CN114998332B

CN114998332B - Method and device for determining karyotype abnormal cells

Info

Publication number: CN114998332B
Application number: CN202210918890.9A
Authority: CN
Inventors: 邝英兰; 何妙燕; 范献军; 叶莘; 陈成苑; 陈鑫
Original assignee: Zhuhai Hengqin Shengao Yunzhi Technology Co ltd
Current assignee: Zhuhai Hengqin Shengao Yunzhi Technology Co ltd
Priority date: 2022-08-02
Filing date: 2022-08-02
Publication date: 2022-11-22
Anticipated expiration: 2042-08-02
Also published as: CN114998332A

Abstract

The application provides a method and a device for determining karyotype abnormal cells, wherein the method comprises the following steps: the method comprises the steps of determining a target karyotype association characteristic value corresponding to a cell to be identified based on a nuclear microscopic image of the cell to be identified, obtaining a candidate karyotype association characteristic set by screening from an initial karyotype association characteristic set based on a Relieff algorithm and an mRMR algorithm, secondarily screening and determining the candidate karyotype association characteristic set based on the mRMR algorithm and a backward selection method, inputting the target karyotype association characteristic value corresponding to the cell to be identified into a trained karyotype abnormal cell determination model, outputting a karyotype abnormal cell identification result, and ensuring the efficiency and accuracy of karyotype abnormal cell identification.

Description

Method and device for determining karyotype abnormal cells

Technical Field

The application relates to the technical field of medical image processing, in particular to a method and a device for determining karyotype abnormal cells.

Background

The target cells currently used for tumor-related cell interpretation in blood are monocytes and non-granulocytes, and therefore, before tumor-related cell interpretation is performed, non-interpretation target cells in a blood sample need to be filtered out to eliminate the influence on the accuracy of interpretation results. However, due to the influence of environmental factors, some of the non-interpretation target cells in the blood sample cannot be completely filtered out.

In order to avoid the influence of the non-interpretation target cells on the interpretation result, the prior art generally determines whether the cell karyotype is abnormal by manually observing and rejecting the cell karyotype as the non-interpretation target cells when the tumor-related cells are interpreted. However, this method is time-consuming and labor-consuming, and has low accuracy, and cannot ensure the efficiency and accuracy of tumor-related cell interpretation.

Disclosure of Invention

The application provides a method and a device for identifying karyotype abnormal cells, which are used for improving the identification efficiency and accuracy of the karyotype abnormal cells and ensuring the interpretation efficiency and accuracy of tumor-related cells.

The application provides a method for determining karyotype abnormal cells, which comprises the following steps:

determining a target karyotype association characteristic value corresponding to the cell to be identified based on a cell nucleus microscopic image corresponding to the cell to be identified; the target karyotype correlation characteristics are determined after screening a candidate karyotype correlation characteristic set from an initial karyotype correlation characteristic set based on a Relieff algorithm and a maximum correlation minimum redundancy (mRMR) algorithm, and performing secondary screening on the candidate karyotype correlation characteristic set based on an mRMR algorithm and a backward selection method; the initial karyotype associated feature set comprises a morphological feature subset, a texture feature subset, a color feature subset and an edge feature subset, and the candidate karyotype associated feature set comprises at least one candidate morphological feature, at least one candidate texture feature, at least one candidate color feature and at least one candidate edge feature;

inputting the target karyotype association characteristic value corresponding to the cell to be identified into a trained karyotype abnormal cell determination model, and outputting a karyotype abnormal cell identification result corresponding to the target karyotype association characteristic value;

the karyotype abnormal cell determination model is obtained by training a target karyotype association characteristic value sample and a predetermined karyotype abnormal cell identification result label.

According to the method for determining the karyotype abnormal cell provided by the application, the step of acquiring the cell nucleus microscopic image corresponding to the cell to be identified comprises the following steps:

segmenting the cell microscopic image corresponding to the blood sample based on a preset segmentation network to obtain an initial cell nucleus microscopic image corresponding to the blood sample;

determining candidate cell nucleuses with over segmentation based on the position relation of each cell nucleus and the number of fluorescence signal points in the cell nucleus in initial cell nucleus microscopic images corresponding to different fluorescence channels of the blood sample, and correcting the candidate cell nucleuses based on a convex hull detection algorithm and a convex defect algorithm to obtain cell nucleus corrected microscopic images corresponding to the blood sample;

and taking the cell nucleus correction microscopic image corresponding to the blood sample as the cell nucleus microscopic image corresponding to the cell to be identified.

According to the method for determining the karyotype abnormal cell, the step of correcting the candidate cell nucleus based on the convex hull detection algorithm and the convex defect algorithm specifically comprises the following steps:

determining a cell nucleus set formed by target cell nuclei of which the sum of the number of fluorescence signal points is a preset value in the candidate cell nuclei;

merging the target cell nucleuses in the cell nucleus set to obtain a corresponding corrected cell nucleus;

determining a contour point set which is farthest away from a convex hull on the contour of the corrected cell nucleus based on a convex hull detection algorithm and a convex defect algorithm, and determining a target contour point in the contour point set based on a preset distance threshold, wherein the target contour point is a contour point in the contour point set, and the distance between the contour point set and the convex hull exceeds the preset distance threshold;

determining whether the corrected cell nucleus is correct based on the number of the target contour points, and deleting the incorrect corrected cell nucleus.

According to the method for determining the karyotype abnormal cells, the candidate karyotype association feature set is obtained by screening from the initial karyotype association feature set based on the Relieff algorithm and the maximum correlation minimum redundancy (mRMR) algorithm, and the method specifically comprises the following steps:

respectively performing feature screening on each feature subset in the initial karyotype associated feature set based on a Relieff algorithm and an mrMR algorithm to obtain at least one candidate morphological feature, at least one candidate texture feature, at least one candidate color feature and at least one candidate edge feature;

and combining the candidate morphological characteristics, the candidate texture characteristics, the candidate color characteristics and the candidate edge characteristics to obtain the candidate karyotype association characteristic set.

According to the method for determining the karyotype abnormal cells, the feature screening is respectively performed on each feature subset in the initial karyotype-associated feature set based on the ReliefF algorithm and the mRMR algorithm, and the method specifically includes:

determining a target feature subset, and acquiring a feature value corresponding to each sample in a training sample set and the target feature subset;

determining the weight corresponding to each feature in the target feature subset through a Relieff algorithm based on the feature value corresponding to each sample in the training sample set and the target feature subset;

screening the features in the target feature subset based on the weight threshold corresponding to the target feature subset, and determining a first feature combination corresponding to the target feature subset;

and determining first average mutual information between each feature and the class C in the first feature combination and second average mutual information between each feature based on an mRMR algorithm, and screening the features in the first feature combination based on the first average mutual information and the second average mutual information to obtain a second feature combination corresponding to the target feature subset.

According to the method for determining karyotypic abnormal cells provided by the application, the features in the second feature combination are the features in the first feature combination, which maximize the difference or quotient between the first average mutual information and the second average mutual information.

According to the method for determining the karyotype abnormal cells, provided by the application, the secondary screening is performed on the candidate karyotype association feature set based on the mRMR algorithm and the backward selection method, and specifically the method comprises the following steps:

re-screening the features in the candidate karyotype associated feature set based on an mRMR algorithm, and determining a potential karyotype associated feature set;

and selecting the features in the potential karyotype association feature set based on a backward selection method, and determining the target karyotype association features.

According to the method for determining the karyotype abnormal cells, the determining, by a ReliefF algorithm, the weight corresponding to each feature in the target feature subset based on the feature value corresponding to each sample in the training sample set and the target feature subset includes:

determining the difference between the characteristic values corresponding to the target sample and the similar nearest neighbor samples with the preset number and the difference between the characteristic values corresponding to the target sample and the different nearest neighbor samples with the preset number based on the characteristic values corresponding to the samples in the training sample set and the target characteristic subset;

and updating the weight corresponding to each feature in the target feature subset based on the difference between the feature values corresponding to the target sample and the similar nearest neighbor samples in the preset number and the difference between the feature values corresponding to the target sample and the different nearest neighbor samples in the preset number until the updating frequency reaches a preset threshold value.

According to the determination method of the karyotype abnormal cell provided by the application, the target karyotype-related characteristic comprises: the cell nucleus perimeter, the cell nucleus area, the cell nucleus circularity, the cell nucleus elongation, the cell nucleus aspect ratio, the cell nucleus rectangularity, the cell nucleus firmness, the cell nucleus orientation degree, the cell nucleus long axis, the cell nucleus short axis, the cell nucleus microscopic image sharpness, the cell nucleus microscopic image total signal intensity, the cell nucleus microscopic image average signal intensity, the cell nucleus microscopic image signal intensity variance and the cell nucleus microscopic image corresponding Zernike moment.

The present application also provides an apparatus for determining a karyotype abnormal cell, including:

the target karyotype association characteristic value determining module is used for determining a target karyotype association characteristic value corresponding to the cell to be identified based on the cell nucleus microscopic image corresponding to the cell to be identified; the target karyotype correlation characteristics are determined after screening a candidate karyotype correlation characteristic set from an initial karyotype correlation characteristic set based on a Relieff algorithm and a maximum correlation minimum redundancy (mRMR) algorithm, and performing secondary screening on the candidate karyotype correlation characteristic set based on an mRMR algorithm and a backward selection method; the initial karyotype associated feature set comprises a morphological feature subset, a texture feature subset, a color feature subset and an edge feature subset, and the candidate karyotype associated feature set comprises at least one candidate morphological feature, at least one candidate texture feature, at least one candidate color feature and at least one candidate edge feature;

the karyotype abnormal cell determination module is used for inputting the target karyotype associated characteristic value corresponding to the cell to be identified into the trained karyotype abnormal cell determination model and outputting a karyotype abnormal cell identification result corresponding to the target karyotype associated characteristic value;

the karyotype abnormal cell determination model is obtained by training a target karyotype correlation characteristic value sample and a predetermined karyotype abnormal cell identification result label.

The present application further provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method for acquiring the karyotype abnormal cell determination described in any one of the above methods when executing the program.

The present application also provides a non-transitory computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps of the method for determining a karyotypically abnormal cell as described in any one of the above.

The present application also provides a computer program product comprising a computer program, which when executed by a processor, implements the steps of the method for determining karyotypic abnormal cells as described in any one of the above.

The method and the device for determining the abnormal karyotype cells provided by the application determine a target karyotype association feature value corresponding to the cells to be identified based on a microscopic nuclear image corresponding to the cells to be identified, the target karyotype association feature is determined by screening a candidate karyotype association feature set from an initial karyotype association feature set based on a Relieff algorithm and a maximum correlation minimum redundancy mRMR algorithm, and then secondary screening is performed on the candidate karyotype association feature set based on an mRMR algorithm and a backward selection method, the initial karyotype association feature set comprises a morphological feature subset, a texture feature subset, a color feature subset and an edge feature subset, the candidate karyotype association feature set comprises at least one candidate morphological feature, at least one candidate texture feature, at least one candidate color feature and at least one candidate edge feature, the target karyotype association feature value corresponding to the cells to be identified is input into a trained abnormal karyotype cell determination model, a karyotype abnormal cell identification result corresponding to the target karyotype association feature value is output, the abnormal karyotype identification result corresponding to the target karyotype association feature value is obtained based on a target karyotype association feature value, the abnormal cell identification result is obtained by automatically determining the efficiency, and the abnormal karyotype identification efficiency is ensured, and the accuracy of the abnormal karyotype identification result is obtained by using the target karyotype identification model.

Drawings

In order to more clearly illustrate the technical solutions in the present application or the prior art, the drawings needed for the description of the embodiments or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.

FIG. 1 is a schematic flow chart of a method for determining karyotypic abnormal cells as provided herein;

FIG. 2 is a schematic illustration of the nuclear morphology of a karyotype normal cell provided herein;

FIG. 3 is a schematic representation of the nuclear morphology of the karyotypically abnormal cells provided herein;

FIG. 4 is a schematic structural diagram of a device for determining karyotype abnormal cells provided herein;

fig. 5 is a schematic structural diagram of an electronic device provided in the present application.

Detailed Description

To make the purpose, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be clearly and completely described below with reference to the drawings in the present application, and it is obvious that the described embodiments are some, but not all embodiments of the present application. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.

In the prior art, when the interpretation of the tumor-related cells in blood is performed, in order to ensure the interpretation accuracy, a preset number of images of the interpretation target cells (i.e., monocytes and granulocytes) need to be obtained through a microscope, and then a corresponding interpretation process is performed. However, due to environmental factors, some of the non-interpretative target cells in the blood sample, including eosinophils, basophils, and neutrophils, cannot be completely filtered out. However, when the microscope collects the cell image, it cannot automatically identify the non-interpretation target cell, and it needs to manually observe and determine whether the karyotype in the cell image is abnormal, and then determine whether the cell image is the non-interpretation target cell, and exclude the cell image when the tumor-related cell is interpreted. The method is time-consuming and labor-consuming, has low accuracy, and cannot ensure the efficiency and accuracy of subsequent tumor-related cell interpretation. Based on this, the embodiment of the application provides a method and a device for determining a karyotype abnormal cell, so as to accurately and efficiently determine the karyotype abnormal cell, and further ensure efficiency and accuracy of tumor-related cell interpretation.

Fig. 1 is a schematic flow chart of a method for determining karyotypic abnormal cells provided in the present application, as shown in fig. 1, the method includes:

step 110, determining a target karyotype association characteristic value corresponding to a cell to be identified based on a cell nucleus microscopic image corresponding to the cell to be identified; the target karyotype association feature is determined after a candidate karyotype association feature set is obtained by screening from an initial karyotype association feature set based on a Relieff algorithm and a maximum correlation minimum redundancy (mRMR) algorithm, and then secondary screening is carried out on the candidate karyotype association feature set based on an mRMR algorithm and a backward selection method; the initial karyotype associated feature set comprises a morphological feature subset, a texture feature subset, a color feature subset and an edge feature subset, and the candidate karyotype associated feature set comprises at least one candidate morphological feature, at least one candidate texture feature, at least one candidate color feature and at least one candidate edge feature.

Specifically, the cell to be identified is a cell corresponding to a blood sample, in the embodiment of the present application, a cell microscopic image corresponding to the blood sample is firstly acquired through a microscope, and the number of the cell to be identified in the cell microscopic image is determined by the resolution of a microscope acquisition camera. The blood sample is stained in advance by a fluorescence in situ hybridization technology, and correspondingly, the cell microscopic image corresponds to four fluorescence channels of red, green, blue and gold. After the cell microscopic image is obtained, the cell nucleus in the cell microscopic image needs to be segmented to obtain a cell nucleus microscopic image corresponding to the cell to be identified. After obtaining the cell nucleus microscopic image corresponding to the cell to be identified, the device for determining the abnormal karyotype cell can determine the target karyotype associated characteristic value corresponding to the cell to be identified based on a predetermined characteristic extraction algorithm.

The prior art artificially determines the karyotype abnormal cells by the following method: the shape of the cell nucleus in the cell nucleus microscopic image is observed by naked eyes to carry out visual judgment, and the method can roughly distinguish granulocytes from non-granulocytes so as to determine the abnormal karyotype cells. Fig. 2 is a schematic diagram of the nuclear morphology of the karyotype normal cells provided by the present application, and fig. 3 is a schematic diagram of the nuclear morphology of the karyotype abnormal cells provided by the present application, as shown in fig. 2-3, the nuclear shape of the karyotype normal cells (i.e., the anucleate cells) is approximately circular, and the nuclear shape of the karyotype abnormal cells (i.e., the granulocytes) is petal-shaped. However, due to the influence of the environment, the resolution of a microscope acquisition camera and other factors, the morphological difference between the nuclei of the normal karyotype cells and the nuclei of the abnormal karyotype cells is often difficult to be directly distinguished by naked eyes, so that the efficiency and the precision of identifying the abnormal karyotype cells in the prior art are extremely low, and the efficiency and the precision of subsequently judging the tumor-related cells are further reduced. Based on this, the embodiment of the application adopts a predetermined feature extraction algorithm to determine the target karyotype associated feature value corresponding to the cell to be identified, and performs automatic identification of the karyotype abnormal cell based on the target karyotype associated feature value corresponding to the cell to be identified.

Specifically, the inventors of the present application found through research that features (i.e., karyotype-associated features) in a nuclear microscopic image that are closely related to nuclear karyotype determination include four types, which are: the cell nucleus morphological characteristics, the cell nucleus texture characteristics, the color characteristics and the edge characteristics; the morphological characteristics of the cell nucleus further comprise: the contour perimeter of the nucleus, the contour area of the nucleus, the contour circularity of the nucleus, the contour elongation of the nucleus, the aspect ratio (namely the ratio of the long axis and the short axis of the minimum circumscribed rectangle of the contour of the nucleus), the rectangle degree (namely the ratio of the minimum circumscribed rectangle of the contour of the nucleus and the area of the connected domain), the firmness (namely the ratio of the area of the connected domain of the contour of the nucleus to the area of the convex hull), the direction degree (namely the angle of the ellipse fitting of the contour of the nucleus), the short axis (namely the long axis of the ellipse fitting of the contour of the nucleus), and the long axis (namely the short axis of the ellipse fitting of the contour of the nucleus); the nuclear texture features further include: total signal intensity (i.e. sum of pixel values in the cell nucleus contour), average signal intensity (i.e. ratio of sum of pixel values in the cell nucleus contour to number of pixel points), signal intensity variance (i.e. variance of pixel values in the cell nucleus contour), inverse difference moment, sharpness, contrast; the color features further include: the background average signal intensity of each fluorescence channel and the background signal intensity variance of each fluorescence channel; the edge feature further includes: zernike moments of orders 2-9.

For the above four types of features, if a full analysis is performed to identify the abnormal karyotype cells, a huge workload is brought to feature extraction, so that the efficiency of identifying the abnormal karyotype cells is greatly reduced. Specifically, in the embodiment of the application, a candidate karyotype-associated feature set is obtained by screening an initial karyotype-associated feature set (i.e., a karyotype-associated feature set jointly composed of the morphological feature subset, the texture feature subset, the color feature subset and the edge feature subset) based on a ReliefF algorithm and a maximum correlation minimum redundancy mRMR algorithm, a preset number of features with the highest correlation with karyotype abnormal cells can be respectively obtained from each feature subset through the ReliefF algorithm, and then redundant features are removed based on the mRMR algorithm, so that the candidate karyotype-associated feature set can be obtained. It should be noted that, in the actual operation process, the features in a certain feature subset may be completely excluded due to the error execution of the ReliefF algorithm and/or the mRMR algorithm, thereby affecting the accuracy of the identification of the karyotype abnormal cells. Based on this, the candidate karyotype associated feature set obtained by screening from the initial karyotype associated feature set based on the ReliefF algorithm and the mRMR algorithm in the embodiments of the present application must be guaranteed to include at least one candidate morphological feature, at least one candidate texture feature, at least one candidate color feature, and at least one candidate edge feature.

After the candidate karyotype associated feature set is determined, secondary screening is carried out on the candidate karyotype associated feature set based on an mRMR algorithm and a backward selection method so as to further remove redundant features among feature subsets, and further screening is carried out to obtain a preset number of target karyotype associated features with high correlation with karyotype abnormal cells.

Step 120, inputting the target karyotype association characteristic value corresponding to the cell to be identified into a trained karyotype abnormal cell determination model, and outputting a karyotype abnormal cell identification result corresponding to the target karyotype association characteristic value;

Specifically, after the target karyotype-associated feature value corresponding to the cell to be recognized is determined, the target karyotype-associated feature value corresponding to the cell to be recognized is input into a trained karyotype abnormal cell determination model, and a karyotype abnormal cell recognition result corresponding to the target karyotype-associated feature value is predicted and output through the karyotype abnormal cell determination model. The karyotype abnormal cell determination model is obtained by training based on a target karyotype correlation characteristic value sample and a predetermined karyotype abnormal cell identification result label. It is understood that the karyotypically abnormal cells are recognized as karyotypically abnormal or karyotypically normal.

Furthermore, after the karyotype abnormal cell identification result is obtained through the karyotype abnormal cell determination model, the embodiment of the application may further determine the proportion of the cells to be identified in the cell microscopic image, which are considered as karyotype abnormal cells, determine the abnormal grade of the cells to be identified based on the proportion of the karyotype abnormal cells in the cell microscopic image, and further determine whether to reacquire the cell microscopic image, that is, determine whether to give up subsequent interpretation operation on the cells to be identified in the current cell microscopic image, reacquire the cell microscopic images of other cells to be identified for subsequent interpretation, and avoid that the efficiency of subsequent tumor-related cell interpretation is affected due to excessive karyotype abnormal cells in the current cell microscopic image.

According to the method provided by the embodiment of the application, a target karyotype association feature value corresponding to a cell to be recognized is determined based on a nuclear microscopic image corresponding to the cell to be recognized, the target karyotype association feature is determined by screening a candidate karyotype association feature set from an initial karyotype association feature set based on a Relieff algorithm and a maximum correlation minimum redundancy mRMR algorithm, and then secondary screening is performed on the candidate karyotype association feature set based on the mRMR algorithm and a backward selection method, the initial karyotype association feature set comprises a morphological feature subset, a texture feature subset, a color feature subset and an edge feature subset, the candidate karyotype association feature set comprises at least one candidate morphological feature, at least one candidate texture feature, at least one candidate color feature and at least one candidate edge feature, the target karyotype association feature value corresponding to the cell to be recognized is input into a trained karyotype abnormal cell determination model, a karyotype abnormal cell identification result corresponding to the target karyotype association feature value is output, the karyotype abnormal cell identification result corresponding to the target karyotype association feature value is obtained based on a target karyotype association feature value sample and a previously determined, and the abnormal cell identification efficiency is ensured, and the accuracy and the abnormal cell identification result of the karyotype is obtained by using a high-type abnormal cell identification efficiency and the abnormal cell identification efficiency is ensured.

Based on the above embodiment, the step of acquiring the cell nucleus microscopic image corresponding to the cell to be identified includes:

Specifically, the conventional cell nucleus segmentation method is to segment the cell nucleus in the cell microscopic image based on a deep neural network (e.g., mask-RCNN, etc.), and this method can accurately segment the cell nucleus of the granulocytic cell. However, the inventor of the present application finds, through research, that since the nucleus of the granulocyte is usually petaloid, the nucleus of the granulocyte is easily segmented into a plurality of different nuclei (i.e., over-segmentation) by using the conventional manner of performing the nucleus segmentation based on the deep neural network, and a nucleus segmentation error will seriously affect the identification of the subsequent abnormal karyotype cells, which results in a reduction in the identification accuracy of the abnormal karyotype cells and a statistical error in the number of cells, and further results in a decision error as to whether to acquire a cell microscopic image again, which affects the efficiency and accuracy of the subsequent tumor-related cell interpretation. Based on this, after segmenting the cell microscopic image corresponding to the blood sample, the embodiment of the present application further determines and corrects the over-segmented cell nucleus, and the specific steps are as follows:

firstly, a cell microscopic image corresponding to the blood sample is segmented based on a predetermined segmentation network (i.e. the aforementioned deep neural network) to obtain an initial cell nucleus microscopic image corresponding to the blood sample. It is understood that the initial nuclear microscopy images correspond to the four different fluorescence channels described above.

After the initial cell nucleus microscopic image corresponding to the blood sample is obtained, determining candidate cell nuclei with over-segmentation based on the position relation of each cell nucleus and the number of fluorescence signal points in the cell nuclei in the initial cell nucleus microscopic image corresponding to different fluorescence channels of the blood sample. It should be noted that if there is an over-segmentation situation in the nuclei of a certain target granulocyte, the nuclei of the corresponding target granulocyte in the initial nucleus microscopic images corresponding to different fluorescence channels of the blood sample will be over-segmented (for example, the nuclei of the target granulocyte are segmented into two sub-nuclei, and the nuclei of the target granulocyte in the initial nucleus microscopic images corresponding to four fluorescence channels will be segmented into two sub-nuclei). In addition, since the over-segmentation is to segment one target cell nucleus into a plurality of cell nuclei, the segmented cell nuclei are adjacent to each other. It is understood that the positional adjacency may be either a direct adjacency (i.e., two cell nucleus outlines are in direct contact) or an indirect adjacency (i.e., multiple cell nuclei are in direct proximity in turn, wherein cell nuclei that are not in direct contact with two outlines are referred to as indirect adjacency). Meanwhile, based on the principle of fluorescence in situ hybridization technology, it is known that for each fluorescence channel, two fluorescence signal points should be included in the same cell nucleus. Based on the principle, if multiple adjacent cell nuclei exist in the same area in the initial cell nucleus microscopic images corresponding to different fluorescence channels of the blood sample, and the sum of the number of fluorescence signal points in the multiple adjacent cell nuclei is 2, the multiple adjacent cell nuclei are over-segmented cell nuclei, and at the moment, the correction can be completed only by merging the multiple adjacent cell nuclei. This judgment method is applicable to the case of over-segmentation of single granulocytes.

However, since there may be cell aggregation during the preparation of the blood sample, there may be a case where the anucleated cells and the nuclei of granulocytes are adjacent in the initial cell nucleus microscopic image corresponding to the blood sample, and at this time, the above-mentioned method for determining the nucleus that is over-segmented cannot accurately determine the target nucleus that is over-segmented, such as adjacent nuclei a, B, and C, where the nucleus a is the nucleus of an anucleated cell and the nuclei B and C are two nuclei over-segmented from granulocytes, it can be understood that there are two fluorescence signal points in the nucleus a, and it is assumed that there are two fluorescence signal points in the nucleus B and no fluorescence signal point in the nucleus C. At this time, it is not possible to determine whether the over-divided cell nuclei are a and C or B and C based on the above determination manner. Meanwhile, in the initial nucleus microscopic image corresponding to the blood sample, there may be a situation that nuclei of two granulocytes are adjacent, and at this time, the above-mentioned method for determining the nucleus excessively segmented cannot accurately determine the nucleus excessively segmented, such as adjacent nuclei D, E, F, and G, where nuclei D and E are two nuclei excessively segmented by granulocyte M, and nuclei F and G are two nuclei excessively segmented by granulocyte N, and it is assumed that there are two fluorescence signal points in both nuclei D and F, and there is no fluorescence signal point in nuclei E and G. At this time, it is not possible to determine whether the over-divided cell nuclei are combined into DE and FG or DG and FE based on the above-described determination method. If the cell judgment of over-segmentation is performed and the cell judgment is combined only based on the sum of the number of fluorescent signal points in the adjacent cell nuclei being 2 at this time, the cell nucleus segmentation error will be further caused. Based on this, the embodiment of the present application first determines that there is an over-segmented candidate nucleus, i.e., the nuclei a-C or D-G with adjacent positions in the foregoing example, based on the position relationship of each nucleus and the number of fluorescence signal points in the nucleus in the initial nucleus microscopic image corresponding to different fluorescence channels of the blood sample, and it can be understood from the foregoing example that if there are a plurality of adjacent nuclei in the same area in the initial nucleus microscopic image corresponding to different fluorescence channels of the blood sample, and the sum of the numbers of fluorescence signal points of at least two nuclei in the plurality of adjacent nuclei is 2, then the plurality of adjacent nuclei will be determined as there is an over-segmented candidate nucleus. After the candidate cell nucleus is determined, the candidate cell nucleus is corrected based on a convex hull detection algorithm and a convex defect algorithm, so that the accuracy of cell nucleus correction can be ensured.

And after correcting all the candidate cell nucleuses with excessive segmentation, obtaining a cell nucleus correction microscopic image corresponding to the blood sample, and taking the cell nucleus correction microscopic image as a cell nucleus microscopic image corresponding to the cell to be identified. It is understood that in the cell nucleus corrected microscopic image, the over-segmented cell nuclei have been corrected, i.e. the cell nuclei in the cell nucleus corrected microscopic image can be identified as accurately segmented cell nuclei. Based on the method, the accuracy of the subsequent identification result of the karyotype abnormal cells can be ensured to the maximum extent.

According to the method provided by the embodiment of the application, the cell microscopic image corresponding to the blood sample is segmented based on a preset segmentation network so as to obtain an initial cell nucleus microscopic image corresponding to the blood sample; determining over-segmented cell nuclei and correcting the over-segmented cell nuclei to obtain cell nucleus corrected microscopic images corresponding to the blood sample based on the position relation of each cell nucleus and the number of fluorescent signal points in the cell nuclei in initial cell nucleus microscopic images corresponding to different fluorescent channels of the blood sample; and taking the cell nucleus correction microscopic image corresponding to the blood sample as the cell nucleus microscopic image corresponding to the cell to be identified, so that the accuracy of cell nucleus segmentation can be ensured, and the accuracy of a subsequent karyotype abnormal cell identification result is further ensured.

Based on the above embodiment, the step of correcting the candidate cell nucleus based on the convex hull detection algorithm and the convex defect algorithm specifically includes:

determining a contour point set which is farthest from a convex hull on the contour of the modified cell nucleus based on a convex hull detection algorithm and a convex defect algorithm, and determining a target contour point in the contour point set based on a preset distance threshold, wherein the target contour point is a contour point in the contour point set, the distance between the contour point set and the convex hull of which exceeds the preset distance threshold;

determining whether the modified cell nuclei are correct based on the number of the target contour points, and deleting incorrect modified cell nuclei.

Specifically, as can be seen from the foregoing embodiment, the preset value is 2, and it can be understood that, since the sum of the number of fluorescence signal points of a plurality of groups of target cell nuclei may exist is 2, the set of cell nuclei may be one or more. In one case (i.e., corresponding to over-segmentation of single granulocytes), the target nuclei in the set of nuclei may be merged to obtain corresponding modified nuclei. However, in the case of a plurality of cell nuclei, it can be known from the foregoing embodiment that it is not possible to accurately determine which cell nucleus set corresponds to the target cell nucleus that is an over-segmented cell nucleus, and therefore, it is necessary to further identify the merged modified cell nucleus (in this case, a plurality of modified cell nuclei are generated), so as to ensure the accuracy of the cell nucleus modification. However, the inventors of the present application found, through research, that due to the specificity of the shape of the nuclei of the granulocytes, there is a significant difference between the contours of the correctly merged corrected nuclei (corresponding to the case where a plurality of nuclei of the same granulocytes over-segmented merge) and the incorrectly merged corrected nuclei (corresponding to the case where the nuclei of the granulocytes over-segmented merge with the nuclei of the agranulocytes and the case where a plurality of nuclei of the granulocytes over-segmented merge across), and the difference can be determined by the set of contour points on the contour of the corrected nuclei that are farthest from the convex hull, that is, if the number of target contour points on the contour of the corrected nuclei that are farthest from the convex hull exceeds the preset distance threshold exceeds the preset number threshold, the corrected nuclei can be determined to be correct. The preset distance threshold is preferably 900 pixels, the preset number threshold is preferably 4, and the preset distance threshold and the preset number threshold may be reasonably adjusted according to an actual application scenario, which is not specifically limited in the embodiment of the present application.

Based on the above, in the embodiment of the application, the contour point set which is farthest away from the convex hull on the contour of each corrected cell nucleus is determined based on the convex hull detection algorithm and the convex defect algorithm, the target contour points in the contour point set are determined based on the preset distance threshold, whether the corrected cell nucleus is correct or not can be determined based on the number of the target contour points, and the corrected cell nucleus which is incorrect is deleted, so that the cell nucleus corrected microscopic image corresponding to the blood sample can be obtained. It is understood that the deleting of the incorrectly corrected cell nucleus refers to deleting a corresponding cell nucleus contour in the cell nucleus microscopic image, which may be implemented by a corresponding algorithm, and the implementation manner of the algorithm is not specifically limited in the embodiments of the present application. As for the convex hull detection algorithm and the convex defect algorithm, which are conventional algorithms in the art, the detailed operation steps are not described herein.

It should also be understood that any existing contour merging algorithm may be used to implement the merging correction process, and this is not particularly limited in this embodiment of the present application.

According to the method provided by the embodiment of the application, a cell nucleus set formed by target cell nuclei is determined, wherein the sum of the number of fluorescence signal points in the candidate cell nuclei is a preset value; merging the target cell nucleuses in the cell nucleus set to obtain a corresponding corrected cell nucleus; determining a contour point set which is farthest from a convex hull on the contour of the modified cell nucleus based on a convex hull detection algorithm and a convex defect algorithm, and determining a target contour point in the contour point set based on a preset distance threshold, wherein the target contour point is a contour point in the contour point set, the distance between the contour point set and the convex hull of which exceeds the preset distance threshold; determining whether the modified cell nuclei are correct based on the number of the target contour points, and deleting incorrect modified cell nuclei. The method can accurately correct the excessively segmented cell nucleus under the condition of cell aggregation, further ensure the accuracy of cell nucleus segmentation and further ensure the accuracy of a subsequent karyotype abnormal cell identification result.

Based on the above embodiment, the screening of the initial karyotype correlation feature set based on the ReliefF algorithm and the maximum correlation minimum redundancy mRMR algorithm to obtain the candidate karyotype correlation feature set specifically includes:

Specifically, in combination with the foregoing embodiments, in order to avoid that all features in a certain feature subset are excluded due to an error in the implementation of the ReliefF algorithm and/or the mRMR algorithm, thereby affecting the accuracy of karyotype abnormal cell identification, it is required to ensure that at least one candidate karyotype-associated feature is retained in each feature subset after screening based on the ReliefF algorithm and the mRMR algorithm. Based on this, in the embodiment of the application, feature screening is performed on each feature subset in the initial karyotype-associated feature set based on a ReliefF algorithm and an mRMR algorithm, so as to obtain at least one candidate morphological feature, at least one candidate texture feature, at least one candidate color feature, and at least one candidate edge feature. Meanwhile, each feature subset is respectively screened, so that the candidate karyotype correlation features corresponding to each feature subset can be guaranteed to be the features with the largest correlation degree and the smallest redundancy with the karyotype abnormal cells, and the efficiency and the accuracy of subsequent karyotype abnormal cell identification are guaranteed.

In the method provided by the embodiment of the application, the screening of the initial karyotype associated feature set based on the ReliefF algorithm and the maximum correlation minimum redundancy mRMR algorithm to obtain the candidate karyotype associated feature set specifically includes: respectively performing feature screening on each feature subset in the initial karyotype associated feature set based on a Relieff algorithm and an mrMR algorithm to obtain at least one candidate morphological feature, at least one candidate texture feature, at least one candidate color feature and at least one candidate edge feature, and combining the candidate morphological feature, the candidate texture feature, the candidate color feature and the candidate edge feature to obtain the candidate karyotype associated feature set, so that the feature screening efficiency is ensured, and the efficiency and the accuracy of subsequent karyotype abnormal cell identification are ensured.

Based on any of the above embodiments, the performing feature screening on each feature subset in the initial karyotype-associated feature set based on the ReliefF algorithm and the mRMR algorithm specifically includes:

Specifically, a training sample set is preset in the embodiment of the present application, where the training sample set includes a first preset number of karyotype normal cell samples and a second preset number of karyotype abnormal cell samples. After a target feature subset is determined, feature values corresponding to samples in the training sample set and the target feature subset are obtained, weights corresponding to features in the target feature subset are determined through a Relieff algorithm based on the feature values corresponding to the samples in the training sample set and the target feature subset, specifically, the sampling frequency m is determined in advance, and the following steps are executed in a circulating mode until the sampling frequency m is reached: randomly taking a sample R from a training sample set, then finding out k adjacent samples of R from a sample set which is similar to the sample R (if the sample set is a karyotype abnormal cell sample, the sample set which is similar to the karyotype abnormal cell sample), finding out k adjacent samples from a sample set which is different from the sample R (if the sample set is a karyotype abnormal cell sample, the sample set which is different from the karyotype normal cell sample), and then updating the weight of each feature through a feature weight calculation formula of a Relieff algorithm based on a feature value corresponding to the same feature of the sample R and the adjacent samples. As for the characteristic weight calculation formula of the ReliefF algorithm, it is well known in the art and will not be described herein.

After determining the weight corresponding to each feature in the target feature subset, screening the features in the target feature subset based on the weight threshold corresponding to the target feature subset, and determining a first feature combination corresponding to the target feature subset. It is to be understood that the features in the first combination of features are features having a weight greater than the weight threshold. It should be noted that, normally, in consideration of ensuring that the screening criteria are unified, the weight threshold corresponding to each feature subset should be the same, but based on the foregoing embodiments, since the ReliefF algorithm may be executed with an error, so that the weight of each feature in the target feature subset is smaller than the weight threshold, in view of ensuring the accuracy of identifying abnormal karyotype cells, in the embodiment of the present application, when the weight of each feature in the target feature subset is smaller than the weight threshold, the weight threshold corresponding to the target feature subset is appropriately reduced, so as to ensure that at least one feature is included in the first feature combination. It can be understood that, since the weights of all the features in the target feature subset are updated every time sampling is performed, even if the ReliefF algorithm is performed with errors, the weights of the features are biased along the same trend (i.e., biased to be larger or smaller), and based on this, even if the weight threshold is adjusted, the features with low correlation are not introduced, so that the accuracy of feature screening is ensured.

After the first feature combination corresponding to the target feature subset is determined, in the embodiment of the present application, first average mutual information between each feature in the first feature combination and the class C (i.e., karyotype abnormal cells) and second average mutual information between each feature are determined based on an mRMR algorithm, and features in the first feature combination are screened based on the first average mutual information and the second average mutual information to obtain a second feature combination corresponding to the target feature subset, so that redundant features in the target feature subset can be deleted based on the second feature combination. As for the calculation formulas of the first average mutual information and the second average mutual information, they are well known in the art and will not be described herein again. It is understood that when there is only one feature in the first feature combination, it will be retained as a second feature combination corresponding to the target feature subset, based on which, it can be ensured that at least one feature is included in the second feature combination, thereby ensuring the accuracy of subsequent karyotype abnormal cell identification.

In the method provided by the embodiment of the present application, the performing feature screening on each feature subset in the initial karyotype-associated feature set based on the ReliefF algorithm and the mRMR algorithm specifically includes: determining a target feature subset, obtaining feature values of samples in a training sample set corresponding to the target feature subset, determining weights corresponding to the features in the target feature subset through a Relieff algorithm based on the feature values of the samples in the training sample set corresponding to the target feature subset, screening the features in the target feature subset based on a weight threshold corresponding to the target feature subset, determining a first feature combination corresponding to the target feature subset, determining first average mutual information between each feature and a category C and second average mutual information between each feature in the first feature combination based on an mRMR algorithm, screening the features in the first feature combination based on the first average mutual information and the second average mutual information to obtain a second feature combination corresponding to the target feature subset, and ensuring the efficiency and accuracy of subsequent karyotype abnormal cell identification.

Based on any of the above embodiments, the feature in the second feature combination is a feature in the first feature combination that maximizes a difference or quotient between the first average mutual information and the second average mutual information.

Specifically, based on the principle of the mRMR algorithm, the feature with the maximum difference or quotient between the first average mutual information and the second average mutual information is the feature with the maximum correlation with the category C and the minimum redundancy, so that the efficiency and accuracy of subsequent karyotype abnormal cell identification can be ensured.

In the method provided by the embodiment of the present application, the feature in the second feature combination is a feature in the first feature combination that maximizes the difference or quotient between the first average mutual information and the second average mutual information, which can ensure that the correlation between the feature in the second feature combination and the abnormal karyotype cells is maximized and the redundancy is minimized, thereby ensuring the efficiency and accuracy of identifying the subsequent karyotype abnormal cells.

Based on any of the above embodiments, the secondary screening of the candidate karyotype-associated feature set based on the mRMR algorithm and the backward selection method specifically includes:

re-screening the features in the candidate karyotype associated feature set based on an mRMR algorithm to determine a potential karyotype associated feature set;

Specifically, based on the foregoing embodiment, it can be known that, in the candidate karyotype-associated feature set obtained by screening from the initial karyotype-associated feature set based on the ReliefF algorithm and the mRMR algorithm, the candidate feature (i.e., the second feature combination) corresponding to each feature subset is the feature with the largest correlation with the karyotype abnormal cells and the smallest redundancy in the corresponding feature subsets. However, in practical applications, the inventors of the present application find that feature redundancy may also exist among feature subsets, and based on this, in the embodiments of the present application, after determining a candidate karyotype-associated feature set (i.e., a set formed by second feature combinations corresponding to the feature subsets), features in the candidate karyotype-associated feature set may be re-screened based on an mRMR algorithm to remove redundant features among different second feature combinations, so as to obtain a potential karyotype-associated feature set.

After the potential karyotype association feature set is obtained, features in the potential karyotype association feature set are selected based on a backward selection method, and the target karyotype association feature is determined, so that the efficiency and accuracy of identification of the karyotype abnormal cells can be guaranteed to the maximum extent through the target karyotype association feature. It is understood that the forward selection method may also be used to screen the set of potential karyotype-associated features, which has the same effect as the backward selection method.

In the method provided by the embodiment of the present application, the secondary screening of the candidate karyotype correlation feature set based on the mRMR algorithm and the backward selection method specifically includes: and re-screening the features in the candidate karyotype associated feature set based on an mRMR algorithm to determine a potential karyotype associated feature set, selecting the features in the potential karyotype associated feature set based on a backward selection method to determine the target karyotype associated feature, and further screening the karyotype associated feature to ensure the identification efficiency and accuracy of the karyotype abnormal cells to the maximum extent.

Based on any one of the embodiments above, the determining, by a ReliefF algorithm, a weight corresponding to each feature in the target feature subset based on the feature value corresponding to each sample in the training sample set and the target feature subset includes:

determining the difference between the feature values corresponding to the target sample and the similar nearest neighbor samples with the preset number and the difference between the feature values corresponding to the target sample and the different nearest neighbor samples with the preset number based on the feature values corresponding to the samples in the training sample set and the target feature subset;

and updating the weight corresponding to each feature in the target feature subset based on the difference between the feature values corresponding to the target sample and the similar nearest neighbor samples with the preset number and the difference between the feature values corresponding to the target sample and the different nearest neighbor samples with the preset number until the updating times reach a preset threshold value.

Specifically, based on the foregoing embodiment, after determining the feature value corresponding to each sample in the training sample set and the target feature subset, the weight corresponding to each feature in the target feature subset may be calculated, specifically, based on the feature value corresponding to each sample in the training sample set and the target feature subset, the difference between the feature value corresponding to the target sample and the same-type nearest neighbor samples of the preset number (i.e., k in the foregoing embodiment) and the difference between the feature value corresponding to the target sample and different-type nearest neighbor samples of the preset number are determined, and then based on the difference between the feature value corresponding to the target sample and the same-type nearest neighbor samples of the preset number and the difference between the feature value corresponding to the target sample and different-type nearest neighbor samples of the preset number, the weight corresponding to each feature in the target feature subset is updated until the update number reaches the preset threshold (i.e., the foregoing number m), the weight corresponding to each feature in the target feature subset can be quickly determined. As for the specific updating process of the weights, the foregoing embodiments have been described, and are not described herein again. It can be understood that the values of k and m can be set according to actual needs, and this is not specifically limited in this embodiment of the present application.

In the method provided by the embodiment of the present application, determining, by using a ReliefF algorithm, a weight corresponding to each feature in the target feature subset based on a feature value corresponding to each sample in the training sample set and the target feature subset includes: based on the characteristic values corresponding to the samples in the training sample set and the target characteristic subset, determining the difference between the characteristic values corresponding to the target sample and the similar nearest neighbor samples with the preset number, and the difference between the characteristic values corresponding to the target sample and the different nearest neighbor samples with the preset number, and updating the weight corresponding to each feature in the target characteristic subset until the updating frequency reaches a preset threshold value, so that the weight of each feature in each target characteristic subset can be rapidly determined, and the efficiency and the accuracy of feature screening are ensured.

Based on any of the above embodiments, the target karyotype correlation features include: the cell nucleus perimeter, the cell nucleus area, the cell nucleus circularity, the cell nucleus elongation, the cell nucleus aspect ratio, the cell nucleus rectangularity, the cell nucleus firmness, the cell nucleus direction degree, the cell nucleus long axis, the cell nucleus short axis, the cell nucleus microscopic image sharpness, the cell nucleus microscopic image total signal intensity, the cell nucleus microscopic image average signal intensity, the cell nucleus microscopic image signal intensity variance and the cell nucleus microscopic image corresponding Zernike moment.

Specifically, in the embodiment of the present application, a candidate karyotype associated feature set is obtained by screening from an initial karyotype associated feature set based on a ReliefF algorithm and an mRMR algorithm, and a target karyotype associated feature determined after performing secondary screening on the candidate karyotype associated feature set based on the mRMR algorithm and a backward selection method includes: the cell nucleus perimeter, the cell nucleus area, the cell nucleus circularity, the cell nucleus elongation, the cell nucleus aspect ratio, the cell nucleus rectangularity, the cell nucleus firmness, the cell nucleus direction degree, the cell nucleus long axis, the cell nucleus short axis, the cell nucleus microscopic image sharpness, the cell nucleus microscopic image total signal intensity, the cell nucleus microscopic image average signal intensity, the cell nucleus microscopic image signal intensity variance and the cell nucleus microscopic image corresponding Zernike moment. It can be understood that the Zernike moments corresponding to the nuclear microscopic images, namely the aforementioned 2-9 order Zernike moments, can be considered based on the efficiency and accuracy of abnormal karyotype cell identification. Meanwhile, it can be seen that the finally obtained target karyotype associated features do not include color features, because the features in the color feature subset are redundant with the other three feature subsets, and therefore, the features in the color feature subset are removed. When the candidate karyotype associated feature set is obtained by screening from the initial karyotype associated feature set based on the Relieff algorithm and the mRMR algorithm, as the feature subsets are respectively screened, the comprehensiveness of the features in the candidate karyotype associated feature set is ensured, the mistaken deletion of useful features is avoided, the candidate karyotype associated feature set is secondarily screened through the mRMR algorithm and a backward selection method, redundant features among the feature subsets are eliminated, the identification accuracy of karyotype abnormal cells is ensured, and the identification efficiency is also considered.

In the method provided by the embodiment of the present application, the target karyotype correlation characteristic includes: the cell nucleus perimeter, the cell nucleus area, the cell nucleus circularity, the cell nucleus elongation, the cell nucleus width-to-height ratio, the cell nucleus rectangularity, the cell nucleus firmness, the cell nucleus direction degree, the cell nucleus long axis, the cell nucleus short axis, the cell nucleus microscopic image sharpness, the cell nucleus microscopic image total signal intensity, the cell nucleus microscopic image average signal intensity, the cell nucleus microscopic image signal intensity variance and the cell nucleus microscopic image corresponding Zernike moment, so that the identification accuracy of the nuclear abnormal cells is guaranteed, and the identification efficiency is also considered.

The device for determining abnormal karyotype cells provided by the present application is described below, and the device for determining abnormal karyotype cells described below and the method for determining abnormal karyotype cells described above may be referred to in correspondence with each other.

Based on any of the above embodiments, fig. 4 is a schematic structural diagram of the device for determining a karyotypic abnormal cell provided by the present application, as shown in fig. 4, the device includes:

the target karyotype association characteristic value determining module 410 is configured to determine a target karyotype association characteristic value corresponding to a cell to be identified based on a cell nucleus microscopic image corresponding to the cell to be identified; the target karyotype correlation characteristics are determined after screening a candidate karyotype correlation characteristic set from an initial karyotype correlation characteristic set based on a Relieff algorithm and a maximum correlation minimum redundancy (mRMR) algorithm, and performing secondary screening on the candidate karyotype correlation characteristic set based on an mRMR algorithm and a backward selection method; the initial karyotype associated feature set comprises a morphological feature subset, a texture feature subset, a color feature subset and an edge feature subset, and the candidate karyotype associated feature set comprises at least one candidate morphological feature, at least one candidate texture feature, at least one candidate color feature and at least one candidate edge feature;

a karyotype abnormal cell determining module 420, configured to input the target karyotype associated characteristic value corresponding to the cell to be identified into the trained karyotype abnormal cell determining model, and output a karyotype abnormal cell identification result corresponding to the target karyotype associated characteristic value;

The device provided by the embodiment of the application determines a target karyotype association feature value corresponding to a cell to be identified based on a cell nucleus microscopic image corresponding to the cell to be identified, wherein the target karyotype association feature value is determined by screening a candidate karyotype association feature set from an initial karyotype association feature set based on a Relieff algorithm and a maximum correlation minimum redundancy mRMR algorithm, and then determined by secondary screening of the candidate karyotype association feature set based on an mRMR algorithm and a backward selection method, the initial karyotype association feature set comprises a morphological feature subset, a texture feature subset, a color feature subset and an edge feature subset, the candidate karyotype association feature set comprises at least one candidate morphological feature, at least one candidate texture feature, at least one candidate color feature and at least one candidate edge feature, the target karyotype association feature value corresponding to the cell to be identified is input into a trained karyotype abnormal cell determination model, the karyotype abnormal cell identification result corresponding to the target karyotype association feature value is output, the high-accuracy of the karyotype abnormal cell identification is achieved, and the accuracy of the karyotype abnormal cell identification is guaranteed, and the high-accuracy of the karyotype abnormal cell identification efficiency is further achieved.

Based on the above embodiment, the apparatus further includes a candidate karyotype associated feature set determination module, where the candidate karyotype associated feature set determination module is configured to perform the following steps:

Based on any of the above embodiments, the apparatus further comprises a secondary screening module, the secondary screening module is configured to perform the following steps:

and selecting the features in the potential karyotype association feature set based on a backward selection method, and determining the target karyotype association feature.

Based on any of the above embodiments, the determining, by a ReliefF algorithm, the weight corresponding to each feature in the target feature subset based on the feature value corresponding to each sample in the training sample set and the target feature subset includes:

Based on any of the above embodiments, the target karyotype association characteristics include: the cell nucleus perimeter, the cell nucleus area, the cell nucleus circularity, the cell nucleus elongation, the cell nucleus aspect ratio, the cell nucleus rectangularity, the cell nucleus firmness, the cell nucleus direction degree, the cell nucleus long axis, the cell nucleus short axis, the cell nucleus microscopic image sharpness, the cell nucleus microscopic image total signal intensity, the cell nucleus microscopic image average signal intensity, the cell nucleus microscopic image signal intensity variance and the cell nucleus microscopic image corresponding Zernike moment.

Fig. 5 illustrates a physical structure diagram of an electronic device, which may include, as shown in fig. 5: a processor (processor) 510, a communication Interface (Communications Interface) 520, a memory (memory) 530, and a communication bus 540, wherein the processor 510, the communication Interface 520, and the memory 530 communicate with each other via the communication bus 540. Processor 510 may call logic instructions in memory 530 to perform the method for determining a karyotypically abnormal cell provided by the above methods, the method comprising: determining a target karyotype association characteristic value corresponding to the cell to be identified based on a cell nucleus microscopic image corresponding to the cell to be identified; the target karyotype association feature is determined after a candidate karyotype association feature set is obtained by screening from an initial karyotype association feature set based on a Relieff algorithm and a maximum correlation minimum redundancy (mRMR) algorithm, and then secondary screening is carried out on the candidate karyotype association feature set based on an mRMR algorithm and a backward selection method; the initial karyotype associated feature set comprises a morphological feature subset, a texture feature subset, a color feature subset and an edge feature subset, and the candidate karyotype associated feature set comprises at least one candidate morphological feature, at least one candidate texture feature, at least one candidate color feature and at least one candidate edge feature; inputting the target karyotype association characteristic value corresponding to the cell to be identified into a trained karyotype abnormal cell determination model, and outputting a karyotype abnormal cell identification result corresponding to the target karyotype association characteristic value; the karyotype abnormal cell determination model is obtained by training a target karyotype association characteristic value sample and a predetermined karyotype abnormal cell identification result label.

In addition, the logic instructions in the memory 530 may be implemented in the form of software functional units and stored in a computer readable storage medium when the logic instructions are sold or used as a stand-alone product. Based on such understanding, the technical solution of the present application or portions thereof that substantially contribute to the prior art may be embodied in the form of a software product stored in a storage medium and including instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.

In another aspect, the present application further provides a computer program product, the computer program product includes a computer program, the computer program can be stored on a non-transitory computer readable storage medium, when the computer program is executed by a processor, the computer can execute the method for determining karyotypic abnormal cells provided by the above methods, the method includes: determining a target karyotype association characteristic value corresponding to the cell to be identified based on the cell nucleus microscopic image corresponding to the cell to be identified; the target karyotype association feature is determined after a candidate karyotype association feature set is obtained by screening from an initial karyotype association feature set based on a Relieff algorithm and a maximum correlation minimum redundancy (mRMR) algorithm, and then secondary screening is carried out on the candidate karyotype association feature set based on an mRMR algorithm and a backward selection method; the initial karyotype associated feature set comprises a morphological feature subset, a texture feature subset, a color feature subset and an edge feature subset, and the candidate karyotype associated feature set comprises at least one candidate morphological feature, at least one candidate texture feature, at least one candidate color feature and at least one candidate edge feature; inputting the target karyotype association characteristic value corresponding to the cell to be identified into a trained karyotype abnormal cell determination model, and outputting a karyotype abnormal cell identification result corresponding to the target karyotype association characteristic value; the karyotype abnormal cell determination model is obtained by training a target karyotype association characteristic value sample and a predetermined karyotype abnormal cell identification result label.

In yet another aspect, the present application further provides a non-transitory computer-readable storage medium, on which a computer program is stored, the computer program being implemented by a processor to perform the method for determining a karyotype abnormal cell provided by the above methods, the method including: determining a target karyotype association characteristic value corresponding to the cell to be identified based on a cell nucleus microscopic image corresponding to the cell to be identified; the target karyotype correlation characteristics are determined after screening a candidate karyotype correlation characteristic set from an initial karyotype correlation characteristic set based on a Relieff algorithm and a maximum correlation minimum redundancy (mRMR) algorithm, and performing secondary screening on the candidate karyotype correlation characteristic set based on an mRMR algorithm and a backward selection method; the initial karyotype associated feature set comprises a morphological feature subset, a texture feature subset, a color feature subset and an edge feature subset, and the candidate karyotype associated feature set comprises at least one candidate morphological feature, at least one candidate texture feature, at least one candidate color feature and at least one candidate edge feature; inputting the target karyotype association characteristic value corresponding to the cell to be identified into a trained karyotype abnormal cell determination model, and outputting a karyotype abnormal cell identification result corresponding to the target karyotype association characteristic value; the karyotype abnormal cell determination model is obtained by training a target karyotype association characteristic value sample and a predetermined karyotype abnormal cell identification result label.

The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.

Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware. With this understanding in mind, the above-described technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments.

Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present application.

Claims

1. A method for determining karyotypically abnormal cells, comprising:

determining a target karyotype association characteristic value corresponding to the cell to be identified based on the cell nucleus microscopic image corresponding to the cell to be identified; the target karyotype correlation characteristics are determined after screening a candidate karyotype correlation characteristic set from an initial karyotype correlation characteristic set based on a Relieff algorithm and a maximum correlation minimum redundancy (mRMR) algorithm, and performing secondary screening on the candidate karyotype correlation characteristic set based on an mRMR algorithm and a backward selection method; the initial karyotype associated feature set comprises a morphological feature subset, a texture feature subset, a color feature subset and an edge feature subset, and the candidate karyotype associated feature set comprises at least one candidate morphological feature, at least one candidate texture feature, at least one candidate color feature and at least one candidate edge feature;

the karyotype abnormal cell determination model is obtained by training a target karyotype correlation characteristic value sample and a predetermined karyotype abnormal cell identification result label;

the step of acquiring the cell nucleus microscopic image corresponding to the cell to be identified comprises the following steps:

determining candidate cell nucleuses with over-segmentation based on the position relation of each cell nucleus and the number of fluorescent signal points in the cell nucleus in initial cell nucleus microscopic images corresponding to different fluorescent channels of the blood sample, and correcting the candidate cell nuclei based on a convex hull detection algorithm and a convex defect algorithm to obtain cell nucleus corrected microscopic images corresponding to the blood sample;

taking the cell nucleus correction microscopic image corresponding to the blood sample as a cell nucleus microscopic image corresponding to the cell to be identified;

the step of correcting the candidate cell nucleus based on the convex hull detection algorithm and the convex defect algorithm specifically comprises the following steps:

2. The method for determining the karyotype abnormal cell according to claim 1, wherein the screening of the initial karyotype associated feature set based on the ReliefF algorithm and the maximum-correlation-minimum-redundancy mRMR algorithm to obtain the candidate karyotype associated feature set specifically includes:

3. The method for determining the karyotypic abnormal cells according to claim 2, wherein the feature screening is performed on each feature subset in the initial karyotype-associated feature set based on a ReliefF algorithm and an mRMR algorithm, specifically including:

4. The method of determining karyotypically abnormal cells according to claim 3, wherein the feature in the second combination of features is a feature in the first combination of features that maximizes the difference or quotient between the first average mutual information and the second average mutual information.

5. The method for determining the karyotypically abnormal cell according to claim 1, wherein the secondary screening of the candidate karyotype-associated feature set based on the mRMR algorithm and the backward selection method specifically includes:

6. The method according to claim 3, wherein the determining, by a ReliefF algorithm, the weight corresponding to each feature in the target feature subset based on the feature value corresponding to each sample in the training sample set and the target feature subset comprises:

7. The method of determining karyotypically abnormal cells according to claim 1, wherein said target karyotypically-associated characteristics include: the cell nucleus perimeter, the cell nucleus area, the cell nucleus circularity, the cell nucleus elongation, the cell nucleus aspect ratio, the cell nucleus rectangularity, the cell nucleus firmness, the cell nucleus direction degree, the cell nucleus long axis, the cell nucleus short axis, the cell nucleus microscopic image sharpness, the cell nucleus microscopic image total signal intensity, the cell nucleus microscopic image average signal intensity, the cell nucleus microscopic image signal intensity variance and the cell nucleus microscopic image corresponding Zernike moment.

8. An apparatus for determining karyotype abnormal cells, comprising:

the karyotype abnormal cell determining module is used for inputting the target karyotype associated characteristic value corresponding to the cell to be identified into the trained karyotype abnormal cell determining model and outputting a karyotype abnormal cell identification result corresponding to the target karyotype associated characteristic value;

9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the method for determining karyotypically abnormal cells as claimed in any one of claims 1 to 7 when executing the program.

10. A non-transitory computer-readable storage medium, on which a computer program is stored, wherein the computer program, when executed by a processor, implements the steps of the method for determining karyotypic abnormal cells according to any one of claims 1 to 7.