CN109671049A - A kind of medical image processing method, system, equipment, storage medium - Google Patents
A kind of medical image processing method, system, equipment, storage medium Download PDFInfo
- Publication number
- CN109671049A CN109671049A CN201811320629.9A CN201811320629A CN109671049A CN 109671049 A CN109671049 A CN 109671049A CN 201811320629 A CN201811320629 A CN 201811320629A CN 109671049 A CN109671049 A CN 109671049A
- Authority
- CN
- China
- Prior art keywords
- lesion
- medical image
- image
- probability
- abnormal
- 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
- 238000003672 processing method Methods 0.000 title claims abstract description 29
- 230000003902 lesion Effects 0.000 claims abstract description 149
- 230000002159 abnormal effect Effects 0.000 claims abstract description 84
- 238000012545 processing Methods 0.000 claims abstract description 10
- 230000011218 segmentation Effects 0.000 claims description 24
- 238000013528 artificial neural network Methods 0.000 claims description 20
- 238000000034 method Methods 0.000 claims description 10
- 238000007635 classification algorithm Methods 0.000 claims description 8
- 238000010801 machine learning Methods 0.000 claims description 8
- 239000000284 extract Substances 0.000 claims description 7
- 239000003814 drug Substances 0.000 claims description 6
- 238000004891 communication Methods 0.000 claims description 3
- 238000004458 analytical method Methods 0.000 abstract description 6
- 230000001575 pathological effect Effects 0.000 abstract description 3
- 230000002062 proliferating effect Effects 0.000 description 25
- 210000000416 exudates and transudate Anatomy 0.000 description 22
- 206010002329 Aneurysm Diseases 0.000 description 20
- 206010012689 Diabetic retinopathy Diseases 0.000 description 16
- 210000001210 retinal vessel Anatomy 0.000 description 14
- 201000010099 disease Diseases 0.000 description 9
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 9
- 238000005070 sampling Methods 0.000 description 8
- 230000000740 bleeding effect Effects 0.000 description 7
- 238000012549 training Methods 0.000 description 7
- 238000013527 convolutional neural network Methods 0.000 description 6
- 206010012601 diabetes mellitus Diseases 0.000 description 6
- 210000001525 retina Anatomy 0.000 description 6
- 201000004569 Blindness Diseases 0.000 description 4
- 208000001344 Macular Edema Diseases 0.000 description 4
- 206010025415 Macular oedema Diseases 0.000 description 4
- 201000010230 macular retinal edema Diseases 0.000 description 4
- 230000002207 retinal effect Effects 0.000 description 4
- 208000009857 Microaneurysm Diseases 0.000 description 3
- 210000001367 artery Anatomy 0.000 description 3
- 210000004204 blood vessel Anatomy 0.000 description 3
- 230000004927 fusion Effects 0.000 description 3
- 210000005036 nerve Anatomy 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 210000003462 vein Anatomy 0.000 description 3
- 206010028980 Neoplasm Diseases 0.000 description 2
- 206010029113 Neovascularisation Diseases 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 239000000835 fiber Substances 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000035699 permeability Effects 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000002792 vascular Effects 0.000 description 2
- 208000002249 Diabetes Complications Diseases 0.000 description 1
- 206010012655 Diabetic complications Diseases 0.000 description 1
- 102000004895 Lipoproteins Human genes 0.000 description 1
- 108090001030 Lipoproteins Proteins 0.000 description 1
- 208000017442 Retinal disease Diseases 0.000 description 1
- 206010057430 Retinal injury Diseases 0.000 description 1
- 206010038923 Retinopathy Diseases 0.000 description 1
- 206010038934 Retinopathy proliferative Diseases 0.000 description 1
- 238000009825 accumulation Methods 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 210000003050 axon Anatomy 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 239000008280 blood Substances 0.000 description 1
- 210000004369 blood Anatomy 0.000 description 1
- 230000017531 blood circulation Effects 0.000 description 1
- 230000000747 cardiac effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000004438 eyesight Effects 0.000 description 1
- 210000000887 face Anatomy 0.000 description 1
- 238000005562 fading Methods 0.000 description 1
- 239000012634 fragment Substances 0.000 description 1
- 230000005802 health problem Effects 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 230000001788 irregular Effects 0.000 description 1
- 230000000302 ischemic effect Effects 0.000 description 1
- 210000003734 kidney Anatomy 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 239000007788 liquid Substances 0.000 description 1
- 210000002189 macula lutea Anatomy 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 239000012528 membrane Substances 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000035755 proliferation Effects 0.000 description 1
- 102000004169 proteins and genes Human genes 0.000 description 1
- 108090000623 proteins and genes Proteins 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 208000024725 retina neoplasm Diseases 0.000 description 1
- 210000003994 retinal ganglion cell Anatomy 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- 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
-
- 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
- G06T2207/30096—Tumor; Lesion
-
- 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
- G06T2207/30101—Blood vessel; Artery; Vein; Vascular
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (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)
- Eye Examination Apparatus (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of medical image processing methods, system, equipment, storage medium, on the one hand the multiple first abnormal probability are obtained by lesion image, on the other hand the multiple second abnormal probability are obtained by medical image and lesion image, further according to the multiple first abnormal probability, multiple second abnormal probability and different weight coefficients obtain the final probability that medical image belongs to different images intensity of anomaly according to different images intensity of anomaly, and using the image abnormity degree of maximum final probability as the image abnormity degree of medical image, the intensity of anomaly of medical image is analyzed in realization, overcome and exists in the prior art to pathological image processing, analysis is by naked eyes, the technical issues of inefficiency.
Description
Technical field
The present invention relates to field of image processing, especially a kind of medical image processing method, system, equipment, storage medium.
Background technique
The rapid growth of diabetes causes the concern of all circles, the finger of disease caused by diabetes during past decades
Number, which increases, has become the huge challenge that current healthcare industry faces.And unfortunately, with disease caused by diabetes
Patient populations' still sustainable growth at an amazing speed.There are also more allowing people to worry, only about 70% patient realizes
This disease is suffered to them.From the point of view of medical angle, diabetes are considered as the basis of many health problems and later period obstacle,
I.e. diabetes can cause a series of lesion and complication, comprising: lead to adverse cardiac, diabetic retinopathy (DR)
With kidney problems etc..DR is one of most common diabetic complication, it is considered as one of the main reason of blindness.Research
Show that the diabetes problem in each area in the world is all at an amazing speed regardless of the size of population or socioeconomic background
Increase.In addition, living condition and treatment facility existing defects of the studies have shown that due to developing country, so nearly 75%
DR patient belong to developing country, and the blindness possibility of DR patient is almost 25 times higher than the people for not suffering from DR.But
Currently lack the medical expert of diabetic retinopathy, most DR patient cannot obtain effective lesion in time in the world
Detection and treatment, most of patient, which only works as retinopathy and has evolved to treatment, to be become highly complex and hardly may be used sometimes
It just can appreciate that when capable degree and seek to treat.However, DR can achieve 90% in initial stage cure rate, therefore find early
DR and being effectively treated, which can substantially reduce DR, leads to the risk of blindness.So DR detection technique is particularly important, doctor
Learning image analysis is to cause one of scientist and the research field of doctor's great interest at present.
DR can be divided into two major classes i.e. non incremental algorithm (NPDR) and proliferative retinopathy
(PDR).There are three subclass by NPDR: slight NPDR, moderate NPDR and severe NPDR.Non-proliferative type is lesion early stage, lesion limitation
In in retina, showing as microaneurysm, bleeding, hardness and soft exudate, arteria retina and Lesion of vein.Proliferous type disease
Fading to rare part extension is more than internal limiting membrane, the mark that new vessels are proliferous type occurs.The lesion of DR is described in detail as follows:
(1) aneurysms
Aneurysms represents the sign that perceives of retinal damage most original, and the abnormal permeability of retinal vessel causes
The formation of aneurysms.On the medical image, it is small, circular, and is had usually in the dark red color spot of macular ilm
Point.Aneurysms can be regarded as a red dot, clear-cut margin, and it is big to be similar to CD between 20 μm to 200 μm for size
Small 8.25%.
(2) hard exudate
Different from aneurysms, hard exudate is that lipoprotein and some other protein leak out institute's shape from retinal vessel
At.Visually, it looks like small white or yellow-white deposit, has apparent edge.Hard exudate is usual
The tissue in the form of annular form, usually occurs in ectoretina.Hard exudate is usually irregular and has light
Pool, and it was found that the edge for being closely located to aneurysms or macular edema.
(3) soft exudate
In general, soft exudate is formed since parteriole occludes.The blood flow reduction for flowing to retina is led
Retinal nerve fibre layer (RNFL) ischemic is caused, axoplasmic flow is finally influenced, to accumulate on retinal ganglion cell axon
Axoplasm fragment.This accumulation can be visualized as the fluffy white lesion in RNFL, this is commonly known as soft exudate.
(4) bleeding
Bleeding due to weak blood vessel leakage and occur, the lesion of bleeding is to have the red of different densities and uneven edge
Point form, and it is found in the range of 125 μm.In broad terms, bleeding is divided into two classes: flame and Dot blot go out
Blood, wherein the first seed type is originating from capillaries anterior arteries and appears on nerve fibre.Second of type Dot blot
Bleeding is circular and is less than aneurysms.Dot blot bleeding can appear on the retina of different level, still, it
In most cases appear in the passages through which vital energy circulates end of capillary.
(5) neovascularization
Neovascularization usually indicates that the atypia for appearing in the new blood vessel on retina inner surface occurs.New blood vessel is very thin
Small and penetrate into vitreous chamber repeatedly, this can reduce visual capacity and keep its significant fuzzy, eventually lead to blindness.
(6) macular edema
Macular edema is accredited as the tumor of retina, is typically due to the permeability of abnormal retinal capillary
And occur.Macular edema causes liquid or other solutes around macula lutea to leak and seriously affect eyesight.
Therefore, in the prior art, for the lesion patient including diabetic retinopathy (DR), doctor is only relied on
Stranger's eye judges whether generation lesion and lesion degree by observing pathological image, by visually observing, analyze image very
Holding time and energy, and inefficiency, there is also otherness, this problems demand is solved for the judgement of different doctors.
Summary of the invention
The present invention is directed to solve at least some of the technical problems in related technologies.For this purpose, of the invention
One purpose is to provide a kind of medical image processing method, system, equipment, storage medium, for improving the place to medical image
Reason, analysis efficiency.
The technical scheme adopted by the invention is that:
In a first aspect, the present invention provides a kind of medical image processing method, comprising the following steps:
Lesion image obtaining step extracts lesion image according to medical image；
Score generation step obtains multiple lesion scores according to the lesion image, and the lesion is scored at the lesion
Image belongs to the first abnormal probability of different images intensity of anomaly；
Second abnormal probability obtaining step, obtains the medical image according to the medical image and the lesion image
Multiple second abnormal probability, the described second abnormal probability are the probability that the medical image belongs to different images intensity of anomaly；
Classifying step, according to the multiple described first abnormal probability, multiple described second abnormal probability and different weight systems
Number obtains the final probability that the medical image belongs to different images intensity of anomaly according to different images intensity of anomaly, and will be maximum
The final probability image abnormity degree of the image abnormity degree as the medical image.
Further, the medical image includes fundus photograph.
Further, the lesion image obtaining step includes:
A variety of lesion mask images are obtained according to the medical image and segmentation neural network；
According to the lesion mask images and a variety of lesion images of the medical image acquisition.
Further, the score generation step includes:
The focus characteristic of a variety of lesion images is extracted respectively, and the focus characteristic includes color or shape；
Splice multiple focus characteristics and be inputted the multiple lesions of the first machine learning classification algorithm acquisition and obtains
Point.
Further, the described second abnormal probability obtaining step includes:
The medicine figure is obtained according to the medical image, a variety of lesion images and the second machine learning classification algorithm
The abnormal probability of multiple the second of picture.
Further, described image intensity of anomaly includes normal, the slight first abnormal, the first exception of moderate, severe first
Abnormal and the second exception.
Further, the weight coefficient of the described first abnormal probability is 0.2, and the weight coefficient of the described second abnormal probability is
0.8。
Second aspect, the present invention provide a kind of magic magiscan, comprising:
Lesion image acquiring unit, for extracting lesion image according to medical image；
Score generation unit, for obtaining multiple lesion scores according to the lesion image, the lesion is scored at described
Lesion image belongs to the first abnormal probability of different images intensity of anomaly；
Second abnormal probability acquiring unit, for obtaining the medicine figure according to the medical image and the lesion image
The abnormal probability of multiple the second of picture, the described second abnormal probability are that the medical image belongs to the general of different images intensity of anomaly
Rate；
Taxon, for according to the multiple described first abnormal probability, multiple described second abnormal probability and different power
Weight coefficient obtains the final probability that the medical image belongs to different images intensity of anomaly according to different images intensity of anomaly, and will
Image abnormity degree of the image abnormity degree of the maximum final probability as the medical image.
The third aspect, the present invention provide a kind of Medical Image Processing equipment, comprising:
At least one processor；And
The memory being connect at least one described processor communication；Wherein,
The memory is stored with the instruction that can be executed by least one described processor, and described instruction is by described at least one
A processor executes, so that at least one described processor is able to carry out the medical image processing method.
Fourth aspect, the present invention provide a kind of computer readable storage medium, the computer-readable recording medium storage
There are computer executable instructions, the computer executable instructions are used to that computer to be made to execute the Medical Image Processing side
Method.
The beneficial effects of the present invention are:
One aspect of the present invention obtains the multiple first abnormal probability by lesion image, on the other hand passes through medical image and disease
Stove image obtains the multiple second abnormal probability, further according to the multiple first abnormal probability, multiple second abnormal probability and different power
Weight coefficient obtains the final probability that medical image belongs to different images intensity of anomaly according to different images intensity of anomaly, and will be maximum
Final probability image abnormity degree of the image abnormity degree as medical image, realize to the intensity of anomaly of medical image into
Row analysis overcomes and exists in the prior art to pathological image processing, analysis by naked eyes, the technical issues of inefficiency.
In addition, the present invention also passes through a variety of lesion mask images that segmentation neural network obtains medical image, with further
Obtain a variety of lesion images, it is ensured that accurately extract lesion image.
Detailed description of the invention
Fig. 1 is an a kind of specific embodiment flow chart of medical image processing method in the present invention；
Fig. 2 is an a kind of specific embodiment schematic diagram of the segmentation neural network of medical image processing method in the present invention.
Specific embodiment
It should be noted that in the absence of conflict, the features in the embodiments and the embodiments of the present application can phase
Mutually combination.
Embodiment 1
A kind of medical image processing method, comprising the following steps:
Lesion image obtaining step extracts lesion image according to medical image；
Score generation step obtains multiple lesion scores according to lesion image, and lesion is scored at lesion image and belongs to difference
The abnormal probability of the first of image abnormity degree, for example, image abnormity degree includes normal and two kinds abnormal, then score generation step
It is to obtain lesion image to belong to normal or abnormal probability, obtains two first abnormal probability, and so on；
Second abnormal probability obtaining step obtains multiple second exceptions of medical image according to medical image and lesion image
Probability, the second abnormal probability are the probability that medical image belongs to different images intensity of anomaly；
Classifying step, according to the multiple first abnormal probability, multiple second abnormal probability and different weight coefficients according to not
Medical image is obtained with image abnormity degree and belongs to the final probability of different images intensity of anomaly, and by maximum final probability
Image abnormity degree of the image abnormity degree as medical image.I.e. by the first of same image abnormity degree the abnormal probability and the
Two abnormal probability are added the final probability for obtaining the image abnormity degree according to different weight coefficients, due to there is multiple and different figures
As intensity of anomaly, then multiple final probability of medical image can be obtained.
On the one hand a kind of medical image processing method obtains the multiple first abnormal probability by lesion image, on the other hand
The multiple second abnormal probability are obtained by medical image and lesion image, further according to the multiple first abnormal probability, multiple second different
Normal probability and different weight coefficients obtain medical image according to different images intensity of anomaly and belong to different images intensity of anomaly
Final probability, and using the image abnormity degree of maximum final probability as the image abnormity degree of medical image, it realizes to doctor
The intensity of anomaly for learning image is analyzed, and is improved processing and the analysis efficiency of medical image, is overcome and exist in the prior art to disease
The technical issues of reason image procossing, analysis rely on visually, inefficiency.
Further, medical image includes but is not limited to fundus photograph, in the present embodiment, to medicine by taking fundus photograph as an example
Image processing method is illustrated, the processing method of the medical image of remaining type, is referred to the following place to fundus photograph
The specific descriptions of reason method, it is similar to the processing method of image, it repeats no more.And image abnormity degree includes but is not limited to just
Often, slightly it is first abnormal, moderate first is abnormal, severe first is abnormal and second is abnormal, wherein by taking fundus photograph as an example, then the
One it is abnormal be non-proliferative lesion, second it is abnormal be proliferative lesion, then the image abnormity degree of fundus photograph includes but unlimited
In normal, slight non-proliferative lesion, moderate non-proliferative lesion, severe non-proliferative lesion and proliferative lesion.With reference to figure
1, Fig. 1 is an a kind of specific embodiment flow chart of medical image processing method in the present invention；Below to Medical Image Processing side
Method illustrates:
Step 1: being pre-processed to data.The data that the present invention uses are two-dimensional color fundus photograph, original picture size
It is 3000 × 3000, picture need to be pre-processed, first puncture the redundances such as the background of picture, by the size of picture
It is adjusted to 128 × 128.
Step 2: lesion image obtaining step, comprising:
A variety of lesion mask images are obtained according to medical image and segmentation neural network；
According to lesion mask images and a variety of lesion images of medical image acquisition.
Wherein, the main task for dividing neural network is to realize a variety of lesion segmentations and extraction, in the present embodiment, lesion kind
Class includes aneurysms, hard exudate and neonatal retinal vascular, and aneurysms, hard exudate belong to non-proliferative venereal disease
Become, neonatal retinal vascular belongs to proliferative lesion.Picture to be measured is input in trained segmentation neural network, by dividing
It cuts each pixel on neural network prediction picture and belongs to background, aneurysms, hard exudate or retinal vessel, and
And export the result predicted at three masks, it is aneurysms mask, hard exudate mask and retinal vessel respectively
Mask.Mask is specifically to be made of 0 and 1, background 0, target 1, such as when generation aneurysms mask, passes through segmentation nerve
Whether each pixel of neural network forecast belongs to aneurysms region, if the location of pixels is labeled as 1, the otherwise location of pixels mark
It is denoted as 0, finally obtains the aneurysms mask artwork for being 128 × 128 with the equirotal size of original image.Mask can be with accurate description
Mask is corresponded to original image by the position of target, image namely lesion image after can extracting each lesion segmentation.
Reference table 1 and Fig. 2, Fig. 2 are a kind of the one specific of the segmentation neural network of medical image processing method in the present invention
Embodiment schematic diagram；The realization of segmentation neural network is based primarily upon multi-task learning frame, divides the segmentation task of neural network
There are three, it is segmentation aneurysms, segmentation hard exudate and segmentation retinal vessel respectively.The first part of network is known as
" inclusion layer ", the weight of these layers are all general for these three tasks.It then, is the certain layer of each task, referred to as
" task certain layer ", the parameter between these layers all independently calculate, and it is shared to be not involved in cross-layer.In these particular task layers
In, information of the e-learning specific to task.These each independent task layers can generate one and individually export as a result, tool
Body is the segmentation mask for exporting aneurysms, hard exudate, retinal vessel.
Table 1 divides neural network level structure
Above-mentioned segmentation neural network can be partitioned into a variety of lesion images simultaneously, be not limited to the present embodiment in three kinds of lesion figures
Picture.And using before dividing neural network, it needs to be trained network, specifically, then initialization model first inputs
The sample for having marked mask flag data is trained, and is trained when training using whole picture, is sampled without patch,
And experiments have shown that more efficient using full figure.After inputting picture, carries out cubic convolution respectively and down-sampling obtains the small figure (5 of feature
× 5 × 128) convolution then, is carried out again and obtains feature (1 × 1 × 1024), is then carried out three groups and is connected up to three entirely
The feature of business, three tasks carry out deconvolution and up-sampling respectively, obtain with original image big figure of a size, these figures are exactly point
The mask for cutting out.Segmentation neural network model is continued to optimize in the training process, and adjusting parameter is until convergence, the target of optimization
It is the continuous difference reduced between the segmentation figure (lesion image) and sample labeled data that neural network forecast comes out, that is, minimizes loss
Function L1:
In above formula, N is number of training, LMAIndicate the Pixel-level loss of aneurysms segmentation, LHESIndicate hard exudate
Segmentation loss, LNVIndicate the segmentation loss of retinal vessel；F (x, θ) is the segmentation result for dividing neural network prediction, wherein
X is a pixel of sample, and θ is learning rate；S1, S2, S3It is the sample of aneurysms, hard exudate, retinal vessel respectively
Labeling position；Φ (θ) is regular terms.
A variety of lesion mask images of fundus photograph, including fine motion can be then obtained using trained segmentation neural network
Arteries and veins tumor mask images, hard exudate mask images and retinal vessel mask images；According to lesion mask images and medicine
Image can be divided to obtain a variety of lesion images, be respectively aneurysms lesion image, hard exudate lesion image and
Retinal vessel lesion image.
Step 3: score generation step, comprising:
The focus characteristic of a variety of lesion images is extracted respectively, and focus characteristic includes color or shape；
Splice multiple focus characteristics and is inputted the first machine learning classification algorithm to obtain multiple lesion scores.
Specifically, after obtaining each lesion image, color, shape of each lesion etc. is extracted such as table according to lesion image
Feature shown in 2, each lesion image can extract 86 dimensional features, and the focus characteristic of 3 lesion images is pressed fine motion
Arteries and veins tumor lesion image, hard exudate lesion image, retinal vessel lesion image sequential concatenation get up as 258 dimensional features
Vector, and feature vector is input to and can be obtained multiple lesions in score generator (i.e. the first machine learning classification algorithm) and obtain
It is divided intoI is the classification of image abnormity degree, and X is the number of medical image, i.e. acquisition lesion image belongs to different figures
Picture intensity of anomaly (including normal, slight non-proliferative lesion, moderate non-proliferative lesion, severe non-proliferative lesion and proliferation
Venereal disease become) probability, it can obtain 5 lesion scores.
2 focus characteristic of table extracts list
| Feature | Feature description |
| f1 | Mean value on RGB image |
| f2 | Variance on RGB image |
| f3-f12 | Color moment on RGB image |
| f13 | The perimeter of target area and |
| f14 | The area of target area and |
| f15-f74 | The LBP feature of target area |
| f75-f86 | HSV feature |
In the present embodiment, score generator uses random forests algorithm.Score generator is instructed before the use
Practice, specifically, the lesion degree that 258 dimensional feature vector of focus characteristic of training sample and the sample mark is input to score
It is trained in generator.Trained score generator is by calculating it can be concluded that the lesion of input belongs to 5 kinds of image abnormities
Degree (i.e. normal, slight non-proliferative lesion, moderate non-proliferative lesion, severe non-proliferative lesion and proliferative lesion)
Probability, take lesion score of this probability value as the sample.
Step 4: the second abnormal probability obtaining step, comprising:
Multiple the second of medical image are obtained according to medical image, a variety of lesion images and the second machine learning classification algorithm
Abnormal probability.
In the present embodiment, image abnormity degree includes normal, slight non-proliferative lesion, moderate non-proliferative lesion, again
Spend non-proliferative lesion and proliferative lesion；Second machine learning classification algorithm is designed based on 3D convolutional neural networks
, 3D convolutional neural networks can input original medical image (i.e. fundus photograph) and aneurysms lesion image, hard simultaneously
Property exudate lesion image and retinal vessel lesion image, feature can be extracted from various dimensions, then carries out 3D convolution, to capture
The characteristic information obtained in the multiple images.Specifically, reference table 3,3D convolutional neural networks include that a real connecting line layer is (i.e. hard
Articulamentum), 4 convolutional layers, 3 down-sampling layers and a full articulamentum.The cube of each convolution nuclear convolution be include original image,
Microaneurysm lesion image, retinal vessel lesion image, hard exudate lesion image 4 pictures, every picture it is big
Small is 128 × 128.In first layer, the core for applying a fixed real line goes to handle the figure of input, generates multiple
Then the information in channel is handled multiple channels respectively, finally again combine the information in all channels to obtain final spy
Sign description.This solid line layer is actually the priori knowledge encoded to lesion, this gets well than random initializtion performance.Every figure
The information for extracting five channels is respectively: the gradient of R, G, B color channel and the direction x and y.So there is 4 × 5=20 disease
Stove figure.Then convolution is carried out respectively in each channel in five channels with one 9 × 9 × 2 3D convolution kernel again.In order to increase
The number of characteristic pattern extracts different features, uses two different convolution kernels, such L in each position1It is wrapped in layer
Two characteristic pattern groups are contained, every group all includes (4-2+1) × 5=15 characteristic pattern.It is followed by down-sampling layer L3Layer, in L2
Down-sampling is carried out with 3 × 3 windows in the characteristic pattern of layer, can thus obtain same number but the lesion of spatial resolution reduction
Figure.L4It is that 9 × 9 × 2 3D convolution kernel is respectively adopted in 5 channels to be calculated.In order to increase characteristic pattern number, each
Position all uses three different convolution kernels, can be obtained by 6 groups of different characteristic patterns in this way, and every group has (4-2+1-2+1) × 5
=10 characteristic patterns.L5Layer be 2 × 2 down-sampling window, then carry out again 2D convolution sum down-sampling twice obtain 128 ×
1 × 1 feature vector, and 128 dimensions are according to empirically determined in the past.It inputs after medical image by under the convolution sum of multilayer
Sampling, the multiple images of input are all converted into the feature vector of one 128 dimension, this feature vector captures retina eyeground
The characteristic information of photo.The feature vector of obtain 128 dimensions is input to softmax layers again, softmax layers of number of nodes and figure
As the class number of intensity of anomaly is consistent, and each node and L9In 128 nodes connect entirely.Final softmax
Layer can obtain a series of predicted values, i.e., the medical image probability value that belongs to 5 kinds of image abnormity degree, i.e., the second abnormal probability are false
If this value isWherein, i (i=0,1,2,3,4) is image abnormity degree classification (i.e. normal, slight non-proliferative
Lesion, moderate non-proliferative lesion, severe non-proliferative lesion and proliferative lesion), X is the medical image number of input.
3 3D convolutional neural networks hierarchical structure of table
Before using 3D convolutional neural networks, needs to be trained it, first initialization model, then input and marked
The sample of note intensity of anomaly is trained, original image, microaneurysm lesion image, the hard exudate lesion of input sample when training
Image, retinal vessel lesion image 4 pictures be trained.3D convolutional neural networks model is constantly excellent in the training process
Change, adjusting parameter is until convergence, the target of optimization are that the classification results that continuous diminution neural network forecast comes out and sample mark are abnormal
Difference between degree, i.e. minimum loss function L2:
In above formula, N is number of training, LsoftmaxIndicate network class loss, labeled data；F (x, θ) is neural network forecast
Classification results, wherein x be a sample, θ is learning rate；C is mark classification；Φ (θ) is regular terms.
Step 5: score integrated classification is specifically obtaining the second abnormal probability(have 5 in the present embodiment
A second abnormal probability) and the first abnormal probability(having 5 first abnormal probability, i.e. lesion score in the present embodiment),
According to different weight coefficient is arranged to obtain final abnormal probability, in the present embodiment, the weight coefficient of the first abnormal probability
It is 0.2, the weight coefficient of the second abnormal probability is 0.8, then final abnormal probability is to get dividing the fusion value to be (having 5 score fusion values in the present embodiment), in the result of calculating, which image
The score fusion value highest of intensity of anomaly classification, then the category is the affiliated image abnormity degree classification of medical image, that is, seeks C
=argmaxiFi(X)。
A kind of medical image processing method proposes score syncretizing mechanism, the lesion score obtained by each lesion image is melted
Enter sorter network, improves susceptibility and classification accuracy to each lesion.And score integrated classification frame of the invention is not only
It is confined to detect in the present invention three kinds of lesions, to be checked to be expanded when measuring more lesions, system suitability is strong.
Embodiment 2
A kind of magic magiscan, comprising:
Lesion image acquiring unit, for extracting lesion image according to medical image；
Score generation unit, for obtaining multiple lesion scores according to lesion image, lesion is scored at lesion image and belongs to
The abnormal probability of the first of different images intensity of anomaly；
Second abnormal probability acquiring unit, for obtaining multiple the second of medical image according to medical image and lesion image
Abnormal probability, the second abnormal probability are the probability that medical image belongs to different images intensity of anomaly；
Taxon, for being pressed according to the multiple first abnormal probability, multiple second abnormal probability and different weight coefficients
The final probability that medical image belongs to different images intensity of anomaly is obtained according to different images intensity of anomaly, and will be maximum final general
Image abnormity degree of the image abnormity degree of rate as medical image.
The specific work process and beneficial effect of magic magiscan please refer to 1 traditional Chinese medicine image processing method of embodiment
The specific descriptions of method, repeat no more.
Embodiment 3
A kind of Medical Image Processing equipment, comprising:
At least one processor；And
The memory being connect at least one described processor communication；Wherein,
The memory is stored with the instruction that can be executed by least one described processor, and described instruction is by described at least one
A processor executes, so that at least one described processor is able to carry out the medical image processing method.At medical image
The specific descriptions of reason method are repeated no more referring to the description in embodiment 1.
Embodiment 4
A kind of computer readable storage medium, the computer-readable recording medium storage have computer executable instructions,
The computer executable instructions are used to that computer to be made to execute the medical image processing method.Medical image processing method
It specifically describes referring to the description in embodiment 1, repeats no more.
It is to be illustrated to preferable implementation of the invention, but the invention is not limited to the implementation above
Example, those skilled in the art can also make various equivalent variations on the premise of without prejudice to spirit of the invention or replace
It changes, these equivalent deformations or replacement are all included in the scope defined by the claims of the present application.
Claims (10)
1. a kind of medical image processing method, which comprises the following steps:
Lesion image obtaining step extracts lesion image according to medical image；
Score generation step obtains multiple lesion scores according to the lesion image, and the lesion is scored at the lesion image
Belong to the first abnormal probability of different images intensity of anomaly；
Second abnormal probability obtaining step, obtains the multiple of the medical image according to the medical image and the lesion image
Second abnormal probability, the described second abnormal probability are the probability that the medical image belongs to different images intensity of anomaly；
Classifying step is pressed according to the multiple described first abnormal probability, multiple described second abnormal probability and different weight coefficients
The final probability that the medical image belongs to different images intensity of anomaly is obtained according to different images intensity of anomaly, and by maximum institute
State image abnormity degree of the image abnormity degree as the medical image of final probability.
2. medical image processing method according to claim 1, which is characterized in that the medical image includes that eyeground is shone
Piece.
3. medical image processing method according to claim 1, which is characterized in that the lesion image obtaining step packet
It includes:
A variety of lesion mask images are obtained according to the medical image and segmentation neural network；
According to the lesion mask images and a variety of lesion images of the medical image acquisition.
4. medical image processing method according to claim 3, which is characterized in that the score generation step includes:
The focus characteristic of a variety of lesion images is extracted respectively, and the focus characteristic includes color or shape；
Splice multiple focus characteristics and be inputted the first machine learning classification algorithm and obtains multiple lesion scores.
5. medical image processing method according to claim 3, which is characterized in that the described second abnormal probability obtaining step
Include:
The medical image is obtained according to the medical image, a variety of lesion images and the second machine learning classification algorithm
Multiple second abnormal probability.
6. medical image processing method according to any one of claims 1 to 5, which is characterized in that described image exception journey
Degree includes normal, slight first exception, moderate first is abnormal, severe first is abnormal and second is abnormal.
7. medical image processing method according to any one of claims 1 to 5, which is characterized in that described first is abnormal general
The weight coefficient of rate is 0.2, and the weight coefficient of the described second abnormal probability is 0.8.
8. a kind of magic magiscan characterized by comprising
Lesion image acquiring unit, for extracting lesion image according to medical image；
Score generation unit, for obtaining multiple lesion scores according to the lesion image, the lesion is scored at the lesion
Image belongs to the first abnormal probability of different images intensity of anomaly；
Second abnormal probability acquiring unit, for obtaining the medical image according to the medical image and the lesion image
Multiple second abnormal probability, the described second abnormal probability are the probability that the medical image belongs to different images intensity of anomaly；
Taxon, for according to the multiple described first abnormal probability, multiple described second abnormal probability and different weight systems
Number obtains the final probability that the medical image belongs to different images intensity of anomaly according to different images intensity of anomaly, and will be maximum
The final probability image abnormity degree of the image abnormity degree as the medical image.
9. a kind of Medical Image Processing equipment characterized by comprising
At least one processor；And
The memory being connect at least one described processor communication；Wherein,
The memory is stored with the instruction that can be executed by least one described processor, and described instruction is by described at least one
It manages device to execute, so that at least one described processor is able to carry out Medical Image Processing as described in any one of claim 1 to 7
Method.
10. a kind of computer readable storage medium, which is characterized in that the computer-readable recording medium storage has computer can
It executes instruction, the computer executable instructions are for making computer execute medicine figure as described in any one of claim 1 to 7
As processing method.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811320629.9A CN109671049B (en) | 2018-11-07 | 2018-11-07 | Medical image processing method, system, equipment and storage medium |
| PCT/CN2018/124660 WO2020093563A1 (en) | 2018-11-07 | 2018-12-28 | Medical image processing method, system, device, and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811320629.9A CN109671049B (en) | 2018-11-07 | 2018-11-07 | Medical image processing method, system, equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109671049A true CN109671049A (en) | 2019-04-23 |
| CN109671049B CN109671049B (en) | 2024-03-01 |
Family
ID=66142041
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811320629.9A Active CN109671049B (en) | 2018-11-07 | 2018-11-07 | Medical image processing method, system, equipment and storage medium |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN109671049B (en) |
| WO (1) | WO2020093563A1 (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111144296A (en) * | 2019-12-26 | 2020-05-12 | 湖南大学 | Retina fundus picture classification method based on improved CNN model |
| CN112652394A (en) * | 2021-01-14 | 2021-04-13 | 浙江工商大学 | Multi-focus target detection-based retinopathy of prematurity diagnosis system |
| WO2021189855A1 (en) * | 2020-09-22 | 2021-09-30 | 平安科技（深圳）有限公司 | Image recognition method and apparatus based on ct sequence, and electronic device and medium |
| CN115816834A (en) * | 2023-02-20 | 2023-03-21 | 常熟理工学院 | Method and system for monitoring printing quality of printer in real time |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115311188B (en) * | 2021-05-08 | 2023-12-22 | 数坤科技股份有限公司 | Image recognition method and device, electronic equipment and storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106548192A (en) * | 2016-09-23 | 2017-03-29 | 北京市商汤科技开发有限公司 | Based on the image processing method of neutral net, device and electronic equipment |
| US20170287137A1 (en) * | 2016-03-31 | 2017-10-05 | Adobe Systems Incorporated | Utilizing deep learning for boundary-aware image segmentation |
| DE202017104953U1 (en) * | 2016-08-18 | 2017-12-04 | Google Inc. | Processing fundus images using machine learning models |
| CN108010021A (en) * | 2017-11-30 | 2018-05-08 | 上海联影医疗科技有限公司 | A kind of magic magiscan and method |
| CN108230294A (en) * | 2017-06-14 | 2018-06-29 | 北京市商汤科技开发有限公司 | Image detecting method, device, electronic equipment and storage medium |
| CN108634934A (en) * | 2018-05-07 | 2018-10-12 | 北京长木谷医疗科技有限公司 | The method and apparatus that spinal sagittal bit image is handled |
Family Cites Families (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP5935146B2 (en) * | 2011-08-30 | 2016-06-15 | 大日本印刷株式会社 | Ophthalmic disease image analysis apparatus, ophthalmic image analysis method, and ophthalmic image analysis program |
| JP2016002380A (en) * | 2014-06-18 | 2016-01-12 | キヤノン株式会社 | Image processing system, operation method for the same, and program |
| CN106530290A (en) * | 2016-10-27 | 2017-03-22 | 朱育盼 | Medical image analysis method and device |
| CN106530295A (en) * | 2016-11-07 | 2017-03-22 | 首都医科大学 | Fundus image classification method and device of retinopathy |
| CN108771530B (en) * | 2017-05-04 | 2021-03-30 | 深圳硅基智能科技有限公司 | Fundus lesion screening system based on deep neural network |
| CN107145756A (en) * | 2017-05-17 | 2017-09-08 | 上海辉明软件有限公司 | A kind of stroke types Forecasting Methodology and device |
| CN107679525B (en) * | 2017-11-01 | 2022-11-29 | 腾讯科技（深圳）有限公司 | Image classification method and device and computer readable storage medium |
| CN108734209A (en) * | 2018-05-16 | 2018-11-02 | 上海鹰瞳医疗科技有限公司 | Feature recognition based on more images and equipment |
-
2018
- 2018-11-07 CN CN201811320629.9A patent/CN109671049B/en active Active
- 2018-12-28 WO PCT/CN2018/124660 patent/WO2020093563A1/en active Application Filing
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170287137A1 (en) * | 2016-03-31 | 2017-10-05 | Adobe Systems Incorporated | Utilizing deep learning for boundary-aware image segmentation |
| DE202017104953U1 (en) * | 2016-08-18 | 2017-12-04 | Google Inc. | Processing fundus images using machine learning models |
| CN106548192A (en) * | 2016-09-23 | 2017-03-29 | 北京市商汤科技开发有限公司 | Based on the image processing method of neutral net, device and electronic equipment |
| CN108230294A (en) * | 2017-06-14 | 2018-06-29 | 北京市商汤科技开发有限公司 | Image detecting method, device, electronic equipment and storage medium |
| CN108010021A (en) * | 2017-11-30 | 2018-05-08 | 上海联影医疗科技有限公司 | A kind of magic magiscan and method |
| CN108634934A (en) * | 2018-05-07 | 2018-10-12 | 北京长木谷医疗科技有限公司 | The method and apparatus that spinal sagittal bit image is handled |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111144296A (en) * | 2019-12-26 | 2020-05-12 | 湖南大学 | Retina fundus picture classification method based on improved CNN model |
| CN111144296B (en) * | 2019-12-26 | 2023-04-18 | 湖南大学 | Retina fundus picture classification method based on improved CNN model |
| WO2021189855A1 (en) * | 2020-09-22 | 2021-09-30 | 平安科技（深圳）有限公司 | Image recognition method and apparatus based on ct sequence, and electronic device and medium |
| CN112652394A (en) * | 2021-01-14 | 2021-04-13 | 浙江工商大学 | Multi-focus target detection-based retinopathy of prematurity diagnosis system |
| CN115816834A (en) * | 2023-02-20 | 2023-03-21 | 常熟理工学院 | Method and system for monitoring printing quality of printer in real time |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109671049B (en) | 2024-03-01 |
| WO2020093563A1 (en) | 2020-05-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109671049A (en) | A kind of medical image processing method, system, equipment, storage medium | |
| Tian et al. | Multi-path convolutional neural network in fundus segmentation of blood vessels | |
| Sheng et al. | Retinal vessel segmentation using minimum spanning superpixel tree detector | |
| Xiang et al. | Automatic segmentation of retinal layer in OCT images with choroidal neovascularization | |
| Abràmoff et al. | Retinal imaging and image analysis | |
| Akram et al. | Automated detection of dark and bright lesions in retinal images for early detection of diabetic retinopathy | |
| Xiong et al. | An approach to evaluate blurriness in retinal images with vitreous opacity for cataract diagnosis | |
| CN109493954A (en) | A kind of SD-OCT image retinopathy detection system differentiating positioning based on classification | |
| Hu et al. | Multimodal retinal vessel segmentation from spectral-domain optical coherence tomography and fundus photography | |
| CN108986106A (en) | Retinal vessel automatic division method towards glaucoma clinical diagnosis | |
| CN110786824B (en) | Coarse marking fundus oculi illumination bleeding lesion detection method and system based on bounding box correction network | |
| Jan et al. | Retinal image analysis aimed at blood vessel tree segmentation and early detection of neural-layer deterioration | |
| Fu et al. | Optic disc segmentation by U-net and probability bubble in abnormal fundus images | |
| Zhu et al. | Digital image processing for ophthalmology: Detection of the optic nerve head | |
| Vimala et al. | Automatic detection of optic disk and exudate from retinal images using clustering algorithm | |
| Suero et al. | Locating the Optic Disc in Retinal Images Using Morphological Techniques. | |
| Wu et al. | Diabetic macular edema grading based on improved Faster R-CNN and MD-ResNet | |
| Suriyasekeran et al. | Algorithms for diagnosis of diabetic retinopathy and diabetic macula edema-a review | |
| CN112957005A (en) | Automatic identification and laser photocoagulation region recommendation algorithm for fundus contrast image non-perfusion region | |
| Wang et al. | Accurate disease detection quantification of iris based retinal images using random implication image classifier technique | |
| Karkuzhali et al. | Robust intensity variation and inverse surface adaptive thresholding techniques for detection of optic disc and exudates in retinal fundus images | |
| Maqsood et al. | Detection of macula and recognition of aged-related macular degeneration in retinal fundus images | |
| Fu et al. | Fovea localization by blood vessel vector in abnormal fundus images | |
| Lamba et al. | Machine learning based segmentation and classification algorithms for glaucoma detection | |
| Zhu et al. | Automated framework for intraretinal cystoid macular edema segmentation in three-dimensional optical coherence tomography images with macular hole |
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 | ||
| CB03 | Change of inventor or designer information | ||
| CB03 | Change of inventor or designer information |
Inventor after: Xu Yong Inventor after: Luo Xiaoling Inventor after: Pu Zuhui Inventor after: Mou Lisha Inventor after: Hu Jiying Inventor before: Xu Yong Inventor before: Luo Xiaoling Inventor before: Pu Zhihui Inventor before: Mou Lisha Inventor before: Hu Jiying |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |