CN1206534C - Method of identifying fluorescence dyeing sperm by nerve net - Google Patents

Abstract

The present invention belongs to the technical field of digital image nerve net recognition in biomedicine, which relates to a method of recognizing fluorescence dyeing sperms by a nerve net. The present invention comprises: fluorescent staining is carried out to sperm samples which are to be measured; a micrographic video camera instrument shoots a fluorescent staining sample which becomes the signals of original images; according to the nerve net recognition technique, pattern classification is carried out; a sequence signal of a fluorescent original image belongs to one mode of classification modes, which is determined in advance; the fluorescent original image is adjusted; quality control inspection is carried out; 5 to 20 visual fields are selected from the sequence signals of the original image obtained by using the micrographic video camera instrument, and each visual field has 3 to 16 frames; a trichromatic original drawing and a two-value reduced drawing for class selection are obtained, and dynamic and static statistical counting of sperms is carried out; the sperm number is adjusted, and a final dynamic and static recognition result of the sperms is obtained. The present invention can separate the motility rate and the survival rate of the sperms strictly so that the detection of sperm density achieves the accuracy of more than 95 %.

Description

A kind of method of nerve net identification fluorescent dye sperm

Technical field the invention belongs to digital image nerve net recognition technology field in the biomedicine, particularly is used for the sperm automatic checkout system has higher intelligent and stronger functional nerve net identification fluorescent dye sperm than gray scale identification method.

Technical background is at present both at home and abroad in the automatic analyzing and testing of sperm system, the optic probe that has is that the double frequency that adopts dodges principle, what the system that has adopted without exception is the microimaging technology, in the sperm automatic checkout system, its camera (CCD) is no matter be color system, or black-and-white system, its recognition technology nearly all is to adopt the method for gray scale, just the sperm in the seminal fluid is for black and is made as 1, background is a white, be made as 0, this method is difficult to accurately distinguish sperm and non-sperm composition; When the red white corpuscle in the seminal fluid and the identification of the reproduction cell that comes off are gone up, cause sperm concentration excessive; When sperm identification not being gone up, cause the density of sperm low excessively; Particularly when the yardstick of non-sperm composition is the same with sperm with gray scale, during the same non-sperm composition of grey level, method therefor can't be distinguished sperm and the non-sperm composition in the seminal fluid, can't distinguish at all.That is to say and to reject the impurity similar to sperm, cause all measured results of analytic system all to have very big error, in order to address this problem existing utilization the " probe is learned in infrared double frequency flash of light ", but also can only solve partial impurities, still can not deal with problems at all.

Summary of the invention the objective of the invention is can't distinguish sperm and non-sperm composition in the seminal fluid for solving existing method, cause all measured results of analytic system all to have the problem of very big error, a kind of method of nerve net identification fluorescent dye sperm is proposed, adopted the fluorescent dye sperm, and combine " nerve net recognition technology ", the present invention can strictly distinguish the motility rate and the survival rate of sperm, makes the detection of sperm concentration reach accuracy more than 95%.

The method of a kind of nerve net identification fluorescent dye sperm that the present invention proposes may further comprise the steps:

1). seminal fluid sample to be measured is carried out fluorescent dye;

2) become original image signal by microimaging instrument production fluorescent dye sample, and be presented on the video screen;

3) according to the nerve net recognition technology fluoroscopic image of shown various different brightness is learnt all sidedly, and pattern classification is carried out in memory further; Give and determine that earlier fluorescence raw image sequence signal belongs to a kind of pattern in the described classification mode;

4) the fluorescence raw image is carried out color, brightness, contrast, tone and adjust, make it become clearly trichromatic diagram and show;

5) adjusted three look original graph being removed ground unrest becomes the denoising sound spectrogram, again the denoising sound spectrogram is carried out binary conversion treatment and become binary picture, and the sperm that will be identified is marked; This three looks original graph and the binary picture of sperm being made sign are compared, if the sperm count of the two differs more than 5%, then reselect classification, again carry out 3)-5) step, sperm count until the two differs in 5%, then is as the criterion with this type of other three look original graph binary picture corresponding with it;

6) choose 5-20 the visual field from the raw image sequence signal of microimaging, each visual field 3-16 frame obtains three look original graph and binary pictures of selected classification, carries out sperm sound attitude statistical counting;

7) adjustment of sperm count: check whether testing sample is individual layer, if not, then will not belong to the unnecessary fuzzy sperm deletion of individual layer sample, if the sperm that not fogging Chu is not identified is augmented, recomputate sperm count, get sperm sound attitude recognition result to the end.

Recognition technology principle of the present invention:

The present invention has adopted the method for fluorescent dye sperm, makes it have only the sperm in the seminal fluid to be colored, and other all non-sperm compositions are refused to dye, and the method for a broad so just is provided to recognition technology.Because seminal fluid is through after dyeing, except two kinds of colors that sperm presented (dead smart for the red and sperm of living for green), also exist the color of other many compositions, so be difficult to accurately distinguish the sperm composition, so the present invention combines existing " nerve net recognition technology ", learn one by one for Protean picture, and further grouping memory, can select suitable group down to carry out according to the situation of picture when operation detects, this have just guaranteed the precision of testing result.

Characteristics of the present invention:

1. dyeing back Necrospermia takes on a red color, and the sperm of living is green, thus utilize the present invention the life or death sperm strictly can be distinguished, and do not influence the activity characteristic of sperm.

2. according to above-mentioned characteristics, the present invention can strictly distinguish the motility rate and the survival rate of sperm.

3. according to characteristics 1, so the detection of sperm concentration reaches the accuracy more than 95%.

Because the present invention can distinguish the Necrospermia strictness, thus no longer be that prior art is divided into level Four in the vigor classification of sperm, but Pyatyi.

4. the nerve net recognition technology is much better than the discriminance analysis system of its gray scale in the accuracy in detection to aspermia or oligospermia (having only 1-2 sperm in each visual field), because have only chromospermism, so utilize the nerve net recognition technology can detect sperm exactly.

Since neural source recognition technology have very high intelligent, so that it can be described the initial trace figure of sperm motility and trace thereof is very clear.

Description of drawings

Fig. 1 is a method flow block diagram of the present invention.

Fig. 2 is a Fuzzy BP nerve net synoptic diagram.

Fig. 3 is used for the automatic analyzing detecting method FB(flow block) of sperm for the recognition result that adopts the inventive method.

The automatic analyzing detecting method of a kind of nerve net identification fluorescent dye sperm that embodiment the present invention proposes reaches embodiment in conjunction with the accompanying drawings and is described in detail as follows:

Method flow of the present invention may further comprise the steps as shown in Figure 1:

The method of a kind of nerve net identification fluorescent dye sperm that the present invention proposes may further comprise the steps:

1) individual layer seminal fluid sample to be measured is carried out beginning to detect after the fluorescent dye;

2) become the fluorescence original image signal by microimaging instrument production fluorescent dye sample, and be presented on the video screen;

3) pattern classification: according to the nerve net recognition technology fluorescence original image of shown various different brightness is learnt all sidedly, and memory classification further, present embodiment is divided into (1,2,3) three quasi-modes; Give and determine that earlier fluorescence raw image sequence signal belongs to a kind of pattern in the described classification mode;

4) input fluorescence raw image sequence is carried out color, brightness, contrast, tone to the fluorescence raw image and is adjusted, and makes it become clearly trichromatic diagram and shows;

5) Quality Control inspection: adjusted three look original graph are removed ground unrest becomes the denoising sound spectrogram, again the denoising sound spectrogram is carried out binary conversion treatment and become binary picture, and the sperm that will be identified is marked; This three looks original graph and the binary picture of sperm being made sign are compared, if the sperm count of the two differs more than 5%, then reselect classification, again carry out 3)-5) step, sperm count until the two differs in 5%, then is as the criterion with this type of other three look original graph binary picture corresponding with it;

6) from the raw image sequence signal of microimaging, choose 5-20 the visual field, each visual field 3-16 frame, obtain three look original graph and binary pictures of selected classification, carry out sperm sound attitude statistical counting (dead/sperm count of living), obtain the sound attitude recognition result of sperm;

7) adjustment of sperm count: check whether testing sample is individual layer, if not, then will not belong to the unnecessary fuzzy sperm deletion of individual layer sample, augmented as the sperm that is not identified because of the not fogging Chu of sperm, recomputate sperm count, get sperm sound attitude recognition result to the end.Said adjustment function is meant that after obtaining recognition result this has just solved density and the indeterminable problem of each technical indicator;

Above-mentioned steps 3) the nerve net recognition technology described in is to adopt green down sperm, the red Necrospermia of living of Fuzzy BP nerve net identification fluorescence, and its cardinal principle is described in detail as follows:

Structure is just the same but input feature vector is expressed the classification independent study training of different Fuzzy BP nerve nets by two with the knowledge of relevant red, marennin composition in the actual fluoroscopic image, note green respectively and redness becomes subchannel with the weights distribution form, thereby reach the purpose of identification, as shown in Figure 2, from left to right be divided among the figure: input layer R, G, B be respectively the image area arbitrfary point (I, red, green, blue tristimulus intensity (gray scale) J) is pressed the input component of green under the fluorescence or red feature representation; The second layer is the obfuscation language description, carries out parameter set { a respectively by triple channel according to Fig. 2 _i, 0≤a _i≤ 255I, a _i≤ a _I+1, i={1,2, ∧, 12}} have determined this layer to reach the structure (each passage is got different parameter sets) of one deck down; The 3rd layer is the fuzzy rule layer, presses

Mode is determined; Output layer is the green or red two-valued variable of differentiating (being realized by two BP nets respectively).

If:

The ground floor input quantity is

Second layer neuron is

The 3rd layer of neuron is The 4th layer of output quantity is O.Get λ, ε, Q (t) are respectively study step-length, critical error and sample, and following algorithm is then arranged:

Input M，N，λ，ε

Input M，N，λ，ε

Initialize?ω _ij ⁽¹⁾，ω _i ⁽²⁾

For t＝1 To t＝N Step+1{

For s＝1 To s＝M Step+1{

δ＝Q(s)-O

$\mathrm{\Δ}{\mathrm{\ω}}_i^{\left(2\right)}=2\·\mathrm{\λ}\·\mathrm{\δ}\·{\mathrm{\ω}}_i^{\left(2\right)}\·\frac{1}{1+e^{-S_i}};{\mathrm{\ω}}_i^{\left(2\right)}={\mathrm{\ω}}_i^{\left(2\right)}+\mathrm{\Δ}{\mathrm{\ω}}_i^{\left(2\right)};$

$\mathrm{\Δ}{\mathrm{\ω}}_{\mathrm{ij}}^{\left(1\right)}=\mathrm{\λ}\·O\·\frac{e^{-S_1}}{{(1+e^{-S_i})}^2};{\mathrm{\ω}}_{\mathrm{ij}}^{\left(1\right)}={\mathrm{\ω}}_{\mathrm{ij}}^{\left(1\right)}+\mathrm{\Δ}{\mathrm{\ω}}_{\mathrm{ij}}^{\left(1\right)};$

E＝E+δ

}

If(E＝E/M)≤ε Then?End?of?Learning

Last color character testing program is by the weights after the study of above-mentioned two BP net and output constructing variable distribution statistics regression model as a result, discerns green and red pixel in the fluoroscopic image simultaneously and puts and realize.

The result of recognition methods of the present invention can be used for as shown in Figure 3, specifically can may further comprise the steps in the automatic analyzing detecting method of sperm:

1) according to the sound attitude recognition result of sperm, relevance to correlation analysis of sequence image three-colo(u)r and judgement sperm motility thereof: comprise setting, binary conversion treatment and the correlation analysis thereof of the dynamic pre-value of sperm motility, related computing or the like, adopt existing gray analysis method to carry out (introduction that its detailed calculated situation is seen the 38 chapter " sperm sound attitude automated image analysis " in " human sperm " book);

2). draw the initial trace figure and trace of sperm motility: system just clearly is plotted in initial trace figure, the trace diagram of sperm on the screen after a series of correlation analyses and related computing, checks (employing prior art) so that manually observe further;

3) calculating of three kinds of speed: according to the movement locus figure of sperm, system just with three kinds of sperm motility (averaged curve movement velocity＜VCL 〉, average point-to-point speed＜VSL, average path movement velocity＜VAP (employing prior art);

4) calculate quiet, the dynamic perfromance statistics of exporting sperm: wherein, the static result of calculating is identical with prior art; Dynamic perfromance result also can count except that all types that has comprised prior art: Necrospermia number accurately.

Because the present invention can distinguish the Necrospermia strictness, thus no longer be that prior art is divided into level Four in the vigor classification of sperm, but Pyatyi is A, and B, C, D, E (E is pure to be Necrospermia) is as shown in the table:

(μ m/s) presses VAP in the quick V of A level 〉=26

The dull at a slow speed 16≤V forward of B level≤26 (μ m/s) press VAP

C level 6≤V＜16 (μ m/s) is at a slow speed pressed VAP

(μ m/s) presses VAP in the motionless 0≤V of D level＜6

Dead smart 0 (the μ m/s) of E level presses VAP

The not only simple digital arrangement of meaning that divides Pyatyi, the more important thing is that it gets real life or death sperm very accurate, the sperm of various motion states is got very clear, and real Necrospermia proportion is also got very clear, be convenient to so very much clinical doctors and diagnose exactly, suit the remedy to the case.This has broken the fuzzy classification method of gray scale recognition system fully, and this itself is exactly an invention.

Claims (1)

1, a kind of method of nerve net identification fluorescent dye sperm may further comprise the steps:

1). seminal fluid sample to be measured is carried out fluorescent dye;

2). become original image signal by microimaging instrument production fluorescent dye sample, and be presented on the video screen;

3). according to the nerve net recognition technology fluoroscopic image of shown various different brightness is learnt all sidedly, and pattern classification is carried out in memory further; Give and determine that earlier fluorescence raw image sequence signal belongs to a kind of pattern in the described classification mode;

4). the fluorescence raw image is carried out color, brightness, contrast, tone adjust, make it become clearly trichromatic diagram and show;

5). adjusted three look original graph are removed ground unrest becomes the denoising sound spectrogram, again the denoising sound spectrogram is carried out binary conversion treatment and become binary picture, and the sperm that will be identified is marked; This three looks original graph and the binary picture of sperm being made sign are compared, if the sperm count of the two differs more than 5%, then reselect classification, again carry out 3)-5) step, sperm count until the two differs in 5%, then is as the criterion with this type of other three look original graph binary picture corresponding with it;

6). choose 5-20 the visual field from the raw image sequence signal of microimaging, each visual field 3-16 frame obtains three look original graph and binary pictures of selected classification, carries out sperm sound attitude statistical counting;

7). the adjustment of sperm count: check whether testing sample is individual layer, if not, then will not belong to the unnecessary fuzzy sperm deletion of individual layer sample, if the sperm that not fogging Chu is not identified is augmented, recomputate sperm count, get sperm sound attitude recognition result to the end.