CN210167123U

CN210167123U - Screening system for fundus image lesions

Info

Publication number: CN210167123U
Application number: CN201921469212.9U
Authority: CN
Inventors: 胡志钢; 陈志�; 陈意; 白玉婧
Original assignee: Shenzhen Sibionics Intelligent Technology Co Ltd
Current assignee: Shenzhen Sibionics Intelligent Technology Co Ltd
Priority date: 2018-12-26
Filing date: 2019-09-04
Publication date: 2020-03-20
Anticipated expiration: 2029-09-04
Also published as: CN109411086A

Abstract

The utility model provides a screening system of fundus image pathological change, a serial communication port, include: a storage module deployed in a cloud server; the acquisition module is used for acquiring more than two fundus images from different eyes of the same person and uploading the fundus images to the storage module; the input module is used for inputting the information of the detected person corresponding to the fundus image, associating the information with the fundus image and uploading the information of the detected person to the storage module; and the screening module is used for extracting the fundus image and the information of the examinee from the storage module and automatically judging whether the associated fundus image has the lesion or not by using a deep learning method based on an artificial neural network so as to generate a screening result. According to the utility model discloses, can utilize the degree of depth learning method based on artificial neural network to assist the doctor to carry out the judgement that pathological change discernment was carried out to the fundus image to can improve the screening efficiency of fundus image pathological change effectively.

Description

Screening system for fundus image lesions

Technical Field

The utility model relates to a screening system, in particular to screening system of fundus image pathological change.

Background

Diabetes is a group of metabolic diseases characterized by hyperglycemia. Hyperglycemia is due to defects in insulin secretion or impaired biological action. Chronic hyperglycemia can lead to chronic damage or dysfunction of various tissues, particularly the eye, kidneys, heart, blood vessels, etc., especially damage to the eye. Statistically, blood vessels in the eyes of a patient with diabetes for many years are damaged, and in severe cases, eyeground bleeding may be caused to blur the vision of the patient or completely lose the vision.

At present, more than 3000 thousands of diabetic retinopathy (abbreviated as 'diabetes mellitus') patients in China exist, more than 80 percent of diabetic patients can suffer from the diabetes mellitus, and the blindness risk of the diabetic retinopathy is 25 times higher than that of normal people. While the diabetic retinopathy is an eye disease that can avoid blindness, and retinopathy is not obvious in an early stage, if fundus examination can be performed regularly at the initial stage of onset, the risk of blindness due to diabetic retinopathy can be reduced by 90% or more. Therefore, early screening, early diagnosis and early treatment are the key points for vision preservation of patients with the diabetic network.

However, the screening rate of the diabetes mellitus in China is less than 10%. In addition, the sugar network screening system has many problems, for example, the intelligent sugar network disease analysis software depends on computer hardware equipment, the carrying and the transferring are inconvenient for basic level screening, and each screening equipment or screening point needs to be independently configured with the intelligent sugar network disease analysis software and the computer hardware equipment, so that the cost is high. Moreover, for a huge diabetes group, the number of screening doctors is seriously insufficient, the manual interpretation method is time-consuming and labor-consuming, the subjectivity is strong, and misdiagnosis and other conditions are easy to occur.

Disclosure of Invention

The present invention has been made in view of the above-mentioned prior art, and an object thereof is to provide a fundus image lesion screening system capable of effectively improving the screening efficiency of fundus image lesions.

Therefore, the utility model provides a screening system of fundus image pathological change, it includes: a storage module deployed in a cloud server; the acquisition module is used for acquiring more than two fundus images from different eyes of the same person and uploading the fundus images to the storage module, wherein the fundus images comprise a target fundus image and a reference fundus image; the input module is used for inputting the information of the detected person corresponding to the fundus image, associating the information with the fundus image and uploading the information of the detected person to the storage module; and a screening module which automatically judges whether the associated fundus image has a lesion or not by using a deep learning method based on an artificial neural network to generate a screening result, wherein the artificial neural network comprises a convolutional neural network structure for processing the fundus image and an output layer for receiving the high-level features output by the convolutional neural network structure and the information of the examinee.

The utility model discloses in, can make things convenient for operating personnel to gather and upload the eye ground image in the place of difference through the cloud ware to be favorable to carrying out eye ground image pathological change examination on a large scale. Moreover, the screening result is generated by interpreting the fundus image by using the deep learning method based on the artificial neural network, so that the screening efficiency of fundus image lesions can be effectively improved.

In addition, in the screening system for fundus image lesions, optionally, the screening system further comprises an output module for outputting the screening result, wherein the output module further outputs the screening result into a report form and uploads the report form to the storage module. Therefore, doctors and users can read the screening results conveniently.

In addition, in the screening system for fundus image lesions according to the present invention, optionally, the subject information includes: name, gender, Identification (ID), and past medical history. This makes it possible to integrate effective information such as subject information and to determine more accurately whether or not a fundus image is diseased.

Additionally, in the fundus image lesion screening system of the present invention, optionally, the fundus image includes at least a optic disc and a macular region of the fundus. Thereby, it is possible to make a determination as to whether or not a lesion has occurred from the effective information of the fundus image.

Additionally, in the fundus image lesion screening system of the present invention, optionally, the acquisition module is a hand-held fundus camera free of mydriasis. Therefore, the portable collecting device can be conveniently carried by an operator or a collector.

Further, in the fundus image lesion screening system according to the present invention, optionally, the convolutional neural network structure includes an input layer that receives the target fundus image and the reference fundus image, respectively, and a plurality of convolution kernels that process data output from the input layer. This enables more efficient processing of the fundus image.

Additionally, in the fundus image pathological examination system of the utility model relates to, optionally, the screening module still will the screening result is uploaded to the storage module. Thus, the screening system can query the storage module for screening results and further process the results.

Additionally, in the fundus image lesion screening system of the present invention, optionally, the screening module is disposed in the cloud server. Therefore, the fundus images can be automatically screened through the cloud server.

Additionally, in the fundus image pathological examination system of the utility model relates to, optionally, the acquisition module still includes and is used for judging acquireing whether qualified judgement unit of fundus image. Therefore, the collected fundus images can be preliminarily judged, and operators can conveniently collect correct fundus images.

Further, in the screening system for fundus image lesions according to the present invention, optionally, the fundus images include two left eye fundus images and two right eye fundus images from the same person. In this case, whether or not a lesion occurs in the fundus image can be determined more accurately.

According to the utility model discloses, can utilize the degree of depth learning method based on artificial neural network to assist the doctor to carry out the judgement that pathological change discernment was carried out to the fundus image to can improve the screening efficiency of fundus image pathological change effectively.

Drawings

Fig. 1 is a system block diagram schematically illustrating an example of a fundus image lesion screening system according to an embodiment of the present invention.

Fig. 2 is a system block diagram schematically illustrating another example of the screening system for fundus image lesions according to the embodiment of the present invention.

Fig. 3 is a block diagram schematically illustrating an acquisition module of a fundus image lesion screening system according to an embodiment of the present invention.

Fig. 4 is a block diagram schematically illustrating an entry module of the fundus image lesion screening system according to the embodiment of the present invention.

Fig. 5 is a block diagram schematic diagram illustrating a screening module of a fundus image lesion screening system according to an embodiment of the present invention.

Fig. 6 is a schematic block diagram illustrating a screening unit of a fundus image lesion screening system according to an embodiment of the present invention.

Fig. 7 is a schematic diagram showing an example of a fundus image according to an embodiment of the present invention.

Fig. 8 is a schematic diagram showing an example of an output result report of the fundus image lesion screening system according to the embodiment of the present invention.

Description of reference numerals:

the system comprises a 1 … screening system, a 2 … cloud server, a 10 … storage module, a 20 … acquisition module, a 21 … acquisition unit, a 22 … judgment unit, a 30 … entry module, a 31 … entry unit, a 32 … verification unit, a 33 … association unit, a 40 … screening module, a 41 … preprocessing unit, a 42 … screening unit, a 420 … convolutional neural network structure, a 421 … input layer, a 422 … plurality of convolutional kernels, a 430 … output layer and a 50 … output module.

Detailed Description

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the following description, the same components are denoted by the same reference numerals, and redundant description thereof is omitted. The drawings are schematic and the ratio of the dimensions of the components and the shapes of the components may be different from the actual ones.

It is noted that the terms "comprises," "comprising," and "having," and any variations thereof, in the present disclosure, such that a process, method, system, article, or apparatus that comprises or has a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include or have other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

In addition, the headings and the like in the following description of the present invention are not intended to limit the scope of the present invention, but are merely provided as a reminder for reading. Such a subtitle should not be understood as a means for segmenting article content, nor should the content under the subtitle be limited to only the scope of the subtitle.

Fig. 1 is a system block diagram schematically showing an example of a fundus image lesion screening system 1 according to an embodiment of the present invention. Fig. 2 is a system block diagram schematically illustrating another example of the fundus image lesion screening system 1 according to the embodiment of the present invention.

A fundus image lesion screening system 1 according to an embodiment of the present invention may include: a storage module 10, an acquisition module 20, an entry module 30, and a screening module 40 (see FIG. 1). In the fundus image lesion screening system 1 according to the present embodiment, as shown in fig. 1, the storage module 10 may be disposed in the cloud server 2; the acquisition module 20 may be configured to acquire two or more fundus images from different eyes of the same person, and may upload the fundus images to the storage module 10, where the fundus images may include a target fundus image and a reference fundus image; the entry module 30 may be configured to enter subject information corresponding to the fundus image, associate the subject information with the fundus image, and upload the subject information to the storage module 10; and the screening module 40 may automatically interpret whether the associated fundus image has a lesion using an artificial neural network-based deep learning method to generate a screening result, the artificial neural network may include a convolutional neural network structure 420 that processes the fundus image, and an output layer 430 that receives the high-level features and subject information output by the convolutional neural network structure 420.

When the fundus image lesion screening system 1 according to the present embodiment is used, an acquirer may acquire a fundus image and upload the acquired fundus image to the storage module 10 using a fundus image acquisition device such as a fundus camera, may enter examinee information simultaneously during acquisition of the fundus image, or may enter the examinee information in advance, upload the examinee information to the storage module 10 disposed in the cloud server 2, and then automatically interpret the information by the screening module 40 of the artificial neural network-based deep learning method, and generate a screening result. Therefore, the fundus image lesion screening method can be beneficial to screening fundus image lesions in a wide range. Moreover, the screening result is generated by interpreting the fundus image by using the deep learning method based on the artificial neural network, so that the screening efficiency of fundus image lesions can be effectively improved.

Further, in some examples, the screening system 1 for fundus image lesions may further include an output module 50 (see fig. 2) for outputting a screening result.

(memory Module 10)

In some examples, the memory module 10 may be a non-volatile memory. In some examples, the Memory module 10 may be a Flash Memory (Flash Memory). However, the present embodiment is not limited to this, and the storage module 10 may be, for example: a ferroelectric random access memory (FeRAM), a Magnetic Random Access Memory (MRAM), a phase change random access memory (PRAM), or a Resistive Random Access Memory (RRAM). This can reduce the possibility of data loss due to sudden power outage.

In other examples, the storage module 10 may also be other types of readable storage media, such as: Read-Only Memory (ROM), Random Access Memory (RAM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), One-time Programmable Read-Only Memory (OTPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), compact-Read-Only Memory (CD-ROM) or other optical disk storage, magnetic disk storage, tape storage, or any other medium readable by a computer that can be used to carry or store data. Thus, an appropriate memory can be selected according to different situations.

As described above, the storage module 10 may be provided within the cloud server 2. Through cloud ware 2, can make things convenient for operating personnel to gather and upload the eye ground image in the place of difference to be favorable to carrying out eye ground image pathological change screening on a large scale. Wherein the cloud server 2 may be leased. This can reduce the maintenance cost of the server. In other examples, the cloud server 2 may be self-built. In this case, the storage module 10 may be provided in a server built by itself, so as to ensure confidentiality of data and prevent leakage of data from a client or a patient.

(obtaining module 20)

Fig. 3 is a block diagram schematically illustrating the acquisition module 20 of the fundus image lesion screening system 1 according to the embodiment of the present invention.

In the present embodiment, the acquisition module 20 may be used to acquire two or more fundus images from different eyes of the same person, and the fundus images may include a target fundus image and a reference fundus image.

In some examples, the obtaining module 20 may include an obtaining unit 21 and a judging unit 22 (see fig. 3). In some examples, the acquisition module 20 may include an acquisition unit 21 that acquires a fundus image and a determination unit 22 that determines whether the fundus image acquired by the acquisition unit 21 is acceptable.

In some examples, the acquisition module 20 may be a handheld fundus camera. Therefore, the portable fundus oculi imaging device can be conveniently carried by an operator or a collecting person, and the operator can conveniently acquire fundus oculi images of a detected person under various conditions. Further, the acquisition module 20 may preferably be a mydriatic-free hand-held fundus camera. Thereby, the fundus image of the subject can be acquired without the need to mydriasis (e.g., drug mydriasis) the subject's eye. Furthermore, in some examples, the acquisition module 20 may also be a desktop fundus camera, preferably a non-mydriatic desktop fundus camera.

Additionally, as shown in fig. 5, in some examples, the fundus image may include a disc and macular region of the fundus. Among them, optic disc is also called optic nerve disc. In other examples, the fundus image may also include arteries and veins of various regions of the fundus. Therefore, whether the fundus is diseased or not can be accurately judged by observing the fundus image.

In some examples, the acquisition unit 21 of the acquisition module 20 may acquire two or more fundus images from different eyes of the same person. Therefore, the reliability of the obtained fundus image is ensured, and the probability of screening errors can be reduced.

In other examples, the fundus images may also include two left eye fundus images and two right eye fundus images from the same person. In this case, it is possible to analyze one of the two left-eye images as a target fundus image, the other as a reference fundus image, and the two right-eye fundus images as reference fundus images. Therefore, the diagnosis process of a doctor can be simulated, and the target image is judged by referring to other fundus images from the same person, so that the accuracy rate of judging the lesion of the fundus images is improved, the fundus images can be mutually referred, the calculated amount is controlled in the minimum range, and the screening efficiency is improved. The two left-eye images (two right-eye images) may be the same fundus image. At this time, the same fundus image is used as both the main analysis target fundus image and the reference fundus image.

In other examples, the fundus images obtained by the acquisition module 20 may also include a plurality of (more than two) fundus images from the same person. In this case, any one of them can be used as a main analysis target fundus image, and the other images can be used as reference images. Thus, whether or not a fundus image is diseased can be determined more accurately. Further, an equal number of fundus images from both the left and right eyes can be used.

In some examples, the fundus image acquired by the acquisition module 20 is not particularly limited, and may be a color image such as an RGB image, a grayscale image, or the like.

In some examples, the fundus image obtained by the acquisition module 20 may be uploaded into the storage module 10, that is, the fundus image may be uploaded into the cloud server 2. Therefore, the images obtained by the acquisition module 20 can be stored or analyzed in time, and meanwhile, the acquisition can be carried out anytime and anywhere, and the acquisition becomes more flexible.

In the present embodiment, the determination unit 22 may be configured to determine whether or not the obtained fundus image is acceptable. In this case, the acquired fundus image can be judged in advance as being acceptable or not by the judgment unit 22. For example, whether the acquired fundus image is clear, whether the angle meets the requirements, and the like. Therefore, the collected fundus images can be preliminarily judged, and operators can conveniently collect correct and effective fundus images.

(recording module 30)

Fig. 4 is a block diagram schematically illustrating the entry module 30 of the fundus image lesion screening system 1 according to the embodiment of the present invention.

In the present embodiment, the entry module 30 may be used to enter subject information corresponding to a fundus image and be associated with the fundus image.

In some examples, the logging module 30 may include a logging unit 31, a verification unit 32, and an association unit 33. In some examples, the entry unit 31 may enter subject information corresponding to a fundus image. The verification unit 32 may verify the information of the subject, such as mismatch of the input identification card information, error of the input bit number, and the like. The association unit 33 may associate the subject information with the fundus image, and thereby may generate an image ID number or the like associated with the fundus image.

In some examples, the entry module 30 may be a cell phone APP. Therefore, the information can be conveniently and rapidly input by the detection personnel or the detected person at any time and any place. However, the present embodiment is not limited thereto, and the input module 30 may also be a computer, a touch screen device or other input devices. In some examples, the subject information may include at least one of name, gender, Identification (ID), past medical history, and the like. Thus, the screening system 1 can more accurately determine whether or not a lesion has occurred in the fundus image based on the basic information of the subject, such as the past medical history.

In addition, in some examples, the entered subject information may be information entered in advance by the subject himself, for example, over the internet. In this case, the time spent by the inspection person on inputting information can be reduced, and the screening system 1 can reserve a serial number for the subject after inputting information in advance, and the inspection person can directly associate the input subject information with the acquired fundus image, greatly improving the work efficiency. In other examples, the entered information may also be retrieved from other databases, such as a database of a hospital or a database of a public security agency. Therefore, the accuracy and reliability of the recorded information can be ensured.

In addition, in some examples, the entry module 30 may also verify whether the data of the subject information is correct through data in other databases. Therefore, the situations of error recording and error recording are avoided.

In some examples, the subject information entered by the entry module 30 may be associated with the fundus image acquired by the acquisition module 20 by the association unit 33. In this case, more reliable determination results can be obtained by combining the basic information of the subject with the fundus image.

In some examples, the subject information entered by the entry module 30 may be uploaded to the storage module 10, that is, to the cloud server 2. Thereby, the screening system 1 is enabled to be associated with the fundus image from the same person in the cloud. In some examples, the subject information entered by the entry module 30 may be collectively uploaded into the storage module 10 after being associated with the acquired fundus image. Thus, the step that the screening system 1 needs to correlate the acquired fundus image with the information of the examinee at the cloud end is eliminated, and the operation efficiency of the screening system 1 is improved.

(screening module 40)

Fig. 5 is a block diagram schematically illustrating the screening module 40 of the fundus image lesion screening system 1 according to the embodiment of the present invention. Fig. 6 is a block diagram schematically illustrating the screening unit 42 of the fundus image lesion screening system 1 according to the embodiment of the present invention. Fig. 7 is a schematic diagram showing an example of a fundus image according to an embodiment of the present invention. Fig. 8 is a schematic diagram showing an example of an output result report of the fundus image lesion screening system 1 according to the embodiment of the present invention.

In the present embodiment, the screening module 40 may be configured to extract the fundus image and the subject information from the storage module 10, and automatically interpret whether the associated fundus image has a lesion by using an artificial neural network-based deep learning method to generate a screening result.

In some examples, screening module 40 may include a pre-processing unit 41 and a screening unit 42 (see fig. 5). In some examples, the pre-processing unit 41 may pre-process the fundus image, such as grayscale processing, normalization, and the like. The screening unit 42 may automatically interpret the pre-processed fundus image using an artificial neural network-based deep learning method.

In some examples, the artificial neural network described above can include a convolutional neural network structure 420, an output layer 430, where the convolutional neural network structure 420 can include an input layer 421 and a plurality of convolutional kernels 422 (e.g., including convolutional kernel 442a, convolutional kernel 442b, convolutional kernel 442c, etc.).

(see FIG. 6)

In some examples, the convolutional neural network structure 420 may be used to process the fundus image to extract high-level features of the fundus image. Specifically, the input layer 421 may be configured to receive the fundus image processed by the preprocessing unit 41, the plurality of convolution kernels 422 may be configured to process the fundus image from the input layer 421 and extract high-level features of the fundus image, and the output layer 430 may be configured to receive the high-level feature set and the subject information. In addition, in some examples, the output layer 430 may also interpret whether or not a lesion exists in the fundus image based on the above-described high-level feature set and subject information. For example, in some examples, when the fundus image has a lesion, the output layer 430 outputs a result of 1 (100%); when the fundus image has no lesion, the output layer 430 outputs the result 0 (0%).

In some examples, the screening module 40 may also perform data amplification on the fundus image pair (including the target fundus image and the reference fundus image) during training when processing the fundus images using the artificial neural network-based deep learning method to increase the amount of data samples for training, thereby increasing the accuracy of determining fundus lesions. Therefore, the influence of human factors can be reduced, the accuracy rate of screening the fundus image is improved, and the efficiency is improved.

In some examples, the preprocessing unit 41 may perform preprocessing such as fundus region detection, image cropping, resizing, normalization, and the like on the fundus image. Therefore, the screening unit 42 can conveniently interpret the fundus image.

In other examples, the screening module 40 may determine the type of fundus lesions that may be detected by identifying image features of the acquired fundus images, such as ① retinal vascular disease, ② acquired maculopathy-related disease, ③ inflammatory disease including non-infectious systemic and infectious (viral, bacterial, fungal) diseases, and other diseases, ④ fundus dystrophy including retinal dystrophy, vitreoretinopathy, and choroidal dystrophy, ⑤ retinal detachment, ⑥ tumors, ⑦ acquired optic nerve disease, ⑧ congenital abnormalities, and the like, all of which may manifest an end-blind, e.g., premature discovery, premature treatment, and a risk of blindness to the patient.

In some examples, the screening module 40 may classify screening results of fundus images as being diseased and non-diseased. Further, in some examples, the screening module 40 may further classify the screening results of the fundus images into a lesion grade and a disease free grade of the fundus lesion. Further, in other examples, the screening results of the screening module 40 on the fundus images may also be classified into types and kinds corresponding to fundus lesions and disease free. Under the condition, the reason and the type of the fundus lesion can be directly obtained through the screening result, so that the subsequent examination process of detection personnel is greatly reduced, the trouble that the patient needs to be repeatedly examined is reduced, and the continuous tracking and examination of medical personnel on the patient are facilitated.

In some examples, the screening results obtained by screening module 40 may be uploaded directly into storage module 10. Thus, screening system 1 can recall the screening results from storage module 10 and further process the results.

In some examples, screening module 40 may be deployed in cloud server 2. In this case, the screening module 40 may directly screen the fundus image stored in the cloud server 2, and the time taken for data transmission can be reduced.

(output module 50)

As shown in FIG. 8, in some examples, the results obtained by screening module 40 may be output in the form of a report by output module 50. Therefore, doctors and users can conveniently inquire the screening results. In some examples, the output module 50 may also upload the report to the storage module 10 after outputting the screening results in a report form. Thus, a user can query the storage module 10 for a screening report.

Additionally, in some examples, output module 50 may retrieve screening result data from storage module 10.

While the present invention has been described in detail in connection with the drawings and the examples, it is to be understood that the above description is not intended to limit the present invention in any way. The present invention may be modified and varied as necessary by those skilled in the art without departing from the true spirit and scope of the invention, and all such modifications and variations are intended to be included within the scope of the invention.

Claims

1. A system for screening fundus image lesions is characterized in that,

the method comprises the following steps:

a storage module deployed in a cloud server;

the acquisition module is used for acquiring more than two fundus images from different eyes of the same person and uploading the fundus images to the storage module, wherein the fundus images comprise a target fundus image and a reference fundus image;

the input module is used for inputting the information of the detected person corresponding to the fundus image, associating the information with the fundus image and uploading the information of the detected person to the storage module; and

a screening module for automatically judging whether the associated fundus image has a lesion by using a deep learning method based on an artificial neural network to generate a screening result, wherein the artificial neural network comprises a convolution neural network structure for processing the fundus image and an output layer for receiving the high-level characteristics output by the convolution neural network structure and the information of the examinee.

2. The screening system of claim 1,

the output module is used for outputting the screening result, and outputting the screening result into a report form and uploading the report form to the storage module.

3. The screening system of claim 1,

the subject information includes: name, sex, identification code, and past medical history.

4. The screening system of claim 1,

the fundus image includes at least a disc and macular region of the fundus.

5. The screening system of claim 1,

the acquisition module is a hand-held eye fundus camera free of mydriasis.

6. The screening system of claim 1,

the convolutional neural network structure includes an input layer that receives the target fundus image and the reference fundus image, respectively, and a plurality of convolution kernels that process data output from the input layer.

7. The screening system of claim 1,

the screening module also uploads the screening results to the storage module.

8. The screening system of claim 1,

the screening module is deployed in the cloud server.

9. The screening system of claim 1,

the acquisition module further includes a judgment unit for judging whether the acquired fundus image is qualified.

10. The screening system of claim 1,

the fundus images include two left eye fundus images and two right eye fundus images from the same person.