A chromosome image recognition method and device, computer equipment and storage medium

Abstract

The application discloses a chromosome image recognition method and device, computer equipment and a storage medium, and relates to the technical field of chromosome image processing. The method comprises the following steps: after obtaining a plurality of independent chromosome monomer images, image feature extraction processing is performed on each chromosome monomer image, corresponding image features are obtained, the image features are subjected to optimization processing based on a white-box neural network ReduNet, corresponding image optimization features are obtained, and finally, the image optimization features are input into a pre-trained chromosome classifier to output corresponding chromosome monomer classification results. Through the vector optimization of the embedding representation after the image feature extraction network, the final chromosome classification recognition accuracy can be significantly improved, and through corresponding experiments, it is found that common evaluation indexes such as accuracy, precision and F value can reach more than 92.39%, and even reach 97.43%.

Description

Technical Field

[0001] This invention belongs to the field of chromosome image processing technology, specifically relating to a chromosome image recognition method, device, computer equipment, and storage medium. Background Technology

[0002] When performing chromosome karyotype analysis, it is often necessary to classify and identify anomalies in the segmented chromosome monosomy images. Currently, the common AI-assisted approach is to first use a deep learning-based segmentation network (generally an instance segmentation network, such as Mask R-CNN) to segment the real microscopic image within the cell nucleus, obtaining an independent image of each chromosome (i.e., a chromosome monosomy image), and then use a recognition network to classify and identify each chromosome.

[0003] Current chromosome image recognition technology requires first capturing images of chromosome slides under a high-power microscope. Before obtaining these high-power images, a preliminary identification is typically performed on images captured under low power. These identified images are then captured again under high power to obtain a high-power image, thus providing a better picture of chromosome division and reducing the workload for doctors. However, since identification is usually done visually by medical staff on the low-power images, it's difficult to assess the severity of chromosome crossing over in these images. This often results in low-quality images, reducing the accuracy of subsequent chromosome classification and identification. Summary of the Invention

[0004] The purpose of this invention is to provide a chromosome image recognition method, apparatus, computer device, and computer-readable storage medium to solve the problem of low accuracy in chromosome classification and recognition in existing chromosome image recognition technologies.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] Firstly, a chromosome image recognition method is provided, including:

[0007] Obtain true microscopic images of cell nuclei;

[0008] The real microscopic image is segmented into chromosome instances to obtain multiple independent images of chromosome monosomy.

[0009] Feature extraction processing is performed on each chromosome monomorph image in the plurality of chromosome monomorph images to obtain the image features of each chromosome monomorph image;

[0010] The image features of each chromosome monosomy image are optimized using a white-box neural network, ReduNet, to obtain the optimized image features of each chromosome monosomy image.

[0011] For each chromosome monosomy image, the corresponding optimized image features are input into a pre-trained chromosome classifier, and the corresponding chromosome monosomy classification result is output.

[0012] Based on the above-mentioned invention, a novel scheme for classifying and identifying chromosomes based on image feature optimization results is provided. This scheme involves obtaining multiple independent images of chromosome monosoms, firstly extracting image features from each individual image to obtain corresponding image features, then optimizing these features using a white-box neural network (ReduNet) to obtain corresponding optimized image features. Finally, these optimized image features are input into a pre-trained chromosome classifier, which outputs the corresponding chromosome monosom classification result. By performing vector optimization of the embedding representation after the image feature extraction network, the final chromosome classification accuracy can be significantly improved. Experiments show that common evaluation metrics such as accuracy, precision, and F-score can reach over 92.39%, even reaching 97.43%. Furthermore, it offers advantages such as eliminating the need for manual identification and rapid automatic identification, making it suitable for practical application and widespread adoption.

[0013] In one possible design, after acquiring a real microscopic image of the cell nucleus and before performing chromosome instance segmentation on the real microscopic image, the method further includes:

[0014] The real microscopic image is subjected to image filtering processing to obtain a real microscopic image with image noise removed;

[0015] The real microscopic image with noise removed is preprocessed to obtain the real microscopic image to be segmented. The preprocessing includes image size uniform adjustment and pixel value standardization. The pixel value standardization refers to uniformly calibrating the image color distribution and / or exposure.

[0016] In one possible design, the real microscopic image is segmented into chromosome instances to obtain multiple independent images of chromosome monosomy, including:

[0017] A deep learning-based instance segmentation algorithm is used to segment the real microscopic image into chromosome instances, resulting in multiple independent images of chromosome monosomy.

[0018] In one possible design, a deep learning-based instance segmentation algorithm is used to segment the real microscopic image into chromosome instances, resulting in multiple independent chromosome monosomy images, including:

[0019] A deep learning-based instance segmentation algorithm is used to first identify chromosome boundary lines in the real microscopic image, and then segment the real microscopic image into multiple independent chromosome monomorphic images based on the chromosome boundary lines.

[0020] In one possible design, feature extraction processing is performed on each chromosome monomorph image among the plurality of chromosome monomorph images to obtain the image features of each chromosome monomorph image, including:

[0021] For each chromosome monomorphic image among the multiple chromosome monomorphic images, the corresponding image is input into an image feature extraction model that is based on a convolutional neural network and a ternary loss function and has been pre-trained in a supervised manner, and the corresponding image features are output.

[0022] In one possible design, an auxiliary task branch is set at the end of the convolutional neural network to determine the chromosome polarity of the input image, so as to output the image features that have been corrected based on the chromosome polarity determination result, wherein the chromosome polarity determination refers to determining whether the chromosome is upright or inverted.

[0023] In one possible design, a white-box neural network, ReduNet, is used to optimize the image features of each chromosome monosomy image, resulting in optimized image features for each chromosome monosomy image, including:

[0024] In the white-box neural network ReduNet, the minimum amount L(Z,ε) required to encode features is estimated using the following formula:

[0025]

[0026] In the formula, Z represents the feature space, ε represents the distortion rate, m represents the total number of images of the multiple chromosome monosomy images, d represents the feature dimension, logdet() represents the clustering calculation function, and I = Λ 1 +Λ 2 +…+Λ j +…+Λ N j represents a positive integer less than or equal to N, N represents the total number of chromosome categories, and Λ j ∈R m×m And R represents the diagonal matrix corresponding to the j-th chromosome category among N chromosome categories. m×m Let Λ be a real square matrix with m×m elements, in the diagonal matrix. j The diagonal element Λ in j(i,i) represents the probability that the i-th chromosome monomorphic image among the plurality of chromosome monomorphic images belongs to the j-th chromosome category, where i represents a positive integer less than or equal to m, and T represents the matrix transpose symbol;

[0027] In the white-box neural network ReduNet, the total spatial compactness index of the image features of each chromosome monomorphic image is measured using the following formula.

[0028] In the white-box neural network ReduNet, the coding rate R of the category feature subspace of the N chromosome categories is calculated using the following formula. C (Z,ε|Λ):

[0029]

[0030] In the formula, Λ represents a set of diagonal matrices and has tr() represents the function to sum the diagonal elements; in the white-box neural network ReduNet, the feature space is progressively transformed and updated using the gradient ascent method. Until the objective function ΔR = R(Z,ε) - R C (Z,ε|Λ) is maximized, where k represents the number of iterations, and Z k Z represents the feature space corresponding to the k-th step. k+1 Let represent the feature space corresponding to the (k+1)th step, and η represent a learning rate variable that is greater than zero. The gradient is represented and calculated according to the following formula:

[0031]

[0032] In the formula, E k =α(I+αZ) k Z k T ) -1 ,

[0033] In the white-box neural network ReduNet, image optimization features of each chromosome monomorphic image are obtained based on the final feature space.

[0034] In a second aspect, a chromosome image recognition device is provided, comprising an image acquisition module, an instance segmentation module, an image feature extraction module, an image feature optimization module, and a classification result determination module that are sequentially connected in communication.

[0035] The image acquisition module is used to acquire real microscopic images of cell nuclei;

[0036] The instance segmentation module is used to perform chromosome instance segmentation processing on the real microscopic image to obtain multiple independent chromosome monosomal images.

[0037] The image feature extraction module is used to perform feature extraction processing on each chromosome monomorph image in the plurality of chromosome monomorph images to obtain the image features of each chromosome monomorph image.

[0038] The image feature optimization module is used to optimize the image features of each chromosome monosomy image using a white-box neural network ReduNet to obtain the optimized image features of each chromosome monosomy image.

[0039] The classification result determination module is used to input the corresponding image optimization features into a pre-trained chromosome classifier for each chromosome monosomy image, and output the corresponding chromosome monosomy classification result.

[0040] Thirdly, the present invention provides a computer device comprising a memory, a processor, and a transceiver connected in sequence for communication, wherein the memory is used to store a computer program, the transceiver is used to send and receive data, and the processor is used to read the computer program and execute the chromosome image recognition method as described in the first aspect or any possible design in the first aspect.

[0041] Fourthly, the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, perform the chromosome image recognition method as described in the first aspect or any possible design of the first aspect.

[0042] Fifthly, the present invention provides a computer program product containing instructions that, when executed on a computer, cause the computer to perform the chromosome image recognition method as described in the first aspect or any possible design in the first aspect.

[0043] The beneficial effects of the above scheme are:

[0044] (1) This invention creatively provides a new scheme for classifying and identifying chromosomes based on image feature optimization results. After obtaining multiple independent chromosome monosomy images, image feature extraction is performed on each chromosome monosomy image to obtain corresponding image features. Then, based on the image features, a white-box neural network ReduNet is used for optimization to obtain corresponding optimized image features. Finally, the optimized image features are input into a pre-trained chromosome classifier to output the corresponding chromosome monosomy classification result. By performing a vector optimization of the embedding representation after the image feature extraction network, the accuracy of the final chromosome classification and recognition can be significantly improved. Through corresponding experiments, it has been found that common evaluation indicators such as accuracy, precision and F-value can reach more than 92.39%, or even 97.43%. It also has the advantages of not requiring manual recognition and fast automatic recognition, which is convenient for practical application and promotion. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a schematic flowchart of the chromosome image recognition method provided in the embodiments of this application.

[0047] Figure 2 This is a schematic diagram of the structure of the chromosome image recognition device provided in the embodiments of this application.

[0048] Figure 3 A schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in conjunction with the accompanying drawings and descriptions of the embodiments or the prior art. Obviously, the following description of the structure of the accompanying drawings is only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0050] It should be understood that although the terms "first" and "second", etc., may be used herein to describe various objects, these objects should not be limited by these terms. These terms are only used to distinguish one object from another. For example, the first object may be referred to as the second object, and similarly, the second object may be referred to as the first object, without departing from the scope of the exemplary embodiments of the invention.

[0051] It should be understood that the term "and / or" that may appear in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, or A and B exist simultaneously. Another example is A, B and / or C, which can mean that any one of A, B, and C or any combination thereof exists. The term " / and" that may appear in this document describes another relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone or A and B exist simultaneously. In addition, the character " / " that may appear in this document generally indicates that the related objects before and after it are in an "or" relationship.

[0052] Example:

[0053] like Figure 1 As shown, the chromosome image recognition method provided in the first aspect of this embodiment can be executed, but is not limited to, by a computer device with certain computing resources, such as a platform server, a personal computer (PC, referring to a multi-purpose computer of a size, price, and performance suitable for personal use; desktop computers, laptops, mini-laptops, tablets, and ultrabooks are all personal computers), a smartphone, a personal digital assistant (PDA), or a wearable device. Figure 1 As shown, the chromosome image recognition method may include, but is not limited to, the following steps S1 to S5.

[0054] S1. Obtain a true microscopic image of the cell nucleus.

[0055] In step S1, the actual microscopic image is the electron microscopic imaging result of the material within the cell nucleus, preferably an electron microscopic imaging result acquired during the mid-stage of cell culture that does not exhibit chromosome overlap. Specifically, the actual microscopic image can be a high-power image, such as a 63x oil immersion image.

[0056] S2. Perform chromosome instance segmentation processing on the real microscopic image to obtain multiple independent chromosome monosomy images.

[0057] Before step S2, to ensure accurate chromosome instance segmentation, preferably, after acquiring a real microscopic image of the cell nucleus and before performing chromosome instance segmentation on the real microscopic image, the method further includes, but is not limited to: first, performing image filtering on the real microscopic image to obtain a real microscopic image with image noise removed; then, preprocessing the real microscopic image with image noise removed to obtain a real microscopic image to be segmented. The preprocessing includes, but is not limited to, image size uniform adjustment processing and pixel value standardization processing. Pixel value standardization processing refers to uniformly calibrating the image color distribution and / or exposure. The purpose of the aforementioned image filtering processing is to filter out impurity imaging such as cell nucleus boundary imaging and cytoplasm imaging, so as to obtain the real microscopic image with image noise removed containing only chromosomes and background color. Furthermore, the specific processes of the image filtering processing, the image size uniform adjustment processing, and the pixel value standardization processing are existing technologies and will not be described in detail here.

[0058] In step S2, specifically, the real microscopic image is subjected to chromosome instance segmentation processing to obtain multiple independent chromosome monosomy images. This includes, but is not limited to, using a deep learning-based instance segmentation algorithm to segment the real microscopic image into chromosome instances, resulting in multiple independent chromosome monosomy images. The aforementioned instance segmentation algorithm is an existing algorithm, such as an instance segmentation algorithm based on the Mask R-CNN instance segmentation network (specifically, but not limited to, CascadeMask R-CNN). More specifically, the real microscopic image is segmented into chromosome instances using a deep learning-based instance segmentation algorithm to obtain multiple independent chromosome monosomy images. This includes, but is not limited to, using a deep learning-based instance segmentation algorithm to first identify chromosome boundary lines in the real microscopic image, and then segmenting the real microscopic image into multiple independent chromosome monosomy images based on the chromosome boundary lines. For example, an auxiliary task can be added to CascadeMask R-CNN and training to calculate edge loss can be introduced to enable it to identify chromosome boundary lines.

[0059] S3. Perform feature extraction processing on each chromosome monomorphic image in the plurality of chromosome monomorphic images to obtain the image features of each chromosome monomorphic image.

[0060] In step S3, specifically, expert visual feature extraction processing is performed on each chromosome monomorphic image among the multiple chromosome monomorphic images to obtain the expert visual features of each chromosome monomorphic image. This includes, but is not limited to, inputting the corresponding image into an image feature extraction model that has undergone pre-supervised recognition training based on a convolutional neural network and a ternary loss function (i.e., based on a convolutional neural network or based on a convolutional neural network and a ternary loss function), and outputting the corresponding image features. The aforementioned convolutional neural network is a type of feedforward neural network that includes convolutional computation and has a deep structure. It is one of the representative algorithms of deep learning, such as the ResNet-50 network. The aforementioned Triplet Loss function was proposed by Google in its 2015 FaceNet paper. (Triplet Loss is defined as minimizing the distance between the anchor point and positive samples with the same identity, and minimizing the distance between the anchor point and negative samples with different identities. The goal of Triplet Loss is to spatially minimize the proximity of features with the same label and spatially minimize the distance between features with different labels. Furthermore, to prevent feature aggregation into a very small space, for two positive examples and one negative example of the same class, the negative example should be at least a preset threshold further away from the positive example.) It is commonly used in image recognition network models. Therefore, the specific training process of the image feature extraction model can be based on a suitable amount of labeled samples and pre-trained using existing supervised recognition training methods. Furthermore, considering that upright and inverted chromosomes would result in the opposite order of the observed features, in order to enable the image feature extraction model to distinguish polarity and improve the accuracy of the feature extraction results, preferably, an auxiliary task branch for determining the chromosome polarity of the input image is set at the end of the convolutional neural network (chromosome polarity annotation needs to be added to the samples before model training), so as to output the image features corrected based on the chromosome polarity determination result, wherein the chromosome polarity determination refers to determining whether the chromosome is upright or inverted; the specific method of correcting the image features based on the chromosome polarity determination result is a conventional method.

[0061] S4. The image features of each chromosome monosomy image are optimized using a white-box neural network ReduNet to obtain the optimized image features of each chromosome monosomy image.

[0062] In step S4, the white-box neural network ReduNet is a deep neural network naturally constructed by deriving the gradient of the maximum coding rate decay (MCR2) target. Each layer of its network can be explained by mathematical operations, and the network parameters are explicitly constructed layer by layer through forward propagation without the need for backpropagation. Therefore, the aforementioned image feature optimization can ensure that the subsequent classifier is more accurate in classification. In order to make the features of the input classifier have the following optimized characteristics: (1) the image features of independent chromosomes of the same type are closer and more compact in the embedding representation space; (2) the image features of independent chromosomes of different types are further apart in the embedding representation space, preferably, the white-box neural network ReduNet is used to optimize the image features of each chromosome monosomal image to obtain the optimized image features of each chromosome monosomal image, including but not limited to the following steps S41 to S45.

[0063] S41. In the white-box neural network ReduNet, the minimum amount L(Z,ε) required to encode features is estimated using the following formula:

[0064]

[0065] In the formula, Z represents the feature space, ε represents the distortion rate, m represents the total number of images of the multiple chromosome monosomy images, d represents the feature dimension, logdet() represents the clustering calculation function, and I = Λ 1 +Λ 2 +…+Λ j +…+Λ N j represents a positive integer less than or equal to N, N represents the total number of chromosome categories, and Λ j ∈R m×m And R represents the diagonal matrix corresponding to the j-th chromosome category among N chromosome categories. m×m Let Λ be a real square matrix with m×m elements, in the diagonal matrix. j The diagonal element Λ in j (i,i) represents the probability that the i-th chromosome monomorphic image among the multiple chromosome monomorphic images belongs to the j-th chromosome category, where i represents a positive integer less than or equal to m, and T represents the matrix transpose symbol.

[0066] In step S41, if image recognition is performed on human chromosomes, the N chromosome categories may specifically include 22 autosomes, the X chromosome, and the Y chromosome.

[0067] S42. In the white-box neural network ReduNet, the total spatial compactness index value of the image features of each chromosome monomorphic image is measured using the following formula.

[0068] S43. In the white-box neural network ReduNet, the coding rate R of the category feature subspace of the N chromosome categories is calculated using the following formula. C (Z,ε|Λ):

[0069]

[0070] In the formula, Λ represents a set of diagonal matrices and has tr() is a function that calculates the sum of the diagonal elements;

[0071] S44. In the white-box neural network ReduNet, the feature space is gradually transformed and updated using the gradient ascent method: Until the objective function ΔR = R(Z,ε) - R C (Z,εΛ) is maximized, where k represents the number of iterations, and Z... k Z represents the feature space corresponding to the k-th step. k+1 Let represent the feature space corresponding to the (k+1)th step, and η represent a learning rate variable that is greater than zero. The gradient is represented and calculated according to the following formula:

[0072]

[0073] In the formula, E k =α(I+αZ) k Z k T ) -1 ,

[0074] In step S44, since the objective function ΔR = R(Z,ε) - R C To increase the total space, make the intra-class spaces (i.e., the abbreviation for class feature subspaces) more compact, and increase the distance between pairs of intra-class spaces, the feature space needs to be gradually transformed and updated until the optimal space transformation result is achieved. Since the objective function is maximized and is concave, the existing gradient ascent method can be used for optimization.

[0075] S45. In the white-box neural network ReduNet, the image optimization features of each chromosome monomorphic image are obtained based on the final feature space.

[0076] In step S45, the specific means of obtaining the image optimization features of each chromosome monomorphic image based on the final obtained feature space are existing conventional means, which will not be described in detail here.

[0077] S5. For each chromosome monosomy image, the corresponding image optimization features are input into a pre-trained chromosome classifier, and the corresponding chromosome monosomy classification result is output.

[0078] In step S5, the chromosome classifier needs to be obtained in advance by routinely processing the initial classifier with the input sample data corresponding to different chromosome categories. The sample data package contains independent chromosome images and image features / image optimization features.

[0079] The experimental results obtained based on the chromosome image recognition method described in steps S1 to S5 are shown in Table 1 below:

[0080] Table 1. Comparison of experimental results between this embodiment and several existing chromosome image recognition technologies.

[0081]

[0082] As shown in Table 1 above, compared with existing chromosome image recognition technologies based on ResNet-50 networks, the basic scheme without the ternary loss function and the optimal scheme of this embodiment can both achieve significant improvements in accuracy and precision. In particular, the optimal scheme based on ResNet-50+Triplet-Loss+ReduNet in this embodiment can achieve very ideal technical results in all four main indicators.

[0083] Therefore, based on the chromosome image recognition method described in steps S1 to S5 above, a new scheme for classifying and recognizing chromosomes based on image feature optimization results is provided. That is, after obtaining multiple independent chromosome monosomy images, image feature extraction processing is first performed on each chromosome monosomy image to obtain the corresponding image features. Then, based on the image features, a white-box neural network ReduNet is used for optimization processing to obtain the corresponding optimized image features. Finally, the optimized image features are input into a pre-trained chromosome classifier to output the corresponding chromosome monosomy classification result. In this way, by performing a vector optimization of the embedding representation after the image feature extraction network, the accuracy of the final chromosome classification and recognition can be significantly improved. Corresponding experiments have shown that common evaluation indicators such as accuracy, precision, and F-value can reach more than 92.39%, or even 97.43%. It also has the advantages of not requiring manual recognition and fast automatic recognition, which is convenient for practical application and promotion.

[0084] like Figure 2 As shown, the second aspect of this embodiment provides a virtual device for implementing the chromosome image recognition method described in the first aspect, including an image acquisition module, an instance segmentation module, an image feature extraction module, an image feature optimization module, and a classification result determination module that are sequentially connected in communication.

[0085] The image acquisition module is used to acquire real microscopic images of cell nuclei;

[0086] The instance segmentation module is used to perform chromosome instance segmentation processing on the real microscopic image to obtain multiple independent chromosome monosomal images.

[0087] The image feature extraction module is used to perform feature extraction processing on each chromosome monomorph image in the plurality of chromosome monomorph images to obtain the image features of each chromosome monomorph image.

[0088] The image feature optimization module is used to optimize the image features of each chromosome monosomy image using a white-box neural network ReduNet to obtain the optimized image features of each chromosome monosomy image.

[0089] The classification result determination module is used to input the corresponding image optimization features into a pre-trained chromosome classifier for each chromosome monosomy image, and output the corresponding chromosome monosomy classification result.

[0090] The working process, working details and technical effects of the aforementioned device provided in the second aspect of this embodiment can be found in the chromosome image recognition method described in the first aspect, and will not be repeated here.

[0091] like Figure 3 As shown, the third aspect of this embodiment provides a computer device for implementing the chromosome image recognition method as described in the first aspect, including a memory, a processor, and a transceiver connected in sequence. The memory stores a computer program, the transceiver transmits and receives data, and the processor reads the computer program and executes the chromosome image recognition method as described in the first aspect. Specifically, the memory may include, but is not limited to, random-access memory (RAM), read-only memory (ROM), flash memory, first-in-first-out (FIFO) memory, and / or first-in-last-out (FILO) memory, etc.

[0092] The working process, working details and technical effects of the aforementioned computer device provided in the third aspect of this embodiment can be found in the chromosome image recognition method described in the first aspect, and will not be repeated here.

[0093] This fourth aspect of the embodiment provides a computer-readable storage medium storing instructions comprising the chromosome image recognition method as described in the first aspect. Specifically, the computer-readable storage medium stores instructions that, when executed on a computer, perform the chromosome image recognition method as described in the first aspect. The computer-readable storage medium refers to a data storage medium, and may include, but is not limited to, floppy disks, optical disks, hard disks, flash memory, USB flash drives, and / or Memory Sticks. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0094] The working process, working details and technical effects of the aforementioned computer-readable storage medium provided in the fourth aspect of this embodiment can be found in the chromosome image recognition method described in the first aspect, and will not be repeated here.

[0095] This fifth aspect of the embodiment provides a computer program product containing instructions that, when executed on a computer, cause the computer to perform the chromosome image recognition method as described in the first aspect. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device.

[0096] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A chromosome image recognition method, characterized in that, include: Obtain true microscopic images of cell nuclei; The real microscopic image is segmented into chromosome instances to obtain multiple independent images of chromosome monosomy. Feature extraction processing is performed on each chromosome monomorph image in the plurality of chromosome monomorph images to obtain the image features of each chromosome monomorph image; The image features of each chromosome monosomy image are optimized using a white-box neural network, ReduNet, to obtain optimized image features for each chromosome monosomy image. Specifically, the minimum amount L(Z,ε) required for encoding features is estimated in the white-box neural network ReduNet using the following formula: In the formula, Z represents the feature space, ε represents the distortion rate, m represents the total number of images of the multiple chromosome monosomy images, d represents the feature dimension, logdet() represents the clustering calculation function, and I = Λ 1 +Λ 2 +…+Λ j +…+Λ N j represents a positive integer less than or equal to N, N represents the total number of chromosome categories, and Λ j ∈R m×m And R represents the diagonal matrix corresponding to the j-th chromosome category among N chromosome categories. m×m Let Λ be a real square matrix with m×m elements, in the diagonal matrix. j The diagonal element Λ in j (i,i) represents the probability that the i-th chromosome monomorphic image among the multiple chromosome monomorphic images belongs to the j-th chromosome category, where i represents a positive integer less than or equal to m, and T represents the matrix transpose. In the white-box neural network ReduNet, the total spatial compactness index of the image features of each chromosome monomorphic image is measured using the following formula. In the white-box neural network ReduNet, the coding rate R of the category feature subspace of the N chromosome categories is calculated using the following formula. C (Z,ε|Λ): In the formula, Λ represents a set of diagonal matrices and has tr() represents the function to sum the diagonal elements; in the white-box neural network ReduNet, the feature space is progressively transformed and updated using the gradient ascent method. Until the objective function ΔR = R(Z,ε) - R C (Z,ε|Λ) is maximized, where k represents the number of iterations, and Z k Z represents the feature space corresponding to the k-th step. k+1 Let represent the feature space corresponding to the (k+1)th step, and η represent a learning rate variable that is greater than zero. The gradient is represented and calculated according to the following formula: In the formula, E k =α(I+αZ) k Z k T ) -1 , In the white-box neural network ReduNet, image optimization features of each chromosome monomorphic image are obtained based on the final feature space. For each chromosome monosomy image, the corresponding optimized image features are input into a pre-trained chromosome classifier, and the corresponding chromosome monosomy classification result is output.

2. The chromosome image recognition method according to claim 1, characterized in that, After acquiring a true microscopic image of the cell nucleus and before performing chromosome instance segmentation on the true microscopic image, the method further includes: The real microscopic image is subjected to image filtering processing to obtain a real microscopic image with image noise removed; The real microscopic image with noise removed is preprocessed to obtain the real microscopic image to be segmented. The preprocessing includes image size uniform adjustment and pixel value standardization. The pixel value standardization refers to uniformly calibrating the image color distribution and / or exposure.

3. The chromosome image recognition method according to claim 1, characterized in that, The real microscopic image is segmented into chromosome instances to obtain multiple independent images of chromosome monosomy, including: A deep learning-based instance segmentation algorithm is used to segment the real microscopic image into chromosome instances, resulting in multiple independent images of chromosome monosomy.

4. The chromosome image recognition method according to claim 3, characterized in that, A deep learning-based instance segmentation algorithm is used to segment the real microscopic image into chromosome instances, resulting in multiple independent images of chromosome monosomy, including: A deep learning-based instance segmentation algorithm is used to first identify chromosome boundary lines in the real microscopic image, and then segment the real microscopic image into multiple independent chromosome monomorphic images based on the chromosome boundary lines.

5. The chromosome image recognition method according to claim 1, characterized in that, Feature extraction processing is performed on each chromosome monomorph image in the plurality of chromosome monomorph images to obtain the image features of each chromosome monomorph image, including: For each chromosome monomorphic image among the multiple chromosome monomorphic images, the corresponding image is input into an image feature extraction model that is based on a convolutional neural network and a ternary loss function and has been pre-trained in a supervised manner, and the corresponding image features are output.

6. The chromosome image recognition method according to claim 5, characterized in that, An auxiliary task branch is set at the end of the convolutional neural network to determine the chromosome polarity of the input image, so as to output the image features that have been corrected based on the chromosome polarity determination result. The chromosome polarity determination refers to determining whether the chromosome is upright or inverted.

7. A chromosome image recognition device, characterized in that, It includes an image acquisition module, an instance segmentation module, an image feature extraction module, an image feature optimization module, and a classification result determination module, all connected in sequence. The image acquisition module is used to acquire real microscopic images of cell nuclei; The instance segmentation module is used to perform chromosome instance segmentation processing on the real microscopic image to obtain multiple independent chromosome monosomal images. The image feature extraction module is used to perform feature extraction processing on each chromosome monomorph image in the plurality of chromosome monomorph images to obtain the image features of each chromosome monomorph image. The image feature optimization module is used to optimize the image features of each chromosome monosomy image using a white-box neural network, ReduNet, to obtain optimized image features for each chromosome monosomy image. Specifically, it includes estimating the minimum amount L(Z,ε) required for encoding features in the white-box neural network ReduNet using the following formula: In the formula, Z represents the feature space, ε represents the distortion rate, m represents the total number of images of the multiple chromosome monosomy images, d represents the feature dimension, logdet() represents the clustering calculation function, and I = Λ 1 +Λ 2 +…+Λ j +…+Λ N j represents a positive integer less than or equal to N, N represents the total number of chromosome categories, and Λ j ∈R m×m And R represents the diagonal matrix corresponding to the j-th chromosome category among N chromosome categories. m×m Let Λ be a real square matrix with m×m elements, in the diagonal matrix. j The diagonal element Λ in j (i,i) represents the probability that the i-th chromosome monomorphic image among the multiple chromosome monomorphic images belongs to the j-th chromosome category, where i represents a positive integer less than or equal to m, and T represents the matrix transpose. In the white-box neural network ReduNet, the total spatial compactness index of the image features of each chromosome monomorphic image is measured using the following formula. In the white-box neural network ReduNet, the coding rate R of the category feature subspace of the N chromosome categories is calculated using the following formula. C (Z,ε|Λ): In the formula, Λ represents a set of diagonal matrices and has tr() represents the function to sum the diagonal elements; in the white-box neural network ReduNet, the feature space is progressively transformed and updated using the gradient ascent method. Until the objective function ΔR = R(Z,ε) - R C (Z,ε|Λ) is maximized, where k represents the number of iterations, and Z k Z represents the feature space corresponding to the k-th step. k+1 Let represent the feature space corresponding to the (k+1)th step, and η represent a learning rate variable that is greater than zero. The gradient is represented and calculated according to the following formula: In the formula, E k =α(I+αZ) k Z k T ) -1 , In the white-box neural network ReduNet, image optimization features of each chromosome monomorphic image are obtained based on the final feature space. The classification result determination module is used to input the corresponding image optimization features into a pre-trained chromosome classifier for each chromosome monosomy image, and output the corresponding chromosome monosomy classification result.

8. A computer device, characterized in that, The device includes a memory, a processor, and a transceiver that are sequentially and communicatively connected, wherein the memory is used to store a computer program, the transceiver is used to send and receive data, and the processor is used to read the computer program and execute the chromosome image recognition method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that... The computer-readable storage medium stores instructions that, when executed on a computer, perform the chromosome image recognition method as described in any one of claims 1 to 6.

Images