CN115985472A - Fundus image labeling method and system based on neural network - Google Patents
Fundus image labeling method and system based on neural network Download PDFInfo
- Publication number
- CN115985472A CN115985472A CN202211524350.9A CN202211524350A CN115985472A CN 115985472 A CN115985472 A CN 115985472A CN 202211524350 A CN202211524350 A CN 202211524350A CN 115985472 A CN115985472 A CN 115985472A
- Authority
- CN
- China
- Prior art keywords
- image
- neural network
- fundus
- fundus image
- initial
- 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
Images
Landscapes
- Eye Examination Apparatus (AREA)
Abstract
The invention discloses a fundus image annotation method and system based on a neural network, wherein the method comprises the following steps: acquiring an initial fundus image of a diabetic patient, and extracting fundus image features from the initial fundus image by using a preset first neural network; searching similar images from a predicted online database based on the characteristics of the fundus images, wherein the preset online database consists of a plurality of fundus images subjected to data preprocessing; and performing focus labeling on the initial fundus image according to the similar image to obtain a labeled image. According to the invention, after the initial fundus image of the diabetic patient is obtained, the first neural network is utilized to extract image characteristics from the initial fundus image, similar images are inquired from the online database based on the image characteristics, and the focus marking is carried out on the initial fundus image according to the similar images for reference of a doctor, so that the diagnosis time consumption is shortened, and the diagnosis efficiency is improved.
Description
Technical Field
The invention relates to the technical field of image annotation, in particular to a fundus image annotation method and system based on a neural network.
Background
Diabetic retinopathy is a complication caused by diabetes mellitus. The common pathological features of the medicine include microangioma, hard exudation, cotton velvet spot, vitreous blood, new blood vessel, etc. The severity of diabetic retinopathy is classified into five grades (0-4) according to the international clinical grading standard. More severe diabetic retinopathy may cause visual impairment and blindness. Therefore, a diabetic needs to go to the eye to examine the fundus periodically in order to find the disease and slow or stop the progress of the disease in time.
The current common examination method is fundus photography examination, fundus lesions are recorded through images, and the images are sent to doctors for diagnosis and follow-up visit.
However, the currently used method has the following technical problems: the traditional examination mode needs a professional doctor to diagnose, is long in time consumption and high in cost, and in view of the current situations that medical resources are unevenly distributed and the number of the professional doctors is limited, the requirement for automatically diagnosing the diabetic retinopathy is increasingly high, and the traditional method cannot meet the application requirement of a user easily.
Disclosure of Invention
The invention provides a fundus image annotation method and system based on a neural network, wherein the method can be used for extracting image characteristics from an initial fundus image by using the neural network after the initial fundus image of a diabetic patient is obtained, inquiring a similar image from an online database based on the image characteristics, and performing focus annotation on the initial fundus image according to the similar image for reference of a doctor so as to shorten the diagnosis time consumption and improve the diagnosis efficiency.
The first aspect of the embodiments of the present invention provides a fundus image annotation method based on a neural network, the method including:
acquiring an initial fundus image of a diabetic patient, and extracting fundus image features from the initial fundus image by using a preset first neural network;
searching similar images from a predicted online database based on the characteristics of the fundus images, wherein the preset online database consists of a plurality of fundus images subjected to data preprocessing;
and performing focus labeling on the initial fundus image according to the similar image to obtain a labeled image.
In a possible implementation manner of the first aspect, the performing lesion labeling on the initial fundus image according to the similar image to obtain a labeled image includes:
predicting case information of the initial fundus image corresponding to the diabetic patient by using a preset second neural network, wherein the case information comprises: gender, age, smoking history, hypertension;
performing result correction on the similar image based on the case information to obtain a corrected image;
performing interpretable auxiliary diagnosis on the corrected image by adopting a class activation labeling technology to obtain a focus area of the image;
and performing activation mapping on the initial fundus image according to the focus region, and highlighting and labeling the focus region corresponding to the initial fundus image in a thermodynamic diagram mode to obtain a labeled image.
In a possible implementation manner of the first aspect, the performing the result correction on the similar image based on the case information to obtain a corrected image includes:
converting the case information into a correction coefficient;
calculating image similarity by using the correction coefficient;
and correcting the similar image by using the image similarity to obtain a corrected image.
In a possible implementation manner of the first aspect, the first neural network is an image processing model for image feature extraction;
the second neural network is an information processing model for information classification judgment.
In one possible implementation manner of the first aspect, the searching for a similar image from a predicted online database based on the fundus image feature includes:
calculating the feature similarity of each fundus image in the predicted on-line database and the features of the fundus images to obtain a plurality of feature similarities;
and screening the characteristic similarity with the maximum value from the plurality of characteristic similarities, and taking the fundus image corresponding to the characteristic similarity with the maximum value as a similar image.
In a possible implementation manner of the first aspect, the labeled image data set used for the preset first neural network training includes auxiliary labeled and artificially labeled image data;
the model training comprises the following steps:
sequentially carrying out data preprocessing, data enhancement, sample resampling and auxiliary labeling on the image data set to obtain a training data set;
and performing model training on the deep neural network by adopting the training data set to obtain a preset first neural network.
In one possible implementation manner of the first aspect, after the step of performing lesion labeling on the initial fundus image according to the similar image, the method further includes:
storing the marked image into a preset image database, and counting the number of images in the preset image database;
and if the number value of the image quantity is larger than a preset quantity threshold value, retraining the preset neural network.
A second aspect of an embodiment of the present invention provides a fundus image annotation system based on a neural network, the system including:
the extraction module is used for acquiring an initial fundus image of a diabetic patient and extracting fundus image features from the initial fundus image by using a preset first neural network;
the searching module is used for searching similar images from a predicted online database based on the characteristics of the fundus images, and the preset online database consists of a plurality of fundus images subjected to data preprocessing;
and the marking module is used for carrying out focus marking on the initial fundus image according to the similar image to obtain a marked image.
In a possible implementation manner of the second aspect, the labeling module is further configured to:
predicting case information of the initial fundus image corresponding to the diabetic patient by using a preset second neural network, wherein the case information comprises: gender, age, smoking history, hypertension;
performing result correction on the similar image based on the case information to obtain a corrected image;
performing interpretable auxiliary diagnosis on the corrected image by adopting a class activation labeling technology to obtain a focus area of the image;
and performing activation mapping on the initial fundus image according to the focus region, and highlighting and labeling the focus region corresponding to the initial fundus image in a thermodynamic diagram mode to obtain a labeled image.
In a possible implementation manner of the second aspect, the labeling module is further configured to:
converting the case information into a correction coefficient;
calculating image similarity by using the correction coefficient;
and correcting the similar image by using the image similarity to obtain a corrected image.
In one possible implementation manner of the second aspect, the first neural network is an image processing model for image feature extraction;
the second neural network is an information processing model for information classification judgment.
In a possible implementation manner of the second aspect, the search module is further configured to:
calculating the feature similarity of each fundus image in the predicted on-line database and the feature of the fundus image to obtain a plurality of feature similarities;
and screening the characteristic similarity with the maximum value from the plurality of characteristic similarities, and taking the fundus image corresponding to the characteristic similarity with the maximum value as a similar image.
In a possible implementation manner of the second aspect, the labeled image data set used by the preset first neural network training includes auxiliary labeled and artificially labeled image data;
the model training comprises the following steps:
sequentially carrying out data preprocessing, data enhancement, sample resampling and auxiliary labeling on the image data set to obtain a training data set;
and performing model training on the deep neural network by adopting the training data set to obtain a preset first neural network.
In a possible implementation manner of the second aspect, the system further includes:
the statistic module is used for storing the marked image into a preset image database and counting the number of images in the preset image database;
and the retraining module is used for retraining the preset neural network if the numerical value of the number of the images is greater than a preset number threshold value.
Compared with the prior art, the fundus image annotation method and system based on the neural network provided by the embodiment of the invention have the beneficial effects that: the invention can extract image characteristics from the initial fundus image by using the neural network after acquiring the initial fundus image of the diabetic, inquire similar images from an online database based on the image characteristics, and label focuses of the initial fundus image according to the similar images for reference of doctors, thereby shortening the diagnosis time consumption and improving the diagnosis efficiency.
Drawings
Fig. 1 is a schematic flowchart of a fundus image annotation method based on a neural network according to an embodiment of the present invention;
FIG. 2 is a flowchart illustrating an operation of a method for labeling fundus images based on a neural network according to an embodiment of the present invention;
fig. 3 is a schematic structural diagram of a fundus image annotation system based on a neural network according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Diabetic retinopathy is a complication caused by diabetes mellitus. The common pathological features of the medicine include microangioma, hard exudation, cotton velvet spot, vitreous blood, new blood vessel, etc. The severity of diabetic retinopathy is classified into five grades (0-4) according to the international clinical grading standard. More severe diabetic retinopathy may cause visual impairment and blindness. Therefore, a diabetic needs to go to the eye to examine the fundus periodically in order to find the disease and slow or stop the progress of the disease in time.
The current common examination method is fundus photography examination, fundus lesion conditions are recorded through images, and the images are sent to doctors for diagnosis and follow-up.
However, the conventional method has the following technical problems: the traditional examination mode needs a professional doctor to diagnose, is long in time consumption and high in cost, and in view of the current situations that medical resources are unevenly distributed and the number of the professional doctors is limited, the requirement for automatically diagnosing the diabetic retinopathy is increasingly high, and the traditional method cannot meet the application requirement of a user easily.
In order to solve the above problem, a fundus image annotation method based on a neural network provided in an embodiment of the present application will be described and explained in detail by the following specific embodiments.
Referring to fig. 1, a schematic flow chart of a fundus image annotation method based on a neural network according to an embodiment of the present invention is shown.
In one embodiment, the fundus image annotation method based on the neural network can be applied to a fundus image annotation system based on the neural network. The system can be installed and applied to a computer terminal.
By way of example, the fundus image labeling method based on the neural network may include:
s11, acquiring an initial fundus image of the diabetic, and extracting fundus image features from the initial fundus image by using a preset first neural network.
In one embodiment, the initial fundus image may be a fundus image of the diabetic retinopathy level of the diabetic patient.
The feature extraction may be to average and pool feature values of the last layer of the model convolution layer of the neural network to obtain image features.
Specifically, the feature extraction may be to extract features of the fundus picture using a trained deep neural network model such as an attention network model, an inclusion-v 3 model, a ResNet50 model as a feature extraction model.
Alternatively, feature data (feature vector) may be generated and stored in the fundus image feature database.
In an optional embodiment, the preset first neural network is a network for model training using an annotated image data set, wherein the annotated image data set includes auxiliary annotation and artificially annotated image data.
Wherein, as an example, the model training may comprise the following sub-steps:
and S111, sequentially carrying out data preprocessing and data enhancement on the image data set, resampling a sample and adding auxiliary labels to obtain a training data set.
And S112, performing model training on the deep neural network by adopting the training data set.
Specifically, the user may preset a database, where an image data set is formed by storing a training data set labeled by the user, an unlabeled training data set, offline feature data, and the like.
Wherein the labeled training data set may be an existing fundus image labeled with the degree of diabetic retinopathy. The picture source is a Kaggle data set. Wherein the data distribution in the training set is, no lesion: mild: medium: and (2) the severity: proliferative diabetic retinopathy =25810:2443:5292:873:708.
there are millions of fundus images in the unlabeled training dataset.
The offline feature data may be feature data of an existing marked fundus image stored.
In an alternative embodiment, the invention may assist in annotation. The auxiliary labeling can be to label the unlabeled training data set by using a trained model, namely to label the five-level label of the diabetic retinopathy.
The manual labeling module provides a non-labeling fundus image for a professional to label.
Before training, the data set needs to be processed, and the processing may include: and (3) preprocessing data, enhancing data, resampling a sample and adding auxiliary labels to obtain a training data set.
Specifically, the data preprocessing includes appropriate cropping, flipping, adding noise, changing the contrast, and the like of the labeled fundus image.
For example, the fundus image may be appropriately cropped, an unnecessary background portion removed, and then the missing value of the image may be complemented and noise may be eliminated.
Data enhancement adopts data enhancement technology to enhance data, and sample resampling is used for increasing data, for example, existing samples with fewer categories are repeatedly sampled or new samples are generated by using image generation technologies such as CycleGan and the like.
And adding auxiliary labels can add auxiliary labels to the data set by using the trained model.
After the above processing is completed, model training and feature extraction may be performed using the processed data set to extract features in the image.
Specifically, a pre-trained model EfficientNet-B0 can be adopted and developed based on PyTorch framework. The classification category is five categories, five grades of diabetic retinopathy. The model was trained using the Adam optimizer, cross entropy loss function. Model performance was evaluated based on the F1 score and accuracy.
And S12, searching similar images from a predicted online database based on the characteristics of the fundus images, wherein the preset online database consists of a plurality of fundus images subjected to data preprocessing.
A plurality of fundus images which are preprocessed through data are stored in the online database, the fundus images can be of previous patients, image annotation can be assisted through the fundus images of the previous patients, a doctor can conveniently perform follow-up diagnosis, and working efficiency is improved.
In an alternative embodiment, step S12 may comprise the following sub-steps:
and S121, calculating the feature similarity of each fundus image in the predicted on-line database and the feature of the fundus image to obtain a plurality of feature similarities.
And S122, screening the characteristic similarity with the maximum value from the plurality of characteristic similarities, and taking the fundus image corresponding to the characteristic similarity with the maximum value as a similar image.
Specifically, the euclidean distance may be used to calculate the similarity between the features of the input image and the features of the database, so as to obtain a plurality of feature similarities.
And then outputting N images of the fundus before ranking and related information of the images according to the similarity to form a result image set, wherein the image with the highest similarity is a similar image. Specifically, a Faiss framework can be adopted to construct feature indexes and sequences.
And S13, performing focus marking on the initial fundus image according to the similar image to obtain a marked image.
Because the similar image is similar to the initial fundus image, the focus position of the initial fundus image can be determined by referring to the focus position in the similar image, and then the focus marking is carried out according to the focus position, so that the doctor can conveniently carry out subsequent diagnosis, and the diagnosis efficiency is improved.
In one embodiment, step S13 may include the following sub-steps:
s131, predicting case information of the initial fundus image corresponding to the diabetic patient by using a preset second neural network, wherein the case information comprises: gender, age, smoking history, hypertension.
And S132, performing result correction on the similar image based on the case information to obtain a corrected image.
And S133, performing interpretable auxiliary diagnosis on the corrected image by adopting a class activation labeling technology to obtain a focus area of the image.
And S134, performing activation mapping on the initial fundus image according to the focus region, and highlighting and labeling the focus region corresponding to the initial fundus image in a thermodynamic diagram mode to obtain a labeled image.
Specifically, the preset prediction input image may correspond to medical record information of sex, age, smoking history, hypertension and the like of the patient, and then a diagnosis result of diabetic retinopathy classification of similar images, basic information of sex, age, smoking history, blood pressure and the like of the patient predicted by the system, and a lesion marking of the ocular fundus image of the eye pattern to be inquired are obtained.
And determining the position of the focus in the similar image according to the diagnosis result, and correcting the similar image according to the position of the focus.
In particular, the correction function may be used to correct the diabetic retinopathy level of similar images.
In one embodiment, step S132 may include:
and S1321, converting the case information into a correction coefficient.
And S1322, calculating the image similarity by adopting the correction coefficient.
And S1323, correcting the similar image by using the image similarity to obtain a corrected image.
Specifically, information such as sex, age, smoking history, hypertension, etc. may be converted into a correction coefficient, into: if the blood pressure is high, the correction coefficient is 2, and the hypotension is 1; age 10-20, correction factor of 2, age 20-30, correction factor of 3, and so on. The specific conversion mode can be adjusted according to actual needs.
Then, each correction coefficient can be used as a calculation weight and substituted into the correction function for calculation, so that the image similarity is obtained.
In one embodiment, the correction function may be expressed as:
image similarity = W1 × X1+ W2 × X2+ W3 × X3+ W4 × X4.
W1 is a correction coefficient of blood pressure (hypertension is 2, hypotension is 1), X1 is a blood pressure value, W2 is a correction coefficient of smoking history (smoking history is 1, no smoking history parameter is 0), X2 is a length of smoking history, W3 is a correction coefficient of gender (male is 1, female is 2), X3 is a constant, W4 is a correction coefficient of age, and X4 is an age value.
The image similarity can be calculated according to the formula, and then the similar images of the images are corrected according to the image similarity to obtain corrected images.
The specific correction method may be to correct the similar image according to the image similarity when the image similarity is smaller than a preset value, and the correction is color, size of a lesion area, and the like.
It should be noted that the neural network model responsible for prediction adopts an inclusion-v 3 neural network, and the two models are respectively trained for continuous value prediction and binary prediction. The continuous value prediction model is used for predicting age and blood pressure and evaluating the performance of the continuous value prediction model based on the average absolute error; the two-classification prediction model is used for predicting gender and smoking state, and the performance of the two-classification prediction model is evaluated based on the AUC value.
Then, the corrected image can be subjected to interpretable auxiliary diagnosis by using a class activation labeling technology to obtain a lesion region of the image.
Specifically, the feature data extracted from the fundus image may be subjected to activation mapping, and the lesion region may be highlighted in the image by means of thermodynamic diagram.
The image may also be roughly feature labeled using class activation thermodynamic diagrams.
Optionally, if the user is a professional physician, the physician may also determine whether there is a misdiagnosis in the diagnosis result according to the self-diagnosis experience, if so, the user inputs the misdiagnosis reason, the system records and corrects the misdiagnosis, and finally, the physician confirms the system correction result again.
Or after the labeling, the information of each medical record of the patient, the diabetic retinopathy label of the fundus image and the focus labeling of the fundus image can be integrated, and a diagnosis report is output, so that a doctor can conveniently check the diagnosis report.
Optionally, the fundus image input by the user can be labeled with lesion grading categories according to the diagnosis result and synchronized to the labeled data module
To further improve the accuracy of neural network labeling, the method may further comprise:
and S14, storing the marked image into a preset image database, and counting the number of images in the preset image database.
And S15, if the number value of the image quantity is larger than a preset quantity threshold value, retraining the preset neural network.
And storing each marked image into the database, retraining the model to improve the performance of the model after the number of the images in the database is increased to a certain number, and applying the trained model to the extraction of the eye fundus image features of the eye pattern to be inquired.
Meanwhile, after the model is updated, feature extraction can be carried out again based on the new model, and the updated offline feature database can be used for measuring the similarity of the eye pattern fundus images to be inquired.
The first neural network of the present invention is an image processing model for image feature extraction, and specifically may be used to extract image features and compare the image features in an image library with the image features.
The second neural network is an information processing model for information classification judgment, and can be specifically used for performing classification judgment according to image features and information, for example, judging whether hypertension exists or not, whether smoking history exists or not, and the like.
Referring to fig. 2, an operation flowchart of a fundus image labeling method based on a neural network according to an embodiment of the present invention is shown.
Specifically, annotated data sets (including manually annotated data sets and assisted-annotated data sets) may be prepared in advance; carrying out augmentation processing on the labeled data set, and carrying out model training on the neural network by using the processed data to obtain a labeled neural network; after obtaining the fundus image of the patient, extracting image characteristics and prediction information by using a neural network, searching for a similar image through the image characteristics, and correcting the similar image by using the prediction information; after correction, the image can be labeled and the pathological grade can be determined. And finally, sending the marked images and integrating various information of the patient to an intelligent terminal of a doctor or displaying the information on a screen of the system for the doctor to refer.
In this embodiment, an embodiment of the present invention provides a fundus image annotation method based on a neural network, which has the following beneficial effects: the invention can extract image characteristics from the initial fundus image by using the neural network after acquiring the initial fundus image of the diabetic, inquire similar images from an online database based on the image characteristics, and label focuses of the initial fundus image according to the similar images for reference of doctors, thereby shortening the diagnosis time consumption and improving the diagnosis efficiency.
The embodiment of the invention also provides a fundus image annotation system based on the neural network, and referring to fig. 3, a schematic structural diagram of the fundus image annotation system based on the neural network provided by the embodiment of the invention is shown.
Wherein, as an example, the fundus image annotation system based on the neural network may include:
the extraction module 301 is used for acquiring an initial fundus image of a diabetic patient and extracting a fundus image feature from the initial fundus image by using a preset first neural network;
a searching module 302, configured to search for a similar image from a predicted online database based on the fundus image features, where the preset online database is composed of a plurality of fundus images subjected to data preprocessing;
and the labeling module 303 is configured to perform lesion labeling on the initial fundus image according to the similar image to obtain a labeled image.
Optionally, the labeling module is further configured to:
predicting case information of the initial fundus image corresponding to a diabetic patient by using a preset second neural network, wherein the case information comprises: sex, age, smoking history, hypertension;
performing result correction on the similar image based on the case information to obtain a corrected image;
performing interpretable auxiliary diagnosis on the corrected image by adopting a class activation labeling technology to obtain a focus area of the image;
and performing activation mapping on the initial fundus image according to the focus region, and highlighting and labeling the focus region corresponding to the initial fundus image in a thermodynamic diagram mode to obtain a labeled image.
Optionally, the labeling module is further configured to:
converting the case information into a correction coefficient;
calculating image similarity by using the correction coefficient;
and correcting the similar image by using the image similarity to obtain a corrected image.
Optionally, the first neural network is an image processing model for image feature extraction;
the second neural network is an information processing model for information classification judgment.
Optionally, the searching module is further configured to:
calculating the feature similarity of each fundus image in the predicted on-line database and the feature of the fundus image to obtain a plurality of feature similarities;
and screening the characteristic similarity with the maximum value from the plurality of characteristic similarities, and taking the fundus image corresponding to the characteristic similarity with the maximum value as a similar image.
Optionally, the labeled image data set used for the preset first neural network training includes image data of auxiliary labeling and artificial labeling;
the model training comprises the following steps:
sequentially carrying out data preprocessing, data enhancement, sample resampling and auxiliary labeling on the image data set to obtain a training data set;
and performing model training on the deep neural network by adopting the training data set to obtain a preset first neural network.
Optionally, the system further comprises:
the statistical module is used for storing the annotation image into a preset image database and counting the number of images in the preset image database;
and the retraining module is used for retraining the preset neural network if the numerical value of the image quantity is greater than a preset quantity threshold value.
It can be clearly understood by those skilled in the art that, for convenience and brevity, the specific working process of the system described above may refer to the corresponding process in the foregoing method embodiment, and is not described herein again.
Further, an embodiment of the present application further provides an electronic device, including: the fundus image annotation method based on the neural network comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor executes the program to realize the fundus image annotation method based on the neural network according to the embodiment.
Further, the present application also provides a computer-readable storage medium storing computer-executable instructions for causing a computer to execute the method for fundus image annotation based on neural network as described in the above embodiment.
While the foregoing is directed to the preferred embodiment of the present invention, it will be understood by those skilled in the art that various changes and modifications may be made without departing from the spirit and scope of the invention.
Claims (10)
1. A fundus image labeling method based on a neural network is characterized by comprising the following steps:
acquiring an initial fundus image of a diabetic patient, and extracting fundus image features from the initial fundus image by using a preset first neural network;
searching similar images from a predicted online database based on the characteristics of the fundus images, wherein the preset online database consists of a plurality of fundus images subjected to data preprocessing;
and performing focus marking on the initial fundus image according to the similar image to obtain a marked image.
2. The fundus image labeling method based on the neural network according to claim 1, wherein said performing lesion labeling on the initial fundus image according to the similar image to obtain a labeled image comprises:
predicting case information of the initial fundus image corresponding to the diabetic patient by using a preset second neural network, wherein the case information comprises: gender, age, smoking history, hypertension;
performing result correction on the similar image based on the case information to obtain a corrected image;
performing interpretable auxiliary diagnosis on the corrected image by adopting a class activation labeling technology to obtain a focus area of the image;
and performing activation mapping on the initial fundus image according to the focus region, and highlighting and labeling the focus region corresponding to the initial fundus image in a thermodynamic diagram mode to obtain a labeled image.
3. The method for labeling a fundus image based on a neural network according to claim 2, wherein said performing a result correction on said similar image based on said case information to obtain a corrected image comprises:
converting the case information into a correction coefficient;
calculating image similarity by using the correction coefficient;
and correcting the similar image by using the image similarity to obtain a corrected image.
4. The fundus image annotation method based on the neural network according to claim 2, characterized in that said first neural network is an image processing model of image feature extraction;
the second neural network is an information processing model for information classification judgment.
5. The method for fundus image annotation based on neural network of claim 1, wherein said finding a similar image from a predicted online database based on said fundus image characteristics comprises:
calculating the feature similarity of each fundus image in the predicted on-line database and the feature of the fundus image to obtain a plurality of feature similarities;
and screening the characteristic similarity with the maximum value from the plurality of characteristic similarities, and taking the fundus image corresponding to the characteristic similarity with the maximum value as a similar image.
6. The fundus image annotation method based on the neural network according to claim 1, wherein the annotated image data set used by the preset first neural network training contains image data of auxiliary annotation and artificial annotation;
the model training comprises the following steps:
sequentially carrying out data preprocessing, data enhancement, sample resampling and auxiliary labeling on the image data set to obtain a training data set;
and performing model training on the deep neural network by adopting the training data set to obtain a preset first neural network.
7. A neural network-based fundus image annotation method according to any one of claims 1-6, wherein after said step of performing lesion annotation on said initial fundus image based on said similar image, said method further comprises:
storing the marked image into a preset image database, and counting the number of images in the preset image database;
and if the number value of the image quantity is larger than a preset quantity threshold value, retraining the preset neural network.
8. A neural network-based fundus image annotation system, the system comprising:
the extraction module is used for acquiring an initial fundus image of a diabetic patient and extracting fundus image characteristics from the initial fundus image by using a preset first neural network;
the searching module is used for searching similar images from a predicted online database based on the characteristics of the fundus images, and the preset online database consists of a plurality of fundus images subjected to data preprocessing;
and the labeling module is used for performing focus labeling on the initial fundus image according to the similar image to obtain a labeled image.
9. An electronic device, comprising: a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the method of neural network based fundus image annotation of any one of claims 1 to 7 when executing the program.
10. A computer-readable storage medium storing computer-executable instructions for causing a computer to perform the neural network-based fundus image annotation method according to any one of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211524350.9A CN115985472B (en) | 2022-12-01 | 2022-12-01 | Fundus image labeling method and fundus image labeling system based on neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211524350.9A CN115985472B (en) | 2022-12-01 | 2022-12-01 | Fundus image labeling method and fundus image labeling system based on neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115985472A true CN115985472A (en) | 2023-04-18 |
| CN115985472B CN115985472B (en) | 2023-09-22 |
Family
ID=85965581
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211524350.9A Active CN115985472B (en) | 2022-12-01 | 2022-12-01 | Fundus image labeling method and fundus image labeling system based on neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115985472B (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107506770A (en) * | 2017-08-17 | 2017-12-22 | 湖州师范学院 | Diabetic retinopathy eye-ground photography standard picture generation method |
| CN108470359A (en) * | 2018-02-11 | 2018-08-31 | 艾视医疗科技成都有限公司 | A kind of diabetic retinal eye fundus image lesion detection method |
| CN110084252A (en) * | 2019-04-29 | 2019-08-02 | 南京星程智能科技有限公司 | Diabetic retinopathy image labeling method based on deep learning |
| CN110490236A (en) * | 2019-07-29 | 2019-11-22 | 武汉工程大学 | Automatic image marking method, system, device and medium neural network based |
| CN110706233A (en) * | 2019-09-30 | 2020-01-17 | 北京科技大学 | Retina fundus image segmentation method and device |
| CN111291765A (en) * | 2018-12-07 | 2020-06-16 | 北京京东尚科信息技术有限公司 | Method and device for determining similar pictures |
| CN111753861A (en) * | 2019-03-28 | 2020-10-09 | 香港纺织及成衣研发中心有限公司 | Automatic image annotation system and method for active learning |
| CN113793301A (en) * | 2021-08-19 | 2021-12-14 | 首都医科大学附属北京同仁医院 | Training method of fundus image analysis model based on dense convolution network model |
-
2022
- 2022-12-01 CN CN202211524350.9A patent/CN115985472B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107506770A (en) * | 2017-08-17 | 2017-12-22 | 湖州师范学院 | Diabetic retinopathy eye-ground photography standard picture generation method |
| CN108470359A (en) * | 2018-02-11 | 2018-08-31 | 艾视医疗科技成都有限公司 | A kind of diabetic retinal eye fundus image lesion detection method |
| CN111291765A (en) * | 2018-12-07 | 2020-06-16 | 北京京东尚科信息技术有限公司 | Method and device for determining similar pictures |
| CN111753861A (en) * | 2019-03-28 | 2020-10-09 | 香港纺织及成衣研发中心有限公司 | Automatic image annotation system and method for active learning |
| CN110084252A (en) * | 2019-04-29 | 2019-08-02 | 南京星程智能科技有限公司 | Diabetic retinopathy image labeling method based on deep learning |
| CN110490236A (en) * | 2019-07-29 | 2019-11-22 | 武汉工程大学 | Automatic image marking method, system, device and medium neural network based |
| CN110706233A (en) * | 2019-09-30 | 2020-01-17 | 北京科技大学 | Retina fundus image segmentation method and device |
| CN113793301A (en) * | 2021-08-19 | 2021-12-14 | 首都医科大学附属北京同仁医院 | Training method of fundus image analysis model based on dense convolution network model |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115985472B (en) | 2023-09-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Liu et al. | A framework of wound segmentation based on deep convolutional networks | |
| US11645753B2 (en) | Deep learning-based multi-site, multi-primitive segmentation for nephropathology using renal biopsy whole slide images | |
| US20220198214A1 (en) | Image recognition method and device based on deep convolutional neural network | |
| CN111986211A (en) | Deep learning-based ophthalmic ultrasonic automatic screening method and system | |
| WO2021114817A1 (en) | Oct image lesion detection method and apparatus based on neural network, and medium | |
| WO2023155488A1 (en) | Fundus image quality evaluation method and device based on multi-source multi-scale feature fusion | |
| CN117392470B (en) | Fundus image multi-label classification model generation method and system based on knowledge graph | |
| CN112132801A (en) | Lung bullae focus detection method and system based on deep learning | |
| Zhang et al. | MRMR optimized classification for automatic glaucoma diagnosis | |
| CN117058676B (en) | Blood vessel segmentation method, device and system based on fundus examination image | |
| CN113222064A (en) | Image target object real-time detection method, system, terminal and storage medium | |
| CN111524093A (en) | Intelligent screening method and system for abnormal tongue picture | |
| CN115578783A (en) | Device and method for identifying eye diseases based on eye images and related products | |
| CN115206478A (en) | Medical report generation method and device, electronic equipment and readable storage medium | |
| CN111383222A (en) | Intervertebral disc MRI image intelligent diagnosis system based on deep learning | |
| CN117010971B (en) | Intelligent health risk providing method and system based on portrait identification | |
| CN117557840A (en) | Fundus lesion grading method based on small sample learning | |
| CN116580801A (en) | Ultrasonic inspection method based on large language model | |
| CN115985472B (en) | Fundus image labeling method and fundus image labeling system based on neural network | |
| CN115170492A (en) | Intelligent prediction and evaluation system for postoperative vision of cataract patient based on AI (artificial intelligence) technology | |
| Ashtari-Majlan et al. | Deep learning and computer vision for glaucoma detection: A review | |
| CN115471512A (en) | Medical image segmentation method based on self-supervision contrast learning | |
| CN114330484A (en) | Method and system for classification and focus identification of diabetic retinopathy through weak supervision learning | |
| CN112634279B (en) | Medical image semantic segmentation method based on attention Unet model | |
| Mazumder et al. | Deep learning approaches for diabetic retinopathy detection by image 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |