CN117934337B - Method for mask repair of blocked chromosome based on unsupervised learning - Google Patents
Method for mask repair of blocked chromosome based on unsupervised learning Download PDFInfo
- Publication number
- CN117934337B CN117934337B CN202410330149.XA CN202410330149A CN117934337B CN 117934337 B CN117934337 B CN 117934337B CN 202410330149 A CN202410330149 A CN 202410330149A CN 117934337 B CN117934337 B CN 117934337B
- Authority
- CN
- China
- Prior art keywords
- chromosome
- mask
- image
- repair
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 210000000349 chromosome Anatomy 0.000 title claims abstract description 298
- 230000008439 repair process Effects 0.000 title claims abstract description 83
- 238000000034 method Methods 0.000 title claims abstract description 39
- 230000008859 change Effects 0.000 claims abstract description 5
- 230000011218 segmentation Effects 0.000 claims description 22
- 230000006870 function Effects 0.000 claims description 19
- 230000031864 metaphase Effects 0.000 claims description 17
- 230000008569 process Effects 0.000 claims description 13
- 238000013528 artificial neural network Methods 0.000 claims description 9
- 238000002372 labelling Methods 0.000 claims description 9
- 238000003062 neural network model Methods 0.000 claims description 9
- 230000000903 blocking effect Effects 0.000 claims description 5
- 238000003709 image segmentation Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a method for mask repair of a blocked chromosome based on unsupervised learning, and relates to the technical field of image processing; determining a shielding relation between the chromosomes by comparing the areas of each chromosome in the overlapped and shielded image blocks after the mask repair, wherein if the area of the mask of the chromosome is predicted to be large by the mask repair model, the chromosome is a shielded chromosome, otherwise, if the area of the mask of the chromosome is not changed or the area change is small after the mask of the chromosome is predicted by the mask repair model, the chromosome is a shielded chromosome; and for the chromosome judged to be blocked, replacing coordinates of the corresponding position in the corresponding chromosome image label by using the repaired chromosome image mask, so that mask repair of the blocked chromosome is realized, and the whole chromosome can be identified when the chromosome is partially blocked.
Description
Technical Field
The invention relates to the technical field of image processing, in particular to a method for mask repair of a blocked chromosome based on unsupervised learning.
Background
At present, the image of the metaphase of the chromosome collected under the high-power microscope contains a large number of chromosomes, and the phenomenon of cross overlapping between each chromosome is unavoidable, so that the structure of the image is complex. In the case of manual labeling of an image segmentation task for an image of a metaphase of a chromosome, it is also difficult to accurately label a mask of a blocked chromosome. Therefore, in the process of performing an image segmentation task on a chromosome image, the phenomenon that a segmentation mask of the chromosome image which is blocked in the image is broken, so that the mask of the segmented image is incomplete is unavoidable. The splitting mask breakage phenomenon of the chromosome image can cause the morphological damage of the chromosome after the chromosome image is split, even one chromosome is identified as a plurality of chromosomes, and the classification and the karyotype analysis effect of the chromosome image are greatly influenced.
Labeling data scarcity: conventional supervised learning methods typically require a large number of labeled samples to train the model. The segmentation labeling task of the overlapping and crossing of the chromosome karyotypes has strong professional and complexity, and requires the expertise of the medical image field to carry out image labeling, so that the economic and time cost for obtaining the high-quality labeling sample is very high. Because the acquired training data of the metaphase image of the chromosome with the cross overlapping phenomenon is less, the existing image segmentation model is difficult to truly learn the segmentation mask characteristics of the overlapping cross region of the chromosome. The prior method is used for dividing the metaphase image of the chromosome, and the problem of mask fracture of the chromosome image caused by crossed overlapping between chromosomes in the image is difficult to solve.
Diversity of cross occlusion modes: in metaphase images of chromosomes, occlusion patterns between chromosomes are very diverse, involving different occlusion shapes and positions. The existing supervised learning methods are not sufficient to capture this diversity.
Problems and considerations in the prior art:
how to solve the technical problem that the whole chromosome can not be identified when the chromosome is partially blocked.
Disclosure of Invention
The invention provides a method for mask repair of a blocked chromosome based on unsupervised learning, which solves the technical problem that the whole chromosome cannot be identified when the chromosome is partially blocked.
In order to solve the technical problems, the technical scheme adopted by the invention is as follows:
a method for mask repair of occluded chromosomes based on unsupervised learning comprises the following steps,
Step S7: determining a shielding relationship between the crossing chromosomes;
Determining a shielding relation between the chromosomes by comparing the areas of each chromosome in the overlapped and shielded image blocks after the mask repair, wherein if the area of the mask of the chromosome is predicted to be large by the mask repair model, the chromosome is a shielded chromosome, otherwise, if the area of the mask of the chromosome is not changed or the area change is small after the mask of the chromosome is predicted by the mask repair model, the chromosome is a shielded chromosome;
step S8: replaced by mask tags that occlude chromosomes;
And for the chromosome judged to be blocked, replacing coordinates of the corresponding position in the corresponding chromosome image label by using the repaired chromosome image mask, so as to realize mask repair of the blocked chromosome.
The further technical proposal is that: in the step S7, the area of the mask of the chromosome image before or after repair is calculated by the formula (8):
(8)
In the formula (8), A is the area of the chromosome mask, and i is the coordinate number in the chromosome mask; n is the number of closed coordinates of the chromosome mask being sought.
The further technical proposal is that: further comprising the following steps before said step S7,
Step S1: preparing a chromosome dataset;
step S2: training a chromosome image instance segmentation model;
Step S3: training a chromosome image mask repair model;
Step S4: predicting chromosome segmentation images;
Step S5: judging crossed occlusion of chromosomes;
Step S6: prediction of cross occlusion chromosome mask repair.
The further technical proposal is that: the step S1 comprises the steps of obtaining an image of a metaphase of a chromosome, obtaining a chromosome dataset D based on an image annotation of the metaphase of the chromosome, the chromosome dataset D comprising mask tags and class tags of visible portions of the chromosome that are not occluded.
The further technical proposal is that: the step S2 includes the steps of inputting the chromosome dataset D obtained in the step S1 into an instance segmentation model and performing model training to obtain a trained chromosome image instance segmentation model f m.
The further technical proposal is that: the step S3 comprises the steps of training and obtaining a trained chromosome image mask repair model f p, and the specific division comprises the following steps:
Step S301: selecting one marked chromosome from the chromosome data set D obtained in the step S1 as a first example chromosome A, randomly selecting the other marked chromosome as a second example chromosome B, and obtaining chromosome images and mask labels of the first example chromosome A and the second example chromosome B;
Step S302: the model training process is divided into two parts, namely a mask repair training part and a regular part; inputting the chromosome image and the mask label selected in the step S301 into a mask repair model for training according to the rules of the two training parts as follows;
Step S3021 mask repair training section: defining a mask B corresponding to a second example chromosome B as an image eraser, randomly shifting the image eraser to the region of the mask A corresponding to the first example chromosome A, erasing part of the mask A by using the mask B, and then obtaining a region mask A\B where the mask A is not shielded by the mask B;
Step S3022 regular part: shifting the mask B to the same position as the random shift in the step S3021, but not erasing the mask A, so as to obtain a region mask B\A where the mask B is not shielded by the mask A;
Step S303: respectively inputting the two training parts in the step S302 into a neural network model for training, inputting the mask repairing training part into a mask A\B, and outputting the mask A under the condition of the mask B; the regular part is input as a mask A, and the condition is mask B\A, so that the mask A is kept and output; the loss function of the model is thus defined as follows:
(1)
(2)
(3)
In the formula (1), L 1 represents a loss function of the mask repair training section; n represents the unsupervised training times in each iterative training process; a, B are chromosome instances in dataset D; l represents a binary cross entropy Binary CrossEntropy loss function, and f p is a mask repair model; m A\B represents a mask A\B; m B represents mask B; i represents an image block of the input neural network; m A represents mask A;
In the formula (2), L 2 represents a loss function of the canonical part; m B\A represents a mask B\A;
in the formula (3), L m represents a final loss function of the model; x corresponds to the bernoulli distribution with a coefficient gamma; gamma is the probability of selecting a mask repair training part in the training process;
Step S304: and optimizing the weight parameters of the model through iterative training according to the neural network model determined in the step S303 and the final loss function L m of the formula (3), and finally obtaining a trained chromosome image mask repair model f p.
The further technical proposal is that: the step S4 includes the steps of inputting the acquired metaphase image of the chromosome into the trained chromosome image example segmentation model f m obtained in the step S2, and predicting to obtain a mask label of a visible part of the chromosome and a category label of the chromosome.
The further technical proposal is that: step S5 comprises the steps of traversing the mid-phase image of the chromosome, the mask label of the visible part of the corresponding chromosome and the class label of the chromosome, judging whether the coordinates of each chromosome mask are crossed or not according to whether a crossed region exists between the closed coordinate curves of the chromosome mask labels by using a vector operation library Shapely algorithm, and judging whether a shielding condition exists between the chromosomes or not; if there is a blocking condition, the maximum and minimum coordinates of the crossed chromosome masks are obtained from the equation (4), the equation (5), the equation (6), and the equation (7), namely:
(4)
(5)
(6)
(7)
(X 1,y1),(x1,y2),(x2,y1),(x2,y2) is the coordinates of the region in the image where the chromosome overlap occlusion occurs, (x i,yi) is the coordinates in the mask labels of all crossed chromosomes; and cutting out the image blocks overlapped and shielded by the chromosome according to the region coordinates overlapped and shielded by the chromosome.
The further technical proposal is that: the step S6 includes steps of traversing all crossed chromosomes in the image blocks overlapped and shielded by the chromosomes in the step S5, respectively inputting the mask repair models f p obtained in the step S3, and predicting to obtain repair masks of all crossed chromosome images.
The further technical proposal is that: in the step S303, the neural network model is selected as a residual neural network ResNet, a dense-connectivity neural network DenseNet, or a U-Net.
The beneficial effects of adopting above-mentioned technical scheme to produce lie in:
Determining a shielding relation between the chromosomes by comparing the areas of each chromosome in the overlapped and shielded image blocks after the mask repair, wherein if the area of the mask of the chromosome is predicted to be large by the mask repair model, the chromosome is a shielded chromosome, otherwise, if the area of the mask of the chromosome is not changed or the area change is small after the mask of the chromosome is predicted by the mask repair model, the chromosome is a shielded chromosome; and for the chromosome judged to be blocked, replacing coordinates of the corresponding position in the corresponding chromosome image label by using the repaired chromosome image mask, so that mask repair of the blocked chromosome is realized, and the whole chromosome can be identified when the chromosome is partially blocked.
See the description of the detailed description section.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a data flow diagram of an unsupervised training process of a mask repair model of a chromosome image;
FIG. 3 is a visual representation of a chromosome image mask broken due to cross occlusion;
Fig. 4 is a visual image of a mask repair achieved by a chromosome image through a mask repair model.
Detailed Description
The mask repair method of the blocked chromosome based on the unsupervised learning refers to a process of performing reasoning repair on the chromosome incomplete mask with the cross blocking area by using the unsupervised learning training method, so that the problem that the chromosome mask is broken after being segmented due to the cross blocking is solved. The method is mainly based on an unsupervised learning method in computer vision.
The following description of the embodiments of the present application will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present application, but not all embodiments. The following description of at least one exemplary embodiment is merely exemplary in nature and is in no way intended to limit the application, its application, or uses. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application, but the present application may be practiced in other ways other than those described herein, and persons skilled in the art will readily appreciate that the present application is not limited to the specific embodiments disclosed below.
As shown in fig. 1, the invention discloses a method for mask repair of an occluded chromosome based on unsupervised learning, which comprises the following steps:
step S1: a chromosome dataset is prepared.
An image of a metaphase of the chromosome is obtained, and a chromosome dataset D is obtained based on the image annotation of the metaphase of the chromosome, the chromosome dataset D including mask tags and class tags for visible portions of the chromosome that are not occluded.
That is, a large number of images including a metaphase of a chromosome are collected by a microscope, and division mask labeling and chromosome class labeling are manually performed on a visible part of the chromosome which is not blocked in the images, so that mask labels of the visible part of the chromosome and class labels of the chromosome are generated, and a chromosome data set D is formed.
Step S2: a chromosome image instance segmentation model is trained.
Based on the chromosome dataset D obtained in the step S1, inputting the chromosome dataset D into an instance segmentation model and performing model training to obtain a trained chromosome image instance segmentation model f m.
That is, using the chromosome dataset D in step S1, an image of the metaphase of the chromosome is input, model training is performed based on an example segmentation model, which is YOLO or MaskRCNN, in order to output a mask tag that obtains the visible portion of the chromosome and a class tag of the chromosome. The weight parameters of the existing pre-trained model are used as initial weights, and then the weight parameters of the model are optimized through iterative training, so that the chromosome image example segmentation model f m is obtained. The example segmentation model is a prior art and will not be described in detail herein.
Step S3: and training a chromosome image mask repair model.
Training and obtaining a trained chromosome image mask repair model f p.
Step S301: one marked chromosome is selected from the chromosome data set D obtained in the step S1 as a first example chromosome A, the other marked chromosome is randomly selected as a second example chromosome B, and chromosome images and mask labels of the first example chromosome A and the second example chromosome B are obtained.
As shown in FIG. 2, input chromosome A is a first example chromosome A, and random chromosome B is a second example chromosome B.
Step S302: the model training process is divided into two parts, namely a mask repair training part and a regular part. The chromosome image and the mask label selected in step S301 are respectively input into the mask repair model for training according to the rules of the following two training parts.
Step S3021 mask repair training section: defining a mask B corresponding to a chromosome B of a second example as an image eraser, randomly shifting the image eraser to the region of the mask A corresponding to the chromosome A of the first example, erasing part of the mask A by using the mask B, and then obtaining a region mask A\B where the mask A is not shielded by the mask B.
Step S3022 regular part: and (3) shifting the mask B to the same position as the random shift in the step S3021, wherein the mask A is not erased, and the area mask B\A where the mask B is not shielded by the mask A is obtained.
Step S303: the two training parts in step S302 are respectively input into the neural network model for training, the mask repair training part is input into the mask a\b, and the condition is the mask B, so as to recover and output the mask a. And the regular part is input as a mask A, and the condition is mask B\A, so that the mask A is kept and output. The neural network model may be selected as a residual neural network ResNet, a densely connected neural network DenseNet, or a U-Net. The loss function of the model is thus defined as follows:
(1)
(2)
(3)
In the formula (1), L 1 represents a loss function of the mask repair training section; n represents the unsupervised training times in each iterative training process; a, B are chromosome instances in dataset D; l represents a binary cross entropy Binary CrossEntropy loss function, and f p is a mask repair model; m A\B represents a mask A\B; m B represents mask B; i represents an image block of the input neural network; m A represents mask A.
In the formula (2), L 2 represents a loss function of the canonical part; m B\A represents mask B\A.
In the formula (3), L m represents a final loss function of the model; x corresponds to the bernoulli distribution with a coefficient gamma; gamma is the probability of selecting a mask repair training portion for the training process.
Step S304: and optimizing the weight parameters of the model through iterative training according to the neural network model determined in the step S303 and the final loss function L m of the formula (3), and finally obtaining a trained chromosome image mask repair model f p.
Step S4: and (3) predicting chromosome segmentation images.
Inputting the acquired metaphase image of the chromosome into a trained chromosome image example segmentation model f m obtained in the step S2, and predicting to obtain a mask label of a visible part of the chromosome and a category label of the chromosome.
Namely, the metaphase image of the chromosome collected under the microscope is input into the trained chromosome image example segmentation model f m obtained in the step S2, and a mask label of the visible part of the chromosome and a category label of the chromosome are obtained through prediction.
Step S5: and judging the crossed occlusion of the chromosome.
Traversing the mid-phase image of the chromosome, the mask label of the visible part of the chromosome and the class label of the chromosome, which are obtained in the step S4, judging whether the coordinates of each chromosome mask are crossed or not according to whether a crossed region exists between the closed coordinate curves of the chromosome mask labels by using a vector operation library Shapely algorithm, and judging whether shielding conditions exist between the chromosomes or not. If there is a blocking condition, the maximum and minimum coordinates of the crossed chromosome masks are obtained from the equation (4), the equation (5), the equation (6), and the equation (7), namely:
(4)
(5)
(6)
(7)
(X 1,y1),(x1,y2),(x2,y1),(x2,y2) is the coordinates of the region in the image where the chromosome overlap occlusion occurs, and (x i,yi) is the coordinates in the mask labels of all chromosomes where the crossover occurs. And cutting out the image blocks overlapped and shielded by the chromosome according to the region coordinates overlapped and shielded by the chromosome.
Step S6: prediction of cross occlusion chromosome mask repair.
Traversing all crossed chromosomes in the image blocks overlapped and shielded by the chromosomes in the step S5, respectively inputting the mask repair model f p obtained in the step S3, and predicting to obtain repair masks of all crossed chromosome images.
Step S7: occlusion relationships between intersecting chromosomes are determined.
The area of the mask of all crossed chromosome images before or after repair in step S6 is then calculated by equation (8):
(8)
In the formula (8), A is the area of the chromosome mask, and i is the coordinate number in the chromosome mask; n is the number of closed coordinates of the chromosome mask being sought.
The occlusion relation between the chromosomes is determined by comparing the areas of each chromosome in the overlapped occlusion image blocks after the mask repair, if the area of the mask of the chromosome is predicted to be large through the mask repair model f p, the chromosome is the occluded chromosome, otherwise, if the area of the mask of the chromosome is not changed or the area change is small after the mask of the chromosome is predicted through the mask repair model f p, the chromosome is the occluded chromosome.
Step S8: replaced by mask tags that occlude the chromosome.
Traversing all the chromosomes judged to be blocked in the step S7, and replacing coordinates of corresponding positions in corresponding chromosome image labels obtained in the step S4 by using the repaired chromosome image masks to realize the mask repair function of the blocked chromosomes.
As shown in fig. 3, the chromosome extending from the lower left corner to the upper right corner breaks the chromosome image mask due to occlusion by the middle chromosome crossover.
As shown in fig. 4, the effect of mask repair achieved by the chromosome image through the mask repair model.
The object is: by the technology, the problem that the mask of the chromosome is broken after the chromosome image is segmented by the example is solved, and the mask of the part shielded by the chromosome is deduced by unsupervised learning, so that the repair of the mask of the chromosome fracture is realized.
The advantages are that:
1. The occluded data need not be annotated: the unsupervised learning method does not require extensive annotation data. For chromosome images, it may be difficult to obtain data with accurate occluded mask tags, so the unsupervised method can avoid the need to annotate the data.
2. Adaptive diversity occlusion mode: mask repair of occluded chromosomes may involve multiple occlusion modes, including different occlusion shapes and positions. The unsupervised learning method can automatically learn and adapt to different shielding modes without explicit label guidance.
3. Potential feature learning: unsupervised learning helps learn potential features and representations from the data. The model can better capture the characteristics of the shape, the texture and the like of the chromosome by independently learning the internal structure of the chromosome image, and better realize mask repair.
4. Generalizing to unknown occlusion cases: because the unsupervised learning method emphasizes the learning of the overall structure of the data, the generalization capability in the unknown occlusion mode is strong. This makes the model more robust in practical applications.
5. The manual intervention is reduced: the unsupervised learning method generally reduces the need for manual labeling and manual intervention, thereby simplifying the overall system construction process.
Claims (6)
1. A method for mask repair of occluded chromosomes based on unsupervised learning, characterized by: comprises the following steps of the method,
Step S1: preparing a chromosome dataset;
The step S1 comprises the steps of obtaining an image of a metaphase of a chromosome, obtaining a chromosome data set D based on image labeling of the metaphase of the chromosome, wherein the chromosome data set D comprises mask tags and category tags of visible parts of the chromosome which are not blocked;
step S2: training a chromosome image instance segmentation model;
the step S2 comprises the steps of inputting the chromosome data set D obtained in the step S1 into an instance segmentation model and performing model training to obtain a trained chromosome image instance segmentation model f m;
Step S3: training a chromosome image mask repair model;
The step S3 comprises the steps of training and obtaining a trained chromosome image mask repair model f p, and the specific division comprises the following steps:
Step S301: selecting one marked chromosome from the chromosome data set D obtained in the step S1 as a first example chromosome A, randomly selecting the other marked chromosome as a second example chromosome B, and obtaining chromosome images and mask labels of the first example chromosome A and the second example chromosome B;
Step S302: the model training process is divided into two parts, namely a mask repair training part and a regular part; inputting the chromosome image and the mask label selected in the step S301 into a mask repair model for training according to the rules of the two training parts as follows;
Step S3021 mask repair training section: defining a mask B corresponding to a second example chromosome B as an image eraser, randomly shifting the image eraser to the region of the mask A corresponding to the first example chromosome A, erasing part of the mask A by using the mask B, and then obtaining a region mask A\B where the mask A is not shielded by the mask B;
Step S3022 regular part: shifting the mask B to the same position as the random shift in the step S3021, but not erasing the mask A, so as to obtain a region mask B\A where the mask B is not shielded by the mask A;
Step S303: respectively inputting the two training parts in the step S302 into a neural network model for training, inputting the mask repairing training part into a mask A\B, and outputting the mask A under the condition of the mask B; the regular part is input as a mask A, and the condition is mask B\A, so that the mask A is kept and output; the loss function of the model is thus defined as follows:
(1)
(2)
(3)
In the formula (1), L 1 represents a loss function of the mask repair training section; n represents the unsupervised training times in each iterative training process; a, B are chromosome instances in dataset D; l represents a binary cross entropy Binary CrossEntropy loss function, and f p is a mask repair model; m A\B represents a mask A\B; m B represents mask B; i represents an image block of the input neural network; m A represents mask A;
In the formula (2), L 2 represents a loss function of the canonical part; m B\A represents a mask B\A;
in the formula (3), L m represents a final loss function of the model; x corresponds to the bernoulli distribution with a coefficient gamma; gamma is the probability of selecting a mask repair training part in the training process;
Step S304: optimizing the weight parameters of the model through iterative training according to the neural network model determined in the step S303 and the final loss function L m of the formula (3), and finally obtaining a trained chromosome image mask repair model f p;
Step S4: predicting chromosome segmentation images;
Step S5: judging crossed occlusion of chromosomes;
Step S6: prediction of cross occlusion chromosome mask repair;
step S7: determining a shielding relationship between the crossing chromosomes;
Determining a shielding relation between the chromosomes by comparing the areas of each chromosome in the overlapped and shielded image blocks after the mask repair, wherein if the area of the mask of the chromosome is predicted to be large by the mask repair model, the chromosome is a shielded chromosome, otherwise, if the area of the mask of the chromosome is not changed or the area change is small after the mask of the chromosome is predicted by the mask repair model, the chromosome is a shielded chromosome;
step S8: replaced by mask tags that occlude chromosomes;
And for the chromosome judged to be blocked, replacing coordinates of the corresponding position in the corresponding chromosome image label by using the repaired chromosome image mask, so as to realize mask repair of the blocked chromosome.
2. A method for masked repair of occluded chromosomes based on unsupervised learning as claimed in claim 1 wherein: in the step S7, the area of the mask of the chromosome image before or after repair is calculated by the formula (8):
(8)
In the formula (8), A is the area of the chromosome mask, and i is the coordinate number in the chromosome mask; n is the number of closed coordinates of the chromosome mask being sought.
3. A method for masked repair of occluded chromosomes based on unsupervised learning as claimed in claim 1 wherein: the step S4 includes the steps of inputting the acquired metaphase image of the chromosome into the trained chromosome image example segmentation model f m obtained in the step S2, and predicting to obtain a mask label of a visible part of the chromosome and a category label of the chromosome.
4. A method for mask repair of occluded chromosomes based on unsupervised learning according to claim 3, wherein: step S5 comprises the steps of traversing the mid-phase image of the chromosome, the mask label of the visible part of the corresponding chromosome and the class label of the chromosome, judging whether the coordinates of each chromosome mask are crossed or not according to whether a crossed region exists between the closed coordinate curves of the chromosome mask labels by using a vector operation library Shapely algorithm, and judging whether a shielding condition exists between the chromosomes or not; if there is a blocking condition, the maximum and minimum coordinates of the crossed chromosome masks are obtained from the equation (4), the equation (5), the equation (6), and the equation (7), namely:
(4)
(5)
(6)
(7)
(X 1,y1),(x1,y2),(x2,y1),(x2,y2) is the coordinates of the region in the image where the chromosome overlap occlusion occurs, (x i,yi) is the coordinates in the mask labels of all crossed chromosomes; and cutting out the image blocks overlapped and shielded by the chromosome according to the region coordinates overlapped and shielded by the chromosome.
5. The method for mask repair of occluded chromosomes based on unsupervised learning of claim 4 wherein: the step S6 includes steps of traversing all crossed chromosomes in the image blocks overlapped and shielded by the chromosomes in the step S5, respectively inputting the mask repair models f p obtained in the step S3, and predicting to obtain repair masks of all crossed chromosome images.
6. A method for masked repair of occluded chromosomes based on unsupervised learning as claimed in claim 1 wherein: in the step S303, the neural network model is selected as a residual neural network ResNet, a dense-connectivity neural network DenseNet, or a U-Net.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202410330149.XA CN117934337B (en) | 2024-03-22 | 2024-03-22 | Method for mask repair of blocked chromosome based on unsupervised learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202410330149.XA CN117934337B (en) | 2024-03-22 | 2024-03-22 | Method for mask repair of blocked chromosome based on unsupervised learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN117934337A CN117934337A (en) | 2024-04-26 |
CN117934337B true CN117934337B (en) | 2024-05-28 |
Family
ID=90766834
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202410330149.XA Active CN117934337B (en) | 2024-03-22 | 2024-03-22 | Method for mask repair of blocked chromosome based on unsupervised learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117934337B (en) |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113658150A (en) * | 2021-08-23 | 2021-11-16 | 西安交通大学 | Chromosome automatic segmentation and classification method based on deep learning |
CN117152147A (en) * | 2023-10-31 | 2023-12-01 | 杭州德适生物科技有限公司 | Online chromosome collaborative analysis method, system and medium |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP3432198B1 (en) * | 2017-07-19 | 2024-04-17 | Tata Consultancy Services Limited | Crowdsourcing and deep learning based segmenting and karyotyping of chromosomes |
TWI765262B (en) * | 2019-06-26 | 2022-05-21 | 長佳智能股份有限公司 | Method for training a separation model for separation of overlapping chromosome based on simulation, and method and system for implementing separation of overlapping chromosome using the separation model |
-
2024
- 2024-03-22 CN CN202410330149.XA patent/CN117934337B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113658150A (en) * | 2021-08-23 | 2021-11-16 | 西安交通大学 | Chromosome automatic segmentation and classification method based on deep learning |
CN117152147A (en) * | 2023-10-31 | 2023-12-01 | 杭州德适生物科技有限公司 | Online chromosome collaborative analysis method, system and medium |
Also Published As
Publication number | Publication date |
---|---|
CN117934337A (en) | 2024-04-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111369572B (en) | Weak supervision semantic segmentation method and device based on image restoration technology | |
CN109919934B (en) | Liquid crystal panel defect detection method based on multi-source domain deep transfer learning | |
CA2598686A1 (en) | Intelligent importation of information from foreign application user interface using artificial intelligence | |
CN112101364B (en) | Semantic segmentation method based on parameter importance increment learning | |
CN114341880A (en) | Techniques for visualizing operation of neural networks | |
CN105528610A (en) | Character recognition method and device | |
CN110084136A (en) | Context based on super-pixel CRF model optimizes indoor scene semanteme marking method | |
CN114258550A (en) | Techniques for modifying the operation of a neural network | |
CN114330109A (en) | Interpretability method and system of deep reinforcement learning model under unmanned scene | |
JP6988995B2 (en) | Image generator, image generator and image generator | |
CN114067268A (en) | Method and device for detecting safety helmet and identifying identity of electric power operation site | |
CN117934337B (en) | Method for mask repair of blocked chromosome based on unsupervised learning | |
CN116758539B (en) | Embryo image blastomere identification method based on data enhancement | |
CN109583584B (en) | Method and system for enabling CNN with full connection layer to accept indefinite shape input | |
CN116228623B (en) | Metal surface defect detection method, equipment and storage medium based on isomorphism regularization self-supervision attention network | |
CN116894113A (en) | Data security classification method and data security management system based on deep learning | |
CN116778164A (en) | Semantic segmentation method for improving deep V < 3+ > network based on multi-scale structure | |
CN115761511A (en) | SAR image target detection method combined with high-confidence knowledge distillation | |
JP7468088B2 (en) | Image processing system and image processing program | |
CN111429436B (en) | Intrinsic image analysis method based on multi-scale attention and label loss | |
CN114443856A (en) | Automatic fault knowledge graph creating method and device for fault tree picture | |
JP7167990B2 (en) | Method, device, system and program for controlling robot, and storage medium | |
CN117593648B (en) | Remote sensing target building extraction method based on weak supervision learning | |
WO2023189662A1 (en) | Image processing/analyzing device, and image processing/analyzing method | |
CN112434748B (en) | Interactive data marking method under weak supervision environment |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant |