CN109903296A - A kind of digital pcr drop detection method based on LBP-Adaboost algorithm - Google Patents
A kind of digital pcr drop detection method based on LBP-Adaboost algorithm Download PDFInfo
- Publication number
- CN109903296A CN109903296A CN201910118093.0A CN201910118093A CN109903296A CN 109903296 A CN109903296 A CN 109903296A CN 201910118093 A CN201910118093 A CN 201910118093A CN 109903296 A CN109903296 A CN 109903296A
- Authority
- CN
- China
- Prior art keywords
- image
- sample
- drop
- lbp
- digital pcr
- 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.)
- Granted
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 32
- 238000004422 calculation algorithm Methods 0.000 title claims abstract description 14
- 239000007788 liquid Substances 0.000 claims abstract description 88
- 108090000623 proteins and genes Proteins 0.000 claims abstract description 23
- 239000011159 matrix material Substances 0.000 claims abstract description 17
- 238000000034 method Methods 0.000 claims abstract description 16
- 238000012549 training Methods 0.000 claims abstract description 13
- 238000007847 digital PCR Methods 0.000 claims description 18
- 238000003752 polymerase chain reaction Methods 0.000 claims description 4
- 238000012216 screening Methods 0.000 claims description 4
- YVPYQUNUQOZFHG-UHFFFAOYSA-N amidotrizoic acid Chemical compound CC(=O)NC1=C(I)C(NC(C)=O)=C(I)C(C(O)=O)=C1I YVPYQUNUQOZFHG-UHFFFAOYSA-N 0.000 claims description 3
- 238000003384 imaging method Methods 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 230000000877 morphologic effect Effects 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 abstract description 9
- 238000010801 machine learning Methods 0.000 abstract description 6
- 230000011218 segmentation Effects 0.000 abstract description 2
- 239000002245 particle Substances 0.000 abstract 1
- 239000000523 sample Substances 0.000 description 39
- 238000006243 chemical reaction Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 238000001914 filtration Methods 0.000 description 3
- 230000004927 fusion Effects 0.000 description 3
- 230000003321 amplification Effects 0.000 description 2
- 230000001186 cumulative effect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000003199 nucleic acid amplification method Methods 0.000 description 2
- 238000012634 optical imaging Methods 0.000 description 2
- 238000013139 quantization Methods 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 238000012408 PCR amplification Methods 0.000 description 1
- 239000013614 RNA sample Substances 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000011304 droplet digital PCR Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 239000007850 fluorescent dye Substances 0.000 description 1
- 239000012634 fragment Substances 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 108020004707 nucleic acids Proteins 0.000 description 1
- 150000007523 nucleic acids Chemical class 0.000 description 1
- 102000039446 nucleic acids Human genes 0.000 description 1
- 239000002773 nucleotide Substances 0.000 description 1
- 125000003729 nucleotide group Chemical group 0.000 description 1
- 238000005191 phase separation Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000002035 prolonged effect Effects 0.000 description 1
- 238000011002 quantification Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Landscapes
- Image Analysis (AREA)
- Measuring Or Testing Involving Enzymes Or Micro-Organisms (AREA)
- Investigating Or Analysing Biological Materials (AREA)
- Image Processing (AREA)
Abstract
The digital pcr drop detection method based on LBP-Adaboost algorithm that the present invention provides a kind of generates cascade classifier including drop positive negative sample of the S1:Adaboost training based on LBP feature；S2: based on co-occurrence matrix segmentation drop " effective coverage ", binary image is generated；S3: Liquid particle image is divided into (border overlay) subgraph, and each subgraph loads cascade classifier respectively, calculates position and the gray value of tested drop；S4: the radius and volume of drop are calculated according to " effective coverage " interior droplet position；S5: the total volume and target gene concentration of all drops are calculated.The present invention is based on machine learning and image processing techniques, can fast and accurately position drip point, improve the accuracy rate for calculating target gene concentration using the drop picture of industrial high-definition camera shooting.
Description
Technical Field
The invention relates to the technical field of machine learning technology and image processing, and aims to identify and position liquid drops generated in a gene detection image and accurately calculate the concentration of a target gene.
Background
The PCR technique used for gene detection is to dilute and disperse a DNA or RNA sample into tens of thousands or even millions of independent reaction units, each reaction unit contains (or does not contain) 1 or more target molecules (DNA or RNA templates), perform molecular template PCR amplification on the target sequences of all the micro-reaction units, after amplification is completed, analyze negative or positive fluorescent signals of each reaction unit and perform statistical analysis (poisson distribution), and finally detect the concentration of nucleotides in the sample.
The droplet digital PCR system uses droplets obtained by oil-water two-phase separation as a micro-reaction unit, and after amplification is completed, the droplets can be labeled by a fluorescent probe for labeling a specific gene fragment. And judging whether the liquid drop is positive or negative according to whether the fluorescent signal is contained.
In an optical imaging system with a large field of view, the phenomenon of nonuniform brightness can be caused by the loss of light energy, so that the shooting effect of liquid drops at the edge of an image is deviated; a high-definition gene detection picture comprises 2 ten thousand of liquid drop points, and the time of detecting all the liquid drop points by using an image processing technology takes more than 90 seconds; in the liquid drop chip, the liquid drops are detected by mistake or missing due to factors such as uneven liquid drop size, liquid drop fusion, random distribution of liquid drop points, large brightness change of the positive and negative liquid drops and the like, and the real number of the liquid drops is influenced.
Disclosure of Invention
As mentioned above, the misdetection or omission of the liquid drops due to the factors of non-uniform size of the liquid drops, liquid drop fusion, random distribution of liquid drop points, large variation of brightness values of yin and yang liquid drops, and high brightness noise points affects the determination of the true number and the positivity and negativity of the liquid drops. Accurate detection of the number of droplets and calculation of the volume of a single droplet are the basis for the success of digital PCR genetic testing.
The invention aims to quickly and accurately position the dropping point and improve the accuracy of calculating the concentration of a target gene.
The method utilizes the liquid drop picture shot by an industrial high-definition camera, and detects the number of positive liquid drops, the total number of liquid drops, the volume of single liquid drop and the concentration of target genes contained in the liquid drop picture based on machine learning and image processing technology.
The invention is realized by the following technical scheme:
the invention provides a digital PCR (polymerase chain reaction) droplet detection method based on an LBP-Adaboost algorithm, which comprises the following steps of:
s1: adaboost trains a droplet positive and negative sample based on LBP characteristics to generate a cascade classifier;
s2: segmenting an effective area of the liquid drop based on the co-occurrence matrix to generate a binary image;
s3: dividing the liquid drop image into (boundary overlapping) sub-images, respectively loading a cascade classifier on each sub-image, and calculating the position and the gray value of the detected liquid drop;
s4: calculating the radius and the volume of the liquid drop according to the position of the liquid drop in the effective area;
s5: and calculating the total volume and the target gene concentration of all detected liquid drops.
It should be understood that the present invention is not limited to the above steps, and may also include other additional steps, for example, before step S1, between steps S1 and S2, between steps S2 and S3, between steps S3 and S4, between steps S4 and S5, and after step S5, without departing from the scope of the present invention.
Preferably, the step S1 includes:
s11, shooting gray level images with different exposure times and different concentration gradients under the condition of a stable imaging system;
s12, intercepting a positive sample containing complete liquid drops, normalizing the positive sample to be 24 x 24 with the same size as shown in figure 2, and carrying out Gaussian filtering on the positive sample; the negative sample contains no droplets or contains droplets in an area smaller thanThe negative sample size is larger than the positive sample;
s13, extracting LBP characteristics of the sample;
s14, generating a linear weighted strong classifier according to an Adaboost algorithm;
and S15, training a plurality of strong classifiers to form a cascade classifier.
It is understood that additional steps may be included before, during, or after the above-described steps without departing from the scope of the present invention.
Preferably, the step S2 includes:
s21, preprocessing an image;
s22, segmenting a liquid drop area by a symbiotic matrix;
s23, generating a binary imageWherein the "threshold" is set to 0.5, B(x，y)Representing the gray value at the binary image coordinates (x, y).
It is understood that additional steps may be included before, during, or after the above-described steps without departing from the scope of the present invention.
Preferably, the step S3 includes:
s31, graying the image;
s32, dividing the image into sub-blocks which are overlapped with each other;
s33, screening the drop points which are positioned for multiple times.
It is understood that additional steps may be included before, during, or after the above-described steps without departing from the scope of the present invention.
Preferably, the step S4 includes:
s41, mapping all the drop points positioned in the step three to the binary image generated in the step two, and screening the drop points in the effective area;
s42, counting droplets with the distance D (i) less than 50 pixels,calculating the average value of the distances between them Wherein t represents the total number of droplets D (i) less than 50 pixels from the droplet, and the radius of D (i) isWherein a is 0.4;
S43.D(i)radius ofDenotes the radius of the i-th droplet, D (i)Volume ofThe volume of the ith droplet is shown, wherein pi is 3.14;
s44. traverse all drops, calculate the radius and volume of each drop, fig. 10 shows a circle marked with the radius of the locating drop.
It is understood that additional steps may be included before, during, or after the above-described steps without departing from the scope of the present invention.
Preferably, the step S5 includes:
s51, calculating the total volume of the target genesWherein DN represents the number of all valid droplets;
s52, calculating the concentration of the target geneWherein q represents the negative rate of the target gene.
It is understood that additional steps may be included before, during, or after the above-described steps without departing from the scope of the present invention.
Further preferably, the step S13 includes:
s131, setting a coordinate point (x, y) to represent a sample position with the size of 24 x 24, and P (x, y) to represent a pixel value of the position, wherein x is more than or equal to 1, and y is less than or equal to 22;
s132, taking the point (x, y) as a center, comparing the gray values of the adjacent eight pixels with a threshold value P (x, y), if the values of the surrounding pixels are greater than the threshold value P (x, y), marking the position of the pixel point as 1, otherwise, marking the position as 0. 8 points in the eight neighborhood can produce an 8-bit binary number (usually converted to a decimal number or LBP code, 256 in total), i.e., the LBP value for point (x, y);
and S133, traversing pixel points at other positions to generate 484 LBP values, and combining the LBP values to generate a sample LBP characteristic, as shown in FIG. 3.
It is understood that additional steps may be included before, during, or after the above-described steps without departing from the scope of the present invention.
Further preferably, the step S14 includes:
s141. positive and negative sample sets (x) for a given droplet1，y1)，(x2，y2)，...，(xi，yi)，...，(xn，yn) Wherein x isiDenotes the ith sample, yi0 means that it is a negative sample, yi1 denotes that it is a positive sample, and n denotes the total number of positive and negative samples;
s142, initializing the weight of the weight,and represents the weight of positive samples, l being the total number of positive samples;representing weights of positive and negative samples, m being negative samplesTotal number;
and S143, generating the strong classifier through T iterations.
It is understood that additional steps may be included before, during, or after the above-described steps without departing from the scope of the present invention.
Further preferably, the step S15 includes:
s151, defining the minimum recognition rate and the maximum false detection rate of each level of strong classifier as d and F respectively, and defining the false detection rate of the cascade classifier as FTarget；
S152. define DiAnd FiRespectively representing the recognition rate and the error rate of the cascade classifier, niDenotes the number of weak classes in the ith strong class, P denotes the positive sample set, N denotes the negative sample set, initialization i is 0, D0＝1.0，F0＝1.0；
S153. when Fi＞FTargetThen, the iteration is as follows:
i＝i+1；
ni＝0，Fi＝Fi-1；
when F is presenti＞f*Fi-1Then, the following processes are performed:
ni＝ni+1；
under the conditions of P and N, N is obtained based on Adaboost algorithm trainingiA strong classifier of the weak classifier;
evaluating the current cascade classification DiAnd Fi；
Reduce the strong classifier threshold of the ith layer so that DiAt least equal to D x Di-1；
Setting N as an empty set;
Fi＞FtargetThen, the negative sample is evaluated by the current cascade classifier, andthe misrecognized samples are placed in negative samples.
It is understood that additional steps may be included before, during, or after the above-described steps without departing from the scope of the present invention.
Further preferably, the step S21 includes:
s211, selecting 3 x 3 square kernels, and performing morphological gradient on the image;
s212, image contrast enhancement.
It is understood that additional steps may be included before, during, or after the above-described steps without departing from the scope of the present invention.
Further preferably, the step S22 includes:
s221, dividing the droplet image into sub-images with the size of 4 x 4, wherein the number of the sub-images is equal to that of the sub-imagesWherein "height" represents the height of the image and "width" represents the width of the image;
and S222, calculating the contrast of the co-occurrence matrix of each sub-image.
It is understood that additional steps may be included before, during, or after the above-described steps without departing from the scope of the present invention.
The technique used in the present invention is defined as follows:
"Local Binary Pattern (LBP)" is a valid feature descriptor.
"Adaboost" is a machine learning algorithm, training different classifiers (weak classifiers) aiming at the same training set, and then combining the weak classifiers to form a stronger final classifier (strong classifier);
"Digital PCR, or Digital PCR (dPCR)", is an absolute nucleic acid molecular quantification technique.
The invention has obvious technical progress:
one gene detection picture (3376 × 2704) approximately comprises about 2 ten thousand of liquid drop points, the time for positioning all liquid drops by the image processing technology is about 90 seconds, the technical scheme divides one picture into 3860 sub-image blocks, the liquid drops are detected in parallel based on a cascade classifier, the detection time is about 20 seconds, and the time is obviously shortened.
In an optical imaging system with a large field of view, the loss of light energy causes uneven brightness of a gene detection image, and firstly, the method adopts methods such as histogram equalization, contrast enhancement, filtering and the like to preprocess the image; secondly, the characteristics of the liquid drop are extracted by utilizing the gray scale invariance of the LBP.
In the liquid drop chip, the sizes of liquid drops are not uniform, the liquid drops are fused, liquid drop points are randomly distributed, and the brightness values of the positive and negative liquid drops are greatly changed.
In the prior art, the same type of patents process the liquid drop pictures based on an image processing method, and the patent trains a large sample to detect the liquid drop pictures based on a machine learning method; the method combining image processing, co-occurrence matrix segmentation, parallel detection and the like is different from other inventions innovatively, and remarkable technical effects are achieved.
Drawings
FIG. 1 is a flow chart of the technical solution of the present invention as shown in FIG. 1.
Fig. 2 is a diagram of step s12. a positive sample containing intact droplets, normalized to the same size 24 x 24.
Fig. 3 is a step s133, traversing pixel points at all positions of the sample to generate 484 LBP values in total, and combining the LBP values to generate the sample LBP feature.
FIG. 4 is a step S2222. extract gray level co-occurrence matrix map of the sub image blocks in four directions of 0, 45, 90, 135, and the pixel pairs in the coarse quantization image<0，2>Is 2, and the element value R corresponding to the horizontal co-occurrence matrix coordinate point (0, 2)(0，2)Is 2.
FIG. 5 shows, step S23. generating a binarized imageWherein the "threshold" is set to 0.5, B(x，y)Representing the gray value at the binary image coordinates (x, y).
Fig. 6 is a diagram of step s32. the image is divided into sub-blocks overlapping each other, the black area indicates the overlap, and the width of the overlap area is 30 pixels.
Fig. 7 shows each sub-block image separately loaded with a cascade classifier, locating the contained drop points of each sub-block.
Fig. 8 shows that the larger the number of divided blocks, the shorter the detection time.
Fig. 9 shows the effect of positioning the image part area of the drop.
Figure 10 shows a circle of radius marks of the located droplets.
Detailed Description
The invention is further illustrated by the following examples, which are not intended to limit the scope of the invention.
The key points of the implementation of the invention are as follows:
(1) the liquid drop image is divided into 3840 parts of subblock images, the LBP extracts the liquid drop characteristics of the subblock images (other characteristic extraction algorithms are not as fast as the LBP time), and the cascade classifier is utilized to detect the liquid drops in parallel, so that the liquid drop detection time is prolonged.
(2) The symbiotic matrix calculates the contrast of the subimages in different directions, and divides a droplet area and a non-droplet area of the gene detection image by adopting a standard difference of the contrast, so that the error division of the droplet boundary caused by the nonuniformity of the light path is reduced, the droplet boundary area is accurately positioned, and non-droplet points and irregular droplet areas are quickly eliminated.
(3) According to the conditions of non-uniform droplet size, droplet fusion, random droplet point distribution, large brightness value change of yin and yang droplets and the like, positive and negative samples with large characteristic difference are manufactured based on LBP characteristics, the number of the positive samples is 4.7 ten thousand, the number of the negative samples is 7.3 ten thousand, the number of weak classifications is 100, 19 generated cascade classifiers are manufactured, the time spent in one training is about 35 minutes, and the correct characteristic selection is suitable and appropriate training parameters, so that the sample training time is reduced.
(4) The gray values of the drop photos with different exposure times and different concentration gradients are greatly different, and firstly, the characteristics of the drops are extracted by utilizing LBP gray scale invariance (the influence of nonuniform light paths is weakened); and secondly, processing the image area with larger or smaller brightness by histogram equalization.
(5) Counting all adjacent liquid drops within a certain range from the central liquid drop, counting the distance between the adjacent liquid drops and the central liquid drop, calculating the radius of the liquid drop, deleting abnormal liquid drops with overlarge or undersize radius, and calculating the volume of the liquid drops according to a sphere volume formula.
Referring to fig. 1, a basic flow chart of the technical solution of the present invention is shown.
A digital PCR liquid drop detection method based on LBP-Adaboost algorithm can accurately detect the position, radius and quantity of liquid drops in a gene image and calculate the liquid drop volume and the target gene concentration, and the method comprises the following steps:
s1: adaboost trains a droplet positive and negative sample based on LBP characteristics to generate a cascade classifier;
s2: segmenting an effective area of the liquid drop based on the co-occurrence matrix to generate a binary image;
s3: dividing the liquid drop image into (boundary overlapping) sub-images, respectively loading a cascade classifier on each sub-image, and calculating the position and the gray value of the detected liquid drop;
s4: calculating the radius and the volume of the liquid drop according to the position of the liquid drop in the effective area;
s5: the total volume and target gene concentration of all droplets were calculated.
The detailed process of the technical scheme of the invention is as follows:
in an embodiment of the present invention, S1.adaboost trains droplet positive and negative samples based on LBP features to generate a cascade classifier, where step S1 includes:
s11, shooting gray level images with different exposure times and different concentration gradients under the condition of a stable imaging system;
s12, intercepting a positive sample containing complete liquid drops, normalizing the positive sample to be 24 x 24 with the same size as shown in figure 2, and carrying out Gaussian filtering on the positive sample; the negative sample contains no droplets or contains droplets in an area smaller thanThe negative sample size is larger than the positive sample;
s13, extracting LBP characteristics of the sample;
in an embodiment of the present invention, the step S13 includes:
s131, setting a coordinate point (x, y) to represent a sample position with the size of 24 x 24, and P (x, y) to represent a pixel value of the position, wherein x is more than or equal to 1, and y is less than or equal to 22;
s132, taking the point (x, y) as a center, comparing the gray values of the adjacent eight pixels with a threshold value P (x, y), if the values of the surrounding pixels are greater than the threshold value P (x, y), marking the position of the pixel point as 1, otherwise, marking the position as 0. 8 points in the eight neighborhood can produce an 8-bit binary number (usually converted to a decimal number or LBP code, 256 in total), i.e., the LBP value for point (x, y);
s133, traversing pixel points at other positions to generate 484 LBP values, and combining the LBP values to generate a sample LBP characteristic, as shown in FIG. 3;
s14, generating a linear weighted strong classifier according to an Adaboost algorithm;
in an embodiment of the present invention, the step S14 includes:
s141. positive and negative sample sets (x) for a given droplet1，y1)，(x2，y2)，...，(xi，yi)，...，(xn，yn) Wherein x isiDenotes the ith sample, yi0 means that it is a negative sample, yi1 denotes that it is a positive sample, and n denotes the total number of positive and negative samples;
s142, initializing the weight of the weight,and represents the weight of positive samples, l being the total number of positive samples;representing the weight of the positive and negative samples, m being the total number of negative samples;
and S143, generating the strong classifier through T iterations.
In an embodiment of the present invention, the step S143 includes:
s1431, normalizing the weight,wherein T is more than or equal to 0 and less than or equal to T;
s1432, for each feature f, training a corresponding weak classifier hfCalculate hfAnd find the weighted error rate with the minimum error rate epsilontAs an optimal classifier ht，εt＝min∑wi|h(xi)-yi|；
S1433, the weight is updated,whereinεi0 represents xiIs divided correctly, otherwise is divided wrongly;
s1434, the finally generated strong classifier:wherein,
s15, training a plurality of strong classifiers to form a cascade classifier;
in an embodiment of the present invention, the step S15 includes:
s151, defining the minimum recognition rate and the maximum false detection rate of each level of strong classifier as d and F respectively, and defining the false detection rate of the cascade classifier as FTarget；
S152. define DiAnd FiRespectively representing the recognition rate and the error rate of the cascade classifier, niDenotes the number of weak classes in the ith strong class, P denotes the positive sample set, N denotes the negative sample set, initialization i is 0, D0＝1.0，F0＝1.0；
S153. when Fi＞FTargetThen, the iteration is as follows:
i＝i+1：
ni＝0，Fi＝Fi-1；
when F is presenti＞f*Fi-1Then, the following processes are performed:
ni＝ni+1；
under the conditions of P and N, N is obtained based on Adaboost algorithm trainingiA strong classifier of the weak classifier;
evaluating the current cascade classification DiAnd Fi；
Reduce the strong classifier threshold of the ith layer so that DiAt least equal to D x Di-1；
Setting N as an empty set;
Fi＞FtargetThen the negative examples are evaluated by the current cascade classifier and the misrecognized examples are placed in the negative examples.
In the embodiment of the present invention, S2. a droplet "effective region" is segmented based on a co-occurrence matrix to generate a binarized image, and the step S2 includes:
s21, preprocessing an image;
in an embodiment of the present invention, the step S21 includes:
s211, selecting 3 x 3 square kernels, and performing morphological gradient on the image;
s212, enhancing image contrast;
in an embodiment of the present invention, the step S212 includes:
s2121, setting the maximum gray value and the minimum gray value of an image as V respectivelymaxAnd VminProbability density of gray value i ofWherein N represents the total number of pixels, NiIndicates the number of pixels having a gray value of i, Vmin≤i≤Vmax；
S2122, calculating the cumulative probability density of the image
S2123. calculating Ci0.1 and Ci0.9 cumulative histogram corresponding pixel value VIs low inAnd VHeight of；
S2124, traversing the whole image, and updating the gray value of the imageWherein V(x，y)Indicates the gray value, F, corresponding to the coordinate point (x, y) of the original image(x，y)Representing the updated pixel values;
s22, segmenting a liquid drop area by a symbiotic matrix;
in an embodiment of the present invention, the step S22 includes:
s221, dividing the droplet image into sub-images with the size of 4 x 4, wherein the number of the sub-images is equal to that of the sub-imagesWherein "height" represents the height of the image and "width" represents the width of the image;
s222, calculating the contrast ratio of the co-occurrence matrix of each sub-image;
in an embodiment of the present invention, the step S222 includes:
s2221, carrying out gray level coarse quantization processing,wherein F(x，y)Representing the gray value of the image, x is more than or equal to 0 and less than or equal to 3, y is more than or equal to 0 and less than or equal to 3, and the gray level of the coarsened image is [0, 8%]；
S2222, extracting gray level co-occurrence matrixes of the sub image blocks in four directions of 0 degree, 45 degrees, 90 degrees and 135 degrees, wherein statistics of the co-occurrence matrixes in the horizontal direction are as follows: let (m, n) denote a coordinate point of the co-occurrence matrix,<m，n>representing a coarsely quantized image pixel pair (0. ltoreq. m, n. ltoreq.7) and (m, n) corresponding element values R(m，n)Representing imagesCoarsely quantized pixel pair<m，n>The number of pixel pairs in the coarse quantized image, as shown in FIG. 4<0，2>Is 2, and the element value R corresponding to the horizontal co-occurrence matrix coordinate point (0, 2)(0，2)Is 2;
s2223, contrast values in four directions of 0 degrees, 45 degrees, 90 degrees and 135 degrees are respectively calculated, and the formula is as follows: where i denotes four different directions, i ∈ [1, 4 ]]；
S2224, based on the contrast values of the four directions of 0 degree, 45 degrees, 90 degrees and 135 degrees, calculating the standard deviation of the contrastWherein
S23, generating a binary imageWherein the "threshold" is set to 0.5, B(x，y)Representing the gray values at the coordinates (x, y) of the binarized image as shown in fig. 5.
In the embodiment of the invention, S3, the droplet image is divided into sub-images, each sub-image is respectively loaded with a cascade classifier, and the position and the gray value of the detected droplet are calculated. The step S3 includes:
s31, graying the image;
s32, dividing the image into mutually overlapped subblocks, wherein as shown in FIG. 6, a black area represents overlapping, the width of the overlapping area is 30 pixels, and each subblock image is respectively loaded with a cascade classifier to locate all droplet areas Size [ D (i) ]subblocksx，D(i)y，D(i)h，D(i)w]Wherein D (i)xAnd D (i)yDenotes the coordinate position of the upper left corner of the ith droplet, D (i)hAnd D (i)wThe height and width of the ith droplet are shown, and as shown in FIG. 7, the coordinates of the center of the ith droplet are shownWhereinFIG. 8 shows that the larger the number of divided blocks, the shorter the detection time;
s33. the droplets of the overlapping area may be positioned several times if the distance D between two positioning points D (i) and D (j)ijAnd if the number of the pixels is less than 12, combining two positioning drops into one positioning point, taking the coordinates of the positioning point as the average value of the coordinates D (i) and D (j), and taking the distance between D (i) and D (j) as the distance
In an embodiment of the present invention, S4. calculating the radius and the volume of the droplet according to the position of the droplet in the "effective area", the step S4 includes:
s41, mapping all the drop points positioned in the third step to the binary image generated in the second step, if coordinate points are located in the binary imageIf the corresponding pixel value is 255, the droplet point is valid, otherwise, the droplet point is deleted, and finally all the droplets are positioned, and fig. 9 shows the positioning effect of the partial region of the droplet image;
s42, counting the distance liquid drops D (i) and liquid drops smaller than 50 pixels, and calculating the average value of the distances among the liquid drops Wherein t represents the total number of droplets D (i) less than 50 pixels from the droplet, and the radius of D (i) isWherein α is 0.4;
S43.D(i)radius ofDenotes the radius of the i-th droplet, D (i)Volume ofThe volume of the ith droplet is shown, wherein pi is 3.14;
s44, traversing all the liquid drops, and calculating the radius and the volume of each liquid drop, wherein a circle marked by the radius of the positioned liquid drop is shown in the figure 10;
in the embodiment of the present invention, S5, calculating the total volume and the target gene concentration of all droplets, where the step S5 includes:
s51, calculating the total volume of all positioned liquid dropsWherein DN represents the number of all valid droplets;
s52, calculating the concentration of the target geneWherein q represents the negative rate of the target gene.
The invention utilizes the liquid drop picture shot by the industrial high-definition camera, and based on machine learning and image processing technology, the invention can detect the number of positive liquid drops, the total number of liquid drops, the volume of single liquid drop and the concentration of target genes contained in the liquid drop image, and can quickly and accurately position the liquid drop point, thereby improving the accuracy rate of calculating the concentration of the target genes.
The above embodiments are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be construed as equivalents thereof, and all such changes, modifications, substitutions, combinations, and simplifications are intended to be included in the scope of the present invention.
Claims (11)
1.A digital PCR (polymerase chain reaction) droplet detection method based on an LBP-Adaboost algorithm is characterized by comprising the following steps of:
s1: adaboost trains a droplet positive and negative sample based on LBP characteristics to generate a cascade classifier;
s2: segmenting an effective area of the liquid drop based on the co-occurrence matrix to generate a binary image;
s3: dividing the liquid drop image into sub-images, respectively loading a cascade classifier on each sub-image, and calculating the position and the gray value of the detected liquid drop;
s4: calculating the radius and the volume of the liquid drop according to the position of the liquid drop in the effective area;
s5: and calculating the total volume and the target gene concentration of all detected liquid drops.
2. The digital PCR droplet detection method of claim 1, wherein the step S1 includes:
s11, shooting gray level images with different exposure times and different concentration gradients under the condition of a stable imaging system;
s12, intercepting a positive sample containing complete liquid drops, and normalizing the positive sample into a positive sample with the same size of 24 x 24;
s13, extracting LBP characteristics of the sample;
s14, generating a linear weighted strong classifier according to an Adaboost algorithm;
and S15, training a plurality of strong classifiers to form a cascade classifier.
3. The digital PCR droplet detection method of claim 1, wherein the step S2 includes:
s21, preprocessing an image;
s22, segmenting a liquid drop area by a symbiotic matrix;
and S23, generating a binary image.
4. The digital PCR droplet detection method of claim 1, wherein the step S3 includes:
s31, graying the image;
s32, dividing the image into sub-blocks which are overlapped with each other;
s33, screening the drop points which are positioned for multiple times.
5. The digital PCR droplet detection method of claim 1, wherein the step S4 includes:
s41, mapping all the drop points positioned in the step three to the binary image generated in the step two, and screening the drop points in the effective area;
s42, counting droplets with the distance D (i) less than 50 pixels, and calculating the average value of the distances among the droplets;
S43.D(i)radius ofDenotes the radius of the i-th droplet, D (i)Volume ofThe volume of the ith droplet is shown, wherein pi is 3.14;
s44, traversing all the liquid drops, and calculating the radius and the volume of each liquid drop.
6. The digital PCR droplet detection method of claim 1, wherein the step S5 includes:
s51, calculating the total volume of the detected liquid drops;
s52, calculating the concentration of the target gene.
7. The digital PCR droplet detection method of claim 2, wherein the step S13 includes:
s131, setting a coordinate point (x, y) to represent the pixel position of the sample image;
s132, taking the point (x, y) as a center, comparing the gray values of the eight adjacent pixels with a threshold value P (x, y) to generate a binary LBP value;
and S133, traversing pixel points at other positions, generating an LBP value of the sample image, and combining the LBP values to generate a sample LBP characteristic.
8. The digital PCR droplet detection method of claim 2, wherein the step S14 includes:
s141. positive and negative sample sets (x) for a given droplet1,y1),(x2,y2),…,(xi,yi),…,(xn,yn) Wherein x isiDenotes the ith sample, yi0 means that it is a negative sample, yiTable 1 (the attached drawings)It is shown as a positive sample, n represents the total number of positive and negative samples;
s142, initializing the weight of the weight,and represents the weight of positive samples, l being the total number of positive samples;representing the weight of the positive and negative samples, m being the total number of negative samples;
and S143, generating the strong classifier through T iterations.
9. The digital PCR droplet detection method of claim 2, wherein the step S15 includes:
s151, defining the minimum recognition rate and the maximum false detection rate of each level of strong classifier as d and F respectively, and defining the false detection rate of the cascade classifier as FTarget；
S152. define DiAnd FiRespectively representing the recognition rate and the error rate of the cascade classifier, niDenotes the number of weak classes in the ith strong class, P denotes the positive sample set, N denotes the negative sample set, initialization i is 0, D0＝1.0，F0＝1.0；
S153. when Fi＞FTargetThen, the iteration is as follows:
i＝i+1；
ni＝0，Fi＝Fi-1；
when F is presenti＞f*Fi-1Then, the following processes are performed:
ni＝ni+1；
under the conditions of P and N, N is obtained based on Adaboost algorithm trainingiA strong classifier of the weak classifier;
evaluating the current cascade classification DiAnd Fi；
Reduce the strong classifier threshold of the ith layer so that DiAt least equal to D x Di-1；
Setting N as an empty set;
Fi＞FtargetThen the negative examples are evaluated by the current cascade classifier and the misrecognized examples are placed in the negative examples.
10. The digital PCR droplet detecting method according to claim 3, wherein the step S21 includes:
s211, selecting 3 x 3 square kernels, and performing morphological gradient on the image;
s212, image contrast enhancement.
11. The digital PCR droplet detecting method according to claim 3, wherein the step S22 includes:
s221, dividing the droplet image into sub-images with the size of 4 x 4, wherein the number of the sub-images is equal to that of the sub-imagesWherein "height" represents the height of the image and "width" represents the width of the image;
and S222, calculating the contrast ratio of the co-occurrence matrix of each sub-image in different directions.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910118093.0A CN109903296B (en) | 2019-02-15 | 2019-02-15 | LBP-Adaboost algorithm-based digital PCR (polymerase chain reaction) droplet detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910118093.0A CN109903296B (en) | 2019-02-15 | 2019-02-15 | LBP-Adaboost algorithm-based digital PCR (polymerase chain reaction) droplet detection method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109903296A true CN109903296A (en) | 2019-06-18 |
| CN109903296B CN109903296B (en) | 2021-06-01 |
Family
ID=66944756
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910118093.0A Active CN109903296B (en) | 2019-02-15 | 2019-02-15 | LBP-Adaboost algorithm-based digital PCR (polymerase chain reaction) droplet detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109903296B (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110734854A (en) * | 2019-09-25 | 2020-01-31 | 中国科学院苏州生物医学工程技术研究所 | -body rapid detection system for real-time fluorescence quantitative analysis of ultrahigh-flux single-cell nucleic acid molecules |
| CN111986175A (en) * | 2020-08-19 | 2020-11-24 | 北京科技大学 | Method for measuring particle size of liquid drop sprayed by industrial nozzle |
| CN112070711A (en) * | 2020-06-04 | 2020-12-11 | 清华大学 | Analysis method of micro-droplets in micro-droplet image detection method |
| CN113846149A (en) * | 2021-09-28 | 2021-12-28 | 领航基因科技（杭州）有限公司 | Digital PCR real-time analysis method of micropore array chip |
| CN114038506A (en) * | 2021-11-09 | 2022-02-11 | 领航基因科技（杭州）有限公司 | Micro-drop type digital PCR high-concentration detection method |
| CN114549528A (en) * | 2022-04-26 | 2022-05-27 | 浙江大学 | Micro-droplet digital PCR (polymerase chain reaction) droplet detection method and system |
| CN114540469A (en) * | 2022-01-11 | 2022-05-27 | 深圳大学 | Digital nucleic acid quantification method based on non-uniform volume liquid drops and image processing |
| CN114638832A (en) * | 2022-05-19 | 2022-06-17 | 深圳市中科先见医疗科技有限公司 | DPCR liquid drop fluorescence detection method based on watershed algorithm |
| CN114648529A (en) * | 2022-05-19 | 2022-06-21 | 深圳市中科先见医疗科技有限公司 | DPCR liquid drop fluorescence detection method based on CNN network |
| CN117095362A (en) * | 2023-10-20 | 2023-11-21 | 中国科学院苏州生物医学工程技术研究所 | Liquid drop monitoring method, system and storage medium of cell sorter |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102800107A (en) * | 2012-07-06 | 2012-11-28 | 浙江工业大学 | Motion target detection method based on improved minimum cross entropy |
| CN103673872A (en) * | 2012-09-10 | 2014-03-26 | 中国科学院空间科学与应用研究中心 | Measurement method and measurement system of liquid drop volume |
| CN104077613A (en) * | 2014-07-16 | 2014-10-01 | 电子科技大学 | Crowd density estimation method based on cascaded multilevel convolution neural network |
| CN104143047A (en) * | 2014-07-21 | 2014-11-12 | 华北电力大学（保定） | Automatic tissue calibration method for IVUS gray-scale image |
| CN105139011A (en) * | 2015-09-07 | 2015-12-09 | 浙江宇视科技有限公司 | Method and apparatus for identifying vehicle based on identification marker image |
| CN106030608A (en) * | 2013-11-06 | 2016-10-12 | 理海大学 | Diagnostic system and method for biological tissue analysis |
| CN106683073A (en) * | 2015-11-11 | 2017-05-17 | 杭州海康威视数字技术股份有限公司 | License plate detection method, camera and server |
| CN107844683A (en) * | 2017-10-17 | 2018-03-27 | 华东医药（杭州）基因科技有限公司 | A kind of computational methods of digital pcr concentration of liquid drops |
| US20180165552A1 (en) * | 2016-12-12 | 2018-06-14 | National Chung Shan Institute Of Science And Technology | All-weather thermal-image pedestrian detection method |
| CN108171157A (en) * | 2017-12-27 | 2018-06-15 | 南昌大学 | The human eye detection algorithm being combined based on multiple dimensioned localized mass LBP histogram features with Co-HOG features |
| CN108897984A (en) * | 2018-05-07 | 2018-11-27 | 上海理工大学 | Based on correlation analysis between CT images group feature and lung cancer gene expression |
| CN109145559A (en) * | 2018-08-02 | 2019-01-04 | 东北大学 | A kind of intelligent terminal face unlocking method of combination Expression Recognition |
-
2019
- 2019-02-15 CN CN201910118093.0A patent/CN109903296B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102800107A (en) * | 2012-07-06 | 2012-11-28 | 浙江工业大学 | Motion target detection method based on improved minimum cross entropy |
| CN103673872A (en) * | 2012-09-10 | 2014-03-26 | 中国科学院空间科学与应用研究中心 | Measurement method and measurement system of liquid drop volume |
| CN106030608A (en) * | 2013-11-06 | 2016-10-12 | 理海大学 | Diagnostic system and method for biological tissue analysis |
| CN104077613A (en) * | 2014-07-16 | 2014-10-01 | 电子科技大学 | Crowd density estimation method based on cascaded multilevel convolution neural network |
| CN104143047A (en) * | 2014-07-21 | 2014-11-12 | 华北电力大学（保定） | Automatic tissue calibration method for IVUS gray-scale image |
| CN105139011A (en) * | 2015-09-07 | 2015-12-09 | 浙江宇视科技有限公司 | Method and apparatus for identifying vehicle based on identification marker image |
| CN106683073A (en) * | 2015-11-11 | 2017-05-17 | 杭州海康威视数字技术股份有限公司 | License plate detection method, camera and server |
| US20180165552A1 (en) * | 2016-12-12 | 2018-06-14 | National Chung Shan Institute Of Science And Technology | All-weather thermal-image pedestrian detection method |
| CN107844683A (en) * | 2017-10-17 | 2018-03-27 | 华东医药（杭州）基因科技有限公司 | A kind of computational methods of digital pcr concentration of liquid drops |
| CN108171157A (en) * | 2017-12-27 | 2018-06-15 | 南昌大学 | The human eye detection algorithm being combined based on multiple dimensioned localized mass LBP histogram features with Co-HOG features |
| CN108897984A (en) * | 2018-05-07 | 2018-11-27 | 上海理工大学 | Based on correlation analysis between CT images group feature and lung cancer gene expression |
| CN109145559A (en) * | 2018-08-02 | 2019-01-04 | 东北大学 | A kind of intelligent terminal face unlocking method of combination Expression Recognition |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110734854A (en) * | 2019-09-25 | 2020-01-31 | 中国科学院苏州生物医学工程技术研究所 | -body rapid detection system for real-time fluorescence quantitative analysis of ultrahigh-flux single-cell nucleic acid molecules |
| CN112070711A (en) * | 2020-06-04 | 2020-12-11 | 清华大学 | Analysis method of micro-droplets in micro-droplet image detection method |
| CN111986175A (en) * | 2020-08-19 | 2020-11-24 | 北京科技大学 | Method for measuring particle size of liquid drop sprayed by industrial nozzle |
| CN113846149A (en) * | 2021-09-28 | 2021-12-28 | 领航基因科技（杭州）有限公司 | Digital PCR real-time analysis method of micropore array chip |
| CN114038506A (en) * | 2021-11-09 | 2022-02-11 | 领航基因科技（杭州）有限公司 | Micro-drop type digital PCR high-concentration detection method |
| CN114540469A (en) * | 2022-01-11 | 2022-05-27 | 深圳大学 | Digital nucleic acid quantification method based on non-uniform volume liquid drops and image processing |
| CN114549528A (en) * | 2022-04-26 | 2022-05-27 | 浙江大学 | Micro-droplet digital PCR (polymerase chain reaction) droplet detection method and system |
| CN114638832A (en) * | 2022-05-19 | 2022-06-17 | 深圳市中科先见医疗科技有限公司 | DPCR liquid drop fluorescence detection method based on watershed algorithm |
| CN114648529A (en) * | 2022-05-19 | 2022-06-21 | 深圳市中科先见医疗科技有限公司 | DPCR liquid drop fluorescence detection method based on CNN network |
| CN117095362A (en) * | 2023-10-20 | 2023-11-21 | 中国科学院苏州生物医学工程技术研究所 | Liquid drop monitoring method, system and storage medium of cell sorter |
| CN117095362B (en) * | 2023-10-20 | 2024-02-02 | 中国科学院苏州生物医学工程技术研究所 | Liquid drop monitoring method, system and storage medium of cell sorter |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109903296B (en) | 2021-06-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109903296B (en) | LBP-Adaboost algorithm-based digital PCR (polymerase chain reaction) droplet detection method | |
| CN108596166B (en) | Container number identification method based on convolutional neural network classification | |
| CN113160192B (en) | Visual sense-based snow pressing vehicle appearance defect detection method and device under complex background | |
| Sari et al. | Road crack detection using support vector machine (SVM) and OTSU algorithm | |
| CN107545239B (en) | Fake plate detection method based on license plate recognition and vehicle characteristic matching | |
| CN106934795B (en) | A kind of automatic testing method and prediction technique of glue into concrete beam cracks | |
| CN107066933B (en) | Road sign identification method and system | |
| CN110956169B (en) | License plate recognition method and device and electronic equipment | |
| CN107633226B (en) | Human body motion tracking feature processing method | |
| CN107103317A (en) | Fuzzy license plate image recognition algorithm based on image co-registration and blind deconvolution | |
| CN111626249B (en) | Method and device for identifying geometric figure in topic image and computer storage medium | |
| CN112580383B (en) | Two-dimensional code identification method and device, electronic equipment and storage medium | |
| CN111222507B (en) | Automatic identification method for digital meter reading and computer readable storage medium | |
| Zhang et al. | Automatic detection of road traffic signs from natural scene images based on pixel vector and central projected shape feature | |
| CN112307919A (en) | Improved YOLOv 3-based digital information area identification method in document image | |
| CN111259893A (en) | Intelligent tool management method based on deep learning | |
| CN115457044B (en) | Pavement crack segmentation method based on class activation mapping | |
| Akbarzadeh et al. | Design and matlab simulation of Persian license plate recognition using neural network and image filtering for intelligent transportation systems | |
| CN112101283A (en) | Intelligent identification method and system for traffic signs | |
| Çavdaroğlu et al. | A character segmentation method to increase character recognition accuracy for Turkish number plates | |
| Yang et al. | Recognizing license plates in real-time | |
| Harikiran et al. | A new method of gridding for spot detection in microarray images | |
| CN114384073B (en) | Subway tunnel crack detection method and system | |
| CN108710915B (en) | Multi-feature fusion gastroscope image processing method based on multi-kernel learning | |
| CN114038506A (en) | Micro-drop type digital PCR high-concentration detection method |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20211202 Address after: 518101 g1316, Lianxing building, building B, Yihua new village, district 46, Haifu community, Xin'an street, Bao'an District, Shenzhen City, Guangdong Province Patentee after: Pilot medical technology (Shenzhen) Co.,Ltd. Address before: 310000 room 1114, Jin Jun Road, 341 Shui Xiang Road, Jianggan District, Hangzhou, Zhejiang. Patentee before: PILOT GENE TECHNOLOGIES (HANGZHOU) Co.,Ltd. |