CN103020488B - A kind of Subcellular Localization method based on fluorescence microscope images - Google Patents
A kind of Subcellular Localization method based on fluorescence microscope images Download PDFInfo
- Publication number
- CN103020488B CN103020488B CN201210588110.5A CN201210588110A CN103020488B CN 103020488 B CN103020488 B CN 103020488B CN 201210588110 A CN201210588110 A CN 201210588110A CN 103020488 B CN103020488 B CN 103020488B
- Authority
- CN
- China
- Prior art keywords
- pixel
- image
- subcellular localization
- value
- subcellular
- 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.)
- Expired - Fee Related
Links
Abstract
The invention discloses a kind of Subcellular Localization method based on fluorescence microscope images.Each secondary Subcellular Localization image is calculated its pixel value all pixel averages u more than 30, to the pixel value imparting null value less than u 30, obtains pretreated subcellular fraction image.Calculate each gray value pixel more than zero and its 8 abutment points differences in image, being labeled as 1 more than the abutment points of this center pixel point value, be otherwise labeled as 0, each central pixel point 8 field be 1 number be respectively 0 to 8.It is 9 classes that number according to the abutment points that this pixel surrounding markings is 1 incorporates this pixel into.Count the pixel number that each class of image comprises.Matrix of differences model according to image slices vegetarian refreshments, adjoining on the occasion of sum, negative value sum, absolute value sum of available every class pixel, altogether obtain 36 characteristic quantities.Finally, according to the 36 groups of subcellular fraction image feature amount extracted, utilize algorithm of support vector machine, image is trained prediction.
Description
Technical field
The invention belongs to technical field of biological information, particularly to a kind of Subcellular Localization side based on fluorescence microscope images
Method.
Background technology
Epr gene has led the quick growth of gene and protein sequencing, and now, the concern power of people has turned to coding
The function of protein.Subcellular Localization refers to that certain albumen or expression product are at intracellular concrete Present site.As only at core
Existing in interior existence, or kytoplasm, or exist on cell membrane, this has very important significance in bioinformatics.The most certainly
The development of dynamic fluorescence microscopy so that Protein Subcellular can position in the case of high-throughout in imaging, it is therefore desirable to fast
The automatic computing technique of speed and subcellular fraction image is effectively quantified by effective algorithm, identification and classification.
At present, people are applied to field of biology computer technology, and the part mankind's is mental to replace to utilize computer
The bio information to magnanimity of working carries out identification and classification, therefrom draws valuable information [2].Image statistics has turned out
Distinguish location aspect and have very successful application, but its conventional method calculates relatively slow, and it is frequently necessary to from experimental image
Individually select cell, thus limit disposal ability and the scope of potential application.
Summary of the invention
It is an object of the invention to provide a kind of Subcellular Localization method based on fluorescence microscope images.
The technical scheme is that a kind of Subcellular Localization method based on fluorescence microscope images comprises the following steps:
Step 1, first carries out pretreatment to Subcellular Localization image based on fluorescence microscope images, to each secondary subcellular fraction
Location image, calculates average u of its pixel value all pixels more than 30, then value is less than the pixel value of the pixel of u-30
It is assigned to zero, obtains pretreated Subcellular Localization image,
Step 2, calculates subcellular fraction framing pixel and is adjacent difference a little, obtain pretreatment the most in step 1
After Subcellular Localization image in, to each pixel value pixel more than zero, calculate the picture of 8 pixels being adjacent
Element difference, being labeled as 1 more than the abutment points of this pixel point value, is otherwise labeled as 0, then with this pixel surrounding markings for 1
The number of abutment points this pixel is incorporated into is 9 classes;
Step 3, extracts Subcellular Localization characteristics of image, i.e. by step 2, first counts this Subcellular Localization image every
The pixel number that one class comprises, secondly, according to the matrix of differences model of Subcellular Localization image slices vegetarian refreshments, adds up each class
The value summation more than zero of Subcellular Localization image slices vegetarian refreshments and minus value summation, finally further according to each picture of each apoplexy due to endogenous wind
The absolute value sum of vegetarian refreshments difference, can obtain the variance of each class, the most altogether obtains 36 stack features amounts;
Step 4, according to the 36 groups of Subcellular Localization image feature amount extracted, utilizes algorithm of support vector machine, to subcellular fraction
Location image is trained prediction, it is achieved Subcellular Localization based on fluorescence microscope images.
Subcellular Localization image carried out pre-treatment step include described in step 1,
Firstly, for Subcellular Localization image, filter out the subcellular fraction image intermediate value pixel value more than 30, and it is equal to calculate it
Value u, as shown in Equation 1:
Sum Value=∑ (I > 30) .(1)
u=mean(Sum Value).
Then this image intermediate value is assigned to zero less than the pixel value of u-30, as shown in Equation 2,
I=0, if (I <u-30) (2)
Subcellular fraction framing pixel described in step 2 is adjacent the step of difference a little and includes,
According to the Subcellular Localization image obtained after pretreatment, its pixel value range is 0, u-30~255, at image I > 0
Pixel, calculate the difference of itself and abutment points, as shown in Equation 3, difference is labeled as 1 more than zero, other be labeled as 0,
9 kinds of model matrixs in following table, in each pixel correspondence table, a kind of situation, obtains (0)-(8) totally 9 kinds of characteristic quantities,
I_b=I (i-1:i+1, j-1:j+1)-I (i, j) (3)
Extraction Subcellular Localization image characteristic step described in step 3 includes,
After abutment points mathematic interpolation, count the pixel number that this each class of subcellular fraction image comprises,
Secondly, according to the matrix of differences of subcellular fraction image slices vegetarian refreshments, this point is asked respectively more than the minus difference of zero-sum
With, then seek the absolute value sum of this difference, as shown in Equation 4,
Add up the value summation more than zero of the subcellular fraction image slices vegetarian refreshments of each class and minus value summation the most respectively,
Finally further according to the absolute value sum of each pixel difference of each apoplexy due to endogenous wind, the variance of each class can be obtained, the most always there are
To 36 stack features amounts,
Sum1=∑ (I_b > 0)
Sum2=∑ (I_b < 0) (4)
sum3=sum1-sum2
After so by adding up, sum1 and sum2 of subcellular fraction image slices vegetarian refreshments in the case of every kind, and the side of sum3
Difference, in the case of adding every kind, total number of subcellular fraction image slices vegetarian refreshments, can extract altogether the 9x4=36 of subcellular fraction image
Stack features amount.
The step for the training prediction of Subcellular Localization image described in step 4 includes,
Use the endogenous protein of the specific cells device including 10 class organelles of fluorescent antibody or other spy agent detections
Or key element, each image is by additional the redying with DNA specific stain 4', 6-diamidino-2-phenylindone (DAPI) with
Cell image, the nucleus position of each cell in this stain saliency maps picture,
Secondary endogenous location, 502 collected image, is fixing HeLa cell image, takes from 60 multiplying power under oil immersion, should
Image is 8 gray level images of 768x512 pixel, and every width contains up to 13 cells.
The present invention, by each secondary fluorescence microscope images is extracted feature, utilizes support vector machine (SVM) algorithm thin to Asia
Born of the same parents' image carries out classifying and predicting, the results show, and the present invention extracts Subcellular Localization characteristics of image algorithm arithmetic speed relatively
Hurry up, recognition correct rate is high, and the highest to the accuracy rate of subcellular fraction image classification.
Accompanying drawing explanation
The subcellular fraction image distinguished by thresholding in Fig. 1 embodiment of the present invention.
The sample image of the protein of 10 class endogenous expressions in Fig. 2 present invention.
Fig. 3 feature of present invention extraction algorithm is to endogenous (Endogenously) Subcellular Localization image set classification accuracy rate
Table.
Original Subcellular Localization image in Fig. 4 embodiment of the present invention.
Pretreated Subcellular Localization image in Fig. 5 embodiment of the present invention.
In Fig. 6 present invention, pixel is adjacent a difference statistical model.
Detailed description of the invention
The present invention proposes the algorithm of a kind of subcellular fraction image characteristics extraction, each secondary subcellular fraction image is carried out feature and carries
Take, according to the feature extracted subcellular fraction image classified and position.This algorithm need not image is carried out cutting, not only
Can quickly process all subcellular fraction images, and there is the highest accuracy.The present invention realizes passing through techniques below
Scheme:
1, first subcellular fraction image is carried out pretreatment.To each secondary subcellular fraction image, calculate its pixel value and be more than 30
Average u of all pixels, the pixel value of the pixel then value being less than u-30 is assigned to zero, obtains pretreated subcellular fraction figure
Picture.
2, calculate subcellular fraction image slices vegetarian refreshments and be adjacent difference a little.Obtain in pretreated image in step 1,
To each pixel value pixel more than zero, calculating the pixel value difference of 8 pixels being adjacent, we are more than this picture
The abutment points of vegetarian refreshments value is labeled as 1, is otherwise labeled as 0, then with the number of abutment points that this pixel surrounding markings is 1 this
It is 9 classes that pixel incorporates into.
3, feature is extracted.By step 2, we can count the pixel that this each class of subcellular fraction image comprises
Number.Secondly, this point, according to the matrix of differences of subcellular fraction image slices vegetarian refreshments, is asked respectively by we more than the minus difference of zero-sum
With, then seek the absolute value sum of this difference, then we add up the value more than zero of the subcellular fraction image slices vegetarian refreshments of each class
Summation and minus value summation, finally further according to the absolute value sum of each pixel difference of each apoplexy due to endogenous wind, can obtain every
The variance of one class, so we obtain 36 stack features amounts altogether.
4, conclusion.The 36 groups of subcellular fraction image feature amount extracted according to the present invention, utilize algorithm of support vector machine, to image
Being trained prediction, the results show, it is very fast that the present invention extracts subcellular fraction characteristics of image algorithm arithmetic speed, accuracy simultaneously
Higher.
The preprocess method that the present invention uses:
First, it is illustrated in figure 4 original Subcellular Localization image, filters out the subcellular fraction image intermediate value pixel more than 30
Value, and calculate its average u, shown in equation below 1:
Sum Value=∑ (I > 30) .(1)
u=mean(Sum Value).
Then this sub-picture intermediate value is assigned to zero less than the pixel value of u-30, shown in equation below 2, the image obtained such as figure
Shown in 5:
I=0, if (I <u-30) (2)
The algorithm calculating adjacent point value of present invention design:
According to the image I obtained after pretreatment, its pixel value range is 0, u-30~255.At the pixel of image I > 0,
Calculate the difference of itself and abutment points, shown in equation below 3.Difference more than zero be labeled as 1, other be labeled as 0.We are permissible
Obtain following 9 kinds of model matrixs, be illustrated in fig. 6 shown below.A kind of situation in each pixel corresponding diagram 6, obtains (0)-(8) totally 9
Plant characteristic quantity.
I_b=I (i-1:i+1, j-1:j+1)-I (i, j) (3)
The present invention designs and extracts characteristics of image algorithm:
After abutment points mathematic interpolation, we can count the pixel number that this each class of subcellular fraction image comprises.
Secondly, this point, according to the matrix of differences of subcellular fraction image slices vegetarian refreshments, is sued for peace respectively by we more than the minus difference of zero-sum, then
Seek the absolute value sum of this difference, shown in equation below 4.Then we add up the subcellular fraction image slices vegetarian refreshments of each class respectively
The value summation more than zero and minus value summation, finally further according to each pixel difference of each apoplexy due to endogenous wind absolute value it
With, the variance of each class can be obtained, so we obtain 36 stack features amounts altogether.
sum1=∑(I_b>0)
Sum2=∑ (I_b < 0) (4)
sum3=sum1-sum2
So by after cumulative, we can draw in Fig. 4 in the case of every kind the sum1 of subcellular fraction image slices vegetarian refreshments and
Sum2, and the variance of sum3, total number of subcellular fraction image slices vegetarian refreshments in the case of adding every kind, so we are the most permissible
Extract the 9x4=36 stack features amount of subcellular fraction image.
The algorithm of image of the present invention training prediction:
Image set used by the present invention is with fluorescent antibody or the endogenous protein of the specific cells device of other spy agent detections
Or key element (10 class organelle).Each image is by use DNA specific stain 4', the 6-diamidino-2-phenyl Yin additional with
The cell image that diindyl (DAPI) is redyed, the nucleus position of each cell in this stain energy saliency maps picture, finally we collect
To 502 secondary endogenous location images.All of is all fixing HeLa cell image, takes from 60 multiplying power under oil immersion.These are all
Being 8 gray level images of 768x512 pixel, every width contains up to 13 cells.This data set can be downloaded from LOCATE site
To (http://locate.imb.uq.edu.au/)。
The time test of this experiment is to complete on the computer that CPU is Intel (R) Xeon (R) 2.00GHz, Matlab's
Version is MatlabR2010b.
After to subcellular fraction Image semantic classification, utilize the extraction characteristics algorithm that the present invention proposes to 503 secondary subcellular fraction figures
As just feature extraction, 6 minutes 12 seconds this step used time of feature extraction, then 36 stack features extracted are normalized place
After reason, so that it may it is trained and predicts classification.That use here is algorithm of support vector machine (SVM), secondary sub-from 503
In cell image, extracting 400 sub-pictures the most immediately as SVM classifier training sample, other 103 sub-pictures are as grader
Forecast sample, altogether circulation 1000 times, the Average Accuracy finally obtaining classification is 96.8%.Wherein the realization of SVM uses
It is libsvm workbox (libsvm-mat). this software kit can be free at http://www.csie.ntu.edu.tw/ ~ cjlin
Obtain.
The present invention, we have proposed threshold value and adjoins statistical method, its essence is and image is carried out thresholding process and passes through
One given threshold value is calculated over the pixel quantity of threshold value.As it is shown in figure 1, endoplasmic reticulum (a) and microtubule skeleton (b) image lead to
Crossing thresholding (a ' and b ') makes its intensity pixel more than u-30 be shown as white, and u is every width cell image image pixel intensities
Meansigma methods.Although image a is similar with vision with the quality of image b, but the most more can make a distinction from image a ' and b '.Figure
As a ' contains more entity white portion, image b ' then shows more internal waviness and emergence edge.
In order to test the effect of this algorithm, we create a subcellular fraction fluorescence microscope images collection, each subcellular fraction
Location image set comprises about 50 sub-pictures.This image set is with fluorescent antibody or the specific cells device of other spy agent detections
Endogenous protein or key element (10 class organelles, 503 sub-pictures), sample image (scale is 10 microns), its point as shown in Figure 2
Not Dai Biao (a) micro-pipe (Microtuble), (b) Golgi body (Golgi), (c) plasma membrane (Plasma membrane), (d) flesh move
Albuminous cell skeleton (Actin cytoskeleton), (e) core (Nucleus), body (Endosome) in (f), (g) endoplasmic reticulum
(ER), (h) mitochondrion (Mitochondria), (i) peroxisome (Peroxisome), (j) lysosome (Lysosome).
The method of the present invention shows, in the case of without image cropping, under Matlab (R2010b) environment, utilizes support vector machine
Algorithm, in the case of cycle calculations 1000 times, distinguishes endogenous (Endogenously) Subcellular Localization image set
With average accuracy during classification up to 96.8%, experimental result is better than the method proposed in [3], detailed classification information such as Fig. 3
Shown in, chart sub-cellular image set is 10 classes, it can be seen that the classification accuracy rate of each class subcellular fraction image and wrong point of rate thereof,
By figure it is also seen that this algorithm is to Endosome, ER, Mitochondria, Nucleus, PM in this 10 class subcellular fraction image,
Actin-Cytoskeleton, Microtubule and Peroxisome8 class image set classification accuracy rate is higher, the most wrong point of rate
Less.
Fig. 4 is the secondary original Subcellular Localization image of in Endogenously image set, and its pixel value range is 0-
255.Through first step pretreatment, first calculate all pixel values pixel average u more than 30, then in whole image
Pixel value is assigned to 0 less than u-30's, and not changing of other obtains the most pretreated image of Fig. 5.Then pixel in Fig. 5,
If pixel value be more than zero, then calculate its adjacent pixels matrix of differences, difference more than zero be labeled as 1, otherwise be then labeled as 0, as
Shown in Fig. 6, it is altogether 9 classes that all of pixel can be classified as (0)-(8), then can count the pixel of each class of this sub-picture
Number.According to the adjacent matrix of differences of each pixel, count respectively its value sum sum1 more than zero and minus value it
And sum2, and absolute value sum sum3, shown as shown in Equation 4.The variance of each stack features is can get again by sum3.And then I
Obtain 36 stack features amounts.The characteristic quantity extracted finally according to our algorithm, utilizes algorithm of support vector machine, fixed to subcellular fraction
Bit image is trained prediction, and this process circulates 1000 times.Test result indicate that, inventive feature extraction algorithm arithmetic speed
Comparatively fast, the Average Accuracy of the Subcellular Localization image classification created in the present invention is reached 96.8%.
Claims (2)
1. a Subcellular Localization method based on fluorescence microscope images, it is characterised in that comprise the following steps:
Step 1, first carries out pretreatment to Subcellular Localization image based on fluorescence microscope images, to each width Subcellular Localization
Image, calculates average u of its pixel value all pixels more than 30, and the pixel value of the pixel then value being less than u-30 is assigned to
Zero, obtain pretreated Subcellular Localization image;
Step 2, calculates Subcellular Localization image slices vegetarian refreshments and is adjacent difference a little, obtain pretreated the most in step 1
In Subcellular Localization image, to each pixel value pixel more than zero, the pixel calculating 8 pixels being adjacent is poor
Value, being labeled as 1 more than the abutment points of this pixel point value, is otherwise labeled as 0, is then the neighbour of 1 with this pixel surrounding markings
It is 9 classes that the number of contact incorporates this pixel into;
Step 3, extracts Subcellular Localization characteristics of image, i.e. by step 2, first counts this each class of Subcellular Localization image
The pixel number comprised, secondly, according to the matrix of differences model of Subcellular Localization image slices vegetarian refreshments, adds up the sub-thin of each class
Born of the same parents position the value summation more than zero of image slices vegetarian refreshments and minus value summation, finally further according to each pixel of each apoplexy due to endogenous wind
The absolute value sum of difference, obtains the variance of each class, the most altogether obtains 36 characteristic quantities;
Step 4, according to 36 the Subcellular Localization image feature amount extracted, utilizes algorithm of support vector machine, to Subcellular Localization
Image is trained prediction, it is achieved Subcellular Localization based on fluorescence microscope images;
Wherein, Subcellular Localization image carried out pre-treatment step include described in step 1:
Firstly, for Subcellular Localization image, filter out the Subcellular Localization image intermediate value pixel value more than 30, and it is equal to calculate it
Value u, as shown in formula (1):
Then this image intermediate value is assigned to zero less than the pixel value of u-30, as shown in formula (2),
I=0, if (I < u-30) (2).
2. Subcellular Localization method based on fluorescence microscope images as claimed in claim 1, it is characterised in that institute in step 2
The calculating Subcellular Localization image slices vegetarian refreshments stated is adjacent the step of difference a little and includes,
According to the Subcellular Localization image obtained after pretreatment, its pixel value range is 0, u-30~255, at the picture of image I > 0
Vegetarian refreshments, calculates the difference of itself and abutment points, as shown in Equation 3,
Difference is labeled as 1 more than zero, other be labeled as 0, obtain 9 kinds of model matrixs in following table, each pixel is corresponding
In table, a kind of situation, obtains (0)-(8) totally 9 kinds of characteristic quantities,
I_b=I (i-1:i+1, j-1:j+1)-I (i, j) (3)
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201210588110.5A CN103020488B (en) | 2012-12-29 | 2012-12-29 | A kind of Subcellular Localization method based on fluorescence microscope images |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201210588110.5A CN103020488B (en) | 2012-12-29 | 2012-12-29 | A kind of Subcellular Localization method based on fluorescence microscope images |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103020488A CN103020488A (en) | 2013-04-03 |
CN103020488B true CN103020488B (en) | 2016-10-05 |
Family
ID=47969086
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201210588110.5A Expired - Fee Related CN103020488B (en) | 2012-12-29 | 2012-12-29 | A kind of Subcellular Localization method based on fluorescence microscope images |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103020488B (en) |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104881857B (en) * | 2014-02-28 | 2017-06-06 | 中国长江三峡集团公司中华鲟研究所 | A kind of image analysis calculation method of fish-egg mother cell polar core deviant |
CN107607511B (en) * | 2017-10-21 | 2020-09-22 | 云南中烟工业有限责任公司 | Method for detecting influence of cigarette smoke on aquaporin 5 subcellular localization |
CN107513552B (en) * | 2017-10-21 | 2021-02-02 | 云南中烟工业有限责任公司 | Method for detecting influence of cigarette smoke on aquaporin 5 cell expression quantity |
CN108550148A (en) * | 2018-04-13 | 2018-09-18 | 重庆大学 | Nucleus in histotomy micro-image divides automatically and classifying identification method |
CN112184696B (en) * | 2020-10-14 | 2023-12-29 | 中国科学院近代物理研究所 | Cell nucleus and organelle counting and area calculating method and system thereof |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101176116A (en) * | 2005-05-13 | 2008-05-07 | 三路影像公司 | Methods of chromogen separation-based image analysis |
US7430319B2 (en) * | 2001-12-19 | 2008-09-30 | Fuji Xerox Co., Ltd. | Image collating apparatus for comparing/collating images before/after predetermined processing, image forming apparatus, image collating method, and image collating program product |
-
2012
- 2012-12-29 CN CN201210588110.5A patent/CN103020488B/en not_active Expired - Fee Related
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7430319B2 (en) * | 2001-12-19 | 2008-09-30 | Fuji Xerox Co., Ltd. | Image collating apparatus for comparing/collating images before/after predetermined processing, image forming apparatus, image collating method, and image collating program product |
CN101176116A (en) * | 2005-05-13 | 2008-05-07 | 三路影像公司 | Methods of chromogen separation-based image analysis |
Non-Patent Citations (3)
Title |
---|
Fast automated cell phenotype image classification;Nicholas A Hamilton等;《BMC Bioinformatics》;20070330;第2页第1行-第7页右栏倒数第4行 * |
Fusion of systems for automated cell phenotype image classification;Loris Nanni等;《Expert Systems with Applications》;20101231;第37卷;第1556页左栏第1节第1行-1562页左栏第6行及摘要 * |
蛋白质亚细胞定位预测的机器学习方法;张树波等;《计算机科学》;20090415;第36卷(第4期);第29页左栏第1节第1行-第32页左栏第1-22行 * |
Also Published As
Publication number | Publication date |
---|---|
CN103020488A (en) | 2013-04-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109034159B (en) | Image information extraction method and device | |
Wahab et al. | Two-phase deep convolutional neural network for reducing class skewness in histopathological images based breast cancer detection | |
CN103020488B (en) | A kind of Subcellular Localization method based on fluorescence microscope images | |
Sommer et al. | Learning-based mitotic cell detection in histopathological images | |
Chen et al. | Semi-automatic segmentation and classification of pap smear cells | |
Veta et al. | Automatic nuclei segmentation in H&E stained breast cancer histopathology images | |
CN105512683B (en) | Object localization method and device based on convolutional neural networks | |
Radhakrishnan et al. | Machine learning for nuclear mechano-morphometric biomarkers in cancer diagnosis | |
Li et al. | Deep convolutional neural networks for detecting secondary structures in protein density maps from cryo-electron microscopy | |
CN112508850B (en) | Deep learning-based method for detecting malignant area of thyroid cell pathological section | |
Ersoy et al. | HEp-2 cell classification in IIF images using shareboost | |
Saha et al. | Multi-lingual scene text detection and language identification | |
Nanni et al. | A reliable method for cell phenotype image classification | |
Veta et al. | Detecting mitotic figures in breast cancer histopathology images | |
Dos Santos et al. | Computer vision for virus image classification | |
JPWO2016017533A1 (en) | Identification device and identification method | |
Yousef Kalafi et al. | Automated identification of Monogeneans using digital image processing and K-nearest neighbour approaches | |
Khan et al. | Malaria parasite classification framework using a novel channel squeezed and boosted CNN | |
Alharbi et al. | Detection of peripheral malarial parasites in blood smears using deep learning models | |
Nateghi et al. | Automatic detection of mitosis cell in breast cancer histopathology images using genetic algorithm | |
Tahir et al. | Subcellular localization using fluorescence imagery: utilizing ensemble classification with diverse feature extraction strategies and data balancing | |
CN110826612A (en) | Training and identifying method for deep learning | |
Chen et al. | Neuron segmentation using deep complete bipartite networks | |
Sadanandan et al. | Feature augmented deep neural networks for segmentation of cells | |
Du et al. | Automatic classification of cells in microscopic fecal images using convolutional neural networks |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20161005 Termination date: 20201229 |
|
CF01 | Termination of patent right due to non-payment of annual fee |