A lung ventilation-perfusion imaging region three-dimensional blood flow distribution abnormality analysis method

Abstract

The application discloses a lung ventilation-perfusion imaging area three-dimensional blood flow distribution abnormality analysis method, relates to the technical field of lung function quantitative evaluation, and comprises the following steps: acquiring lung CT image data of a patient to be analyzed, performing structure segmentation on a plurality of sections of the lung CT image data to obtain ventilation phase gray scale images and blood flow perfusion phase gray scale images; on the basis of the lung CT image data, accurately extracting a non-black lung region by threshold segmentation, respectively calculating the ratio of ventilation CT values and blood flow CT values in the region, and generating lung region segmentation data; based on the three-dimensional V / Q ratio atlas, identifying abnormal sub-regions in which local ventilation blood flow ratios deviate from physiological ranges; and obtaining a blood flow distribution abnormality analysis report by performing structured analysis on the abnormal sub-regions. The application divides irregular sub-regions by performing a K-means clustering algorithm on gray scale values in each section, realizes adaptive segmentation of lung function heterogeneity structures, and improves the rationality of local ratio calculation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of quantitative evaluation of lung function, and particularly relates to a three-dimensional blood flow distribution abnormality analysis method for lung ventilation-perfusion imaging regions. BACKGROUND

[0002] The three-dimensional blood flow distribution abnormality analysis method for lung ventilation-perfusion imaging regions belongs to the technical field of medical image quantitative analysis, and current conventional methods are mostly based on two-dimensional cross-sectional images to calculate ventilation and perfusion function ratios, rely on manual delineation of regions or global mean estimation, can realize basic lung function evaluation, and are suitable for clinical preliminary screening and qualitative judgment.

[0003] However, the conventional methods lack three-dimensional continuity integration capability in the spatial dimension, and it is difficult to realize accurate spatial positioning of abnormal regions; meanwhile, the local functional heterogeneity is not structurally clustered and analyzed in the quantitative level, resulting in weak correlation between volume evaluation and function distribution, and affecting accurate identification and clinical interpretation of abnormal regions. SUMMARY

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

[0005] Therefore, the present application provides a three-dimensional blood flow distribution abnormality analysis method for lung ventilation-perfusion imaging regions to solve the problem of inaccurate positioning and quantification of lung blood flow abnormal regions.

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

[0007] The present application provides a three-dimensional blood flow distribution abnormality analysis method for lung ventilation-perfusion imaging regions, which comprises,

[0008] Obtaining lung CT image data of a patient to be analyzed, performing structure segmentation on a plurality of cross sections of the lung CT image data to obtain ventilation phase gray scale images and blood flow perfusion phase gray scale images;

[0009] Based on the lung CT image data, accurately extracting a non-black lung region by threshold segmentation to calculate the ratio of ventilation CT value to blood flow CT value in the region, and generating lung region segmentation data;

[0010] Inputting the lung region segmentation result into a U-Net model for lung segmentation processing to obtain the gray scale value in the ventilation phase image and the gray scale value in the blood flow perfusion phase image;

[0011] Performing K-means clustering algorithm on the gray scale value in each cross section, and dividing the left and right lungs into a plurality of irregular sub-regions according to spatial proximity and functional gray scale similarity;

[0012] Calculate the average ventilation gray value and the average blood perfusion gray value of all pixels inside each irregular sub-region, and determine the local ventilation-blood flow ratio of the sub-region;

[0013] Reposition and integrate the local ventilation-blood flow ratio of all irregular sub-regions in each section in three-dimensional space through spatial coordinates to generate a three-dimensional V / Q ratio map;

[0014] Based on the three-dimensional V / Q ratio map, identify abnormal sub-regions with local ventilation-blood flow ratio deviating from the physiological range, and obtain a blood flow distribution abnormality analysis report through structured analysis of the abnormal sub-regions.

[0015] As a preferred scheme of the lung ventilation-perfusion imaging region three-dimensional blood flow distribution abnormality analysis method,

[0016] The lung ventilation-perfusion imaging region three-dimensional blood flow distribution abnormality analysis method is characterized in that: the lung CT image data of the patient to be analyzed is obtained, and the specific steps are as follows,

[0017] Based on the DICOM protocol interface, receive the ventilation phase SPECT data and perfusion phase SPECT data synchronously collected by the patient, and analyze through normalization processing to obtain the lung CT image;

[0018] Perform mask processing by segmenting and quantifying the lung CT image, and perform pixel-level cropping of the ventilation phase and perfusion phase images through the mask to generate lung CT image data of the ventilation phase gray image and blood flow perfusion phase gray image that retains the lung tissue region.

[0019] The lung ventilation-perfusion imaging region three-dimensional blood flow distribution abnormality analysis method is characterized in that: the lung CT image data is segmented into multiple sections to obtain the ventilation phase gray image and the blood flow perfusion phase gray image, and the specific steps are as follows,

[0020] Based on the synchronously collected SPECT ventilation phase and perfusion phase combined lung CT image data, the spatial dimension is processed through normalization to obtain an accurately matched three-dimensional joint data set;

[0021] According to the registered three-dimensional joint data set, the ventilation phase and perfusion phase images are pixel-level cropped using the lung parenchyma segmentation mask to generate the ventilation phase gray image and the blood flow perfusion phase gray image.

[0022] The lung ventilation-perfusion imaging region three-dimensional blood flow distribution abnormality analysis method is characterized in that: based on the lung CT image data, the non-black lung region is accurately extracted through threshold segmentation, and the ratio of the ventilation CT value and the blood flow CT value in the region is calculated to generate lung region segmentation data, and the specific steps are as follows,

[0023] According to the lung region segmentation data, threshold segmentation is adopted to preliminarily segment the CT image data, so as to obtain a mask containing only lung tissue, and to obtain ventilation phase and perfusion phase SPECT gray scale images;

[0024] According to the mask, pixel mask extraction is performed on the ventilation phase and perfusion phase SPECT gray scale images collected synchronously, and through the lung pixel data of the lung parenchyma region, lung region segmentation data is generated.

[0025] The lung ventilation-perfusion imaging region three-dimensional blood flow distribution abnormality analysis method is characterized in that the lung region segmentation result is input into a U-Net model for lung segmentation processing, so as to obtain the gray scale values in the ventilation phase image and the gray scale values in the blood flow perfusion phase image, and the specific steps are as follows,

[0026] Based on the registered CT image data, upsampling is performed through the decoder of the U-Nt model, and combined with the multi-scale feature fusion of the CT image data, pixel-by-pixel extraction is performed on the ventilation phase and blood flow perfusion phase gray scale images, so as to obtain the lung ventilation function gray scale distribution map and the lung blood flow perfusion function gray scale distribution map;

[0027] According to the lung ventilation function gray scale distribution map and the lung blood flow perfusion function gray scale distribution map, high-precision contour recognition of the lung lobe boundary is performed through the network architecture of the U-Net model, so as to generate a lung lobe segmentation mask;

[0028] Based on the lung lobe segmentation mask, a mask superposition method is used to act on the ventilation phase and blood flow perfusion phase gray scale images respectively, so as to obtain the gray scale values in the ventilation phase image and the gray scale values in the blood flow perfusion phase image.

[0029] The lung ventilation-perfusion imaging region three-dimensional blood flow distribution abnormality analysis method is characterized in that the K-means clustering algorithm is performed on the gray scale values in each section, and according to the spatial proximity and functional gray scale similarity, the left and right lungs are respectively divided into a plurality of irregular sub-regions, and the specific steps are as follows,

[0030] Based on the gray scale images of the ventilation phase and the blood flow perfusion phase, combined with the corresponding spatial coordinate information, a multi-dimensional feature vector of each pixel is constructed;

[0031] By calculating the Euclidean distance between the multi-dimensional feature vectors of each pixel, an iterative optimization method is used to cluster and divide the left and right lungs respectively, so as to obtain an initial lung sub-region distribution map;

[0032] According to the lung sub-region distribution map, the position of each cluster center is dynamically updated to be the mean value of all pixel feature vectors in the corresponding region, and the left and right lungs are each divided into a plurality of irregular sub-regions through repeated iteration.

[0033] The lung ventilation-perfusion imaging region three-dimensional blood flow distribution abnormality analysis method is characterized in that: the average ventilation gray value and the average blood flow perfusion gray value of all pixels in each irregular sub-region are calculated, and the local ventilation blood flow ratio of the sub-region is determined, and the specific steps are as follows,

[0034] Based on the lung region segmentation result, the lung pixels in the ventilation phase gray scale image and the blood flow perfusion phase gray scale image are registered to obtain the ventilation and perfusion gray scale values corresponding in space position;

[0035] According to the ventilation and perfusion gray scale values, the gray scale values of the pixels in each sub-region are fused and pixel value calculation is performed on the ventilation CT image by using weighted fusion to obtain the local ventilation blood flow ratio of the sub-region.

[0036] The lung ventilation-perfusion imaging region three-dimensional blood flow distribution abnormality analysis method is characterized in that: the local ventilation blood flow ratio of all irregular sub-regions in each section is repositioned and integrated in three-dimensional space through spatial coordinates to generate a three-dimensional V / Q ratio map, and the specific steps are as follows,

[0037] Based on the lung region segmentation results of the three sections of the coronal, sagittal and axial sections after registration, the center coordinates of each sub-region in each section and the corresponding local ventilation blood flow ratio are spatially position coded to obtain accurate position information of each sub-region in the pixel space;

[0038] According to the three-dimensional position information and the ratio data, K-means clustering analysis is used to perform spatial clustering analysis on the sub-regions between adjacent sections to obtain the ventilation and perfusion conditions of the discrete regions covering the complete lung volume;

[0039] The ventilation and perfusion conditions of the discrete regions are distributed to the ventilation blood flow ratio of each pixel in the V / Q ratio of the CT image by a color mapping method to generate a three-dimensional V / Q ratio map.

[0040] The lung ventilation-perfusion imaging region three-dimensional blood flow distribution abnormality analysis method is characterized in that: based on the three-dimensional V / Q ratio map, the abnormal sub-regions with the local ventilation blood flow ratio deviating from the physiological range are identified, and the specific steps are as follows,

[0041] Based on the three-dimensional V / Q ratio map, all pixels are preliminarily screened by screening to obtain a preliminary abnormal pixel sub-region;

[0042] According to the preliminary abnormal pixel sub-region, the spatially adjacent abnormal pixels are aggregated by using a pixel decoder to generate an independent three-dimensional abnormal pixel sub-region;

[0043] The automatic analysis of medical images by the U-Net model of deep learning performs quantitative analysis on each three-dimensional abnormal pixel sub-region, and identifies the abnormal sub-region deviating from the physiological range of local ventilation-blood flow ratio.

[0044] The lung ventilation-perfusion imaging region three-dimensional blood flow distribution abnormality analysis method is characterized in that: through structured analysis of the abnormal sub-region, an abnormal blood flow distribution analysis report is obtained, and the specific steps are as follows,

[0045] Based on the three-dimensional V / Q ratio map, the pixel regions continuously deviating from the physiological range are morphologically connected to obtain a three-dimensional abnormal sub-region set;

[0046] The three-dimensional abnormal sub-region set is subjected to threshold ventilation-blood flow ratio spatial positioning and volume quantification to obtain an abnormal blood flow distribution analysis report.

[0047] The present application has the beneficial effects that: by performing K-means clustering algorithm on the gray value in each section to divide irregular sub-regions, the adaptive segmentation of the functional heterogeneity structure in the lung is realized, and the rationality of local ratio calculation is improved; and then through three-dimensional spatial repositioning and integration, a three-dimensional V / Q ratio map is generated, so that the spatial distribution of the abnormal region is visualized and volume quantification is supported, and finally the dual beneficial effects of accurate positioning and clinically interpretable analysis are achieved. BRIEF DESCRIPTION OF DRAWINGS

[0048] 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.

[0049] Fig. 1 The flowchart of the lung ventilation-perfusion imaging region three-dimensional blood flow distribution abnormality analysis method.

[0050] Fig. 2 The flowchart of the lung CT image data processing and segmentation.

[0051] Fig. 3 The flowchart of the three-dimensional V / Q ratio map generation.

[0052] Fig. 4 The flowchart of the blood flow distribution abnormality analysis report generation. DETAILED DESCRIPTION

[0053] 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.

[0054] 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.

[0055] Secondly, the "one embodiment" or "an embodiment" referred to herein means a specific feature, structure, characteristic, or combination of features and / or characteristics described herein that can be included in at least one implementation of the present application. The various appearances of "in one embodiment" or "an embodiment" in the specification do not all refer to the same embodiments, although they can.

[0056] Referring to Figs. 1-4 For one embodiment of the present application, the embodiment provides a lung ventilation-perfusion imaging region three-dimensional blood flow distribution anomaly analysis method, comprising the following steps:

[0057] S1, acquiring lung CT image data of a patient to be analyzed, performing structure segmentation on the lung CT image data in multiple sections to obtain ventilation phase gray scale images and blood flow perfusion phase gray scale images.

[0058] Based on the DICOM protocol interface, ventilation phase SPECT data and perfusion phase SPECT data synchronously collected by the patient are received, and normalization processing is performed for analysis to obtain lung CT images;

[0059] Further, ventilation phase SPECT data and perfusion phase SPECT data are obtained from a medical imaging device using a DICOM protocol interface, ensuring that the two sets of data are collected synchronously in time and aligned in space coordinates, and then normalization processing is performed on the ventilation phase SPECT data and the perfusion phase SPECT data respectively, so that the respective data ranges are unified to the same scale. Linear scaling is used for normalization processing to map the minimum value to zero and the maximum value to one. The normalized ventilation phase SPECT data and perfusion phase SPECT data are input into an image fusion algorithm. Pixel weighted summation is used for image fusion, and the weighting coefficient is calculated from the pixel value. The fusion result generates a lung CT image.

[0060] It should be noted that the pixel weighted summation in the present application performs weighted accumulation on the pixel gray value or ventilation blood flow ratio value according to the inverse of the spatial distance as the weight, ensures that the spatial proximity dominates the calculation result, and makes the local ventilation blood flow ratio, pixel ventilation blood flow ratio mapping value and abnormal region representative value all reflect the weighted average characteristics of the spatial distribution, finally supports the generation of three-dimensional V / Q ratio map and the output of blood flow distribution abnormality analysis report; in the present application, the pixel value calculation realizes the proportional transparent superposition of CT image and mask, makes the lung tissue profile visible in a translucent form on the CT image, performs value range clipping on each pixel of the fused image, limits the pixel value in the range of 0 to 255, ensures that the output image conforms to the standard display format, and finally generates the fused image as the lung CT image.

[0061] By performing mask processing on the segmented and quantized lung CT image, the ventilation phase gray scale image and the blood perfusion phase gray scale image data of the lung CT image are generated by performing pixel-level clipping on the ventilation phase and perfusion phase images through the mask;

[0062] Further, by performing mask processing on the segmented and quantized lung CT image, and identifying the lung tissue region in the lung CT image through image segmentation, a mask image corresponding to the lung tissue region is generated, the pixel value of the lung tissue region in the mask image is a valid value, and the pixel value of the non-lung tissue region is an invalid value; then the mask image is applied to the ventilation phase image and the perfusion phase image respectively, and a logical AND operation is performed pixel by pixel, so that only the pixels corresponding to the valid value positions in the mask image are retained in the ventilation phase image and the perfusion phase image, and the rest of the positions are set to invalid values; finally, the ventilation phase gray scale image and the blood perfusion phase gray scale image both only contain the lung tissue region, and the non-lung tissue region is completely excluded, realizing accurate retention of the lung tissue region.

[0063] It should be noted that the mask processing in the present application performs image segmentation on the lung CT image, and the image segmentation identifies the pixel set belonging to the lung tissue region according to the distribution characteristics of the pixel gray value in the lung CT image (such as normal aerated lung parenchyma, lung blood vessels and bronchial wall, extrathoracic tissue, etc.); the pixels meeting the density characteristics of the lung tissue are screened out by setting the gray value range to form a preliminary lung tissue region profile; the mask processing result is used for subsequent lung segmentation, gray value extraction and sub-region division, ensuring that all calculations and analyses are limited within the anatomical range of the lung tissue, and improving the spatial accuracy and physiological relevance of the ventilation blood flow ratio calculation.

[0064] It should be noted that the setting of the gray value range is based on the physical density difference of different tissues in the lung CT image, and the typical gray scale performance of lung tissue, air, soft tissue and background in the CT image is observed, combined with clinical anatomy and imaging standards, to determine the effective gray interval for segmentation and calculation;

[0065] Based on the SPECT ventilation phase and perfusion phase combined lung CT image data, the accurate matching three-dimensional joint data set is obtained by normalizing the spatial dimension;

[0066] Further, the spatial resampling is performed on the SPECT ventilation phase and perfusion phase combined lung CT image data respectively, so that the physical spacing of each pixel in the three spatial axes is consistent, the pixel value is redistributed by normalizing processing, then the SPECT ventilation phase image and the perfusion phase combined lung CT image are aligned in the same spatial coordinate system, the position and angle deviation is adjusted, the mutual information is used as the similarity measurement criterion in the registration process, the transformation parameters are iteratively optimized until convergence, and finally the spatial dimension normalized accurate matching three-dimensional joint data set is output.

[0067] According to the registered three-dimensional joint data set, the ventilation phase and perfusion phase images are pixel-level cropped by using the lung parenchyma segmentation mask to generate the ventilation phase gray scale image and the blood perfusion phase gray scale image;

[0068] Further, the lung parenchyma segmentation mask and the ventilation phase image are pixel by pixel aligned in the spatial position, the ventilation phase image pixels corresponding to the non-zero region in the lung parenchyma segmentation mask are retained, and the rest of the region values are zero to form the ventilation phase gray scale image; the lung parenchyma segmentation mask and the perfusion phase image are pixel by pixel aligned in the spatial position, the perfusion phase image pixels corresponding to the non-zero region in the lung parenchyma segmentation mask are retained, and the rest of the region values are zero to form the blood perfusion phase gray scale image.

[0069] S2, based on the lung CT image data, the non-black lung region is accurately extracted by threshold segmentation, and the ratio of the ventilation CT value to the blood flow CT value in the region is calculated to generate the lung region segmentation data.

[0070] According to the lung region segmentation data, the CT image data is preliminarily segmented by threshold segmentation to obtain a mask containing only lung tissue, and the ventilation phase and perfusion phase SPECT gray scale images are obtained;

[0071] Further, the gray scale value of each pixel in the CT image data is read, and the pixels with the gray scale value falling within the gray scale value interval are retained by the corresponding gray scale value interval of the lung tissue in the CT image, and the pixels with the gray scale value falling outside the gray scale value interval are set to zero to generate a mask containing only lung tissue. The mask is applied to the ventilation phase SPECT gray scale image and the perfusion phase SPECT gray scale image respectively, and the same spatial position mask is performed on the ventilation phase SPECT gray scale image and the perfusion phase SPECT gray scale image, only the gray scale value of the corresponding position of the lung tissue region identified by the mask is retained, and the rest of the position gray scale value is set to zero to obtain the ventilation phase SPECT gray scale image and the perfusion phase SPECT gray scale image limited by the lung tissue.

[0072] According to the mask, pixel mask extraction is performed on the synchronously collected ventilation phase and perfusion phase SPECT gray scale images, and through the lung pixel data of the lung parenchyma region, lung region segmentation data is generated.

[0073] Further, when performing pixel mask extraction on the synchronously collected ventilation phase and perfusion phase SPECT gray scale images, according to the region indicated by the mask, it is judged whether the corresponding positions in the ventilation phase SPECT gray scale image and the perfusion phase SPECT gray scale image belong to the lung parenchyma region pixel by pixel (for example, at a certain pixel coordinate (x, y, z), if the CT mask value is 1, it indicates that the position is located in the anatomical lung tissue, then the radioactivity counts (such as ventilation count 850 and perfusion count 1200) of the point are extracted from the ventilation phase SPECT image and the perfusion phase SPECT image respectively, which are used to calculate the local ventilation blood flow ratio; if the mask value is 0 (such as located in the rib or mediastinum region), even if the SPECT image has a signal at this position, it is not belong to the lung parenchyma and is excluded, ensuring that all functional analysis is strictly limited within the true lung tissue range, improving the physiological accuracy and clinical reliability of the V / Q ratio.); According to the pixel position marked as the lung parenchyma region in the mask, the gray values of the pixel positions in the ventilation phase SPECT gray scale image and the perfusion phase SPECT gray scale image are retained, otherwise the pixel position gray value is set to invalid; according to the ventilation phase SPECT gray scale image and the perfusion phase SPECT gray scale image, image data containing only lung parenchyma region pixels is extracted; the extracted lung parenchyma region pixel data is taken as the lung region segmentation data.

[0074] S3, input the lung region segmentation result into the U-Net model for lung segmentation processing, and obtain the gray value in the ventilation phase image and the gray value in the blood flow perfusion phase image.

[0075] Based on the registered CT image data, the decoder of the U-Net model is used for upsampling, and the ventilation phase and blood flow perfusion phase gray scale images are extracted pixel by pixel by combining the multi-scale feature fusion of the CT image data, to obtain the lung ventilation function gray distribution map and the lung blood flow perfusion function gray distribution map;

[0076] Further, starting from the low-resolution feature map output by the encoder, transposed convolution is performed layer by layer to gradually restore the spatial resolution, and at each upsampling stage, the multi-scale features of the CT image data extracted from the corresponding level of the decoder of the U-Net model are spliced to the current decoder feature map to realize the fusion of spatial details and semantic information. Through pixel-by-pixel extraction of the ventilation phase and the blood perfusion phase grayscale images at the final output layer of the decoder, the extraction process directly maps the grayscale value corresponding to each spatial position output by the decoder to form a lung ventilation function grayscale distribution map and a lung blood perfusion function grayscale distribution map. The spatial dimensions of the two distribution maps are consistent with the registered CT image data, and the lung ventilation function grayscale distribution map and the lung blood perfusion function grayscale distribution map are obtained.

[0077] It should be noted that the training of the U-Net model in the present application generally includes the following steps: first, prepare a paired input-label data set, use the lung CT image as the input, and the manually or semi-automatically labeled lung parenchyma mask as the label; and pre-process the data (such as normalization, cropping, data enhancement, etc.) to improve the generalization ability; construct the U-Net network structure, the encoder path extracts multi-scale features through convolution and downsampling, and the decoder path gradually restores spatial details through upsampling and skip connection; during training, a pixel-level loss function is used to measure the difference between the predicted mask and the true mask, and the network parameters are iteratively updated through backpropagation and optimizer to obtain an accurate lung tissue segmentation U-Net model; the entire processing process maintains the spatial alignment of the input and output, ensuring that the ventilation function gray distribution map and the blood perfusion function gray distribution map strictly correspond in pixel-level spatial position, providing anatomical structure constraints and spatial consistency guarantees for subsequent sub-region division and ventilation-blood flow ratio calculation; the gray value corresponding to each spatial position output by the decoder is obtained by inputting the registered CT image data into the encoder path of the U-Net model, and the CT image data is a multi-channel image, each channel corresponding to gray information of different sections or functional phases; the encoder extracts spatial features of the image through consecutive convolution layers and downsampling layers, and the size of the feature map output at each level is halved and the number of channels is doubled, forming a multi-scale feature pyramid from shallow to deep; at the output of each encoder, the spatial feature map retains the corresponding spatial resolution and local anatomical structure details; the decoder path starts from the deepest feature map and restores the spatial resolution through transposed convolution or upsampling at each level; after each upsampling, the decoder concatenates the feature map of the current layer with the feature map of the corresponding layer of the encoder, which has the same spatial size, in the channel dimension, to realize skip connection, fuse deep semantic information and shallow spatial details, and the output layer of the decoder compresses the number of channels of the spatial feature map to the same number as the input image through a 1x1 convolution kernel, to obtain a spatial feature map with the same spatial size as the original input; the pixel value of each spatial position in the spatial feature map, the gray value output by the decoder at the spatial position, and the gray value distribution map correspond to the ventilation function gray distribution map and the blood perfusion function gray distribution map in the lung, respectively.

[0078] According to the ventilation function gray distribution map and the blood perfusion function gray distribution map in the lung, the network architecture of the U-Net model is used to perform high-precision contour recognition on the lung lobe boundary to generate a lung lobe segmentation mask.

[0079] Further, by taking two functional gray scale distribution maps as input, sending them into the network architecture of the U-Net model, the encoder part of the U-Net model extracts the spatial features of the lung lobe boundary layer by layer and reduces the resolution, and the decoder part gradually recovers the spatial dimension through upsampling and fuses the feature maps of the corresponding encoder layer, generates a lung lobe segmentation mask consistent with the size of the input image, the lung lobe segmentation mask covers the complete lung lobe range, and the value of each pixel position thereof reflects the probability that the pixel position belongs to the lung lobe region, and the high-precision contour recognition of the lung lobe boundary is completed through pixel-by-pixel classification, and a lung lobe segmentation mask covering the complete lung lobe range is generated.

[0080] It should be noted that in the present application, the decoder is composed of multiple decoding levels, each decoding level including one upsampling and one convolution; the upsampling adopts bilinear interpolation or transposed convolution method to expand the spatial size of the current layer feature map by one time, recovering the spatial resolution lost due to the downsampling of the encoder; after each upsampling, the decoder extracts the feature map of the same spatial size from the corresponding level of the encoder, fuses the feature maps after convolution and upsampling through channel dimension splicing, and realizes multi-scale feature fusion; the spliced feature map is then subjected to two consecutive 3x3 convolution layers, followed by a nonlinear activation function, for integrating context semantic information and local spatial details, while eliminating the checkerboard effect or fuzzy boundary of upsampling; the number of channels of the feature map output by each decoding level is halved step by step, and the spatial size is multiplied step by step, gradually approaching the spatial resolution of the original lung CT image data; the output layer of the decoder compresses the number of channels to be consistent with the number of segmentation categories through 1x1 convolution, generating a probability value of each pixel belonging to the lung lobe structure; the probability map and the original lung CT image data are spatially aligned at the voxel level.

[0081] Based on the lung lobe segmentation mask, the mask superposition method is used to act on the ventilation phase and blood perfusion phase gray scale images respectively to obtain the gray scale values in the ventilation phase image and the gray scale values in the blood perfusion phase image.

[0082] Further, the spatial positions corresponding to the non-zero regions in the lung lobe segmentation mask are used to extract the gray scale values of the corresponding positions in the ventilation phase gray scale image to form a set of gray scale values in the ventilation phase image; the spatial positions corresponding to the non-zero regions in the lung lobe segmentation mask are used to extract the gray scale values of the corresponding positions in the blood perfusion phase gray scale image to form a set of gray scale values in the blood perfusion phase image; the gray scale values are obtained by matching each pixel position to obtain the gray scale values in the ventilation phase image and the gray scale values in the blood perfusion phase image.

[0083] S4, performing K-means clustering algorithm on the gray scale values in each section, and dividing the left and right lungs into multiple irregular sub-regions according to spatial proximity and functional gray scale similarity.

[0084] Based on the gray images of the ventilation phase and the blood perfusion phase, combined with the corresponding spatial coordinate information, a multi-dimensional feature vector of each pixel is constructed;

[0085] Further, the gray value of each pixel point in the ventilation phase gray image is extracted, the gray value of the pixel point at the corresponding position in the blood perfusion phase gray image is extracted, and the X coordinate value, Y coordinate value and Z coordinate value of the pixel point in the three-dimensional space are extracted from the spatial coordinate information. The five values of the ventilation phase gray image are arranged in a fixed order to form a five-dimensional vector, and the five-dimensional vector is the multi-dimensional feature vector of each pixel.

[0086] By calculating the Euclidean distance between the multi-dimensional feature vectors of each pixel, an iterative optimization method is used to cluster and divide the left and right lungs respectively to obtain an initial lung sub-region distribution map.

[0087] Further, the corresponding multi-dimensional feature vector of each pixel in the left and right lung regions is extracted, the Euclidean distance between the multi-dimensional feature vectors of any two pixels is calculated, the Euclidean distance is used as a similarity measure, and an iterative optimization method is used to cluster and divide all pixels. In each iteration process, the class to which the pixel belongs is adjusted to minimize the intra-class distance and maximize the inter-class distance. The cluster center is updated by iteration and the pixel is re-assigned until the displacement of all cluster centers in the multi-dimensional feature space is less than the minimum intra-class distance. The initial lung sub-region distribution map corresponding to the left and right lungs is generated.

[0088] It should be noted that the iterative optimization method gradually enhances the spatial proximity and functional gray similarity of each pixel in the present application. When the displacement of all pixel centers in the multi-dimensional feature space in the two consecutive iterations is less than the fixed convergence standard, the iteration is stopped. The final output of the pixel clustering attribution result constitutes the initial lung sub-region distribution map. Each sub-region corresponds to a cluster, and the pixels in the sub-region have high consistency in spatial position and ventilation / perfusion gray.

[0089] According to the lung sub-region distribution map, the position of each cluster center is dynamically updated to be the mean value of the feature vectors of all pixels in the corresponding region, and the left and right lungs are each generated by a plurality of irregular sub-regions through repeated iteration.

[0090] Further, by extracting the pixel set corresponding to each cluster in the lung sub-region distribution map, the feature vectors of all pixels in each cluster are arithmetically averaged, and the average is taken as the new center position of the cluster; based on the updated cluster center positions, each pixel is reassigned to the region to which the nearest cluster center belongs; the pixel reassignment and cluster center position updating steps are repeated until the cluster center positions no longer change in consecutive iterations; the division results of the left and right lungs each composed of multiple irregular sub-regions are obtained, each irregular sub-region corresponds to a stable convergent cluster center, and the average of the feature vectors of all pixels inside each irregular sub-region is equal to its cluster center position, and the left and right lungs each composed of multiple irregular sub-regions are generated.

[0091] S5, calculate the average ventilation gray value and the average blood flow perfusion gray value of all pixels inside each irregular sub-region, and determine the local ventilation blood flow ratio of the sub-region.

[0092] Based on the lung region segmentation result, the lung pixels in the ventilation phase gray scale image and the blood flow perfusion phase gray scale image are registered to obtain ventilation and perfusion gray values that strictly correspond in space;

[0093] Further, the pixel sets belonging to the lung region in the ventilation phase gray scale image and the blood flow perfusion phase gray scale image are extracted using the lung region segmentation result, a registration method based on mutual information is used to align the spatial positions of the lung region pixels in the two images, the mutual information of the joint distribution of the lung region gray values in the two images is maximized to optimize the spatial transformation parameters, and the blood flow perfusion phase gray scale image is resampled according to the optimized spatial transformation parameters, so that each lung pixel in the resampled blood flow perfusion phase gray scale image completely coincides with the corresponding lung pixel in the ventilation phase gray scale image in spatial coordinates, and the ventilation and perfusion gray values that strictly correspond in space are obtained.

[0094] According to the ventilation and perfusion gray values, the gray values of the pixels in each sub-region are fused and pixel value calculations are performed on the ventilation CT image using weighted fusion to obtain the local ventilation blood flow ratio of the sub-region;

[0095] Further, by extracting the registered gray value and the gray value of the corresponding position pixel in the ventilation CT image for each pixel in each sub-region, the two sets of gray values are fused by a predetermined weight coefficient, and the weight coefficient is determined according to the anatomical position and physiological characteristics of the sub-region, and the weight coefficient can be more specifically and clearly expanded through clinical experience data statistical analysis by collecting a large amount of clinical image and functional examination data (such as SPECT / CT, lung function test, arterial blood gas analysis, etc.) through the system, and the ventilation capacity, blood perfusion level, ventilation-blood flow ratio (V / Q) and variation law of different lung sub-regions are quantitatively analyzed; the pixel gray value after weighted fusion is used to represent the local ventilation-blood flow state of the pixel position, and the weighted fusion gray value of all pixels in the sub-region is spatially averaged, and the obtained value is the local ventilation-blood flow ratio of the sub-region.

[0096] It should be noted that in the present application, in each irregular sub-region, the spatial Euclidean distance reciprocal of the pixel and the geometric center of the sub-region is used as the weight, and the ventilation phase gray value and the blood perfusion phase gray value are respectively pixel weighted and summed to obtain the weighted ventilation gray sum and the weighted blood perfusion gray sum, and the difference between the two is the local ventilation-blood flow ratio of the sub-region, which reflects the matching relationship between the spatial weighted average ventilation and perfusion function in the sub-region through the ratio, and ensures that the central region pixels contribute more to the result, and the edge pixels contribute decreasingly, which improves the spatial representativeness and physiological rationality of the local ventilation-blood flow ratio.

[0097] S6, the local ventilation-blood flow ratio of all irregular sub-regions in each section is repositioned and integrated in three-dimensional space through spatial coordinates to generate a three-dimensional V / Q ratio map.

[0098] Based on the segmentation results of the lung region in the registered coronal, sagittal and axial sections, the center coordinates of each sub-region and the corresponding local ventilation-blood flow ratio in each section are spatially encoded to obtain the accurate position information of each sub-region in the pixel space;

[0099] Further, based on the segmentation results of the lung region in the registered coronal, sagittal and axial sections, the center coordinates of each sub-region are extracted, the center coordinate values are obtained by calculating the geometric center of the pixel points contained in each sub-region in the respective section coordinate system, the center coordinates of the coronal, sagittal and axial sections are mapped to a unified pixel space coordinate system according to the spatial registration relationship, the three-dimensional spatial alignment of the center coordinates of the three sections is realized, and the corresponding local ventilation-blood flow ratio of each sub-region and the aligned center coordinates in the pixel space are one-to-one bound to form a joint coding data structure containing spatial position and physiological parameters, which is recorded in a color coding data structure, and the accurate position information of each sub-region in the pixel space is obtained.

[0100] It should be noted that, in the present application, according to the ventilation-perfusion ratio value corresponding to each pixel in the three-dimensional V / Q ratio map, the color coding is mapped to the color value, and different colors are assigned in turn from low to high according to the ventilation-perfusion ratio through the color value. The low ratio area is mapped to the blue system, the medium ratio area is mapped to the green or yellow system, and the high ratio area is mapped to the red system. Each pixel looks up the corresponding color value in the scale according to the ventilation-perfusion ratio value and assigns the pixel, so that the different ventilation and perfusion matching states in the three-dimensional V / Q ratio map are directly presented through color difference, which is convenient for visually identifying the ventilation and perfusion balanced area, the ventilation dominant area and the perfusion dominant area. The color coding process keeps one-to-one correspondence between the value and the color, ensuring the spatial continuity and color transition smoothness of the ventilation-perfusion ratio distribution in the three-dimensional space.

[0101] According to the three-dimensional position information and the ratio data, K-means clustering analysis is adopted to perform spatial clustering analysis on the sub-regions between adjacent sections, and the ventilation and perfusion conditions of the discrete regions covering the complete lung volume are obtained; through the color mapping method, the ventilation and perfusion conditions of the discrete regions are distributed to the ventilation-perfusion ratio of each pixel in the CT image, and a three-dimensional V / Q ratio map is generated;

[0102] Further, by obtaining three-dimensional position information and ratio data, the three-dimensional position information is composed of the spatial coordinates of the CT image pixels, and the ratio data is the ventilation-perfusion ratio value corresponding to each pixel; the three-dimensional position information and the ratio data are combined to form a feature vector, each feature vector contains spatial coordinates and ventilation-perfusion ratio value; K-means clustering analysis is performed on the sub-regions between adjacent sections, and the feature vector is taken as the input in the clustering process, and the clustering clusters are divided according to the spatial proximity and the ventilation-perfusion ratio. The number of clustering clusters is determined by the natural partition of the lung anatomy structure, and the clustering center is updated by iterative calculation until convergence; the clustering result forms discrete regions covering the complete lung volume, and each discrete region has a unified ventilation and perfusion condition label; a color mapping method is used to map the ventilation and perfusion condition label of each discrete region to a specific color value, and the color value is linearly distributed according to the ventilation-perfusion ratio value range; according to the color value, each pixel in the V / Q ratio of the CT image is assigned, and the color value of the pixel is determined by the ventilation and perfusion condition label of the discrete region to which it belongs, and finally a three-dimensional V / Q ratio map is generated.

[0103] It should be noted that in the present application, K-means clustering analysis is performed on the sub-regions between adjacent sections based on the multi-dimensional feature vector composed of three-dimensional position information and ratio data as input, the number of clusters is set, the cluster center points are randomly initialized, the Euclidean distance between the feature vector of each pixel and each cluster center point is calculated, and the pixel is assigned to the nearest cluster. After one round of assignment, the cluster center point is recalculated as the mean value of all pixel feature vectors in the cluster, and the assignment and center updating steps are repeated until the cluster center changes less than the convergence standard. The discrete regions covering the complete lung volume are obtained, and the pixels inside each discrete region have high consistency in spatial position and ventilation-perfusion ratio value, which are used to represent the unified ventilation and perfusion conditions and provide a structured partition basis for color mapping and three-dimensional V / Q ratio map generation.

[0104] S7, based on the three-dimensional V / Q ratio map, identifying abnormal sub-regions with local ventilation-perfusion ratio deviating from the physiological range, and obtaining a blood flow distribution abnormality analysis report through structured analysis of the abnormal sub-regions.

[0105] Based on the three-dimensional V / Q ratio map, all pixels are screened to obtain a preliminary abnormal pixel sub-region.

[0106] Further, in the three-dimensional V / Q ratio map, according to the continuity characteristics of the pixels in spatial distribution and value distribution, the pixels with ventilation-perfusion ratio V / Q deviating from the normal range 1.0±0.5 or SPECT signal intensity significantly lower than the adjacent region are screened, and the spatial region with abnormal performance is manually selected on the visualization interface according to the pixel set. The selected region is the preliminary abnormal pixel sub-region.

[0107] It should be noted that in the present application, the physiological reference interval of ventilation-perfusion ratio is set on the three-dimensional V / Q ratio map visualization interface, and the ventilation-perfusion ratio of each pixel in the map is compared according to the physiological reference interval. The pixels outside the spatial region range are automatically marked as candidate abnormal pixels, ensuring that the abnormal region screening result meets both the physiological standard and the imaging experience.

[0108] According to the preliminary abnormal pixel sub-region, the spatially adjacent abnormal pixels are aggregated using a pixel decoder to generate an independent three-dimensional abnormal pixel sub-region.

[0109] Further, according to each pixel point in the preliminary abnormal pixel sub-region being marked as an abnormal pixel point as a starting point, six directly adjacent pixel positions in the three-dimensional space are traversed by the pixel decoder, if the adjacent pixel also belongs to the preliminary abnormal pixel sub-region, it is included in the independent three-dimensional abnormal pixel sub-region being currently constructed, and the current independent three-dimensional abnormal pixel sub-region is no longer expanded in space, the next abnormal pixel point not belonging to any independent three-dimensional abnormal pixel sub-region is selected as a new starting point, the abnormal pixel sub-region traversal and aggregation process is repeated, until all the pixels in the preliminary abnormal pixel sub-region are included in a certain independent three-dimensional abnormal pixel sub-region, and the independent three-dimensional abnormal pixel sub-region is generated.

[0110] The U-Net model based on deep learning is used for automatic analysis of medical images to quantitatively analyze each three-dimensional abnormal pixel sub-region, and identify an abnormal sub-region in which a local ventilation-blood flow ratio deviates from a physiological range.

[0111] Further, the U-Net model based on deep learning is used, the U-Net model based on deep learning outputs a segmentation probability map of each pixel point, a pixel set in a three-dimensional space is extracted according to the segmentation probability map, the pixel set is continuous and has a probability higher than a physiological normal distribution interval, a three-dimensional abnormal pixel sub-region is formed, a statistical distribution feature of a ventilation-blood flow ratio corresponding to all the pixel points in each three-dimensional abnormal pixel sub-region is calculated, the statistical distribution feature is compared with the physiological range point by point, a pixel point in which the ventilation-blood flow ratio falls outside the physiological range is marked, and an abnormal sub-region in which a local ventilation-blood flow ratio deviates from a physiological range is obtained according to a region composed of the pixel points.

[0112] It should be noted that in the present application, the statistical distribution feature of the ventilation-blood flow ratio corresponding to all the pixel points in each three-dimensional abnormal pixel sub-region is calculated, and the calculation process is as follows: the ventilation-blood flow ratios of all the pixel points in each three-dimensional abnormal pixel sub-region are extracted to form a one-dimensional numerical sequence, the arithmetic mean value of the one-dimensional numerical sequence is calculated, the sum of the ventilation-blood flow ratios of all the pixel points is divided by the total number of the pixel points, and the average ventilation-blood flow ratio of the three-dimensional abnormal pixel sub-region is obtained.

[0113] Based on the three-dimensional V / Q ratio map, a morphological connection is performed on a pixel region continuously deviating from the physiological range to obtain a three-dimensional abnormal sub-region set.

[0114] Further, by identifying all the pixel values deviating from the physiological range in the three-dimensional V / Q ratio map, the pixels adjacent in space and deviating from the physiological range are merged into an independent connected region, each independent connected region is a three-dimensional abnormal sub-region, and the three-dimensional abnormal sub-region set is obtained.

[0115] The three-dimensional abnormal sub-region set is subjected to threshold ventilation blood flow ratio spatial positioning and volume quantification to obtain a blood flow distribution abnormality analysis report.

[0116] The embodiment also provides a computer device suitable for the lung ventilation-perfusion imaging region three-dimensional blood flow distribution abnormality analysis method, which comprises a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to realize the lung ventilation-perfusion imaging region three-dimensional blood flow distribution abnormality analysis method proposed in the above embodiment.

[0117] The computer device can be a terminal, which comprises a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be achieved through WIFI, an operator network, NFC (Near Field Communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0118] The embodiment also provides a storage medium having a computer program stored thereon, which is executed by a processor to realize the lung ventilation-perfusion imaging region three-dimensional blood flow distribution abnormality analysis method proposed in the above embodiment. The storage medium can be realized by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic storage, a flash memory, a magnetic disk or an optical disk.

[0119] To sum up, the present application divides the irregular sub-regions by performing K-means clustering algorithm on the gray value in each section, realizes adaptive segmentation of the functional heterogeneity structure in the lung, and improves the rationality of local ratio calculation; then generates a three-dimensional V / Q ratio map through three-dimensional space relocation and integration, visualizes the spatial distribution of abnormal regions and supports volume quantification, and finally achieves the dual beneficial effects of precise positioning and clinically interpretable analysis.

[0120] It should be noted that the above examples are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the present application, and they should be covered in the scope of the claims of the present application.

Claims

1. A method of analyzing abnormal three-dimensional blood flow distribution in a lung ventilation-perfusion imaging region, characterized by: The application relates to a lung blood flow distribution analysis method and device. The lung CT image data of a patient to be analyzed is acquired, and structure segmentation of multiple sections of the lung CT image data is performed to obtain ventilation phase grayscale images and blood perfusion phase grayscale images. Based on the lung CT image data, non-black lung region is accurately extracted through threshold segmentation, and the ratio of ventilation CT value to blood flow CT value in the region is calculated to generate lung region segmentation data. The lung region segmentation result is input into a U-Net model for lung segmentation processing to obtain the grayscale value in the ventilation phase image and the grayscale value in the blood perfusion phase image. K-means clustering algorithm is performed on the grayscale value in each section to divide the left and right lungs into multiple irregular sub-regions according to spatial proximity and functional grayscale similarity. The average ventilation grayscale value and the average blood perfusion grayscale value of all pixels in each irregular sub-region are calculated to determine the local ventilation-blood flow ratio of the sub-region. The local ventilation-blood flow ratio of all irregular sub-regions in each section is repositioned and integrated in three-dimensional space through spatial coordinates to generate a three-dimensional V / Q ratio atlas. Based on the three-dimensional V / Q ratio atlas, abnormal sub-regions with a local ventilation-blood flow ratio deviating from the physiological range are identified, and a blood flow distribution abnormality analysis report is obtained through structured analysis of the abnormal sub-regions.

2. The lung ventilation-perfusion visualization region three-dimensional blood flow distribution abnormality analysis method according to claim 1, characterized by: The lung CT image data of a patient to be analyzed is acquired, and structure segmentation of multiple sections of the lung CT image data is performed to obtain ventilation phase grayscale images and blood perfusion phase grayscale images. Based on the lung CT image data, non-black lung region is accurately extracted through threshold segmentation, and the ratio of ventilation CT value to blood flow CT value in the region is calculated to generate lung region segmentation data. The lung region segmentation result is input into a U-Net model for lung segmentation processing to obtain the grayscale value in the ventilation phase image and the grayscale value in the blood perfusion phase image.

3. The lung ventilation-perfusion visualization region three-dimensional blood flow distribution abnormality analysis method according to claim 2, characterized by: K-means clustering algorithm is performed on the grayscale value in each section to divide the left and right lungs into multiple irregular sub-regions according to spatial proximity and functional grayscale similarity. The average ventilation grayscale value and the average blood perfusion grayscale value of all pixels in each irregular sub-region are calculated to determine the local ventilation-blood flow ratio of the sub-region. The local ventilation-blood flow ratio of all irregular sub-regions in each section is repositioned and integrated in three-dimensional space through spatial coordinates to generate a three-dimensional V / Q ratio atlas.

4. The lung ventilation-perfusion visualization region three-dimensional blood flow distribution abnormality analysis method according to claim 3, characterized by: Based on the three-dimensional V / Q ratio atlas, abnormal sub-regions with a local ventilation-blood flow ratio deviating from the physiological range are identified, and a blood flow distribution abnormality analysis report is obtained through structured analysis of the abnormal sub-regions. The lung CT image data of a patient to be analyzed is acquired, and structure segmentation of multiple sections of the lung CT image data is performed to obtain ventilation phase grayscale images and blood perfusion phase grayscale images. Based on the lung CT image data, non-black lung region is accurately extracted through threshold segmentation, and the ratio of ventilation CT value to blood flow CT value in the region is calculated to generate lung region segmentation data.

5. The lung ventilation-perfusion visualization region three-dimensional blood flow distribution abnormality analysis method according to claim 4, characterized by: The lung region segmentation result is input into a U-Net model for lung segmentation processing to obtain the grayscale value in the ventilation phase image and the grayscale value in the blood perfusion phase image. K-means clustering algorithm is performed on the grayscale value in each section to divide the left and right lungs into multiple irregular sub-regions according to spatial proximity and functional grayscale similarity. The average ventilation grayscale value and the average blood perfusion grayscale value of all pixels in each irregular sub-region are calculated to determine the local ventilation-blood flow ratio of the sub-region. The local ventilation-blood flow ratio of all irregular sub-regions in each section is repositioned and integrated in three-dimensional space through spatial coordinates to generate a three-dimensional V / Q ratio atlas. Based on the three-dimensional V / Q ratio atlas, abnormal sub-regions with a local ventilation-blood flow ratio deviating from the physiological range are identified, and a blood flow distribution abnormality analysis report is obtained through structured analysis of the abnormal sub-regions. Based on the registered CT image data, the decoder of the U-Nt model is used for upsampling, and the multi-scale feature fusion of the CT image data is combined to extract the ventilation phase and the blood perfusion phase gray scale image pixel by pixel to obtain the lung ventilation function gray scale distribution map and the lung blood perfusion function gray scale distribution map; According to the lung ventilation function gray scale distribution map and the lung blood perfusion function gray scale distribution map, the network architecture of the U-Net model is used for high-precision contour recognition of the lung lobe boundary to generate a lung lobe segmentation mask; Based on the lung lobe segmentation mask, a mask superposition method is used to act on the ventilation phase and the blood perfusion phase gray scale image respectively to obtain the gray scale value in the ventilation phase image and the gray scale value in the blood perfusion phase image.

6. The lung ventilation-perfusion visualization region three-dimensional blood flow distribution abnormality analysis method according to claim 5, characterized by: K-means clustering algorithm is performed on the gray scale value in each section, and the left and right lungs are respectively divided into multiple irregular sub-regions according to spatial proximity and functional gray scale similarity, and the specific steps are as follows, Based on the gray scale images of the ventilation phase and the blood perfusion phase, combined with the corresponding spatial coordinate information, a multi-dimensional feature vector of each pixel is constructed; By calculating the Euclidean distance between the multi-dimensional feature vectors of each pixel, an iterative optimization method is used to cluster and divide the left and right lungs respectively to obtain an initial lung sub-region distribution map; According to the lung sub-region distribution map, the position of each cluster center is dynamically updated to be the mean value of all pixel feature vectors in the corresponding region, and the left and right lungs are respectively generated from multiple irregular sub-regions through repeated iteration.

7. The lung ventilation-perfusion visualization region three-dimensional blood flow distribution abnormality analysis method according to claim 6, characterized by: The average ventilation gray scale value and the average blood perfusion gray scale value of all pixels in each irregular sub-region are calculated to determine the local ventilation blood flow ratio of the sub-region, and the specific steps are as follows, Based on the lung region segmentation result, the lung pixels in the ventilation phase gray scale image and the blood perfusion phase gray scale image are registered to obtain the ventilation and perfusion gray scale values corresponding in space; According to the ventilation and perfusion gray scale values, the gray scale values and ventilation CT images of the pixels in each sub-region are fused and pixel value calculation is performed by using weighted fusion to obtain the local ventilation blood flow ratio of the sub-region.

8. The lung ventilation-perfusion visualization region three-dimensional blood flow distribution abnormality analysis method according to claim 7, characterized by: The local ventilation blood flow ratio of all irregular sub-regions in each section is repositioned and integrated in three-dimensional space through spatial coordinates to generate a three-dimensional V / Q ratio map, and the specific steps are as follows, Based on the lung region segmentation results of the three sections of the coronal, sagittal and axial sections after registration, the center coordinates of each sub-region and the corresponding local ventilation blood flow ratio in each section are spatially encoded to obtain the accurate position information of each sub-region in the pixel space; According to the three-dimensional position information and ratio data, K-means clustering analysis is used to perform spatial clustering analysis on the sub-regions between adjacent sections to obtain the ventilation and perfusion conditions of the discrete regions covering the complete lung volume; The ventilation and perfusion conditions of the discrete regions are distributed to the ventilation blood flow ratio of each pixel in the V / Q ratio of the CT image by color mapping method to generate a three-dimensional V / Q ratio map.

9. The lung ventilation-perfusion visualization region three-dimensional blood flow distribution abnormality analysis method according to claim 8, characterized by: Based on the three-dimensional V / Q ratio map, abnormal sub-regions with local ventilation blood flow ratio deviating from the physiological range are identified, and the specific steps are as follows, Based on the three-dimensional V / Q ratio map, all pixels are preliminarily screened by screening to obtain a preliminary abnormal pixel sub-region. According to the preliminary abnormal pixel sub-region, the spatially adjacent abnormal pixels are aggregated by using a pixel decoder to generate an independent three-dimensional abnormal pixel sub-region; Through the automatic analysis of the medical image by the U-Net model of deep learning, each three-dimensional abnormal pixel sub-region is quantitatively analyzed, and an abnormal sub-region deviating from the physiological range of the local ventilation-blood flow ratio is identified.

10. The lung ventilation-perfusion visualization region three-dimensional blood flow distribution abnormality analysis method according to claim 9, characterized by: Through the structured analysis of the abnormal sub-region, an abnormal blood flow distribution analysis report is obtained, and the specific steps are as follows, Based on the three-dimensional V / Q ratio map, the pixel regions continuously deviating from the physiological range are morphologically connected to obtain a three-dimensional abnormal sub-region set; The three-dimensional abnormal sub-region set is subjected to threshold ventilation-blood flow ratio spatial positioning and volume quantification to obtain an abnormal blood flow distribution analysis report.

Images