CN107169556A - stem cell automatic counting method based on deep learning - Google Patents
stem cell automatic counting method based on deep learning Download PDFInfo
- Publication number
- CN107169556A CN107169556A CN201710339326.0A CN201710339326A CN107169556A CN 107169556 A CN107169556 A CN 107169556A CN 201710339326 A CN201710339326 A CN 201710339326A CN 107169556 A CN107169556 A CN 107169556A
- Authority
- CN
- China
- Prior art keywords
- image
- cell
- stem cell
- candidate
- stem
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06M—COUNTING MECHANISMS; COUNTING OF OBJECTS NOT OTHERWISE PROVIDED FOR
- G06M11/00—Counting of objects distributed at random, e.g. on a surface
- G06M11/02—Counting of objects distributed at random, e.g. on a surface using an electron beam scanning a surface line by line, e.g. of blood cells on a substrate
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10056—Microscopic image
- G06T2207/10061—Microscopic image from scanning electron microscope
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30242—Counting objects in image
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Investigating Or Analysing Biological Materials (AREA)
- Image Analysis (AREA)
Abstract
A kind of stem cell automatic counting method based on deep learning, cytometry field comprises the following steps：S11：The cell for needing to count is shot using phase contrast microscope, stem cell image is generated.S12：Image preprocessing；Noise reduction process is carried out to cell image and illumination equalization processing obtains the stem cell image of uniform illumination.S13：Remove the cell artifact produced in phase contrast microscope shooting process.S14：The stem cell image for removing pseudo- movie queen is split, multiple candidate stem cells images are obtained.S21：Manual markings are carried out to the multiple candidate cell images being partitioned into, training set is set up.S22：Training set input CNN is trained.S23：Count stem cell count result.The present invention solves the shortcoming that Traditional Man cell count consumes a large amount of manpowers, and the defect of cell growth environment can be destroyed by also overcoming streaming counting method, be counted by the cell image of shooting, with it is stable, efficient, automatic, lossless the features such as.
Description
Technical field
The present invention relates to cytometry field, more particularly to the stem cell automatic technique method based on deep learning.
Background technology
With the progress of human sciences, increasing scientific research personnel is directed to disclosing secrets of life.Stem cell is can
The class cell bred with self-replacation, on other occasions, can be divided into other kinds of cell.Therefore for dry thin
Growth, increment, the differentiation research of born of the same parents are an important research directions in cell biology.
Research work of the cell count to growth and the division of cell is accurately carried out to have great significance, it is existing to pass
System cell research mode, is mainly directly observed cell sample by researcher by microscope, not only consumes a large amount of
Time, program is complicated, also in the presence of serious subjectivity.In addition, in order to study the division of stem cell, breed and broke up
Journey, is often dyed and is added fluorescent technique to cell, and certain influence can be caused on cytoactive etc..Therefore, how
Natural activity state of the multipotential stem cell under artificial culture environment is kept, makes its splitting into for primary study without intervention
Problem.Using microexamination technology, by computer vision treatment technology, turning into the quantitative analysis to stem cell image can
Energy.
Therefore, all things considered, in the prior art the method to cell count have three kinds, below will be to this three classes cell count
Method carries out shortcoming explanation.
The first：Coulter counter method.Coulter counter method is loaded in pipe is determinedElectrolyte solution, population is mixed
It is suspended in electrolyte solution, determining to have between a pore, pore electrod on tube wall has certain voltage, is suspended inElectrolyteIn particle with
When electrolyte is by small bore tube, replace the electrolyte of same volume, make curent change and record remembering because resistance changes
Record on device, finally electric signal can be converted into particle diameter.Size distribution can be tried to achieve with this method.Coulter counter method can be by cell
It is considered as particle, therefore is widely used on the blood-counter system of automation, medically there is great meaning.But,
The method but and is not applied to for stem-cell research.First, it is impossible to make stem cell on culture medium by calculating instrument, otherwise
The stable environment of ring stem cell growth can be broken, the result of research is influenceed.Second, this method is to be obtained by statistical means in solution
TCS, and cell count needs to be accurate to position for stem-cell research, therefore this method and does not apply to.
Second：Artificial statistic law；By manually being counted to the cell image under microscope, there is higher counting accurate
True property.The workload of this method is huge, a large amount of valuable human resources of consumption, and this method is almost not carried out when data volume is larger
Possibility；Additionally due to artificial subjectivity, although counting relatively accurate, inconsistent due to observer's professional standards, having can
The state of cell can be judged by accident.
The third：Counted by flow cytometry degree cell.Flow cytometer is than being used for cell earlier
The instrument of quantity quantitative analysis.Flow cytometer is counted in the state of cell flowing, makes what these flowed first
Cell passes through the detection zone of flow cytometer, while cell passes through this region, has specificity fluorescent with laser excitation
The antibody of mark, the fluorescence antibody being now excited launches the fluorescence of certain wavelength, and cell instrument detects these exciting lights simultaneously
And these optical signals are converted into electric signal, while also some physics and biochemical character parameter of measurement cell, according to exciting light
Feature and these Physical and chemical characteristics parameters, different types of cell is made a distinction, and respectively to different types of thin
Born of the same parents' quantity is counted.But, the real space position into the sample of flow cytometer has been destroyed, so cell instrument
Measurement lacks spatial positional information, and the real space positional information of sample has very important significance for the research tool of cell,
Because in the microenvironment of tissue, a class cell is moved to another position from a position in microenvironment and is likely to
The serious state for having influence on tissue or organ.The sample before cell instrument is entered prepares to need very big artificial participation simultaneously
Amount.Therefore, drain cell calculating instrument is also not particularly suited for the counting of stem cell.
The content of the invention
It is an object of the invention to：Understood according in background technology, in the prior art in the presence of following technical problem：(1) that
Special counting method can break the stable environment of the stem cell growth on ring culture medium, influence the result of research；(2) artificial statistic law work
Amount is huge, a large amount of valuable human resources of consumption, the possibility that this method is almost not carried out when data volume is larger, and manually counts
Number methods have certain subjectivity, inconsistent due to observer's professional standards, it is possible to cell although counting relatively accurate
State judge by accident；(3) the real space position of sample can be destroyed by being counted using flow cytometry, and enter cell
Sample before instrument prepares to need very big artificial participation amount.To solve these three problems, the present invention provides a kind of for stem cell
The stem cell count method based on deep learning of digital picture.
Technical scheme is as follows：
A kind of stem cell automatic counting method based on deep learning, comprises the following steps：
S1：The cell for needing to count is shot using phase contrast microscope, stem cell image is generated；
S2：Stem cell image is split by cell segmentation technology, potential multiple candidate stem cells images are obtained；
Specifically, S2 comprises the following steps：
S21：Image preprocessing；Noise reduction process is carried out to stem cell image and illumination equalization processing obtains uniform illumination
Stem cell image；
S22：Remove the cell artifact produced in phase contrast microscope shooting process；
S23：The stem cell image for removing pseudo- movie queen is split, multiple candidate stem cells images are obtained；
S3：Cell recognition；Specifically, S3 comprises the following steps：
S31：Manual markings are carried out to the multiple candidate stem cells images being partitioned into, training set is set up；
S32：Training set input CNN is trained, to each candidate stem cells image, CNN can export a knot
Fruit represents whether it is cell；
S32 comprises the following steps：
S33：Stem cell population in all multiple candidate stem cells images of statistics, draws stem cell count result.
Preferably, the noise reduction process to stem cell image uses Gaussian filter, and illumination equalization processing is subtracted using background
Division.
Specifically, S22 detailed process is：
If g is image, f is original image, and H is imaging array, and C is background model, is imaged according to phase contrast microscope
Principle, a linear model is defined as by imaging process：
g≈Hf+C
If background has been removed, model can be reduced to：
g≈Hf
The process for going artifact is exactly to recover f process from g, and f is recovered using a constraint quadratic function：
Wherein, L is the Laplacian Matrix of definition space pixel neighbours' similarity, and Λ is a Positive diagonal matrix, ws、wr
It is to learn to obtain the regularization term of weighted by grid search；
Above-mentioned formula be not closed solution, can only numerical radius, therefore constraint reconstruction f have non-negative solution, formula is converted
To seek optimization problem：
O (f)=fTQf+2bTf+c s.t.f≥0
Q=HTH+wsL+wt∑
B=-HTg-wt∑Tf(t)+wrdiag(Λ)/2
Wherein, c is constant term wtFor the weighted factor of time consistency regularization, f(t)For the corresponding f of t, s, r, Q,
B is the habituation statement in constraint quadratic programming problem to weight factor；F is obtained using non-negative multiplication more new algorithm, is gone
Except the stem cell image of pseudo- movie queen：
Specifically, S23 detailed process is：
S231：Binary conversion treatment is carried out to the stem cell image for removing pseudo- movie queen, sharp-edged stem cell image is obtained；
S232：The area of each connected domain in stem cell image is calculated again；
S233；Give up the part that area is less than threshold value, remaining as candidate cell region, threshold value is according to figure cell image
Resolution ratio is set；
S234：Candidate cell region is cut, candidate stem cells image is obtained.
Specifically, S31 is concretely comprised the following steps：
S311：The stem cell image that a part is partitioned into sets tag along sort, is divided into cell and the class of acellular two；
S312：To be categorized as the sample of cell as positive sample, the sample to be categorized as acellular is aligned as negative sample
Sample carries out rotation expansion, and every pictures are carried out with 90 °, 180 °, 270 ° of rotations；
S313：Minute surface conversion is carried out to each original image in positive sample and postrotational image, by a positive sample figure
As being extended for 8, positive sample and its expansion image and negative sample are collectively constituted into training set.
Specifically, S32 is concretely comprised the following steps：
S321：Input data is the matrix that 24 × 24 pixels are constituted, and first feature figure layer uses 5 × 5 window pair
Input picture carries out convolution, and the size for obtaining each characteristic pattern is 20 × 20；
S322：The characteristic pattern obtained in first down-sampling layer to convolution carries out down-sampling operation, obtains same amount of
Characteristic pattern, size is reduced to 10 × 10；
S323：A convolution operation and down-sampling operation are carried out again, finally give the output result of one 1 × 2, i.e. candidate
Image is cell or acellular；
S324：Output result label corresponding with training set is contrasted, comparing result is fed back into CNN, used
Back-propagation algorithm is updated to the weights of neuron in CNN, repeats this process, and the accuracy of identification can gradually rise directly
To stablizing relatively, a CNN that can be used for candidate image to classify is finally obtained.
Specifically, S33 detailed process is：All candidate cell images input CNN of one micro-image is divided
Class, to each candidate image, CNN can export a result and represent whether it is stem cell, then count all and sentenced by CNN
It is set to the candidate image quantity of cell image, obtains inputting the stem cell sum of micro-image.
After such scheme, beneficial effects of the present invention are as follows：The present invention mainly make use of Digital image processing technique,
Auto-counting of Cells is carried out to the stem cell image that phase contrast microscope is shot, Traditional Man cell count consumption is thoroughly solved big
The shortcoming of manpower is measured, the defect of cell growth environment can be destroyed by also overcoming streaming counting method, be entered by the cell image of shooting
Row count, with it is stable, efficient, automatic, lossless the features such as.
Brief description of the drawings
Fig. 1 is, by noise reduction and after going background, removes the image before artifact in the present invention；
Fig. 2 is the image of the pseudo- movie queen of removal in the present invention；
Fig. 3 is convolutional neural networks model structure schematic diagram；
Fig. 4 is the full connection of local sensing principle in specific embodiment and local connection diagram；
Fig. 5 is that the weights in specific embodiment share schematic diagram；
Fig. 6 is convolutional neural networks structure chart.
Embodiment
The solution of the present invention is further described in detail with reference to specific embodiment.
Stem cell automatic counting method based on deep learning, comprises the following steps：
S1：The stem cell for needing to count is shot using phase contrast microscope, stem cell image is generated；
S2：Stem cell image is split by cell segmentation technology, potential multiple candidate stem cells images are obtained；
Specifically, S2 comprises the following steps：
S21：Image preprocessing；Noise reduction is carried out to stem cell image and illumination equalization obtains the stem cell figure of uniform illumination
Picture；Avoid because noise and uneven illumination are even and produce influence to the accurate segmentation of cell.
Noise reduction process is carried out to stem cell image using Gaussian filter in the present embodiment, Gaussian filter is widely used
In the noise reduction process of image procossing, by the pixel in image with being replaced after other pixel value weighted averages in itself and field
Change, it is highly effective for the noise of suppression Normal Distribution.
One-dimensional zero-mean gaussian function is：
Wherein, σ is the parameter of Gaussian function, and physical meaning is the variance of Gaussian function, needs artificially to set in actual use
Put, the width of Gaussian function is determined by σ.The two-dimentional discrete Gaussian function of zero-mean is used frequently as smoothing filter and carried out at image
Manage, two-dimensional Gaussian function is：
Then image is carried out by background subtraction method removing Lighting operations, reaches illumination effect in a balanced way.Assuming that background C is
Binomial multinomial model, i.e.,
C (u, v)=k0+k1u+k2v+k3u2+k4uv+k5v2
Wherein, u, v are the transverse and longitudinal coordinate of pixel, and k0-k5 is the coefficient of the model.If it is intended to estimating k, it is necessary to know
Road background pixel gc.To obtain gc, it is necessary to it is partitioned into background.Therefore, regard all pixels as background first, then based on minimum
Square law estimates k value：
k*=(ATA)-1ATg
Wherein, A is image array, obtains after restored map and segmentation result, can improve background by iterating and estimate
One pictures are calculated background C=Ak by evaluation*, further according to g ← g-C by background removal, obtaining illumination, stem cell is schemed in a balanced way
Picture.
S22：Remove the cell artifact produced in phase contrast microscope shooting process；First, artifact Producing reason is with differing
Microscopical image-forming principle is relevant.Cell in culture dish, due to without dyeing, substantially close to fully transparent, making
With common optical microphotograph means, effective image information can not almost be obtained by directly observing by the naked eye or shooting, therefore,
Need a kind of special optical instrument：Phase contrast microscope.Phase contrast microscope is to have installed ring-shaped light additional in ordinary optical microscope
What two special devices of grid and phase-plate were constituted, because cell has certain thickness, when light is by cell, phase can be produced
Difference, phase contrast microscope can be converted to this phase difference in the visible difference of vibration of human eye (light intensity difference), shooting process, can be thin
Aperture shape cell artifact is produced around born of the same parents, segmentation work is interfered.
Remove cell artifact concrete mode step be：
If g is image, f is original image, and H is imaging array, and C is background model, is imaged according to phase contrast microscope
Principle, a linear model is defined as by imaging process：
g≈Hf+C
If background has been removed, model can be reduced to：
g≈Hf
The process for going artifact is exactly to recover f process from g, and f is recovered using a constraint quadratic function：
Wherein, L is the Laplacian Matrix of definition space pixel neighbours' similarity, and Λ is a Positive diagonal matrix, ws、wr
It is to learn to obtain the regularization term of weighted by grid search；
Above-mentioned formula be not closed solution, can only numerical radius, therefore constraint reconstruction f have non-negative solution, formula is converted
To seek optimization problem：
O (f)=fTQf+2bTf+c s.t.f≥0
Q=HTH+wsL+wt∑
B=-HTg-wt∑Tf(t)+wrdiag(Λ)/2
F is obtained using non-negative multiplication more new algorithm, obtains removing the cell image of pseudo- movie queen, Fig. 1 is original image, Fig. 2
To remove the image of pseudo- movie queen.
S23：Segmentation is carried out to the stem cell image for removing pseudo- movie queen and obtains candidate stem cells image；
S23 detailed process is：
S231：Binary conversion treatment is carried out to the stem cell image for removing pseudo- movie queen, sharp-edged stem cell image is obtained；
S232：The area of each connected domain in stem cell image is calculated again；
S233；Give up the part that area is less than threshold value, remaining as candidate cell region, threshold value is according to stem cell image
Resolution ratio is set, and in the present embodiment, sets minimum area threshold value as 10 pixels；
S234：Candidate cell region is cut, candidate stem cells image is obtained.
Before description step S2, first convolutional neural networks are explained, convolutional neural networks CNN has turned into current speech to be known
Not, the study hotspot of the artificial intelligence field such as image recognition, video analysis, with strong applicability, feature extraction and classifying simultaneously
Carry out, global optimization training parameter is few etc., the study hotspot as current machine learning areas.The basic net of convolutional neural networks
Network structure can be divided into five parts：Input layer, feature extraction layer (convolutional layer), Feature Mapping layer (down-sampling is layer by layer), full chain
Connect layer and output layer.Each neuron local experiences domain corresponding with upper strata of feature extraction layer is connected, by wave filter and
Nonlinear transformation extracts the feature in local experiences domain.Down-sampling layer carries out dimensionality reduction to characteristic vector, while increasing the anti-of model
Distortion ability.Simple convolutional neural networks structure is by two convolutional layer (C1,C3) and two down-sampling layer (S2,S4) alternate group
Into as shown in Figure 3.In order to reduce number of parameters in convolutional neural networks, to reduce system complexity, present invention employs part
Perception principle, reason is as follows：, need to be comprising 1,000,000 hidden neurons, if adopted for the image of 1000 × 1000 pixels
Full attachment structure is used, then has 1000 × 1000 × 1000000=1012Individual connection, correspondence 1012Individual weighting parameter, according to part
Attachment structure, portion's receptive field of setting a trap is 10 × 10, and each receptive field of hidden layer need to only be connected with this 10 × 10 topography, then
Only need 108Individual parameter, it is possible to decrease 4 orders of magnitude, this can significantly simplify training process.The area of full connection drawn game portion connecting structure
It is not as shown in Figure 4.
In addition to reducing parameter using local attachment structure, it can also share to reduce of training parameter by weights
Number.Weights are shared, i.e., the neuron in same Feature Mapping uses identical weighting parameter, and the shared schematic diagram of weights is as schemed
Shown in 5.It is assumed that in locally connection network, each neuron connects 10 × 10 image-region, i.e., each neuron is right
Answer 100 Connecting quantities.It is same equivalent to allowing each neuron to use if this 100 Connecting quantities are arranged into same
Convolution kernel carries out image convolution, then only needs 100 parameters just to cover whole image visible range, is ensureing effectiveness of information
Under the premise of number of parameters is greatly reduced.
Next, the step of continuing above describes：
S3：Cell recognition；Specifically, S2 comprises the following steps：
S31：Manual markings are carried out to the stem cell image being partitioned into, training set is set up；S31's concretely comprises the following steps：
S311：The stem cell image that a part is partitioned into sets tag along sort, is divided into cell and the class of acellular two；
S312：Sample using label as cell is as positive sample, and the sample using label as acellular is aligned as negative sample
Sample carries out rotation expansion, and every pictures are carried out with 90 °, 180 °, 270 ° of rotations；
S313：Minute surface conversion is carried out to each original image in positive sample and postrotational image, by a positive sample figure
As being extended for 8, positive sample and its expansion image and negative sample are collectively constituted into training set.
S32：Training set input CNN is trained；S22's as shown in Figure 6 concretely comprises the following steps：
S321：Input data is the matrix that 24 × 24 pixels are constituted, and first feature figure layer uses 5 × 5 window pair
Input picture carries out convolution, and the size for obtaining each characteristic pattern is 20 × 20；
S322：The characteristic pattern obtained in first down-sampling layer to convolution carries out down-sampling operation, obtains same amount of
Characteristic pattern, size is reduced to 10 × 10；
S323：A convolution operation and down-sampling operation are carried out again, finally give the output result of one 1 × 2, i.e. candidate
Image is cell or acellular；
S324：Output result label corresponding with training set is contrasted, comparing result (error) is fed back to
The weights of neuron in CNN are updated by CNN using back-propagation algorithm, repeat this process, and the accuracy of identification can be by
Edge up high until stablizing relatively, finally obtain a CNN that can be used for candidate image to classify.
S33：Count cell counts.All candidate stem cells images input CNN of one micro-image is divided
Class, to each candidate stem cells image, CNN can export a result and represent whether it is cell, and it count
To cell quantity.
After the method for the present embodiment, enter for the stem cell image that Chinese Academy of Sciences's biological medicine is provided with health research institute
Go experiment, it is as a result as shown in the table:
| Picture number | Artificial counting | Candidate cell number | Redundancy rate % | CNN is counted | Accuracy % |
| 200 | 15 | 14 | 93.33 | 14 | 93.33 |
| 250 | 37 | 39 | 105.41 | 36 | 97.30 |
| 300 | 81 | 90 | 111.11 | 77 | 95.06 |
| 350 | 138 | 151 | 109.42 | 130 | 94.20 |
It is of the invention thoroughly to solve the shortcoming that Traditional Man cell count consumes a large amount of manpowers, also overcome streaming counting method
The defect of cell growth environment can be destroyed, is counted by the cell image of shooting, with stable, efficient, automatic, lossless etc.
Feature, as can be seen from the above table, accuracy rate are higher.
The present invention is not limited to above-mentioned specific embodiment, it will be appreciated that one of ordinary skill in the art is without creative
Work just can make many modifications and variations according to the design of the present invention.In a word, all technical staff in the art are according to this
The design of invention passes through the available technical side of logical analysis, reasoning, or a limited experiment on the basis of existing technology
Case, all should be in the protection domain being defined in the patent claims.
Claims (7)
1. the stem cell automatic counting method based on deep learning, it is characterised in that comprise the following steps：
S1：The cell for needing to count is shot using phase contrast microscope, stem cell image is generated；
S2：Stem cell image is split by cell segmentation technology, potential multiple candidate stem cells images are obtained；Specifically
Ground, S2 comprises the following steps：
S21：Image preprocessing；Noise reduction process is carried out to stem cell image and illumination equalization processing obtains the dry thin of uniform illumination
Born of the same parents' image；
S22：Remove the cell artifact produced in phase contrast microscope shooting process；
S23：The stem cell image for removing pseudo- movie queen is split, multiple candidate stem cells images are obtained；
S3：Cell recognition；Specifically, S3 comprises the following steps：
S31：Manual markings are carried out to the multiple candidate stem cells images being partitioned into, training set is set up；
S32：Training set input CNN is trained, to each candidate stem cells image, CNN can export a result table
Whether show it is cell；
S33：Stem cell population in all multiple candidate stem cells images of statistics, draws stem cell count result.
2. the stem cell automatic counting method according to claim 1 based on deep learning, it is characterised in that right in S21
The noise reduction process of stem cell image uses Gaussian filter, and illumination equalization processing uses background subtraction method.
3. the stem cell automatic counting method according to claim 1 based on deep learning, it is characterised in that S22 tool
Body process is：
If g is image, f is original image, and H is imaging array, and C is background model, according to phase contrast microscope image-forming principle,
Imaging process is defined as a linear model：
g≈Hf+C
If background has been removed, model can be reduced to：
g≈Hf
The process for going artifact is exactly to recover f process from g, and f is recovered using a constraint quadratic function：
<mrow>
<mi>O</mi>
<mrow>
<mo>(</mo>
<mi>f</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mo>|</mo>
<mo>|</mo>
<mi>H</mi>
<mi>f</mi>
<mo>-</mo>
<mi>g</mi>
<mo>|</mo>
<msubsup>
<mo>|</mo>
<mn>2</mn>
<mn>2</mn>
</msubsup>
<mo>+</mo>
<msub>
<mi>w</mi>
<mi>s</mi>
</msub>
<msup>
<mi>f</mi>
<mi>T</mi>
</msup>
<mi>L</mi>
<mi>f</mi>
<mo>+</mo>
<msub>
<mi>w</mi>
<mi>r</mi>
</msub>
<mo>|</mo>
<mo>|</mo>
<mi>&Lambda;</mi>
<mi>f</mi>
<mo>|</mo>
<msub>
<mo>|</mo>
<mn>1</mn>
</msub>
</mrow>
Wherein, L is the Laplacian Matrix of definition space pixel neighbours' similarity, and Λ is a Positive diagonal matrix, ws、wrBe
The lower weight factor for learning to obtain by grid search of different regularization rules (takes w in testing hereins=1, wr=0.01)；
Above-mentioned formula be not closed solution, can only numerical radius, therefore constraint reconstruction f have non-negative solution, formula is converted into and asked
Optimization problem：
O (f)=fTQf+2bTf+c s.t.f≥0
Q=HTH+wsL+wt∑
B=-HTg-wt∑Tf(t)+wrdiag(Λ)/2
Wherein, c is constant term, wt(w is taken for the weighted factor of time consistency regularization in testing hereint=0.1), f(t)For t
Moment corresponding f；F is obtained using non-negative multiplication more new algorithm, obtains removing the stem cell image of pseudo- movie queen.
4. the stem cell automatic counting method according to claim 1 based on deep learning, it is characterised in that S23 tool
Body process is：
S231：Binary conversion treatment is carried out to the stem cell image for removing pseudo- movie queen, sharp-edged stem cell image is obtained；
S232：The area of each connected domain in stem cell image is calculated again；
S233；Give up the part that area is less than threshold value, remaining as candidate cell region, threshold value is differentiated according to figure cell image
Rate is set；
S234：Candidate cell region is cut, candidate stem cells image is obtained.
5. the stem cell automatic counting method according to claim 1 based on deep learning, it is characterised in that S31 tool
Body step is：
S311：The stem cell image that a part is partitioned into sets tag along sort, is divided into cell and the class of acellular two；
S312：To be categorized as the sample of cell as positive sample, to be categorized as the sample of acellular as negative sample, to positive sample
Rotation expansion is carried out, every pictures are carried out with 90 °, 180 °, 270 ° of rotations；
S313：Minute surface conversion is carried out to each original image in positive sample and postrotational image, a positive sample image is expanded
Fill for 8, positive sample and its expansion image and negative sample are collectively constituted into training set.
6. the stem cell automatic counting method according to claim 1 based on deep learning, it is characterised in that S32 tool
Body step is：
S321：Input data is the matrix that 24 × 24 pixels are constituted, and first feature figure layer is using 5 × 5 window to input
Image carries out convolution, and the size for obtaining each characteristic pattern is 20 × 20；
S322：The characteristic pattern obtained in first down-sampling layer to convolution carries out down-sampling operation, obtains same amount of feature
Figure, size is reduced to 10 × 10；
S323：A convolution operation and down-sampling operation are carried out again, finally give the output result of one 1 × 2, i.e. candidate image
It is cell or acellular；
S324：Output result label corresponding with training set is contrasted, comparing result CNN is fed back into, using reverse
Propagation algorithm is updated to the weights of neuron in CNN, repeats this process, and the accuracy of identification can be gradually risen until phase
To stable, a CNN that can be used for candidate image to classify is finally obtained.
7. the stem cell automatic counting method according to claim 1 based on deep learning, it is characterised in that S33 tool
Body process is：All candidate cell images input CNN of one micro-image is classified, to each candidate image, CNN
A result will be exported and represent whether it is stem cell, all candidate images for being determined as cell image by CNN are then counted
Quantity, obtains inputting the stem cell sum of micro-image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710339326.0A CN107169556A (en) | 2017-05-15 | 2017-05-15 | stem cell automatic counting method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710339326.0A CN107169556A (en) | 2017-05-15 | 2017-05-15 | stem cell automatic counting method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107169556A true CN107169556A (en) | 2017-09-15 |
Family
ID=59815252
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710339326.0A Pending CN107169556A (en) | 2017-05-15 | 2017-05-15 | stem cell automatic counting method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107169556A (en) |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107832838A (en) * | 2017-11-28 | 2018-03-23 | 北京羽医甘蓝信息技术有限公司 | The method and apparatus for evaluating cell smear sample satisfaction |
| CN108052984A (en) * | 2017-10-30 | 2018-05-18 | 中国科学院计算技术研究所 | Method of counting and device |
| CN108074231A (en) * | 2017-12-18 | 2018-05-25 | 浙江工业大学 | Magnetic sheet surface defect detection method based on convolutional neural network |
| CN108615236A (en) * | 2018-05-08 | 2018-10-02 | 上海商汤智能科技有限公司 | A kind of image processing method and electronic equipment |
| CN109166100A (en) * | 2018-07-24 | 2019-01-08 | 中南大学 | Multi-task learning method for cell count based on convolutional neural networks |
| CN109697450A (en) * | 2017-10-20 | 2019-04-30 | 曦医生技股份有限公司 | Cell sorting method |
| CN109781732A (en) * | 2019-03-08 | 2019-05-21 | 江西憶源多媒体科技有限公司 | A kind of small analyte detection and the method for differential counting |
| CN110188592A (en) * | 2019-04-10 | 2019-08-30 | 西安电子科技大学 | A kind of urinary formed element cell image disaggregated model construction method and classification method |
| CN110501278A (en) * | 2019-07-10 | 2019-11-26 | 同济大学 | A kind of method for cell count based on YOLOv3 and density estimation |
| CN110647876A (en) * | 2019-11-28 | 2020-01-03 | 北京小蝇科技有限责任公司 | Blood smear full-field intelligent analysis method |
| GB2576191A (en) * | 2018-08-08 | 2020-02-12 | Sphere Fluidics Ltd | Droplet processing methods and systems |
| WO2020134532A1 (en) * | 2018-12-29 | 2020-07-02 | 北京市商汤科技开发有限公司 | Deep model training method and apparatus, electronic device, and storage medium |
| CN111919117A (en) * | 2018-03-20 | 2020-11-10 | 株式会社岛津制作所 | Cell image analysis device, cell image analysis system, learning data generation method, learning data generation program, manufacturing method, and learning model generation method |
| JP2020182412A (en) * | 2019-05-08 | 2020-11-12 | Ｓｔｅｍｃｅｌｌ株式会社 | Counting method learning device, cell counting device, counting method learning method and cell counting method, programs for counting method learning method and cell counting method, and recording medium on which those programs are recorded |
| CN112135912A (en) * | 2018-06-01 | 2020-12-25 | 株式会社前沿制药 | Image processing method, drug sensitivity test method, and image processing device |
| CN114067315A (en) * | 2021-10-23 | 2022-02-18 | 广州市艾贝泰生物科技有限公司 | Cell counting method, cell counting device, computer device, and storage medium |
| CN116645384A (en) * | 2023-07-27 | 2023-08-25 | 天津医科大学第二医院 | Stem cell area rapid segmentation method based on artificial intelligence |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101750272A (en) * | 2008-12-18 | 2010-06-23 | 鞍山钢铁集团公司 | Blood cell image recognition counting method |
| WO2014040184A1 (en) * | 2012-09-13 | 2014-03-20 | The Governing Council Of The University Of Toronto | System and method for fetal and maternal red blood cell counting |
| CN106600577A (en) * | 2016-11-10 | 2017-04-26 | 华南理工大学 | Cell counting method based on depth deconvolution neural network |
-
2017
- 2017-05-15 CN CN201710339326.0A patent/CN107169556A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101750272A (en) * | 2008-12-18 | 2010-06-23 | 鞍山钢铁集团公司 | Blood cell image recognition counting method |
| WO2014040184A1 (en) * | 2012-09-13 | 2014-03-20 | The Governing Council Of The University Of Toronto | System and method for fetal and maternal red blood cell counting |
| CN106600577A (en) * | 2016-11-10 | 2017-04-26 | 华南理工大学 | Cell counting method based on depth deconvolution neural network |
Non-Patent Citations (5)
| Title |
|---|
| ZHAOZHENG YIN 等: "Understanding the phase contrast optics to restore artifact-free microscopy images for segmentation", 《MEDICAL IMAGE ANALYSIS》 * |
| 杨吉成 等: "《医用细胞工程》", 30 June 2001, 上海：上海交通大学出版社 * |
| 陈勇 主编: "《变电站交流回路智能检验系统设计与实现》", 31 December 2015, 长春：吉林人民出版社 * |
| 韦康博: "《人工智能 比你想象的更具颠覆性的智能革命》", 30 September 2016, 北京：现代出版社 * |
| 黄福珍 等: "《人脸检测》", 30 April 2006, 上海：上海交通大学出版社 * |
Cited By (25)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109697450A (en) * | 2017-10-20 | 2019-04-30 | 曦医生技股份有限公司 | Cell sorting method |
| CN109697450B (en) * | 2017-10-20 | 2023-04-07 | 曦医生技股份有限公司 | Cell sorting method |
| CN108052984B (en) * | 2017-10-30 | 2019-11-08 | 中国科学院计算技术研究所 | Method of counting and device |
| CN108052984A (en) * | 2017-10-30 | 2018-05-18 | 中国科学院计算技术研究所 | Method of counting and device |
| CN107832838A (en) * | 2017-11-28 | 2018-03-23 | 北京羽医甘蓝信息技术有限公司 | The method and apparatus for evaluating cell smear sample satisfaction |
| CN108074231A (en) * | 2017-12-18 | 2018-05-25 | 浙江工业大学 | Magnetic sheet surface defect detection method based on convolutional neural network |
| CN108074231B (en) * | 2017-12-18 | 2020-04-21 | 浙江工业大学 | Magnetic sheet surface defect detection method based on convolutional neural network |
| CN111919117A (en) * | 2018-03-20 | 2020-11-10 | 株式会社岛津制作所 | Cell image analysis device, cell image analysis system, learning data generation method, learning data generation program, manufacturing method, and learning model generation method |
| CN108615236A (en) * | 2018-05-08 | 2018-10-02 | 上海商汤智能科技有限公司 | A kind of image processing method and electronic equipment |
| CN112135912A (en) * | 2018-06-01 | 2020-12-25 | 株式会社前沿制药 | Image processing method, drug sensitivity test method, and image processing device |
| CN109166100A (en) * | 2018-07-24 | 2019-01-08 | 中南大学 | Multi-task learning method for cell count based on convolutional neural networks |
| GB2576191B (en) * | 2018-08-08 | 2022-11-16 | Sphere Fluidics Ltd | Droplet processing methods and systems |
| GB2576191A (en) * | 2018-08-08 | 2020-02-12 | Sphere Fluidics Ltd | Droplet processing methods and systems |
| WO2020134532A1 (en) * | 2018-12-29 | 2020-07-02 | 北京市商汤科技开发有限公司 | Deep model training method and apparatus, electronic device, and storage medium |
| CN109781732A (en) * | 2019-03-08 | 2019-05-21 | 江西憶源多媒体科技有限公司 | A kind of small analyte detection and the method for differential counting |
| CN110188592B (en) * | 2019-04-10 | 2021-06-29 | 西安电子科技大学 | Urine formed component cell image classification model construction method and classification method |
| CN110188592A (en) * | 2019-04-10 | 2019-08-30 | 西安电子科技大学 | A kind of urinary formed element cell image disaggregated model construction method and classification method |
| JP2020182412A (en) * | 2019-05-08 | 2020-11-12 | Ｓｔｅｍｃｅｌｌ株式会社 | Counting method learning device, cell counting device, counting method learning method and cell counting method, programs for counting method learning method and cell counting method, and recording medium on which those programs are recorded |
| CN110501278B (en) * | 2019-07-10 | 2021-04-30 | 同济大学 | Cell counting method based on YOLOv3 and density estimation |
| CN110501278A (en) * | 2019-07-10 | 2019-11-26 | 同济大学 | A kind of method for cell count based on YOLOv3 and density estimation |
| CN110647876B (en) * | 2019-11-28 | 2020-06-26 | 北京小蝇科技有限责任公司 | Blood smear full-field intelligent analysis method |
| CN110647876A (en) * | 2019-11-28 | 2020-01-03 | 北京小蝇科技有限责任公司 | Blood smear full-field intelligent analysis method |
| CN114067315A (en) * | 2021-10-23 | 2022-02-18 | 广州市艾贝泰生物科技有限公司 | Cell counting method, cell counting device, computer device, and storage medium |
| CN116645384A (en) * | 2023-07-27 | 2023-08-25 | 天津医科大学第二医院 | Stem cell area rapid segmentation method based on artificial intelligence |
| CN116645384B (en) * | 2023-07-27 | 2023-10-03 | 天津医科大学第二医院 | Stem cell area rapid segmentation method based on artificial intelligence |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107169556A (en) | stem cell automatic counting method based on deep learning | |
| CN109800736B (en) | Road extraction method based on remote sensing image and deep learning | |
| Huang et al. | Urban land-use mapping using a deep convolutional neural network with high spatial resolution multispectral remote sensing imagery | |
| Misra et al. | SpikeSegNet-a deep learning approach utilizing encoder-decoder network with hourglass for spike segmentation and counting in wheat plant from visual imaging | |
| CA2948499C (en) | System and method for classifying and segmenting microscopy images with deep multiple instance learning | |
| Rahaman et al. | An efficient multilevel thresholding based satellite image segmentation approach using a new adaptive cuckoo search algorithm | |
| CN109977780A (en) | A kind of detection and recognition methods of the diatom based on deep learning algorithm | |
| CN106485251A (en) | Egg embryo classification based on deep learning | |
| Ghani et al. | Dendritic spine classification using shape and appearance features based on two-photon microscopy | |
| Pan et al. | Mitosis detection techniques in H&E stained breast cancer pathological images: A comprehensive review | |
| Jambhekar | Red blood cells classification using image processing | |
| Chen et al. | Agricultural remote sensing image cultivated land extraction technology based on deep learning | |
| CN112581450B (en) | Pollen detection method based on expansion convolution pyramid and multi-scale pyramid | |
| CN110853070A (en) | Underwater sea cucumber image segmentation method based on significance and Grabcut | |
| CN111898677A (en) | Plankton automatic detection method based on deep learning | |
| CN114821296A (en) | Underground disease ground penetrating radar image identification method and system, storage medium and terminal | |
| US20230267703A1 (en) | Hierarchical constraint (hc)-based method and system for classifying fine-grained graptolite images | |
| Li et al. | FecalNet: Automated detection of visible components in human feces using deep learning | |
| Li et al. | Identification and localization of grape diseased leaf images captured by UAV based on CNN | |
| CN116452820B (en) | Method and device for determining environmental pollution level | |
| Sun et al. | Lake algal bloom monitoring via remote sensing with biomimetic and computational intelligence | |
| Beck et al. | Application of neural networks to quantify the utility of indices of biotic integrity for biological monitoring | |
| CN116403211A (en) | Segmentation and clustering method and system based on single-cell pathology image cell nuclei | |
| Marouf et al. | Fine-tuning MobileNetV2 for Sea Animal Classification | |
| Dawoud et al. | Edge-based self-supervision for semi-supervised few-shot microscopy image cell segmentation |
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: 20170915 |
|
| RJ01 | Rejection of invention patent application after publication |