CN115496716A

CN115496716A - Single and double micronucleus cell image detection method based on deep learning and related equipment

Info

Publication number: CN115496716A
Application number: CN202211076978.7A
Authority: CN
Inventors: 杨炼; 陈忠雄; 崔玉峰
Original assignee: Shanghai Beion Pharmaceutical Technology Co ltd
Current assignee: Shanghai Beion Pharmaceutical Technology Co ltd
Priority date: 2022-09-05
Filing date: 2022-09-05
Publication date: 2022-12-20

Abstract

The invention discloses a method for detecting single and double micronucleus cell images based on deep learning and related equipment, wherein the method comprises the following steps: an obtaining step, obtaining a true positive micronucleus omics image, wherein the true positive micronucleus omics image carries key indexes; a detection step, namely inputting the true positive micronucleus images into a micronucleus detection model for processing so as to remove false positive micronucleus samples which do not accord with a preset area relationship, and outputting micronucleus model images; and calculating by the micronucleus detection model by using the micronucleus area of the output micronucleus model image and the micronucleus area of the true positive micronucleus model image. And a counting step, counting the detection data of the key indexes. The invention provides a microkernel image detection method based on deep learning and an image processing algorithm, which realizes large-scale, multi-index, high-efficiency, high-precision and visual detection of microkernels and effectively detects cancer risks.

Description

Single-double micronucleus cell image detection method based on deep learning and related equipment

Technical Field

The invention relates to a DNA analysis technology, in particular to a method for detecting a single-double micronucleus cell image based on deep learning and related equipment.

Background

Micronucleus is not only widely used for the type and pattern of genetic effects of exogenous chemical substances (such as drugs, food additives, cosmetics, environmental pollutants, etc.), but also plays an important role in cancer risk screening and risk prediction.

The corresponding detection of micronuclomics can comprehensively show various problems of DNA damage and repair, chromosome breakage or loss, gene division instability, double centromere chromosomes, apoptosis, necrosis, cell growth inhibition and the like, wherein the ratio of the NPBs (nuclear proton bridge proteins) to the MNI (microribonucleic acid) can be used as a biomarker for judging chromosome breakage loss and genetic damage repair, and the micronucleus rate of cells is the proportion of cells containing micronucleus in 2000 dinuclear cells, so that the detection of the BN (binuclear cells) is very important for counting micronucleus.

The existing micronucleus omics detection method for binuclear cells is a counting method of a computer image analysis system based on an image processing algorithm, wherein hardware such as a microscope and a camera is used for obtaining images, and then preprocessing such as noise reduction and edge detection is carried out on the images by using filtering algorithms such as sharpening. And then, extracting the region of interest by adopting algorithms such as improved watershed, seed region growth, iterative threshold segmentation and the like according to characteristic rules such as the size, the shape, the length-width ratio, the relative concave-convex depth, the color and the like of the microkernel, and realizing the counting of the microkernel. Meanwhile, the area of the micro-core needs to meet 1/20 to 1/3 of the area of the main core, and if the target detection model is only used for detection, the detected target cannot pass the secondary verification of the image processing method after being detected.

Therefore, the technical problems that in the prior art, the true positive micronucleus images in the micronucleus images cannot be accurately screened and the detection precision cannot be guaranteed exist.

Disclosure of Invention

The invention aims to achieve the technical purpose of improving the accuracy and precision of single and double micronucleus cell detection.

A method for detecting a single-double micronucleus cell image based on deep learning comprises the following steps:

the method comprises the following steps:

an obtaining step, namely obtaining a true positive micronucleus omics image, wherein the true positive micronucleus omics image carries key index information;

a detection step, namely inputting the true positive micronucleus images into a micronucleus detection model for processing so as to eliminate false positive micronucleus samples which do not accord with a preset area relation, and outputting micronucleus model images;

and calculating by the micronucleus detection model by using the micronucleus area of the output micronucleus model image and the micronucleus area of the true positive micronucleus image.

And a counting step, counting the detection data of the key indexes.

Preferably, the calculating step is implemented specifically as:

and respectively segmenting the main nucleus area and the micronucleus area in the computing areas of the output image and the true positive micronucleus image by using an image processing algorithm, and respectively comparing the main nucleus area and the micronucleus area.

Preferably, the key indicators include micronucleus MNI, nucleoplasmic bridge NPB, nuclear bud nbud and normal binuclear cell BN.

Preferably, the method further comprises the following steps: a pretreatment step comprising:

carrying out position marking on the key indexes in the true positive micronucleus images and/or the microkaryotries images to be detected, wherein the position marking is based on image grid units;

and performing image segmentation and data enhancement operations on the microkaryotic image.

Preferably, image segmentation and data enhancement operations are performed on the microkaryotic image, specifically implemented as:

slicing the entire microkaryotic image according to a predetermined size by using a sliding window method to obtain a plurality of sub-image pixel blocks;

and performing rotation, movement, turnover and scaling operations on the sub-image pixel blocks.

Preferably, the present invention further comprises: the model training step is specifically realized as follows:

obtaining model parameters of key indexes of the true positive microkernel image, wherein the model parameters comprise: a loss function, the loss function comprising: a confidence loss function and a bounding box regression loss function;

fitting the positions of all key indexes of the micronucleus by adopting a deep learning technology, wherein the deep learning technology can adopt a DropBlock mode to spatially reduce image features;

inputting the preprocessed micronucleus omics image to be detected, and extracting by adopting a hierarchical feature fusion strategy; the CSP module constructs a main neural network to extract image characteristics, and training is completed by intercepting gradient flow;

when a new true positive micronucleus image detection requirement exists, inputting the preprocessed true positive micronucleus image into a micronucleus detection model and inputting the pretreated true positive micronucleus image into a micronucleus detection model so as to obtain the output micronucleus model image.

Preferably, the CSP module constructs a main neural network to extract image features, and includes:

the method comprises the steps that (1) feature information of a high layer is transmitted and fused in a top-down sampling mode through a feature pyramid network FPN;

and adding a bottom-up Path Aggregation Network (PAN) at the output end of the FPN to supplement the position characteristics of the single and double micronucleus cells and transmitting the low-level strong positioning characteristics upwards.

Preferably, the detection data is: micronucleus MNI, nuclear-cytoplasmic-bridge NPB, nuclear bud burst and normal binuclear cell BN number, micronucleus cell rate and ratio of nuclear-cytoplasmic-bridge NPB to micronuclei.

The invention provides a deep learning-based single-double micronucleus cell image detection method, which inputs a preprocessed true-positive micronucleus image into a deep learning-based micronucleus detection model for training so as to adapt to the characteristics of each index of micronucleus, respectively obtains model parameters of each index, and inputs the preprocessed micronucleus image to be detected into the trained micronucleus detection model so as to obtain the output of a micronucleus model image. The image processing algorithm is used for comparing the micronucleus area of the output image with the micronucleus area of the true-positive micronucleus omics image, eliminating false-positive micronucleus samples which do not accord with a preset area relation, counting the screened micronucleus index data, and forming a visual analysis report, so that the high-efficiency and high-precision visual detection of large-scale and multi-index micronucleus and the effective detection of cancer risk are realized.

Drawings

The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the invention and not to limit the invention. In the drawings:

FIG. 1a is a schematic structural diagram of a structure of a micronucleus detection model senet attention mechanism in a deep learning-based single-double micronucleus image detection method according to an embodiment of the present invention;

FIG. 1b is a flowchart illustrating a deep learning-based method for detecting mononuclear cell and mononuclear cell images according to an embodiment of the present invention;

FIG. 1c is a schematic diagram of a key index of a deep learning-based mononuclear cell image detection method according to an embodiment of the present invention;

FIG. 1d is a schematic diagram illustrating the detection effect of each index micronucleus region area of the deep learning-based image detection method for single and double micronucleus cells according to an embodiment of the present invention;

FIG. 1e is a schematic diagram showing the microkaryotic detection accuracy effect of the deep learning-based image detection method for single and double micronucleus cells according to one embodiment of the present invention;

FIG. 2 is a flowchart illustrating a method for detecting mononuclear cell and mononuclear cell images based on deep learning according to an embodiment of the present invention;

FIG. 3 is a flowchart illustrating the image segmentation and data enhancement processes in the deep learning-based mononuclear cell image detection method according to an embodiment of the present invention;

fig. 4 is a diagram illustrating operation steps of data rotation, movement, flipping, scaling, and the like in the deep learning-based single and double micronucleus image detection method according to an embodiment of the present invention.

Detailed Description

In order to make the technical problems, technical solutions and advantageous effects to be solved by the present invention clearer and clearer, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not limit the invention.

In recent years, deep learning has made a great progress in medical image analysis, providing new means and techniques for medical research. Rather than forming a quantitative indicator, qualitative analysis is a generalization of the properties and characteristics of the microkernel images. Therefore, the results of qualitative analysis cannot be reproduced and are greatly influenced by subjective factors. The quantitative analysis means that a mathematical model is established according to statistical data, and various quantitative indexes of micronucleus are calculated.

The traditional single and double micronucleus cell analysis method relies on manual analysis, a deep neural network model is built by adopting a deep learning method, the training of a color micronucleus cell detection model is completed through a large number of single and double micronucleus cell samples, and the analysis efficiency can be greatly improved by adopting the model to carry out color micronucleus cell analysis. The method for automatically detecting the estimated single and double micronucleus cell dosage can replace the conventional manual analysis and dosage estimation method, and the error caused by automatic analysis is smaller and the analysis speed is 30 times faster.

Traditional machine learning algorithms and deep learning algorithms can be used for quantitative analysis of microkernel images, but rely on artificial design of feature expression, namely extraction and selection of the shape, size and texture of an image, elimination of redundant features and acquisition of an optimal feature set. Furthermore, the manual selection of various micronuclomics indices depends on a lot of expertise, especially common features of the micronucleus and nuclear bud processes. Each of the indices of micronuclei, nuclear bud processes, and nuclear bridges has different characteristics in shape, color, size, and refractive index. Even under the condition that index types are the same, images still have certain differences in size, structure and color depth, and the manually designed extraction rules are difficult to cover the comprehensive features of the images, so that the use of the extraction rules is limited.

Deep learning through convolutions, pools, and other hierarchical network structures, deep learning can transport a large number of labeled microkernel images from an input layer to an output layer. The network autonomous learning can obtain the high-dimensional and low-dimensional features of the microkernel image. Compared with other machine learning algorithms, the deep learning algorithm has stronger feature extraction capability on large data samples. Based on the micro-nuclear big data microscopic image accumulated in a laboratory, the deep micro-nuclear analysis can fully exert the advantages of the deep micro-nuclear analysis in a big data sample, promote the development of the quantitative and qualitative analysis of the micro-nuclear and is beneficial to promoting the application of the micro-nuclear detection in the research such as early cancer screening, risk assessment and the like.

Therefore, based on the reasons, the invention provides a microkaryotic image detection method based on deep learning and an image processing algorithm, so that large-scale, multi-index, efficient, high-precision and visual detection of microkaryotics is realized, and the cancer risk is effectively detected.

Before the embodiment is performed, it should be noted that, in the network constructed by the present invention, referring to fig. 1a, a deep convolutional neural network model and a one-stage target detection model are adopted, and a Selayer attention mechanism is added; according to the method, the CSP module is adopted to construct the trunk neural network to extract the characteristics, so that the deep characteristics of the mononuclear cell and the mononuclear cell can be more effectively extracted, meanwhile, the gradient flow is cut off to prevent excessive repeated gradient information from being used for model training, and the extraction capability of the trunk neural network on the characteristics of the mononuclear cell and the mononuclear cell is improved by adopting a hierarchical characteristic fusion strategy.

Meanwhile, for convenience of explanation, the cell micronucleus image is a lymphocyte micronucleus image, but the method for detecting a single-double micronucleus image can also be applied to other cell micronucleus images, and is not limited to this.

Specifically, referring to fig. 1b, a method for detecting a mononuclear cell image based on deep learning includes the following steps:

s11, an obtaining step, namely obtaining a true positive micronucleus image, wherein the true positive micronucleus image carries key index information;

alternatively, the microkernel image to be detected may be obtained through an automatic scanning microscope, a digital section scanner, a CCD optical microscope, and the like, and the true positive microkernel image may be preprocessed according to the method for preprocessing the true positive microkernel image in the foregoing embodiment.

Preferably, the key indicators include micronucleus MNi, nucleoplasmic bridge NPBs, nuclear bud NBUD and normal binuclear cell BN, see fig. 1c.

S12, a detection step, namely inputting the true positive micronucleus images into a micronucleus detection model for processing so as to eliminate false positive micronucleus samples which do not accord with a preset area relation, and outputting micronucleus model images;

the output micronuclomic model image may be considered as an image with a high degree of similarity to a true positive micronuclomic image.

In the embodiment of the invention, a target detection model is constructed by utilizing a deep learning technology, and a deep convolution neural network model for detecting the micronucleus index is trained. After the model is trained, the position of each micronucleus index in the micronucleus image can be obtained, so that the number of each index can be counted, the counting speed is high, and the detection efficiency is high.

And S13, calculating by using the micronucleus area of the output micronucleus model image and the micronucleus area of the true positive micronucleus model image by using the micronucleus detection model.

Preferably, the calculating step is implemented specifically as:

By way of example, referring to fig. 1d, a schematic diagram of the detection effect obtained for each index microkernel area according to the embodiment of the present invention is shown; by using the preset area relation that the micronucleus area accounts for 1/20 to 1/3 of the main nucleus area, if only the target detection model is used for detection, the detected target cannot be verified twice by the detected image processing method, so that the true positive micronucleus images in the micronucleus images to be detected cannot be screened accurately, and the detection precision cannot be guaranteed. Therefore, in the present embodiment, the accuracy of micronucleus image detection is further improved by using an image processing algorithm to eliminate false positive micronucleus samples that do not conform to the predetermined area relationship.

It will be appreciated that micronucleus screening is required prior to counting the number of micronucleus indices. Because the relation between the micronucleus area and the preset area of the main nucleus area is in a certain range, extracting the micronucleus cell target area output by the target detection model, and utilizing an image processing algorithm to divide and calculate the micronucleus area and the main nucleus area, eliminating false positive micronucleus samples which do not conform to the area relation, and ensuring the accuracy of each index statistic of the micronucleus.

And S14, a step of statistics, namely statistics of the detection data of the key indexes.

And counting the selected micronucleus index data and forming a visual analysis report. FIG. 1e is a schematic diagram of the effect of micronucleus detection accuracy provided by an embodiment of the present invention:

in the specific implementation process, the accuracy of each index data of the micronucleus can be detected firstly. For example, in this embodiment, the average detection accuracy of the micronucleus MNI, the nucleoplasmic bridge NPB and the nuclear bud nbud is 0.982, and the detection accuracy of the normal binuclear cell BN is 0.961, and then the filtration index data of the micronucleus group can be counted to form a visual analysis report.

Specifically, the data of each index of the screened micronucleus comprise: the number of indices, micronuclear cell rate and the ratio of nuclear bridges NPB to micronuclei. According to the statistical data of each index of the micronuclomics, the micronucleus number of the binuclear cells in 1000 binuclear cells is the micronucleus cell rate, and the ratio of the nuclear-cytoplasmic bridge NPB number to the micronucleus number is the ratio of the nuclear-cytoplasmic bridge NPB number to the micronucleus. A visual analysis report is formed from more than one sentence of information to assess cancer risk.

s21, obtaining model parameters of key indexes of the true positive microkernel image, wherein the model parameters comprise: a loss function, the loss function comprising: a confidence loss function and a bounding box regression loss function;

the loss function is used to estimate the inconsistency between the predicted and actual values of the model. It is a non-negative real-valued function. The smaller the loss function, the better the robustness of the model.

A loss function of model training is divided into a confidence coefficient loss function and a boundary frame regression loss function, the learning capability of the model to a characteristic complex region in an input image is enhanced by adding influence factors gamma and alpha in the confidence coefficient loss function, and the accuracy of regression prediction of the model to a double single-double-micro nuclear cell boundary frame is improved according to the width-to-height ratio consistency of a real boundary frame and a predicted boundary frame by adding the influence factor delta in the boundary frame regression loss function. Finally, the proportion of the two loss values in the total loss value is balanced through a weight coefficient lambda, and the designed loss function can train a single-double micronucleus cell detection model more effectively.

S22, fitting the positions of all key indexes of the micronucleus by adopting a deep learning technology, wherein the deep learning technology can adopt a DropBlock mode to spatially reduce the image characteristics;

the traditional Drop out mode does not consider the spatial characteristics of image characteristics when performing characteristic reduction, so that the effect of improving the robustness of a model is not obvious.

S23, inputting the preprocessed micronucleus images to be detected, and extracting by adopting a hierarchical feature fusion strategy; the CSP module constructs a main neural network to extract image characteristics, and completes training by intercepting a gradient stream;

and S24, when a new true positive micronucleus image detection requirement exists, inputting the preprocessed true positive micronucleus image into the micronucleus detection model and inputting the input micronucleus detection model into the micronucleus detection model so as to obtain the output micronucleus model image.

Training the data in an iterative mode to obtain a model prediction test data set, and calculating the precision of the deep learning model on the test data set. The loss value of the training data set gradually decreases as the number of training increases. As the loss value decreases, the range tends to stabilize and the model training succeeds.

In this embodiment, the preprocessed true positive microkernel image is input into a microkernel detection model based on deep learning to be trained so as to adapt to the position of each index of the microkernel, model parameters of each index are respectively obtained, and the microkernel image to be detected is preprocessed and input into the trained microkernel detection model to obtain an output image. The image processing algorithm is used for comparing the microkernel area of the output image with the microkernel area of the true positive microkernel image, eliminating false positive microkernel samples which do not accord with the preset area relation, counting the screened microkernel index data and forming a visual analysis report, and is beneficial to evaluating cancer risks.

And (3) a model training step, namely combining a deep learning technology with an image processing algorithm to construct a target detection network model, automatically finding the position of each index of the micronucleus group in the micronucleus image, extracting the micronucleus cell area on the basis, and performing secondary screening by using the image processing algorithm. The method can carry out numerical detection on various micronucleus indexes in the micronucleus image, and calculate key information such as micronucleus cell rate, nuclear proton bridge NPB and micronucleus ratio and the like.

Preferably, the CSP module constructs a backbone neural network to extract image features, including:

the characteristics are extracted by adopting the spatial pyramid pooling operation, so that the receiving range of the subsequent network structure on the trunk characteristics is expanded, and the characteristics of the mononuclear cells and the mononuclear cells with different sizes are fused, so that the model designed by the invention can have higher detection accuracy on the mononuclear cells and the mononuclear cells with different sizes.

And adding a bottom-up path aggregation network PAN at the output end of the FPN to supplement the position characteristics of the single and double micronucleus cells and transmitting the low-level strong localization characteristics upwards.

The method comprises the steps of adopting a feature pyramid network FPN to transmit and fuse feature information of a high layer in an up-sampling mode from top to bottom, then adding a bottom-up path aggregation network PAN at the output end of the FPN to supplement the position features of single and double micronucleus cells, and transmitting the strong positioning features of a low layer upwards. By adopting a feature fusion strategy combining FPN and PAN, the accuracy of the model for detecting the single and double micronucleus cells is greatly improved.

Referring to fig. 3, before inputting the true positive micronucleus image into the micronucleus detection model, the method further comprises: the method comprises a preprocessing step, wherein the preprocessing step of the true positive micronucleus images comprises the following steps:

s31, carrying out position marking on the key indexes in the true positive micronucleus image and/or the microkaryotype image to be detected, wherein the position marking is based on an image grid unit;

the method quantifies the position characteristics of the micronucleus cells, divides the micronucleus cell image to be analyzed into S multiplied by S grid units, quickly analyzes whether the micronucleus cells are single or double in the image by predicting the probability of the existence of the micronucleus cells in each grid unit, and then predicts the central coordinates and width and height information of the micronucleus cells relative to the grid units to further determine the bounding boxes of the micronucleus cells.

And S32, performing image segmentation and data enhancement operation on the micronucleus images.

It should be noted that image segmentation includes, but is not limited to: a threshold-based segmentation method, a region-based segmentation method, an edge-based segmentation method, a segmentation method based on a specific theory such as cluster analysis, gene coding, wavelet transform, neural network, and the like. In particular, the Otsu algorithm (Otsu method or the maximum inter-class variance method) can be used for binary segmentation of micronucleus images.

Referring to fig. 4, the image segmentation and data enhancement operations are performed on the microkaryotic image, which are specifically implemented as:

s41, slicing the whole micronucleus image according to a preset size by using a sliding window method to obtain a plurality of sub-image pixel blocks;

the image slicing operation includes: the entire microkaryotic image is sliced according to a predetermined size by using a sliding window method to obtain a plurality of sub-image blocks, and the predetermined size employs 512 × 512 pixels.

And S42, performing rotation, movement, turnover and scaling operations on the sub-image pixel blocks.

The data enhancement operation includes: performing rotation, move, flip, and zoom operations on the sub-image tiles, and saving all sub-image tile data before and after the data enhancement operation.

The technical advantages of the above embodiments are as follows:

1. the traditional single and double micronucleus cell analysis method relies on manual analysis, a deep neural network model is built by adopting a deep learning method, the training of a color micronucleus cell detection model is completed through a large number of single and double micronucleus cell samples, and the analysis efficiency can be greatly improved by adopting the model to carry out color micronucleus cell analysis. The method for automatically detecting the estimated single and double micronucleus cell dosage can replace the conventional manual analysis and dosage estimation method, and the error caused by automatic analysis is smaller and the analysis speed is 30 times faster.

2. The method quantifies the position characteristics of the micronucleus cells, divides the micronucleus cell image to be analyzed into S multiplied by S grid units, quickly analyzes whether the micronucleus cells are single or double in the image by predicting the probability of the existence of the micronucleus cells in each grid unit, and then predicts the central coordinates and width and height information of the micronucleus cells relative to the grid units to further determine the bounding boxes of the micronucleus cells.

3. According to the method, the CSP module is adopted to construct the trunk neural network to extract the characteristics, so that the deep characteristics of the mononuclear cells and the binuclear cells can be more effectively extracted, the excessive repeated gradient information is prevented from being used for training the model by intercepting the gradient flow, and the extraction capability of the trunk neural network on the characteristics of the mononuclear cells and the binuclear cells is improved by adopting a hierarchical characteristic fusion strategy.

4. The traditional Drop out mode does not consider the spatial characteristics of image characteristics when performing characteristic reduction, so that the effect of improving the robustness of the model is not obvious.

5. The characteristics are extracted by adopting the spatial pyramid pooling operation, so that the receiving range of the subsequent network structure on the trunk characteristics is expanded, and the characteristics of the mononuclear cells and the mononuclear cells with different sizes are fused, so that the model designed by the invention can have higher detection accuracy on the mononuclear cells and the mononuclear cells with different sizes.

6. The model designed by the invention adopts a feature pyramid network FPN to transmit and fuse feature information of a high layer in an up-sampling mode from top to bottom, and then adds a bottom-up path aggregation network PAN at the output end of the FPN to supplement the position features of the mononuclear cells and the mononuclear cells, so as to transmit the strong positioning features of a low layer upwards. By adopting a feature fusion strategy combining FPN and PAN, the accuracy of the model for detecting the single and double micronucleus cells is greatly improved.

7. The method for detecting the mononuclear cells and the binuclear cells can realize the end-to-end automatic analysis of the mononuclear cell and the binuclear cell images by inputting the mononuclear cell and the binuclear cell images to be analyzed, directly mark the mononuclear cells and the binuclear cells on the input images, and count the number of the mononuclear cells and the binuclear cells. The single and double micronucleus cell detection method has the advantages of higher speed, higher detection accuracy and stronger robustness.

Meanwhile, the invention also discloses a hardware carrier supporting the method in the embodiment, which comprises the following steps: a computing device, at least one processor; and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform any one of the methods for deep learning based image detection of mononuclear cells as described above. And a readable medium storing computer executable instructions for executing the above method for detecting single and double micronucleus images based on deep learning.

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

While the preferred embodiments of the present application have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including the preferred embodiment and all changes and modifications that fall within the scope of the present application. It will be apparent to those skilled in the art that various changes and modifications may be made in the present application without departing from the spirit and scope of the application. Thus, if such modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is intended to include such modifications and variations as well.

Claims

1. A method for detecting single and double micronucleus cell images based on deep learning is characterized by comprising the following steps: the method comprises the following steps:

a detection step, namely inputting the true positive micronucleus images into a micronucleus detection model for processing so as to remove false positive micronucleus samples which do not accord with a preset area relationship, and outputting micronucleus model images;

calculating, by the micronucleus detection model, the micronucleus area of the output micronucleus model image and the micronucleus area of the true positive micronucleus model image;

and a counting step, counting the detection data of the key indexes.

2. The method for detecting the image of the mononuclear and binuclear cells based on the deep learning of claim 1, wherein the calculating step is specifically realized as follows:

3. The deep learning-based image detection method for mononuclear and binuclear mononuclear cells according to claim 1 or 2, wherein the key indicators include micronucleus MNI, nuclear bridge NPB, nuclear bud nbud and normal binuclear cell BN.

4. The deep learning-based single-double micronucleus image detection method of claim 1, characterized in that it further comprises: a pretreatment step comprising:

5. The deep learning-based single and double micronucleus image detection method of claim 4, characterized in that image segmentation and data enhancement operations are performed on micronucleus images, specifically implemented as:

and performing rotation, movement, flipping and scaling operations on the sub-image pixel blocks.

6. The deep learning-based single-double micronucleus image detection method of claim 1, characterized in that it further comprises: the model training step is specifically realized as follows:

obtaining model parameters of key indexes of the true positive micronucleus image, wherein the model parameters comprise: a loss function, the loss function comprising: a confidence loss function and a bounding box regression loss function;

inputting the preprocessed micronucleus images to be detected, and extracting by adopting a hierarchical feature fusion strategy; the CSP module constructs a main neural network to extract image characteristics, and completes training by intercepting a gradient stream;

when a new true positive micronucleus image detection requirement exists, inputting the preprocessed true positive micronucleus image into a micronucleus detection model and inputting the preprocessed true positive micronucleus image into a micronucleus omics detection model so as to obtain the output micronucleus model image.

7. The deep learning-based single-double micronucleus image detection method according to claim 3, wherein the CSP module constructs a trunk neural network for image feature extraction, and the method comprises the following steps:

8. The method for detecting mononuclear and binuclear cell images based on deep learning of claim 3, wherein the detection data is: micronucleus MNI, nuclear-cytoplasmic-bridge NPB, nuclear bud burst and normal binuclear cell BN number, micronucleus cell rate and ratio of nuclear-cytoplasmic-bridge NPB to micronuclei.

9. A computing device characterized by at least one processor; and

a memory communicatively coupled to the at least one processor; wherein, the first and the second end of the pipe are connected with each other,

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-8.

10. A readable medium storing computer-executable instructions for performing the method for detecting mononuclear cell and binuclear mononuclear cell images based on deep learning according to claims 1 to 8.