CN114863163A - Method and system for cell classification based on cell image - Google Patents
Method and system for cell classification based on cell image Download PDFInfo
- Publication number
- CN114863163A CN114863163A CN202210347243.7A CN202210347243A CN114863163A CN 114863163 A CN114863163 A CN 114863163A CN 202210347243 A CN202210347243 A CN 202210347243A CN 114863163 A CN114863163 A CN 114863163A
- Authority
- CN
- China
- Prior art keywords
- cell
- deep learning
- learning model
- image
- database
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 48
- 238000013136 deep learning model Methods 0.000 claims abstract description 85
- 238000012549 training Methods 0.000 claims abstract description 35
- 238000004364 calculation method Methods 0.000 claims abstract description 14
- 238000012545 processing Methods 0.000 claims description 13
- 238000002372 labelling Methods 0.000 claims description 11
- 238000005259 measurement Methods 0.000 claims description 11
- 238000001514 detection method Methods 0.000 claims description 9
- 238000000605 extraction Methods 0.000 claims description 8
- 238000005070 sampling Methods 0.000 claims description 8
- 238000010586 diagram Methods 0.000 description 8
- 230000006870 function Effects 0.000 description 8
- 238000013135 deep learning Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 4
- 238000013528 artificial neural network Methods 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 2
- 239000013598 vector Substances 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000003064 k means clustering Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000000691 measurement method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 231100000915 pathological change Toxicity 0.000 description 1
- 230000036285 pathological change Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/69—Microscopic objects, e.g. biological cells or cellular parts
- G06V20/698—Matching; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/761—Proximity, similarity or dissimilarity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
- G06V10/763—Non-hierarchical techniques, e.g. based on statistics of modelling distributions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Multimedia (AREA)
- Computing Systems (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method and a system for cell classification based on a cell image, wherein the embodiment of the invention adopts a deep learning model obtained by training to classify cells in the cell image, and the method comprises the following steps: providing a cell image, extracting cell characteristics from the cell image, and inputting the cell characteristics into a deep learning model obtained through training; the deep learning model respectively carries out similarity calculation on the cell characteristics and various different cell characteristics in a cell database to obtain similarity values between the cell characteristics and the different cell characteristics, wherein the different cell characteristics in the cell database are obtained by carrying out cell type matching and clustering based on cell image samples when the deep learning model is trained; and according to the high-low order of the similarity values, taking a set number of categories corresponding to the category cell characteristics as cell classification results in the cell images. Therefore, when the deep learning model obtained by training is adopted to classify the cells in the cell image, the accuracy of cell classification is improved.
Description
Technical Field
The present application relates to the field of computer image processing technologies, and in particular, to a method and a system for cell classification based on cell images.
Background
In recent years, due to the strong modeling capability of deep learning and neural networks and the characteristics of large information amount and rich features of medical images, the application of a deep learning technology to the medical images is becoming mature, and an image processing technology based on the deep learning and the neural networks is adopted to analyze cell images and assist doctors in screening pathological changes of cells, so that the deep learning technology and the neural networks are gradually a research hotspot in the world.
Thanks to the publication of large-scale image datasets and the development of high-performance computing, deep learning-based cell image recognition methods have made great progress. At present, the mainstream medical image reading system based on deep learning performs imaging processing on a cell smear of a specific part, performs cell detection and positioning based on the obtained cell image, and accurately identifies whether a cell is diseased or not to obtain a cell detection result. In a cell image recognition task in the field of medical images, a convolutional neural network is the most common classification method, but due to the difference of a slice making mode of a cell smear and scanning equipment, the diversity and irregularity of cell forms, cell overlapping and the like, the superposition of various factors makes the recognition of cell images difficult.
Although the specific task of cell image recognition by adopting the deep learning model greatly exceeds that of the machine learning model adopted before, the problems of poor generalization, low repeatability and the like exist in the practical application process of many deep learning models in the industry at present, and the application scenarios are similar to those of the traditional application scenarios. In the field of medical image identification, the cell image difference is large under the influence of different slice making modes and scanning equipment, when a set of basic deep learning model is applied to cell classification of the cell image, the cell image classification method is difficult to adapt to all slice making modes, and the accuracy rate of cell classification is low. Further, if the adaptive deep learning model is trained on cell images of different classes of cells, separate training is required on a variety of different training data, and the efficiency of training the deep learning model is low.
Disclosure of Invention
In view of the above, embodiments of the present disclosure provide a method for classifying cells based on a cell image, which can improve accuracy of cell classification when classifying cells in the cell image.
The embodiment of the invention also provides a system for classifying cells based on the cell image, which can improve the accuracy of cell classification when the cells in the cell image are classified.
The embodiment of the application is realized as follows:
a method of cell classification based on cell images, the method comprising:
providing a cell image, extracting cell characteristics from the cell image, and inputting the cell characteristics into a deep learning model obtained through training;
the deep learning model respectively carries out similarity calculation on the cell characteristics and various different cell characteristics in a cell database to obtain similarity values between the cell characteristics and the different cell characteristics, wherein the different cell characteristics in the cell database are obtained by carrying out cell type matching and clustering based on cell image samples when the deep learning model is trained;
and according to the sequence of the similarity values, taking a set number of categories corresponding to the category cell characteristics as cell classification results in the cell images.
Preferably, the different cell types of the cell database are cell type matching based on cell image samples when training the deep learning model, and the cell type matching comprises:
labeling cell types for cell features in a cell database;
extracting cell sample characteristics from the cell image;
and measuring the similarity of the cell sample characteristics and the cell characteristics marked with the cell types to obtain a difference value between the cell sample characteristics and the cell characteristics marked with the cell types.
Preferably, in the training of the deep learning model, the cell class matching based on the cell image sample is performed by a metric learning module in the deep learning model.
Preferably, the different classes of cell features in the cell database are clustered based on cell image samples when training the deep learning model, and the clustering includes:
and uniformly sampling the cell characteristics under the corresponding cell types based on the difference value between the cell characteristics marked with the cell types, and performing clustering processing based on the selected clustering center to obtain the cell characteristics of each cell type.
Preferably, the different classes of cell features in the cell database are clustered by a feature clustering module in the deep learning model based on cell image samples when the deep learning model is trained.
Preferably, said labeling cell classes for cell features in a cell database comprises:
segmenting a cell image sample obtained by scanning a cell smear sample into a plurality of image sample slices;
detecting the characteristics of a single cell sample from a plurality of image sample slices by adopting a cell detection method;
and carrying out cell type labeling on the characteristics of the single cell sample obtained by detection.
Preferably, the deep learning model calculates similarity between the cell features and a plurality of different types of cell features in a cell database, to obtain similarity values between the cell features and the different types of cell features, and the similarity values are obtained by a trained feature matching network in the deep learning model.
A system for cell classification based on cell images, the system comprising: an extraction module, a feature matching network module and a classification result module, wherein,
the extraction module is used for extracting cell characteristics from a provided cell image and inputting the cell characteristics into a deep learning model obtained through training;
the characteristic matching network module is used for respectively carrying out similarity calculation on the cell characteristics and a plurality of different types of cell characteristics in a cell database by the deep learning model to obtain similarity values between the cell characteristics and the different types of cell characteristics, wherein the different types of cell characteristics in the cell database are obtained by carrying out cell type matching and clustering based on cell image samples when the deep learning model is trained;
and the classification result module is used for taking the classes of the cell features of the corresponding classes with set quantity as the cell classification results in the cell images according to the high-low sequence of the similarity values.
Preferably, the system further comprises:
the measurement learning module is also used for marking cell types for the cell characteristics in the cell database; extracting cell sample characteristics from the cell image; and measuring the similarity of the cell sample characteristics and the cell characteristics marked with the cell types to obtain a difference value between the cell sample characteristics and the cell characteristics marked with the cell types.
Preferably, the system further comprises: and the characteristic clustering module is used for uniformly sampling the cell characteristics under the corresponding cell types based on the difference value between the cell characteristics marked with the cell types and carrying out clustering processing based on the selected clustering center to obtain the cell characteristics of each cell type.
The embodiment of the invention adopts a deep learning model obtained by training to classify cells in cell images, and comprises the following steps: providing a cell image, extracting cell characteristics from the cell image, and inputting the cell characteristics into a deep learning model obtained through training; the deep learning model respectively carries out similarity calculation on the cell characteristics and various different cell characteristics in a cell database to obtain similarity values between the cell characteristics and the different cell characteristics, wherein the different cell characteristics in the cell database are obtained by carrying out cell type matching and clustering based on cell image samples when the deep learning model is trained; and according to the high-low order of the similarity values, taking a set number of categories corresponding to the category cell characteristics as cell classification results in the cell images. Therefore, when the deep learning model obtained by training is adopted to classify the cells in the cell image, the accuracy of cell classification is improved.
Drawings
Fig. 1 is a flowchart of a method for cell classification based on a cell image according to an embodiment of the present disclosure;
FIG. 2 is a schematic diagram of a process architecture for performing cell feature matching and classification by using a deep learning model according to an embodiment of the present disclosure
Fig. 3 is a schematic diagram of a process of metric learning performed by a metric learning module in the deep learning model according to the embodiment of the present application;
fig. 4 is a schematic diagram of an architecture for performing cell classification based on a cell image by using a deep learning model obtained through training according to an embodiment of the present disclosure;
fig. 5 is a schematic structural diagram of a system for cell classification based on a cell image according to an embodiment of the present disclosure.
Detailed Description
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments obtained by a person of ordinary skill in the art based on the embodiments in the present application without making any creative effort belong to the protection scope of the present application.
The terms "first," "second," "third," "fourth," and the like in the description and in the claims of the present application and in the drawings described above, if any, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the application described herein are, for example, capable of operation in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprising" and "having," as well as any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements explicitly listed, but may include other steps or elements not explicitly listed or inherent to such process, method, article, or apparatus.
The technical solution of the present application will be described in detail with specific examples. Several of the following embodiments may be combined with each other and some details of the same or similar concepts or processes may not be repeated in some embodiments.
In order to improve the accuracy of cell classification when classifying cells in a cell image, the embodiment of the present application adopts a deep learning model obtained by training to perform cell classification in a cell image, including: providing a cell image, extracting cell characteristics from the cell image, and inputting the cell characteristics into a deep learning model obtained through training; the deep learning model respectively carries out similarity calculation on the cell characteristics and various different cell characteristics in a cell database to obtain similarity values between the cell characteristics and the different cell characteristics, wherein the different cell characteristics in the cell database are obtained by carrying out cell type matching and clustering based on cell image samples when the deep learning model is trained; and according to the high-low order of the similarity values, taking a set number of categories corresponding to the category cell characteristics as cell classification results in the cell images.
Therefore, when the deep learning model obtained by training is adopted to classify the cells in the cell image, the accuracy of cell classification is improved.
Fig. 1 is a flowchart of a method for classifying cells based on a cell image according to an embodiment of the present disclosure, which includes the following specific steps:
102, respectively carrying out similarity calculation on the cell characteristics and a plurality of different types of cell characteristics in a cell database by the deep learning model to obtain similarity values between the cell characteristics and the different types of cell characteristics, wherein the different types of cell characteristics in the cell database are obtained by carrying out cell type matching and clustering based on cell image samples when the deep learning model is trained;
and 103, taking the categories of the cell features corresponding to the set number of categories as cell classification results in the cell images according to the high-low sequence of the similarity values.
In the embodiment of the present application, the set number is set as needed, and may also be set to 1. In this case, the cell class with the highest similarity value is selected as the cell classification result in the cell image.
In the embodiment of the present application, the cell characteristics of the cell database under different cell classes are optimized based on the labeled cell classes. And optimizing and training the feature matching network in the deep learning model in the optimization process. Specifically, the different cell types of the cell features in the cell database are cell type matching based on cell image samples when the deep learning model is trained, and the cell type matching comprises the following steps:
labeling cell types for cell features in a cell database;
extracting cell sample characteristics from the cell image;
and measuring the similarity of the cell sample characteristics and the cell characteristics marked with the cell types to obtain a difference value between the cell sample characteristics and the cell characteristics marked with the cell types.
In the above process, when the deep learning model is trained, the cell class matching based on the cell image sample is implemented by training of a metric learning module in the deep learning model.
The measurement learning module in the deep learning model can be realized by adopting a multi-element loss function, similarity measurement is carried out on the cell sample characteristics, and then clustering is carried out according to the similarity measurement, so that the generalization of the cell database is improved. In the implementation stage of cell classification, similarity calculation can be performed one by one based on multiple cell types in the cell database, and the cell classification result is determined according to the similarity calculation, so that the accuracy of cell classification is improved.
In the above process, the cell characteristics are labeled in the cell database by cell types:
segmenting a cell image sample obtained by scanning a cell smear sample into a plurality of image sample slices;
detecting the characteristics of a single cell sample from a plurality of image sample slices by adopting a cell detection method;
and marking the cell type of the detected single cell sample characteristics as cell sample characteristics.
In this embodiment of the present application, the clustering based on cell image samples for different cell features in the cell database when training the deep learning model includes:
and uniformly sampling the cell characteristics under the corresponding cell types based on the difference value between the cell characteristics marked with the cell types, and performing clustering processing based on the selected clustering center to obtain the cell characteristics of each cell type.
Specifically, when the deep learning model is trained, the clustering based on the cell image samples is completed by a feature clustering module in the deep learning model.
In the embodiment of the application, the deep learning model calculates the similarity of the cell features and various different types of cell features in the cell database respectively to obtain the similarity between the cell features and the different types of cell features, and the similarity is realized by a feature matching network obtained by training in the deep learning model.
It can be seen that the deep learning model of the embodiment of the invention can be optimized only by the feature matching network without retraining, so that the accuracy of cell classification can be rapidly improved.
It can be seen from the embodiment of the present application that the embodiment of the present application includes a training stage and an implementation stage of a deep learning model, and the training stage optimizes the feature matching network and simultaneously adopts a cell feature matching method based on metric learning to accurately optimize the cell features of different cell types in the cell database, which is described in detail in a step-by-step manner with reference to fig. 2. Fig. 2 is a schematic diagram of a process architecture for performing cell feature matching and classification by using a deep learning model according to an embodiment of the present disclosure.
Step 1) cell image acquisition
In this step, the cell smear is scanned in an overlapping manner using an automatic scanner to obtain a cell image.
In this step, the process of cell image acquisition can be applied to the training phase and the implementation phase of the deep learning model.
Step 2) labeling cell types of cell characteristics
Firstly, segmenting a cell image obtained by scanning a cell smear into a plurality of image segments (patch);
secondly, detecting and obtaining single cell characteristics from a plurality of image slices by adopting a cell detection method;
and finally, carrying out cell type labeling on the detected single cell characteristics.
In the step, cell characteristics are set in advance under each cell type in a cell database, and the cell type marking is carried out on the single cell characteristics in a mode.
Step 3) the metric learning module carries out the metric learning processing procedure
Extracting cell sample characteristics from the cell image, carrying out similarity measurement on the cell sample characteristics by adopting a multi-element loss function based on the cell characteristics marked with the cell types, and learning the difference between the cell characteristics marked with the cell types.
The process of step 3) is shown in fig. 3, and fig. 3 is a schematic diagram of a process of metric learning performed by a metric learning module in the deep learning model provided in the embodiment of the present application, specifically, after extracting features of a cell sample from a cell image (feature extractor), performing mapping (embedding) operation and sampling (sampling) in the metric learning module, and then performing similarity measurement based on the cell features labeled with cell classes by using a multi-element loss function (loss) to learn differences between the cell features labeled with the cell classes.
4) Clustering process of feature clustering module
In this step, after the cell features are uniformly sampled in the corresponding cell categories based on the difference values between the cell features labeled with the cell categories, the cell features of each cell category are obtained by performing clustering processing based on the selected clustering center.
The step is solved when the deep learning model trains the feature matching network, so that the generalization of the cell database is improved.
5) Cell class matching for feature matching networks
The step is completed in the implementation stage of the deep learning model, and similarity calculation is respectively carried out on cell characteristics extracted from a cell image and various different types of cell characteristics in a cell database; and according to the high-low order of the similarity values, taking a set number of categories corresponding to the category cell characteristics as cell classification results in the cell images.
The implementation architecture of this step is shown in fig. 4, and fig. 4 is a schematic diagram of an implementation process of performing cell class matching by using a feature matching network of a deep learning model according to the embodiment of the present application.
It can be seen that, in the embodiment of the present application, the cell database is customized in a specialized manner, the cell feature clustering is performed for the cell types, and the cell features are matched, so that the deep learning model does not need to be retrained, and the learning of the cell feature increment can be rapidly realized only by optimizing the feature matching network in the deep learning model, thereby improving the accuracy of cell classification.
As a specific example, the embodiments of the present application are described in detail in a step-by-step manner.
Step 1) cell image acquisition
In this step, the cell smear is scanned in an overlapping manner using an automatic scanner to obtain a cell image.
In this step, the process of cell image acquisition can be applied to the training phase and the implementation phase of the deep learning model.
Step 2) labeling cell types of cell characteristics
Firstly, segmenting a cell image obtained by scanning a cell smear into a plurality of patches;
secondly, detecting and obtaining single cell characteristics from a plurality of patches by adopting a cell detection method;
and finally, carrying out cell type labeling on the detected single cell characteristics.
In the step, cell characteristics are set in advance under each cell type in a cell database, and the cell type marking is carried out on the single cell characteristics in a mode.
Step 3) the metric learning module carries out the metric learning processing procedure
In this step, cell sample features are extracted from the cell image, and the cell sample features are subjected to similarity measurement based on the cell features labeled with the cell types by using a multi-component loss function, so that differences between the cell features labeled with the cell types are learned.
Specifically, the present application aims to achieve accurate cell classification by comparing cell features to be tested with cell features in an existing cell database without retraining a deep learning model. The core method is to construct a feature extraction network with strong discrimination and a similarity measurement method, so that effective feature comparison can be realized. The measurement learning module uses a convolutional neural network as a backbone network to extract cell characteristics, amplifies the cell characteristic similarity of cells of the same category by adopting a multi-element loss function, reduces the cell characteristic similarity of cells of different categories, and distinguishes different categories by a similarity measurement mode.
In this step, the multi-tuple loss function is designed as follows:
wherein x in the above formula i Representing an input image of the cell; f represents a feature extraction network;
representing the cell characteristics after network output;
the euclidean metric distance between feature vectors representing cells of the same class,
representing Euclidean distance between feature vectors of different cell types, wherein alpha is the minimum interval between the two distances, and alpha is a hyper-parameter and can be adjusted manually.
The loss function shows that the similarity between the cell characteristics of the same type is larger, the similarity between the cells of different types is smaller, and the loss function is smaller.
Step 4) clustering process of feature clustering module
In this step, after the cell features are uniformly sampled in the corresponding cell categories based on the difference values between the cell features labeled with the cell categories, the cell features of each cell category are obtained by performing clustering processing based on the selected clustering center.
Specifically, the cell features of each cell type in the constructed cell database are clustered through the feature matching network learned by the metric learning module, and in order to ensure the representativeness of each cell type in the cell database and improve the generalization of the cell database, the specific way of selecting the cell features is as follows:
1) dividing the cell data set into basic matching libraries according to the cell types according to the cell data set marked with the types;
2) performing feature clustering on the cell features of each type in the basic matching library, wherein the clustering method adopts K-means;
3) the cell characteristics under each category can be divided into N clustering centers after clustering;
4) and taking Top-K cells near the N clustering centers as typical cell characteristics of the class, and finally, incorporating the typical cell characteristics of each class into a final cell database.
Here, the used clustering method is K-means clustering, and the specific implementation manner is as follows:
1) selecting initialized k samples as initial clustering centers: f (x) 1 )，f(x 2 )，f(x k )；
2) For each sample f (x) in the dataset i ) Calculating the distances from the cluster centers to the k cluster centers and dividing the cluster centers into classes corresponding to the cluster centers with the minimum distances;
3) for each class a i Recalculating its cluster center
4) And repeating the steps until the clustering center is not changed, and stopping clustering.
Step 5) cell class matching of feature matching networks
The step is completed in the implementation stage of the deep learning model, and similarity calculation is respectively carried out on cell characteristics extracted from a cell image and a plurality of different types of cell characteristics in a cell database; and according to the high-low order of the similarity values, taking a set number of categories corresponding to the category cell characteristics as cell classification results in the cell images.
The module can be used for fast optimization of a deep learning model, the model does not need to be retrained, only specific cell characteristics need to be added, and a special cell database is constructed, so that fast optimization can be fast realized, the accuracy of cell classification is improved, and then the module is adapted to various film making methods and scanning equipment, an optimal model system under a certain consumable material is fast customized, the model training cost is effectively saved, and the large-scale deployment and application of the deep learning model are facilitated.
Fig. 5 is a schematic structural diagram of a system for cell classification based on cell images according to an embodiment of the present application, where the system includes: an extraction module, a feature matching network module and a classification result module, wherein,
the extraction module is used for extracting cell characteristics from a provided cell image and inputting the cell characteristics into a deep learning model obtained through training;
the characteristic matching network module is used for respectively carrying out similarity calculation on the cell characteristics and various different types of cell characteristics in a cell database by the deep learning model to obtain similarity values between the cell characteristics and the different types of cell characteristics, wherein the different types of cell characteristics in the cell database are obtained by carrying out cell type matching and clustering based on cell image samples when the deep learning model is trained;
and the classification result module is used for taking the classes of the cell features of the corresponding classes with set quantity as the cell classification results in the cell images according to the high-low sequence of the similarity values.
In the above system, further comprising: the measurement learning module is also used for marking cell types for the cell characteristics in the cell database; extracting cell sample characteristics from the cell image; and measuring the similarity of the cell sample characteristics and the cell characteristics marked with the cell types to obtain a difference value between the cell sample characteristics and the cell characteristics marked with the cell types.
In the above system, the system further comprises: and the characteristic clustering module is used for uniformly sampling the cell characteristics under the corresponding cell types based on the difference value between the cell characteristics marked with the cell types and carrying out clustering processing based on the selected clustering center to obtain the cell characteristics of each cell type.
It can be seen that the tuning of the deep learning model can be effectively realized only by constructing a special cell database, and the accuracy of cell classification is improved. The embodiment of the application effectively saves the cost of deep learning model training and is beneficial to large-scale deployment and application of the deep learning model.
The embodiment of the application can be applied to medical images such as cell images and other general scenes, and has high practical value.
Those skilled in the art will appreciate that various combinations and/or combinations of features recited in the various embodiments and/or claims of the present disclosure can be made, even if such combinations or combinations are not explicitly recited in the present application. In particular, the features recited in the various embodiments and/or claims of the present application may be combined and/or coupled in various ways, all of which fall within the scope of the present disclosure, without departing from the spirit and teachings of the present application.
The principles and embodiments of the present application are explained herein using specific examples, which are provided only for the purpose of facilitating understanding of the method and the core idea of the present application and are not intended to limit the present application. It will be appreciated by those skilled in the art that changes may be made in this embodiment and its broader aspects and without departing from the principles, spirit and scope of the invention, and that all such modifications, equivalents, improvements and equivalents as may be included within the scope of the invention are intended to be protected by the claims.
Claims (10)
1. A method for cell classification based on a cell image, the method comprising:
providing a cell image, extracting cell characteristics from the cell image, and inputting the cell characteristics into a deep learning model obtained through training;
the deep learning model respectively carries out similarity calculation on the cell characteristics and various different cell characteristics in a cell database to obtain similarity values between the cell characteristics and the different cell characteristics, wherein the different cell characteristics in the cell database are obtained by carrying out cell type matching and clustering based on cell image samples when the deep learning model is trained;
and according to the high-low order of the similarity values, taking a set number of categories corresponding to the category cell characteristics as cell classification results in the cell images.
2. The method of claim 1, wherein the different classes of cell features in the cell database are cell class matching based on cell image samples when training the deep learning model comprises:
labeling cell types for cell features in a cell database;
extracting cell sample characteristics from the cell image;
and measuring the similarity of the cell sample characteristics and the cell characteristics marked with the cell types to obtain a difference value between the cell sample characteristics and the cell characteristics marked with the cell types.
3. The method of claim 2, wherein the performing cell class matching based on cell image samples is performed by a metric learning module in the deep learning model when the deep learning model is trained.
4. The method of claim 2, wherein the different classes of cell features in the cell database are clustered based on cell image samples when training the deep learning model, comprising:
and uniformly sampling the cell characteristics under the corresponding cell types based on the difference value between the cell characteristics marked with the cell types, and performing clustering processing based on the selected clustering center to obtain the cell characteristics of each cell type.
5. The method of claim 4, wherein the different classes of cell features in the cell database are clustered based on cell image samples by a feature clustering module in the deep learning model when training the deep learning model.
6. The method of claim 2, wherein labeling cell classes for cell features in a cell database comprises:
segmenting a cell image sample obtained by scanning a cell smear sample into a plurality of image sample slices;
detecting the characteristics of a single cell sample from a plurality of image sample slices by adopting a cell detection method;
and carrying out cell type labeling on the characteristics of the single cell sample obtained by detection.
7. The method of claim 1, wherein the deep learning model performs similarity calculation on the cell features and a plurality of different types of cell features in a cell database respectively to obtain similarity values between the cell features and the different types of cell features, and the similarity values are obtained by a trained feature matching network in the deep learning model.
8. A system for cell classification based on cell images, the system comprising: an extraction module, a feature matching network module and a classification result module, wherein,
the extraction module is used for extracting cell characteristics from a provided cell image and inputting the cell characteristics into a deep learning model obtained through training;
the characteristic matching network module is used for respectively carrying out similarity calculation on the cell characteristics and a plurality of different types of cell characteristics in a cell database by the deep learning model to obtain similarity values between the cell characteristics and the different types of cell characteristics, wherein the different types of cell characteristics in the cell database are obtained by carrying out cell type matching and clustering based on cell image samples when the deep learning model is trained;
and the classification result module is used for taking the classes of the cell features of the corresponding classes with set quantity as the cell classification results in the cell images according to the high-low sequence of the similarity values.
9. The system of claim 8, wherein the system further comprises:
the measurement learning module is also used for marking cell types for the cell characteristics in the cell database; extracting cell sample characteristics from the cell image; and measuring the similarity of the cell sample characteristics and the cell characteristics marked with the cell types to obtain a difference value between the cell sample characteristics and the cell characteristics marked with the cell types.
10. The system of claim 9, wherein the system further comprises: and the characteristic clustering module is used for uniformly sampling the cell characteristics under the corresponding cell types based on the difference value between the cell characteristics marked with the cell types and carrying out clustering processing based on the selected clustering center to obtain the cell characteristics of each cell type.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210347243.7A CN114863163A (en) | 2022-04-01 | 2022-04-01 | Method and system for cell classification based on cell image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210347243.7A CN114863163A (en) | 2022-04-01 | 2022-04-01 | Method and system for cell classification based on cell image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114863163A true CN114863163A (en) | 2022-08-05 |
Family
ID=82630553
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210347243.7A Pending CN114863163A (en) | 2022-04-01 | 2022-04-01 | Method and system for cell classification based on cell image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114863163A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116313115A (en) * | 2023-05-10 | 2023-06-23 | 浙江大学 | Drug action mechanism prediction method based on mitochondrial dynamic phenotype and deep learning |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109101602A (en) * | 2018-08-01 | 2018-12-28 | 腾讯科技（深圳）有限公司 | Image encrypting algorithm training method, image search method, equipment and storage medium |
| CN110276283A (en) * | 2019-06-10 | 2019-09-24 | 腾讯科技（深圳）有限公司 | Image identification method, Model of Target Recognition training method and device |
| WO2020106010A1 (en) * | 2018-11-19 | 2020-05-28 | 노을 주식회사 | Image analysis system and analysis method |
| CN111783517A (en) * | 2020-05-13 | 2020-10-16 | 北京达佳互联信息技术有限公司 | Image recognition method and device, electronic equipment and storage medium |
| CN111931751A (en) * | 2020-10-13 | 2020-11-13 | 深圳市瑞图生物技术有限公司 | Deep learning training method, target object identification method, system and storage medium |
-
2022
- 2022-04-01 CN CN202210347243.7A patent/CN114863163A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109101602A (en) * | 2018-08-01 | 2018-12-28 | 腾讯科技（深圳）有限公司 | Image encrypting algorithm training method, image search method, equipment and storage medium |
| WO2020106010A1 (en) * | 2018-11-19 | 2020-05-28 | 노을 주식회사 | Image analysis system and analysis method |
| CN110276283A (en) * | 2019-06-10 | 2019-09-24 | 腾讯科技（深圳）有限公司 | Image identification method, Model of Target Recognition training method and device |
| CN111783517A (en) * | 2020-05-13 | 2020-10-16 | 北京达佳互联信息技术有限公司 | Image recognition method and device, electronic equipment and storage medium |
| CN111931751A (en) * | 2020-10-13 | 2020-11-13 | 深圳市瑞图生物技术有限公司 | Deep learning training method, target object identification method, system and storage medium |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116313115A (en) * | 2023-05-10 | 2023-06-23 | 浙江大学 | Drug action mechanism prediction method based on mitochondrial dynamic phenotype and deep learning |
| CN116313115B (en) * | 2023-05-10 | 2023-08-15 | 浙江大学 | Drug action mechanism prediction method based on mitochondrial dynamic phenotype and deep learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhao et al. | Cloud shape classification system based on multi-channel cnn and improved fdm | |
| EP3478728B1 (en) | Method and system for cell annotation with adaptive incremental learning | |
| CN105389593B (en) | Image object recognition methods based on SURF feature | |
| CN112069921A (en) | Small sample visual target identification method based on self-supervision knowledge migration | |
| CN110633758A (en) | Method for detecting and locating cancer region aiming at small sample or sample unbalance | |
| CN109410238B (en) | Wolfberry identification and counting method based on PointNet + + network | |
| CN108596038B (en) | Method for identifying red blood cells in excrement by combining morphological segmentation and neural network | |
| CN112633382A (en) | Mutual-neighbor-based few-sample image classification method and system | |
| CN110909643B (en) | Remote sensing ship image small sample classification method based on nearest neighbor prototype representation | |
| CN108009567B (en) | Automatic excrement character distinguishing method combining image color and HOG and SVM | |
| CN109033944A (en) | A kind of all-sky aurora image classification and crucial partial structurtes localization method and system | |
| CN111027443A (en) | Bill text detection method based on multitask deep learning | |
| CN110348494A (en) | A kind of human motion recognition method based on binary channels residual error neural network | |
| CN109993187A (en) | Modeling method, robot and storage device for identifying object types | |
| CN111126361A (en) | SAR target identification method based on semi-supervised learning and feature constraint | |
| CN114782948B (en) | Global interpretation method and system for cervical fluid-based cytological smear | |
| CN115909006A (en) | Mammary tissue image classification method and system based on convolution Transformer | |
| CN114863163A (en) | Method and system for cell classification based on cell image | |
| CN115359264A (en) | Intensive distribution adhesion cell deep learning identification method | |
| Khamael et al. | Using adapted JSEG algorithm with fuzzy C mean for segmentation and counting of white blood cell and nucleus images | |
| CN113628252A (en) | Method for detecting gas cloud cluster leakage based on thermal imaging video | |
| CN103268494A (en) | Parasite egg identifying method based on sparse representation | |
| CN117611901A (en) | Small sample image classification method based on global and local contrast learning | |
| CN103093239B (en) | A kind of merged point to neighborhood information build drawing method | |
| CN115619784B (en) | Unsupervised defect detection method and terminal based on depth feature memory library |
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 |