Chest x-ray image processing method for ai small shadow detection rate for pneumoconiosis

Abstract

The present application relates to the technical field of pneumoconiosis image detection, and discloses a chest X-ray image processing method for AI small shadow detection rate of pneumoconiosis. The method first generates a basic mask through an initial segmentation network, and then corrects the accurate lung profile by a hierarchical optimization unit. A multi-scale feature extraction module analyzes the gray scale features in the profile to construct an initial small shadow probability distribution graph. A morphological structure analyzer identifies discrete shadow areas and labels suspicious shadow clusters, and a dynamic confidence feedback mechanism evaluates the reliability thereof to generate an optimized feature map. A spatial relationship modeling network establishes a shadow cluster topology correlation matrix, matches the target shadow area with a pneumoconiosis pathological feature library, and performs screening. A three-dimensional reconstruction engine generates a small shadow volume density heat map, a hierarchical fusion module integrates the heat map and the original image spatial coordinates, and outputs an enhanced distribution atlas, and finally generates a structured diagnosis report data.

Description

Technical Field

[0001] This invention relates to the field of pneumoconiosis imaging detection technology, specifically to a chest X-ray image processing method for improving the detection rate of small shadows in pneumoconiosis using AI. Background Technology

[0002] Chest X-ray images are widely used in the diagnosis of pneumoconiosis, and the detection of small opacities is an important basis for judging the condition. Traditionally, the detection of small opacities mainly relies on manual interpretation of images. The diagnostic results are easily affected by subjective factors such as physician experience and fatigue. Different physicians may have different interpretations of the same image, resulting in low diagnostic consistency.

[0003] With the development of computer technology, some automatic detection methods based on image processing have emerged. However, existing methods often face the problem of inaccurate lung region segmentation when processing chest X-ray images, with blurred boundaries between the lungs and surrounding tissues, leading to inaccurate regions in subsequent small shadow analysis. In the feature extraction stage, most methods use single-scale analysis, which makes it difficult to fully capture the gray-scale distribution features of small shadows, resulting in a lot of noise in the initial probability distribution map, and the true shadows are easily masked.

[0004] Meanwhile, existing technologies lack in-depth morphological analysis of small shadows, making it difficult to effectively identify discrete shadow regions and suspicious shadow clusters. Furthermore, the lack of a dynamic evaluation mechanism for the reliability of shadow clusters easily leads to misclassification of normal tissue as shadows or omission of true lesion areas. In addition, there are shortcomings in the spatial relationship modeling of shadow regions, failing to accurately establish topological associations between shadow clusters, resulting in low matching with pathological feature databases and affecting the accuracy of target shadow region selection. Finally, the generated diagnostic reports are mostly unstructured data, which is detrimental to subsequent statistical analysis. Summary of the Invention

[0005] The purpose of this invention is to provide a chest X-ray image processing method for improving the detection rate of small shadows in pneumoconiosis using AI, in order to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a chest X-ray image processing method for improving the detection rate of small opacities in pneumoconiosis using AI, the method comprising:

[0007] Acquire the original chest X-ray image to be processed, perform initial segmentation network processing on the original image, and generate a base mask containing the lung region;

[0008] The basic mask is refined and corrected using a hierarchical optimization unit to output an accurate lung contour;

[0009] The gray-scale distribution features within the precise lung contour are analyzed using a multi-scale feature extraction module to construct an initial small shadow probability distribution map.

[0010] The initial small shadow probability distribution map is processed based on the morphological structure analyzer to identify discrete shadow regions and mark suspicious shadow clusters.

[0011] The morphological feature confidence of the suspected shadow clusters is evaluated using a dynamic confidence feedback mechanism to generate an optimized small shadow feature map;

[0012] The optimized small shadow feature map is input into the spatial relationship modeling network to establish the topological association matrix between shadow clusters;

[0013] The topological correlation matrix is ​​matched with a preset pneumoconiosis pathological feature database to screen target shadow areas that conform to clinical pathological features.

[0014] The target shadow region is processed using a 3D reconstruction engine to generate a small shadow volume density heatmap.

[0015] The hierarchical fusion module integrates the spatial coordinate information of the small shadow volume density heatmap and the original image to output an enhanced small shadow distribution map.

[0016] Structured diagnostic report data is generated based on the enhanced small shadow distribution map.

[0017] Preferably, the initial segmentation network processing specifically includes: constructing a dual-channel convolutional neural network as the segmentation backbone network, wherein the segmentation backbone network includes a spatial path and a context path that work in parallel; extracting high-resolution edge features of the original image through the spatial path to generate an edge feature tensor; capturing global semantic features of the original image using the context path to generate a semantic feature tensor; merging the edge feature tensor and the semantic feature tensor using an adaptive feature fusion processor to generate a fused feature map; performing multi-level supervised training on the fused feature map and updating the weight parameters of the segmentation backbone network through a backpropagation algorithm; and using the binarized image output by the segmentation backbone network after training convergence as the base mask.

[0018] Preferably, the operation of the hierarchical optimization unit includes: receiving the base mask and executing a contour tracking algorithm to extract the lung boundary pixel chain; calculating the local curvature anomalies of the lung boundary pixel chain using a curvature analysis model; constructing a boundary correction vector field based on the local curvature anomalies; optimizing the boundary correction vector field using an iterative energy minimization algorithm to drive the lung boundary pixel chain to move towards the anatomical standard position; and outputting accurate lung contour data containing a smooth and continuous contour.

[0019] Preferably, the operation steps of the multi-scale feature extraction module are as follows: establishing a multi-resolution pyramid structure within the precise lung contour; processing the lung region of the original image using a Gaussian difference filter bank to generate a multi-scale gradient magnitude map; extracting local texture features from the multi-scale gradient magnitude map to construct a rotation-invariant texture descriptor; calculating the similarity matrix between the rotation-invariant texture descriptor and the standard pneumoconiosis texture template; and fusing the spatial weights of the similarity matrix and the multi-scale gradient magnitude map to generate initial small shadow probability distribution data.

[0020] Preferably, the morphological structure analyzer performs the following operations: performs adaptive threshold segmentation on the initial small shadow probability distribution map to obtain binarized candidate regions; processes the binarized candidate regions using a connected component labeling algorithm to generate discrete shadow region identifiers; calculates the geometric morphological feature vector for each discrete shadow region, the geometric morphological feature vector including area irregularity and boundary fractal dimension; analyzes the spatial clustering characteristics of the geometric morphological feature vector based on a density clustering algorithm, and labels suspicious shadow clusters that satisfy clustering constraints.

[0021] Preferably, the dynamic confidence feedback mechanism includes: constructing a shadow feature evaluation network, which includes a feature encoder and a confidence predictor; inputting the original image patch of the suspected shadow cluster into the feature encoder to generate a deep feature representation; calculating the morphological confidence score of the deep feature representation through the confidence predictor; using a non-maximum suppression algorithm to filter suspected shadow clusters whose morphological confidence scores are higher than a critical threshold; aligning the filtering results with the original image coordinates to generate an optimized small shadow feature matrix.

[0022] Preferably, the operation of the spatial relationship modeling network includes: receiving the optimized small shadow feature matrix and extracting spatial location coordinates;

[0023] Construct a graph neural network model with each shadow cluster as a node and the Euclidean distance between nodes as the edge weight; learn the spatial dependencies between nodes through graph convolutional layers and update the node feature representations; calculate the topological association strength between the updated node feature representations and neighboring nodes; output the topological association matrix data containing the connection strength between nodes.

[0024] Preferably, the pathological feature database matching process includes: pre-constructing a standard pneumoconiosis shadow feature database, which stores the shadow distribution patterns of typical pneumoconiosis cases; extracting key path feature vectors from the topological correlation matrix; calculating the cosine similarity between the key path feature vectors and the standard pneumoconiosis shadow feature database; filtering topological subgraphs with cosine similarity greater than a matching threshold; mapping the topological subgraphs back to the original image space, and outputting the target shadow region coordinate set.

[0025] Preferably, the operation steps of the three-dimensional reconstruction engine are as follows: acquiring an X-ray image sequence corresponding to the target shadow region on a continuous layer; processing the X-ray image sequence using a voxel reconstruction algorithm to generate three-dimensional volume data; calculating the gray-level gradient features of each voxel in the three-dimensional volume data; optimizing the signal-to-noise ratio of the three-dimensional volume data based on an anisotropic diffusion filter; and generating three-dimensional heatmap data with labeled small shadow spatial distribution.

[0026] Preferably, the hierarchical fusion module performs the following: receiving the spatial coordinate transformation parameters of the three-dimensional heatmap data and the original image; establishing a conversion function from small shadow density to color mapping; projecting the three-dimensional heatmap data onto the two-dimensional diagnostic plane using an image registration algorithm; fusing the anatomical structural features of the projected data and the original image to generate an enhanced small shadow distribution map; and generating a structured report containing the number and distribution location of shadows based on the density partitioning statistical results of the enhanced small shadow distribution map.

[0027] Compared with the prior art, the beneficial effects of the present invention are:

[0028] Through the collaborative operation of multiple modules, accurate detection and analysis of small opacities in pneumoconiosis are achieved. After the initial segmentation network generates a basic mask, the hierarchical optimization unit refines and corrects its boundaries, resulting in a more accurate lung contour and reducing interference from surrounding tissues in subsequent analysis. The multi-scale feature extraction module extracts grayscale distribution features from the accurate lung contour, and the constructed initial small opacity probability distribution map can more comprehensively reflect potential opacity information, avoiding feature omissions caused by single-scale analysis.

[0029] A morphological structure analyzer processes the initial probability distribution map, effectively identifying discrete shadow regions and marking suspicious shadow clusters, making the distribution of small shadows clearer. A dynamic confidence feedback mechanism evaluates the reliability of the morphological features of suspicious shadow clusters; the optimized small shadow feature map reduces the possibility of misjudgment and missed judgment, making the shadow information closer to reality. A spatial relationship modeling network establishes a topological correlation matrix between shadow clusters, matching it with a pneumoconiosis pathological feature database to screen target shadow regions that conform to clinicopathological characteristics, making the selected regions more clinically valuable.

[0030] The 3D reconstruction engine processes the small shadow volume density heatmap generated from the target shadow area, which can intuitively display the density distribution of the shadow. The hierarchical fusion module integrates the spatial coordinate information of the heatmap and the original image, and the output enhanced small shadow distribution map combines the density and spatial location information of the shadow, making the distribution of small shadows more intuitive and comprehensive. The structured presentation of diagnostic report data makes it easy for physicians to quickly obtain key information and also facilitates comparison between different cases.

[0031] The entire process begins with precise segmentation of the lung region, followed by feature extraction, shadow recognition, credibility assessment, spatial relationship modeling, matching with the pathology database, 3D reconstruction and atlas fusion, and finally the generation of a structured report. This forms a complete processing chain, with each step working together to improve the overall effectiveness of small shadow detection and provide more reliable image evidence for the diagnosis of pneumoconiosis. Attached Figure Description

[0032] Figure 1 This is a schematic diagram illustrating the working principle of the chest X-ray image processing method for improving the detection rate of small shadows in pneumoconiosis as described in this invention.

[0033] Figure 2 A flowchart for hierarchical optimization of unit operations;

[0034] Figure 3 A flowchart illustrating the operation of the multi-scale feature extraction module;

[0035] Figure 4 A flowchart of the dynamic confidence feedback mechanism;

[0036] Figure 5 A flowchart for modeling network operations for spatial relationships;

[0037] Figure 6 Flowchart for the detection and processing of small shadows in pneumoconiosis. Detailed Implementation

[0038] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0039] Please see Figure 1 This invention provides a chest X-ray image processing method for improving the detection rate of small shadows in pneumoconiosis using AI, the method comprising:

[0040] The process begins by acquiring the original chest X-ray image and performing initial segmentation network processing to generate a base mask containing the lung region. A hierarchical optimization unit refines the boundaries of this base mask, outputting a precise lung contour. A multi-scale feature extraction module analyzes the grayscale distribution features within this precise lung contour to construct an initial small shadow probability distribution map. A morphological structure analyzer processes this initial small shadow probability distribution map, identifying discrete shadow regions and marking suspicious shadow clusters. A dynamic confidence feedback mechanism evaluates the morphological feature confidence of these suspicious shadow clusters, generating an optimized small shadow feature map. This optimized small shadow feature map is then input into a spatial relationship modeling network to establish a topological correlation matrix between shadow clusters. This topological correlation matrix is ​​matched against a pre-defined pneumoconiosis pathological feature database to select target shadow regions that match clinical pathological characteristics. A 3D reconstruction engine processes these target shadow regions, generating a small shadow volume density heatmap. A hierarchical fusion module integrates the small shadow volume density heatmap with the spatial coordinate information of the original image, outputting an enhanced small shadow distribution atlas. Structured diagnostic report data is then generated based on this enhanced small shadow distribution atlas.

[0041] Raw chest X-ray images are acquired using medical imaging equipment and stored in the data processing system. Initial segmentation involves a trained neural network model, with hierarchical optimization units using boundary correction vector fields for fine-tuning. The multi-scale feature extraction module incorporates multi-resolution structures when processing grayscale distribution features. A morphological structure analyzer employs adaptive thresholding and connected component algorithms. A dynamic confidence feedback mechanism uses a shadow feature evaluation network to calculate morphological confidence scores. A spatial relationship modeling network constructs a graph neural network model. Pathological feature library matching is based on cosine similarity calculation. The 3D reconstruction engine applies voxel reconstruction algorithms when processing X-ray image sequences. A hierarchical fusion module performs image registration algorithms.

[0042] See Figure 2 This embodiment involves the operation of the initial segmentation network processing and the hierarchical optimization unit, and is applied to the processing flow of raw chest X-ray images.

[0043] The process begins with the input of a raw chest X-ray image to be processed. This image is captured by a medical imaging device and stored in a digital format, such as DICOM, with a pixel resolution of 1024×1024 pixels and a grayscale value range of 0-255. The initial segmentation network constructs a dual-channel convolutional neural network as the segmentation backbone. This backbone network employs a pre-trained architecture. Specifically, the spatial path consists of three convolutional layers: the first layer uses 32 3×3 convolutional kernels to process the input image and extract high-resolution edge information; the second layer uses 64 3×3 convolutional kernels to further refine the edge features; and the third layer outputs an edge feature tensor with 128 3×3 convolutional kernels, which preserves the spatial details of the original image. The context path includes a downsampling module: the initial layer uses 64 7×7 convolutional kernels to capture the global context; subsequent layers combine max pooling operations to reduce the resolution and expand to 256 channels to output a semantic feature tensor; this path integrates a global average pooling layer to summarize image-level semantics. These two paths process the input image in parallel: the spatial path focuses on local edges, while the contextual path focuses on the overall structure. An adaptive feature fusion unit then merges the edge feature tensors generated by the spatial path and the semantic feature tensors generated by the contextual path. The fusion process incorporates a channel attention mechanism: the average pooling value of the feature maps is calculated as weights, and higher weights are assigned to salient regions in the edge features; element-wise addition is used to combine the weighted tensors to generate a fused feature map, which is a three-dimensional array with a size equal to the input image size multiplied by the number of feature channels. Multi-level supervised training is performed on the fused feature map, using a dataset of images labeled lung regions, with 100 epochs of training iterations. Backpropagation updates the weight parameters of the segmentation backbone network, using binary cross-entropy as the loss function, calculating the error between the predicted mask and the ground truth label; the Adam optimizer is used with an initial learning rate of 0.001, employing an exponential decay strategy to reduce the learning rate later in training. Multi-level supervision involves adding auxiliary losses in intermediate layers to ensure stable convergence during training. After training, the binarized image output by the segmentation backbone network is used as the base mask. This mask identifies the lung region in a black and white binary format, with a foreground pixel value of 1 and a background value of 0.

[0044] The hierarchical optimization unit receives the base mask as input and executes a contour tracking algorithm to extract the lung boundary pixel chain. The contour tracking algorithm scans the binarized image of the base mask using a boundary tracking method, identifying the outer boundary points of all connected regions. It uses Freeman chain code to encode the positional information of the boundary pixels, generating an ordered list of pixels, i.e., the lung boundary pixel chain, which is represented as a sequence of coordinate points. Subsequently, a curvature analysis model is used to calculate local curvature anomalies in the lung boundary pixel chain. Local curvature calculation moves a sliding window of 9 pixels along the pixel chain, evaluating the curvature at each point. By comparing the directional changes of vectors at adjacent points, locations with curvature exceeding a preset threshold of 0.1 are detected as anomalies; these anomalies correspond to discontinuous or jagged regions on the boundary. A boundary correction vector field is constructed based on the local curvature anomalies; each anomaly is assigned a correction vector, with the vector direction pointing to a reference line at an anatomically standard position. The reference line is predefined based on the lung anatomy, and the vector magnitude is calculated from the distance difference. The boundary correction vector field is initialized as a field map covering the entire image region. The boundary correction vector field is optimized using an iterative energy minimization algorithm. This algorithm defines an energy function comprising two terms: a boundary smoothing term that measures the curvature consistency of the boundary points, and a boundary constraint term that ensures the boundary does not deviate too far from its original position. The optimization process employs gradient descent iterations, with 10 iterations. In each iteration, the gradient of the energy function is calculated, and the direction and magnitude of the correction vector are adjusted. Ultimately, the optimization process drives the lung boundary pixel chain to smoothly move towards its anatomically standard position, eliminating abnormal protrusions or depressions. After optimization, accurate lung contour data is output, stored as a list of coordinate points, each containing x and y coordinate values, ensuring a continuous, smooth contour without breaks or overfitting.

[0045] In the overall implementation, the hardware platform utilizes a graphics processing unit (GPU) to accelerate computation; the segmentation backbone network is implemented using the PyTorch framework; the input image size is standardized to 512×512 pixels to speed up processing; the channel attention mechanism of the adaptive feature fusion unit is implemented using the SE module; the contour tracking algorithm uses Python library functions to implement chaincode operations; the threshold setting of the curvature analysis model is based on experimentally validated standard values; the reference line for the boundary correction vector field is taken from a lung template database; the optimization iteration step size is set to 0.01 to ensure convergence; data structures such as feature tensors are stored as NumPy arrays, and the list of coordinate points is in CSV format. The entire processing flow is executed sequentially, with intermediate results stored in a memory cache to avoid redundant calculations and improve efficiency. After outputting accurate lung contours, the system stores the data as a file or transfers it to subsequent modules.

[0046] The above implementation details the complete operation sequence of the initial segmentation network processing and hierarchical optimization unit, ensuring accurate segmentation of the lung region in clinical applications. The initial segmentation network processing takes approximately 500 milliseconds to process image data, and the hierarchical optimization unit operation takes approximately 200 milliseconds, running smoothly on standard hardware. Parameter tuning of the segmentation backbone network is based on cross-validation, with the dataset divided into training and validation sets. The contour tracking algorithm processes approximately 1000 boundary points per scan. The curvature analysis model uses a sliding window with a one-pixel movement interval, analyzing each point individually. During the boundary correction vector field optimization process, gradient calculation uses numerical methods to approximate the derivative. The convergence condition of the energy minimization algorithm is an energy change less than 0.001 or reaching the maximum number of iterations. The coordinate points of the accurate lung contour output are evenly spaced, with the spacing controlled within 1 pixel. After system integration, it functions as an independent processing unit, with input / output interfaces compatible with standard medical image formats. Training data augmentation techniques include random rotation and scaling of the original images to enhance model robustness. A batch normalization layer is introduced in the feature fusion stage to stabilize the training process. The list of coordinate points is serialized and stored as a binary file, saving space. The execution engine is based on multi-threaded parallel optimization calculations. During contour drawing, B-spline interpolation is used to smooth the output path. The input and output verification of the hierarchical optimization unit passes unit tests to ensure correct alignment. Finally, the processing results are integrated into the system log, recording the execution status of key steps.

[0047] See Figure 3 This embodiment involves the operation process of a multi-scale feature extraction module and a morphological structure analyzer, which is applied to the analysis and processing of precise lung contour regions.

[0048] The precise lung contour data output from Example 1 is obtained. This data exists in the form of a coordinate point sequence, formatted as an N×2 array, where N represents the number of boundary points. Based on this contour coordinate, lung region pixel blocks are extracted from the original chest X-ray image and cropped into a rectangular region image, with the size automatically adapted to the maximum bounding rectangle. The multi-scale feature extraction module establishes a multi-resolution pyramid structure within this rectangular region. The pyramid structure is implemented with three resolution levels: the first layer maintains the original resolution, such as 1024×1024 pixels; the second layer generates a 512×512 pixel image through a downsampling factor of 2, using a bilinear interpolation algorithm; the third layer is downsampled again to a resolution of 256×256 pixels. Each layer of the image is processed independently.

[0049] A Gaussian difference filter bank was used to process the lung region of the original image to generate multi-scale gradient magnitude maps. The Gaussian difference filter bank consists of three independent filters: the first filter convolves the original layer image with a Gaussian kernel of σ=1.0, outputting a smoothed image G1; the second filter processes the same layer image with σ=1.6 to obtain G2; and the third filter with σ=2.0 yields G3. The difference between adjacent Gaussian smoothed images was calculated: D1=|G1-G2|, D2=|G2-G3|, forming two difference results. The Sobel gradient operator was applied to D1 and D2 to extract the derivatives in the X and Y directions, respectively, synthesizing gradient magnitude maps M1 and M2. M1 and M2 were normalized to the range of 0-255 to serve as scale-dependent gradient feature maps.

[0050] Local texture features are extracted from the gradient magnitude map using a sliding window method. The window size is set to 16×16 pixels, with a step size of 8 pixels. At each window position, gray-level co-occurrence matrix attributes, including contrast, correlation, and energy values, are calculated. Texture patterns are calculated based on rotation-invariant variants of Local Binary Patterns (LBP). The binary pattern of the 8-neighborhood of the window's center pixel is calculated, and the smallest decimal value is obtained through cyclic shifting as the invariant code. The gray-level co-occurrence matrix attributes and LBP encoded values ​​are combined to construct a 128-dimensional rotation-invariant texture descriptor vector.

[0051] Calculate the similarity matrix between the descriptor vector and the standard pneumoconiosis texture template. A pre-built database of standard pneumoconiosis texture templates contains the mean vectors of texture features from 1000 pneumoconiosis cases. Similarity is calculated using cosine similarity: the current descriptor vector V is multiplied by the database template vector T, and then divided by the modulus product; the result is stored in a similarity score matrix, where rows correspond to pixel window positions and columns correspond to different template indices. The highest similarity value at each position is taken as the representative score S(x,y).

[0052] Spatial weights are fused from the similarity matrix and the multi-scale gradient magnitude map. Spatial weight calculation is based on the saliency of the gradient magnitude map: M1 and M2 are weighted and summed, with weight factors set to 0.6 and 0.4 respectively; the summation result is normalized to obtain the spatial weight map W(x,y). At the same spatial location, the texture similarity score S(x,y) is multiplied by the spatial weight W(x,y), and the product is mapped to the probability value P(x,y) using the sigmoid function. Bicubic interpolation is performed at all window locations to generate an initial small shadow probability distribution map with the same resolution as the original image, with probability values ​​ranging from 0 to 1.

[0053] The morphological structure analyzer operates on the probability distribution map. Adaptive thresholding segmentation uses a local window algorithm: the image is divided into 32×32 pixel sub-blocks, and a threshold is calculated independently for each sub-block. The threshold calculation formula is based on the mean and standard deviation of the probability values ​​within the block, with the threshold set to the mean plus 0.5 times the standard deviation. The binarization operation sets pixels above the threshold to 1, otherwise to 0, generating a binarized candidate region map.

[0054] A connected component labeling algorithm is used to process binary candidate regions. The algorithm performs a four-neighbor scan: traversing the binary image from left to right and from top to bottom. When a foreground pixel (value = 1) is found, a seed filling algorithm is initiated to label connected regions: starting from that pixel, the upper, lower, left, and right adjacent pixels are recursively checked; all connected foreground pixels are assigned the same label value. Label values ​​are assigned sequentially starting from 1, and the final output is a label matrix containing multiple discrete shadow regions.

[0055] Calculate the geometric feature vector for each labeled region. The region area A is calculated as the total number of pixels within the region; the region boundary perimeter P is precisely calculated using a boundary tracing algorithm. The area irregularity index R = (P²) / (4πA) is obtained. Boundary fractal dimension calculation: at different scales s, the box-covering method is used to count the number of boxes N(s) required to cover the boundary; the slope of a fitted line in the log-log coordinate system is used as the dimension D. Each region outputs a feature vector containing the coordinate centroid, area, perimeter, irregularity, and fractal dimension.

[0056] The spatial distribution of feature vectors is analyzed based on the density clustering algorithm. The DBSCAN algorithm is implemented with the following input parameters: neighborhood radius ε = 5 pixels, minimum number of points minPts = 10. The algorithm execution process is as follows: the centroid coordinates of each region are used as data points; the Euclidean distance between points is calculated; the core point is defined as a point within the ε-neighborhood of at least minPts; clusters are formed by expanding connected regions from the core point. Sets of points satisfying the clustering conditions are marked as suspected shadow clusters. Each cluster is assigned a unique identifier ID and records the original region label, centroid coordinates, feature mean, and spatial range. The final output is a list of suspected shadow clusters in a structured array format, containing ID, number of members, geometric statistical feature values, and bounding box coordinates.

[0057] During implementation, all calculations utilize single-precision floating-point operations. Original images are loaded in 16-bit grayscale format. Pyramid generation employs GPU parallel downsampling. Gaussian convolution uses separable convolution for performance optimization. Texture descriptor calculation is accelerated using SIMD instructions. The similarity matrix is ​​stored in a sparse format to save memory. The probability distribution map uses a multi-layer pyramid fusion strategy to enhance detail. The connected component labeling algorithm is optimized to a two-pass scanning method. Geometric feature calculation incorporates region convex hull detection to improve accuracy. Density clustering spatial indexing constructs a kd-tree to improve search efficiency. The system includes an anomaly handling mechanism: when the area of ​​a region exceeds a threshold, the region segmentation strategy is automatically triggered to recalculate features. Parameters are configurable, with interfaces providing variables such as ε and minPts for user adjustment. Data output is encapsulated in JSON format for compatibility with subsequent processing modules. Processing results are cached in an in-memory database, supporting breakpoint continuation functionality. The overall processing time of the morphological structure analyzer is controlled within 1.5 seconds to complete typical image analysis.

[0058] Example 3: See Figure 4 This embodiment relates to the operation flow of a dynamic confidence feedback mechanism, a key component in the image processing workflow responsible for evaluating the morphological feature confidence of suspicious shadow clusters. The implementation begins with receiving the list of suspicious shadow clusters output from Embodiment 2. This list includes the ID identifier, number of member regions, geometric feature statistics, and spatial bounding box coordinates for each cluster. The shadow feature evaluation network is constructed using a modular design, comprising two main components: a feature encoder and a confidence predictor. The feature encoder architecture is based on a convolutional neural network. The input layer receives 64×64 pixel image patches, which are cropped from the original chest X-ray image. The cropping center is determined by the centroid coordinates of the suspicious shadow cluster. When a shadow cluster spans multiple regions, the cropped region automatically adapts to the minimum bounding square.

[0059] The feature encoder's network structure comprises five convolutional layers, configured as follows: the first convolutional layer uses 32 5×5 convolutional kernels with a stride of 2 and employs the ReLU activation function; the second convolutional layer uses 64 3×3 convolutional kernels with a stride of 2 and also employs ReLU activation; the third to fifth convolutional layers each use 128 3×3 convolutional kernels with a stride of 1, and the activation function continues the ReLU design. Each convolutional layer is followed by a batch normalization layer to stabilize the training process. Max pooling is applied after the third and fifth convolutional layers, with a pooling window of 2×2 and a stride of 2. A global average pooling layer is added at the end of the network to compress the feature map into a fixed-length vector representation. When processing input image patches, this feature encoder progressively extracts and abstracts visual features through hierarchical convolutional operations, ultimately outputting a 512-dimensional vector that encapsulates the morphological feature information of the image patch.

[0060] The confidence predictor design comprises two fully connected layers: the first fully connected layer has 256 neurons and uses ReLU activation; the second fully connected layer outputs a single neuron, and the output value is compressed to the (0,1) interval using the sigmoid function, representing the morphological confidence score. The score calculation process can be represented as follows:

[0061]

[0062] in, This represents the input 512-dimensional deep feature vector. and These represent the weight matrix and bias term of the first fully connected layer, respectively. Represents the ReLU activation function. and These are the parameters of the second fully connected layer. This represents the sigmoid function, which ultimately outputs a confidence score. This score quantifies the confidence level that the current image patch contains pneumoconiosis shadows; a higher score indicates a higher degree of match between the morphological features and typical pneumoconiosis shadows.

[0063] The non-maximum suppression algorithm employs a spatial overlap criterion to handle multiple suspicious shadow clusters. The algorithm first sorts all shadow clusters in descending order of their confidence scores. When processing the queue, the highest-scoring cluster is selected as the baseline, and the intersection-over-union (IoU) ratio between its bounding box and the bounding boxes of the remaining clusters is calculated. When the IoU value exceeds a preset threshold of 0.3, the corresponding low-scoring cluster is marked as suppressed. This process is iterated until all clusters have been processed. The final retained shadow clusters constitute the filtering result set; these clusters have sufficient spatial discriminative power while maintaining high morphological confidence.

[0064] The coordinate alignment operation maps the filtered shadow clusters back to the original image space. This process is based on a spatial transformation matrix, which records the transformation relationship from the cropped image patch coordinates to the original image coordinates. For each retained shadow cluster, its centroid coordinates are repositioned to their precise location in the original image through an inverse transformation. Simultaneously, all relevant geometric feature parameters (such as area and boundary shape) are adjusted accordingly based on the coordinate transformation ratio to ensure consistency of feature values. The final optimized small shadow feature matrix is ​​stored in a structured format, containing fields such as: cluster ID, original image coordinates (x, y), confidence score, area adjustment value, and boundary complexity index.

[0065] At the implementation level, hardware acceleration utilizes the CUDA parallel computing architecture. The convolutional operations of the feature encoder are optimized using the cuDNN library, with a batch size set to 32. The confidence predictor's parameter initialization employs the Xavier method to avoid the vanishing gradient problem. The IoU calculation of the non-maximum suppression algorithm is accelerated through vectorization operations. Bilinear interpolation is used in the coordinate alignment process to maintain geometric accuracy. For memory management, deep feature representations use float16 half-precision storage to save space. TensorRT optimization is enabled during the network inference stage to reduce computational latency. Preprocessing of input image patches includes grayscale normalization and local contrast enhancement. An anomaly handling mechanism monitors the distribution range of confidence scores, triggering a recalculation process when extreme values ​​occur. The parameter configuration interface allows adjustment of the confidence threshold and the non-maximum suppression overlap threshold. The data output format is compatible with the DICOM standard and includes original coordinate reference information. The processing log records the detailed computation path of each shadow cluster, supporting result traceability. Performance optimization measures include asynchronous data transmission and pipelined parallel processing to improve overall throughput. A quality control module periodically verifies coordinate alignment accuracy to ensure spatial positioning accuracy. The system resource monitoring system tracks GPU memory usage in real time to prevent overflow errors. During implementation, the weight parameters of the feature encoder are initialized through pre-training and transfer learning is performed on the pneumoconiosis image dataset. The confidence predictor employs a balanced sampling strategy to ensure an appropriate ratio of positive to negative samples. The overall processing time of the dynamic confidence feedback mechanism is controlled within 800 milliseconds per image, meeting clinical real-time requirements. The results visualization component overlays optimized small shadow features onto the original image, using color gradients to represent confidence levels, assisting physicians in review. System integration testing verifies image compatibility with X-ray equipment from different manufacturers, ensuring broad applicability. The maintenance interface provides online update functionality for the feature encoder, supporting iterative model optimization. Security mechanisms implement encrypted data transmission and access control to protect patient privacy.

[0066] See Figure 5 This embodiment involves the operational flow of spatial relationship modeling networks and pathological feature database matching, illustrating its implementation process through a specific example. Taking a clinical chest X-ray image as an example, the image size is 2048×2048 pixels, with a bit depth of 16 bits. It has been processed using the aforementioned embodiment to obtain an optimized small shadow feature matrix, containing feature data for 6 suspicious shadow clusters. The specific contents of the shadow feature matrix are shown in the table below:

[0067]

[0068] After receiving the feature matrix, the spatial relationship modeling network first extracts the spatial coordinates of all shadow clusters. Taking cluster C1 (456, 789) as an example, the system establishes a graph structure data model, treating each shadow cluster as a graph node. Node attributes include coordinate position, confidence score, and morphological features. Edge connections between nodes are established based on the principle of spatial proximity. The Euclidean distance between all node pairs is calculated, and a connection edge is established when the distance is less than 100 pixels. The edge weight is set to the reciprocal of the distance, ensuring a stronger connection between nodes that are closer together. In this example, cluster C1 is 58 pixels away from C2 and 34 pixels away from C3, both satisfying the connection condition; therefore, two edges are established with weights of 0.017 and 0.029, respectively.

[0069] The graph neural network model employs a two-layer graph convolutional architecture to process the graph data. The first graph convolutional layer takes 5 input node features (including coordinates, confidence, area, boundary complexity, and grayscale value) and expands the output dimension to 32. The feature update process for each node aggregates the weighted features of its neighboring nodes, with weights derived from edge connection strength. The second graph convolutional layer compresses the feature dimension to 16, highlighting key spatial relationships. After graph convolution processing, the feature vector of node C1 is updated to a 16-dimensional representation containing spatial context information, where the values ​​of each dimension reflect the node's relative position and feature associations within the overall spatial distribution.

[0070] The topological association matrix is ​​generated based on the processed node features. The cosine similarity of each pair of node features is calculated, with a similarity threshold set at 0.7. Node pairs with a similarity higher than this value are considered to have a significant topological association. In this example, clusters C1 and C2 have a similarity of 0.83, and clusters C1 and C3 have a similarity of 0.79, both exceeding the threshold, and are therefore marked as strongly associated in the matrix. However, clusters C1 and C4 have a similarity of only 0.32, so no association is established. The final generated topological association matrix is ​​a 6×6 symmetric matrix, with each element value being a binary label (1 indicating association, 0 indicating no association).

[0071] During the pathological feature database matching phase, the system accesses a pre-built standard pneumoconiosis shadow feature database. This database contains three typical pneumoconiosis shadow distribution patterns: diffuse, nodular, and mixed. Each pattern stores a standard topological graph structure and feature vectors. For example, the feature vector for the nodular pattern emphasizes the roundness index and spatial clustering density of small shadows. Key path features are extracted from the current topological association matrix. Specifically, the system identifies the largest connected subgraph, which in this example is a triangular connection structure composed of {C1, C2, C3}; and calculates the feature parameters of this subgraph, including average node degree (2.0), clustering coefficient (0.87), and average path length (1.0).

[0072] The extracted sub-image features are compared with a standard library for similarity. The similarity calculation uses a feature vector-by-feature comparison method. The current example sub-image shows an 82% match with the nodular pattern in the standard library, exceeding the preset 75% matching threshold. Therefore, this group of shadow clusters is determined to conform to the pathological characteristics of pneumoconiosis. The system outputs a set of coordinates for the target shadow region, including the precise location information of clusters C1, C2, and C3. Simultaneously, the matched pattern type is recorded as nodular, serving as a reference for subsequent diagnostic analysis.

[0073] In terms of implementation details, spatial relationship modeling is implemented using the PyTorchGeometric graph neural network framework. The pathological feature library is stored in a graph database, supporting fast topological queries. Node feature normalization ensures consistent numerical ranges across dimensions. Distance thresholds for edge connectivity conditions are dynamically adjusted based on image resolution. Mean pooling is used as the aggregation function in the graph convolutional layers. The topological association matrix is ​​stored in a sparse format to optimize space efficiency. Feature comparison during the matching process is accelerated through parallel computation. An anomaly handling mechanism monitors the numerical validity of node features to prevent overflow errors. The results visualization component marks associated shadow clusters with the same color, intuitively displaying spatial distribution patterns. The quality control module periodically verifies the connectivity of the graph structure to ensure the accuracy of topological analysis. The system parameter configuration interface allows for adjustment of distance and similarity thresholds. The data processing pipeline is automated, requiring no manual intervention from feature matrix input to target region output. During implementation, node feature updates are incrementally processed to reduce memory consumption. Edge weight calculations are cached to avoid redundant computations. The pathological feature library supports dynamic expansion, allowing the addition of new standard patterns at any time. A security mechanism implements access control to protect sensitive patient data. Performance monitoring records the time consumed in each step, which is used for system optimization analysis.

[0074] This embodiment relates to the operation process of the 3D reconstruction engine and the hierarchical fusion module.

[0075] A continuous sequence of X-ray images is acquired, containing multi-angle projection images of the patient during inspiration and expiration. The system loads the spatial coordinates of the target shadow region and parses the device scanning parameters using DICOM metadata. The 3D reconstruction engine processes the image sequence using a voxel reconstruction algorithm. The algorithm first generates an empty voxel space with a resolution of 512×512×256 voxel units, each initialized to zero. All input images are traversed, and ray paths are calculated based on projection geometry parameters. Distance-weighted interpolation is applied as the rays traverse the voxel space, and the attenuation coefficient of the penetrating ray is accumulated for each voxel value. A multi-resolution strategy is implemented during the reconstruction process: a low-resolution approach is used initially to quickly construct the global structure, and after two iterations, a high-resolution approach is used for refined processing. The axial sampling interval automatically adapts to the device slice thickness parameters, with a typical value of 0.5 mm.

[0076] Post-processing of the 3D volumetric data includes grayscale gradient calculation and anisotropic diffusion optimization. Grayscale gradient features are extracted using a 3D Sobel operator, with 3×3×3 convolution weights configured in the X, Y, and Z directions of the operator kernel. The magnitude and direction of the gradient vector are calculated at each voxel location, and the results are stored as a four-dimensional array (X, Y, Z, Mag). Anisotropic diffusion filters are implemented based on this gradient data, with the diffusion coefficient function dynamically adjusted according to the gradient magnitude. The filter undergoes five iterations: the first iteration smooths uniform regions with a large diffusion scale; subsequent iterations gradually shrink the diffusion range; and the final iteration smooths only weak gradient regions. The diffusion equation is solved using implicit discretization, and the boundary conditions are set to Neumann zero-flux conditions. The signal-to-noise ratio of the processed 3D data is optimized, and the boundary continuity of small shadow regions is enhanced.

[0077] The generation of the small shadow volume density heatmap establishes a grayscale-density conversion model. The model defines a piecewise mapping relationship between gradient magnitude and density value: magnitude 0-50 corresponds to density 0-0.3; 50-150 maps to 0.3-0.7; and values ​​above 150 map to 0.7-1.0. This mapping is performed on each voxel to generate density distribution volume data. The heatmap color encoding uses HSV color space conversion: density value is used as the hue channel parameter, with a blue-green-yellow-red gradient representing density from low to high; saturation is fixed at 90%; the brightness channel is adjusted according to the original grayscale value. The generated 3D heatmap dataset contains a three-channel array of voxel coordinates, density values, and RGB color values.

[0078] The hierarchical fusion module performs spatial coordinate transformation operations. This module loads the original reference X-ray image and analyzes its DICOM coordinate system parameters. It calculates the projection transformation matrix from the 3D thermal map space to the 2D image plane. It transforms and integrates the device's geometric parameters: source-image distance, source-object distance, and detector tilt angle. Using a perspective projection model, the 3D voxel coordinates are converted to 2D homogeneous coordinates, and then aligned to the original image grid through affine transformation. The projection process implements a ray tracing algorithm: virtual rays are emitted from the X-ray source point, passing through the 3D voxel space, accumulating the density and color values ​​along the path to generate an intermediate projection map.

[0079] The fusion process employs multi-channel synthesis. The original image serves as the background layer, preserving grayscale anatomical structure; the projected heatmap serves as the foreground layer, adding pseudo-color information. The fusion strategy utilizes alpha blending: low-density areas have a transparency of 70%, while high-density areas have a transparency of 30%. Color space conversion ensures the fused image retains the grayscale contrast of the original image. An enhanced small shadow distribution map is generated, featuring a dual-view display mode: the main view displays the fused image, while the secondary view provides a 3D heatmap for rotational viewing.

[0080] The generation of structured diagnostic reports is based on density zoning statistics. A grid density counting system is established in the projected coordinate system: the lung field is divided into upper, middle, and lower zones, and each zone is further divided into inner, middle, and outer bands. The system automatically identifies the distribution of small opacities within each zone: continuous areas with a density value greater than 0.3 are considered valid opacities. Statistical indicators include the total number of pixels of opacities in each zone, the largest continuous opacity size, the average density value, and the coordinates of the distribution centroid. The report data is structured in XML format, including patient information, device parameters, statistical data, and image reference links. The report generation module also creates a JPG format distribution map thumbnail, embedded in the XML document for easy viewing.

[0081] The system employs a layered architecture. The 3D reconstruction engine utilizes CUDA kernel functions to accelerate voxel computation. Heatmap generation utilizes OpenGL shaders for real-time rendering. The coordinate transformation module integrates the ITK library for spatial registration. An anomaly handling mechanism monitors for ray projection anomalies during voxel reconstruction. A quality control subsystem verifies 3D reconstruction accuracy through phantom testing. The processing parameter configuration interface provides manual adjustment functions for diffusion intensity and color mapping curves. Performance monitoring logs record the processing time for each image, with an average time of 3.5 seconds for the reconstruction phase and 1.2 seconds for the fusion phase. Result data storage employs a layered compression strategy, with original projection data archived independently. A security mechanism implements patient information anonymization, complying with medical data protection standards. The output interface is compatible with PACS system transmission protocols, supporting automatic uploading of diagnostic reports and distribution maps. The system maintenance module provides offline calibration functionality for the reconstruction engine, ensuring long-term operational accuracy and stability.

[0082] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0083] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A chest X-ray image processing method for AI small shadow detection rate of pneumoconiosis, characterized by, The method comprises the following steps: acquiring a chest X-ray original image to be processed, performing initial segmentation network processing on the original image to generate a basic mask containing a lung region; using a hierarchical optimization unit to perform boundary refinement correction on the basic mask to output an accurate lung profile; analyzing the gray scale distribution features in the accurate lung profile by a multi-scale feature extraction module to construct an initial small shadow probability distribution map; processing the initial small shadow probability distribution map based on a morphological structure analyzer to identify discrete shadow regions and label suspicious shadow clusters; using a dynamic confidence feedback mechanism to evaluate the morphological feature confidence of the suspicious shadow clusters to generate an optimized small shadow feature map; inputting the optimized small shadow feature map into a spatial relationship modeling network to establish a topological correlation matrix between the shadow clusters; matching the topological correlation matrix with a pre-set dust pneumo-pathological feature library to screen target shadow regions that meet the clinical pathological features; processing the target shadow regions using a three-dimensional reconstruction engine to generate a small shadow volume density heat map; integrating the small shadow volume density heat map and the spatial coordinate information of the original image by a hierarchical fusion module to output an enhanced small shadow distribution atlas; generating a structured diagnostic report data based on the enhanced small shadow distribution atlas; The operation process of the dynamic confidence feedback mechanism is: constructing a shadow feature evaluation network, which includes a feature encoder and a confidence predictor; inputting the original image block of the suspicious shadow cluster into the feature encoder to generate a deep feature representation; calculating the morphological confidence score of the deep feature representation by the confidence predictor, wherein the score value quantifies the confidence level of the current image block containing dust pneumo-small shadows, and the higher the value, the higher the matching degree of the morphological features with the typical dust pneumo-shadows; using a non-maximum suppression algorithm to screen suspicious shadow clusters with a morphological confidence score higher than a critical threshold; aligning the screening results with the original image coordinates to generate an optimized small shadow feature matrix.

2. The AI small shadow detection rate of chest X-ray image processing method for pneumoconiosis according to claim 1, characterized in that, The initial segmentation network processing specifically comprises: constructing a dual-channel convolutional neural network as a segmentation backbone network, which includes a spatial path and a context path working in parallel; extracting high-resolution edge features of the original image through the spatial path to generate an edge feature tensor; capturing global semantic features of the original image using the context path to generate a semantic feature tensor; merging the edge feature tensor and the semantic feature tensor using an adaptive feature fusioner to generate a fused feature map; performing multi-level supervised training on the fused feature map to update the weight parameters of the segmentation backbone network through a back propagation algorithm; outputting the binary image of the trained and converged segmentation backbone network as the basic mask.

3. The pneumoconiosis-oriented AI small shadow detection rate chest X-ray image processing method of claim 2, characterized by, The operation of the hierarchical optimization unit comprises: receiving the basic mask and performing a contour tracking algorithm to extract a lung boundary pixel chain; calculating local curvature abnormal points of the lung boundary pixel chain using a curvature analysis model; constructing a boundary correction vector field based on the local curvature abnormal points; The boundary correction vector field is optimized by an iterative energy minimization algorithm to drive the lung boundary pixel chain to move towards an anatomical standard position; Precise lung profile data containing a smooth continuous contour is outputted.

4. The AI small shadow detection rate of chest X-ray image processing method for pneumoconiosis according to claim 1, characterized in that, The operation steps of the multi-scale feature extraction module are: A multi-resolution pyramid structure is established within the precise lung profile; A Gaussian difference filter set is used to process the lung region of the original image to generate a multi-scale gradient amplitude map; Local texture features of the multi-scale gradient amplitude map are extracted to construct a rotation-invariant texture descriptor; A similarity matrix of the rotation-invariant texture descriptor and a standard pneumoconiosis texture template is calculated; The similarity matrix and the spatial weight of the multi-scale gradient amplitude map are fused to generate initial small shadow probability distribution data.

5. The AI small shadow detection rate of a chest X-ray image processing method for pneumoconiosis according to claim 1, characterized in that, The morphological structure analyzer performs the following operations: Adaptive threshold segmentation is performed on the initial small shadow probability distribution map to obtain a binary candidate region; A connected component labeling algorithm is used to process the binary candidate region to generate discrete shadow region labels; The geometric and morphological feature vector of each discrete shadow region is calculated, which includes area irregularity and boundary fractal dimension; The spatial clustering characteristics of the geometric and morphological feature vector are analyzed based on a density clustering algorithm to label suspicious shadow clusters that meet the clustering constraints.

6. The AI small shadow detection rate of chest X-ray image processing method for pneumoconiosis according to claim 1, characterized in that, The operation of the spatial relationship modeling network includes: The optimized small shadow feature matrix is received and the spatial position coordinates are extracted; A graph neural network model is constructed, with each shadow cluster as a node and the Euclidean distance between nodes as the edge weight; The spatial dependence relationship between nodes is learned through a graph convolution layer to update the node feature representation; The topological correlation strength between the updated node feature representation and adjacent nodes is calculated; A topological correlation matrix data containing the connection strength between nodes is outputted.

7. The AI small shadow detection rate of chest X-ray image processing method for pneumoconiosis according to claim 1, characterized in that, The pathological feature library matching process includes: A standard pneumoconiosis shadow feature library is pre-constructed, which stores the shadow distribution patterns of typical pneumoconiosis cases; Key path feature vectors in the topological correlation matrix are extracted; The cosine similarity of the key path feature vectors and the standard pneumoconiosis shadow feature library is calculated; Topological subgraphs with a cosine similarity greater than a matching threshold are screened; The topological subgraphs are mapped back to the original image space to output a target shadow region coordinate set.

8. The AI small shadow detection rate of chest X-ray image processing method for pneumoconiosis according to claim 1, characterized in that, The operation steps of the three-dimensional reconstruction engine are: An X-ray image sequence corresponding to the target shadow region is obtained for consecutive layers; A voxel reconstruction algorithm is used to process the X-ray image sequence to generate three-dimensional body data; The gray gradient features of each voxel in the three-dimensional body data are calculated; The signal-to-noise ratio of the three-dimensional body data is optimized based on an anisotropic diffusion filter; A three-dimensional heat map data labeled with small shadow spatial distribution is generated.

9. The pneumoconiosis-oriented AI small shadow detection rate chest X-ray image processing method of claim 8, characterized by, The hierarchical fusion module performs: The three-dimensional heat map data and the spatial coordinate transformation parameters of the original image are received; A conversion function for mapping small shadow density to color is established; The three-dimensional heat map data is projected to a two-dimensional diagnostic plane through image registration algorithm; The projection data and the anatomical structure features of the original image are fused to generate an enhanced small shadow distribution atlas; According to the density partition statistical result of the enhanced small shadow distribution map, a structured report containing shadow quantity and distribution position is generated.

Images