Search

CN-121999019-A - Cell image registration method and device and computer equipment

CN121999019ACN 121999019 ACN121999019 ACN 121999019ACN-121999019-A

Abstract

The embodiment of the application is suitable for the fields of biomedical technology and image processing technology, and provides a cell image registration method, a device and computer equipment, wherein the method comprises the steps of obtaining feature images corresponding to two cell images to be registered, wherein the feature images comprise a foreground probability image and a cell flow field image; generating two groups of cell instance point sets based on the foreground probability map and the cell flow field map, performing feature matching on the two groups of cell instance point sets to obtain target matching combinations, determining first affine transformation of the two cell images based on the target matching combinations, and registering the two cell images according to the cell flow field map and the first affine transformation. By adopting the method, the cell images can be registered by combining the cell flow field diagram, and the accuracy of image registration is improved.

Inventors

  • CHEN YUCHAO
  • ZHANG FENG

Assignees

  • 利德健康科技(广州)有限公司

Dates

Publication Date
20260508
Application Date
20251225

Claims (10)

  1. 1. A method of cell image registration, comprising: Acquiring feature images corresponding to two cell images to be registered, wherein the feature images comprise a foreground probability image and a cell flow field image; Generating two sets of cell instance point sets based on the foreground probability map and the cell flow field map; performing feature matching on the two cell instance point sets to obtain a target matching combination; Determining a first affine transformation of two of the cell images based on the target matching combination; and registering the two cell images according to the cell flow field diagram and the first affine transformation.
  2. 2. The method of claim 1, wherein the generating two sets of cell instance points based on the foreground probability map and the cell flow field map comprises: initializing a coordinate map, wherein the coordinate map is used for representing the coordinate position of each coordinate point flow direction corresponding to an input image, and the value in the coordinate map represents the coordinate value of each coordinate point; Updating the coordinate map by superposing the cell flow field map in the coordinate map to obtain a target coordinate map, wherein any cell point in the target coordinate map flows to the center point of the cell to which the cell belongs; And processing a plurality of cell points in the target coordinate map according to the foreground probability map to obtain two cell instance point sets which are respectively in one-to-one correspondence with the two cell images.
  3. 3. The method according to claim 2, wherein said processing the plurality of cell points in the target coordinate map according to the foreground probability map to obtain two sets of cell instance points corresponding to the two cell images one to one respectively includes: for any target coordinate map, eliminating cell center points corresponding to the center points with probability values lower than a first threshold in the foreground probability map in the target coordinate map; Merging a plurality of cell points into a plurality of point sets according to a plurality of cell center points remained in the target coordinate map; And eliminating a plurality of point sets containing the cell points with the number smaller than a second threshold value to obtain a cell instance point set.
  4. 4. A method according to any one of claims 1 to 3, wherein said feature matching of two sets of said cell instance points results in a target matching combination comprising: Respectively determining shape characteristics of two groups of cell instance point sets, and performing characteristic matching on the shape characteristics to obtain a plurality of groups of matching combinations, wherein any matching combination has corresponding similarity; And determining target matching combinations from a plurality of groups of matching combinations according to the similarity.
  5. 5. The method of claim 4, wherein said separately determining shape characteristics of two sets of said cell instance points comprises: and respectively calculating HU invariant moments of the two cell instance point sets to obtain shape features corresponding to the two cell instance point sets.
  6. 6. The method according to any one of claims 1 to 3 or 5, wherein said determining a first affine transformation of two of said cell images based on said target matching combination comprises: calculating the displacement of each corresponding cell center point in the target matching combination; determining a first affine transformation of two cell images according to the displacement amounts of the cell center points.
  7. 7. The method of claim 6, wherein said registering two of said cell images from said cell flow field map and said first affine transformation comprises: Aligning the cell flow field diagrams corresponding to the two cell images according to the first affine transformation; calculating a second affine transformation that satisfies a direction-based flow field consistency constraint; And registering the two cell images according to the second affine transformation.
  8. 8. The method of claim 7, wherein the calculating a second affine transformation comprises: constructing a flow field consistency loss function based on the direction; An optimal second affine transformation is calculated that minimizes a loss value of the flow field consistency loss function based on direction.
  9. 9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the computer device is caused to implement the method of any one of claims 1 to 8 when the processor executes the computer program.
  10. 10. A computer program product comprising a computer program which, when run, causes the method of any one of claims 1 to 8 to be performed.

Description

Cell image registration method and device and computer equipment Technical Field The embodiment of the application belongs to the technical field of biomedicine and image processing, and particularly relates to a cell image registration method, device and computer equipment. Background Accurate registration of cell images is a key technique in biomedical image analysis, with the goal of spatially aligning two or more cell images acquired at different times, different viewing angles, or under different conditions. The technology has important value in the fields of cytokinetic research, drug response tracking, disease diagnosis, super-resolution image reconstruction and the like. The existing cell image registration technology mainly comprises two major types of registration methods based on characteristics and registration methods based on image gray scale. In cell images, there is significant non-rigid deformation and appearance change in the image content due to vital movements such as cell division, movement, apoptosis, etc., and changes in staining, imaging conditions. The registration method based on the traditional features is often difficult to extract stable and repeatable feature points in the environment, so that the mismatching rate is high, and registration failure is easy to occur. In addition, because areas such as cytoplasm and the like generally lack abundant texture information, the image gray level-based registration method has weak texture or uniform characteristics, is easy to sink into local optimization in the areas, causes stagnation of the optimization process, cannot realize accurate alignment, and also can influence the final registration effect. In recent years, with the development of deep learning technology, a registration method based on deep learning has become a research hotspot. Such methods may use convolutional neural networks (Convolutional Neural Network, CNN) to directly learn the mapping from image pairs to spatial transformations, helping to improve registration efficiency. However, according to the registration method based on deep learning, the network model is trained on a specific data set, and the registration performance of new data with different imaging modes, cell types and fluorescent markers is obviously reduced. The generalization ability of the model is strongly dependent on the diversity and representativeness of training data, which severely limits the practical application of the method in wide biomedical research. Disclosure of Invention In view of this, the embodiments of the present application provide a method, an apparatus, and a computer device for registering cell images, which can register cell images in combination with a cell flow field map, so as to help to improve accuracy of image registration, expand generalization capability of the registration method, and be applicable to registration of various cell images. A first aspect of an embodiment of the present application provides a cell image registration method, including: Acquiring feature images corresponding to two cell images to be registered, wherein the feature images comprise a foreground probability image and a cell flow field image; Generating two sets of cell instance point sets based on the foreground probability map and the cell flow field map; performing feature matching on the two cell instance point sets to obtain a target matching combination; Determining a first affine transformation of two of the cell images based on the target matching combination; and registering the two cell images according to the cell flow field diagram and the first affine transformation. Optionally, the generating two sets of cell instance points based on the foreground probability map and the cell flow field map includes: initializing a coordinate map, wherein the coordinate map is used for representing the coordinate position of each coordinate point flow direction corresponding to an input image, and the value in the coordinate map represents the coordinate value of each coordinate point; Updating the coordinate map by superposing the cell flow field map in the coordinate map to obtain a target coordinate map, wherein any cell point in the target coordinate map flows to the center point of the cell to which the cell belongs; And processing a plurality of cell points in the target coordinate map according to the foreground probability map to obtain two cell instance point sets which are respectively in one-to-one correspondence with the two cell images. Optionally, the processing the plurality of cell points in the target coordinate map according to the foreground probability map to obtain two cell instance point sets corresponding to two cell images one by one respectively includes: for any target coordinate map, eliminating cell center points corresponding to the center points with probability values lower than a first threshold in the foreground probability map in the target coordinate map; Me