial map with the bubble spatial distribution map for analysis, filter out bubble voxel clusters with abnormal bubble number and continuous voxel bubble length, and generate a voxel bubble aggregation core map. S5: Based on the core diagram of the voxel bubble aggregation, calculate the leakage index of each bubble voxel aggregation cluster. If it exceeds the risk judgment threshold, mark the leakage failure area of ​​the phantom and generate a bone density detection phantom leakage failure early warning information table.

2. The method for detecting and warning of bone mineral density phantom leakage failure according to claim 1, characterized in that, The bubble spatial distribution map includes the bubble region range, voxel coordinates, and three-dimensional boundary relationships. The edge aggregation index specifically includes the proportion of bubble voxels, the voxel values ​​near the container wall, and the spatial distribution degree. The bubble aggregation spatial map includes the cluster number, voxel density, and cluster boundary coordinates. The voxel bubble aggregation core map specifically refers to the abnormal cluster number, the number of bubbles, and the length of continuous voxels. The bone density detection phantom leakage failure early warning information table includes the failure area number, the spatial range of the failure area, and the leakage index.

3. The method for detecting and warning of bone mineral density phantom leakage failure according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Acquire all pixel data of the CT scan image of the bone density phantom, record the gray value and spatial coordinates of each pixel, and sort them according to the spatial location in three-dimensional space to generate pixel spatial data; S102: The pixel spatial data is input into the 3D U-Net network structure. The local features of adjacent pixels are extracted through the convolutional layer, the spatial texture information is summarized by the downsampling layer, and then the spatial relationship of the region is restored through the upsampling layer. The probability distribution of each pixel is calculated to obtain the spatial texture probability. S103: Based on the spatial texture probability, an attention gating mechanism is introduced to perform weighted calculations on the interrelationships between multiple pixels, aggregate the probability values ​​of the associated regions, mark the spatial range of the bubble area and map it in three-dimensional coordinates to generate a bubble spatial distribution map.

4. The method for detecting and warning of bone mineral density phantom leakage failure according to claim 1, characterized in that, The specific steps of S2 are as follows: S201: Obtain bubble voxel coordinates based on the bubble spatial distribution map, record the position point of each bubble voxel in three-dimensional space, and store them in index order to obtain a bubble voxel coordinate set; S202: Call the bubble voxel coordinate set and obtain the container wall boundary coordinates. Calculate the straight-line distance from each bubble voxel to the container wall using the shortest path and summarize the results to obtain the bubble wall distance value. S203: Based on the bubble wall distance value, filter out the number of bubble voxels that have not reached the distance threshold, calculate the ratio of the filtered voxel number to the total number of voxels in the bubble voxel coordinate set, and generate the edge aggregation index.

5. The method for detecting and warning of bone mineral density phantom leakage failure according to claim 4, characterized in that, The distance threshold is set by obtaining all the shortest distance values ​​from the bubble voxel to the container wall, then statistically analyzing the distribution range and central tendency of the values, and finally combining them with the overall scale range of the container wall.

6. The method for detecting and warning of bone mineral density phantom leakage failure according to claim 1, characterized in that, The specific steps for S3 are as follows: S301: Based on the edge aggregation index and bubble voxel coordinates, the bubble voxels are spatially clustered using the DBSCAN clustering algorithm. The cluster radius is set, and the number density of bubble voxels in each cluster region is calculated to obtain the bubble voxel density distribution. S302: Based on the bubble voxel density distribution, filter bubble voxel clusters whose density exceeds a set density threshold, record and mark the bubble voxel coordinates within the clusters, and obtain bubble cluster markers. S303: Based on the bubble cluster markers, identify the aggregation regions of bubble voxels in three-dimensional space, map the spatial distribution information of the clusters with the coordinates of the bubble voxels, and obtain a bubble aggregation spatial map.

7. The method for detecting and warning of bone mineral density phantom leakage failure according to claim 6, characterized in that, The density threshold is set by calculating the number of neighboring voxels of each bubble voxel within the cluster radius and statistically analyzing the number density, then performing distribution statistics on the number density of all voxels and extracting the concentration interval, and combining the differences of multiple density intervals.

8. The method for detecting and warning of bone mineral density phantom leakage failure according to claim 1, characterized in that, The specific steps of S4 are as follows: S401: Based on the bubble aggregation spatial map and the bubble spatial distribution map, the two types of spatial data are superimposed and analyzed in a three-dimensional coordinate system. The voxel distribution under the same spatial coordinates is compared and the difference in the number of voxels is statistically analyzed to obtain the difference value of the number of bubble voxels. S402: Based on the difference in the number of bubble voxels, calculate the continuous voxel length of the bubble in the coordinate axis direction for the arrangement of adjacent voxels in each cluster, and compare it with the set continuous length threshold to filter out clusters with abnormal length and obtain the continuous voxel length. S403: Call the difference value between the length of the continuous voxel and the number of bubble voxels to identify the selected abnormal clusters, map the spatial distribution range to three-dimensional coordinates, and establish a core map of voxel bubble aggregation; The continuous length threshold is set by calculating the length distribution of consecutive adjacent voxels within the cluster, statistically analyzing the maximum and average continuous spacing between voxels, and taking into account the overall structural characteristics of the cluster.

9. The method for detecting and warning of bone mineral density phantom leakage failure according to claim 1, characterized in that, The specific steps of S5 are as follows: S501: Obtain the core map of the voxel bubble aggregation, accumulate the index values ​​of each cluster using the risk assessment index method and normalize them, calculate the risk score of multiple clusters, and generate the leakage index. S502: Call the leakage index, compare the values ​​of multiple clusters with the set risk judgment threshold one by one, mark the clusters that exceed the risk judgment threshold, and obtain the leakage failure area of ​​the phantom. S503: Based on the amount of leakage failure area of ​​the phantom, and combined with the three-dimensional spatial coordinate information, establish a corresponding early warning table structure, and record the cluster number and coordinate range of the failure area to establish a bone density detection phantom leakage failure early warning information table. The risk assessment threshold is set based on the leakage index distribution characteristics of the bubble voxel clusters.

10. A bone mineral density phantom leakage failure detection and early warning system, characterized in that, The system is used to implement the bone mineral density phantom leakage failure detection and early warning method according to any one of claims 1-9, and the system includes: The image recognition module acquires all pixels based on the CT scan image of the bone density phantom and analyzes them using 3D U-Net. It also calculates the interrelationships between pixels and identifies bubble regions by combining an attention gating mechanism, generating a bubble spatial distribution map and transmitting it to the edge aggregation module. The edge clustering module obtains the coordinates of bubble voxels and container walls based on the bubble spatial distribution map, calculates the distance from each bubble voxel to the container wall, filters out the number of bubble voxels that do not reach the distance threshold and calculates the proportion of the total number of bubble voxels, which is used as the edge clustering index and passed to the clustering analysis module. The clustering analysis module, based on the edge clustering index and bubble voxel coordinates, calculates the number density of bubble voxels in the cluster radius using the DBSCAN clustering algorithm, filters and marks bubble voxel clusters, generates a bubble clustering spatial map, and transmits it to the anomaly screening module. The anomaly screening module overlays and analyzes the bubble aggregation spatial map and the bubble spatial distribution map, filters out bubble voxel clusters with abnormal bubble number and continuous voxel bubble length, generates a voxel bubble aggregation core map and transmits it to the risk warning module. The risk warning module calculates the leakage index of each bubble voxel cluster based on the core image of the voxel bubble aggregation using the risk assessment index method. If the leakage index exceeds the risk judgment threshold, the leakage failure area of ​​the phantom is marked and an early warning is issued, generating a bone density detection phantom leakage failure early warning information table.

Images