CN110458806A - A kind of analysis method and system of eye picture and attribute information - Google Patents
A kind of analysis method and system of eye picture and attribute information Download PDFInfo
- Publication number
- CN110458806A CN110458806A CN201910605621.5A CN201910605621A CN110458806A CN 110458806 A CN110458806 A CN 110458806A CN 201910605621 A CN201910605621 A CN 201910605621A CN 110458806 A CN110458806 A CN 110458806A
- Authority
- CN
- China
- Prior art keywords
- user
- picture
- eye
- attribute information
- eye picture
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20076—Probabilistic image processing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30041—Eye; Retina; Ophthalmic
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Radiology & Medical Imaging (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Medical Informatics (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Quality & Reliability (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to the analysis methods and system of a kind of eye picture and attribute information, obtain the attribute information of eye picture and the user of the user with keratitis；According to the inflammatory infiltration and/or ulcer shape of the eye picture, the classification information of the eye picture is differentiated；According to the attribute information and the classification information, the probability that the keratitis that the user is suffered from belongs to a different category is obtained.Oculist of the present invention without profession, eye picture and attribute information need to only be uploaded, it can know the probability of the suffered from keratitis type of user, dicision of diagnosis and treatment is made for user or doctor, and reliable reference data is provided, and it can be generalized to basic hospital, more base crowds are popularized, the defect that existing medical resource is unevenly distributed is made up, the oculist for profession reduces the heavy burdens.
Description
Technical field
The present invention relates to medical image and medical information processing technology field, more particularly, to a kind of eye picture and
The analysis method and system of attribute information.
Background technique
Keratitis is global common diseases causing blindness, and corneal blindness occupy the blinding eye disease second in China.Keratitis
Early diagnosis, early treatment can avoid causing the serious consequence such as perforation of cornea, entophthamia.Many kinds of, the clinical manifestation of keratitis
Complexity, but typical clinical manifestations and sign are shown in keratitis early stage.Many experienced doctors can be according to keratitis
These typical clinical manifestations and inquiry related history data correct diagnosis is made to patient.Clinically common inspection at present
Checking method has slit lamp examination, cornea to strike off microscope, fungal culture, tissue biopsy, PCR, Laser Scanning Confocal Microscope etc..But due to
China's medical resource distributed pole is uneven, and many remote districts and township hospital oculist are deficient, or do not buy expensive eye
Section's instrument.Therefore correct keratitis diagnosis can not be obtained in time and timely treat by resulting in patient.In addition to this, own
The inspection result of instrument must be interpreted, be diagnosed by medical practitioner, could be patient's progress further treatment, these because
Element ultimately increases the risk that more conditions of patients aggravate.The method of one ideal keratitis diagnosis in addition to should have it is sensitive,
Accuracy is high outer, should also have quick, inexpensive advantage.However existing diagnostic method be not obviously able to satisfy still it is above just
Prompt, low cost requirement.
Summary of the invention
The present invention is directed to overcome at least one defect (deficiency) of the above-mentioned prior art, a kind of eye picture and attribute are provided
The analysis method and system of information need to only upload eye picture and attribute information, it can know without the oculist of profession
The probability of the suffered from keratitis type of user makes dicision of diagnosis and treatment and provides reliable reference data for user or doctor.
The technical solution adopted by the present invention is that:
A kind of analysis method of eye picture and attribute information, comprising the following steps:
Obtain the attribute information of eye picture and the user of the user with keratitis；
According to the inflammatory infiltration and/or ulcer shape of the eye picture, the classification information of the eye picture is differentiated；
According to the attribute information and the classification information, obtain the keratitis that the user is suffered from belong to a different category it is general
Rate.
First the eye picture to user with keratitis carries out preliminary classification and obtains classification information, then classification information is combined
The attribute information of user carries out comprehensive analysis together, obtains the keratitis that user is suffered from and belongs to different classes of probability.Pass through
The attribute information of eye appearance picture and the user of the comprehensive analysis user with keratitis, can more comprehensively, more accurately
The class probability for analyzing the suffered from keratitis of user, makes dicision of diagnosis and treatment and provides reliable reference data for user or doctor.This
Embodiment is simple and easy, and perhaps training operating technology personnel can allow user or amateur eye to the oculist without profession
The basic hospital staff of section doctor knows the probability of the type of the suffered from keratitis of user, makes up existing medical resource distribution not
Equal defect.
Further, according to the inflammatory infiltration and/or ulcer shape of the eye picture, differentiate point of the eye picture
Category information, specifically:
According to the inflammatory infiltration and/or ulcer shape of the eye picture, the eye picture is differentiated by deep learning algorithm
Classification information.
Further, according to the attribute information and the classification information, the keratitis category that the user is suffered from is obtained
In different classes of probability, specifically:
According to the attribute information and the classification information, the use is obtained by NB Algorithm or random forests algorithm
The probability that the keratitis that family is suffered from belongs to a different category.
Further, the attribute information includes trauma history, wound medium, eye treatment history, eyelid feature, the past medication
Wherein one or more of history, other site disorders in addition to eye.
Further, the classification information include whether for similar round paste lesion, whether be satellite stove, whether formed and exempted from
Epidemic disease ring, whether formed pseudopodium, whether in dendroid, whether in map shape, it is whether rounded, whether rough surface, whether surface
Completely, whether edge blurry, whether the clean and tidy one or more of them in edge.
A kind of analysis system of eye picture and attribute information, comprising:
Data acquisition module suffers from the eye picture of the user of keratitis and the attribute information of the user for obtaining；
Picture preprocessing module differentiates the eye figure for the inflammatory infiltration and/or ulcer shape according to the eye picture
The classification information of piece；
Picture classification module, for obtaining the cornea that the user is suffered from according to the attribute information and the classification information
The probability that inflammation belongs to a different category.
Eye picture progress preliminary classification of the picture preprocessing module first to user with keratitis obtains classification information, schemes
The attribute information of classification information combination user is carried out comprehensive analysis again by piece categorization module together, obtains the keratitis that user is suffered from
Belong to different classes of probability.The attribute of eye appearance picture and the user by comprehensive analysis user with keratitis
Information can more comprehensively, more accurately analyze the class probability of the suffered from keratitis of user, make diagnosis and treatment for user or doctor and determine
Plan provides reliable reference data.The present embodiment is simple and easy, without professional oculist or trains operating technology personnel,
The basic hospital staff of user or amateur oculist can be allowed to know the probability of the type of the suffered from keratitis of user,
Make up the defect that existing medical resource is unevenly distributed.
Further, the picture preprocessing module, specifically for according to the inflammatory infiltration of the eye picture and/or bursting
Ulcer shape differentiates the classification information of the eye picture by deep learning algorithm.
Further, the picture classification module is specifically used for being passed through according to the attribute information and the classification information
NB Algorithm or random forests algorithm obtain the probability that the keratitis that the user is suffered from belongs to a different category.
Further, the system also includes:
Picture is presorted module, for obtaining the eye picture of user, is presorted out by deep learning algorithm with keratitis
User eye picture.
Further, the system also includes:
Eye locating module is positioned by deep learning algorithm and is cut out in the picture for obtaining the picture of user
The eye picture of eye formation user.
Compared with prior art, the invention has the benefit that
(1) user or doctor can be corresponding with the eye picture and input that each Terminal Type uploads under online or off-line state
People's data forms attribute information, it can obtains the probability for the affiliated type of keratitis that user is suffered from, is that user or doctor make
Dicision of diagnosis and treatment provides reliable reference data out；
(2) operation of the present invention is simple, without the oculist or operator of profession, can be generalized to basic hospital, popularizes more
More base crowds makes up the defect that existing medical resource is unevenly distributed, and the oculist for profession reduces the heavy burdens；
(3) present invention is improved by machine learning algorithm such as deep learning algorithm, NB Algorithm, random forests algorithm
The Accuracy and high efficiency of comprehensive analysis is carried out to eye picture and attribute information.
Detailed description of the invention
Fig. 1 is the method flow diagram of the embodiment of the present invention 1.
Fig. 2 is another method flow diagram of the embodiment of the present invention 1.
Fig. 3 is that the first systematic of the embodiment of the present invention 2 forms figure.
Fig. 4 is that second system of the embodiment of the present invention 2 forms figure.
Fig. 5 is that the third system of the embodiment of the present invention 2 forms figure.
Specific embodiment
Attached drawing of the present invention only for illustration, is not considered as limiting the invention.It is following in order to more preferably illustrate
Embodiment, the certain components of attached drawing have omission, zoom in or out, and do not represent the size of actual product；For art technology
For personnel, the omitting of some known structures and their instructions in the attached drawings are understandable.
Embodiment 1
As shown in Figure 1, the present embodiment provides the analysis methods of a kind of eye picture and attribute information, comprising the following steps:
Obtain the attribute information of eye picture and the user of the user with keratitis；
According to the inflammatory infiltration and/or ulcer shape of the eye picture, the classification information of the eye picture is differentiated；
According to the attribute information and the classification information, obtain the keratitis that the user is suffered from belong to a different category it is general
Rate.
First the eye picture to user with keratitis carries out preliminary classification and obtains classification information, then classification information is combined
The attribute information of user carries out comprehensive analysis together, obtains the keratitis that user is suffered from and belongs to different classes of probability.Pass through
The attribute information of eye appearance picture and the user of the comprehensive analysis user with keratitis, can more comprehensively, more accurately
The class probability for analyzing the suffered from keratitis of user, makes dicision of diagnosis and treatment and provides reliable reference data for user or doctor.This
Embodiment is simple and easy, and perhaps training operating technology personnel can allow user or amateur eye to the oculist without profession
The basic hospital staff of section doctor knows the probability of the type of the suffered from keratitis of user, makes up existing medical resource distribution not
Equal defect.
In the present embodiment, the attribute information includes trauma history, eye treatment history, eyelid feature, the past medication history, both
Wherein one or more of other site disorders toward eye medical history, in addition to eye.
Specifically, the attribute information include the following:
Trauma history includes: a. scratch；B. through and through wound c. is without trauma history d. chemical burn.
The wound medium of trauma history includes: a. plant；B. other.
Eye treatment history includes: a. to wear contact lens history；B. cornea refractive surgery history；C. corneal graft；D. intraocular
History of operation.
Eyelid feature includes: a. blear-eye；B. trichiasis；C. entropion；D. ectropion.
The past medication history includes: a. using steroids history；B. other.
The past eye disease medical history includes: a. lacrimal passage disease；B. keratitis；, c. dry eyes.
The course of disease of the past eye disease medical history: a. is acute≤and 2 days；B. 2 days chronic >.
Other site disorders in addition to eye: a. diabetes；B. immune deficiency.
In specific implementation process, doctor judges that user or user judge whether oneself has the above attribute information
Wherein one or more, to form the attribute information of the user.For example, the user is the trauma history for having through and through wound, and wound
Medium is plant, then the attribute information of the user is plant to have through and through wound history and wound medium.
In the present embodiment, the classification information include whether for similar round paste lesion, whether be satellite stove, whether shape
At immune ring, whether formed pseudopodium, whether in dendroid, whether in map shape, it is whether rounded, whether rough surface, whether
Surface is clean, whether edge blurry, whether the clean and tidy one or more of them in edge.
In specific implementation process, according to the inflammatory infiltration and/or ulcer shape of the eye picture, judge it in these points
In category information whether being, to form classification information.For example, the inflammatory infiltration and/or ulcer shape of the eye picture are
Similar round paste lesion, form pseudopodium, be unclean in dendroid, rough surface, surface, be that edge is fuzzy, edge is untidy, then divide
Category information are as follows: be similar round paste lesion, be not satellite stove, be not to be formed immune ring, be to form pseudopodium, be in dendroid, no
Be in map shape, be not it is rounded, be rough surface, be not surface it is clean, be edge blurry, be not that edge is clean and tidy.
Specifically, the classification of the suffered from keratitis of user may include fungoid, it is bacillary, viral.
In the present embodiment, according to the inflammatory infiltration and/or ulcer shape of the eye picture, differentiate the eye picture
Classification information, specifically:
According to the inflammatory infiltration and/or ulcer shape of the eye picture, the eye picture is differentiated by deep learning algorithm
Classification information.
In specific implementation process, the eye picture is inputted in trained first deep learning model, is obtained described
The classification information of eye picture.First deep learning model is the eye by normal cornea picture and all kinds of infected keratitis
Picture is trained built-up, and all kinds of infected keratitis can be gone out with preliminary classification and obtain classification information.First deep learning
Model can specifically use ResNet model.
In the present embodiment, according to the attribute information and the classification information, the cornea that the user is suffered from is obtained
The probability that inflammation belongs to a different category, specifically:
According to the attribute information and the classification information, the use is obtained by NB Algorithm or random forests algorithm
The probability that the keratitis that family is suffered from belongs to a different category.
The probability that the keratitis that the user is suffered from belongs to a different category is obtained by NB Algorithm, specifically
Are as follows: assuming that the attribute information of the user and classification information are set x={ a1, a2... ..., am, the classification of keratitis is y=
{y1, y2... ..., yn, wherein n is the sum of attribute information and classification information, and m is the sum of keratitis classification.Set of computations x
In the corresponding conditional probability P (y of each elementj| x), wherein i=1,2 ... ..., m；J=1,2 ... ..., n.If P (yk| x)=
max{P(y1| x), P (y2| x) ... ..., P (ym| x) }, then the keratitis classification that the user is suffered from belongs to yk。
Conditional probability P (yj| it can x) obtain by the following method:
Categorized good attribute information and classification information are formed into training sample set, calculate each attribute information and each
The conditional probability of classification information: P (a1|y1), P (a2|y1) ... ..., P (an|y1)；P(a1|y2), P (a2|y2) ... ..., P (an|
y2)；……；P(a1|ym), P (a2|ym) ... ..., P (an|ym)。
When each attribute information and classification information are independent, conditional probability P (y can be obtained according to Bayes' theoremj|
X):
The probability that the keratitis that the user is suffered from belongs to a different category is obtained by random forests algorithm, specifically:
In the random forest that the attribute information of the user and classification information input are built, carried out by the decision tree in random forest
Ballot, the probability to be belonged to a different category according to the keratitis classification that the determining user is suffered from of voting.
Categorized good attribute information and classification information are formed into training sample set, it is assumed that N indicates training sample number,
The sum of M presentation class information and attribute information, the building process of random forest are as follows:
(1) input feature vector number m, for determining the result of decision of a node on decision tree；Wherein m should be much smaller than M.
(2) from N number of training sample in a manner of sampling with replacement, n times is sampled, form a training set, and with not being extracted into
Sample is predicted, its error is assessed.
(3) for each node, m feature is randomly choosed, the decision of each node is all based on these features on decision tree
Determining.According to m feature, its optimal divisional mode is calculated.
(4) each tree all can completely grow up without beta pruning, this is possible to after having built a normal tree classifier to be adopted
With.
(5) it repeats the above steps, constructs other many decision trees, until reaching a group decision tree of predetermined number, i.e.,
Random forest is built.
Preferably, after obtaining the probability that the keratitis that the user is suffered from belongs to a different category, the user is built
When view carries out Laser Scanning Confocal Microscope inspection and cultivation of cervical secretions further to make a definite diagnosis, the first deep learning mould is added in confirmed result
The training sample of type, NB Algorithm or random forests algorithm is concentrated, so that having certainly with method provided by embodiment
The ability that adaptive learning updates.
As shown in Fig. 2, in the present embodiment, in the attribute for obtaining eye picture and the user of the user with keratitis
It is further comprising the steps of before information:
The eye picture for obtaining user is presorted out by deep learning algorithm and suffers from the eye picture of the user of keratitis.
In specific implementation process, the eye picture of user is inputted in trained second deep learning model, therefrom in advance
Sort out the eye picture of the user with keratitis.Second deep learning model is by known cornea scraping blade cultivation results
Eye picture is trained built-up, can identify whether eye picture is the eye picture with keratitis.Second depth
Study module can specifically use CNN model.
In the present embodiment, it in the eye picture for obtaining user, is presorted out by deep learning algorithm with keratitis
User eye picture before, it is further comprising the steps of:
The picture for obtaining user positions by deep learning algorithm and cuts out the eye that the eye in the picture forms user
Picture.
Before the eye picture to user is presorted, eye positioning first is carried out to the picture of user, is cut according to positioning
Eye picture is taken, subsequent presort to eye picture and and the other accuracy of anticipation can be improved.
In specific implementation process, the picture of user is inputted in trained third deep learning model, to orient
Eye.Third deep learning model be by marked the eye picture of eye locations be trained it is built-up.Third
Deep learning model can specifically use Faster-RCNN model.
In specific implementation process, user or base doctor can upload eye under online or off-line state with each Terminal Type
Picture and attribute information, so that eye picture and attribute information are acquired.The terminal includes smart phone, tablet computer, pen
Remember the various kinds of equipment such as sheet, desktop computer, screening equipment, wherein screening equipment is to refer to directly shoot eye appearance picture and energy
The equipment for inputting customer attribute information.
Embodiment 2
As shown in figure 3, the present embodiment provides the analysis systems of a kind of eye picture and attribute information, comprising:
Data acquisition module 10 suffers from the eye picture of the user of keratitis and the attribute information of the user for obtaining；
Picture preprocessing module 20 differentiates the eye for the inflammatory infiltration and/or ulcer shape according to the eye picture
The classification information of picture；
Picture classification module 30, for obtaining the angle that the user is suffered from according to the attribute information and the classification information
The probability that film inflammation belongs to a different category.
Eye picture progress preliminary classification of the picture preprocessing module 20 first to user with keratitis obtains classification information,
The attribute information of classification information combination user is carried out comprehensive analysis again by picture classification module 30 together, obtains the angle that user is suffered from
Film inflammation belongs to different classes of probability.Eye appearance picture and the user by comprehensive analysis user with keratitis
Attribute information can more comprehensively, more accurately analyze the class probability of the suffered from keratitis of user, pay a home visit for user or doctor
It treats decision and reliable reference data is provided.The present embodiment is simple and easy, oculist or training operating technology without profession
Personnel can allow the basic hospital staff of user or amateur oculist to know the type of the suffered from keratitis of user
Probability makes up the defect that existing medical resource is unevenly distributed.
In the present embodiment, the attribute information includes trauma history, eye treatment history, eyelid feature, the past medication history, both
Wherein one or more of other site disorders toward eye medical history, in addition to eye.
Specifically, the attribute information include the following:
Trauma history includes: a. scratch；B. through and through wound c. is without trauma history d. chemical burn.
The wound medium of trauma history includes: a. plant；B. other.
Eye treatment history includes: a. to wear contact lens history；B. cornea refractive surgery history；C. corneal graft；D. intraocular
History of operation.
Eyelid feature includes: a. blear-eye；B. trichiasis；C. entropion；D. ectropion.
The past medication history includes: a. using steroids history；B. other.
The past eye disease medical history includes: a. lacrimal passage disease；B. keratitis；, c. dry eyes.
The course of disease of the past eye disease medical history: a. is acute≤and 2 days；B. 2 days chronic >.
Other site disorders in addition to eye: a. diabetes；B. immune deficiency.
In specific implementation process, doctor judges that user or user judge whether oneself has the above attribute information
Wherein one or more, to form the attribute information of the user.For example, the user is the trauma history for having through and through wound, and wound
Medium is plant, then the attribute information of the user is plant to have through and through wound history and wound medium.
In the present embodiment, the classification information include whether for similar round paste lesion, whether be satellite stove, whether shape
At immune ring, whether formed pseudopodium, whether in dendroid, whether in map shape, it is whether rounded, whether rough surface, whether
Surface is clean, whether edge blurry, whether the clean and tidy one or more of them in edge.
In specific implementation process, picture preprocessing module 20 is according to the inflammatory infiltration and/or ulcer shape of the eye picture
Shape, judge its in these classification informations whether being, to form classification information.For example, the inflammation of the eye picture is soaked
Profit and/or ulcer shape are similar round paste lesions, form pseudopodium, is unclean in dendroid, rough surface, surface, are edge moulds
Paste, edge it is untidy, then picture preprocessing module 20 judges its classification information are as follows: be similar round paste lesion, be not satellite stove,
Be not to be formed immune ring, be to be formed pseudopodium, be in dendroid, be not in map shape, be not rounded, be rough surface, be not
Surface is clean, be edge blurry, be not that edge is clean and tidy.
Specifically, the classification of the suffered from keratitis of user may include fungoid, it is bacillary, viral.
In the present embodiment, picture preprocessing module 20, specifically for according to the inflammatory infiltration of the eye picture and/or
Ulcer shape differentiates the classification information of the eye picture by deep learning algorithm.
In specific implementation process, the eye picture is inputted trained first deep learning by picture preprocessing module 20
In model, the classification information of the eye picture is obtained.First deep learning model is by normal cornea picture and all kinds of senses
The eye picture of metachromia keratitis be trained it is built-up, can be gone out with preliminary classification all kinds of infected keratitis obtain classification letter
Breath.First deep learning model can specifically use ResNet model.
In the present embodiment, picture classification module 30 is specifically used for being led to according to the attribute information and the classification information
It crosses NB Algorithm or random forests algorithm obtains the probability that the keratitis that the user is suffered from belongs to a different category.
Picture classification module 30 obtains the keratitis that the user is suffered from by NB Algorithm and belongs to inhomogeneity
Other probability, specifically: assuming that the attribute information of the user and classification information are set x={ a1, a2... ..., am, keratitis
Classification be y={ y1, y2... ..., yn, wherein n is the sum of attribute information and classification information, and m is the total of keratitis classification
Number.The corresponding conditional probability P (y of each element in set of computations xj| x), wherein i=1,2 ... ..., m；J=1,2 ... ..., n.
If P (yk| x)=max { P (y1| x), P (y2| x) ... ..., P (ym| x) }, then the keratitis classification that the user is suffered from belongs to yk。
Conditional probability P (yj| it can x) obtain by the following method:
Categorized good attribute information and classification information are formed into training sample set, calculate each attribute information and each
The conditional probability of classification information: P (a1|y1), P (a2|y1) ... ..., P (an|y1)；P(a1|y2), P (a2|y2) ... ..., P (an|
y2)；……；P(a1|ym), P (a2|ym) ... ..., P (an|ym)。
When each attribute information and classification information are independent, conditional probability P (y can be obtained according to Bayes' theoremj|
X):
Picture classification module 30 obtains the keratitis that the user is suffered from by random forests algorithm and belongs to a different category
Probability, specifically: by the attribute information of the user and the random forest that builds of classification information input, pass through random forest
In decision tree vote, determine probability that the keratitis classification suffered from of the user belongs to a different category according to voting.
Categorized good attribute information and classification information are formed into training sample set, it is assumed that N indicates training sample number,
The sum of M presentation class information and attribute information, the building process of random forest are as follows:
(1) input feature vector number m, for determining the result of decision of a node on decision tree；Wherein m should be much smaller than M.
(2) from N number of training sample in a manner of sampling with replacement, n times is sampled, form a training set, and with not being extracted into
Sample is predicted, its error is assessed.
(3) for each node, m feature is randomly choosed, the decision of each node is all based on these features on decision tree
Determining.According to m feature, its optimal divisional mode is calculated.
(4) each tree all can completely grow up without beta pruning, this is possible to after having built a normal tree classifier to be adopted
With.
(5) it repeats the above steps, constructs other many decision trees, until reaching a group decision tree of predetermined number, i.e.,
Random forest is built.
As shown in Figure 4, it is preferable that the system also includes model modification modules 60；
After obtaining the probability that the keratitis that the user is suffered from belongs to a different category, the user, which is proposed, carries out copolymerization coke
When microexamination and cultivation of cervical secretions are further to make a definite diagnosis, the first deep learning is added in confirmed result by model modification module 60
Model, NB Algorithm/random forests algorithm training sample are concentrated, so that having certainly with method provided by embodiment
The ability that adaptive learning updates.
As shown in figure 5, in the present embodiment, the system also includes:
Picture is presorted module 40, for obtaining the eye picture of user, is presorted out by deep learning algorithm with cornea
The eye picture of scorching user.
In specific implementation process, the eye picture of user is inputted trained second depth by picture module 40 of presorting
It practises in model, the eye picture for the user with keratitis that therefrom presorts out.Second deep learning model is by known angle
The eye picture of film scraping blade cultivation results is trained built-up, can identify whether eye picture is the eye with keratitis
Portion's picture.Second deep learning module can specifically use CNN model.
In the present embodiment, the system also includes:
Eye locating module 50 is positioned by deep learning algorithm and is cut out in the picture for obtaining the picture of user
Eye formed user eye picture.
It presorts before module 40 presorts to the eye picture of user in picture, eye locating module 50 is first to user
Picture carry out eye positioning, according to positioning intercept eye picture, can be improved subsequent pictures presort module 40 and picture it is pre-
The accuracy that processing module 20 analyzes eye picture.
In specific implementation process, the picture of user is inputted trained third deep learning model by eye locating module 50
In, to orient eye.Third deep learning model is trained by having marked the eye picture of eye locations
It is built-up.Third deep learning model can specifically use Faster-RCNN model.
In specific implementation process, user or base doctor can upload eye under online or off-line state with each Terminal Type
Picture and attribute information, so that data acquisition module 10 gets eye picture and attribute information.The terminal includes intelligent hand
The various kinds of equipment such as machine, tablet computer, notebook, desktop computer, screening equipment, wherein screening equipment is to refer to directly shoot eye
Equipment portion appearance picture and customer attribute information can be inputted.Module can be set in Cloud Server some or all of in system
In.
Obviously, the above embodiment of the present invention is only intended to clearly illustrate technical solution of the present invention example, and
It is not the restriction to a specific embodiment of the invention.It is all made within the spirit and principle of claims of the present invention
Any modifications, equivalent replacements, and improvements etc., should all be included in the scope of protection of the claims of the present invention.
Claims (10)
1. a kind of analysis method of eye picture and attribute information, which comprises the following steps:
Obtain the attribute information of eye picture and the user of the user with keratitis；
According to the inflammatory infiltration and/or ulcer shape of the eye picture, the classification information of the eye picture is differentiated；
According to the attribute information and the classification information, obtain the keratitis that the user is suffered from belong to a different category it is general
Rate.
2. the analysis method of a kind of eye picture and attribute information according to claim 1, which is characterized in that according to described
The inflammatory infiltration and/or ulcer shape of eye picture, differentiate the classification information of the eye picture, specifically:
According to the inflammatory infiltration and/or ulcer shape of the eye picture, the eye picture is differentiated by deep learning algorithm
Classification information.
3. the analysis method of a kind of eye picture and attribute information according to claim 1, which is characterized in that according to described
Attribute information and the classification information obtain the probability that the keratitis that the user is suffered from belongs to a different category, specifically:
According to the attribute information and the classification information, the use is obtained by NB Algorithm or random forests algorithm
The probability that the keratitis that family is suffered from belongs to a different category.
4. the analysis method of a kind of eye picture and attribute information according to claim 1, which is characterized in that the attribute
Information includes trauma history, wound medium, eye treatment history, eyelid feature, the past medication history, other position diseases in addition to eye
Wherein one or more of disease.
5. the analysis method of a kind of eye picture and attribute information according to claim 1, which is characterized in that the classification
Information include whether for similar round paste lesion, whether be satellite stove, whether formed immune ring, whether formed pseudopodium, whether be in
Dendroid, whether in map shape, it is whether rounded, whether rough surface, whether surface it is clean, whether edge blurry, whether side
The clean and tidy one or more of them of edge.
6. a kind of analysis system of eye picture and attribute information characterized by comprising
Data acquisition module suffers from the eye picture of the user of keratitis and the attribute information of the user for obtaining；
Picture preprocessing module differentiates the eye figure for the inflammatory infiltration and/or ulcer shape according to the eye picture
The classification information of piece；
Picture classification module, for obtaining the cornea that the user is suffered from according to the attribute information and the classification information
The probability that inflammation belongs to a different category.
7. the analysis system of a kind of eye picture and attribute information according to claim 6, which is characterized in that the picture
Preprocessing module is sentenced specifically for the inflammatory infiltration and/or ulcer shape according to the eye picture by deep learning algorithm
The classification information of the not described eye picture.
8. the analysis system of a kind of eye picture and attribute information according to claim 6, which is characterized in that the picture
Categorization module is specifically used for passing through NB Algorithm or random forest according to the attribute information and the classification information
Algorithm obtains the probability that the keratitis that the user is suffered from belongs to a different category.
9. the analysis system of a kind of eye picture and attribute information according to claim 6, which is characterized in that further include:
Picture is presorted module, for obtaining the eye picture of user, is presorted out by deep learning algorithm with keratitis
User eye picture.
10. the analysis system of a kind of eye picture and attribute information according to claim 9, which is characterized in that further include:
Eye locating module is positioned by deep learning algorithm and is cut out in the picture for obtaining the picture of user
The eye picture of eye formation user.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910605621.5A CN110458806A (en) | 2019-07-05 | 2019-07-05 | A kind of analysis method and system of eye picture and attribute information |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910605621.5A CN110458806A (en) | 2019-07-05 | 2019-07-05 | A kind of analysis method and system of eye picture and attribute information |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110458806A true CN110458806A (en) | 2019-11-15 |
Family
ID=68482307
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910605621.5A Pending CN110458806A (en) | 2019-07-05 | 2019-07-05 | A kind of analysis method and system of eye picture and attribute information |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110458806A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111652841A (en) * | 2020-04-24 | 2020-09-11 | 温州医科大学附属眼视光医院 | Corneal injury detection method based on image processing |
| WO2022142368A1 (en) * | 2020-12-29 | 2022-07-07 | Aimomics (Shanghai) Intelligent Technology Co., Ltd | Rapid screen system based on eye region image |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100204584A1 (en) * | 2009-02-12 | 2010-08-12 | Alcon Research, Ltd. | Method and apparatus for ocular surface imaging |
| CN109344808A (en) * | 2018-07-24 | 2019-02-15 | 中山大学中山眼科中心 | A kind of eyes image processing system based on deep learning |
| CN109636796A (en) * | 2018-12-19 | 2019-04-16 | 中山大学中山眼科中心 | A kind of artificial intelligence eye picture analyzing method, server and system |
-
2019
- 2019-07-05 CN CN201910605621.5A patent/CN110458806A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100204584A1 (en) * | 2009-02-12 | 2010-08-12 | Alcon Research, Ltd. | Method and apparatus for ocular surface imaging |
| CN109344808A (en) * | 2018-07-24 | 2019-02-15 | 中山大学中山眼科中心 | A kind of eyes image processing system based on deep learning |
| CN109636796A (en) * | 2018-12-19 | 2019-04-16 | 中山大学中山眼科中心 | A kind of artificial intelligence eye picture analyzing method, server and system |
Non-Patent Citations (2)
| Title |
|---|
| 杜婷: "基于机器学习的角膜炎图像辅助诊断研究与实现", 《中国优秀博硕士学位论文全文数据库（硕士）信息科技辑》》 * |
| 甘天圣: "基于卷积神经网络的细粒度角膜炎图像分类研究", 《中国优秀博硕士学位论文全文数据库（硕士）信息科技辑》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111652841A (en) * | 2020-04-24 | 2020-09-11 | 温州医科大学附属眼视光医院 | Corneal injury detection method based on image processing |
| WO2022142368A1 (en) * | 2020-12-29 | 2022-07-07 | Aimomics (Shanghai) Intelligent Technology Co., Ltd | Rapid screen system based on eye region image |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhang et al. | Automatic cataract grading methods based on deep learning | |
| Jawaheer et al. | Endonasal versus external dacryocystorhinostomy for nasolacrimal duct obstruction | |
| Huang et al. | Glaucoma detection using adaptive neuro-fuzzy inference system | |
| CN105513077B (en) | A kind of system for diabetic retinopathy screening | |
| Huang et al. | Development and comparison of automated classifiers for glaucoma diagnosis using Stratus optical coherence tomography | |
| JP6594033B2 (en) | Image processing apparatus, image processing method, and program | |
| Sim et al. | Automated retinal image analysis for diabetic retinopathy in telemedicine | |
| Seoud et al. | Automatic grading of diabetic retinopathy on a public database | |
| Treder et al. | Using deep learning in automated detection of graft detachment in Descemet membrane endothelial keratoplasty: a pilot study | |
| Lin et al. | A review of machine learning techniques for keratoconus detection and refractive surgery screening | |
| CN109858540B (en) | Medical image recognition system and method based on multi-mode fusion | |
| Turuwhenua et al. | Fully automated montaging of laser scanning in vivo confocal microscopy images of the human corneal subbasal nerve plexus | |
| Kauppi | Eye fundus image analysis for automatic detection of diabetic retinopathy | |
| Perdomo et al. | Convolutional network to detect exudates in eye fundus images of diabetic subjects | |
| CN110458806A (en) | A kind of analysis method and system of eye picture and attribute information | |
| Balasubramanian et al. | Automatic classification of focal lesions in ultrasound liver images using principal component analysis and neural networks | |
| Ali et al. | Glaucoma detection based on local binary patterns in fundus photographs | |
| Songire et al. | Automated detection of cholesterol presence using iris recognition algorithm | |
| Tahvildari et al. | Application of artificial intelligence in the diagnosis and management of corneal diseases | |
| Camara et al. | A comprehensive review of methods and equipment for aiding automatic glaucoma tracking | |
| CN112863699B (en) | ESD preoperative discussion system based on mobile terminal | |
| Lever et al. | The Paediatric glaucoma diagnostic ability of optical coherence tomography: a comparison of macular segmentation and peripapillary retinal nerve fibre layer thickness | |
| Wang et al. | An intelligent optical coherence tomography-based system for pathological retinal cases identification and urgent referrals | |
| CN113539476A (en) | Stomach endoscopic biopsy Raman image auxiliary diagnosis method and system based on artificial intelligence | |
| CN108573267A (en) | A kind of method and device of liver organization textural classification |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191115 |