CN114723674A - Glaucoma auxiliary screening system based on decoupling training and reasoning - Google Patents
Glaucoma auxiliary screening system based on decoupling training and reasoning Download PDFInfo
- Publication number
- CN114723674A CN114723674A CN202210257364.2A CN202210257364A CN114723674A CN 114723674 A CN114723674 A CN 114723674A CN 202210257364 A CN202210257364 A CN 202210257364A CN 114723674 A CN114723674 A CN 114723674A
- Authority
- CN
- China
- Prior art keywords
- fundus image
- training
- module
- neural network
- glaucoma
- 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
- 238000012549 training Methods 0.000 title claims abstract description 54
- 208000010412 Glaucoma Diseases 0.000 title claims abstract description 43
- 238000012216 screening Methods 0.000 title claims abstract description 37
- 238000007781 pre-processing Methods 0.000 claims abstract description 24
- 238000013528 artificial neural network Methods 0.000 claims abstract description 15
- 208000024891 symptom Diseases 0.000 claims abstract description 12
- 238000003062 neural network model Methods 0.000 claims description 40
- 238000012545 processing Methods 0.000 claims description 20
- 238000000034 method Methods 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 8
- 238000012360 testing method Methods 0.000 claims description 6
- 238000012795 verification Methods 0.000 claims description 6
- 230000015556 catabolic process Effects 0.000 claims description 4
- 238000006731 degradation reaction Methods 0.000 claims description 4
- 230000008034 disappearance Effects 0.000 claims description 4
- 230000004927 fusion Effects 0.000 claims description 4
- 238000002372 labelling Methods 0.000 claims description 4
- 230000008602 contraction Effects 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 3
- 230000002265 prevention Effects 0.000 abstract 1
- 230000006870 function Effects 0.000 description 14
- 238000010586 diagram Methods 0.000 description 5
- 201000010099 disease Diseases 0.000 description 4
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 230000004913 activation Effects 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000004438 eyesight Effects 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 230000002207 retinal effect Effects 0.000 description 2
- 201000004569 Blindness Diseases 0.000 description 1
- 206010061323 Optic neuropathy Diseases 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000012790 confirmation Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000007850 degeneration Effects 0.000 description 1
- 230000003111 delayed effect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000002059 diagnostic imaging Methods 0.000 description 1
- 230000002427 irreversible effect Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000004126 nerve fiber Anatomy 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 210000003733 optic disk Anatomy 0.000 description 1
- 208000020911 optic nerve disease Diseases 0.000 description 1
- 238000012014 optical coherence tomography Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 210000001525 retina Anatomy 0.000 description 1
- 210000003994 retinal ganglion cell Anatomy 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000009978 visual deterioration Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
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
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B3/00—Apparatus for testing the eyes; Instruments for examining the eyes
- A61B3/10—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions
- A61B3/12—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions for looking at the eye fundus, e.g. ophthalmoscopes
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B3/00—Apparatus for testing the eyes; Instruments for examining the eyes
- A61B3/10—Objective types, i.e. instruments for examining the eyes independent of the patients' perceptions or reactions
- A61B3/14—Arrangements specially adapted for eye photography
-
- 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
-
- 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
- G06N3/08—Learning methods
-
- 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
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30041—Eye; Retina; Ophthalmic
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Theoretical Computer Science (AREA)
- Medical Informatics (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Public Health (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Heart & Thoracic Surgery (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Veterinary Medicine (AREA)
- Computational Linguistics (AREA)
- Animal Behavior & Ethology (AREA)
- Artificial Intelligence (AREA)
- Surgery (AREA)
- General Engineering & Computer Science (AREA)
- Ophthalmology & Optometry (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Primary Health Care (AREA)
- Epidemiology (AREA)
- Pathology (AREA)
- Databases & Information Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a glaucoma auxiliary screening system based on decoupling training and reasoning, which comprises a control module, a fundus image acquisition module, a fundus image preprocessing module, a neural network reasoning module and an output module, wherein the control module is used for acquiring a fundus image; the fundus image acquisition module is used for acquiring a fundus image of a user; the fundus image preprocessing module is used for preprocessing the acquired fundus image; the neural network reasoning module is used for classifying and screening the fundus images; the output module is used for judging whether the user has glaucoma symptoms or not according to the classified screening result and giving a treatment suggestion; the control module is used for controlling the fundus image acquisition, the preprocessing, the classification screening and the result output of each module. The invention improves the mobility of the glaucoma screening system, meets the requirement of large-scale glaucoma screening, enables the glaucoma screening to be anytime and anywhere, has low cost and low threshold, and has important significance for early prevention of glaucoma.
Description
Technical Field
The invention relates to the field of artificial intelligence and medical imaging, in particular to a glaucoma auxiliary screening system based on decoupling training and reasoning.
Background
Glaucoma is an optic neuropathy characterized by progressive degeneration of retinal ganglion cells, visual field loss and visual deterioration, and is one of the main causes of irreversible blindness in China. According to statistics, the number of glaucoma patients in the country is as many as 1582 thousands, and scholars predict that the number of glaucoma patients in the world will reach 1 hundred million 1182 thousands in 2040 years.
Glaucoma is found in time in the early stage and reasonable treatment means are adopted, so that most patients can stop the progress of the disease and keep certain eyesight. However, the disease often has no obvious symptoms in the early stage, and many patients cannot be diagnosed until the late stage, so that treatment is delayed and vision is lost.
Clinical examination methods for the retinal retina include fundus images, optical coherence tomography, and the like. The fundus images have the advantages of low cost, simplicity in operation and strong mobility, but the fundus images are collected through a fundus camera clinically, and the traditional fundus camera is large in size and not suitable for large-scale screening. The fundus images need to be diagnosed by a professional doctor, the change of the forms of the fundus optic disk and the retinal nerve fiber layer in the fundus images of the glaucoma patients at the early stage is difficult to be distinguished by naked eyes, and misdiagnosis is easy to occur if the quality of the acquired fundus images is not high enough.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a glaucoma auxiliary screening system based on decoupling training and reasoning by combining the actual requirements of glaucoma screening.
In order to achieve the above object, on one hand, the invention provides a glaucoma auxiliary screening system based on decoupling training and reasoning, which comprises a control module, and a fundus image acquisition module, a fundus image preprocessing module, a neural network reasoning module and an output module which are respectively connected with the control module;
the fundus image acquisition module is used for acquiring images of the fundus of the user;
the fundus image preprocessing module is used for preprocessing the acquired fundus images so as to enhance the image quality and enable the neural network reasoning module to identify the image information more easily;
the neural network reasoning module is used for classifying and screening the preprocessed fundus images;
the output module is used for judging whether the user has glaucoma symptoms or not according to the classification screening result and giving treatment suggestions when the user has the glaucoma symptoms;
the control module is used for controlling the fundus image acquisition, preprocessing, classification screening and result output of the modules.
The fundus image preprocessing module comprises a color processing unit, a fundus image space exchange unit and an image enhancement processing unit;
the color processing unit is used for extracting RBG characteristics of the fundus images and adjusting the brightness, contrast, saturation and sharpness of the images;
the fundus image space exchange unit is used for cutting, rotating, turning and zooming fundus images;
the fundus image preprocessing module comprises a color processing unit, a fundus image space exchange unit and an image enhancement processing unit;
the color processing unit is used for extracting RBG characteristics of the fundus images and adjusting the brightness, contrast, saturation and sharpness of the images;
the fundus image space exchange unit is used for cutting, rotating, overturning and scaling the fundus image;
the image enhancement processing unit is used for carrying out noise reduction processing on the image so as to improve the image quality.
The fundus image acquisition module acquires a fundus image of a user through a fundus image acquisition camera.
The construction process of the neural network model is as follows:
step 1, acquiring a large number of fundus image samples by fundus image acquisition equipment, and labeling and classifying the acquired fundus image samples;
step 2, preprocessing the fundus image sample;
and 3, training the neural network model by utilizing the preprocessed fundus image samples, decoupling the training model and the reasoning model by using a structural parameterization algorithm of the obtained neural network model, and compressing to obtain the target neural network model.
In the step 1, a large number of fundus image samples including fundus images of glaucoma patients with different symptom degrees and fundus images of normal people are collected, after the fundus image samples are labeled, all the image samples are divided into three types, namely training samples, test samples and verification samples, wherein the training samples account for half of the total number of samples, and the test samples and the verification samples respectively account for one fourth of the total number of samples.
The training process of the neural network model in the step 3 is as follows:
s31, training a neural network by using the preprocessed fundus image data, wherein the training model adopts a multi-branch contraction type neural network model to realize fusion of multi-scale features, improve the feature utilization rate and simultaneously prevent the problems of gradient disappearance, degradation and feature loss along with the deepening of the network layer number;
s32, the loss function uses a cross entropy function, and after the convergence of the loss function, the training is finished to obtain a trained neural network model;
and S33, using the expanded structural parameterization algorithm to equate the trained neural network model parameters into a simple single-branch neural network model for reasoning.
The loss function in step S32 uses a cross entropy function, the expression of which is
Wherein p is the actual value of the sample, and q is the predicted value.
The structural parameterization algorithm in the step S33 obtains the beneficial effect of the invention according to the linear operation rule of convolution operation
The glaucoma auxiliary screening system based on the decoupling training and reasoning adopts the multi-branch shrinkage type neural network model for training, and the multi-scale features are mined and fused in the training process, so that the feature utilization rate is improved. After training is finished, the parameters of the multi-branch contractive neural network model are equivalently transformed into a single-branch simple neural network model by using and expanding a structural reparameterization algorithm for reasoning, so that the complexity of the neural network model is reduced while the accuracy of the neural network model is ensured, and the operation speed is accelerated. The portable glaucoma auxiliary screening system based on the multi-branch shrinkage network capable of decoupling training and reasoning adopts a deep learning method, and the strong characteristic extraction capability of the system is utilized to solve the problem that early glaucoma symptoms are not obvious and diagnosis omission is easily caused, so that non-professionals can use the system to screen glaucoma.
Drawings
FIG. 1 is a schematic diagram of the components of a decoupled training and reasoning based glaucoma screening system of the present invention;
FIG. 2 is a flow chart of the present invention for constructing a neural network model;
FIG. 3 is a flow chart of the fundus image sample pre-processing in an embodiment of the present invention;
FIG. 4 is a diagram illustrating the effect of partial color processing according to an embodiment of the present invention;
FIG. 5 is a diagram illustrating the effect of partial spatial transformation according to an embodiment of the present invention;
FIG. 6 is a flow chart of neural network model training and structure reparameterization in an embodiment of the present invention;
FIG. 7 is a schematic diagram of a TrapNet network structure according to an embodiment of the present invention;
FIG. 8 is a schematic structural diagram of a neural network after structural parameterization processing according to an embodiment of the present invention;
fig. 9 is a flowchart of a glaucoma screening system according to the present invention applied to follow-up monitoring of a patient.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, 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 some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without any inventive step, are within the scope of the present invention.
Example (b): see fig. 1-9.
As shown in fig. 1, the glaucoma auxiliary screening system based on decoupling training and reasoning of the present invention comprises a control module, and a fundus image acquisition module, a fundus image preprocessing module, a neural network reasoning module, and an output module, which are respectively connected to the control module;
the fundus image acquisition module is used for acquiring a fundus image of a user;
the fundus image preprocessing module is used for preprocessing the acquired fundus images so as to enhance the image quality and enable the neural network reasoning module to identify the image information more easily;
the neural network reasoning module is used for classifying and screening the preprocessed fundus images;
the output module is used for judging whether the user has glaucoma symptoms or not according to the classification screening result and giving treatment suggestions when the user has the glaucoma symptoms;
the control module is used for controlling the fundus image acquisition, the preprocessing, the classification screening and the result output of each module.
As shown in fig. 3, the fundus image preprocessing module includes a color processing unit, a fundus image space exchanging unit, and an image enhancement processing unit;
as shown in fig. 4, the color processing unit is used for RBG feature extraction and image brightness, contrast, saturation, sharpness adjustment of the fundus image;
as shown in fig. 5, the fundus image space exchanging unit is used to crop, rotate, flip, and zoom a fundus image;
the image enhancement processing unit is used for carrying out noise reduction processing on the image so as to improve the image quality.
The fundus image acquisition module acquires a fundus image of a user through a fundus image acquisition camera.
As shown in fig. 2, the neural network model is constructed as follows:
step 1, acquiring a large number of fundus image samples by fundus image acquisition equipment, and labeling and classifying the acquired fundus image samples;
step 2, preprocessing the fundus image sample;
and 3, training the neural network model by utilizing the preprocessed fundus image samples, decoupling the training model and the reasoning model by using a structural parameterization algorithm of the obtained neural network model, and compressing to obtain the target neural network model.
In the step 1, a large number of fundus image samples including fundus images of glaucoma patients with different symptom degrees and fundus images of normal people are collected, after the fundus image samples are labeled, all the image samples are divided into three types, namely training samples, test samples and verification samples, wherein the training samples account for half of the total number of samples, and the test samples and the verification samples respectively account for one fourth of the total number of samples.
As shown in fig. 6, the training process of the neural network model in step 3 is as follows:
s31, training a neural network by using the preprocessed fundus image data, wherein the training model adopts a multi-branch contraction type neural network model to realize fusion of multi-scale features, improve the feature utilization rate and simultaneously prevent the problems of gradient disappearance, degradation and feature loss along with the deepening of the network layer number;
s32, using a cross entropy function as a loss function, and finishing training after the loss function is converged to obtain a trained neural network model;
and S33, using the expanded structural parameterization algorithm to equate the trained neural network model parameters into a simple single-branch neural network model for reasoning.
Specifically, in order to realize the fusion of multi-scale features and improve the feature utilization rate, a shallow-layer network block Trap V1 comprises a plurality of branches containing 1x1 convolutions and different numbers of 3x3 convolutions (wherein the effect of two continuous 3x3 convolutions can be equivalent to 5x5 convolution, and the effect of three continuous 3x3 convolutions can be equivalent to 7x7 convolution), so as to extract features of different scales, each branch uses 1x1 convolution compression channels first, the network efficiency is prevented from being influenced by the number of excessively thick channels, uniform data distribution of a BN layer (Batch Normalization) is introduced behind each convolution module, and the training speed is accelerated. And each network block is added with a jump connection, so that the gradient is transmitted while the low-level characteristics are kept, and the problems of disappearance of the gradient and degradation along with the deepening of the network layer number are prevented. As the network layer number increases, the receptive fields of a plurality of continuous 3x3 convolutions are larger, and characteristics are easily lost, so that three continuous 3x3 convolutions are discarded in Trap V2, two continuous 3x3 convolutions are discarded in Trap V3, and the network structure of Trap net is shown in fig. 7.
The loss function in step S32 uses a cross entropy function, an expression of which
Wherein p is the actual value of the sample, and q is the predicted value.
Specifically, in the process of back propagation, if the initial derivative of the activation function is small, the update amplitude of the whole gradient is small, the convergence speed is slow, and the gradient of the cross entropy loss function for the last layer is not related to the derivative of the activation function and is only related to the difference between the output value and the true value, so that the update speed of the weight is accelerated by using the cross entropy loss function, and the convergence speed is further accelerated.
The structural reparameterization algorithm in step S33 is obtained according to the linear operation rule of convolution operation.
Specifically, for the structure reparameterization algorithm, there are:
for an NxN convolution with an input channel A and an output channel B, the parameters are a fourth-order tensor We RB ×A×N×NWith a bias of b ∈ RB. Assuming that its input is I ∈ RA×H×WIts output is O ∈ RB×H′×W′Copy bias b to REP (b) ε RB×H′×W′As used herein
To represent a convolution operator, the NxN can be expressed as
The output at (h, w) of the ith channel is
The formula shows that the convolution operation is actually a linear operation, and the linear operation conforms to the combination law and the distribution law, so that the combination law and the distribution law of the convolution operation can be expressed in the following forms
Therefore, the method of the structure reparameterization is provided by utilizing the linear operation rule of the convolution operation.
Further, after the decoupling training and reasoning process of the structural parameterization, the neural network model is divided into a training model and a reasoning model, and the structure of the whole network model is shown in fig. 8. The training model first uses a Trap V1 network block containing multi-scale branches, and gradually uses Trap V2 and Trap V3 with smaller receptive fields as the layer number grows deeper to prevent feature loss. After the training is finished, the TrapNet module is equivalent to the convolution of 3x3 by using a structural parameterization method, so that the parameter quantity is greatly reduced, and the operation cost during reasoning is reduced.
As shown in fig. 9, which is a flowchart of the glaucoma screening system applied to the follow-up monitoring of the patient, when the user is diagnosed with glaucoma by the system, the staff will recommend the user to go to the hospital for confirmation, if the user is further diagnosed in the hospital, the staff will build files according to the detection condition, and store the cup-to-disk ratio of the user calculated from the fundus image into the file, and at the same time, the user who builds the files regularly takes fundus images, and judge whether the disease condition of the user deteriorates by comparing the cup-to-disk ratio recorded each time, and if the disease condition deteriorates, recommend the user to go to the hospital for treatment, thereby saving the time and examination cost of the user.
In conclusion, the invention utilizes the multi-branch contractive neural network model for training, excavates and fuses multi-scale features in the training process, improves the feature utilization rate, equivalently transforms the parameters of the multi-branch contractive neural network model into the single-branch simple neural network model for reasoning by using and expanding the structure reparameterization algorithm after the training is finished, reduces the complexity of the neural network model while ensuring the precision of the neural network model, and accelerates the model operation speed
The above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (8)
1. A glaucoma auxiliary screening system based on decoupling training and reasoning is characterized by comprising a control module, and a fundus image acquisition module, a fundus image preprocessing module, a neural network reasoning module and an output module which are respectively connected with the control module;
the fundus image acquisition module is used for acquiring a fundus image of a user;
the fundus image preprocessing module is used for preprocessing the acquired fundus images so as to enhance the image quality and enable the neural network reasoning module to identify the image information more easily;
the neural network reasoning module is used for classifying and screening the preprocessed fundus images;
the output module is used for judging whether the user has glaucoma symptoms or not according to the classification screening result and giving treatment suggestions when the user has the glaucoma symptoms;
the control module is used for controlling the fundus image acquisition, preprocessing, classification screening and result output of the modules.
2. The glaucoma-aided screening system based on decoupleable training and reasoning according to claim 1, wherein the fundus image preprocessing module comprises a color processing unit, a fundus image space exchanging unit and an image enhancement processing unit;
the color processing unit is used for extracting RBG characteristics of the fundus images and adjusting the brightness, contrast, saturation and sharpness of the images;
the fundus image space exchange unit is used for cutting, rotating, overturning and scaling the fundus image;
the image enhancement processing unit is used for carrying out noise reduction processing on the image so as to improve the image quality.
3. The glaucoma auxiliary screening system based on decoupled training and reasoning of claim 1 wherein the fundus image capturing module captures a fundus image of the user via a fundus image capturing camera.
4. The decoupled training and reasoning based glaucoma auxiliary screening system of claim 3 wherein said neural network model is constructed as follows:
step 1, acquiring a large number of fundus image samples by fundus image acquisition equipment, and labeling and classifying the acquired fundus image samples;
step 2, preprocessing the fundus image sample;
and 3, training the neural network model by utilizing the preprocessed fundus image samples, decoupling the training model and the reasoning model by using a structural parameterization algorithm of the obtained neural network model, and compressing to obtain the target neural network model.
5. The glaucoma auxiliary screening system based on the decoupleable training and reasoning of claim 6, wherein a large number of fundus image samples including fundus images of glaucoma patients with different symptom degrees and fundus images of normal persons are collected in step 1, and after labeling of the fundus image samples, all the image samples are divided into three types, namely training samples, testing samples and verification samples, wherein the training samples account for half of the total samples, and the testing samples and the verification samples each account for one fourth of the total samples.
6. The glaucoma-aided screening system based on decoupleable training and reasoning according to claim 6, wherein the training process for the neural network model in step 3 is as follows:
s31, training a neural network by using the preprocessed fundus image data, wherein the training model adopts a multi-branch contraction type neural network model to realize fusion of multi-scale features, improve the feature utilization rate and simultaneously prevent the problems of gradient disappearance, degradation and feature loss along with the deepening of the network layer number;
s32, using a cross entropy function as a loss function, and finishing training after the loss function is converged to obtain a trained neural network model;
and S33, using the expanded structural parameterization algorithm to equate the trained neural network model parameters into a simple single-branch neural network model for reasoning.
7. The decoupled training and reasoning based glaucoma aiding screening system of claim 6 wherein the loss function in step S32 uses a cross entropy function expressed as:
wherein p is the actual value of the sample, and q is the predicted value.
8. The decoupled training and reasoning based glaucoma auxiliary screening system of claim 6 wherein said structural reparameterization algorithm of step S33 is derived from the linear operation rule of convolution operation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210257364.2A CN114723674A (en) | 2022-03-16 | 2022-03-16 | Glaucoma auxiliary screening system based on decoupling training and reasoning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210257364.2A CN114723674A (en) | 2022-03-16 | 2022-03-16 | Glaucoma auxiliary screening system based on decoupling training and reasoning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114723674A true CN114723674A (en) | 2022-07-08 |
Family
ID=82238646
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210257364.2A Pending CN114723674A (en) | 2022-03-16 | 2022-03-16 | Glaucoma auxiliary screening system based on decoupling training and reasoning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114723674A (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111476283A (en) * | 2020-03-31 | 2020-07-31 | 上海海事大学 | Glaucoma fundus image identification method based on transfer learning |
| CN113077889A (en) * | 2021-03-31 | 2021-07-06 | 武开寿 | Artificial intelligence ophthalmopathy screening service method and system |
| CN113657124A (en) * | 2021-07-14 | 2021-11-16 | 内蒙古工业大学 | Multi-modal Mongolian Chinese translation method based on circulation common attention Transformer |
| CN114021603A (en) * | 2021-10-25 | 2022-02-08 | 哈尔滨工程大学 | Radar signal modulation mode identification method based on model reparameterization |
-
2022
- 2022-03-16 CN CN202210257364.2A patent/CN114723674A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111476283A (en) * | 2020-03-31 | 2020-07-31 | 上海海事大学 | Glaucoma fundus image identification method based on transfer learning |
| CN113077889A (en) * | 2021-03-31 | 2021-07-06 | 武开寿 | Artificial intelligence ophthalmopathy screening service method and system |
| CN113657124A (en) * | 2021-07-14 | 2021-11-16 | 内蒙古工业大学 | Multi-modal Mongolian Chinese translation method based on circulation common attention Transformer |
| CN114021603A (en) * | 2021-10-25 | 2022-02-08 | 哈尔滨工程大学 | Radar signal modulation mode identification method based on model reparameterization |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Li et al. | Automatic detection of diabetic retinopathy in retinal fundus photographs based on deep learning algorithm | |
| Ran et al. | Cataract detection and grading based on combination of deep convolutional neural network and random forests | |
| Haloi | Improved microaneurysm detection using deep neural networks | |
| Khan et al. | Cataract detection using convolutional neural network with VGG-19 model | |
| CN109345538A (en) | A kind of Segmentation Method of Retinal Blood Vessels based on convolutional neural networks | |
| Nasir et al. | Deep DR: detection of diabetic retinopathy using a convolutional neural network | |
| Hassan et al. | Exploiting the transferability of deep learning systems across multi-modal retinal scans for extracting retinopathy lesions | |
| Bilal et al. | Diabetic retinopathy detection using weighted filters and classification using CNN | |
| Wu et al. | Automatic cataract detection with multi-task learning | |
| Triyadi et al. | Deep learning in image classification using vgg-19 and residual networks for cataract detection | |
| CN117338234A (en) | Diopter and vision joint detection method | |
| Yamuna et al. | Detection of abnormalities in retinal images | |
| CN117426748A (en) | MCI detection method based on multi-mode retina imaging | |
| Khan et al. | Screening fundus images to extract multiple ocular features: A unified modeling approach | |
| Nair et al. | Multi-labelled ocular disease diagnosis enforcing transfer learning | |
| Brancati et al. | Segmentation of pigment signs in fundus images for retinitis pigmentosa analysis by using deep learning | |
| Ali et al. | Cataract disease detection used deep convolution neural network | |
| Patra et al. | Diabetic Retinopathy Detection using an Improved ResNet50-InceptionV3 Structure | |
| Calderon et al. | CNN-based quality assessment for retinal image captured by wide field of view non-mydriatic fundus camera | |
| CN114723674A (en) | Glaucoma auxiliary screening system based on decoupling training and reasoning | |
| Deepa et al. | Pre-Trained Convolutional Neural Network for Automated Grading of Diabetic Retinopathy | |
| Kazi et al. | Processing retinal images to discover diseases | |
| Nguyen et al. | Cataract Detection using Hybrid CNN Model on Retinal Fundus Images | |
| Anggraeni et al. | Detection of the emergence of exudate on the image of retina using extreme learning machine method | |
| Suwandi et al. | A Systematic Literature Review: Diabetic Retinopathy Detection Using Deep 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 |