CN113435493A - Deep migration learning-based label-free leukocyte classification system and method - Google Patents
Deep migration learning-based label-free leukocyte classification system and method Download PDFInfo
- Publication number
- CN113435493A CN113435493A CN202110691552.1A CN202110691552A CN113435493A CN 113435493 A CN113435493 A CN 113435493A CN 202110691552 A CN202110691552 A CN 202110691552A CN 113435493 A CN113435493 A CN 113435493A
- Authority
- CN
- China
- Prior art keywords
- white blood
- training
- blood cell
- leukocyte
- image
- 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
- 210000000265 leukocyte Anatomy 0.000 title claims abstract description 174
- 238000000034 method Methods 0.000 title claims abstract description 55
- 230000005012 migration Effects 0.000 title claims abstract description 34
- 238000013508 migration Methods 0.000 title claims abstract description 34
- 238000012549 training Methods 0.000 claims abstract description 79
- 238000001514 detection method Methods 0.000 claims abstract description 32
- 238000003384 imaging method Methods 0.000 claims abstract description 25
- 238000010186 staining Methods 0.000 claims abstract description 18
- 238000012360 testing method Methods 0.000 claims abstract description 16
- 238000007781 pre-processing Methods 0.000 claims abstract description 11
- 238000013135 deep learning Methods 0.000 claims abstract description 9
- 230000011218 segmentation Effects 0.000 claims abstract description 8
- 238000005259 measurement Methods 0.000 claims abstract description 7
- 238000005457 optimization Methods 0.000 claims abstract description 7
- 238000012545 processing Methods 0.000 claims abstract description 5
- 210000004027 cell Anatomy 0.000 claims description 25
- 230000006870 function Effects 0.000 claims description 16
- 230000003044 adaptive effect Effects 0.000 claims description 5
- 230000002708 enhancing effect Effects 0.000 claims description 4
- 210000003743 erythrocyte Anatomy 0.000 claims description 4
- 238000002073 fluorescence micrograph Methods 0.000 claims description 4
- 238000002372 labelling Methods 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 230000003321 amplification Effects 0.000 claims description 3
- 210000004369 blood Anatomy 0.000 claims description 3
- 239000008280 blood Substances 0.000 claims description 3
- 239000007850 fluorescent dye Substances 0.000 claims description 3
- 238000003709 image segmentation Methods 0.000 claims description 3
- 238000003199 nucleic acid amplification method Methods 0.000 claims description 3
- 238000007865 diluting Methods 0.000 claims description 2
- 238000013526 transfer learning Methods 0.000 abstract description 6
- 230000008569 process Effects 0.000 abstract description 5
- 210000003855 cell nucleus Anatomy 0.000 description 4
- 239000011521 glass Substances 0.000 description 2
- 210000003714 granulocyte Anatomy 0.000 description 2
- 230000000877 morphologic effect Effects 0.000 description 2
- 230000035790 physiological processes and functions Effects 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 206010061218 Inflammation Diseases 0.000 description 1
- 210000003651 basophil Anatomy 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 210000000601 blood cell Anatomy 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 239000003795 chemical substances by application Substances 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000004043 dyeing Methods 0.000 description 1
- 210000003979 eosinophil Anatomy 0.000 description 1
- 210000002865 immune cell Anatomy 0.000 description 1
- 230000003832 immune regulation Effects 0.000 description 1
- 230000028993 immune response Effects 0.000 description 1
- 230000004054 inflammatory process Effects 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 230000007794 irritation Effects 0.000 description 1
- 210000004698 lymphocyte Anatomy 0.000 description 1
- 239000003550 marker Substances 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000001616 monocyte Anatomy 0.000 description 1
- 210000000440 neutrophil Anatomy 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 230000000638 stimulation Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/2163—Partitioning the feature space
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Investigating Or Analysing Biological Materials (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to the field of image processing and the field of deep learning classification, and discloses a system and a method for classifying unmarked white blood cells based on deep migration learning, which comprises a training stage and a detection stage, wherein the training stage comprises white blood cell sorting, white blood cell staining, white blood cell microscopic imaging, white blood cell image preprocessing and segmentation, a white blood cell image training data set is established, and the training data set is enhanced and the network training and testing optimization of the migration learning is realized; the detection stage comprises the steps of sorting white blood cells, performing flow-type imaging on the white blood cells in the microfluidic channel, segmenting the white blood cell image, establishing a white blood cell image actual measurement data set and applying a migration learning network to perform actual sample detection. The invention solves the problem that the white blood cells need to be stained and marked before detection, and reduces the operation time of staining and marking and the influence of the staining and marking on the morphology and the state of the white blood cells. The invention adopts a transfer learning method, and can improve the speed and accuracy during training in the training process.
Description
Technical Field
The invention belongs to the field of image processing and the field of deep learning classification, and particularly relates to a system and a method for classifying unmarked white blood cells based on deep migration learning.
Background
Leukocytes are important immune cells of the human body and are mainly involved in immune response and regulation of inflammation. Leukocytes are generally classified into three types, granulocytes, monocytes, and lymphocytes, according to the difference in cell nucleus. Furthermore, granulocytes can be classified into neutrophils, eosinophils, and basophils.
There are two main methods for classifying leukocytes: one is a manual classification method. The method is carried out in a laboratory by well-trained personnel, and comprises the steps of staining white cell nuclei, observing morphological differences of the white cell nuclei through a microscope, and detecting, classifying and counting the white cells according to the morphological differences of the white cell nuclei. The other is an automated analytical device method, such as a coulter counter and a flow cytometer, which are automated analytical devices for blood cells based on electrical properties or laser scattering properties. However, the above two methods have obvious disadvantages, and the manual classification method requires manual operation of professionals and the dyeing process is complicated. Manual operation to classify leukocytes can cause many errors, including errors in sampling and errors in accuracy of results due to statistical probability. The use of staining agents can cause irritation to the white blood cells, resulting in changes in the physiological state or morphology of the white blood cells. The physiological state will change from the inactive form to the active form, and the form will be different from the original form. The disadvantages of automated analytical devices are high instrument costs, complex operation, etc.
In order to solve the disadvantages of the above leukocyte classification methods, a series of intelligent leukocyte classification methods based on deep learning have been proposed in recent years, such as CN106248559A a leukocyte five classification method based on deep learning, and CN110084150A an automatic leukocyte classification method and system based on deep learning. And they are mainly directed to the classification of stained-labeled leukocytes. Therefore, the invention provides a method and a system for classifying unmarked white blood cells based on deep migration learning. This method is directed to the classification of bright field leukocytes without markers, avoiding the stimulation of leukocytes due to marker staining and the complexity of the staining procedure. The method for transfer learning in deep learning not only solves the defects of large volume and complex operation of automatic equipment, but also can improve the accuracy and reduce the training time.
Disclosure of Invention
The invention aims to provide a system and a method for classifying unmarked white blood cells based on deep migration learning, which aim to solve the technical problems that marked staining can stimulate white blood cells and staining operation is complex.
In order to solve the technical problems, the specific technical scheme of the system and the method for classifying the unmarked white blood cells based on the deep migration learning is as follows:
a label-free white blood cell classification system based on deep migration learning comprises a training device and a detection device, wherein the training stage device comprises a white blood cell sorting module, a white blood cell staining module, a white blood cell microscopic imaging module, a white blood cell image preprocessing and segmenting module, a white blood cell image training data set establishing module, a training data set enhancing module and a migration learning network training and testing optimizing module; the detection device comprises a leukocyte sorting module, a leukocyte flow type imaging module, a leukocyte image segmentation module, a leukocyte image actual measurement data set establishment module and an actual sample detection module;
the leukocyte sorting module of the training device adopts a double-helix microfluidic chip, the first stage of the double-helix microfluidic chip adopts a rectangular section to realize cell focusing, and the second stage adopts a trapezoidal section to realize cell sorting;
the leucocyte microscopic imaging module adopts a 100X objective lens;
the leukocyte sorting module of the detection device adopts a double-helix microfluidic chip to sort and then leads into a microfluidic single channel.
The invention also discloses a label-free leukocyte classification method based on deep migration learning, which comprises a training stage and a detection stage, wherein the training stage comprises leukocyte sorting, leukocyte staining, leukocyte microscopic imaging, leukocyte image preprocessing and segmentation, a leukocyte image training data set is established, and the training data set is enhanced and is subjected to network training and test optimization of migration learning; the detection stage comprises the steps of sorting white blood cells, performing flow-type imaging on the white blood cells in the microfluidic channel, segmenting white blood cell images, establishing a white blood cell image actual measurement data set and applying a migration learning network to perform actual sample detection.
Furthermore, the leukocyte sorting is to dilute the blood sample and then separate the unmarked red blood cells and the leukocytes by using a double-helix microfluidic chip; the first stage of the double-helix microfluidic chip adopts a rectangular section to realize cell focusing, and the second stage adopts a trapezoidal section to realize cell sorting.
Further, the leukocyte staining is to perform fluorescent staining on the sorted high-purity leukocytes, and then place the stained cells on a slide glass.
Further, the leukocyte microscopic imaging adopts a 100 x objective lens to perform microscopic imaging on the leukocytes in the same field of view in turn in a bright field and fluorescence mode, and then performs real type labeling on each leukocyte according to the fluorescence image characteristics corresponding to the bright field leukocyte image.
Further, the pretreatment and segmentation of the white blood cell image are carried out, firstly, the obtained white blood cell microscopic image is pretreated, and a bright field microscopic image containing a plurality of white blood cells is subjected to self-adaptive threshold segmentation algorithm and open/close operation processing to obtain a binary image; the adaptive threshold algorithm is Otsu's method, and the parameter formula of Otsu's method to the threshold is
Wherein
Indicating the threshold parameter that is finally found,
which represents the percentage of background pixels,
which represents the ratio of the foreground pixels to each other,
the average gray value of the background is represented,
an average gray value representing the foreground; then obtaining the coordinates of each white blood cell through the preprocessed binary image, and dividing the white blood cells in a bright field data set according to the coordinate values, wherein the size of the division is 200-200 pixels, and a series of images with single white blood cell are formed.
Furthermore, the establishing of the leukocyte image training data set is to firstly screen the segmented leukocytes, remove the cells on the edge or a plurality of adhered cells, label different types of the screened leukocyte images according to the corresponding fluorescent image characteristics, namely establishing the unmarked leukocyte training data set.
Furthermore, the enhancement of the training data set is realized through multi-angle rotation, the data set is interpolated and amplified firstly, and then is rotated, the amplification factor is related to the rotation angle, and the cell image can be restored to the pixel size of 200 x 200 by cutting off the black edge after the rotation.
Further, the migration learning training network in the migration learning network training and test optimization adopts a Resnet-50 network for fine tuning training, the language used is python 2.7, and the framework used is Tensorflow;
the last layer of the pre-training model is adjusted through a fine adjustment method, the output of 1000 classes is changed into the output of 3 or 5 classes, and the three-class or five-class targets are consistent; a learning rate decay method is adopted in the training of the loss function: the basic learning rate is set to be 0.001, the learning rate is reduced to be 0.99 times of the basic learning rate every 100 steps, and the steps are set to be 2000 during training; adopting a random gradient descent algorithm and a dropout algorithm when obtaining the weight value and the deviation value in iteration; wherein root mean square value (RMSProp) algorithm is used as the optimizer for gradient descent, and the value of dropout is set to 0.5;
using relu as an activation function requires minimizing an objective function such as a formula
Wherein
The value of the (loss function) is the smallest,
label (real sample) representing the model input,
the input of the representation model is represented by,
showing the predicted value of the output of the model,
the weight values of the model are represented by,
representing the new weight value of the updated model;
using a loss function such as the formula
Where M represents the number of categories, N represents the total number of samples,
indicating an indicator variable that is 1 if the current class is the same as the sample i class, 0 otherwise,
representing the prediction probability that sample i belongs to class sample c.
Furthermore, the white blood cell sorting in the detection stage is the same as the training stage, the micro-fluidic single channel is introduced after the double-helix micro-fluidic chip sorting, the flow-type bright-field microscopic imaging is carried out, the flow-type cell image is preprocessed and segmented by adopting the same method in the training stage, a test data set is obtained, and the data set test is carried out through the deep learning network optimized in the training stage, so that the white blood cell classification detection can be realized.
The invention has the beneficial effects that:
1. compared with the existing white blood cell classification, the system and the method for classifying the unmarked white blood cells solve the problem that the white blood cells need to be dyed and marked before detection, and reduce the operation time of the dyed and marked white blood cells and the influence of the dyed and marked white blood cells on the shape and the state of the white blood cells.
2. The invention adopts a transfer learning method, and can improve the speed and accuracy during training in the training process.
Drawings
FIG. 1 is a block diagram of the process for the classification of unlabeled leukocytes according to the invention.
Detailed Description
For better understanding of the objects, structure and functions of the present invention, the system and method for classifying unlabeled leukocytes based on deep migration learning will be described in detail below with reference to the accompanying drawings.
A label-free white blood cell classification system based on deep migration learning comprises a training device and a detection device. The training stage device comprises a leukocyte sorting module, a leukocyte staining module, a leukocyte microscopic imaging module, a leukocyte image preprocessing and segmenting module, a leukocyte image training data set establishing module, a training data set enhancing module and a migration learning network training and testing optimizing module. The detection device comprises a leukocyte sorting module, a leukocyte flow type imaging module, a leukocyte image segmentation module, a leukocyte image actual measurement data set establishment module and an actual sample detection module.
A leukocyte sorting module of the training device adopts a double-helix microfluidic chip, the first stage of the double-helix microfluidic chip adopts a rectangular section to realize cell focusing, and the second stage adopts a trapezoidal section to realize cell sorting.
The leucocyte microscopic imaging module adopts a 100 x objective lens.
A leukocyte sorting module of the detection device adopts a double-helix microfluidic chip to sort and then leads into a microfluidic single channel.
A method for classifying unlabeled leukocytes based on deep migration learning comprises a training stage and a detection stage. Wherein the training phase comprises the following steps: sorting white blood cells, staining the white blood cells, carrying out white blood cell microscopic imaging, preprocessing and segmenting a white blood cell image, establishing a white blood cell image training data set, enhancing the training data set, and carrying out transfer learning network training and test optimization; the detection stage comprises the following steps: and (3) sorting white blood cells, performing flow imaging on the white blood cells in the microfluidic channel, segmenting the white blood cell image, establishing a white blood cell image actual measurement data set, and performing actual sample detection by using a migration learning network.
And (3) leukocyte sorting, namely diluting the blood sample, and then separating unmarked red blood cells and leukocytes by using a double-helix microfluidic chip to achieve the aim of high-purity leukocyte sorting. The first stage of the double-helix microfluidic chip adopts a rectangular section to realize cell focusing, and the second stage adopts a trapezoidal section to realize cell sorting and remove interfering red blood cells.
And (3) staining the white blood cells, performing fluorescent staining on the sorted high-purity white blood cells, and placing the stained cells on a glass slide.
And (3) carrying out microscopic imaging on the white blood cells under the same field of view by adopting a 100X objective lens in turn in a bright field and fluorescence mode. Then, the real type marking can be carried out on each leukocyte according to the fluorescent image characteristics corresponding to the bright field leukocyte image.
Preprocessing and segmenting the white blood cell image, preprocessing the obtained white blood cell microscopic image, and processing a bright field microscopic image containing a plurality of white blood cells by an adaptive threshold segmentation algorithm and an open-close operation to obtain a binary image. The method of adaptive thresholding used is Otsu. The parameter formula of Otsu method on the threshold is
Wherein
Indicating the threshold parameter that is finally found,
which represents the percentage of background pixels,
representing the ratio of foreground pixels.
The average gray value of the background is represented,
representing the average gray value of the foreground. And obtaining the coordinates of each leukocyte through the preprocessed binary image, and dividing the image in a bright field data set according to the coordinate values, wherein the size of the division is 200-200 pixels, and a series of images with single leukocyte are formed.
Establishing a leukocyte image training data set, screening the segmented leukocytes, removing cells on the edge or adhered to the edge, labeling different types of the screened leukocyte images according to the characteristics of the corresponding fluorescence images, and establishing a non-labeled leukocyte training data set.
The training data set is enhanced, and the image enhancement is to process the images with less training set so as to increase the number of the images. The data enhancement comprises the following steps: rotation, scaling, adding gaussian noise, etc. However, the operation of carrying out scale transformation or adding noise on the image destroys the details of the original image, and the rotation transformation is more suitable. Thus, the training data set enhancement is achieved by multi-angle rotation. In order to remove the black edge, the data set is interpolated and amplified, and then rotated, wherein the amplification factor is related to the rotation angle, so that the cell image can be restored to the pixel size of 200 x 200 by cutting off the black edge after rotation.
And (4) training and testing optimization of the transfer learning network, wherein the transfer learning training network adopts a Resnet-50 network to perform fine tuning training. The language used was python 2.7 and the framework used was Tensorflow.
The last layer of the pre-training model is adjusted by a fine-tuning method, and the output of 1000 classes is changed into the output of 3 or 5 classes, which is specifically consistent with the three-class or five-class target to be achieved. The Resnet-50 is suitable for three-channel images, and the adopted white blood cell images are single-channel gray-scale images, so that data sets are required to be overlapped by the number of channels, and the single-channel gray-scale images are changed into three-channel gray-scale images. And a method of learning rate decline is adopted when the loss function is trained. The basic learning rate was set to 0.001, the learning rate was decreased to 0.99 times the basic learning rate every 100 steps, and the steps were set to 2000 during training. And adopting a random gradient descent algorithm and a dropout algorithm when obtaining the weight value and the deviation value in iteration. Where the root mean square value (RMSProp) algorithm is used as the gradient descent optimizer and the value of dropout is set to 0.5.
Using relu as an activation function requires minimizing an objective function such as a formula
In which we wish to
The value of the (loss function) is the smallest,
label (real sample) representing the model input,
the input of the representation model is represented by,
and showing the predicted value of the model output.
The weight values of the model are represented by,
representing the new weight values of the model after updating.
Using a loss function such as the formula
Where M represents the number of categories and N represents the total number of samples.
Indicating an indicator variable that is 1 if the current class is the same as the sample i class, 0 otherwise,
representing the prediction probability that sample i belongs to class sample c.
The white blood cell sorting in the detection stage is the same as the training stage, the micro-fluidic single channel is introduced after the double-helix micro-fluidic chip sorting, the flow-type bright-field microscopic imaging is carried out, the flow-type cell image is preprocessed and segmented by adopting the same method in the training stage to obtain a test data set, and the data set test is carried out through the deep learning network optimized in the training stage, so that the white blood cell classification detection can be realized.
It is to be understood that the present invention has been described with reference to certain embodiments, and that various changes in the features and embodiments, or equivalent substitutions may be made therein by those skilled in the art without departing from the spirit and scope of the invention. In addition, many modifications may be made to adapt a particular situation or material to the teachings of the invention without departing from the essential scope thereof. Therefore, it is intended that the invention not be limited to the particular embodiment disclosed, but that the invention will include all embodiments falling within the scope of the appended claims.
Claims (10)
1. A label-free white blood cell classification system based on deep migration learning comprises a training device and a detection device, and is characterized in that the training stage device comprises a white blood cell sorting module, a white blood cell staining module, a white blood cell microscopic imaging module, a white blood cell image preprocessing and segmenting module, a white blood cell image training data set establishing module, a training data set enhancing module and a migration learning network training and testing optimizing module; the detection device comprises a leukocyte sorting module, a leukocyte flow type imaging module, a leukocyte image segmentation module, a leukocyte image actual measurement data set establishment module and an actual sample detection module;
the leukocyte sorting module of the training device adopts a double-helix microfluidic chip, the first stage of the double-helix microfluidic chip adopts a rectangular section to realize cell focusing, and the second stage adopts a trapezoidal section to realize cell sorting;
the leucocyte microscopic imaging module adopts a 100X objective lens;
the leukocyte sorting module of the detection device adopts a double-helix microfluidic chip to sort and then leads into a microfluidic single channel.
2. A label-free white blood cell classification method based on deep migration learning comprises a training stage and a detection stage, and is characterized in that the training stage comprises white blood cell sorting, white blood cell staining, white blood cell microscopic imaging, white blood cell image preprocessing and segmentation, a white blood cell image training data set is established, the training data set is enhanced, and network training and test optimization of migration learning is carried out; the detection stage comprises the steps of sorting white blood cells, performing flow-type imaging on the white blood cells in the microfluidic channel, segmenting white blood cell images, establishing a white blood cell image actual measurement data set and applying a migration learning network to perform actual sample detection.
3. The method for classifying the unlabeled leukocytes based on deep migration learning of claim 2, wherein the leukocyte sorting is performed by diluting a blood sample and then separating the unlabeled erythrocytes and leukocytes by using a double-helix microfluidic chip; the first stage of the double-helix microfluidic chip adopts a rectangular section to realize cell focusing, and the second stage adopts a trapezoidal section to realize cell sorting.
4. The method for classifying unlabeled leukocytes based on deep migration learning according to claim 3, wherein the leukocyte staining is performed by fluorescent staining of the sorted high-purity leukocytes, and then the stained leukocytes are placed on a slide.
5. The method for classifying unlabeled leukocytes based on deep migration learning of claim 4, wherein the leukocyte microscopic imaging is performed by using a 100 × objective lens, and performing microscopic imaging on leukocytes in the same field of view alternately in a bright field and a fluorescence mode, and then performing real type labeling on each leukocyte according to fluorescence image characteristics corresponding to bright field leukocyte images.
6. The method for classifying unlabeled leukocytes based on deep migration learning according to claim 5, wherein the leukocyte image preprocessing and segmentation comprises preprocessing an obtained leukocyte microscopic image, and processing a bright field microscopic image containing a plurality of leukocytes by an adaptive threshold segmentation algorithm and an open/close operation to obtain a binarized image; the adaptive threshold algorithm is Otsu's method, and the parameter formula of Otsu's method to the threshold is
Wherein
Indicating the threshold parameter that is finally found,
which represents the percentage of background pixels,
which represents the ratio of the foreground pixels to each other,
the average gray value of the background is represented,
an average gray value representing the foreground; then obtaining the coordinates of each white blood cell through the preprocessed binary image, and dividing the white blood cells in a bright field data set according to the coordinate values, wherein the size of the division is 200-200 pixels, and a series of images with single white blood cell are formed.
7. The method according to claim 6, wherein the creating of the leukocyte image training dataset comprises selecting the segmented leukocytes, removing the cells on the edges or adhering to the segmented leukocytes, labeling the selected leukocyte images with different types according to their corresponding fluorescence image characteristics, and creating the unlabeled leukocyte training dataset.
8. The method according to claim 7, wherein the enhancement of the training data set is achieved by multi-angle rotation, the data set is interpolated and amplified, and then rotated, the amplification factor is related to the rotation angle, and the removal of the black border after rotation can restore the cell image to a pixel size of 200 x 200.
9. The method for classifying unlabeled leukocytes based on deep migration learning of claim 8, wherein the migration learning training network in the migration learning network training and test optimization is fine-tuned and trained by using a Resnet-50 network, the language used is python 2.7, and the framework used is Tensorflow;
the last layer of the pre-training model is adjusted through a fine adjustment method, the output of 1000 classes is changed into the output of 3 or 5 classes, and the three-class or five-class targets are consistent; a learning rate decay method is adopted in the training of the loss function: the basic learning rate is set to be 0.001, the learning rate is reduced to be 0.99 times of the basic learning rate every 100 steps, and the steps are set to be 2000 during training; adopting a random gradient descent algorithm and a dropout algorithm when obtaining the weight value and the deviation value in iteration; wherein root mean square value (RMSProp) algorithm is used as the optimizer for gradient descent, and the value of dropout is set to 0.5;
using relu as an activation function requires minimizing an objective function such as a formula
Wherein
The value of the (loss function) is the smallest,
label (real sample) representing the model input,
the input of the representation model is represented by,
showing the predicted value of the output of the model,
the weight values of the model are represented by,
representing the new weight value of the updated model;
using a loss function such as the formula
Where M represents the number of categories, N represents the total number of samples,
indicating an indicator variable that is 1 if the current class is the same as the sample i class, 0 otherwise,
representing the prediction probability that sample i belongs to class sample c.
10. The method for classifying unlabeled leukocytes based on deep migration learning according to claim 9,
the white blood cell sorting in the detection stage is the same as the training stage, the micro-fluidic single channel is introduced after the double-helix micro-fluidic chip sorting, the flow-type bright-field microscopic imaging is carried out, the flow-type cell image is preprocessed and segmented by adopting the same method in the training stage, a test data set is obtained, and the data set test is carried out through the deep learning network optimized in the training stage, so that the white blood cell classification detection can be realized.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110691552.1A CN113435493A (en) | 2021-06-22 | 2021-06-22 | Deep migration learning-based label-free leukocyte classification system and method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110691552.1A CN113435493A (en) | 2021-06-22 | 2021-06-22 | Deep migration learning-based label-free leukocyte classification system and method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113435493A true CN113435493A (en) | 2021-09-24 |
Family
ID=77756953
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110691552.1A Pending CN113435493A (en) | 2021-06-22 | 2021-06-22 | Deep migration learning-based label-free leukocyte classification system and method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113435493A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114152557A (en) * | 2021-11-16 | 2022-03-08 | 深圳元视医学科技有限公司 | Image analysis based blood cell counting method and system |
| CN116503859A (en) * | 2023-06-27 | 2023-07-28 | 成都云芯医联科技有限公司 | Data enhancement three-classification leukemia algorithm based on deep learning |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200171488A1 (en) * | 2018-11-15 | 2020-06-04 | Massachusetts Institute Of Technology | Multi-Dimensional Double Spiral Device and Methods of Use Thereof |
-
2021
- 2021-06-22 CN CN202110691552.1A patent/CN113435493A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200171488A1 (en) * | 2018-11-15 | 2020-06-04 | Massachusetts Institute Of Technology | Multi-Dimensional Double Spiral Device and Methods of Use Thereof |
Non-Patent Citations (2)
| Title |
|---|
| XIWEI HUANG等: "Deep-Learning Based Label-Free Classification of Activated and Inactivated Neutrophils for Rapid Immune State Monitoring", 《SENSORS》 * |
| 黄汐威等: "无透镜微流控成像流动细胞检测与计数系统", 《传感器与微系统》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114152557A (en) * | 2021-11-16 | 2022-03-08 | 深圳元视医学科技有限公司 | Image analysis based blood cell counting method and system |
| CN114152557B (en) * | 2021-11-16 | 2024-04-30 | 深圳元视医学科技有限公司 | Image analysis-based blood cell counting method and system |
| CN116503859A (en) * | 2023-06-27 | 2023-07-28 | 成都云芯医联科技有限公司 | Data enhancement three-classification leukemia algorithm based on deep learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108021903B (en) | Error calibration method and device for artificially labeling leucocytes based on neural network | |
| CN106248559B (en) | A kind of five sorting technique of leucocyte based on deep learning | |
| Ahirwar et al. | Advanced image analysis based system for automatic detection and classification of malarial parasite in blood images | |
| Jambhekar | Red blood cells classification using image processing | |
| CN108961208A (en) | A kind of aggregation leucocyte segmentation number system and method | |
| CN103984939B (en) | A kind of sample visible component sorting technique and system | |
| Shkolyar et al. | Automatic detection of cell divisions (mitosis) in live-imaging microscopy images using convolutional neural networks | |
| CN110473167B (en) | Deep learning-based urinary sediment image recognition system and method | |
| CN113435493A (en) | Deep migration learning-based label-free leukocyte classification system and method | |
| Rawat et al. | Review of leukocyte classification techniques for microscopic blood images | |
| EP3455789A1 (en) | A method to combine brightfield and fluorescent channels for cell image segmentation and morphological analysis using images obtained from imaging flow cytometer (ifc) | |
| Mundhra et al. | Analyzing microscopic images of peripheral blood smear using deep learning | |
| CN112580748B (en) | Method for counting classified cells of stain image | |
| WO1997005563A1 (en) | Robustness of classification measurement | |
| CN114283407A (en) | Self-adaptive automatic leukocyte segmentation and subclass detection method and system | |
| NL2024777B1 (en) | Blood leukocyte segmentation method based on color component combination and contour fitting | |
| Davidson et al. | Automated detection and staging of malaria parasites from cytological smears using convolutional neural networks | |
| CN114332855A (en) | Unmarked leukocyte three-classification method based on bright field microscopic imaging | |
| CN112001315B (en) | Bone marrow cell classification and identification method based on migration learning and image texture characteristics | |
| Kovalev et al. | Robust recognition of white blood cell images | |
| Meimban et al. | Blood cells counting using python opencv | |
| Evangeline et al. | Computer aided system for human blood cell identification, classification and counting | |
| CN110414317B (en) | Full-automatic leukocyte classification counting method based on capsule network | |
| Diouf et al. | Convolutional neural network and decision support in medical imaging: case study of the recognition of blood cell subtypes | |
| Sapna et al. | Techniques for segmentation and classification of leukocytes in blood smear images-a review |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210924 |
|
| RJ01 | Rejection of invention patent application after publication |