CN-121982298-A - Blood vessel image segmentation method and related equipment

CN121982298ACN 121982298 ACN121982298 ACN 121982298ACN-121982298-A

Abstract

The application discloses a blood vessel image segmentation method and related equipment, the method comprises the steps of performing multidirectional feature extraction processing on acquired blood vessel image data to be segmented through the scheme to obtain a plurality of different levels of direction sensing feature data serving as multi-scale direction sensing feature data, performing feature decoding on the obtained multi-scale direction sensing feature data to obtain a plurality of different levels of space features serving as space feature data, performing feature fusion on the obtained multi-scale direction sensing feature data and the space feature data to obtain different levels of fusion feature data serving as fusion data, performing up-sampling reconstruction processing on the obtained fusion data to obtain reconstruction data, and performing convolution activation processing on the obtained reconstruction data to obtain a blood vessel segmentation result. The embodiment of the application can improve the segmentation accuracy and continuity of the blood vessel. The application can be widely applied to the technical field of image data processing.

Inventors

XIAO CHI
WU FENGRUI
PAN JIAYU
HUANG GANGHUA
JIANG YAN
LU ZHIKANG
LI ANAN
LUO QINGMING

Assignees

海南大学
海南大学三亚研究院

Dates

Publication Date: 20260505
Application Date: 20251216

Claims (10)

1. A method of segmenting a blood vessel image, the method comprising: acquiring blood vessel image data to be segmented, performing multi-directional feature extraction on the blood vessel image data to be segmented, and determining multi-scale direction sensing feature data, wherein the multi-scale direction sensing feature data comprises a plurality of different levels of direction sensing feature data; performing feature decoding on the multi-scale direction perception feature data to determine spatial feature data, wherein the spatial feature data comprises a plurality of spatial features of different levels; performing feature fusion according to the multi-scale direction sensing feature data and the space feature data to determine fusion data, wherein the fusion data comprises a plurality of fusion feature data of different levels; and carrying out up-sampling reconstruction on the fusion data, determining reconstruction data, carrying out convolution activation processing on the reconstruction data, and determining a blood vessel segmentation result.
2. The method according to claim 1, wherein the multi-directional feature extraction is performed on the blood vessel image data to be segmented to determine multi-scale directional perception feature data, specifically comprising: Three-dimensional direction feature extraction is carried out on the blood vessel image data to be segmented, direction feature data are determined, and pooling treatment is carried out on the direction feature data to obtain first feature data; and carrying out three-dimensional residual convolution processing on the first characteristic data, determining semantic characteristic data, carrying out characteristic combination on the semantic characteristic data and the first characteristic data, and determining the multi-scale direction perception characteristic data.
3. The method according to claim 2, wherein the three-dimensional direction feature extraction is performed on the blood vessel image data to be segmented, and the direction feature data is determined, specifically including: Taking the blood vessel image data to be segmented as input data, and carrying out multi-branch convolution processing on the input data to determine a plurality of branch output data, wherein the branch output data comprises directional characteristic data in the X direction, directional characteristic data in the Y direction, directional characteristic data in the Z direction and global space context information; fusing a plurality of branch output data according to channel dimension, determining branch fusion data, performing nonlinear activation and channel normalization processing on the branch fusion data, and determining first output data; and carrying out residual connection according to the first output data and the input data, and determining the direction characteristic data.
4. The method according to claim 2, wherein the performing three-dimensional residual convolution processing on the first feature data to determine semantic feature data specifically includes: Extracting the first characteristic data, determining local spatial characteristic data, carrying out semantic enhancement on the local spatial characteristic data, and determining trunk path data; performing feature mapping on the first feature data, determining mapping data, performing channel alignment on the mapping data, and determining projection path data; And carrying out characteristic connection on the trunk path data and the projection path data, and determining the semantic characteristic data.
5. The method according to claim 1, wherein the upsampling decoding of the multi-scale directional perceptual feature data to determine spatial feature data comprises: Taking the multi-scale direction sensing characteristic data as second input data, carrying out three-dimensional direction characteristic extraction on the second input data, and determining second characteristic data; or, carrying out three-dimensional residual convolution processing on the second input data to determine third characteristic data; and carrying out up-sampling processing on the second characteristic data or the third characteristic data to determine the spatial characteristic data.
6. The method according to claim 1, wherein the feature fusion is performed according to the multi-scale directional perception feature data and the spatial feature data, and determining fusion data specifically includes: Performing feature stitching according to the multi-scale direction perception feature data and the space feature data, determining stitching feature data, performing channel weighting processing on the stitching feature data according to normalized weights, and determining a weighted feature map; Carrying out multi-branch convolution on the weighted feature map to determine a plurality of branch convolution data, and carrying out data splicing on the plurality of branch convolution data to determine spliced data; Performing three-dimensional convolution on the spliced data, determining convolution characteristic data, performing activation processing on the convolution characteristic data, and determining characteristic output data; and carrying out residual connection according to preset attention weight, the sub-feature map and the weighted feature map, and determining the fusion data.
7. A vascular image segmentation system, the system comprising: The device comprises a feature extraction module, a feature extraction module and a feature extraction module, wherein the feature extraction module is used for acquiring blood vessel image data to be segmented, performing multi-direction feature extraction on the blood vessel image data to be segmented, and determining multi-scale direction sensing feature data, wherein the multi-scale direction sensing feature data comprises a plurality of different levels of direction sensing feature data; The feature decoding module is used for performing feature decoding on the multi-scale direction sensing feature data to determine spatial feature data, wherein the spatial feature data comprises a plurality of spatial features of different levels; the feature fusion module is used for carrying out feature fusion according to the multi-scale direction sensing feature data and the space feature data to determine fusion data, wherein the fusion data comprises a plurality of different levels of fusion feature data; and the segmentation output module is used for carrying out up-sampling reconstruction on the fusion data, determining reconstruction data, carrying out convolution activation processing on the reconstruction data and determining a blood vessel segmentation result.
8. An electronic device, comprising: At least one processor; At least one memory for storing at least one program; the at least one program, when executed by the at least one processor, causes the at least one processor to implement the method of any one of claims 1 to 6.
9. A computer readable storage medium storing a computer program, characterized in that the computer program, when executed by a processor, implements the method of any one of claims 1 to 6.
10. A computer program product comprising a computer program, characterized in that the computer program, when executed by a processor, implements the method of any one of claims 1 to 6.

Description

Blood vessel image segmentation method and related equipment Technical Field The application relates to the technical field of image data processing, in particular to a blood vessel image segmentation method and related equipment. Background MRA (magnetic resonance vascular imaging) is a noninvasive vascular imaging technology, provides important support for diagnosis and treatment of vascular diseases, has obvious anisotropy due to the fact that intracranial blood vessels have complex three-dimensional tree structures, and therefore blood vessels are difficult to identify from an MRA image for segmentation, and meanwhile, image signals are easy to lose due to hemodynamic effects, so that accuracy and continuity of blood vessel segmentation are reduced. Disclosure of Invention The embodiment of the application mainly aims to provide a blood vessel image segmentation method and related equipment, which can improve the blood vessel segmentation precision and continuity. To achieve the above object, an aspect of an embodiment of the present application provides a blood vessel image segmentation method, including: acquiring blood vessel image data to be segmented, performing multi-directional feature extraction on the blood vessel image data to be segmented, and determining multi-scale direction sensing feature data, wherein the multi-scale direction sensing feature data comprises a plurality of different levels of direction sensing feature data; performing feature decoding on the multi-scale direction perception feature data to determine spatial feature data, wherein the spatial feature data comprises a plurality of spatial features of different levels; performing feature fusion according to the multi-scale direction sensing feature data and the space feature data to determine fusion data, wherein the fusion data comprises a plurality of fusion feature data of different levels; and carrying out up-sampling reconstruction on the fusion data, determining reconstruction data, carrying out convolution activation processing on the reconstruction data, and determining a blood vessel segmentation result. In some embodiments, the multi-directional feature extraction is performed on the to-be-segmented blood vessel image data to determine multi-scale directional perception feature data, which specifically includes: Three-dimensional direction feature extraction is carried out on the blood vessel image data to be segmented, direction feature data are determined, and pooling treatment is carried out on the direction feature data to obtain first feature data; and carrying out three-dimensional residual convolution processing on the first characteristic data, determining semantic characteristic data, carrying out characteristic combination on the semantic characteristic data and the first characteristic data, and determining the multi-scale direction perception characteristic data. In some embodiments, the three-dimensional direction feature extraction is performed on the blood vessel image data to be segmented, and the determining direction feature data specifically includes: Taking the blood vessel image data to be segmented as input data, and carrying out multi-branch convolution processing on the input data to determine a plurality of branch output data, wherein the branch output data comprises directional characteristic data in the X direction, directional characteristic data in the Y direction, directional characteristic data in the Z direction and global space context information; fusing a plurality of branch output data according to channel dimension, determining branch fusion data, performing nonlinear activation and channel normalization processing on the branch fusion data, and determining first output data; and carrying out residual connection according to the first output data and the input data, and determining the direction characteristic data. In some embodiments, the performing three-dimensional residual convolution processing on the first feature data to determine semantic feature data specifically includes: Extracting the first characteristic data, determining local spatial characteristic data, carrying out semantic enhancement on the local spatial characteristic data, and determining trunk path data; performing feature mapping on the first feature data, determining mapping data, performing channel alignment on the mapping data, and determining projection path data; And carrying out characteristic connection on the trunk path data and the projection path data, and determining the semantic characteristic data. In some embodiments, the upsampling decoding the multi-scale directional perceptual feature data to determine spatial feature data specifically includes: Taking the multi-scale direction sensing characteristic data as second input data, carrying out three-dimensional direction characteristic extraction on the second input data, and determining second characteristic data; or, carrying out three-dimen