A lung ventilation-perfusion visualization region image fusion method

Abstract

The application discloses a lung ventilation-perfusion development region image fusion method, relates to the technical field of image fusion, and comprises the following steps: arranging lung medical image data according to slice sequences to obtain a first image data set; performing data processing on the first image data set to generate a second image data set, inputting the second image data set into a U-Net model to perform lung segmentation processing, and obtaining a lung region mask; performing fusion processing on ventilation development images and perfusion development images after data processing to obtain a first fusion image, and generating a plurality of lung sub-regions; extracting ventilation development intensity and perfusion development intensity of each lung sub-region respectively, calculating ventilation-perfusion ratio values of the lung sub-regions, performing color coding mapping on the lung sub-regions according to the ventilation-perfusion ratio values, and generating a second fusion image. The application solves the image fusion problem of accurate color coding and visualization of ventilation-perfusion ratio values of lung adaptive sub-regions.

Description

Technical Field

[0001] This invention relates to the field of image fusion technology, and in particular to a method for image fusion of lung ventilation-perfusion imaging regions. Background Technology

[0002] Lung ventilation and perfusion imaging images are typically acquired separately using VQ radionuclide or CT-MRI. Clinically, these two images need to be spatially registered and then overlaid for analysis to assess the matching of lung function regions. Current mainstream methods mostly employ rigid registration combined with simple channel overlay, relying on manual delineation of regions of interest, which makes it difficult to achieve refined visualization of anatomical-functional coupling. Traditional methods, when performing multimodal fusion, do not rely on lung parenchymal structural constraints, easily introducing interference from non-lung tissues; furthermore, region segmentation depends on fixed thresholds or coarse-grained segmentation, failing to adaptively extract functionally homogeneous subregions, thus limiting the spatial accuracy of ventilation and perfusion image fusion.

[0003] However, conventional methods have limitations in pixel-level fusion accuracy and quantitative mapping of regional functions: firstly, they lack high-precision automatic segmentation of lung parenchyma regions and are easily affected by non-lung tissues such as the chest wall and mediastinum, leading to blurred boundaries of the fusion region; secondly, they do not achieve quantitative calculation and visual encoding of the ventilation-perfusion ratio at the anatomical sub-region level, making it difficult to intuitively reflect the local ventilation-perfusion matching status. This invention enhances the spatial mapping capability of functional parameters and improves clinical interpretation efficiency through U-Net-driven precise segmentation and K-means-guided sub-region division. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an image fusion method for lung ventilation-perfusion imaging regions to solve the problem of accurate color encoding and visualization of the ventilation-perfusion ratio in adaptive lung sub-regions. To solve the above technical problems, this invention provides the following technical solution:

[0006] This invention provides a method for image fusion of lung ventilation-perfusion imaging regions, comprising,

[0007] Lung medical imaging data are arranged according to a slice sequence to obtain a first image dataset; the lung medical imaging data includes ventilation imaging images and perfusion imaging images; the first image dataset is processed to generate a second image dataset; the second image dataset includes the processed ventilation imaging images and perfusion imaging images; the second image dataset is input into a U-Net model for lung segmentation processing to obtain a lung region mask; the processed ventilation imaging images and perfusion imaging images are fused using the lung region mask to obtain a first fused image, and the first fused image is further divided into multiple lung sub-regions; the ventilation imaging intensity and perfusion imaging intensity of each lung sub-region are extracted respectively, and the ventilation-perfusion ratio of each lung sub-region is calculated; the lung sub-regions are color-coded and mapped according to the ventilation-perfusion ratio to generate a second fused image.

[0008] As a preferred embodiment of the lung ventilation-perfusion imaging region image fusion method of the present invention, the step of arranging the lung medical image data according to the slice sequence to obtain a first image dataset includes the following steps: extracting the spatial location information and slice number of the lung medical image data; and sorting the lung medical image data according to the anatomical order based on the spatial location information and the slice number to generate the first image dataset.

[0009] As a preferred embodiment of the lung ventilation-perfusion imaging region image fusion method of the present invention, the specific steps for processing the first image dataset to generate a second image dataset are as follows: converting the format of the first image dataset to obtain standard format image data; normalizing the standard format image data to obtain normalized image data; and identifying and cropping the lung regions of the normalized image data to generate the second image dataset.

[0010] As a preferred embodiment of the lung ventilation-perfusion imaging region image fusion method of the present invention, the step of inputting the second image dataset into a U-Net model for lung segmentation processing to obtain a lung region mask includes the following steps: inputting the second image dataset into the encoder of the U-Net model for downsampling and feature extraction; upsampling through the decoder of the U-Net model, and combining skip connections to fuse multi-scale features to output a segmentation probability map; and binarizing and morphologically processing the segmentation probability map to generate the lung region mask.

[0011] As a preferred embodiment of the lung ventilation-perfusion imaging region image fusion method of the present invention, the following steps are taken: The ventilation imaging image and the perfusion imaging image after data processing are fused using the lung region mask to obtain a first fused image. Specifically, the effective lung pixel positions are determined according to the lung region mask; pixel values ​​at corresponding positions are extracted from the ventilation imaging image and the perfusion imaging image after data processing; a first transparency weight of the ventilation imaging image and a second transparency weight of the perfusion imaging image are set; the ventilation pixel value and the perfusion pixel value at each effective pixel position are fused and calculated; the fused pixel values ​​are filled into the corresponding positions to generate the first fused image; wherein each pixel contains a weighted combination of ventilation and perfusion information.

[0012] As a preferred embodiment of the lung ventilation-perfusion imaging region image fusion method of the present invention, the specific steps for dividing the first fused image into regions are as follows: extracting the horizontal coordinate, vertical coordinate, and fused gray value of each pixel from the first fused image to construct a three-dimensional feature vector; setting the number of clusters, initializing the three-dimensional feature vector with K-means clustering, and selecting initial cluster centers; iteratively calculating the distance from each pixel to each cluster center, assigning the pixel to the nearest cluster center, and updating the cluster center position until convergence, to obtain the clustering result; and dividing pixels belonging to the same cluster into the same lung sub-region according to the clustering result.

[0013] As a preferred embodiment of the lung ventilation-perfusion imaging region image fusion method of the present invention, the specific steps for extracting the ventilation imaging intensity and perfusion imaging intensity of each lung sub-region are as follows: Based on the pixel location information of each lung sub-region, a spatial range index is established for each lung sub-region; based on the spatial range index, the gray values ​​of all pixels within each lung sub-region are extracted from the data-processed ventilation imaging image, and the statistical average of the gray values ​​is calculated as the ventilation imaging intensity of the lung sub-region; the gray values ​​of all pixels within the corresponding lung sub-region are extracted from the data-processed perfusion imaging image, and the statistical average of the gray values ​​is calculated as the perfusion imaging intensity of the lung sub-region.

[0014] As a preferred embodiment of the lung ventilation-perfusion imaging region image fusion method of the present invention, the specific steps for calculating the ventilation-perfusion ratio of each lung sub-region are as follows: based on the ventilation imaging intensity and the corresponding perfusion imaging intensity of each lung sub-region, an initial ventilation-perfusion ratio is obtained through ratio calculation; the initial ventilation-perfusion ratio is normalized to standardize the ratio range to a preset interval to obtain the ventilation-perfusion ratio of each lung sub-region.

[0015] As a preferred embodiment of the lung ventilation-perfusion imaging region image fusion method of the present invention, the specific steps of color encoding mapping of each lung sub-region according to the ventilation-perfusion ratio are as follows: establishing a mapping relationship table between the ventilation-perfusion ratio and color values, setting warm color tones for low ratio regions and cool color tones for high ratio regions; querying the mapping relationship table according to the ventilation-perfusion ratio of each lung sub-region, and assigning a corresponding RGB color value to each lung sub-region.

[0016] As a preferred embodiment of the lung ventilation-perfusion imaging region image fusion method of the present invention, the specific steps for generating the second fused image are as follows: according to the assigned RGB color values, all pixels in each lung sub-region are filled with color; all lung sub-regions that have been filled with color are stitched together according to their original spatial positions to form a complete second fused image.

[0017] The beneficial effects of this invention are as follows: by accurately guiding the weighted fusion of ventilation and perfusion images through lung region masking, the dual-mode information is synchronously embedded in the anatomical structure, ensuring the spatial consistency of physiological function data; furthermore, by automatically dividing lung sub-regions through three-dimensional feature K-means clustering, the ventilation-perfusion ratio calculation is based on functionally homogeneous regions, thereby improving the fusion of color-coded lung ventilation images, and finally generating a fused image that intuitively reflects the regional ventilation-perfusion matching status. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A flowchart of an image fusion method for lung ventilation-perfusion imaging regions.

[0020] Figure 2 This is the first fused image obtained after weighted fusion.

[0021] Figure 3 This is a schematic diagram of lung lobe segmentation. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-3 As an embodiment of the present invention, this embodiment provides a method for fusion of lung ventilation-perfusion imaging regions, comprising the following steps:

[0026] S1: Arrange the lung medical image data according to the slice sequence to obtain the first image dataset.

[0027] In this embodiment of the invention, the lung medical imaging data is a DICOM format file obtained from a medical image management system. Specifically, the invention extracts a DICOM file containing ThorRoutine1.0B70s from the image sequence generated by the CT scanning device, and simultaneously obtains two functional imaging data, LungPerfusionTomo[Recon-NoAc] and LungventilationTomo[Recon-NoAc], which correspond to perfusion imaging images and ventilation imaging images, respectively.

[0028] S1.1: Extract the spatial location information and slice number of the lung medical imaging data.

[0029] This invention extracts the spatial location information of each slice by parsing the header information of a DICOM file.

[0030] The position coordinates of the slice in three-dimensional space, the position of the slice along the scanning axis, and the slice number in the sequence can all be obtained from DICOM metadata. This embodiment processes data from three standard anatomical sections: axial, coronal, and sagittal, with each section containing the same number of slice images.

[0031] S1.2: Based on spatial location information and slice number, sort the lung medical imaging data in anatomical order to generate the first image dataset.

[0032] Based on the extracted spatial location information, this invention reorganizes lung medical imaging data according to anatomical order. Specifically, for axial sections, they are arranged from head to foot; for coronal sections, they are arranged from front to back; and for sagittal sections, they are arranged from left to right.

[0033] In practice, this invention constructs a data structure containing nine subdirectories: dicom1, dicom2, dicom3, img_p1, img_p2, img_p3, img_v1, img_v2, and img_v3. dicom1, dicom2, and dicom3 store DICOM source files for the coronal, sagittal, and axial planes, respectively; img_p1, img_p2, and img_p3 store perfusion imaging images for the corresponding sections; and img_v1, img_v2, and img_v3 store ventilation imaging images for the corresponding sections. The ventilation and perfusion imaging images arranged anatomically in the nine subdirectories are combined to form the first image dataset. The first image dataset contains complete lung image sequences arranged anatomically, with each sequence position corresponding to both a ventilation and perfusion imaging image.

[0034] S2: Process the first image dataset to generate the second image dataset.

[0035] The second image dataset includes ventilated and perfusion images after data processing.

[0036] S2.1: Convert the format of the first image dataset to obtain standard format image data.

[0037] This invention converts DICOM format files in a first image dataset to PNG format. The pixel data of the DICOM file is read using the DicomToPngConverter tool and converted into the standard PNG image format. The format conversion process preserves the grayscale information of the original image while mapping the 16-bit depth data of the lung medical images in the first image dataset to an 8-bit image format suitable for display and processing. After conversion, standard format image data is obtained.

[0038] S2.2: Normalize the standard format image data to obtain normalized image data.

[0039] It should be noted that the normalization process in this invention uses the Min-Max normalization method to map the pixel values ​​of standard format image data to a uniform numerical range.

[0040] Specifically, this invention extracts the minimum and maximum pixel values ​​of each image in standard format image data, and normalizes all pixel values ​​to the [0,1] interval through linear transformation. For multi-channel images, each channel is normalized independently. Normalization eliminates intensity differences caused by different scanning parameters, ensuring that ventilation and perfusion images have comparable numerical ranges, thereby generating normalized image data.

[0041] S2.3: Identify and crop the lung region from the normalized image data to generate a second image dataset.

[0042] This invention identifies lung regions from normalized image data and locates the boundaries of lung tissue by analyzing the grayscale distribution characteristics of the image.

[0043] Specifically, since lung tissue has a lower CT value due to the presence of air, after identifying the lung region, the image is cropped according to the lung boundary coordinates to remove surrounding non-lung tissue and retain the smallest rectangular area containing the complete lung. Then, the cropped ventilation imaging image and perfusion imaging image are reorganized to form the second image dataset.

[0044] Therefore, by using step S2.3, each image in the generated second image dataset focuses on the lung region, which can reduce the amount of computation required for processing and improve the accuracy of the analysis.

[0045] S3: Input the second image dataset into the U-Net model for lung segmentation processing to obtain the lung region mask.

[0046] S3.1: Input the second image dataset into the encoder of the U-Net model for downsampling and feature extraction.

[0047] The encoder extracts features from ventilation and perfusion images in the second image dataset using 3×3 convolution operations. The convolution kernel slides across the image, extracting basic features such as edges, texture, and local shape by performing inner product operations with local image regions. After the convolution operation, the ReLU activation function is used to set values ​​less than 0 in the convolution output to 0, while values ​​greater than 0 remain unchanged. Then, max pooling is performed with a 2×2 window, and the maximum value within the window is used as the output, achieving downsampling. Downsampling reduces the spatial size of the image to one-half of its original size. As the encoder's encoding path progresses, the number of channels increases from 1 to 64, 128, 256, 512, and finally to 1024, and the extracted features shift from focusing on small-scale edges to focusing on larger-scale tissue contours.

[0048] S3.2: Upsampling is performed through the decoder of the U-Net model, and multi-scale features are fused by skip connections to output a segmentation probability map.

[0049] The decoder performs upsampling using a 2×2 transposed convolution, doubling the feature space size of the encoder output while reducing the number of channels from 1024 to 512, 256, 128, and 64. The upsampled features are then further refined and fused using a 3×3 convolution and ReLU activation. Skip connections directly pass features from the corresponding encoder layer to the same resolution layer in the decoder, compensating for information loss caused by downsampling. The decoder concatenates the upsampled features with the features passed from the skip connections and then fuses them using a 3×3 convolutional layer. Finally, a 1×1 convolution adjusts the number of output channels to correspond to the number of classes in the segmentation task. For binary classification, the output has two channels, each corresponding to the probability distribution of one class, resulting in a segmentation probability map.

[0050] Features of the encoder layer: The features of the encoder layer refer to the spatial semantic representations extracted and compressed by each layer through convolution and pooling operations in the downsampling path. Specifically, the shallow encoder (close to the input end) mainly captures high-resolution local structural features, such as lung edge contours, vascular textures, and subtle anatomical boundaries. Its receptive field is small, preserving rich spatial detail information and ensuring dual optimization of lung segmentation in terms of boundary accuracy and structural integrity.

[0051] S3.3: Binarize and morphologically process the segmentation probability map to generate a lung region mask.

[0052] After initial lung segmentation using the U-Net model, this invention performs post-processing on the segmentation probability map to improve the accuracy of the segmentation results. Specifically, binarization is performed, converting the segmentation probability map into a binary map containing only the target region and background, distinguished by 0 and 1. Morphological processing, as a classic image post-processing technique, optimizes the regions segmented by U-Net through a series of morphological operations.

[0053] U-Net is a convolutional neural network architecture specifically designed for image segmentation tasks. Its structure resembles the letter "U," hence its name. It mainly consists of two parts: an encoder (contraction path) and a decoder (expansion path), with information passed between them via skip connections.

[0054] Encoder: The encoder part is similar to a traditional convolutional neural network, using a series of convolutional and pooling layers to progressively extract high-level semantic features from the image and reduce the spatial resolution of the feature map. In this process, each convolution operation increases the number of feature channels in the image to capture richer image features; while each pooling operation halves the size of the feature map, such as the common 2×2 max pooling, allowing the U-Net model to focus on more abstract and representative features in the image, thereby obtaining contextual information. For example, when segmenting medical brain images, the encoder can extract feature representations of the approximate structures of brain tissue and ventricles from the original image through continuous downsampling.

[0055] Decoder: The decoder, symmetrical to the encoder, primarily functions to gradually restore the spatial resolution of the image through upsampling operations, transforming the abstract features extracted by the encoder into concrete segmentation results. During upsampling, transposed convolution (deconvolution) is typically used to enlarge the feature map size, while simultaneously combining high-resolution feature information from corresponding levels in the encoder (obtained through skip connections) to refine the segmentation results. For example, when segmenting tumors in medical images, the decoder utilizes features from different levels passed from the encoder to gradually restore the detailed information of the tumor, accurately delineating its boundaries.

[0056] Skip connections: Skip connections are a key design feature of U-Net. They concatenate and fuse low-resolution feature maps from different levels in the encoder with high-resolution feature maps from the corresponding upsampling steps in the decoder. The advantage of this is that it fully utilizes the low-level spatial details extracted during the encoding stage during decoding, compensating for information loss caused by downsampling. This helps the decoder more accurately locate and segment target regions, improving segmentation precision and accuracy. For example, when segmenting cell images, skip connections can pass detailed features such as cell edges extracted earlier in the encoder to the decoder, resulting in clearer and more accurate cell boundaries in the final segmentation.

[0057] At the end of a convolutional neural network, a 1×1 convolutional layer is typically used to map the fused feature map to the required number of classes. Classification is then performed using a softmax or sigmoid activation function to determine the probability of each pixel belonging to a different class, thus completing the image segmentation task. The entire U-Net model can be trained end-to-end, generating segmentation maps directly from the original image. During training, the U-Net model automatically adjusts the weights to minimize the loss function (such as the commonly used pixel-level cross-entropy loss function, which measures the difference between the true and predicted labels of each pixel), thereby learning the image features and segmentation patterns.

[0058] Problems with U-Net segmentation:

[0059] Despite U-Net's significant achievements in image segmentation tasks, some issues remain due to the complexity of images in real-world applications and the inherent limitations of the U-Net model itself.

[0060] Noise issues: Images may be introduced with various types of noise during acquisition, transmission, or processing, such as Gaussian noise and salt-and-pepper noise. When processing these noisy images, U-Net may misidentify noise as part of the target region, resulting in isolated small noise areas in the segmentation results. For example, in medical X-ray image segmentation, the noise in the image itself may cause U-Net to misidentify noise points as part of bone or lesion areas, affecting the accuracy of diagnosis.

[0061] Edge discontinuity: When segmenting target objects, the U-Net model sometimes exhibits discontinuous and jagged segmentation boundaries. This is because during the downsampling and upsampling processes of the U-Net model, information loss and feature fusion may not be precise enough, leading to inaccurate localization of target object edges. For example, when segmenting coastlines in satellite images, the segmented coastlines may appear discontinuous, affecting the accurate measurement of information such as coastline length.

[0062] Region fragmentation: For objects with complex shapes and irregular structures, the U-Net model may exhibit region fragmentation, where a single, complete target region is divided into multiple small pieces. This is because the model struggles to fully capture the overall structure and contextual information of complex targets, leading to incorrect segmentation decisions. For example, when segmenting complex tumors in medical images, the tumor region may be divided into multiple unconnected parts, making it impossible to accurately assess the size and shape of the tumor.

[0063] Poor segmentation performance for small objects: When the target object occupies a small proportion of the image, i.e., when small objects exist, the segmentation performance of the U-Net model is often unsatisfactory. This is because during the model's downsampling process, information about small objects is easily ignored or lost, leading to the inability to accurately recover and identify small objects during subsequent upsampling and segmentation processes. For example, when segmenting tiny organelles in a cell image, there may be cases of missed segmentation or inaccurate segmentation.

[0064] Implementation process demonstration:

[0065] Original Image: The original lung CT image exhibits uneven grayscale distribution, some blurring of the boundary between the lung region and surrounding tissues, and the presence of noise points. These factors all pose challenges to accurate lung region segmentation. For example, in some low-contrast areas, the lung tissue is not clearly distinguishable from other thoracic tissues, which can easily lead to segmentation errors.

[0066] U-Net Segmentation Results: The trained U-Net model was used to segment the original lung CT image, yielding preliminary segmentation results. It can be seen that U-Net can roughly segment the lung region, but some obvious problems exist. Some lung edges exhibit jagged discontinuities, due to the limitations of the U-Net model in handling complex lung edge features. Additionally, the segmentation results contain some isolated small noise regions, which are mis-segmented as part of the lung tissue and may interfere with subsequent diagnostic analysis.

[0067] Morphological Processing: Morphological processing is performed to address the issues with the U-Net segmentation results. First, erosion is used to remove isolated small noise regions. A 3×3 square structuring element is selected, and erosion is applied to the segmentation results. After erosion, isolated small noise points, whose surrounding pixels do not belong to the lung region, fail to meet the erosion conditions and are thus removed, resulting in a cleaner image. Next, dilation is performed to smooth the lung edges. Using a circular structuring element (with a radius of 5 pixels), dilation expands the lung region boundaries outward, initially connecting broken edges and making them more continuous. Then, closing is performed to further optimize the edges. Closing involves dilation followed by erosion. After dilation connects the edges, erosion removes the excess connections caused by dilation, restoring the lung region to a reasonable size and making the edges smoother and more accurate.

[0068] Erosion operation: mainly used for binary or grayscale images. Its core purpose is to shrink or thin the foreground region in the image, thereby removing isolated noise points, separating stuck targets, or smoothing object boundaries.

[0069] Erosion conditions: Binarizing and morphologically processing the segmentation probability map to generate a lung region mask can improve the anatomical accuracy of the mask, providing a cleaner lung region basis for subsequent image fusion and sub-region segmentation. This is a key preprocessing step to ensure the reliability of functional parameter calculation.

[0070] Final optimized segmentation results: After morphological processing, the segmentation results of the lung CT images were significantly optimized. The lung edges became smoother and more continuous, and the previously existing jagged edges were effectively resolved. At the same time, isolated small noise regions were successfully removed, and the segmentation results more accurately reflected the true shape and boundaries of the lungs, providing doctors with clearer and more accurate image information for diagnosing lung diseases. (Comparison images of the original image, U-Net segmentation results, morphological processing, and final optimized segmentation results can be inserted here as needed to more intuitively demonstrate the processing effect.)

[0071] Ideally, the U-Net segmentation result is clean and complete, fully segmenting the lungs without noise. After binarization and morphological processing, a lung region mask is finally generated, providing accurate spatial localization for the fusion process in step S4 using the lung region mask.

[0072] S4: The ventilation imaging image and perfusion imaging image after data processing are fused using a lung region mask to obtain a first fused image. After dividing the first fused image into regions, multiple lung sub-regions are generated.

[0073] In this invention, step S4, which uses a lung region mask to fuse the processed ventilation imaging image and perfusion imaging image to obtain a first fused image, includes the following steps A1 to A3:

[0074] Step A1: Determine the effective lung pixel positions based on the lung region mask, and extract the pixel values ​​at the corresponding positions from the ventilation imaging image and perfusion imaging image after data processing.

[0075] This invention matches a lung region mask with the processed ventilation and perfusion images in a second image dataset. In the lung region mask, a value of 1 represents the lung region, and a value of 0 represents the background region. Based on the lung region mask, pixel values ​​corresponding to the lung region positions in the ventilation and perfusion images are extracted to obtain the effective lung pixel values ​​for ventilation and perfusion.

[0076] Step A2: Set the first transparency weight of the ventilation-developed image and the second transparency weight of the perfusion-developed image, and perform a fusion calculation on the ventilation pixel value and perfusion pixel value at each effective pixel location.

[0077] Specifically, this invention employs the AlphaBlending transparency mixing method, setting a first transparency weight (represented by alpha in this embodiment) for the ventilation imaging image and a second transparency weight (represented by beta in this embodiment) for the perfusion imaging image. For each effective lung pixel location, the ventilation pixel value and the perfusion pixel value are fused using a weighted summation method. During the fusion process, the values ​​of alpha and beta are set such that the two types of images each account for 50%, achieving a balanced combination of ventilation and perfusion information.

[0078] For example, weighted fusion of transparency in a chest CT scan cross-sectional image can be performed. Alpha blending is used to add a semi-transparent image to one image, resulting in a visible combination. In OpenCV, the `addWeighted` function might be needed. The function parameters are the weights of the two images, such as alpha, beta, and gamma. The original image and the mask are blended in a 50 / 50 ratio, i.e., alpha=0.5, beta=0.5, gamma=0. The output image is the weighted sum of the two. However, it's important to note that the two images must be the same size and have the same number of channels; the mask's channel count needs to be converted first. The `cvtColor` or `merge` function can be used to convert a single channel to a three-channel image. The blending mode determines whether to multiply or blend the images based on the underlying color, using NumPy vector operations to process each pixel. The mask and CT image are then combined using algorithms such as the FLANN (Fast Nearest Neighbor) matcher and homography calculation. This method is used to extract or analyze information from specific regions (such as the lungs). A weighted fusion is performed by adjusting the transparency of the two images, using the following formula:

[0079] output=img1*alpha+img2*beta+gamma;

[0080] Where output represents the output, img1 represents chest CT scan cross-sectional image 1, and img2 represents chest CT scan cross-sectional image 2.

[0081] Step A3: Fill the corresponding positions with the pixel values ​​after fusion calculation to generate the first fused image.

[0082] Each pixel contains a weighted combination of ventilation and perfusion information.

[0083] Specifically, the pixel values ​​obtained from the fusion calculation in step A2 are filled into the corresponding lung region positions, while the background region remains 0, resulting in the following: Figure 2The first fused image is shown. Each lung pixel in the first fused image contains a weighted combination of ventilation imaging and perfusion imaging information, achieving pixel-level fusion of the two modalities.

[0084] Step S4 involves dividing the first fused image into regions, specifically including steps B1 to B4:

[0085] Step B1: Extract the x-coordinate, y-coordinate, and fused gray value of each pixel from the first fused image to construct a three-dimensional feature vector.

[0086] This invention extracts the spatial location and intensity information of each lung pixel from a first fused image. Specifically, it extracts the pixel's x-coordinate, y-coordinate, and the fused grayscale value, combining these three to construct a three-dimensional feature vector. This three-dimensional feature vector simultaneously contains the pixel's spatial location information and the fused ventilation-perfusion intensity information.

[0087] Step B2: Set the number of clusters, initialize the three-dimensional feature vectors with K-means clustering, and select the initial cluster centers.

[0088] This invention sets the number of clusters to 9, dividing the lung into 9 sub-regions. The K-means clustering algorithm initializes the constructed 3D feature vector, selecting 9 initial cluster centers. These initial cluster centers are distributed in the feature space, serving as the starting point for iteration.

[0089] Step B3: Iteratively calculate the distance from each pixel to each cluster center, assign each pixel to the nearest cluster center, and update the cluster center positions until convergence, thus obtaining the clustering result.

[0090] It should be noted that, in this embodiment of the invention, the K-means clustering algorithm iteratively performs the following operations: calculates the Euclidean distance from the three-dimensional feature vector of each pixel to the nine cluster centers, and assigns the pixel to the cluster to which the nearest cluster center belongs. The feature mean is recalculated for each cluster as the new cluster center. This process is repeated until the cluster centers no longer change significantly, at which point the algorithm converges. The clustering process considers both spatial proximity and the similarity of the merged grayscale values.

[0091] Step B4: Based on the clustering results, pixels belonging to the same cluster are divided into the same lung sub-region.

[0092] Based on the final result of K-means clustering, all pixels belonging to the same cluster are labeled as the same lung sub-region. For example... Figure 3 As shown, the lung is divided into 9 irregular sub-regions, and the pixels in each region have high consistency in spatial location and fused gray value.

[0093] S5: Extract the ventilation imaging intensity and perfusion imaging intensity of each lung sub-region, and calculate the ventilation-perfusion ratio of each lung sub-region.

[0094] In step S5, the ventilation contrast intensity and perfusion contrast intensity of each lung sub-region are extracted respectively. The specific steps include steps C1 to C3:

[0095] Step C1: Based on the pixel location information of each lung sub-region, establish the spatial range index of each lung sub-region.

[0096] It should be noted that, based on the nine lung sub-regions obtained by K-means clustering in step S4, this invention records the region affiliation of each pixel and uses the region affiliation as a spatial range index. The spatial range index determines the locations of all pixels contained in each lung sub-region.

[0097] Specifically, after the K-means clustering algorithm completes the segmentation of the first fused image, each lung pixel has been assigned to a unique sub-region. Based on the clustering results of the K-means clustering algorithm, a spatial range index is constructed for each lung sub-region. The spatial range index records the coordinate information of all pixels belonging to a lung sub-region in the image plane, thus forming an association list for precise localization.

[0098] Step C2: Based on the spatial range index, extract the gray values ​​of all pixels in each lung sub-region from the data-processed ventilation imaging image, and calculate the statistical average of the gray values ​​as the ventilation imaging intensity of the lung sub-region.

[0099] Using the spatial range index established in step C1, traverse the pixel coordinates within the i-th lung sub-region. For each coordinate, extract the corresponding pixel grayscale value from the processed ventilation imaging image in the second image dataset. After collecting the grayscale values ​​of all pixels within the i-th lung sub-region, calculate the statistical average and use it as the ventilation imaging intensity of the i-th lung sub-region. The ventilation imaging intensity is the quantitative information of ventilation, corresponding to V in the ventilation-perfusion ratio (VQ ratio), where V represents ventilation.

[0100] Step C3: Extract the gray values ​​of all pixels in the corresponding lung sub-region from the perfusion imaging image after data processing, and calculate the statistical average of the gray values ​​as the perfusion imaging intensity of the lung sub-region.

[0101] Specifically, using the same method as in step C2, pixel grayscale values ​​of each lung sub-region are extracted from the perfusion imaging image after data processing, and the statistical average is calculated as the perfusion imaging intensity. The perfusion imaging intensity is the quantitative information of perfusion, corresponding to Q in the ventilation-perfusion ratio (VQ ratio), where Q represents perfusion.

[0102] Step S5 calculates the ventilation-perfusion ratio for each lung sub-region, specifically including steps D1 and D2:

[0103] Step D1: Based on the ventilation imaging intensity and the corresponding perfusion imaging intensity of each lung sub-region, the initial ventilation-perfusion ratio is obtained through ratio calculation.

[0104] This invention calculates the ratio of ventilation imaging intensity to perfusion imaging intensity for each lung sub-region, and uses the result as the initial ventilation-perfusion ratio. The initial ventilation-perfusion ratio is the initial VQ ratio, reflecting the degree of matching between ventilation and blood perfusion.

[0105] Step D2: Normalize the initial ventilation-perfusion ratio to standardize the ratio range to a preset interval to obtain the ventilation-perfusion ratio for each lung sub-region.

[0106] To ensure the comparability of initial ventilation-perfusion ratios obtained from different patients or scanning conditions, the initial ratios of all lung sub-regions are normalized. Specifically, the minimum and maximum values ​​of the initial ratios of all sub-regions in the current batch are determined, and the initial ratios of each lung sub-region are scaled proportionally so that the final values ​​are constrained within a fixed range. Thus, the ventilation-perfusion ratios of each lung sub-region are obtained after normalization.

[0107] It should be noted that the preset range is based on the normal range of the ventilation-perfusion ratio, 0.8-1.2. A ventilation-perfusion ratio below 0.8 indicates relative over-perfusion or under-perfusion, while a ventilation-perfusion ratio above 1.2 indicates relative over-perfusion or under-perfusion.

[0108] S6: Color-coded mapping of each lung sub-region is performed based on the ventilation-perfusion ratio to generate a second fused image.

[0109] Step S6 involves color-coding and mapping each lung sub-region based on the ventilation-perfusion ratio, specifically including steps E1 and E2:

[0110] Step E1: Establish a mapping table between ventilation-perfusion ratio and color value, setting warm colors for low ratio areas and cool colors for high ratio areas.

[0111] This invention establishes a mapping table based on the ventilation-perfusion ratio of each lung sub-region calculated in step S5. The mapping table divides the ventilation-perfusion ratio range into multiple intervals, each interval corresponding to an RGB color value. For example, a ventilation-perfusion ratio below 0.8 corresponds to a red color, indicating relatively excessive perfusion or insufficient ventilation; a ventilation-perfusion ratio above 1.2 corresponds to a blue color, indicating relatively excessive ventilation or insufficient perfusion; and a ventilation-perfusion ratio in the range of 0.8-1.2 corresponds to a green color, indicating normal ventilation-perfusion matching.

[0112] Step E2: Based on the ventilation-perfusion ratio of each lung sub-region, query the mapping table and assign a corresponding RGB color value to each lung sub-region.

[0113] The ventilation-perfusion ratio of each lung sub-region is used as input and queried from the mapping table established in step E1. A unique RGB color value is determined based on the numerical range of the ventilation-perfusion ratio. To ensure color encoding consistency, this embodiment uses Normalize and ScalarMappable to create color mappings: to avoid color deviations caused by manual assignment, this invention constructs an automated color mapping system using the Normalize and ScalarMappable classes from the matplotlib library, ensuring a unique correspondence between the ratio range and RGB colors and cross-scene consistency.

[0114] Step S6 generates the second fused image, specifically including steps F1 and F2:

[0115] Step F1: Fill all pixels in each lung sub-region with color according to the assigned RGB color values.

[0116] Using the spatial range index established in step C1, each lung sub-region is traversed. For each lung sub-region, the RGB color values ​​obtained in step E2 are applied to all pixels contained within the lung sub-region. All pixels belonging to the same lung sub-region are uniformly filled with the assigned RGB color values. Color filling converts the original grayscale information into color information reflecting the ventilation-perfusion ratio.

[0117] Step F2: Stitch together all the lung sub-regions that have been filled with color according to their original spatial positions to form a complete second fused image.

[0118] After all lung sub-regions are filled with color, the colored areas are recombine based on the coordinates of each lung sub-region's pixels in the original image. The background area remains black. Each colored lung sub-region is then drawn onto a new image canvas according to its corresponding coordinates in the first fused image. Once all lung sub-regions are stitched together, a second fused image is formed. The second fused image visually displays the ventilation-perfusion ratio of each lung sub-region through color, presenting the spatial distribution of lung function status.

[0119] This embodiment also provides a computer device applicable to the lung ventilation-perfusion imaging region image fusion method, comprising: 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 implement the lung ventilation-perfusion imaging region image fusion method proposed in the above embodiment.

[0120] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0121] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the lung ventilation-perfusion imaging region image fusion method proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0122] In summary, this invention precisely guides the weighted fusion of ventilation and perfusion images through lung region masking, achieving synchronous embedding of dual-mode information within the anatomical structure and ensuring spatial consistency of physiological functional data. Furthermore, it automatically divides lung sub-regions through three-dimensional feature K-means clustering, enabling ventilation-perfusion ratio calculation based on functionally homogeneous regions, thereby improving the fusion of color-coded lung ventilation images and ultimately generating a fused image that intuitively reflects the regional ventilation-perfusion matching status.

[0123] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A lung ventilation-perfusion visualization region image fusion method, characterized by: The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model.

2. The lung ventilation-perfusion opacification region image fusion method of claim 1, wherein: The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model.

3. The lung ventilation-perfusion opacification region image fusion method of claim 2, wherein: The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model.

4. The lung ventilation-perfusion opacification region image fusion method of claim 3, wherein: The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model.

5. The lung ventilation-perfusion opacification region image fusion method of claim 4, wherein: The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model.

6. The lung ventilation-perfusion opacification region image fusion method of claim 5, wherein: The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U-Net model. The application relates to a lung region segmentation method and device based on a U Extracting the horizontal coordinate, vertical coordinate and fused gray value of each pixel from the first fusion image to construct a three-dimensional feature vector; Setting the number of clusters, K-means clustering initialization is performed on the three-dimensional feature vector, and the cluster center is selected; Through iterative calculation of the distance of each pixel point to each cluster center, the pixel point is assigned to the nearest cluster center, and the cluster center position is updated until convergence, and the clustering result is obtained; According to the clustering result, the pixel points belonging to the same cluster are divided into the same lung sub-region.

7. The lung ventilation-perfusion opacification region image fusion method of claim 6, wherein: Respectively extracting the ventilation imaging intensity and perfusion imaging intensity of each lung sub-region, the specific steps are as follows, According to the pixel position information of each lung sub-region, the spatial range index of each lung sub-region is established; Based on the spatial range index, the gray values of all pixels in each lung sub-region are extracted from the ventilation imaging image after data processing, and the statistical average value of the gray values is calculated as the ventilation imaging intensity of the lung sub-region; From the perfusion imaging image after data processing, the gray values of all pixels in the corresponding lung sub-region are extracted, and the statistical average value of the gray values is calculated as the perfusion imaging intensity of the lung sub-region.

8. The lung ventilation-perfusion opacification region image fusion method of claim 7, wherein: The ventilation perfusion ratio of each lung sub-region is calculated, and the specific steps are as follows, Based on the ventilation imaging intensity and the corresponding perfusion imaging intensity of each lung sub-region, the initial ventilation perfusion ratio is obtained by ratio operation; The initial ventilation perfusion ratio is normalized to standardize the ratio range to a preset interval, and the ventilation perfusion ratio of each lung sub-region is obtained.

9. The lung ventilation-perfusion opacification region image fusion method of claim 8, wherein: According to the ventilation perfusion ratio, each lung sub-region is color-coded and mapped, and the specific steps are as follows, Establishing the mapping relationship table of the ventilation perfusion ratio and the RGB color value, setting the low ratio area to correspond to the warm color tone, and the high ratio area to correspond to the cold color tone; According to the ventilation perfusion ratio of each lung sub-region, the mapping relationship table is queried, and the corresponding RGB color value is assigned to each lung sub-region.

10. The lung ventilation-perfusion opacification region image fusion method of claim 9, wherein: The second fusion image is generated, and the specific steps are as follows, According to the assigned RGB color value, all pixels in each lung sub-region are color-filled; All lung sub-regions that have completed color filling are spliced according to the original spatial position to form a complete second fusion image.

Images