A lung ventilation-perfusion map image segmentation and quantitative analysis system

Abstract

The application discloses a lung ventilation-perfusion imaging area image segmentation and quantitative analysis system and relates to the technical field of medical image analysis.The system comprises a data acquisition module, which acquires SPECT V / Q image data and case information corresponding to the SPECT V / Q image data, and generates a structured data set; a feature extraction module, which performs format conversion and multi-scale feature extraction on the SPECT V / Q image data in the structured data set, and generates a compressed feature map data set; a lung lobe segmentation module, which inputs the compressed feature map data set into a U-Net model, segments lung lobe areas, performs post-processing, and generates a lung lobe area mask image; and an image fusion module, which performs alpha blending fusion on the lung lobe area mask image and the SPECT V / Q image data, and generates a color fusion image.The application captures lung alveolus-to-lung lobe structure by fusing four branch features of a hollow pyramid network, and reduces GPU memory consumption.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical image analysis, in particular to a lung ventilation-perfusion imaging region image segmentation and quantitative analysis system. BACKGROUND

[0002] In recent years, the field of medical image analysis has made significant progress driven by deep learning technology, especially in the diagnosis of lung diseases. Segmentation models such as U-Net have achieved automatic division of lung lobe regions. Existing technologies mainly extract features from a single image modality and combine traditional clustering algorithms to quantify ventilation blood flow ratio (V / Q).

[0003] However, conventional methods rely on isolated image data (such as SPECT V / Q imaging), do not integrate clinical medical record information, resulting in a disconnect between functional analysis and pathological background; the feature extraction network lacks a multi-scale fusion mechanism, shallow edge features and deep semantic features interact insufficiently, and ignores feature distillation compression, causing waste of computing resources and missed detection of small lesions. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides a lung ventilation-perfusion imaging region image segmentation and quantitative analysis system to solve the problem of limited lung function partition quantitative analysis precision caused by the lack of cooperation between multi-modal medical images and clinical information.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] In a first aspect, the present application provides a lung ventilation-perfusion imaging region image segmentation and quantitative analysis system, comprising,

[0008] A data acquisition module acquires SPECT V / Q image data and medical record information corresponding to the SPECT V / Q image data, and generates a structured data set;

[0009] A feature extraction module performs format conversion and multi-scale feature extraction on the SPECT V / Q image data in the structured data set, and generates a compressed feature map data set;

[0010] A lung lobe segmentation module inputs the compressed feature map data set into a U-Net model to segment lung lobe regions and perform morphological post-processing, generating a lung lobe region mask image;

[0011] An image fusion module performs alpha blending fusion of the lung lobe region mask image and the SPECT V / Q image data, generating a color fusion image;

[0012] The feature enhancement module performs Min-Max normalization processing and luminance feature value calculation on each channel of the color fusion image, highlights useful information while suppressing noise interference, and generates an enhanced feature tensor by performing standardization and luminance-based nonlinear transformation on the pixel values of each channel and performing feature enhancement.

[0013] The clustering analysis module divides the enhanced feature tensor into sub-regions by using a K-means clustering algorithm, calculates the V / Q ratio of each sub-region and maps it to the CT image, and generates a quantitative analysis report.

[0014] As a preferred scheme of the lung ventilation-perfusion imaging region image segmentation and quantitative analysis system, the format conversion and multi-scale feature extraction of the SPECT V / Q image data to generate a compressed feature map dataset include the following steps,

[0015] The SPECT V / Q image data is converted into a PNG format image;

[0016] The PNG format image is input into a hollow pyramid network to capture a multi-scale receptive field and output a multi-scale feature map;

[0017] The shallow edge feature map and the deep semantic feature map in the multi-scale feature map are extracted and spliced through a feature fusion channel to obtain a fusion feature map;

[0018] The fusion feature map is compressed by a depth separable convolution layer for feature distillation to generate a compressed feature map dataset.

[0019] As a preferred scheme of the lung ventilation-perfusion imaging region image segmentation and quantitative analysis system, the format conversion and multi-scale feature extraction of the SPECT V / Q image data to generate a compressed feature map dataset include the following steps,

[0020] The encoder path processes the compressed feature map dataset to generate a reduced dimension feature map;

[0021] The decoder path up-samples the reduced dimension feature map and fuses it with the encoder feature jump connection to output a probability distribution map;

[0022] The probability distribution map is morphologically post-processed to output a lung lobe region mask image.

[0023] As a preferred scheme of the lung ventilation-perfusion imaging region image segmentation and quantitative analysis system, the morphological post-processing includes,

[0024] The erosion operation is performed on the lung lobe region mask image to eliminate isolated noise points to obtain a denoising mask image.

[0025] Perform a dilation operation on the denoised mask image to fill the internal holes and output a continuous region mask image;

[0026] Edge patching is used to repair boundary breaks in continuous region mask images.

[0027] As a preferred embodiment of the lung ventilation-perfusion imaging region image segmentation and quantification analysis system of the present invention, the generation of the color fusion image includes the following steps.

[0028] Convert the masked image of the lung lobe region into a multi-channel grayscale image to generate a grayscale mask image;

[0029] The grayscale mask image is spatially registered with the SPECT V / Q image by feature matching, and the registered image pair is output.

[0030] Based on the anatomical structure of the lung lobes, the registered image pairs are dynamically fused in different regions. Adaptive fusion weights are calculated according to the spatial positions of the left and right lung lobes. The left lung region is weighted using the first weight coefficient, and the right lung region is weighted using the second weight coefficient to generate a color fused image.

[0031] As a preferred embodiment of the lung ventilation-perfusion imaging region image segmentation and quantification analysis system of the present invention, the step of performing Min-Max normalization processing and brightness feature value calculation includes the following steps.

[0032] Calculate the normalization parameters independently for each channel of the color fusion image and output the normalized channel data;

[0033] Luminance feature values ​​are calculated based on normalized channel data to generate a luminance feature map.

[0034] As a preferred embodiment of the lung ventilation-perfusion imaging region image segmentation and quantification analysis system of the present invention, the feature enhancement, generating enhanced feature tensors, includes the following steps:

[0035] Invalid pixel data in the brightness feature map are excluded, and channel-wise Min-Max normalization is performed on the valid pixel data to obtain a normalized pixel matrix.

[0036] Multi-scale convolution operations are applied to the standardized pixel matrix to extract local texture features at different scales and integrate them to generate a texture feature vector;

[0037] Multi-scale global context features are captured from the normalized pixel matrix using spatial pyramid pooling operations, and a context feature vector is output.

[0038] The texture feature vector and the context feature vector are fused, and an enhanced feature tensor is generated by highlighting useful information and suppressing noise interference through nonlinear transformation.

[0039] As a preferred scheme of the lung ventilation-perfusion imaging region segmentation and quantitative analysis system, the step of using the K-means clustering algorithm to divide the enhanced feature tensor into sub-regions comprises the following steps,

[0040] Extracting the spatial coordinates, brightness values and texture feature vectors in the enhanced feature tensor to construct a multi-dimensional feature vector set;

[0041] Setting the number of clusters according to the symmetry of the left and right lung anatomical structures, and initializing the spatially uniformly distributed cluster center points;

[0042] Calculating the Euclidean distance of each feature vector in the multi-dimensional feature vector set to each cluster center, assigning cluster labels based on the nearest neighbor principle, and generating an initial sub-region division map;

[0043] According to the initial sub-region division map, iteratively updating the cluster center position to the mean value of the feature vectors in the cluster, and optimizing the cluster boundary distribution;

[0044] When the cluster boundary distribution of continuous multiple iterations is lower than the boundary threshold, the calculation is terminated, and the cluster label map of the left and right lung function sub-regions is output.

[0045] As a preferred scheme of the lung ventilation-perfusion imaging region segmentation and quantitative analysis system, the step of calculating the V / Q ratio of each region and mapping it to the CT image comprises the following steps,

[0046] Based on the cluster label map, calculating the ventilation and perfusion intensity mean value of each sub-region, and outputting the V / Q ratio;

[0047] Mapping the V / Q ratio to the CT image to generate a discrete color mapping map;

[0048] Converting the discrete color mapping map to a continuous heat map through an interpolation algorithm.

[0049] As a preferred scheme of the lung ventilation-perfusion imaging region segmentation and quantitative analysis system, the CT image refers to a computer tomography image providing a lung anatomical structure reference.

[0050] The present application has the beneficial effects that: the SPECT V / Q imaging is dynamically associated with the electronic medical record, a structured data set is constructed, pathological related features are screened through the attention mechanism, and the differential accuracy of pulmonary embolism and COPD is significantly improved; the hollow pyramid network fuses four branch features to capture the alveolar to pulmonary lobe structure, and the feature dimension is compressed through the depth separable convolution, so that the micro lesions such as ground glass nodules are retained while the GPU memory consumption is reduced; based on the enhanced feature tensor, the functional sub-area is automatically divided, the V / Q thermogram is generated through RANSAC optimization registration and cubic spline interpolation, and the manual setting error is eliminated. BRIEF DESCRIPTION OF DRAWINGS

[0051] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0052] Figure 1 It is a schematic diagram of a lung ventilation-perfusion imaging region segmentation and quantitative analysis system.

[0053] Figure 2 It is a schematic diagram of a structured data set generation.

[0054] Figure 3 It is a feature extraction schematic diagram.

[0055] Figure 4 It is a binary image example one after morphological processing.

[0056] Figure 5 It is a binary image example two after morphological processing.

[0057] Figure 6 It is a binary image example three after morphological processing.

[0058] Figure 7 It is a schematic diagram of color fusion image generation.

[0059] Figure 8 It is a schematic diagram of enhanced feature tensor generation.

[0060] Figure 9 It is a schematic diagram of three-dimensional lung V / Q ratio K-means clustering analysis. DETAILED DESCRIPTION

[0061] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification.

[0062] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. However, it will be apparent to one skilled in the art that the present application can be practiced without the specific details set forth in this description. In other instances, well-known methods, procedures, components, and circuits have not been described in detail so as not to unnecessarily obscure aspects of the present application.

[0063] Secondly, the "one embodiment" or "an embodiment" referred to herein means a specific feature, structure, or characteristic under discussion. Thus, "one embodiment" does not mean a single embodiment nor is it to be taken individually or selectively from other embodiments.

[0064] Reference is made to Figures 1-9 For one embodiment of the present application, the embodiment provides a lung ventilation-perfusion imaging region segmentation and quantitative analysis system, comprising the following steps:

[0065] A data acquisition module acquires SPECT V / Q image data and medical record information corresponding to the SPECT V / Q image data, and generates a structured data set.

[0066] Specifically, referring to Figure 2 The DICOM protocol interface is used to receive raw image data transmitted by the SPECT device in real time, including DICOM sequence files of the lung ventilation phase and DICOM sequence files of the lung perfusion phase, and output uncompressed raw DICOM data sets;

[0067] Based on the patient ID, the associated fields in the electronic medical record are automatically searched, including basic information, medical history characteristics, and laboratory indicators, and a structured case JSON file is output. When the field is missing, the default value NULL is automatically filled.

[0068] The basic information includes age, gender, height / weight (for example, BMI ≥ 18.5 triggers a marker); the medical history characteristics include chronic obstructive pulmonary disease (COPD) staging and pulmonary embolism (PE) history flag features; and the laboratory indicators include arterial oxygen partial pressure (PaO2) and D-dimer concentration (unit: μg / L).

[0069] The above information is obtained with the consent of the user and is used for legal purposes.

[0070] The raw DICOM data set and the structured medical record JSON file are subjected to data consistency verification, including image- medical record association, completeness, and timeliness verification;

[0071] Wherein, the image-medicine record association verification refers to that the patient ID is completely matched in the DICOM file header and the electronic medical record; the integrity verification refers to that the image quantity of the lung ventilation phase / lung perfusion phase is guaranteed to be consistent; the timeliness verification refers to that the image acquisition time and the medical record time difference is less than or equal to 24 hours;

[0072] The state code 200_OK is output after the verification passes, and the error type code (such as E101: ID mismatch) is returned when the verification fails.

[0073] The original DICOM data set and the structured medical record JSON file after the verification pass are associated with the index based on the examination number as the primary key, and the structured data set containing image data and medical record information is output.

[0074] The feature extraction module performs format conversion and multi-scale feature extraction on the SPECT V / Q image data in the structured data set to generate a compressed feature map data set.

[0075] Specifically, referring to Figure 3 , the original DICOM file in the structured data set is read, each DICOM slice is converted to PNG format, the original gray resolution (such as 16bit→8bit linear mapping) is retained during the conversion process, and the key metadata (such as window width / window level) in the DICOM file header is written into the Exif information area of the PNG file, and a standardized PNG image sequence is generated;

[0076] The standardized PNG image sequence is input into the atrous pyramid (ASPP) feature extraction network, and the atrous pyramid (ASPP) feature extraction network processes the standardized PNG image sequence through four parallel branches;

[0077] Among them, branch 1 applies a 3×3 convolution layer with a hole rate of 6 to capture large-scale context features;

[0078] Branch 2 applies a 3×3 convolution layer with a hole rate of 12 to extract super-large-scale semantic features;

[0079] Branch 3 applies a 3×3 convolution layer with a hole rate of 18 to perceive the global anatomical structure;

[0080] Branch 4 generates image-level features by compressing the spatial dimension through a global average pooling layer;

[0081] The output after the four parallel branch processing is aligned in channel through a 1×1 convolution layer, and then spliced and fused to generate a multi-scale fusion feature map;

[0082] The shallow edge feature map (resolution 128x128) of the third residual block is extracted from the multi-scale fusion feature map, and the deep semantic feature map (resolution 32x32) at the end of the network is extracted; the shallow edge feature map is up-sampled to 32x32 resolution by bilinear interpolation, and is spliced in the channel dimension with the deep semantic feature map; the spliced feature map is fused by a 3x3 convolution layer, and an enhanced feature map is output;

[0083] The enhanced feature map is input into a depth separable convolution layer, including a depth convolution stage and a point-wise convolution stage; wherein the depth convolution stage adopts a 3x3 convolution kernel for spatial filtering, and the point-wise convolution stage adopts a 1x1 convolution kernel for channel transformation; high-efficiency calculation is realized by separating space and channel operations, the feature map dimension is compressed, and finally a compressed feature map dataset is output.

[0084] The compressed feature map dataset is input into a U-Net model to segment the lung lobe region and perform morphological post-processing, generating a lung lobe region mask image.

[0085] Specifically, referring to Figures 4-6 The lung lobe region mask image is input into the encoder path of the U-Net model, and the encoder includes a five-level down-sampling structure. Each level extracts local features through two consecutive 3x3 convolution layers and a ReLU activation function, and then reduces the spatial resolution by half while doubling the channel number through a 2x2 max-pooling layer, gradually capturing global semantic features of the lung lobe anatomical structure. After five levels of down-sampling, a low-resolution feature map with a spatial size reduced to 1 / 32 of the original image is output, which contains deep semantic representations of the lung lobe region.

[0086] In the decoder path of the U-Net model, five up-sampling operations are performed on the low-resolution feature map. Each level first expands the low-resolution feature map size by one time through transposed convolution (convolution kernel 2x2, step 2), and then splices the channel dimension with the feature map of the corresponding level of the encoder through a jump connection. The spliced feature map fuses multi-scale information through two 3x3 convolution layers and a ReLU activation function, gradually restores the spatial resolution to the original size, and finally outputs a probability distribution map with the same size as the input image, wherein each pixel value in the probability distribution map represents the probability (0-1 interval) of belonging to the lung lobe region.

[0087] The output probability distribution map is binarized by a fixed threshold: pixels with a probability greater than or equal to the fixed threshold are marked as 1 (lung lobe region), and the rest are marked as 0 (background), generating an initial binary mask image; morphological opening operations, including erosion and dilation operations, are performed on the initial binary mask image.

[0088] It should be noted that the fixed threshold setting is based on the bimodal characteristics of the U-Net model output probability distribution, and the segmentation threshold is dynamically determined by maximizing the inter-class variance between the lung lobe region and the background, to ensure the robust separation of anatomical structures and noise.

[0089] Wherein, the erosion operation is to traverse the mask image with a 3x3 circular structure element to eliminate isolated noise points (such as tracheal artifacts); the dilation operation is to fill the small gaps (such as blood vessel interruption areas) generated by erosion with the same structure element, and output the lung lobe mask image optimized in connectivity;

[0090] The boundary broken area of the optimized mask image is detected, and the region growing algorithm is used for repair: taking the broken end point as the center, searching for the nearest adjacent boundary point along the 8-neighbor direction, generating a linear connection path within a maximum distance of 5 pixels, filling the closed area formed by the linear connection path and smoothing the transition, and outputting the lung lobe region mask image with complete anatomical structure, to ensure the continuity of the lung apex, lung base and interlobar fissure boundary.

[0091] The image fusion module alpha blends the lung lobe region mask image and the SPECT V / Q image data to generate a color fusion image.

[0092] Specifically, referring to Figure 7 , the lung lobe region mask image is converted into a three-channel gray image by a channel expansion algorithm: the lung lobe region pixels with a mask value of 1 are mapped to pure white, and the background region pixels are mapped to pure black, to generate a multi-channel gray mask image with the same size as the SPECT V / Q image, which retains the geometric accuracy and spatial information of the original mask, and ensures the anatomical consistency of subsequent registration; a feature point-based registration algorithm is used to align the multi-channel gray mask image and the original SPECT V / Q image, the lung lobe contour corner points in the gray mask image are extracted as feature points, and the texture feature points at the corresponding anatomical position of the SPECT V / Q image are extracted, the cosine similarity between the lung lobe contour corner points and the texture feature points is calculated by a FLANN (Fast Library for Approximate Nearest Neighbors) matcher to establish a corresponding relationship between the feature points; at the same time, the RANSAC algorithm is used to estimate the best affine transformation parameters to form an affine transformation matrix, to eliminate the interference of outlier matching points;

[0093] Further, the estimation of the optimal affine transformation parameters using the RANSAC algorithm refers to the following process. After the FLANN matcher establishes the correspondence between the feature points, the RANSAC algorithm is used to randomly select a minimum sample subset (e.g., 3 pairs of points) from the feature point pairs to calculate the initial affine transformation parameters and generate a candidate transformation matrix. Then, the spatial position deviation of all feature points under the candidate transformation matrix is calculated, and the points with a deviation less than a deviation threshold are determined as inliers and the number of inliers is counted. The sampling and inlier counting process is iteratively performed, and the calculation is terminated when the proportion of inliers does not improve for N consecutive iterations. The candidate transformation matrix with the largest number of inliers is selected as the optimal solution. Finally, the least squares method is used to fit the accurate affine transformation parameters based on the inlier set of the optimal solution, and a six-degree-of-freedom affine transformation matrix that eliminates the interference of outliers is output, which includes rotation, scaling, and translation parameters.

[0094] It should be noted that the deviation threshold is determined based on the average edge alignment error statistics of the lung parenchyma anatomical structure.

[0095] The affine transformation matrix is applied to resample the grayscale mask image to be fully spatially aligned with the SPECT V / Q image, and an accurately registered image pair is output. The partitioned dynamic Alpha transparency fusion processing is performed on the accurately registered image pair: the spatial positions of the left and right lung lobes are identified based on the lung lobe anatomical structure, the left lung region is weighted and fused using a first weight coefficient, and the right lung region is weighted and fused using a second weight coefficient. The SPECT V / Q image is taken as the base layer, the registration mask image is taken as the superimposed layer, the offset is set to 0 to ensure no brightness offset, the calculation is performed according to the pixel-by-pixel weighting formula, and the preliminary fused pixel value is output.

[0096] The preliminary fused pixel value is subjected to channel enhancement processing: in the lung lobe region, the red channel gain is enhanced to intensify the display, so that the ventilation / perfusion state is superimposed on the anatomical structure in a translucent red color, and the non-lung lobe region retains the original grayscale information, thereby generating a channel-enhanced image.

[0097] The channel-enhanced image is subjected to post-processing optimization: the Laplacian operator is used to intensify the lung lobe boundary profile to improve the recognition of the anatomical structure, the adaptive histogram equalization is used to optimize the contrast to highlight the perfusion abnormal region, and a scale and a direction marker are added, thereby finally generating a 24-bit true color fusion image, in which the low perfusion region is a deep red patch and the high ventilation region is a bright red uniform distribution.

[0098] The feature enhancement module performs Min-Max normalization processing and brightness feature value calculation on each channel based on the color fusion image. Through standardization and brightness-based nonlinear transformation of the pixel values of each channel, useful information is highlighted, noise interference is suppressed, and feature enhancement is performed, thereby generating an enhanced feature tensor.

[0099] Specifically, referring to Figure 8The color fusion image is decomposed into three independent channels of red (R), green (G) and blue (B); an independent normalization operation is performed on each channel: the effective pixel value range of each channel (excluding the background area) is calculated, the minimum and maximum reference points of the channel are determined; the pixel value is mapped to the [0, 1] interval through linear transformation, the relative density gradient characteristics of the lung area are retained, and a normalized three-channel floating-point data matrix is output;

[0100] The normalized three-channel floating-point data matrix is subjected to brightness feature conversion: according to the distribution characteristics of the human eye retina photoreceptor cells, different weight coefficients are given to the R, G and B channels; the weighted calculation result is taken as the pixel brightness value, a single-channel brightness feature map is generated, and a background threshold judgment mechanism is established: the average brightness of the edge area pixels of the image is calculated, the area with an average brightness lower than the average threshold is marked as an invalid area (including the chest cavity background and device artifacts), and a brightness feature map with invalid area marking is output;

[0101] According to the invalid area marking of the brightness feature map, a binary mask is created to identify the effective lung area: all pixel points are traversed, and the area with a brightness value ≥ the background threshold is identified as an effective area; morphological closing operation is applied to fill the small gaps in the effective area, to ensure the continuity of the lung tissue, the original data is filtered based on the binary mask, and the invalid area pixels are removed, and a standardized pixel matrix containing only the lung parenchyma is output;

[0102] A multi-scale convolution operation is performed on the standardized pixel matrix to capture texture features of different anatomical levels, as follows:

[0103] First stage: multi-scale convolution captures anatomical level features;

[0104] The lung CT image in the standardized pixel matrix is taken as input, different sizes of "lenses" (convolution kernels) are used to observe different levels of anatomical structures, 3x3 small-scale convolution kernels are used for small-range scanning to capture subtle gray-scale changes, and microscopic texture features of alveolar level are extracted, including ground glass nodule texture, which appears as a fuzzy shadow on CT like a piece of frosted glass, is an early sign of multiple lung diseases, and a small convolution kernel is most suitable for capturing such subtle texture changes;

[0105] A 5x5 medium-scale convolution kernel is used for medium-range scanning to capture mesoscopic structure features of pulmonary lobule level, and blood vessel branch patterns and accompanying structures are extracted, the pulmonary lobule is the basic structure and functional unit of the lung, and contains a fine bronchus and its accompanying blood vessels, and the 5x5 convolution kernel can well see whether the morphology, running and distribution of these small blood vessels are normal;

[0106] Large-scale 7x7 convolution kernel is used to scan a large range and perceive the macro-distribution characteristics of lung lobes, especially the gradient change pattern between ventilation and perfusion. Pulmonary embolism can cause blood flow (perfusion) to be blocked in the lung, while the airflow (ventilation) may be normal. This "mismatch" will form a specific density distribution gradient on the CT. The large convolution kernel can perceive the macro-distribution change;

[0107] The three sets of convolution outputs are processed by a nonlinear activation function, enhancing the feature expression capability, and integrating into a multi-scale texture feature vector set, comprehensively representing the structural heterogeneity of lung tissue at multiple scales.

[0108] Second stage: spatial pyramid pooling integrates global context;

[0109] The multi-scale texture feature vector set is input into the spatial pyramid pooling layer to realize multi-level global context information integration:

[0110] The first-level maximum pooling divides the feature maps in the multi-scale texture feature vector set into a 4x4 grid, extracts the maximum response value of each local region, captures the context features in the medium range, and highlights the significant areas (such as the most suspected lesion area;

[0111] The second-level average pooling uses a 2x2 grid division to divide the feature maps in the multi-scale texture feature vector set into 4 large blocks, takes the average value of each block, reflects the overall profile of the corresponding area, and captures the context distribution characteristics in a larger range;

[0112] The third-level global average pooling performs a global average operation, takes the average value of the entire feature map in the multi-scale texture feature vector set, generates an image-level summary vector, and represents the overall properties of the entire image;

[0113] The pooling outputs of the three levels are spliced to form a high-dimensional global context feature vector containing the overall lung function distribution characteristics.

[0114] Third stage: cross-modal fusion and gated attention mechanism;

[0115] The high-dimensional global context feature vector (knowing the "overall situation") and the multi-scale texture feature vector set (containing "local details") are interacted through the feature cross-attention mechanism to automatically determine "in a certain global context, which local texture features are more worthy of attention". For example, in the blood vessel distribution area, the texture of the alveoli does not need to be paid much attention to, while the morphological features of the blood vessels themselves are very important, and the response to pathological areas is strengthened;

[0116] The gated attention network is established to dynamically adjust the contribution ratio of global features and local features in the final result through learnable parameters, to give higher weight to the high-risk area of pulmonary embolism (such as the distribution area of pulmonary segmental artery); residual connection is introduced in the fusion process to retain the original feature information and ensure effective gradient propagation, and an enhanced feature tensor is output, in which the pathological area (such as the perfusion defect area caused by pulmonary embolism) presents high-intensity feature activation, which significantly improves the segmentation and identification performance of the downstream task.

[0117] It should be noted that the enhanced feature tensor contains multi-scale information of microscopic, mesoscopic and macroscopic features, realizing the unity of global and local context of knowing details and knowing the whole. Due to the attention mechanism and gating weighting, the pathological area caused by pulmonary embolism, such as the perfusion defect area, will present high-intensity feature activation in the enhanced feature tensor, like being highlighted, making the pathological area significantly enhanced, which is more suitable for downstream tasks. Because the features to be judged by the network have been made very obvious and prominent, as the input of the subsequent classifier or segmentation network, it can greatly improve the accuracy and reliability of pulmonary embolism detection.

[0118] The clustering analysis module uses the K-means clustering algorithm to divide the enhanced feature tensor into sub-regions, calculates the V / Q ratio of each sub-region and maps it to the CT image to generate a quantitative analysis report.

[0119] Specifically, referring to Figure 9 , based on the enhanced feature tensor, spatial coordinates, brightness values and texture feature vectors are extracted to construct a multi-dimensional feature vector set; 18 cluster centers (9 regions in the left lung / 9 regions in the right lung) are preset according to the anatomical symmetry of the left and right lungs, and the center point positions are initialized using a spatial uniform distribution strategy; the Euclidean distance of the feature vector to each center is calculated, and the cluster label is assigned by the nearest neighbor principle to generate an initial sub-region division map; the cluster center is iteratively updated to the feature mean position within the cluster to dynamically optimize the cluster boundary distribution; when the mean square deviation of the cluster boundary displacement of consecutive iterations is less than or equal to the variance threshold pixel, the calculation is terminated, and the final clustering label map conforming to the anatomical functional partition is output.

[0120] Based on the final clustering label map, the pixel set of each sub-region is extracted, the ventilation intensity mean (V) and perfusion intensity mean (Q) are calculated respectively, and the regional V / Q ratio matrix is output; the Jet color mapping rule is constructed:

[0121] Low V / Q area (such as <0.8) is mapped to dark red, indicating perfusion defect; normal area (such as 0.8-1.2) is mapped to yellow; high V / Q area (such as >1.2) is mapped to dark blue, indicating ventilation disorder; through cubic spline interpolation, the discrete ratio is converted into a continuous color gradient, which is superimposed on the CT image to generate a continuous heat map that integrates anatomical positioning and functional evaluation; wherein the CT image refers to the computer tomography image providing the anatomical structure of the lung as a reference;

[0122] Integrating continuous heat map with V / Q ratio matrix to build visual diagnostic framework: rendering heat map profile in coronal, sagittal and axial planes respectively, embedding dynamic table to display V / Q statistical values (mean / standard deviation / extreme value) of each lung lobe, adding interactive slider to adjust window width and window level, assisting in accurate positioning of lesions, outputting HTML5 diagnostic analysis framework supporting multi-dimensional browsing;

[0123] Inputting HTML5 diagnostic framework into Playwright headless browser engine, automatically converting it into PDF document conforming to DICOM SR standard; through OCR recognition, extracting key quantitative indicators in PDF and dynamically associating them with medical record database: associating D-dimer abnormal value of pulmonary embolism patients with low V / Q area; labeling emphysema area of COPD patients with high V / Q correlation, and inserting risk grading prompt (high risk / medium risk / low risk) at the same time, outputting structured initial diagnosis report with clinical decision support labeling;

[0124] Integrating structured initial diagnosis report, patient basic information and laboratory indicators to build final diagnosis report, and embedding patient basic information matrix (age / sex / BMI) in the header of the final diagnosis report, associating heat map with laboratory abnormal indicators (such as PaO2<80mmHg red warning) in the image analysis area, and generating natural language description in the diagnosis conclusion area: such as "2.1cm×1.8cm perfusion defect area (V / Q=0.63) is found in the left lower lobe, combined with elevated D-dimer (624μg / L), which is consistent with the image findings of pulmonary embolism", outputting quantitative analysis report.

[0125] In summary, the present application dynamically associates SPECT V / Q imaging with electronic medical records, builds structured data sets, and filters pathologically relevant features through a gated attention mechanism, significantly improving the accuracy of pulmonary embolism and COPD differentiation; the hollow pyramid network fuses four branch features to capture alveolar to lobar structure, and through deep separable convolution compression feature dimension, it retains micro lesions such as ground glass nodules while reducing GPU memory consumption; based on enhanced feature tensor, it automatically divides functional sub-regions, optimizes registration and cubic spline interpolation through RANSAC to generate anatomically calibrated V / Q heat map, eliminating manual setting errors.

[0126] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit it, although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, which should be covered in the scope of the claims of the present application.

Claims

1. A system for segmenting and quantifying lung ventilation-perfusion imaging regions, characterized in that: include, The data acquisition module acquires SPECT V / Q image data and the corresponding medical record information, and generates a structured dataset. The feature extraction module performs format conversion and multi-scale feature extraction on the SPECT V / Q image data to generate a compressed feature map dataset; The lung lobe segmentation module inputs the compressed feature map dataset into the U-Net model to segment the lung lobe region and performs morphological post-processing to generate a lung lobe region mask image. The image fusion module performs alpha blending on the masked image of the lung region and the SPECT V / Q image data to generate a color fused image; The feature enhancement module performs Min-Max normalization processing and brightness feature value calculation on a channel-by-channel basis based on the color fusion image. By standardizing the pixel values ​​of each channel and performing a nonlinear transformation based on brightness, it highlights useful information, suppresses noise interference, and performs feature enhancement to generate an enhanced feature tensor. The clustering analysis module uses the K-means clustering algorithm to divide the enhanced feature tensor into sub-regions, calculates the V / Q ratio of each sub-region and maps it to the CT image, and generates a quantitative analysis report.

2. The lung ventilation-perfusion imaging region image segmentation and quantification analysis system as described in claim 1, characterized in that: The process of converting the format of the SPECT V / Q image data and extracting multi-scale features to generate a compressed feature map dataset includes the following steps: Convert the SPECT V / Q image data to generate a PNG format image; Input a PNG format image into a hollow pyramid network to capture multi-scale receptive fields and output multi-scale feature maps. The shallow edge feature map and the deep semantic feature map are extracted from the multi-scale feature map and then concatenated through the feature fusion channel to obtain the fused feature map; The fused feature map is compressed by feature distillation through depthwise separable convolutional layers to generate a compressed feature map dataset.

3. The lung ventilation-perfusion imaging region image segmentation and quantification analysis system as described in claim 1, characterized in that: The process of inputting the compressed feature map dataset into the U-Net model to segment the lung lobe region and performing morphological post-processing to generate a lung lobe region mask image includes the following steps: The encoder path processes the compressed feature map dataset to generate a dimensionality-reduced feature map; The decoder path upsamples the dimensionality-reduced feature map and fuses it with the encoder feature skip connections to output a probability distribution map. Morphological post-processing is performed on the probability distribution map to output a masked image of the lung lobe region.

4. The lung ventilation-perfusion imaging region image segmentation and quantification analysis system as described in claim 3, characterized in that: The morphological post-processing includes, An erosion operation is performed on the masked image of the lung lobe region to remove isolated noise points, resulting in a denoised mask image; Perform a dilation operation on the denoised mask image to fill the internal holes and output a continuous region mask image; Edge patching is used to repair boundary breaks in continuous region mask images.

5. The lung ventilation-perfusion imaging region image segmentation and quantification analysis system as described in claim 1, characterized in that: The generation of the color fused image includes the following steps: Convert the masked image of the lung lobe region into a multi-channel grayscale image to generate a grayscale mask image; The grayscale mask image is spatially registered with the SPECT V / Q image by feature matching, and the registered image pair is output. Based on the anatomical structure of the lung lobes, the registered image pairs are dynamically fused in different regions. Adaptive fusion weights are calculated according to the spatial positions of the left and right lung lobes. The left lung region is weighted using the first weight coefficient, and the right lung region is weighted using the second weight coefficient to generate a color fused image.

6. The lung ventilation-perfusion imaging region image segmentation and quantification analysis system as described in claim 1, characterized in that: The process of performing Min-Max normalization and calculating luminance feature values ​​includes the following steps: Calculate the normalization parameters independently for each channel of the color fusion image and output the normalized channel data; Luminance feature values ​​are calculated based on normalized channel data to generate a luminance feature map.

7. The lung ventilation-perfusion imaging region image segmentation and quantification analysis system as described in claim 1, characterized in that: The process of performing feature enhancement and generating enhanced feature tensors includes the following steps: Invalid pixel data in the brightness feature map are excluded, and channel-wise Min-Max normalization is performed on the valid pixel data to obtain a normalized pixel matrix. Multi-scale convolution operations are applied to the standardized pixel matrix to extract local texture features at different scales and integrate them to generate a texture feature vector; Multi-scale global context features are captured from the normalized pixel matrix using spatial pyramid pooling operations, and a context feature vector is output. The texture feature vector and the context feature vector are fused, and an enhanced feature tensor is generated by highlighting useful information and suppressing noise interference through nonlinear transformation.

8. The lung ventilation-perfusion imaging region image segmentation and quantification analysis system as described in claim 1, characterized in that: The process of using the K-means clustering algorithm to partition the enhanced feature tensor into sub-regions includes the following steps. Extract the spatial coordinates, brightness values, and texture feature vectors from the enhanced feature tensor to construct a multidimensional feature vector set; The number of clusters is set according to the symmetry of the left and right lung anatomical structures, and the cluster center points are initialized to be evenly distributed in space. Calculate the Euclidean distance from each feature vector in the multidimensional feature vector set to each cluster center, assign cluster labels based on the nearest neighbor principle, and generate an initial sub-region partitioning map; Iteratively update the cluster center position based on the mean of the feature vectors within the cluster according to the initial sub-region partitioning map, and optimize the cluster boundary distribution; The calculation terminates when the cluster boundary distribution of consecutive iterations falls below the boundary threshold, and the cluster label map of the left and right lung function sub-regions is output.

9. The lung ventilation-perfusion imaging region image segmentation and quantification analysis system as described in claim 1, characterized in that: The calculation of the V / Q ratio for each region and its mapping to the CT image includes the following steps. The average ventilation and irrigation intensity of each sub-region is calculated based on the cluster label map, and the V / Q ratio is output. The V / Q ratio is mapped to the CT image to generate a discrete color map. The discrete color map is converted into a continuous heatmap using an interpolation algorithm.

10. The lung ventilation-perfusion imaging region image segmentation and quantification analysis system as described in claim 1, characterized in that: The CT images refer to computed tomography images that provide a baseline of lung anatomy.

Images