Cell image registration method, device, equipment, storage medium and program product

Abstract

The application discloses a cell image registration method and device, electronic equipment, a storage medium and a computer program product. The method comprises the following steps: image acquisition is performed on fluorescent microspheres for replacing cell samples to obtain a bright field image and a fluorescence image; the bright field image and the fluorescence image are divided to obtain a plurality of regional image pairs; each regional image pair comprises one bright field regional image and one fluorescence regional image corresponding in position; centroid detection is performed on the regional image pairs to obtain a bright field centroid point set and a fluorescence centroid point set of each regional image pair; centroid local matching is performed on the bright field centroid point set and the fluorescence centroid point set to obtain a local offset of each regional image pair; and fusion processing is performed on the local offsets of all the regional image pairs to obtain a global registration offset. The method can solve the problem that matching errors or failures occur in the registration process, thereby affecting the accuracy and stability of the registration result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to a cell image registration method and device, electronic equipment, readable storage medium and computer program product. BACKGROUND

[0002] In the research of cell biology and drug screening, it is often necessary to accurately register bright field images and fluorescence images to fuse cell structure information and molecular marker information, so as to realize comprehensive analysis of cell morphology, positioning and metabolic activity. The registration method widely used at present mainly relies on feature point-based registration method, which extracts significant features such as corner points and edges in the image, establishes the corresponding relationship between the feature points, and calculates the global transformation matrix to realize registration accordingly. However, due to the significant difference in imaging mechanism between bright field images and fluorescence images, their texture features and structural expressions often do not have obvious correspondence, which makes it difficult to extract a sufficient number of matching reliable feature points in the multi-modal scene. This limitation makes the existing method prone to matching errors or failure in the registration process, thereby affecting the accuracy and stability of the registration result. SUMMARY

[0003] In view of the above problems, the present application provides a cell image registration method, device, electronic equipment, readable storage medium and computer program product, which can solve the problem that matching errors or failure occur in the registration process, thereby affecting the accuracy and stability of the registration result.

[0004] In a first aspect, the present application provides a cell image registration method, comprising:

[0005] Image acquisition is performed on the fluorescent microspheres of the substitute cell sample to obtain bright field images and fluorescence images;

[0006] The bright field images and the fluorescence images are divided to obtain a plurality of region image pairs; wherein each region image pair comprises a bright field region image and a fluorescence region image corresponding in position;

[0007] Centroid detection is performed on the region image pairs to obtain a bright field centroid point set and a fluorescence centroid point set of each region image pair;

[0008] Centroid local matching is performed on the bright field centroid point set and the fluorescence centroid point set to obtain a local offset of each region image pair;

[0009] The local offsets of all region image pairs are fused to obtain a global registration offset.

[0010] In the technical solution, the method can effectively solve the problems of difficult feature point extraction and unreliable matching caused by the difference in imaging mechanism between the bright field image and the fluorescence image. Through the centroid detection based on the fluorescence microspheres as a unified reference and the local-global combined offset calculation logic, the accuracy and stability of the cell image registration result are significantly improved, thereby providing reliable data support for the accurate fusion analysis of cell structure information and molecular marker information in cell biology research and drug screening.

[0011] In some embodiments, the plurality of bright field region images in the bright field image have consistent sizes;

[0012] The plurality of fluorescence region images in the fluorescence image have consistent sizes.

[0013] In the technical solution, the method can ensure the uniformity of the region division rules of the two types of images, provide a standardized comparison basis for subsequent centroid detection and local offset calculation of corresponding region image pairs, and avoid the imbalance of local registration benchmarks caused by the difference in region size.

[0014] In some embodiments, the centroid detection on the region image pairs to obtain the bright field centroid point set and the fluorescence centroid point set of each region image pair comprises:

[0015] Performing gray-scale processing on each region image pair to obtain a first processed region image pair;

[0016] Performing binaryzation processing on each first processed region image pair to obtain a second processed region image pair;

[0017] Performing microsphere contour detection on each second processed region image pair to obtain a microsphere contour image pair including a microsphere external contour; wherein the microsphere contour image pair includes one bright field microsphere contour region image and one fluorescence microsphere contour region image corresponding in position;

[0018] Performing centroid detection on the microsphere contour image pair to obtain the bright field centroid point set and the fluorescence centroid point set of each region image pair.

[0019] In the technical solution, the method can systematically eliminate redundant gray scale information and background noise interference in the bright field and fluorescence images through the progressive image processing flow of "gray-scale processing → binaryzation processing → microsphere contour detection → centroid detection", and accurately separate the effective contour region of the fluorescence microspheres. The gray-scale processing and binaryzation processing can simplify the image information dimension and highlight the light-dark difference between the microspheres and the background. The targeted microsphere contour detection can also accurately lock the target region and avoid the interference of the cell structure or other impurities on the centroid identification, thereby ensuring that the bright field centroid point set and the fluorescence centroid point set extracted from the contour image accurately correspond to the actual position of the fluorescence microspheres.

[0020] In some embodiments, the center-of-mass detection on the microsphere contour image pair obtains a bright-field center-of-mass point set and a fluorescent center-of-mass point set of each of the region image pairs, including:

[0021] calculating a microsphere contour geometric feature of each of the region image pairs; wherein the microsphere contour geometric feature includes a first geometric feature of an external contour of each microsphere in the bright-field microsphere contour region image, and a second geometric feature of an external contour of each microsphere in the fluorescent microsphere contour region image, the first geometric feature and the second geometric feature each at least including a zeroth moment and a first moment of the external contour of the microsphere;

[0022] According to the first geometric feature, the center-of-mass coordinates of the external contour of each microsphere in the bright-field microsphere contour region image are calculated to obtain a bright-field center-of-mass point set of the region image pair;

[0023] According to the second geometric feature, the center-of-mass coordinates of the external contour of each microsphere in the fluorescent microsphere contour region image are calculated to obtain a fluorescent center-of-mass point set of the region image pair.

[0024] In the above technical solution, the method can rely on the mathematical properties of the moment method to convert the spatial form information of the microsphere contour into accurate numerical parameters, avoid the dependence on texture and edge in artificial subjective judgment or traditional feature extraction, and fundamentally eliminate the interference of the difference between the bright-field and fluorescent image modalities on the center-of-mass positioning. At the same time, the combination calculation of the zeroth moment (reflecting the contour area / gray sum) and the first moment (reflecting the contour gray distribution barycenter) can also accurately anchor the geometric center of the microsphere contour, so as to ensure that the coordinate values of the bright-field center-of-mass point set and the fluorescent center-of-mass point set strictly correspond to the actual spatial position of the microsphere.

[0025] In some embodiments, the center-of-mass local matching on the bright-field center-of-mass point set and the fluorescent center-of-mass point set obtains a local offset of each of the region image pairs, including:

[0026] obtaining a preset matching algorithm parameter; wherein the matching algorithm parameter at least includes a maximum number of iterations and an inlier judgment threshold;

[0027] According to the matching algorithm parameter, the bright-field center-of-mass point set and the fluorescent center-of-mass point set of each of the region image pairs are subjected to center-of-mass local matching to obtain a local offset of each of the region image pairs.

[0028] In the technical scheme, the method can avoid low efficiency of the matching process caused by falling into local optimization or redundant calculation through setting of the maximum iteration number, and ensure completion of point set matching within a reasonable time. In addition, the method can effectively screen reliable matching pairs (inliers) in the bright field and fluorescent centroid point sets through setting of the inlier judgment threshold, eliminate false matching pairs (outliers) caused by noise and microsphere detection errors, and reduce interference of abnormal data on calculation of local offsets. Finally, the method can also improve output stability of accurate local offsets of each regional image pair based on the parameterized matching logic.

[0029] In some embodiments, the centroid local matching of the bright field centroid point set and the fluorescent centroid point set of each of the regional image pairs according to the matching algorithm parameters to obtain the local offset of each of the regional image pairs comprises:

[0030] initializing a global optimal inlier proportion of each of the regional image pairs;

[0031] obtaining a matching point pair set of each of the regional image pairs from the bright field centroid point set and the fluorescent centroid point set of each of the regional image pairs;

[0032] calculating an affine transformation matrix of each of the regional image pairs according to the matching point pairs;

[0033] performing coordinate transformation on all centroid points in the bright field centroid point set according to the affine transformation matrix to obtain a transformed centroid point set of each of the regional image pairs;

[0034] calculating an Euclidean distance between each transformed centroid point in the transformed centroid point set and its nearest neighbor point in the fluorescent centroid point set to obtain an Euclidean distance set of each of the regional image pairs;

[0035] determining an inlier set of each of the regional image pairs according to the Euclidean distance set and the inlier judgment threshold;

[0036] calculating an inlier proportion of each of the regional image pairs according to an inlier number of the inlier set and a total point number of the transformed centroid point set;

[0037] when the inlier proportion is greater than a currently stored global optimal inlier proportion, storing the inlier proportion as a new global optimal inlier proportion of each of the regional image pairs, and determining an affine transformation matrix corresponding to an iteration round of the inlier proportion as an optimal affine transformation matrix of each of the regional image pairs;

[0038] when the current iteration number reaches the maximum iteration number, determining the finally determined optimal affine transformation matrix as a target affine transformation matrix of each of the regional image pairs;

[0039] Generate a local offset of each of the region image pairs according to the last column of the target affine transformation matrix.

[0040] In the above technical solution, the method can continuously screen out the affine transformation matrix that best reflects the true registration relationship by initializing the global optimal inlier ratio and iteratively updating, thereby avoiding accidental error of single calculation. The centroid coordinate mapping can be realized by means of the affine transformation matrix, and the inliers can be screened by the Euclidean distance and the inlier threshold, so as to accurately exclude the outliers caused by the detection deviation of microspheres and image noise, thereby ensuring the reliability of the matched point pairs. Meanwhile, the local offset is generated based on the optimal affine transformation matrix when the maximum number of iterations is reached, which not only ensures the convergence and efficiency of the matching process, but also makes the local offset strictly fit the actual position corresponding relationship of the microspheres in the region, thereby providing high-credibility local data for the fusion of subsequent global registration offset.

[0041] In some embodiments, the acquiring, from the bright-field centroid point set of each of the region image pairs and the fluorescent centroid point set, a matched point pair set of each of the region image pairs comprises:

[0042] The matched point pair set of each of the region image pairs is obtained by randomly selecting point pairs from the bright-field centroid point set of each of the region image pairs and the fluorescent centroid point set based on the image center Gaussian distribution algorithm.

[0043] Alternatively, the matched point pair set of each of the region image pairs is obtained by randomly selecting multiple groups of nearest neighbor point pairs from the bright-field centroid point set of each of the region image pairs and the fluorescent centroid point set.

[0044] In the above technical solution, the image center Gaussian distribution point selection can preferentially select centroid point pairs close to the core region of the image and more representative, thereby reducing the interference of noise points in the edge region and adapting to the image scene with relatively concentrated cell distribution. The nearest neighbor point selection can directly select point pairs with the strongest position correlation according to the spatial distance, thereby quickly establishing a preliminary matching relationship and adapting to the scene with relatively dispersed microsphere distribution but clear spatial correspondence.

[0045] In some embodiments, the fusion processing of the local offsets of all the region image pairs to obtain a global registration offset comprises:

[0046] Acquiring a weight coefficient of each of the region image pairs;

[0047] Weighted average fusion calculation is performed on the local offsets of all the region image pairs according to the weight coefficients to obtain a global registration offset.

[0048] In the technical solution, the method can dynamically allocate weights according to image quality (such as microsphere definition, noise level), microsphere quantity and other factors of different regions, so that the region with higher matching accuracy and stronger information reliability has a higher weight in global calculation, and the interference of low-quality region data on the global result is reduced.

[0049] In some embodiments, the method further comprises:

[0050] Based on the global registration offset, the local offset of each of the region image pairs is feedback detected, an adaptability score of each of the local offsets is obtained, and a plurality of the adaptability scores are output.

[0051] In the technical solution, the method can realize fine monitoring of the registration result, thereby enhancing the interpretability and debuggability of the method, and helping to quickly locate and solve local registration abnormal problems in actual application.

[0052] In the second aspect, the application provides a cell image registration device, comprising:

[0053] An image acquisition unit is configured to acquire images of fluorescent microspheres as substitutes for cell samples to obtain bright field images and fluorescent images.

[0054] A region division unit is configured to divide the bright field images and the fluorescent images to obtain a plurality of region image pairs. Each of the region image pairs comprises a bright field region image and a fluorescent region image corresponding in position.

[0055] A centroid detection unit is configured to detect centroids of the region image pairs to obtain a bright field centroid point set and a fluorescent centroid point set of each of the region image pairs.

[0056] A centroid local matching unit is configured to perform centroid local matching on the bright field centroid point set and the fluorescent centroid point set to obtain a local offset of each of the region image pairs.

[0057] A fusion unit is configured to perform fusion processing on the local offsets of all the region image pairs to obtain a global registration offset.

[0058] In the technical solution, the device can effectively solve the problems of difficult feature point extraction and unreliable matching caused by the difference in imaging mechanism between the bright field images and the fluorescent images. Through centroid detection and local-global combined offset calculation logic with the fluorescent microspheres as a unified reference, the accuracy and stability of the cell image registration result are significantly improved, and reliable data support is provided for precise fusion analysis of cell structure information and molecular marker information in cell biology research and drug screening.

[0059] In a third aspect, the present application provides an electronic device, comprising a memory and a processor, wherein the memory is configured to store a computer program, and the processor is configured to execute the computer program to enable the electronic device to perform the cell image registration method according to any one of the first aspect.

[0060] In a fourth aspect, the present application provides a readable storage medium, wherein the readable storage medium stores a computer program, and the computer program is executed by a processor to perform the cell image registration method according to any one of the first aspect.

[0061] In a fifth aspect, the present application provides a computer program product, wherein the computer program product comprises a computer program, and the computer program is executed by a processor to perform the cell image registration method according to any one of the first aspect.

[0062] The present application has the beneficial effect that the accurate microsphere centroid point extraction effect can be achieved without the aid of traditional feature point detection algorithm, but only by morphological operation combined with contour analysis and geometric center calculation. Meanwhile, more accurate microsphere contour can be obtained by using independent binary threshold value for processing in each sub-region. Finally, the RANSAC matching is only performed in each corresponding sub-region, and the calculation amount is greatly reduced based on the fact that the number of centroid points in the local region is much smaller than the number of centroid points in the whole image. BRIEF DESCRIPTION OF DRAWINGS

[0063] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments of the present application. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope, and for those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0064] Figure 1 The flowchart of the cell image registration method in some embodiments of the present application;

[0065] Figure 2 The flowchart of the cell image registration method in some embodiments of the present application;

[0066] Figure 3 The schematic diagram of microsphere position deviation in bright field and fluorescence environment in some embodiments of the present application;

[0067] Figure 4 The image division legend in some embodiments of the present application;

[0068] Figure 5 The bright field and fluorescence image processing flowchart example in some embodiments of the present application;

[0069] Figure 6 An example diagram of cell contour centroid detection for brightfield and fluorescent images in some embodiments of the present application;

[0070] Figure 7 A structural schematic diagram of a cell image registration apparatus in some embodiments of the present application;

[0071] Figure 8 A structural schematic diagram of an electronic device in some embodiments of the present application. DETAILED DESCRIPTION

[0072] The embodiments of the technical solutions of the present application will be described in detail below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present application, and therefore only serve as examples, and cannot limit the protection scope of the present application.

[0073] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs; the terms used herein are only for the purpose of describing specific embodiments of the present application, and are not intended to limit the present application; the terms "include" and "have" and any variations thereof in the specification and claims of the present application and the above description of drawings are intended to cover non-exclusive inclusion.

[0074] In the description of the embodiments of the present application, the technical terms "first", "second", etc. are only used to distinguish different objects, and cannot be understood as indicating or implying relative importance or implicitly indicating the number, specific order or primary and secondary relationship of the indicated technical features. In the description of the embodiments of the present application, the meaning of "a plurality of" is two or more (including two), and similarly, "a plurality of groups" means two or more groups (including two groups), and "a plurality of pieces" means two or more pieces (including two pieces) unless otherwise explicitly and specifically limited.

[0075] Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the present application. The appearance of the phrase in various places in the specification does not necessarily all refer to the same embodiment, nor is it necessarily mutually exclusive or alternative to other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0076] In the description of the embodiments of the present application, the term "and / or" is only a description of the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B, which can represent the three cases of A alone, A and B together, and B alone. In addition, the character " / " in this paper generally represents that the front and rear associated objects are a "or" relationship.

[0077] Due to physical offset in the imaging process, the bright field image and the fluorescence image will have spatial position deviation. At present, in order to realize accurate registration of the images, a batch of fluorescent microspheres (which appear as regular circles or bright spots in the image and can be seen in the bright field and fluorescence images) is usually added in the experiment, and the spatial position deviation of the bright field image and the fluorescence image is calculated by detecting the corresponding positions of the microspheres in the bright field image and the fluorescence image, so as to complete the image registration. Among them, the existing microsphere registration methods mainly include manual annotation method and image-based registration method.

[0078] However, the manual annotation method is simple and intuitive, and does not require complex algorithms. However, it has a large workload, low efficiency, and is easily affected by the subjective influence of the operator.

[0079] In the image-based registration method, the feature point-based registration method (such as the classic SIFT and ORB algorithms) usually cannot detect feature points with good correspondence in multi-modal images; the registration method based on deep learning often needs a large amount of training data for model training, and the model trained by it usually has poor interpretability.

[0080] In view of the above technical problems, the embodiments of the present application provide a bright field image and fluorescence image microsphere registration method based on region division and centroid detection, which mainly includes the following four registration processes: region division, centroid detection, local matching of centroids, and fusion of matching results.

[0081] On this basis, the method can realize accurate extraction of microsphere centroid points without the help of traditional feature point detection algorithms, only by morphological operation combined with contour analysis and geometric center calculation. At the same time, using independent binary threshold for processing in each sub-region, more accurate microsphere contours can also be obtained. Finally, RANSAC matching is only performed in each corresponding sub-region, and the calculation amount is greatly reduced based on the fact that the number of centroid points in the local region is much smaller than the number of centroid points in the whole image.

[0082] As shown in Figure 1 Some embodiments of the present application provide a cell image registration method, which comprises:

[0083] S101, image acquisition is performed on fluorescent microspheres of a substitute cell sample to obtain a bright field image and a fluorescence image;

[0084] S102, the bright field image and the fluorescence image are divided to obtain a plurality of region image pairs; each region image pair comprises a bright field region image and a fluorescence region image corresponding in position;

[0085] S103, centroid detection is performed on the regional image pair to obtain a bright field centroid point set and a fluorescent centroid point set of each regional image pair;

[0086] S104, centroid local matching is performed on the bright field centroid point set and the fluorescent centroid point set to obtain a local offset of each regional image pair;

[0087] S105, fusion processing is performed on the local offsets of all regional image pairs to obtain a global registration offset.

[0088] In some embodiments, the fluorescent microspheres refer to small particles with fluorescent properties, uniform size and stability (usually with a diameter of nanometers to microns, appearing as regular circles or bright spots in images, and visible in both bright field and fluorescent images). The core role is that they can be identified in bright field images by morphological / gray level differences and detected in fluorescent images by specific fluorescent signals, thereby building a "bridge" between the two types of modal images.

[0089] In some embodiments, the bright field image refers to a conventional optical imaging image, and the imaging principle is to generate an image by using the difference between absorption and transmission of light when passing through a cell sample. Among them, the high-density structure of the cell (such as the cell nucleus) will absorb more light, showing a darker area; the background and low-density structure will transmit more light, showing a brighter area. This image mainly reflects the overall morphology, contour and other structural information of the cell.

[0090] In some embodiments, the fluorescent image refers to an image obtained based on fluorescent imaging technology, which needs to be used in conjunction with a fluorescent marker (such as the fluorescent microspheres in the present method, or other molecular specific fluorescent probes). Among them, the fluorescent marker will emit fluorescence under the irradiation of a specific wavelength excitation light, and the imaging system captures the fluorescent signal to generate an image. This image can specifically show the position of the marker (such as the distribution of the fluorescent microspheres).

[0091] In some embodiments, the regional image pair refers to a combination of regions corresponding one by one after the bright field image and the fluorescent image are divided according to the same rule, and each combination contains "one bright field regional image" and "one fluorescent regional image". For example, the bright field image is uniformly divided into 10x10 small regions, and the fluorescent image is divided into 10x10 small regions according to the same coordinate range and size, and the "first row, first column region" of the bright field and the "first row, first column region" of the fluorescent image constitute a regional image pair.

[0092] In some embodiments, centroid detection refers to the process of calculating the geometric center coordinates of the target (in this method, the outline of a fluorescent microsphere) using image algorithms. The core of this process is quantization calculation using "image moments" (such as the zeroth moment reflecting the target area and the first moment reflecting the centroid of the target's grayscale distribution). The specific process is as follows: first, the region image is grayscaled and binarized to highlight the microsphere outline; then, the microsphere's range is locked through outline detection; and finally, the centroid coordinates of this outline are calculated. The aim is to transform the "microsphere outline" into precise "point coordinates" (i.e., the centroid point), providing standardized feature points for subsequent point set matching.

[0093] In some embodiments, the bright-field centroid set refers to the set of centroid coordinates of all microspheres corresponding to the bright-field region image obtained by centroid detection in a single region image pair; the fluorescence centroid set refers to the set of centroid coordinates of all microspheres corresponding to the fluorescence region image obtained by centroid detection in a single region image pair.

[0094] In some embodiments, centroid local matching refers to the process of establishing a mapping relationship between corresponding points in the "bright field centroid set" and the "fluorescence centroid set" within a single region image pair using an algorithm (such as the RANSAC algorithm mentioned above). The core is to filter "valid matching point pairs" (inner points), eliminate "incorrect matching point pairs" (outer points, such as the centroids of false positive microspheres caused by noise), and calculate the positional offset relationship between the bright field and fluorescence images within the region based on the valid point pairs.

[0095] In some embodiments, local offset refers to the positional offset parameter (typically a two-dimensional coordinate offset, such as Δx, Δy, or affine transformation parameters including rotation and scaling) of a bright field image relative to a fluorescence image in a single region image pair obtained through "centroid local matching".

[0096] In some embodiments, global registration offset refers to the uniform offset relationship applicable to the entire brightfield image and fluorescence image obtained after weighted averaging and fusing the "local offsets" of all region image pairs.

[0097] In the above embodiments, this method can effectively solve the problems of difficult feature point extraction and unreliable matching caused by the difference in imaging mechanism between bright field images and fluorescence images. By using centroid detection with fluorescent microspheres as a unified reference and offset calculation logic that combines local and global approaches, the accuracy and stability of cell image registration results are significantly improved, providing reliable data support for the precise fusion analysis of cell structure information and molecular marker information in cell biology research and drug screening.

[0098] In some embodiments, multiple bright-field regions in a bright-field image have the same image size;

[0099] The multiple fluorescent region images in the fluorescent image have consistent sizes.

[0100] In some embodiments, the method can divide the bright-field image and the fluorescent image into multiple regions with consistent sizes and independent of each other before feature point detection. Thus, when performing registration, registration is only performed between corresponding images.

[0101] In some embodiments, because the method needs to calculate the distance between each microsphere centroid in the bright-field image and each microsphere in the fluorescent image in the registration stage, the calculation complexity of this step is O(n 2 ). However, after region division, the method only needs to calculate the distance between microspheres in the corresponding region, thereby greatly reducing the calculation amount. Meanwhile, the registration work of each region can also use parallel computing, further accelerating the calculation speed.

[0102] For example, the method actually tests the time for the above-mentioned case: in the case of using the same registration algorithm (RANSAC), the total registration time without region division is 10.5 seconds, and the total registration time after region division (divided into 4 rows and 6 columns, a total of 24 independent regions) is 0.6 seconds, and the time is reduced by about 94.3%.

[0103] In the above embodiments, the method can ensure that the region division rules of the two types of images are uniform, provide a standardized comparison basis for subsequent centroid detection and local offset calculation of corresponding region image pairs, and avoid imbalance of local registration benchmarks due to differences in region sizes.

[0104] In some embodiments, the centroid detection is performed on the region image pairs to obtain a bright-field centroid point set and a fluorescent centroid point set of each region image pair, including:

[0105] Each region image pair is subjected to grayscale processing to obtain a first processed region image pair.

[0106] Each first processed region image pair is subjected to binarization processing to obtain a second processed region image pair.

[0107] Each second processed region image pair is subjected to microsphere contour detection to obtain a microsphere contour image pair including a microsphere external contour; wherein the microsphere contour image pair includes one bright-field microsphere contour region image and one fluorescent microsphere contour region image corresponding in position.

[0108] The centroid detection is performed on the microsphere contour image pair to obtain a bright-field centroid point set and a fluorescent centroid point set of each region image pair.

[0109] In some embodiments, due to the differences in imaging differences, texture structure between bright field images and fluorescence images, the feature points extracted by SIFT, ORB and other methods lack stable correspondence relationship, and reliable matching cannot be achieved. Therefore, the method proposes a microsphere centroid detection method suitable for bright field and fluorescence images. In view of the differences in the structure of the two types of images, a differentiated processing strategy is adopted.

[0110] In some embodiments, the method first performs a grayscale operation on the bright field image and the fluorescence image. At this time, the microsphere edge in the bright field image is black, and the rest is white; the microsphere region in the fluorescence image is white, and the rest is black. Based on this, the method can perform a binary inversion operation (greater than the threshold value is set to black, less than the threshold value is set to white) on the bright field image, so that in the processed binary image, the microsphere edge is converted to white, and the rest is black. And directly binarize (greater than the threshold value is set to white, less than the threshold value is set to black) the fluorescence image, so that in the processed image, the microsphere region is converted to white, and the rest is black.

[0111] On this basis, the method can use the contour detection algorithm to detect the external contour of the microsphere for the binary bright field and fluorescence images. Then, for the external contour of the microsphere, the centroid coordinates of the contour are calculated by calculating the zeroth moment (M 00 ) and the first moment (M 10 and M 01 ).

[0112] .

[0113] In the above embodiments, the method can systematically eliminate redundant gray scale information, background noise interference in bright field and fluorescence images through the progressive image processing flow of "grayscale→binarization→microsphere contour detection→centroid detection", and accurately separate the effective contour area of the fluorescent microsphere. Among them, grayscale and binarization processing can simplify the image information dimension and highlight the light and dark differences between microspheres and background; targeted microsphere contour detection can also accurately lock the target area and avoid interference of cell structure or other impurities on centroid recognition, so as to ensure that the bright field centroid point set and the fluorescence centroid point set extracted from the contour image accurately correspond to the actual position of the fluorescent microsphere.

[0114] In some embodiments, the centroid detection is performed on the microsphere contour image pair to obtain the bright field centroid point set and the fluorescence centroid point set of each region image pair, including:

[0115] calculating a microsphere contour geometric feature of each of the region image pairs; wherein the microsphere contour geometric feature comprises a first geometric feature of an outer contour of each microsphere in the bright field microsphere contour region image, and a second geometric feature of the outer contour of each microsphere in the fluorescent microsphere contour region image, the first geometric feature and the second geometric feature each comprising at least a zeroth moment and a first moment of the outer contour of the microsphere;

[0116] calculating, according to the first geometric feature, a centroid coordinate of the outer contour of each microsphere in the bright field microsphere contour region image, to obtain a bright field centroid point set of the region image pairs;

[0117] calculating, according to the second geometric feature, a centroid coordinate of the outer contour of each microsphere in the fluorescent microsphere contour region image, to obtain a fluorescent centroid point set of the region image pairs.

[0118] In the above embodiments, the method can rely on the mathematical properties of the moment method to convert the spatial form information of the microsphere contour into accurate numerical parameters, avoiding the dependence on texture and edge in artificial subjective judgment or traditional feature extraction, and fundamentally eliminating the interference of the modal difference between the bright field and fluorescent images on the centroid positioning. At the same time, the combination calculation of the zeroth moment (reflecting the contour area / gray sum) and the first moment (reflecting the contour gray distribution barycenter) can also accurately anchor the geometric center of the microsphere contour, thereby ensuring that the coordinate values of the bright field centroid point set and the fluorescent centroid point set strictly correspond to the actual spatial position of the microsphere.

[0119] In some embodiments, the bright field centroid point set and the fluorescent centroid point set are subjected to centroid local matching to obtain a local offset of each region image pair, comprising:

[0120] obtaining a preset matching algorithm parameter; wherein the matching algorithm parameter comprises at least a maximum iteration number and an inlier judgment threshold value;

[0121] performing centroid local matching on the bright field centroid point set and the fluorescent centroid point set of each region image pair according to the matching algorithm parameter, to obtain a local offset of each region image pair.

[0122] In some embodiments, the method can use the RANSAC (Random Sample Consensus) algorithm to realize the matching of the microsphere centroid points in the bright field and fluorescent images.

[0123] In the above embodiment, the method can avoid low efficiency of the matching process due to falling into local optimization or redundant calculation through setting of the maximum iteration number, ensuring completion of point set matching within a reasonable time; and can effectively screen out reliable matching pairs (inliers) in the bright field and fluorescent centroid point sets through setting of the inlier judgment threshold, eliminate false matching pairs (outliers) generated due to noise and microsphere detection error, and reduce interference of abnormal data on calculation of local offsets. Finally, relying on the parameterized matching logic, the output stability of the accurate local offset of each regional image pair can also be improved.

[0124] In some embodiments, the centroid local matching of the bright field centroid point set and the fluorescent centroid point set of each regional image pair is performed according to the matching algorithm parameters, to obtain the local offset of each regional image pair, including:

[0125] Initializing the global optimal inlier proportion of each regional image pair;

[0126] From the bright field centroid point set and the fluorescent centroid point set of each regional image pair, a matching point pair set of each regional image pair is obtained;

[0127] According to the matching point pairs, an affine transformation matrix of each regional image pair is calculated;

[0128] According to the affine transformation matrix, coordinate transformation is performed on all centroid points in the bright field centroid point set, to obtain a transformed centroid point set of each regional image pair;

[0129] The Euclidean distance between each transformed centroid point in the transformed centroid point set and its nearest neighbor point in the fluorescent centroid point set is calculated, to obtain a Euclidean distance set of each regional image pair;

[0130] According to the Euclidean distance set and the inlier judgment threshold, an inlier set of each regional image pair is determined;

[0131] According to the inlier number of the inlier set and the total point number of the transformed centroid point set, an inlier proportion of each regional image pair is calculated;

[0132] When the inlier proportion is greater than the currently stored global optimal inlier proportion, the inlier proportion is stored as a new global optimal inlier proportion of each regional image pair, and the affine transformation matrix corresponding to the iteration round of the inlier proportion is determined as the optimal affine transformation matrix of each regional image pair;

[0133] When the current iteration number reaches the maximum iteration number, the finally determined optimal affine transformation matrix is determined as the target affine transformation matrix of each regional image pair;

[0134] According to the last column of the target affine transformation matrix, a local offset of each regional image pair is generated.

[0135] Exemplarily, assuming that the centroid point set extracted in the bright field image is P, the centroid point set extracted in the fluorescence image is Q, the RANSAC algorithm parameters are set, the maximum iteration number N is set, and the inlier judgment threshold t is set (the threshold t is used to judge whether the distance between the matched points is close enough to be recognized as an inlier. The value needs to be set according to the specific microsphere size, etc. This method uses 2 pixels).

[0136] Based on this, in each iteration, the method randomly selects three pairs of nearest neighbor matched points from the bright field point set P and the fluorescence image point set Q to form a subset {p, q}, which is used to calculate the affine transformation matrix. Among them, three pairs of matched points are selected because the minimum number of points required to calculate the affine transformation matrix in a two-dimensional plane is three pairs of points that are not collinear; at the same time, this step can also randomly select point pairs based on the Gaussian distribution of the image center, so that the points in the center area of the image are more likely to be selected (due to the deviation of the bright field and the fluorescence image, the points in the edge area of the image may be missing, and the centroid points in the center area are more likely to have stable corresponding relationship).

[0137] Then, the method uses the selected three pairs of points to calculate the affine transformation matrix A, which satisfies q≈A·p, and A can transform all the points in the bright field image into the coordinate system of the fluorescence image.

[0138] For all points in the bright field image point set P, the matrix A is used for transformation to obtain the point set For each point in , the nearest neighbor point in the fluorescence image point set Q is searched, and the Euclidean distance d is calculated. If d<t, the point is considered as an “inlier”.

[0139] The number of inliers in this round N inlier is counted, and the proportion of the total number of points is calculated:

[0140] .

[0141] If the inlier proportion of the current model is greater than the optimal model, the current affine matrix is updated as the optimal result.

[0142] After all iterations are completed, the affine transformation matrix with the highest inlier proportion is output as the final matching result, and the last column of the affine transformation matrix is extracted, that is, the offset offset.

[0143] In some embodiments, the method extracts the last column of the affine transformation matrix, which essentially only retains the translation part, and no longer retains other transformation information such as rotation, scaling, and shearing that may be included in the affine transformation.

[0144] In the above embodiment, the method can continuously screen out the affine transformation matrix that best reflects the true registration relationship by initializing the global optimal inner point ratio and iteratively updating, avoiding accidental error of single calculation. The centroid coordinate mapping can be realized by means of the affine transformation matrix, and the outer points generated due to the detection deviation of microspheres and image noise can be accurately removed by screening inner points through the Euclidean distance and inner point threshold, so as to ensure the reliability of the matched point pairs. Meanwhile, the local offset is generated based on the optimal affine transformation matrix when the maximum number of iterations is reached, which can not only ensure the convergence and efficiency of the matching process, but also make the local offset strictly fit the actual position corresponding relationship of the microspheres in the region, thereby providing high-credibility local data for the fusion of subsequent global registration offset.

[0145] In some embodiments, the matched point pair set of each regional image pair is obtained from the bright field centroid point set and the fluorescent centroid point set of each regional image pair, including:

[0146] The matched point pair set of each regional image pair is obtained by randomly selecting point pairs from the bright field centroid point set and the fluorescent centroid point set of each regional image pair based on the image center Gaussian distribution algorithm.

[0147] Or, the matched point pair set of each regional image pair is obtained by randomly selecting multiple groups of nearest neighbor point pairs from the bright field centroid point set and the fluorescent centroid point set of each regional image pair.

[0148] In the above embodiment, the image center Gaussian distribution based point selection can preferentially select centroid point pairs close to the core area of the image and more representative, reduce the interference of noise points in the edge area, and adapt to image scenes with relatively concentrated cell distribution. The nearest neighbor point selection can directly select point pairs with the strongest position correlation according to the spatial distance, quickly establish a preliminary matching relationship, and adapt to scenes with relatively dispersed microsphere distribution but clear spatial correspondence.

[0149] In some embodiments, the local offsets of all regional image pairs are fused to obtain a global registration offset, including:

[0150] The weight coefficient of each regional image pair is obtained.

[0151] The local offsets of all regional image pairs are weighted and averaged to obtain the global registration offset.

[0152] For example, it is assumed that the original image is divided into m x n sub-regions in the above steps, and the RANSAC centroid point registration process is performed on each sub-region to obtain the local offset offset (i,j) of the region. Thus, the following matrix can be obtained:

[0153] offset(1,1); offset(2,1) …… offset(l,m);

[0154] offset(2,1); offset(2,2) …… offset(2,m);

[0155] …………………………………………

[0156] offset(m,1); offset(m,2) …… ofset(n,m).

[0157] At this time, in order to generate the global registration offset of the whole image, the offsets of the regions are fused in a weighted average manner. The specific calculation method is:

[0158] .

[0159] By default, all weights can be set to 1, that is, the contributions of the regions are considered equal;

[0160] If the image has slight distortion, etc., the weight of the central region can be set to be larger, and the weight of the edge region can be set to be smaller.

[0161] In the above embodiment, the method can dynamically allocate weights according to the image quality (such as microsphere clarity, noise level), the number of microspheres, and other factors of different regions, so that the offset of the region with higher matching accuracy and stronger information reliability occupies a higher weight in global calculation, and the interference of low-quality region data on the global result is reduced.

[0162] In some embodiments, the method further comprises:

[0163] Based on the global registration offset, the local offset of each region image pair is feedback detected to obtain an adaptability score of each local offset, and a plurality of adaptability scores are output.

[0164] In some embodiments, a quantification system can be introduced in the method to feedback the local offset for scoring.

[0165] In the above embodiment, the method can realize fine monitoring of the registration result, thereby enhancing the explainability and debuggability of the method, and helping to quickly locate and solve the local registration abnormal problem in actual application.

[0166] In order to make the purposes, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be described clearly and completely below. In some embodiments, as shown in Figure 2 The cell image registration method comprises:

[0167] Figure 3 A schematic diagram of microsphere position deviation in bright field and fluorescence environment is shown.

[0168] S201, the bright field image and the fluorescence image are respectively regionally divided, and the original image is divided into a plurality of equal size independent regions.

[0169] Please refer to Figure 4 , Figure 4 A image division legend is shown.

[0170] S202, different morphological detection methods are used for the bright field image and the fluorescence image to obtain microsphere contours, and then the image matrix of the contours is calculated to obtain the centroid coordinates of the microspheres as registration feature points.

[0171] Please refer to Figure 5 , Figure 5 A bright field and fluorescence image processing flow example diagram is shown.

[0172] Please refer to Figure 6 , Figure 6 A bright field and fluorescence image cell contour centroid detection example diagram is shown.

[0173] S203, in each sub-region, based on the extracted centroid point set in the bright field and fluorescence images, a RANSAC algorithm is used for local registration estimation.

[0174] S204, the local registration results of each region are fused to calculate the overall registration relationship of the whole image.

[0175] In some embodiments, the bright field image and the fluorescence image have obvious differences in imaging mechanism, and the method can design differentiated microsphere detection methods for the two types of images. For the bright field image, an inverted binary + contour detection strategy is adopted. For the fluorescence image, a binary + contour detection strategy is adopted. In this way, the microsphere contour edge in the two types of backgrounds is successfully detected.

[0176] In some embodiments, the method can first use a more targeted adaptive binary threshold for each independent region, and the binary result is more accurate, which is beneficial to subsequent contour detection. Then, in the matching process, the corresponding sub-regions are matched, which can reduce a large amount of calculation time and improve the detection efficiency. Finally, the results of multiple regions are fused, and different weights can be set for each region according to the specific imaging situation for fusion.

[0177] Figure 7 A structural schematic diagram of a cell image registration device is shown, and it should be understood that the device is used in the method of the present application. Figure 1The method executed by the device corresponds to the method, and the device can execute the steps involved in the method. The specific functions and effects of the device can be referred to the description in the foregoing method. To avoid repetition, the detailed description is appropriately omitted here.

[0178] The cell image registration device comprises:

[0179] The image acquisition unit 310 is configured to acquire images of the fluorescent microspheres in the substitute cell sample to obtain a bright field image and a fluorescent image.

[0180] The region division unit 320 is configured to divide the bright field image and the fluorescent image to obtain a plurality of region image pairs. Each region image pair comprises a bright field region image and a fluorescent region image corresponding in position.

[0181] The centroid detection unit 330 is configured to detect centroids of the region image pairs to obtain a bright field centroid point set and a fluorescent centroid point set of each region image pair.

[0182] The centroid local matching unit 340 is configured to perform centroid local matching on the bright field centroid point set and the fluorescent centroid point set to obtain a local offset of each region image pair.

[0183] The fusion unit 350 is configured to perform fusion processing on the local offsets of all the region image pairs to obtain a global registration offset.

[0184] In some embodiments, the plurality of bright field region images in the bright field image are of the same size.

[0185] The plurality of fluorescent region images in the fluorescent image are of the same size.

[0186] In some embodiments, the centroid detection unit 330 comprises:

[0187] The grayscale unit 331 is configured to perform grayscale processing on each region image pair to obtain a first processed region image pair.

[0188] The binarization unit 332 is configured to perform binarization processing on each first processed region image pair to obtain a second processed region image pair.

[0189] The contour detection unit 333 is configured to perform microsphere contour detection on each second processed region image pair to obtain a microsphere contour image pair comprising a bright field microsphere contour region image and a fluorescent microsphere contour region image corresponding in position.

[0190] The centroid detection unit 334 is configured to detect centroids of the microsphere contour image pair to obtain a bright field centroid point set and a fluorescent centroid point set of each region image pair.

[0191] In some embodiments, the centroid detection unit 334 is specifically configured to calculate a microsphere contour geometric feature of each region image pair; wherein the microsphere contour geometric feature comprises a first geometric feature of an external contour of each microsphere in the bright field microsphere contour region image, and a second geometric feature of the external contour of each microsphere in the fluorescent microsphere contour region image, and the first geometric feature and the second geometric feature each at least comprise a zeroth moment and a first moment of the external contour of the microsphere;

[0192] According to the first geometric feature, the centroid coordinates of the external contour of each microsphere in the bright field microsphere contour region image are calculated to obtain a bright field centroid point set of the region image pair;

[0193] According to the second geometric feature, the centroid coordinates of the external contour of each microsphere in the fluorescent microsphere contour region image are calculated to obtain a fluorescent centroid point set of the region image pair.

[0194] In some embodiments, the centroid local matching unit 340 comprises:

[0195] The first acquisition sub-unit 341 is configured to acquire preset matching algorithm parameters; wherein the matching algorithm parameters at least comprise a maximum iteration number and an inlier judgment threshold value;

[0196] The local matching sub-unit 342 is configured to perform centroid local matching on the bright field centroid point set and the fluorescent centroid point set of each region image pair according to the matching algorithm parameters to obtain a local offset of each region image pair.

[0197] In some embodiments, the local matching sub-unit 342 is specifically configured to initialize a global optimal inlier ratio of each region image pair;

[0198] From the bright field centroid point set and the fluorescent centroid point set of each region image pair, a matching point pair set of each region image pair is obtained;

[0199] An affine transformation matrix of each region image pair is calculated according to the matching point pairs;

[0200] According to the affine transformation matrix, coordinate transformation is performed on all centroid points in the bright field centroid point set to obtain a transformed centroid point set of each region image pair;

[0201] The Euclidean distance between each transformed centroid point in the transformed centroid point set and its nearest neighbor point in the fluorescent centroid point set is calculated to obtain a Euclidean distance set of each region image pair;

[0202] According to the Euclidean distance set and the inlier judgment threshold value, an inlier set of each region image pair is determined;

[0203] According to the inlier number of the inlier set and the total point number of the transformed centroid point set, an inlier ratio of each region image pair is calculated;

[0204] When the inlier ratio is greater than the currently stored global optimal inlier ratio, the inlier ratio is stored as a new global optimal inlier ratio of each regional image pair, and an affine transformation matrix corresponding to the iteration round of the inlier ratio is determined as an optimal affine transformation matrix of each regional image pair.

[0205] When the current iteration number reaches the maximum iteration number, the finally determined optimal affine transformation matrix is determined as a target affine transformation matrix of each regional image pair.

[0206] According to the last column of the target affine transformation matrix, a local offset of each regional image pair is generated.

[0207] In some embodiments, the local matching subunit 342 is further configured to obtain a matching point pair set of each regional image pair by randomly selecting point pairs from the bright field centroid point set and the fluorescence centroid point set of each regional image pair based on a Gaussian distribution algorithm based on image centers.

[0208] Or, a matching point pair set of each regional image pair is obtained by randomly selecting multiple groups of nearest neighbor point pairs from the bright field centroid point set and the fluorescence centroid point set of each regional image pair.

[0209] In some embodiments, the fusion unit 350 comprises:

[0210] The second acquisition subunit 351 is configured to acquire a weight coefficient of each regional image pair.

[0211] The fusion subunit 352 is configured to perform weighted average fusion calculation on the local offsets of all regional image pairs according to the weight coefficients, to obtain a global registration offset.

[0212] In some embodiments, the cell image registration device further comprises:

[0213] The detection unit 360 is configured to perform feedback detection on the local offset of each regional image pair based on the global registration offset, to obtain an adaptability score of each local offset, and output multiple adaptability scores.

[0214] As shown in Figure 8 The present application provides an electronic device 400, which comprises a processor 401 and a memory 402, the processor 401 and the memory 402 are interconnected and communicate with each other through a communication bus 403 and / or other forms of connection mechanism (not marked), the memory 402 stores a computer program executable by the processor 401, when the computing device runs, the processor 401 executes the computer program to execute the method in any of the preceding optional implementation manners.

[0215] The application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to perform the method in any of the optional implementation manners.

[0216] The computer readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0217] The application provides a computer program product, which comprises a computer program, and the computer program is executed by a processor to perform the method in any of the optional implementation manners.

[0218] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the application, and not to limit them; although the application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the application, and they should be covered in the scope of the claims and the specification of the application. Especially, as long as there is no conflict, the technical features mentioned in each embodiment can be combined in any way. The application is not limited to the specific embodiments disclosed in the text, but includes all technical solutions falling within the scope of the claims.

Claims

1. A cell image registration method, characterized in that, include: Images of fluorescent microspheres that served as substitute cell samples were acquired, yielding bright-field and fluorescence images; The bright-field image and the fluorescence image are divided to obtain multiple region image pairs; wherein each region image pair includes a bright-field region image and a fluorescence region image corresponding to the location; Centroid detection is performed on the region image pairs to obtain the bright field centroid set and fluorescence centroid set for each region image pair; Obtain preset matching algorithm parameters; wherein, the matching algorithm parameters include at least the maximum number of iterations and the interior point judgment threshold; Based on the matching algorithm parameters, local centroid matching is performed on the bright field centroid set and fluorescence centroid set of each region image pair to obtain the local offset of each region image pair. The local offsets of all the image pairs in the region are fused to obtain the global registration offset; The step of performing local centroid matching on the bright-field centroid set and fluorescence centroid set of each region image pair according to the matching algorithm parameters to obtain the local offset of each region image pair includes: Initialize the global optimal inlier ratio for each region image pair; A Gaussian distribution algorithm based on image center is used to randomly select point pairs from the bright field centroid set and the fluorescence centroid set of each region image pair to obtain a set of matching point pairs for each region image pair; or, multiple sets of nearest neighbor point pairs are randomly selected from the bright field centroid set and the fluorescence centroid set of each region image pair to obtain a set of matching point pairs for each region image pair. Calculate the affine transformation matrix for each region image pair based on the matching point pairs; The coordinates of all centroids in the bright field centroid set are transformed according to the affine transformation matrix to obtain the transformed centroid set of each region image pair; Calculate the Euclidean distance between each transformed centroid in the transformed centroid set and its nearest neighbor in the fluorescent centroid set to obtain the Euclidean distance set for each region image pair; Based on the Euclidean distance set and the interior point determination threshold, determine the interior point set for each region image pair; The proportion of interior points in each region image pair is calculated based on the number of interior points in the interior point set and the total number of points in the transformed centroid point set. When the inlier ratio is greater than the currently stored global optimal inlier ratio, the inlier ratio is stored as the new global optimal inlier ratio for each region image pair, and the affine transformation matrix of the iteration corresponding to the inlier ratio is determined as the optimal affine transformation matrix for each region image pair. When the current iteration count reaches the maximum iteration count, the final optimal affine transformation matrix is ​​determined as the target affine transformation matrix for each region image pair; The local offset of each region image pair is generated based on the last column of the target affine transformation matrix.

2. The cell image registration method according to claim 1, characterized in that, The multiple bright-field regions in the bright-field image have the same image size; The multiple fluorescent regions in the fluorescence image have the same image size.

3. The cell image registration method according to claim 1, characterized in that, The step of performing centroid detection on the region image pairs to obtain the bright-field centroid set and fluorescence centroid set for each region image pair includes: Each of the aforementioned region image pairs is converted to grayscale to obtain a first processed region image pair; Binarize each first processing region image pair to obtain a second processing region image pair; Microsphere contour detection is performed on each pair of images of the second processing region to obtain a pair of microsphere contour images including the outer contour of the microsphere; wherein, the pair of microsphere contour images includes a bright field microsphere contour region image and a fluorescent microsphere contour region image corresponding to the position; Centroid detection is performed on the microsphere contour image pairs to obtain the bright-field centroid set and fluorescence centroid set for each region image pair.

4. The cell image registration method according to claim 3, characterized in that, The centroid detection of the microsphere contour image pairs to obtain the bright-field centroid set and fluorescence centroid set for each region image pair includes: Calculate the microsphere contour geometric features for each of the said region image pairs; wherein, the microsphere contour geometric features include a first geometric feature of the outer contour of each microsphere in the bright field microsphere contour region image, and a second geometric feature of the outer contour of each microsphere in the fluorescent microsphere contour region image, wherein both the first geometric feature and the second geometric feature include at least the zero-order moment and the first-order moment of the outer contour of the microsphere; Based on the first geometric feature, the centroid coordinates of the outer contour of each microsphere in the bright-field microsphere contour region image are calculated to obtain the bright-field centroid point set of the region image pair; Based on the second geometric feature, the centroid coordinates of the outer contour of each microsphere in the image of the fluorescent microsphere contour region are calculated to obtain the set of fluorescent centroid points of the region image pair.

5. The cell image registration method according to claim 1, characterized in that, The process of fusing the local offsets of all the aforementioned region image pairs to obtain the global registration offset includes: Obtain the weight coefficients for each of the region image pairs; The global registration offset is obtained by weighted averaging and fusing the local offsets of all the image pairs in the region according to the weight coefficients.

6. The cell image registration method according to claim 1, characterized in that, The method further includes: Based on the global registration offset, the local offset of each region image pair is detected using feedback, and the fitness score of each local offset is obtained, and multiple fitness scores are output.

7. A cell image registration device, characterized in that, The cell image registration device includes: The image acquisition unit is used to acquire images of fluorescent microspheres that are substitutes for cell samples, and to obtain bright-field images and fluorescence images; A region segmentation unit is used to segment the bright-field image and the fluorescence image to obtain multiple region image pairs; wherein each region image pair includes a bright-field region image and a fluorescence region image corresponding to the location; The centroid detection unit is used to perform centroid detection on the region image pairs to obtain the bright field centroid set and the fluorescence centroid set for each region image pair. A centroid local matching unit is used to perform centroid local matching on the bright field centroid point set and the fluorescence centroid point set to obtain the local offset of each region image pair; The fusion unit is used to fuse the local offsets of all the region image pairs to obtain the global registration offset. The centroid local matching unit includes: The first acquisition subunit is used to acquire preset matching algorithm parameters; wherein, the matching algorithm parameters include at least the maximum number of iterations and the interior point judgment threshold; The local matching subunit is used to perform local centroid matching on the bright field centroid set and fluorescence centroid set of each region image pair according to the matching algorithm parameters, so as to obtain the local offset of each region image pair. The local matching subunit is specifically used to initialize the global optimal inlier ratio for each region image pair. From the bright field centroid set and the fluorescence centroid set of each region image pair, obtain the set of matching point pairs for each region image pair; Calculate the affine transformation matrix for each region image pair based on the matching point pairs; Based on the affine transformation matrix, the coordinates of all centroids in the bright field centroid set are transformed to obtain the transformed centroid set for each region image pair. Calculate the Euclidean distance between each transformed centroid in the transformed centroid set and its nearest neighbor in the fluorescent centroid set to obtain the set of Euclidean distances for each region image pair; Based on the Euclidean distance set and the interior point judgment threshold, determine the interior point set of each region image pair; Calculate the proportion of inliers for each region image pair based on the number of inliers in the inlier set and the total number of points in the transformed centroid set. When the inlier ratio is greater than the currently stored global optimal inlier ratio, the inlier ratio is stored as the new global optimal inlier ratio for each region image pair, and the affine transformation matrix corresponding to the inlier ratio in the iteration round is determined as the optimal affine transformation matrix for each region image pair. When the current iteration count reaches the maximum iteration count, the final optimal affine transformation matrix is ​​determined as the target affine transformation matrix for each region image pair; Based on the last column of the target affine transformation matrix, generate the local offset for each region image pair; Specifically, the local matching subunit is used to randomly select point pairs from the bright field centroid set and the fluorescence centroid set of each region image pair using a Gaussian distribution algorithm based on the image center, so as to obtain the matching point pair set of each region image pair. Alternatively, multiple sets of nearest neighbor point pairs can be randomly selected from the bright-field centroid set and the fluorescence centroid set of each region image pair to obtain the matching point pair set for each region image pair.

8. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor running the computer program to cause the electronic device to perform the cell image registration method according to any one of claims 1 to 6.

9. A readable storage medium, characterized in that, The readable storage medium stores a computer program, which, when executed by a processor, performs the cell image registration method according to any one of claims 1 to 6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, performs the cell image registration method according to any one of claims 1 to 6.

Images