CN108922599A - A kind of accurate mask method of medical image lesion point based on MIL - Google Patents
A kind of accurate mask method of medical image lesion point based on MIL Download PDFInfo
- Publication number
- CN108922599A CN108922599A CN201810677693.6A CN201810677693A CN108922599A CN 108922599 A CN108922599 A CN 108922599A CN 201810677693 A CN201810677693 A CN 201810677693A CN 108922599 A CN108922599 A CN 108922599A
- Authority
- CN
- China
- Prior art keywords
- sample
- lesion point
- model
- medical image
- data set
- 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
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H30/00—ICT specially adapted for the handling or processing of medical images
- G16H30/20—ICT specially adapted for the handling or processing of medical images for handling medical images, e.g. DICOM, HL7 or PACS
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Epidemiology (AREA)
- Medical Informatics (AREA)
- Primary Health Care (AREA)
- Public Health (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of accurate mask method of medical image lesion point based on MIL, includes the following steps：Step 1：Medical images data sets are acquired, positive sample collection and negative sample collection are divided into；Step 2：Carry out the initialization of disaggregated model；Step 3：S example is acquired for each sample standard deviation that positive sample is concentrated, each example is input in sorter model；Record has the maximum example of lesion point probability value, and then the example is added in data set D；Step 4：S example is acquired for each sample standard deviation that negative sample is concentrated, each example is input in sorter model；Record most unlikely has the example of lesion point, and then also the example is added in data set D；Step 5：It is iterated training in neural network classifier model, obtains training pattern；Step 6：Detection and the mark of new samples are carried out according to disaggregated model；The present invention in large quantity can accurately mark lesion point sample, and speed it is fast, it is at low cost, with high efficiency.
Description
Technical field
The present invention relates to a kind of medical image processing methods, and in particular to one kind is based on MIL（Multiple Instance
Learning）The accurate mask method of medical image lesion point.
Background technique
The image labeling of medical domain is：By lesions such as tumour, tubercle, calcification points in the image datas such as DR piece, CT film
Region marked out in the image datas such as dicom by way of human-computer interaction by doctor come；As deep learning is in medicine figure
As the expansion in process field, training just needs the data set of large-scale standard to support；Therefore, the accurate mark of mass data collection
Problem is just particularly important；Different with the mark of normal image, the mark of medical image needs professional knowledge and technical ability
People does；Therefore, it is difficult to obtain a large amount of accurate labeled data collection to learn for deep neural network；Currently, by traditional depth
Habit is widely used in the field of medical imaging, and there are still following problems：First, the image of medical domain, which accurately marks, needs high level
Doctor complete, non-medical field personnel are nonsensical to the mark of medical image；But objectively, the work of China doctor is strong
Degree is big, it is difficult to have energy to carry out the accurate mark of medical image；And there is a large amount of different types of lesion figures in reality
Picture, but the picture much stored does not all have enough time marking and use, and resource utilization is lower；Second, that is, enable a physician to essence
The medical image really marked, but need to eliminate personal subjectivity since data mark, it needs more people repeatedly to mark, takes comprehensive flat
Result；This results in capable of reaching effective labeled data of application level in fact negligible amounts；Third, the image sample of disease
Originally it is limited to the disease incidence of the disease, certain diseases are difficult to construct enough magnitudes there are sample rareness due to being non-common disease
The problem of data set.
The reform of digitlization case history is carried out in current existing medical institutions extensively, having built many PACS systems etc. includes
The medical records database such as medical imaging data, physicochemical data；Diagnosis is contained in these case histories, but not to focal zone
Domain precise marking；Current medical imaging data are mostly non-precision labeled data（It is sick and disease-free that has only been marked）, but not
Accurately provide lesions position coordinate；Therefore current deep learning algorithm is caused to be difficult to directly adapt to.
Summary of the invention
The present invention provides that a kind of speed is fast, at low cost, the accurate side of mark of the high-efficient medical image lesion point based on MIL
Method.
The technical solution adopted by the present invention is that：A kind of accurate mask method of medical image lesion point based on MIL, including with
Lower step：
Step 1：Medical images data sets are acquired, positive sample collection and negative sample collection are divided into；
Step 2：Carry out the initialization of disaggregated model；
Step 3：S example is acquired for each sample standard deviation that positive sample is concentrated, each example is input in sorter model；
Record has the maximum example of lesion point probability value, and then the example is added in data set D；
Step 4：S example is acquired for each sample standard deviation that negative sample is concentrated, each example is input in sorter model；
Record most unlikely has the example of lesion point, is then added to the example in the data set D that step 3 obtains；
Step 5：The data set D that step 4 obtains is iterated training in neural network classifier model, obtains training mould
Type；
Step 6：Detection and the mark of new samples are carried out according to the model that step 5 obtains.
Further, the disaggregated model uses the neural network model based on LeNet, including input layer, the first convolution
Layer, the first pond layer, the second convolutional layer, the second pond layer, the first full articulamentum, activation primitive layer, the second full articulamentum and use
In realize classification and it is softmax layers normalized.
Further, detection and new samples each in annotation process sample S example in the step 6, to each example
Classified with sorter model；If having one and above example having lesion point, determine that the sample is to have lesion point diagram piece；It is no
Then determine the sample for no lesion point diagram piece；Until having detected all new samples.
Further, during the repetitive exercise in the step 5, step 3 is repeated before training each time and step 4 obtains
Take updated data set D.
Further, initialization procedure is to choose the sample with label to carry out supervised learning in the step 2.
The beneficial effects of the invention are as follows：
（1）The present invention is not influenced by variety classes lesion, and mixing to a variety of lesions also can be carried out accurate mark, robust
Property is strong；
（2）The present invention also can be carried out essence to some irregular, uncommon shapes not only to the mark of lesion point Common Shape
Really mark, universality are high；
（3）The present invention is conducive to detection and mark of the hospital to various lesion images, can greatly reduce professional technician to doctor
The workload of image labeling is learned, while being also convenient for the lesion point that doctor quickly checked and diagnosed patient；
（4）The present invention large batch of can accurately mark lesion point, speed is fast, it is at low cost, with high efficiency.
Detailed description of the invention
Fig. 1 is flow diagram of the present invention.
Fig. 2 is the disaggregated model structural schematic diagram that the present invention uses.
Specific embodiment
The present invention will be further described in the following with reference to the drawings and specific embodiments.
A kind of accurate mask method of medical image lesion point based on MIL, includes the following steps：
Step 1：Medical images data sets are acquired, positive sample collection and negative sample collection are divided into；
By Medical Devices camera acquire different types of lesion point image several, therefrom number of images at regular intervals
Select picture clearly image as raw data set；A certain number of pictures are chosen, are made after corresponding image tag is convenient for
It is continuous to be trained.
Step 2：Carry out the initialization of disaggregated model；
It chooses the sample with label and carries out supervised learning, carry out the initialization of disaggregated model；Disaggregated model selection is based on
The model of the medical image lesion point precise classification of LeNet, disaggregated model are not limited to LeNet, it is also an option that other points
Class model can all make the adjustment of relevant parameter, for improving nicety of grading according to training result when training terminates every time；
It can also be using disaggregated models such as following AlexNet, GoogleNet, VGG.
Based on LeNet disaggregated model as shown in Fig. 2, including input layer, the first convolutional layer, the first pond layer, the second convolution
Layer, the second pond layer, the first full articulamentum, activation primitive layer, the second full articulamentum and for realizing classification and it is normalized
Softmax layers.
Input layer：Input layer inputs one 28 × 28 picture；
First convolutional layer：Convolution is carried out to input layer picture respectively using 20 5 × 5 convolution kernels, includes 20 × 5 × 5=500
A parameter weighting parameter；Picture side length is after convolution（28-5+1）/ 1=24,20 24 × 24 map are generated, comprising 20 × 24 ×
24=11520 neurons；
First pond layer：It is down-sampled to upper one layer each 2 × 2 region progress, each maximum regional value is chosen, this layer is not joined
Number；The down-sampled length of each map and width later become original half；
Second convolutional layer：The layer carries out convolution, institute to upper one layer each map respectively using 20 × 50 5 × 5 convolution kernels
To include 20 × 50 × 5 × 5=25000 parameter weighting parameters；Picture side length is after convolution（12-5+1）/ 1=8, so generating
50 8 × 8 map include 50 × 8 × 8=3200 neurons；
Second pond layer：It is similar with first pond layer function, by the down-sampled map for being 4 × 4 of 8 × 8 map；The layer is without ginseng
Number；
First full articulamentum：Upper one layer of all neurons are attached, which contains 500 neurons, so one is shared
50 × 4 × 4 × 500=400000 weighting parameters；
Activation primitive layer, that is, relu layers：It realizes x=max [0, x], this layer of neuron number and upper one layer of identical, no weighting parameter；
Second full articulamentum：Function is similar with the first full articulamentum, which shares 10 neurons, includes 500 × 10=5000
Parameter；
Softmax layers：It realizes classification and normalization, the probability of each example generic is recorded using softmax function, thus
Carry out classification judgement.
Step 3：S example is acquired for each sample standard deviation that positive sample is concentrated, each example is input to classifier mould
In type；Record has the maximum example of lesion point probability value, and then the example is added in data set D.
Step 4：S example is acquired for each sample standard deviation that negative sample is concentrated, each example is input to classifier mould
In type；Record most unlikely has the example of lesion point, is then added to the example in the data set D that step 3 obtains.
Step 5：The data set D that step 4 obtains is iterated training in neural network classifier model, is trained
Model；
Data set is sent in disaggregated model by Program transformation at .npy format using mainstream deep learning frame and platform,
Classification based training is carried out, training pattern is obtained and is saved in local；Carry out multiplicating training, each training will resampling simultaneously
More new data set；It carries out saving last training pattern after repeatedly training.
Step 6：Detection and the mark of sample are carried out according to the model that step 5 obtains；The sample to be marked is sent to
Differentiation is trained in disaggregated model；Each new samples sample S example, are classified to each example with sorter model；
If having one and above example having lesion point, determine that the sample is to have lesion point diagram piece；Otherwise determine the sample for no lesion
Point picture；Until all new samples have been detected, so that the mass for completing new samples accurately marks.
Wherein MIL（Multiple Instance Learning）It indicates multi-instance learning, includes packet（bags）And example
（instance）Two key concepts；Packet is by multiple composition examples, for example in Medical Images Classification, a picture is exactly
One packet, the patches that picture segmentation goes out mean that example.
In training part：It takes sub-fraction training sample to be labeled, is then carried out by the method for certain supervised learning
Training obtains the disaggregated model of an initialization；And then S instance is sampled to each positive sample bag, each
Instance is input in sorter model, has the lesion point maximum instance of probability value using SoftMax function record, most
The instance is added in data set D afterwards；S instance is equally sampled to each negative sample bag, each
Instance is input in sorter model, most unlikely there is the instance of lesion point using SoftMax function record, also will
The instance is added in data set D；It is finally input in sorter model and is trained with obtained data set D again, often
Secondary training will resampling and more new data set D, and save finally trained model.Differentiating part：New sample
Bags equally samples S instance, then classifies for each instance sorter model, and the rule of judgement is：Only
There is an instance to have lesion point, then determines that sample bag is the picture for having lesion point；If all instance are
There is no lesion point, then determines sample bag for the picture of no lesion point；The samples pictures of each Zhang Xin are carried out determining step above
Suddenly, so training, until having detected all new samples.
Multi-instance learning method belongs to the scope of Weakly supervised study, is not have for training the instance of sorter model
Category label, but bags is to have category label, this point and previous all frames are not very identical；In training department
Point, each time when train classification models, the data set loaded all can resampling and more new data set D, with frequency of training
Increase, the nicety of grading of disaggregated model can also greatly improve；After the completion of training, model will be stored in local, convenient for it is subsequent
The accurate mark of sample lesion point is used in the field of medical imaging；Method and step and disaggregated model used are transplanted to medical treatment to set
In standby, different types of lesion point samples pictures are taken by medicine video camera, picture is zoomed to required by MIL algorithm
Picture size, and by program, all pixels value in image is read in the form of multi-dimensional matrix in the algorithm；It loads trained
The disaggregated model of completion, so that mass, rapid, precision detects and marks lesion point picture.
Using the video camera of Medical Devices as collector, different types of lesion point image is acquired, is therefrom selected several
High-visible image carries out disaggregated model by choosing sub-fraction semantic label image as model training data acquisition system
Initialization operation.Then, model training and save using the deep learning frame and platform of mainstream, finally using training
Model new image is detected and is marked, obtain the largely data sets that precisely mark；The present invention is not by variety classes disease
The influence of stove, mixing to a variety of lesions also can be carried out accurate mark, strong robustness；It is not only common for lesion point
The mark of shape, for some irregular, uncommon shape also can be carried out accurate mark, and universality is high；It is imaged by medicine
Head carries out image sampling, and utilizes trained model, automatic to detect lesion point image, to precisely be differentiated to new samples
And mark；Be conducive to detection and mark of the hospital for various lesion images, can greatly reduce professional technician for medicine
The workload of image labeling, while being also convenient for the lesion point that doctor quickly checked and diagnosed patient；It only needs to provide high resolution
Lesion picture, the accurate mark of lesion point in large quantity can be carried out, and speed is fast, it is at low cost, therefore have high effect
It is forthright.
Claims (5)
1. a kind of accurate mask method of medical image lesion point based on MIL, which is characterized in that include the following steps：
Step 1：Medical images data sets are acquired, positive sample collection and negative sample collection are divided into；
Step 2：Carry out the initialization of disaggregated model；
Step 3：S example is acquired for each sample standard deviation that positive sample is concentrated, each example is input in sorter model；
Record has the maximum example of lesion point probability value, and then the example is added in data set D；
Step 4：S example is acquired for each sample standard deviation that negative sample is concentrated, each example is input in sorter model；
Record most unlikely has the example of lesion point, is then added to the example in the data set D that step 3 obtains；
Step 5：The data set D that step 4 obtains is iterated training in neural network classifier model, obtains training mould
Type；
Step 6：Detection and the mark of new samples are carried out according to the model that step 5 obtains.
2. a kind of accurate mask method of medical image lesion point based on MIL according to claim 1, which is characterized in that
The disaggregated model uses the neural network model based on LeNet, including input layer, the first convolutional layer, the first pond layer, second
Convolutional layer, the second pond layer, the first full articulamentum, activation primitive layer, the second full articulamentum and for realizing classifying and normalize
Softmax layer.
3. a kind of accurate mask method of medical image lesion point based on MIL according to claim 1, which is characterized in that
Detection and new samples each in annotation process sample S example in the step 6, are divided with sorter model each example
Class；If having one and above example having lesion point, determine that the sample is to have lesion point diagram piece；Otherwise determine that the sample is disease-free
Stove point picture；Until having detected all new samples.
4. a kind of accurate mask method of medical image lesion point based on MIL according to claim 1, which is characterized in that
During repetitive exercise in the step 5, step 3 is repeated before training each time and step 4 obtains updated data set
D。
5. a kind of accurate mask method of medical image lesion point based on MIL according to claim 1, which is characterized in that
Initialization procedure is to choose the sample with label to carry out supervised learning in the step 2.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810677693.6A CN108922599A (en) | 2018-06-27 | 2018-06-27 | A kind of accurate mask method of medical image lesion point based on MIL |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810677693.6A CN108922599A (en) | 2018-06-27 | 2018-06-27 | A kind of accurate mask method of medical image lesion point based on MIL |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108922599A true CN108922599A (en) | 2018-11-30 |
Family
ID=64423841
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810677693.6A Pending CN108922599A (en) | 2018-06-27 | 2018-06-27 | A kind of accurate mask method of medical image lesion point based on MIL |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108922599A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110111340A (en) * | 2019-04-28 | 2019-08-09 | 南开大学 | The Weakly supervised example dividing method cut based on multichannel |
| CN110148192A (en) * | 2019-04-18 | 2019-08-20 | 上海联影智能医疗科技有限公司 | Medical image imaging method, device, computer equipment and storage medium |
| CN110969626A (en) * | 2019-11-27 | 2020-04-07 | 西南交通大学 | Method for extracting hippocampus of human brain nuclear magnetic resonance image based on 3D neural network |
| CN112634255A (en) * | 2020-12-30 | 2021-04-09 | 清华大学 | Method and device for establishing brain focus detection model and computer equipment |
| WO2021114630A1 (en) * | 2020-05-28 | 2021-06-17 | 平安科技（深圳）有限公司 | Medical image sample screening method, apparatus, computer device, and storage medium |
| CN114926396A (en) * | 2022-04-13 | 2022-08-19 | 四川大学华西医院 | Mental disorder magnetic resonance image preliminary screening model construction method |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106909784A (en) * | 2017-02-24 | 2017-06-30 | 天津大学 | Epileptic electroencephalogram (eeg) recognition methods based on two-dimentional time-frequency image depth convolutional neural networks |
| CN107563123A (en) * | 2017-09-27 | 2018-01-09 | 百度在线网络技术（北京）有限公司 | Method and apparatus for marking medical image |
| CN107680088A (en) * | 2017-09-30 | 2018-02-09 | 百度在线网络技术（北京）有限公司 | Method and apparatus for analyzing medical image |
| CN107784319A (en) * | 2017-09-26 | 2018-03-09 | 天津大学 | A kind of pathological image sorting technique based on enhancing convolutional neural networks |
-
2018
- 2018-06-27 CN CN201810677693.6A patent/CN108922599A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106909784A (en) * | 2017-02-24 | 2017-06-30 | 天津大学 | Epileptic electroencephalogram (eeg) recognition methods based on two-dimentional time-frequency image depth convolutional neural networks |
| CN107784319A (en) * | 2017-09-26 | 2018-03-09 | 天津大学 | A kind of pathological image sorting technique based on enhancing convolutional neural networks |
| CN107563123A (en) * | 2017-09-27 | 2018-01-09 | 百度在线网络技术（北京）有限公司 | Method and apparatus for marking medical image |
| CN107680088A (en) * | 2017-09-30 | 2018-02-09 | 百度在线网络技术（北京）有限公司 | Method and apparatus for analyzing medical image |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110148192A (en) * | 2019-04-18 | 2019-08-20 | 上海联影智能医疗科技有限公司 | Medical image imaging method, device, computer equipment and storage medium |
| CN110111340A (en) * | 2019-04-28 | 2019-08-09 | 南开大学 | The Weakly supervised example dividing method cut based on multichannel |
| CN110111340B (en) * | 2019-04-28 | 2021-05-14 | 南开大学 | Weak supervision example segmentation method based on multi-path segmentation |
| CN110969626A (en) * | 2019-11-27 | 2020-04-07 | 西南交通大学 | Method for extracting hippocampus of human brain nuclear magnetic resonance image based on 3D neural network |
| CN110969626B (en) * | 2019-11-27 | 2022-06-07 | 西南交通大学 | Method for extracting hippocampus of human brain nuclear magnetic resonance image based on 3D neural network |
| WO2021114630A1 (en) * | 2020-05-28 | 2021-06-17 | 平安科技（深圳）有限公司 | Medical image sample screening method, apparatus, computer device, and storage medium |
| CN112634255A (en) * | 2020-12-30 | 2021-04-09 | 清华大学 | Method and device for establishing brain focus detection model and computer equipment |
| CN112634255B (en) * | 2020-12-30 | 2022-12-02 | 清华大学 | Method and device for establishing brain focus detection model and computer equipment |
| CN114926396A (en) * | 2022-04-13 | 2022-08-19 | 四川大学华西医院 | Mental disorder magnetic resonance image preliminary screening model construction method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108922599A (en) | A kind of accurate mask method of medical image lesion point based on MIL | |
| Xue et al. | An application of transfer learning and ensemble learning techniques for cervical histopathology image classification | |
| CN106056595B (en) | Based on the pernicious assistant diagnosis system of depth convolutional neural networks automatic identification Benign Thyroid Nodules | |
| CN108257135A (en) | The assistant diagnosis system of medical image features is understood based on deep learning method | |
| Pan et al. | Mitosis detection techniques in H&E stained breast cancer pathological images: A comprehensive review | |
| CN109325942A (en) | Eye fundus image Structural Techniques based on full convolutional neural networks | |
| Huang et al. | Lesion-based contrastive learning for diabetic retinopathy grading from fundus images | |
| CN110910377B (en) | Cerebral infarction MRI image identification method based on neural network | |
| CN108549912A (en) | A kind of medical image pulmonary nodule detection method based on machine learning | |
| CN111369501B (en) | Deep learning method for identifying oral squamous cell carcinoma based on visual features | |
| CN116188479B (en) | Hip joint image segmentation method and system based on deep learning | |
| Wang et al. | Cataract detection based on ocular B-ultrasound images by collaborative monitoring deep learning | |
| CN110867242A (en) | Capsule endoscope image intelligent screening system | |
| Tang et al. | Lesion segmentation and RECIST diameter prediction via click-driven attention and dual-path connection | |
| CN111028230A (en) | Fundus image optic disc and macula lutea positioning detection algorithm based on YOLO-V3 | |
| Hasan et al. | Dental impression tray selection from maxillary arch images using multi-feature fusion and ensemble classifier | |
| CN114399634A (en) | Three-dimensional image classification method, system, device and medium based on weak supervised learning | |
| CN113011340B (en) | Cardiovascular operation index risk classification method and system based on retina image | |
| CN114140437A (en) | Fundus hard exudate segmentation method based on deep learning | |
| CN110070125A (en) | A kind of liver and gall surgical department's therapeutic scheme screening technique and system based on big data analysis | |
| CN113963199A (en) | Medical waste identification method based on multiple sensor feature fusion and machine learning | |
| Guo et al. | LLTO: towards efficient lesion localization based on template occlusion strategy in intelligent diagnosis | |
| Wang et al. | Optic disc detection based on fully convolutional neural network and structured matrix decomposition | |
| CN101609452A (en) | The fuzzy SVM feedback that is used for the medical image target identification is estimated method | |
| CN116012639A (en) | Quantitative index and staging method for retinal fundus image of premature infant based on meta-learning |
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 |
Application publication date: 20181130 |
|
| RJ01 | Rejection of invention patent application after publication |