Image labeling method and system applied to brain MRI image segmentation

Abstract

The embodiment of the application discloses an image labeling method and system applied to brain MRI image segmentation, the method comprises the following steps: acquiring brain MRI image data set of a target object, the brain MRI image data set comprises a plurality of scanning level original image sequences; performing multi-modal feature fusion processing on the original image sequence to generate an enhanced image feature set; calling a multi-level cascade segmentation network to perform hierarchical feature extraction on the enhanced image feature set to obtain a multi-scale anatomical structure feature map; and performing region boundary optimization processing based on the multi-scale anatomical structure feature map to generate a labeled brain structure segmentation image. Therefore, the boundaries of various structures of the brain can be accurately defined, the segmented image is more accurate and clear, and thus the image segmentation labeling of the brain MRI image can be accurately and clearly realized.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, specifically to an image annotation method and system for segmenting brain MRI images. Background Technology

[0002] In the research and processing of medical images, the segmentation of brain MRI images is crucial. By segmenting brain MRI images, different brain structures can be clearly distinguished. With the continuous development of medical imaging technology, the requirements for the accuracy and precision of brain MRI image segmentation are becoming increasingly stringent. Therefore, how to achieve more accurate and comprehensive brain MRI image segmentation has become a research hotspot in this field.

[0003] Currently, some traditional brain MRI image segmentation techniques struggle to fully integrate multimodal image features when dealing with complex brain structures, resulting in incomplete image information and an inability to deeply explore brain structural features. Other techniques lack effective capture of features at different scales and levels during feature extraction, leading to insufficient detail in the presentation of brain anatomy and an inability to fully reflect the true state of brain structure. Furthermore, their ability to optimize region boundaries during image segmentation and annotation is inadequate, resulting in inaccurate and unclear image boundaries after segmentation and annotation. Therefore, achieving accurate and clear image segmentation and annotation of brain MRI images remains a significant technical challenge that needs to be overcome. Summary of the Invention

[0004] This invention provides an image annotation method and system for brain MRI image segmentation, which can accurately and clearly achieve image segmentation and annotation of brain MRI images.

[0005] In a first aspect, embodiments of the present invention provide an image annotation method for brain MRI image segmentation, applied to an image annotation system. The method includes: acquiring a brain MRI image data set of a target object, the brain MRI image data set including original image sequences at multiple scanning levels; performing multimodal feature fusion processing on the original image sequences to generate an enhanced image feature set; calling a multi-layer cascaded segmentation network to perform hierarchical feature extraction on the enhanced image feature set to obtain a multi-scale anatomical structure feature map; and performing region boundary optimization processing based on the multi-scale anatomical structure feature map to generate an annotated brain structure segmentation image.

[0006] Secondly, embodiments of the present invention provide an image annotation system, comprising:

[0007] processor;

[0008] Storage device, on which computer programs are stored,

[0009] When the computer program is executed by the processor, the processor implements any of the image annotation methods described above for segmenting brain MRI images.

[0010] This invention provides a readable storage medium storing a program or instructions that, when executed by a processor, implement the steps of the image annotation method applied to brain MRI image segmentation.

[0011] Therefore, this invention first acquires a set of brain MRI image data containing original image sequences from multiple scanning levels. Then, it performs multimodal feature fusion processing to effectively integrate features from multiple modalities, thereby generating an enhanced image feature set. This makes the information contained in the images more comprehensive and in-depth, while also uncovering features that are not easily noticed. Furthermore, a multi-layer cascaded segmentation network is used to perform hierarchical feature extraction on the enhanced image feature set, enabling the capture of brain anatomical structural features at different scales and levels, resulting in a multi-scale anatomical structural feature map. Finally, based on the multi-scale anatomical structural feature map, region boundary optimization processing is performed to generate an annotated brain structure segmentation image. This accurately defines the boundaries of various brain structures, making the segmented image more accurate and clear. In this way, accurate and clear image segmentation and annotation of brain MRI images can be achieved. Attached Figure Description

[0012] Figure 1 This is a flowchart of an image annotation method for brain MRI image segmentation provided in an embodiment of the present invention.

[0013] Figure 2 This is a schematic diagram of the basic structure of an image annotation system provided in an embodiment of the present invention. Detailed Implementation

[0014] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0015] See Figure 1 As shown, this figure is a flowchart of an image annotation method for brain MRI image segmentation provided by an embodiment of the present invention. This method can be applied to image annotation systems. Figure 1 As shown, the method may include steps 110-140.

[0016] Step 110: Obtain a set of brain MRI images of the target object, wherein the set of brain MRI images includes raw image sequences of multiple scanning levels.

[0017] In this embodiment, taking a patient suspected of having a brain disease as an example, the patient's brain can be scanned from all angles using existing MRI scanning equipment. During the scan, the equipment acquires image data from multiple different scanning levels according to preset slice spacing and scanning angles. This data is recorded in the form of raw image sequences, each containing detailed information about the brain tissue at that level. For example, the scanning equipment acquires image data from several scanning levels from the top to the bottom of the brain at different time points and angles. Each level of image is a two-dimensional grayscale image. The grayscale values ​​in these images represent the degree of response of different brain tissues to MRI signals. For example, gray matter, white matter, and cerebrospinal fluid will exhibit different grayscale characteristics in the images, thus constituting a set of brain MRI image data.

[0018] Step 120: Perform multimodal feature fusion processing on the original image sequence to generate an enhanced image feature set.

[0019] In this embodiment, multimodal feature fusion processing is performed on the original image sequences in the brain MRI image data set obtained above. The aim is to integrate image features from different modalities to obtain more comprehensive and accurate brain information, thereby generating an enhanced image feature set.

[0020] In an optional embodiment, the step of performing multimodal feature fusion processing on the original image sequence to generate an enhanced image feature set includes:

[0021] Step 121: Perform cross-modal feature alignment processing on each scanning level in the original image sequence to obtain a spatially aligned cross-modal image sequence.

[0022] In this embodiment, for the original image sequence of each scanning level, images from different modalities may differ in spatial location and scale. For example, the acquired image data includes different modalities such as T1-weighted images and T2-weighted images. Taking one scanning level as an example, the brain tissue contours in the T1-weighted image and the T2-weighted image may have slight positional deviations. To enable better feature fusion of these images from different modalities, cross-modal feature alignment processing needs to be performed. Using an image registration algorithm, images from one modality are used as a reference, and operations such as translation, rotation, and scaling are performed on images from other modalities to achieve spatial alignment. For example, using the T1-weighted image as a reference, by calculating the positional differences of feature points of the same anatomical structure in the T2-weighted image and the T1-weighted image, an image transformation algorithm is used to transform the T2-weighted image accordingly, so that the brain structures in the two modalities are accurately aligned, thereby obtaining a spatially aligned cross-modal image sequence.

[0023] Step 122: Perform cross-modal feature transformation processing on the spatially aligned cross-modal image sequence to generate a transformed feature set with target dimension mapping; wherein, the cross-modal feature transformation processing includes performing local texture enhancement and global structure preservation operations on the original image sequence of each scanning level, so that each feature map in the transformed feature set contains cross-modal complementary anatomical information.

[0024] In this embodiment, the spatially aligned cross-modal image sequence obtained after cross-modal feature alignment processing is further subjected to cross-modal feature transformation processing. For the original image sequence at each scanning level, local texture enhancement is performed. For example, a local filtering algorithm is used to analyze each small region in the image. Taking a 3×3 pixel neighborhood as an example, by calculating the grayscale value changes of the pixels within the neighborhood, an appropriate filtering kernel is used for filtering to highlight the texture features of the region.

[0025] For subtle textures in brain tissue, such as the direction of nerve fibers, this local texture enhancement operation can more clearly reveal them. Simultaneously, global structure preservation operations are performed, utilizing global feature extraction algorithms such as principal component analysis to extract the overall structural information of the image, ensuring that the global structure of the image is not destroyed while enhancing local texture. After these operations, the features of different modalities are fused and transformed to generate a transformed feature set with target dimension mapping. Each feature map in this set incorporates complementary anatomical information from different modalities. For example, a single feature map may contain information about brain tissue density from T1-weighted images and information about tissue water content from T2-weighted images, making the feature map more comprehensively reflect the anatomical structure of the brain.

[0026] Step 123: Call the completed attention weight allocation model to assign weights to different modal features in the transformation feature set and generate a modal weight distribution map; perform weighted fusion processing on the transformation feature set according to the modal weight distribution map to generate the enhanced image feature set.

[0027] In this embodiment, a transformation feature set has already been generated. Next, a pre-tuned attention weight allocation model is invoked. This model, trained and optimized with extensive data, assigns weights to different modal features in the transformation feature set based on their importance in reflecting brain structural information. For example, in the analysis of certain brain diseases, features in T1-weighted images may be more important for identifying the boundaries of lesion regions, while features in T2-weighted images may be more crucial for determining the nature of the lesion. The attention weight allocation model assigns corresponding weights to each modal feature based on these characteristics, generating a modality weight distribution map. In this map, each modality feature has a corresponding weight value, the magnitude of which reflects the importance of the modality feature. Then, based on this modality weight distribution map, the transformation feature set undergoes weighted fusion processing. For each feature map in the transformation feature set, weighted calculations are performed according to the corresponding weight values. Finally, the weighted results are stitched together to generate an enhanced image feature set. This enhanced image feature set fully integrates the advantageous information of different modalities, enabling a more accurate reflection of the brain's anatomical structure.

[0028] Step 130: Call the multi-layer cascaded segmentation network to perform hierarchical feature extraction on the enhanced image feature set to obtain multi-scale anatomical structure feature maps.

[0029] In this embodiment, a multi-layer cascaded segmentation network is used to further process the generated enhanced image feature set in order to extract brain anatomical features at different scales, thereby obtaining a multi-scale anatomical feature map.

[0030] As one implementation method, the step of calling a multi-layer cascaded segmentation network to perform hierarchical feature extraction on the enhanced image feature set to obtain multi-scale anatomical structure feature maps includes:

[0031] Step 131: Input the enhanced image feature set into the first encoder of the multi-layer cascaded segmentation network to extract the first-level anatomical structure feature map; input the first-level anatomical structure feature map into the second encoder of the multi-layer cascaded segmentation network to extract the second-level anatomical structure feature map; input the second-level anatomical structure feature map into the third encoder of the multi-layer cascaded segmentation network to extract the third-level anatomical structure feature map.

[0032] In this embodiment, the enhanced image feature set is input into the first encoder of a multi-layer cascaded segmentation network. The first encoder performs a series of convolution and pooling operations on the input feature set. For example, by performing convolution operations with feature maps in the enhanced image feature set using convolution kernels of different sizes, local features at different scales are extracted. For instance, the first encoder uses 3×3 and 5×5 convolution kernels. A 3×3 convolution kernel can extract smaller-scale detailed features, such as subtle textures in brain tissue; a 5×5 convolution kernel can extract relatively larger-scale structural features, such as the approximate outlines of sulci and gyri. After these operations, a first-level anatomical structure feature map is obtained.

[0033] Next, the first-level anatomical feature map is input into the second encoder, which further extracts higher-level features based on the first level. It may use larger convolutional kernels or more convolutional layers to process the first-level feature map and extract more macroscopic anatomical features, such as the approximate boundaries of different lobes of the brain, thus obtaining the second-level anatomical feature map.

[0034] Then, the second-level anatomical feature map is input into the third encoder, which will continue to extract more abstract and higher-level features, such as the overall structural features of the entire brain, to obtain the third-level anatomical feature map.

[0035] Step 132: Perform multi-scale feature fusion processing on the first-level anatomical structure feature map, the second-level anatomical structure feature map, and the third-level anatomical structure feature map to generate the multi-scale anatomical structure feature map; wherein, the multi-scale feature fusion processing includes upsampling the first-level anatomical structure feature map, downsampling the third-level anatomical structure feature map, and adjusting the contribution weights of feature maps at different scales through a channel attention mechanism.

[0036] In this embodiment, after obtaining the first, second, and third-level anatomical structure feature maps, multi-scale feature fusion processing is performed. First, the first-level anatomical structure feature map is upsampled. Since the first-level feature map extracts small-scale detailed features, upsampling, such as bilinear interpolation, enlarges its size to match other feature maps, allowing it to be fused with other feature maps at the same scale. For the third-level anatomical structure feature map, downsampling is performed. Because the third-level feature map extracts large-scale macroscopic features, downsampling allows its size to be compared with other features. Figure 1 To facilitate integration.

[0037] Then, the contribution weights of feature maps at different scales are adjusted using a channel attention mechanism. This mechanism analyzes the importance of each feature map across different channels. For example, in some channels, first-level feature maps may be more important for displaying fine brain structures, while in others, third-level feature maps may be more crucial for understanding the overall brain layout. Based on these analyses, corresponding weights are assigned to each channel of the feature maps at different scales. These weight-adjusted feature maps are then stitched together to generate a multi-scale anatomical feature map. This map integrates brain anatomical features at different scales, providing a more comprehensive reflection of brain structural information.

[0038] Step 140: Based on the multi-scale anatomical feature map, perform region boundary optimization processing to generate a labeled brain structure segmentation image.

[0039] In this embodiment, based on the generated multi-scale anatomical feature map, region boundary optimization processing is carried out to finally generate a labeled brain structure segmentation image, so as to more clearly divide the different structural regions of the brain.

[0040] In a preferred embodiment, the step of optimizing the region boundaries based on the multi-scale anatomical feature map to generate an annotated brain structure segmentation image includes:

[0041] Step 141: Perform initial segmentation processing on the multi-scale anatomical feature map to generate an initial brain structure segmentation mask.

[0042] In this embodiment, initial segmentation processing is performed on the multi-scale anatomical feature map to initially divide the regions of different brain structures and generate an initial brain structure segmentation mask.

[0043] As a preferred technical solution, the initial segmentation processing of the multi-scale anatomical feature map to generate an initial brain structure segmentation mask includes:

[0044] Step 1410: Input the multi-scale anatomical structure feature map into the fully connected decoder of the multi-layer cascaded segmentation network to generate an initial probability distribution map; perform threshold segmentation processing on the initial probability distribution map to obtain a binarized segmentation result; perform connected component analysis processing on the binarized segmentation result to remove isolated noise regions and obtain a denoised binarized segmentation result; perform topological correction processing on the denoised binarized segmentation result based on a preset brain anatomical structure template to generate the initial brain structure segmentation mask.

[0045] In this embodiment, multi-scale anatomical feature maps are input into the fully connected decoder of a multi-layer cascaded segmentation network. The fully connected decoder performs a series of deconvolution and fully connected operations on the input feature maps, converting them into an initial probability distribution map. Each pixel value in this initial probability distribution map represents the probability that the pixel belongs to a different brain structure. For example, in a probability distribution map, a pixel with a probability value of 0.8 indicates that the pixel has an 80% chance of belonging to a gray matter region. Next, threshold segmentation is performed on the initial probability distribution map, setting an appropriate threshold, such as 0.5. Pixels with probability values ​​greater than 0.5 are set to 1, and pixels with probability values ​​less than 0.5 are set to 0, thus obtaining a binarized segmentation result. In this binarized image, white pixels represent regions that may belong to the target brain structure, and black pixels represent other regions.

[0046] However, binarization may generate some isolated noise points that could interfere with subsequent analysis. Therefore, connected component analysis is performed on the binarized segmentation results. A connected component labeling algorithm is used to identify connected regions in the image, and small isolated regions are treated as noise and removed, resulting in a denoised binarized segmentation result. Finally, topological correction is performed on the denoised binarized segmentation result based on a pre-set brain anatomical structure template. The pre-set template contains topological information about normal brain structures, such as the relative positions and shapes of different brain regions. The denoised binarized segmentation result is compared and adjusted with the template, and regions that do not conform to the template's topological structure are corrected, thereby generating an initial brain structure segmentation mask.

[0047] Step 142: Perform boundary confidence analysis on the initial brain structure segmentation mask to determine the confidence distribution map of the blurred boundary region.

[0048] In this embodiment, in order to further optimize the boundary of brain structure segmentation, a boundary confidence analysis is performed on the generated initial brain structure segmentation mask to determine the confidence distribution map of the blurred boundary region.

[0049] In one implementation, the step of performing boundary confidence analysis on the initial brain structure segmentation mask to determine the confidence distribution map of the blurred boundary region includes:

[0050] Step 1420: Calculate the boundary gradient of the initial brain structure segmentation mask to generate a boundary gradient intensity map; evaluate the feature uncertainty of the multi-scale anatomical structure feature map to generate a feature uncertainty distribution map; perform weighted fusion processing on the boundary gradient intensity map and the feature uncertainty distribution map to generate an initial confidence map; perform regional connectivity analysis on the initial confidence map to select continuous regions below a preset confidence threshold as the boundary fuzzy regions; generate the confidence distribution map based on the spatial distribution of the boundary fuzzy regions.

[0051] In this embodiment, boundary gradient calculation is first performed on the initial brain structure segmentation mask. Using gradient calculation algorithms, such as the Sobel operator, the gradient value of each pixel in the mask is calculated, generating a boundary gradient intensity map. In this map, regions with larger gradient values ​​indicate areas of significant boundary changes, which may be the boundaries of different brain structures. For example, at the boundary between gray matter and white matter, the grayscale value of the pixel changes significantly, and the gradient value will also be correspondingly larger. Next, feature uncertainty assessment is performed on the multi-scale anatomical feature map. Using several uncertainty assessment algorithms, the stability and reliability of each feature in the feature map are analyzed, generating a feature uncertainty distribution map. The values ​​in this feature uncertainty distribution map reflect the degree of uncertainty of each feature location.

[0052] Next, the boundary gradient intensity map and the feature uncertainty distribution map are weighted and fused. Based on the importance of the two maps, appropriate weights are assigned; for example, the boundary gradient intensity map has a weight of 0.6, and the feature uncertainty distribution map has a weight of 0.4. The values ​​of corresponding pixels in the two maps are weighted according to their respective weights to generate an initial confidence map. In this initial confidence map, the value of each pixel represents the confidence level of that location as a boundary. Regional connectivity analysis is then performed on the initial confidence map. Through connectivity analysis algorithms, continuous regions in the map with confidence levels below a preset threshold (e.g., 0.3) are identified; these regions are the boundary ambiguity regions.

[0053] Finally, based on the spatial distribution of these blurred boundary regions, a confidence distribution map is generated. In this confidence distribution map, each blurred boundary region has a corresponding confidence value, which can help to optimize the boundary more accurately in the future.

[0054] Step 143: Perform iterative morphological optimization on the initial brain structure segmentation mask based on the confidence distribution map to generate an optimized brain structure segmentation mask; wherein, the iterative morphological optimization operation includes performing dilation-erosion sequence processing on the blurred boundary region and adjusting the size of the dilation kernel based on the confidence distribution map.

[0055] In this embodiment, based on the generated confidence distribution map, an iterative morphological optimization operation is performed on the initial brain structure segmentation mask to further optimize the boundaries of brain structure segmentation and generate an optimized brain structure segmentation mask.

[0056] In an optional embodiment, the step of performing iterative morphological optimization on the initial brain structure segmentation mask based on the confidence distribution map to generate an optimized brain structure segmentation mask includes:

[0057] Step 1430: Determine the corresponding morphological operation intensity parameters based on the confidence values ​​of each region in the confidence distribution map; perform morphological closing operations with a fixed kernel size on the target confidence regions in the initial brain structure segmentation mask; perform morphological opening operations with an adaptive kernel size on the blurred boundary regions, wherein the kernel size is negatively correlated with the confidence value; repeat the morphological closing and opening operations until the confidence values ​​of all regions in the confidence distribution map reach a preset optimization threshold; use the morphological processing result obtained in the last iteration as the optimized brain structure segmentation mask.

[0058] In this embodiment, the morphological operation intensity parameters are first determined based on the confidence values ​​of each region in the confidence distribution map. For target confidence regions with high confidence, whose boundaries are relatively clear, a fixed kernel-size morphological closing operation is performed. Morphological closing can fill small holes, making the boundaries more continuous. For example, a 3×3 structuring element is used to perform the closing operation on these regions. For regions with blurred boundaries, due to their low confidence, an adaptive kernel-size morphological opening operation is performed. The kernel size is negatively correlated with the confidence value; that is, the lower the confidence, the larger the kernel size. For example, for a region with blurred boundaries and a confidence of 0.2, a 5×5 structuring element is used for the opening operation; for a region with a confidence of 0.1, a 7×7 structuring element is used. The opening operation can remove some noise and small protrusions, making the boundaries smoother. Then, the morphological closing and opening operations are repeated. After each iteration, the confidence values ​​of each region in the confidence distribution map are re-evaluated until the confidence values ​​of all regions reach a preset optimization threshold (e.g., 0.6). Finally, the morphological processing results obtained from the last iteration are used as the optimized brain structure segmentation mask.

[0059] Step 144: Overlay and fuse the optimized brain structure segmentation mask with the original image sequence to generate the labeled brain structure segmentation image.

[0060] In this embodiment, the optimized brain structure segmentation mask is superimposed and fused with the original image sequence to generate an annotated brain structure segmentation image, making the segmentation results more intuitive and accurate.

[0061] As a preferred embodiment, the step of overlaying and fusing the optimized brain structure segmentation mask with the original image sequence to generate the annotated brain structure segmentation image includes:

[0062] Step 1440: Smooth the edges of the optimized brain structure segmentation mask to generate a smooth segmentation contour; mix the grayscale values ​​of the smooth segmentation contour with those of the original image sequence to generate a preliminary labeled image; perform color encoding on the boundary regions of different anatomical structures in the preliminary labeled image to generate a pseudo-color annotation layer with discriminative power; spatially align and overlay the pseudo-color annotation layer with the original image sequence to generate the labeled brain structure segmentation image.

[0063] In this embodiment, the optimized brain structure segmentation mask is first subjected to edge smoothing. Smoothing algorithms, such as Gaussian filtering, are used to smooth the mask edges, generating smooth segmentation contours. For example, a Gaussian filter with a standard deviation of 1 is used to filter the mask edges. Then, the smoothed segmentation contours are blended with the grayscale values ​​of the original image sequence using transparency blending. A transparency parameter, such as 0.5, is set, and the smoothed segmentation contours are blended with the grayscale values ​​of the original image sequence according to this transparency, generating a preliminary labeled image. This preliminary labeled image clearly shows both the information from the original image and the approximate location of the segmentation contours. Next, color encoding is performed on the boundary regions of different anatomical structures in the preliminary labeled image.

[0064] In another preferred embodiment, the step of performing color encoding processing on the boundary regions of different anatomical structures in the preliminary labeled image to generate a pseudo-color labeling layer with discriminative power includes:

[0065] Step 14401: Assign a unique color identifier to different anatomical regions according to the preset brain anatomy classification rules.

[0066] In this embodiment, the pre-defined brain anatomical structure classification rules are based on medical anatomy. For example, the brain is divided into different lobes such as the frontal lobe, parietal lobe, temporal lobe, and occipital lobe, as well as different brain tissues such as the thalamus and hippocampus. Each anatomical region is assigned a unique color identifier, such as red for the frontal lobe, blue for the parietal lobe, green for the temporal lobe, yellow for the occipital lobe, purple for the thalamus, and orange for the hippocampus. These color identifiers are carefully selected to ensure sufficient differentiation in subsequent displays, facilitating observation and analysis of brain structures by doctors or researchers.

[0067] Step 14402: Perform category matching processing on each connected region in the optimized brain structure segmentation mask to determine the corresponding target color identifier.

[0068] In this embodiment, for each connected region in the optimized brain structure segmentation mask, its location, shape, and other features in the brain are matched with preset brain anatomical structure classification rules. For example, by analyzing the shape and location of a connected region and finding that it matches the features of the frontal lobe, the target color identifier corresponding to that connected region is determined to be red. This process can use relevant common image analysis algorithms and medical knowledge models to accurately classify each connected region, thereby determining the correct target color identifier.

[0069] Step 14403: Based on the target color identifier, perform gradient color filling processing on the boundary pixels of the connected region to generate the pseudo-color annotation layer; wherein, the color saturation of the gradient color filling processing is positively correlated with the distance from the boundary pixel to the center of the segmentation contour.

[0070] In this embodiment, for each connected region with a defined target color identifier, a gradient color fill is applied to its boundary pixels. Taking a connected region as an example, its target color identifier is red. The gradient color fill begins from the center of the segmentation contour and proceeds towards the boundary pixels. The boundary pixels farther from the center of the segmentation contour have higher color saturation. For example, at boundary pixels closer to the center, the red saturation might be 0.3; as the distance increases, the saturation gradually rises, reaching 0.8 at the boundary pixels farthest from the center. This gradient color fill process generates a distinctive pseudo-color annotation layer, making the boundaries of different anatomical structures clearer and easier to observe and distinguish.

[0071] In an alternative technical solution, the step of spatially aligning and superimposing the pseudo-color annotation layer with the original image sequence to generate the annotated brain structure segmentation image includes:

[0072] Step 14404: Perform normalized coordinate transformation on the pixel coordinates of the pseudo-color annotation layer to generate a normalized coordinate mapping table aligned with the physical coordinate system of the original image sequence; wherein, the normalized coordinate transformation includes mapping the pixel positions of the pseudo-color annotation layer to the scanning interlayer resolution ratio space of the original image sequence, and compensating for interlayer displacement errors through a bilinear interpolation algorithm.

[0073] In this embodiment, since the pseudo-color annotation layer and the original image sequence may differ in coordinate system and resolution, a normalized coordinate transformation is required. First, the pixel positions of the pseudo-color annotation layer are mapped to the inter-layer resolution ratio space of the original image sequence. For example, the original image sequence has a certain ratio in inter-layer resolution, with an inter-layer spacing of 1 mm, and the pixel positions of the pseudo-color annotation layer need to be repositioned according to this inter-layer resolution. Inter-layer displacement errors are compensated using a bilinear interpolation algorithm. For pixels in the pseudo-color annotation layer located between two layers, their accurate position in the original image sequence coordinate system is calculated using bilinear interpolation based on their relative position between the two layers. After these operations, a normalized coordinate mapping table aligned with the physical coordinate system of the original image sequence is generated, ensuring that the pseudo-color annotation layer and the original image sequence correspond accurately in spatial location.

[0074] Step 14405: Perform spatial resampling processing on the color channels of the pseudo-color annotation layer according to the standardized coordinate mapping table to generate a resampled pseudo-color layer that perfectly matches the pixel spatial distribution of the original image sequence; wherein, the spatial resampling processing adopts an isotropic interpolation kernel function, and the kernel width parameter of the isotropic interpolation kernel function is consistent with the dimension of the intra-layer pixel spacing of the original image sequence.

[0075] In this embodiment, the color channels of the pseudo-color annotation layer are spatially resampled based on the generated standardized coordinate mapping table. An isotropic interpolation kernel function is used, which can perform uniform interpolation in all directions. The kernel width parameter is dimensionally consistent with the intra-layer pixel spacing of the original image sequence. For example, if the intra-layer pixel spacing of the original image sequence is 0.5mm, then the kernel width parameter will be set according to this spacing. Through this spatial resampling process, the color channels of the pseudo-color annotation layer are resampled, so that the resampled pseudo-color layer perfectly matches the original image sequence in pixel space distribution. This ensures accurate fusion of color information with the original image during subsequent overlay processes.

[0076] Step 14406: Input the resampled pseudocolor layer into the transparency mixing channel, perform multi-channel pixel-level overlay operation processing to generate a mixed labeled image; wherein, the overlay operation processing includes performing linear brightness preservation transformation on the grayscale values ​​of the original image sequence, and performing dynamic range compression on the RGB color components of the resampled pseudocolor layer, so that the pixel values ​​after overlay do not exceed the preset maximum display dynamic range.

[0077] In this embodiment, the resampled pseudocolor layer is input to the transparency mixing channel for multi-channel pixel-level overlay processing. For the grayscale values ​​of the original image sequence, a linear brightness-preserving transform is performed to retain the brightness information of the original image during overlay, ensuring that the overall brightness distribution of the brain does not change significantly. For example, a linear transform function is used to appropriately scale and offset the grayscale values ​​of the original image to ensure that the brightness is within a reasonable range. Simultaneously, dynamic range compression is performed on the RGB color components of the resampled pseudocolor layer. Since the value range of the RGB color components may be large, compression is necessary to prevent the overlaid pixel values ​​from exceeding the preset maximum display dynamic range (e.g., 0-255). For example, a compression function is used to map the values ​​of the color components to a suitable range. After these operations, a hybrid annotation image is generated, which contains information from both the original image and the color information of the pseudocolor annotation layer.

[0078] Step 14407: Perform edge sharpening and anti-aliasing processing on the hybrid labeled image to generate an intermediate labeled image with enhanced boundaries; wherein, the edge sharpening processing adopts an anisotropic diffusion filtering algorithm, and the gradient threshold parameter of the anisotropic diffusion filtering algorithm matches the gray-level gradient distribution statistics of the original image sequence.

[0079] In this embodiment, edge sharpening and anti-aliasing are performed on the hybrid labeled image. An anisotropic diffusion filtering algorithm is used for edge sharpening. This algorithm can perform targeted diffusion processing based on gradient information in different directions within the image. The gradient threshold parameter is matched with the gray-level gradient distribution statistics of the original image sequence. For example, a suitable gradient threshold is determined by statistical analysis of the gray-level gradient of the original image sequence. When the gradient value in the image is greater than the threshold, the algorithm performs a smaller diffusion process to highlight the edges; when the gradient value is less than the threshold, a larger diffusion process is performed to smooth the image. Simultaneously, anti-aliasing is performed. An anti-aliasing algorithm, such as an interpolation-based algorithm, is used to smooth jagged edges in the image, generating an intermediate labeled image with enhanced boundaries, making the boundaries of brain structures clearer and smoother.

[0080] Step 14408: Perform adaptive contrast adjustment processing based on the boundary intensity distribution of the intermediate labeled image to generate the final displayed brain structure segmentation image; wherein, the adaptive contrast adjustment processing includes performing histogram equalization on low contrast regions and performing nonlinear gamma correction on high gradient regions, and the correction coefficient of the nonlinear gamma correction is negatively correlated with the local gradient magnitude.

[0081] In this embodiment, adaptive contrast adjustment is performed based on the boundary intensity distribution of the intermediate labeled image. For low-contrast regions in the image, histogram equalization is performed. By statistically analyzing the gray-level histogram of low-contrast regions, the gray-level values ​​are redistributed, thereby improving the contrast of these regions. For example, the gray-level values ​​of low-contrast regions are expanded from a narrow range to a wider range, thereby increasing the visual clarity of the image. Simultaneously, nonlinear gamma correction is performed on high-gradient regions. The correction coefficient of nonlinear gamma correction is negatively correlated with the local gradient magnitude; that is, the larger the gradient magnitude, the smaller the correction coefficient. This avoids excessive contrast enhancement in high-gradient regions, which could lead to an image that is too bright or too dark. After these operations, the final brain structure segmentation image is generated, with more appropriate contrast and clearer, more discernible brain structure segmentation.

[0082] Step 14409: Perform color space verification processing on the final displayed brain structure segmentation image to ensure that the chromaticity channel of the pseudo-color annotation layer and the luminance channel of the original image sequence do not overlap or conflict within the display color gamut; wherein, the color space verification processing includes converting the RGB color space to the YUV space and then detecting the orthogonality index of the luminance component and the chromaticity component. If an overlapping area is detected, the chromaticity coordinates are remapped to a preset safe color palette range.

[0083] In this embodiment, the final displayed brain structure segmentation image undergoes color space verification processing. First, the RGB color space is converted to YUV space, which separates color information into luminance (Y) and chrominance (U, V) components. Then, the orthogonality index of the luminance and chrominance components is detected, and the correlation between them is calculated. If overlapping areas are detected, indicating interference between the luminance and chrominance components, the chrominance coordinates need to be remapped to a preset safe color palette range. This preset safe color palette range is determined based on the color gamut of the display device and the requirements for displaying medical images. By mapping the chrominance coordinates to this range, it is ensured that the chrominance channels of the pseudo-color annotation layer and the luminance channels of the original image sequence do not overlap or conflict within the display color gamut, thereby guaranteeing the display quality and accuracy of the image.

[0084] In one independent embodiment, after generating the annotated brain structure segmentation image, the method further includes: performing cross-level three-dimensional voxel fusion processing on the annotated brain structure segmentation image to generate a three-dimensional brain anatomical model; wherein the cross-level three-dimensional voxel fusion processing includes mapping the annotation results of each scanning level to a three-dimensional spatial coordinate system according to the inter-slice resolution ratio, and compensating for inter-slice displacement errors using an isotropic interpolation algorithm; performing surface mesh optimization processing on the three-dimensional brain anatomical model to generate a smooth and continuous three-dimensional visualization model; wherein the surface mesh optimization processing includes eliminating voxel stitching noise based on the Laplacian smoothing algorithm and adjusting the mesh density using an adaptive triangulation algorithm, where the mesh density is negatively correlated with the local radius of curvature; spatially registering the three-dimensional visualization model with a standard brain anatomical atlas and calculating the volume difference coefficient of each anatomical structure; wherein the spatial registration includes aligning based on feature point rigid transformation and optimizing non-rigid deformation parameters using a mutual information maximization algorithm; generating a brain structure abnormality detection report based on the volume difference coefficient, and storing the three-dimensional visualization model and the abnormality detection report together in a medical image database.

[0085] In this embodiment, for the generated annotated brain structure segmentation image, cross-level 3D voxel fusion processing is performed. First, the annotation results of each scanning level are mapped to a 3D spatial coordinate system according to the inter-level resolution ratio. For example, if the inter-level resolution is known to be 1 mm, the position of each voxel in the 2D segmentation image of each scanning level is determined in 3D space according to this resolution. Inter-level displacement errors are compensated by isotropic interpolation algorithms. For voxels in the inter-level transition region, the accurate position and attributes are calculated using interpolation algorithms, thereby generating a 3D brain anatomical structure model.

[0086] Next, surface mesh optimization was performed on the 3D brain anatomical model. A Laplacian smoothing algorithm was used to eliminate voxel stitching noise; this algorithm smooths the mesh surface by adjusting the positions of mesh vertices. Simultaneously, an adaptive triangulation algorithm was employed to adjust the mesh density, determining it based on the local radius of curvature of the model surface. Regions with smaller local radii of curvature, such as some folds in the brain, received a higher mesh density to more accurately represent these complex structures; regions with larger local radii of curvature received a lower mesh density. After these processes, a smooth and continuous 3D visualization model was generated.

[0087] Next, the 3D visualization model is spatially registered with a standard brain anatomical atlas. First, alignment is performed based on rigid transformations of feature points. Key feature points in both the model and the atlas are identified and aligned using rigid transformations such as translation and rotation. Then, the non-rigid deformation parameters are optimized using a mutual information maximization algorithm to further adjust the model's shape and better match the standard atlas. During registration, the volume difference coefficient for each anatomical structure is calculated. For example, the volume difference coefficient is obtained by calculating the difference between the volume of a certain anatomical structure in the model and the corresponding volume in the standard atlas, and then normalizing the result.

[0088] Finally, a brain structural abnormality detection report is generated based on the volume difference coefficient, analyzing which anatomical structures exhibit abnormal volumes. The 3D visualization model is then linked to the abnormality detection report and stored in a medical imaging database for easy access and analysis by doctors and researchers.

[0089] In a standalone embodiment, after generating the annotated brain structure segmentation image, the method further includes: performing a segmentation quality assessment on the annotated brain structure segmentation image to generate a region confidence heatmap; wherein the segmentation quality assessment includes calculating the segmentation probability variance of each pixel and normalizing the variance value to a preset interval as a confidence index; extracting corresponding local image blocks from the original image sequence based on the coordinates of regions below a preset threshold in the region confidence heatmap to generate a set of image blocks to be verified; inputting the set of image blocks to be verified into a debugged segmentation error classification network to output error type labels and correction vectors; wherein the segmentation error classification network adopts a multi-scale convolutional structure and classifies boundary blurring, topological breakage, and modal confusion errors based on the cross-entropy loss function; performing pixel-level geometric deformation correction on the annotated brain structure segmentation image based on the correction vector to generate an optimized segmentation image; wherein the geometric deformation correction includes applying an elastic transformation field to the error region, the transformation field parameters being controlled by the amplitude and direction components of the correction vector.

[0090] In this embodiment, the segmentation quality of the annotated brain structure segmentation image is evaluated. First, the segmentation probability variance of each pixel is calculated. For example, for each pixel, the variance value is calculated based on its probability distribution in different segmentation results. The variance value is normalized to a preset interval as a confidence index. The preset interval is [0, 1]. Through normalization, the variance value is mapped to this interval. The larger the value, the lower the segmentation confidence of the pixel. A region confidence heatmap is generated based on the calculation results. In this heatmap, different colors represent different confidence levels. Then, based on the coordinates of regions below a preset threshold (e.g., 0.3) in the region confidence heatmap, corresponding local image blocks are extracted from the original image sequence. These image blocks may have segmentation errors and are combined into a set of image blocks to be verified.

[0091] Next, the set of image patches to be verified is input into the debugged segmentation error classification network. This network employs a multi-scale convolutional structure, extracting features from the image patches through convolutional kernels of different scales. Based on the cross-entropy loss function, it classifies boundary blurring, topological breakage, and modality confusion errors, outputting error type labels and correction vectors.

[0092] Finally, pixel-level geometric deformation correction is performed on the annotated brain structure segmentation image based on the correction vector. For regions with errors, an elastic transformation field is applied, with the transformation field parameters controlled by the amplitude and direction components of the correction vector. For example, a larger amplitude of the correction vector indicates that a larger deformation is needed to correct the error, while the direction component determines the direction of the deformation. After these operations, an optimized segmentation image is generated, improving the accuracy and quality of the segmentation.

[0093] In another implementation of this invention, acquiring a set of brain MRI image data of the target object is the beginning of the entire process. A full-range scan of a suspected brain disease patient is performed using existing MRI scanning equipment. According to the preset inter-slice spacing and scanning angle, multiple original image sequences at different scanning levels are acquired. The grayscale value of each two-dimensional grayscale image in these sequences reflects the degree of response of different brain tissues to MRI signals, thereby constituting an image data set containing rich brain information.

[0094] Multimodal feature fusion processing of raw image sequences is a crucial step in improving image information quality. Images from different modalities, such as T1-weighted images and T2-weighted images, may differ in spatial location and scale. First, cross-modal feature alignment is performed. Using image registration algorithms, one modality image is used as a reference, and other modalities are translated, rotated, and scaled to achieve spatial alignment, resulting in a spatially aligned cross-modal image sequence. Next, cross-modal feature transformation is performed, applying local texture enhancement and global structure preservation operations to the raw image sequences at each scanning level. Local filtering algorithms highlight subtle textures, such as nerve fiber orientation; global feature extraction algorithms, such as principal component analysis, preserve the overall image structure. These operations generate a transformed feature set with target dimension mapping, where each feature map incorporates complementary anatomical information from different modalities. Finally, an attention weight allocation model is used to assign weights to the different modal features in the transformed feature set, generating a modal weight distribution map. Weighted fusion is then performed based on this map to generate an enhanced image feature set that more accurately reflects the brain's anatomical structure.

[0095] A multi-level cascaded segmentation network is used to perform hierarchical feature extraction on the enhanced image feature set to obtain multi-scale anatomical structure feature maps. The enhanced image feature set is sequentially input into each level encoder of the multi-level cascaded segmentation network. Through operations such as convolution and pooling, anatomical structure features at different scales are extracted to obtain anatomical structure feature maps at each level. Then, multi-scale feature fusion processing is performed on these feature maps. The first-level feature map is upsampled, and the third-level feature map is downsampled to match their sizes. The contribution weights of feature maps at different scales are adjusted through a channel attention mechanism to generate a multi-scale anatomical structure feature map that integrates brain anatomical structure features at different scales.

[0096] Based on multi-scale anatomical feature maps, region boundary optimization is performed to generate annotated brain structure segmentation images. First, initial segmentation processing is performed on the feature maps, including inputting a fully connected decoder to generate an initial probability distribution map, thresholding, connected component analysis for noise reduction, and topological correction based on a preset brain anatomical template, resulting in an initial brain structure segmentation mask. Next, boundary confidence analysis is performed on the mask. This involves calculating boundary gradients, assessing feature uncertainty, weighted fusion to generate an initial confidence map, and performing region connectivity analysis to determine the confidence distribution map of blurred boundary regions. Based on this distribution map, iterative morphological optimization is performed on the initial mask, applying morphological closing and opening operations with different kernel sizes to regions with different confidence levels until the confidence value reaches a preset threshold, resulting in an optimized brain structure segmentation mask. Finally, this mask is overlaid and fused with the original image sequence, undergoing a series of operations including edge smoothing, transparency blending, color encoding, spatial alignment overlay, edge sharpening and anti-aliasing, adaptive contrast adjustment, and color space verification to generate an annotated brain structure segmentation image.

[0097] After generating the annotated brain structure segmentation images, further processing and analysis are performed. Cross-level 3D voxel fusion processing can be performed, mapping the annotation results of each scanning level to a 3D spatial coordinate system. Interpolation algorithms are used to compensate for inter-slice displacement errors, generating a 3D brain anatomical model. Surface mesh optimization is applied to this model to eliminate voxel stitching noise, adjust mesh density, and generate a smooth and continuous 3D visualization model. This model is then spatially registered with standard brain anatomical atlases, calculating the volume difference coefficients of each anatomical structure, generating a brain structural abnormality detection report, and storing the model and report in a medical image database. Segmentation quality assessment can also be performed, calculating the segmentation probability variance of each pixel and normalizing it as a confidence index to generate a region confidence heatmap. Based on the heatmap, a set of image blocks to be validated is extracted, input into a segmentation error classification network, and outputs error type labels and correction vectors. Pixel-level geometric deformation correction is then performed on the segmented image to generate an optimized segmented image.

[0098] Furthermore, high-resolution structural MRI images can be generated through deep learning, enabling virtual multimodal fusion of single-modal PET data. Automated segmentation algorithms based on generative MRI can reduce gray-white matter boundary image recognition errors. For image segmentation model debugging, a cascaded 3D U-Net segmentation framework is built based on real MRI and generated synthetic MRI, using real MRI segmentation labels as the gold standard for debugging data generation. A loss function minimizing multi-scale Dice loss (weight 0.6) and boundary Hausdorff distance loss (weight 0.4) is employed, combined with the Adam optimizer (initial lr = 5 × 10^-4) for debugging until convergence. Applying 3D morphological closing operations (3 × 3 × 3 spherical kernels) to the segmentation edges eliminates voxel-level jagged artifacts. Fusing SUVR heatmaps with synthetic MRI anatomical structures can also generate interactive Aβ deposition distribution labeled images.

[0099] In summary, this invention first acquires a brain MRI image dataset containing original image sequences from multiple scanning levels. Then, it performs multimodal feature fusion processing to effectively integrate features from various modalities, thereby generating an enhanced image feature set. This allows the images to contain more comprehensive and in-depth information, while also uncovering subtle features. Furthermore, a multi-layer cascaded segmentation network is used to extract hierarchical features from the enhanced image feature set, capturing brain anatomical structural features at different scales and levels, resulting in a multi-scale anatomical structure feature map. Finally, based on the multi-scale anatomical structure feature map, region boundary optimization processing is performed to generate an annotated brain structure segmentation image. This accurately defines the boundaries of various brain structures, making the segmented image more accurate and clear. Thus, accurate and clear image segmentation and annotation of brain MRI images can be achieved.

[0100] It is worth noting that the core of this invention is to improve the efficiency and accuracy of medical image analysis through computer image processing technology, and does not directly involve the implementation of disease diagnosis or treatment methods. This invention focuses on establishing a complete brain MRI image processing workflow, specifically including quantifiable technical modules such as multimodal feature fusion, hierarchical feature extraction, and boundary optimization algorithms. Its innovation lies in the mathematical modeling of image data and the computer-automated processing methods. For example, it solves the spatial registration problem of different image sequences through cross-modal feature alignment algorithms, optimizes the multi-scale feature fusion process using channel attention mechanisms, and improves segmentation boundary accuracy by combining morphological operations and confidence analysis. These techniques are essentially technological improvements in the fields of computer vision and image processing, with clear technical implementation paths and industrial application scenarios.

[0101] Furthermore, the effectiveness of this invention relies entirely on the automated processing capabilities of the computer system, without depending on the subjective judgment of doctors or medical diagnostic procedures. Although the object of processing is brain medical images, the innovation of the technical solution lies in the mathematical transformation and feature optimization of the image data. For example, image spatial alignment is achieved through coordinate normalization, edge sharpness is improved by using anisotropic diffusion filtering, and segmentation errors are corrected based on elastic transformation fields. The improvement direction of these technical means always revolves around technical indicators such as image processing accuracy, computational efficiency, and feature representation ability. The result is an optimized segmented image or a three-dimensional visualization model, which is a computer-generated data product rather than a medical diagnostic conclusion. Even if "suspected patients" are mentioned in the example, the technical solution itself is not limited to disease diagnosis scenarios. Its method is also applicable to brain image analysis of healthy individuals, possessing universal technical characteristics.

[0102] Furthermore, the technical contributions of this invention lie in the enhancement and optimization of the functions of medical image processing equipment, belonging to the field of medical equipment technology. For example, it improves feature extraction efficiency through a multi-layer cascaded segmentation network architecture, enhances the accuracy of anatomical structure reconstruction using a three-dimensional voxel fusion algorithm, and improves the visualization effect of human-computer interaction using pseudo-color annotation layer generation technology. These technical achievements can be directly applied to the development and performance improvement of medical image analysis systems. The implementation of the entire technical solution does not involve applying medical measures or interventions to the human body. Its core value lies in providing higher-quality image analysis tools for medical diagnosis, rather than replacing or influencing the doctor's diagnostic decision-making process. Therefore, this invention meets the requirements of the object attribute of the technical solution, belongs to the invention of solving specific technical problems using technical means, and is fundamentally different from disease diagnosis and treatment methods.

[0103] It is worth noting that in practical applications, cross-modal feature alignment can be performed based on existing multimodal registration methods in medical image processing technology (such as open-source tools like ANTs and Elastix), and inter-slice displacement compensation can be achieved through the B-spline elastic registration algorithm.

[0104] For local texture enhancement in cross-modal feature transformation, a direction-controllable Gabor filter bank can be used for multi-scale texture feature extraction, combined with non-local mean filtering to achieve global structure preservation.

[0105] The attention weight allocation model can build a channel attention mechanism based on the Squeeze-and-Exception network architecture, and capture global statistical information of multimodal features through adaptive pooling layers.

[0106] The multi-layer cascaded segmentation network can be built with reference to the 3D U-Net framework. The residual connection module is integrated in the encoding path to prevent gradient vanishing, and the transposed convolution is used in the decoding path to restore the feature map resolution.

[0107] Adaptive kernel size calculation in morphological optimization operations can be achieved by establishing a linear mapping function between confidence values ​​and structuring element sizes, where the upper limit of the kernel size needs to be physically constrained based on the inter-layer resolution of the image.

[0108] The three-dimensional voxel fusion process requires the establishment of a spatial transformation matrix based on the pixel spacing and layer thickness parameters in the DICOM standard, and the use of a trilinear interpolation algorithm to achieve cross-layer data fusion.

[0109] Surface mesh optimization can be achieved using Laplacian smoothing based on the MeshLab open-source toolkit, and the density of triangular patches can be dynamically adjusted by combining local curvature radius calculation.

[0110] In segmentation quality assessment, probability variance calculation can be performed using the Monte Carlo Dropout method to obtain pixel-level predicted distributions, and a sliding window strategy can be used to extract image patches to be validated. The segmentation error classification network can construct cascaded Inception modules to achieve multi-scale feature capture, and the Focal Loss function can be used to alleviate the class imbalance problem.

[0111] In addition, operations involving spatial transformations can be performed using DICOM coordinate system conversion to ensure dimensional uniformity, and the kernel size for morphological operations needs to be converted to millimeters based on the physical resolution of the image to avoid scale confusion.

[0112] Based on the above techniques, those skilled in the art can clearly and completely implement the embodiments of the present invention.

[0113] See Figure 2 As shown in the figure, this is a schematic diagram of the basic structure of an image annotation system 200 provided in an embodiment of the present invention. The image annotation system 200 includes:

[0114] Processor 201;

[0115] Storage device 202, on which computer program 2020 is stored;

[0116] When the computer program 2020 is executed by the processor 201, the processor 201 implements any of the image annotation methods described above for segmenting brain MRI images.

[0117] Based on the above, a readable storage medium is provided, on which a program or instructions are stored, and when the program or instructions are executed by a processor, the steps of the above method are implemented.

[0118] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems or apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and relevant parts can be referred to the method section.

Claims

1. An image labeling method applied to brain MRI image segmentation, characterized in that, The method comprises the following steps: obtain a brain MRI image data set of a target object, the brain MRI image data set comprising a plurality of original image sequences of scanning levels; perform multi-modal feature fusion processing on the original image sequences to generate an enhanced image feature set; call a multi-layer cascaded segmentation network to perform hierarchical feature extraction on the enhanced image feature set to obtain multi-scale anatomical structure feature maps; perform region boundary optimization processing based on the multi-scale anatomical structure feature maps to generate a brain structure segmentation image with completed labeling: perform initial segmentation processing on the multi-scale anatomical structure feature maps to generate an initial brain structure segmentation mask; perform boundary confidence analysis on the initial brain structure segmentation mask to determine a confidence distribution map of a boundary fuzzy region; perform an iterative morphological optimization operation on the initial brain structure segmentation mask based on the confidence distribution map to generate an optimized brain structure segmentation mask; wherein the iterative morphological optimization operation comprises performing a dilation-erosion sequence processing on the boundary fuzzy region, and adjusting the size of the dilation kernel based on the confidence distribution map; and performing superimposed fusion processing on the optimized brain structure segmentation mask and the original image sequences to generate the brain structure segmentation image with completed labeling; the superimposed fusion processing of the optimized brain structure segmentation mask and the original image sequences to generate the brain structure segmentation image with completed labeling comprises: perform edge smoothing processing on the optimized brain structure segmentation mask to generate a smoothed segmentation contour; perform transparency mixing processing on the smoothed segmentation contour and the gray value of the original image sequences to generate a preliminary labeling image; The color coding processing is performed on the boundary regions of different anatomical structures in the preliminary labeled image to generate a pseudo-color labeling layer with distinguishability: unique color identifiers are assigned to different anatomical regions according to a preset brain anatomical structure classification rule; a category matching process is performed on each connected region in the optimized brain structure segmentation mask to determine the corresponding target color identifier; a gradient color filling process is performed on the boundary pixels of the connected region based on the target color identifier to generate the pseudo-color labeling layer; wherein the color saturation of the gradient color filling process is positively correlated with the distance of the boundary pixels to the center of the segmentation contour; the pseudo-color labeling layer is spatially aligned and superimposed with the original image sequence to generate the brain structure segmentation image after labeling is completed: a normalized coordinate conversion process is performed on the pixel coordinates of the pseudo-color labeling layer to generate a standardized coordinate mapping table aligned with the physical coordinate system of the original image sequence; wherein the normalized coordinate conversion process includes mapping the pixel positions of the pseudo-color labeling layer to the inter-scan layer resolution ratio space of the original image sequence, and compensating for the inter-layer displacement error by a bilinear interpolation algorithm; a spatial resampling process is performed on the color channels of the pseudo-color labeling layer according to the standardized coordinate mapping table to generate a resampled pseudo-color layer that completely matches the pixel space distribution of the original image sequence; wherein the spatial resampling process uses an isotropic interpolation kernel function, and the kernel width parameter of the isotropic interpolation kernel function is dimensionally unified with the intra-layer pixel spacing of the original image sequence; the resampled pseudo-color layer is input into a transparency mixing channel to perform a multi-channel pixel-level superposition operation to generate a mixed labeling image; wherein the superposition operation includes performing a linear brightness preservation transformation on the grayscale values of the original image sequence, and performing dynamic range compression on the RGB color components of the resampled pseudo-color layer, so that the pixel values after superposition do not exceed the preset maximum display dynamic range; an edge sharpening and anti-aliasing process is performed on the mixed labeling image to generate a boundary-enhanced intermediate labeling image; wherein the edge sharpening process uses an anisotropic diffusion filtering algorithm, and the gradient threshold parameter of the anisotropic diffusion filtering algorithm matches the grayscale gradient distribution statistics of the original image sequence; an adaptive contrast adjustment process is performed based on the boundary intensity distribution of the intermediate labeling image to generate a final displayed brain structure segmentation image; wherein the adaptive contrast adjustment process includes performing a histogram equalization operation on low-contrast regions, and simultaneously performing a non-linear gamma correction on high-gradient regions, and the correction coefficient of the non-linear gamma correction is negatively correlated with the local gradient amplitude; a color space verification process is performed on the final displayed brain structure segmentation image to ensure that the chroma channels of the pseudo-color labeling layer and the luminance channels of the original image sequence do not overlap in the display color gamut range.The color space verification process includes converting the RGB color space into a YUV space, detecting an orthogonality index of a luminance component and a chroma component after the conversion, and if an overlapping region is detected, remapping chroma coordinates to a preset safe palette interval.

2. The method of claim 1, wherein, the multi-modal feature fusion processing on the original image sequences to generate an enhanced image feature set comprises: perform cross-modal feature alignment processing on each scanning level in the original image sequences to obtain a spatially aligned cross-modal image sequence; perform cross-modal feature conversion processing on the spatially aligned cross-modal image sequence to generate a converted feature set with target dimensional mapping; wherein the cross-modal feature conversion processing comprises performing local texture enhancement and global structure preservation operations on the original image sequence of each scanning level, so that each feature map in the converted feature set contains complementary anatomical information across modalities; call a completed attention weight allocation model to allocate weights to different modal features in the converted feature set to generate a modal weight distribution map; perform weighted fusion processing on the converted feature set according to the modal weight distribution map to generate the enhanced image feature set.

3. The method of claim 2, wherein, the calling of the multi-layer cascaded segmentation network to perform hierarchical feature extraction on the enhanced image feature set to obtain multi-scale anatomical structure feature maps comprises: input the enhanced image feature set into a first encoder of the multi-layer cascaded segmentation network to extract a first-level anatomical structure feature map; input the first-level anatomical structure feature map into a second encoder of the multi-layer cascaded segmentation network to extract a second-level anatomical structure feature map; inputting the second-level anatomical structure feature map into a third encoder of the multi-level cascaded segmentation network to extract a third-level anatomical structure feature map; performing multi-scale feature fusion processing on the first-level anatomical structure feature map, the second-level anatomical structure feature map and the third-level anatomical structure feature map to generate the multi-scale anatomical structure feature map; wherein the multi-scale feature fusion processing comprises upsampling operation on the first-level anatomical structure feature map, downsampling operation on the third-level anatomical structure feature map, and adjusting the contribution weight of different scale feature maps through channel attention mechanism.

4. The method of claim 1, wherein, The initial segmentation processing on the multi-scale anatomical structure feature map to generate an initial brain structure segmentation mask comprises: inputting the multi-scale anatomical structure feature map into a fully connected decoder of the multi-level cascaded segmentation network to generate an initial probability distribution map; performing threshold segmentation processing on the initial probability distribution map to obtain a binary segmentation result; performing connected domain analysis processing on the binary segmentation result to remove isolated noise regions to obtain a denoised binary segmentation result; performing topological correction processing on the denoised binary segmentation result based on a preset brain anatomical structure template to generate the initial brain structure segmentation mask.

5. The method of claim 1, wherein, The boundary confidence analysis on the initial brain structure segmentation mask to determine a confidence distribution map of the boundary fuzzy region comprises: performing boundary gradient calculation on the initial brain structure segmentation mask to generate a boundary gradient intensity map; performing feature uncertainty evaluation on the multi-scale anatomical structure feature map to generate a feature uncertainty distribution map; performing weighted fusion processing on the boundary gradient intensity map and the feature uncertainty distribution map to generate an initial confidence map; performing region connectivity analysis on the initial confidence map to filter out continuous regions below a preset confidence threshold as the boundary fuzzy region; generating the confidence distribution map based on the spatial distribution of the boundary fuzzy region.

6. The method of claim 1, wherein, The iterative morphological optimization operation on the initial brain structure segmentation mask based on the confidence distribution map to generate an optimized brain structure segmentation mask comprises: determining the corresponding morphological operation intensity parameter according to the confidence value of each region in the confidence distribution map; performing morphological closing operation processing with a fixed kernel size on the target confidence region in the initial brain structure segmentation mask; performing morphological opening operation processing with an adaptive kernel size on the boundary fuzzy region, wherein the kernel size is negatively correlated with the confidence value; repeating the morphological closing operation processing and the morphological opening operation processing until the confidence value of all regions in the confidence distribution map reaches a preset optimization threshold; taking the morphological processing result obtained in the last iteration as the optimized brain structure segmentation mask.

7. An image annotation system characterized by comprising: comprise: a processor; a storage device having a computer program stored thereon, when the computer program is executed by the processor, the processor implements the image labeling method for brain MRI image segmentation according to any one of claims 1-6.

Images