CN114595747A - Intelligent identification system for malignant tumors of eyelids based on photos - Google Patents
Intelligent identification system for malignant tumors of eyelids based on photos Download PDFInfo
- Publication number
- CN114595747A CN114595747A CN202210156943.8A CN202210156943A CN114595747A CN 114595747 A CN114595747 A CN 114595747A CN 202210156943 A CN202210156943 A CN 202210156943A CN 114595747 A CN114595747 A CN 114595747A
- Authority
- CN
- China
- Prior art keywords
- eyelid
- tumor
- module
- data set
- malignant
- 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
- 210000000744 eyelid Anatomy 0.000 title claims abstract description 79
- 201000011510 cancer Diseases 0.000 title claims abstract description 34
- 208000004413 Eyelid Neoplasms Diseases 0.000 claims abstract description 184
- 206010050497 Eyelid tumour Diseases 0.000 claims abstract description 184
- 230000004614 tumor growth Effects 0.000 claims abstract description 30
- 238000000034 method Methods 0.000 claims abstract description 21
- 230000003211 malignant effect Effects 0.000 claims description 61
- 206010028980 Neoplasm Diseases 0.000 claims description 54
- 208000025865 Ulcer Diseases 0.000 claims description 50
- 230000036269 ulceration Effects 0.000 claims description 50
- 238000012549 training Methods 0.000 claims description 47
- 230000012010 growth Effects 0.000 claims description 25
- 230000008569 process Effects 0.000 claims description 18
- 238000012360 testing method Methods 0.000 claims description 18
- 238000013135 deep learning Methods 0.000 claims description 17
- 230000036210 malignancy Effects 0.000 claims description 17
- 230000008676 import Effects 0.000 claims description 11
- 208000027439 benign eyelid neoplasm Diseases 0.000 claims description 9
- 208000026043 eyelid cancer Diseases 0.000 claims description 9
- 230000006870 function Effects 0.000 claims description 9
- 238000006073 displacement reaction Methods 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 6
- 230000004807 localization Effects 0.000 claims description 5
- 238000003745 diagnosis Methods 0.000 claims description 4
- 230000001575 pathological effect Effects 0.000 claims description 3
- 238000009966 trimming Methods 0.000 claims description 3
- 238000010586 diagram Methods 0.000 description 6
- 208000035250 cutaneous malignant susceptibility to 1 melanoma Diseases 0.000 description 2
- 208000011061 eyelid melanoma Diseases 0.000 description 2
- 201000001441 melanoma Diseases 0.000 description 2
- 238000004393 prognosis Methods 0.000 description 2
- 201000008407 sebaceous adenocarcinoma Diseases 0.000 description 2
- 210000001519 tissue Anatomy 0.000 description 2
- 206010004146 Basal cell carcinoma Diseases 0.000 description 1
- 206010018691 Granuloma Diseases 0.000 description 1
- 206010027476 Metastases Diseases 0.000 description 1
- 208000007256 Nevus Diseases 0.000 description 1
- 208000009077 Pigmented Nevus Diseases 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 208000031513 cyst Diseases 0.000 description 1
- 230000006378 damage Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 210000001508 eye Anatomy 0.000 description 1
- 208000024519 eye neoplasm Diseases 0.000 description 1
- 230000001815 facial effect Effects 0.000 description 1
- 201000011066 hemangioma Diseases 0.000 description 1
- 230000008595 infiltration Effects 0.000 description 1
- 238000001764 infiltration Methods 0.000 description 1
- 230000003902 lesion Effects 0.000 description 1
- 230000009401 metastasis Effects 0.000 description 1
- 206010061289 metastatic neoplasm Diseases 0.000 description 1
- 201000008106 ocular cancer Diseases 0.000 description 1
- 210000003695 paranasal sinus Anatomy 0.000 description 1
- 230000004083 survival effect Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000004382 visual function Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention provides an eyelid malignant tumor intelligent identification system based on a photo, and belongs to the technical field of eyelid tumor image identification. The intelligent identification system for the malignant tumors of the eyelids comprises: the device comprises a picture importing module, an eyelid tumor positioning module, an eyelid tumor growth condition judging module and a result outputting module. The method solves the problems that the identification of partial good and malignant tumors of the eyelid is difficult and the early identification rate of the malignant tumors of the eyelid is low at present, and has the advantages of medical resource saving and high reliability.
Description
Technical Field
The invention relates to the technical field of eyelid tumor image recognition, in particular to an eyelid malignant tumor intelligent recognition system based on photos.
Background
Eyelid tumors are a common type of ocular tumor that occurs well in middle-aged and elderly women. Eighty percent of eyelid tumors are benign, as: nevi, macular tumors, cysts, granulomas, hemangiomas, and the like. While the remaining twenty percent of eyelid tumors are malignant, including: basal cell carcinoma, sebaceous gland carcinoma, malignant melanoma, metastatic tumor and the like, and at present, the cause of malignant tumor of eyelid is not clear.
Eyelid malignancies pose a significant threat to visual function, facial appearance, and even life by invading tissues adjacent to the eye, brain, and paranasal sinuses. Early detection and treatment of eyelid malignancies can prevent severe tissue structural damage and achieve the most cosmetically and functionally satisfactory results. Furthermore, although eyelid melanoma and sebaceous adenocarcinoma are rare lesions, they can metastasize distally early and therefore have a high mortality rate. However, if these highly malignant eyelid tumors can be detected at an early stage (skin infiltration depth ≦ 0.76 mm) and treated accordingly, their 5-year survival rate can reach over 99%. Therefore, early identification of eyelid malignancy is important to improve prognosis.
However, today, the identification and diagnosis of eyelid tumor types often require an experienced ophthalmologist to perform examination and diagnosis, and the accuracy of identifying eyelid malignancy by young ophthalmologists is low, and the identification of partial eyelid benign malignancy is difficult and the early identification rate of eyelid malignancy is low. For example, certain malignant melanoma of the eyelids is very similar in appearance, morphology, texture, etc. both early and benign pigmented nevi and is easily overlooked by the patient himself or a physician with relatively little clinical experience. However, malignant melanoma has a high probability of early metastasis and poses a great risk to life. And the treatment is carried out after the transfer, the curative effect is often poor, the complications are many, and the prognosis is relatively poor.
Disclosure of Invention
The technical problem solved by the invention is as follows: at present, partial good and malignant tumors of eyelids are difficult to identify, and the early identification rate of the malignant tumors of the eyelids is low.
In order to solve the problems, the technical scheme of the invention is as follows:
a photo-based eyelid malignancy intelligent recognition system is provided, comprising:
a photo import module for importing ordinary digital camera or smart mobile phone shooting photo into system, the photo import module includes: a primary import submodule for importing the eyelid picture for the first time, a re-import submodule for importing the eyelid picture again, an eyelid tumor positioning module for transmitting the output data of the primary import submodule and an eyelid tumor growth condition judging module for transmitting the output data of the re-import submodule,
an eyelid tumor positioning module used for guiding the picture leading-in module into the position of the tumor in the picture for positioning, cutting the picture and outputting the cut eyelid tumor picture, wherein the eyelid tumor positioning module is based on an eyelid tumor positioning model trained through positioning deep learning,
the eyelid tumor benign and malignant identification module is used for identifying the cut tumor picture derived by the eyelid tumor positioning module to obtain the tumor type, the eyelid tumor benign and malignant identification module is based on an eyelid tumor benign and malignant identification model trained through classification deep learning,
an eyelid tumor growth condition determining module for comparing the picture imported by the picture importing module with the picture imported for the first time, wherein the eyelid tumor growth condition determining module comprises: a growth speed judgment submodule for judging the growth speed of the tumor through the picture, a ulceration judgment submodule for judging the ulceration condition of the tumor through the picture, a tumor growth condition output submodule for obtaining the benign and malignant tumor growth condition of the tumor according to the judgment results of the growth speed judgment submodule and the ulceration judgment submodule,
a result output module for outputting the result determined by the eyelid tumor benign/malignant identification module, wherein the result output module comprises: and the second recognition informing submodule reminds the user of diagnosing as soon as possible when the judgment result of the eyelid tumor benign and malignant identification module or the eyelid tumor growth condition judgment module is malignant, reminds the user of uploading the eyelid picture again on a day to further diagnose when the judgment result of the eyelid tumor benign and malignant identification module is benign, uploads the eyelid picture to the eyelid malignant tumor intelligent identification system again when the user uploads the eyelid picture to the eyelid malignant tumor intelligent identification system again, and informs the user of the benign eyelid tumor informing submodule when the judgment result of the eyelid tumor growth condition judgment module is benign.
Further, the intelligent identification system for malignant eyelid tumors further comprises: a user information management module for processing user information, the user information management module comprising: the intelligent identification system comprises a personal information management submodule for processing personal information of a user and a photo management submodule for managing historical eyelid photos of the user, so that the intelligent identification system can record and backtrack the information of the user.
Further, the intelligent identification system for the malignant tumors of the eyelids further comprises: the database is used for storing user information and uploading eyelid photos by a user, and can record and process data generated in the working process of the system.
Further, the training process of the eyelid tumor localization model is as follows:
SA1, collecting eyelid photographs containing malignant eyelid tumor, benign eyelid tumor and non-eyelid tumor to generate an image data set, and dividing the image data set into a training data set and a test data set;
SA2, marking the position of the eyelid tumor in the eyelid picture of the image data set by a rectangular frame;
SA3, constructing an eyelid tumor positioning model through a Faster-RCNN network, defining a loss function, selecting an optimizer, and training the eyelid tumor positioning model through a training data set;
SA4, after training, verifying and adjusting the eyelid tumor positioning model through the test data set, so that the eyelid tumor cutting accuracy of the eyelid tumor positioning model in the picture is more than 90%.
The data processed by the eyelid tumor positioning model can be used as data sources of the eyelid tumor benign and malignant identification model and the eyelid tumor ulceration identification model, so that the model training is facilitated, and the eyelid tumor type can be conveniently judged by the eyelid tumor benign and malignant identification model.
Further, the training process of the eyelid tumor benign and malignant identification model is as follows:
SB1, trimming the eyelid tumor photos cut by the eyelid tumor positioning model, dividing the cut eyelid tumor photos into benign and malignant photos according to pathological results, preprocessing the photos, performing data increment through horizontal and vertical overturning and displacement, taking the processed photos as an image data set, and dividing the image data set into a training data set and a test data set;
SB2, constructing a good and malignant identification model of the eyelid tumor through a deep learning classification network VGG16, defining a loss function, selecting an optimizer, and training the good and malignant identification model of the eyelid tumor through a training data set;
SB3, after the training is finished, verifying and adjusting the eyelid tumor benign and malignant identification model through the test data set, so that the accuracy of the eyelid tumor benign and malignant identification model in the photo is more than 90%.
The eyelid tumor benign and malignant identification model can judge the tumor type according to the eyelid tumor picture, greatly saves the time for patients to go to a hospital for inquiry, has higher reliability, and relieves the shortage of medical resources.
Preferably, the working principle of the growth rate judgment submodule is as follows: the two photos are adjusted in size, so that the sizes of the eyelids in the two photos are the same, then the eyelid tumor positioning module is called to cut the positions of the eyelid tumors in the primary imported photo and the secondary imported photo respectively, the sizes of the two cut eyelid tumor photos are compared, and the growth speed of the tumors is obtained.
Preferably, the ulceration judgment submodule is based on the eyelid tumor ulceration recognition model trained through classified deep learning, and the ulceration recognition model can directly judge the change condition of the tumor and serve as another re-judgment basis of the benign and malignant eyelid tumors, so that the reliability of the result obtained by the intelligent recognition system is higher.
Preferably, the training process of the eyelid tumor ulceration recognition model is as follows:
SC1, collecting eyelid photos containing eyelid tumor ulceration and eyelid tumor non-ulceration, dividing the eyelid photos into ulceration and non-ulceration, preprocessing the photos, performing data increment through horizontal and vertical overturning and displacement, taking the processed photos as an image data set, and dividing the image data set into a training data set and a test data set;
SC2, constructing an eyelid tumor ulceration recognition model through a deep learning classification network VGG16, defining a loss function, selecting an optimizer, and training the eyelid tumor ulceration recognition model through a training data set;
and SC3, after the training is finished, verifying and adjusting the eyelid tumor ulceration recognition model through the test data set, so that the accuracy rate of judging the eyelid tumor type in the picture by the eyelid tumor ulceration recognition model is more than 90%.
Further preferably, the judgment principle of the tumor growth condition output submodule is as follows: when the growth rate of the tumor is more than 30%/month or/and the tumor is broken, the output is high in malignant probability, the user needs to further diagnose and treat the tumor, otherwise, the output is high in benign probability, the observation can be continued, the growth rate and the broken condition of the tumor are used as another re-judgment basis for the benign and malignant conditions of the eyelid tumor, and the reliability of the result obtained by the intelligent recognition system is high.
The invention has the beneficial effects that:
(1) according to the method, the cutting of the eyelid picture and the identification of the eyelid tumor category are completed through the eyelid tumor positioning model and the eyelid tumor benign and malignant identification model, the model trained by a large number of pictures is equivalent to an ophthalmologist with abundant experience, the identification of the tumor category is more accurate and rapid compared with that of a novice doctor, the reference value is higher, and the time and the economic cost for preliminary identification and judgment of the eyelid tumor are saved;
(2) after the identification is finished, the problem of low early identification rate of the eyelid tumor is considered, so that the growth speed and the ulceration of the eyelid tumor are used as auxiliary information on the basis of intelligent identification and become another repeated judgment basis of the benign and malignant eyelid tumor, and the reliability of the result obtained by an intelligent identification system is higher.
Drawings
FIG. 1 is a system framework diagram of embodiment 1;
FIG. 2 is a diagram of a photo lead-in module in example 1;
FIG. 3 is a diagram showing the result output module in example 1;
FIG. 4 is a block diagram of the eyelid tumor growth status determining module in example 1;
FIG. 5 is a system framework diagram of embodiment 2;
FIG. 6 is a block diagram of a user information management module in accordance with embodiment 2;
FIG. 7 is a flow chart of an eyelid tumor localization model training process;
FIG. 8 is a flow chart of a process for training a recognition model of benign and malignant eyelid tumors;
FIG. 9 is a flow chart of an eyelid tumor ulceration recognition module training process;
fig. 10 is a flowchart of the eyelid tumor localization model process.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention clearer, the present invention will be described in further detail with reference to the accompanying drawings, and it is apparent that the described embodiments are only a part of the embodiments of the present invention, 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.
The terminology used in the embodiments of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used in the description of the invention and the appended claims, the singular forms "a", "an", and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise, and "the plural" typically includes at least two.
Example 1
As shown in fig. 1, the present embodiment is a photo-based eyelid malignancy intelligent identification system, including:
as shown in fig. 2, the photo import module is used for importing photos taken by a general digital camera or a smart phone into the system, and includes: a primary import submodule for importing the eyelid picture for the first time, a re-import submodule for importing the eyelid picture again, an eyelid tumor positioning module for transmitting the output data of the primary import submodule and an eyelid tumor growth condition judging module for transmitting the output data of the re-import submodule,
an eyelid tumor positioning module used for guiding the picture leading-in module into the position of the tumor in the picture for positioning, cutting the picture and outputting the cut eyelid tumor picture, wherein the eyelid tumor positioning module is based on an eyelid tumor positioning model trained through positioning deep learning,
the eyelid tumor benign and malignant identification module is used for identifying the cut tumor picture derived by the eyelid tumor positioning module to obtain the tumor type, the eyelid tumor benign and malignant identification module is based on an eyelid tumor benign and malignant identification model trained through classification deep learning,
an eyelid tumor growth condition determining module for comparing the re-imported picture with the primarily imported picture, as shown in fig. 4, the eyelid tumor growth condition determining module includes: a growth speed judgment submodule for judging the growth speed of the tumor through the picture, a ulceration judgment submodule for judging the ulceration condition of the tumor through the picture, a tumor growth condition output submodule for obtaining the benign and malignant tumor growth condition of the tumor according to the judgment results of the growth speed judgment submodule and the ulceration judgment submodule,
a result output module for outputting the result determined by the eyelid tumor benign/malignant identification module, as shown in fig. 3, the result output module includes: and the second recognition informing submodule reminds the user of diagnosing as soon as possible when the judgment result of the eyelid tumor benign and malignant identification module or the eyelid tumor growth condition judgment module is malignant, reminds the user of uploading the eyelid picture again on a day to further diagnose when the judgment result of the eyelid tumor benign and malignant identification module is benign, uploads the eyelid picture to the eyelid malignant tumor intelligent identification system again when the user uploads the eyelid picture to the eyelid malignant tumor intelligent identification system again, and informs the user of the benign eyelid tumor informing submodule when the judgment result of the eyelid tumor growth condition judgment module is benign.
The working principle of the growth speed judgment submodule is as follows: adjusting the sizes of the two photos to enable the sizes of the eyelids in the two photos to be the same, then calling an eyelid tumor positioning module to respectively cut the positions of the eyelid tumors in the primary imported photo and the secondary imported photo, comparing the sizes of the two cut eyelid tumor photos to obtain the growth speed of the tumors, and enabling a ulceration judgment sub-module to be based on an eyelid tumor ulceration recognition model trained through classified deep learning; the judgment principle of the tumor growth condition output submodule is as follows: when the growth rate of the tumor is more than 30%/month or/and the tumor is broken, the output is a malignant probability which is larger, the user needs to further diagnose, otherwise, the output is a benign probability which is larger, and the observation can be continued.
As shown in fig. 7, the training process of the eyelid tumor localization model is:
SA1, collecting eyelid photographs containing malignant eyelid tumor, benign eyelid tumor and non-eyelid tumor to generate an image data set, and dividing the image data set into a training data set and a test data set;
SA2, marking the position of the eyelid tumor in the eyelid picture of the image data set by a rectangular frame;
SA3, constructing an eyelid tumor positioning model through a Faster-RCNN network, defining a loss function, selecting an optimizer, and training the eyelid tumor positioning model through a training data set;
SA4, after training, verifying and adjusting the eyelid tumor positioning model through the test data set, so that the eyelid tumor cutting accuracy of the eyelid tumor positioning model in the picture is more than 90%.
As shown in fig. 8, the training process of the eyelid tumor benign and malignant identification model is as follows:
SB1, trimming the eyelid tumor photos cut by the eyelid tumor positioning model, dividing the cut eyelid tumor photos into benign and malignant photos according to pathological results, preprocessing the photos, performing data increment through horizontal and vertical overturning and displacement, taking the processed photos as an image data set, and dividing the image data set into a training data set and a test data set;
SB2, constructing a good and malignant identification model of the eyelid tumor through a deep learning classification network VGG16, defining a loss function, selecting an optimizer, and training the good and malignant identification model of the eyelid tumor through a training data set;
SB3, after the training is finished, verifying and adjusting the eyelid tumor benign and malignant identification model through the test data set, so that the accuracy of the eyelid tumor benign and malignant identification model in the photo is more than 90%.
As shown in fig. 9, the training process of the eyelid tumor ulceration recognition model is as follows:
SC1, collecting eyelid photos containing eyelid tumor ulceration and eyelid tumor non-ulceration, dividing the eyelid photos into ulceration and non-ulceration, preprocessing the photos, performing data increment through horizontal and vertical turning and displacement, taking the processed photos as an image data set, and dividing the image data set into a training data set and a test data set;
SC2, constructing an eyelid tumor ulceration recognition model through a deep learning classification network VGG16, defining a loss function, selecting an optimizer, and training the eyelid tumor ulceration recognition model through a training data set;
and SC3, after the training is finished, verifying and adjusting the eyelid tumor ulceration recognition model through the test data set, so that the accuracy rate of judging the eyelid tumor type in the picture by the eyelid tumor ulceration recognition model is more than 90%.
The working process of the embodiment is as follows:
s1, the user logs in the eyelid malignant tumor intelligent recognition system, and the photographed eyelid picture is imported to the eyelid tumor positioning module through the primary import submodule;
s2, the eyelid tumor positioning module positions the position of the input picture in the picture of the eyelid tumor through an eyelid tumor positioning model constructed by a Faster-RCNN network, and then cuts the picture according to the positioning to obtain a cut picture, wherein the process is as shown in FIG. 10, and the cut picture is led into the eyelid tumor benign and malignant identification module;
s3, the eyelid tumor benign and malignant identification module identifies the cut photos through an eyelid tumor benign and malignant identification model constructed by a deep learning classification network VGG16, when the result is malignant, the malignant notification submodule of the result output module outputs that the probability is high, the user needs to further diagnose, and when the result is benign, the operation mode comprises the following steps:
s3-1, informing the user through the re-identification informing submodule of the result output module that the eyelid photo is uploaded again after 15 days,
s3-2 and 15 days later, the user uploads the current eyelid picture to the eyelid tumor growth condition judgment module through the reintroduction submodule, the growth speed judgment submodule adjusts the sizes of the two pictures to enable the eyelids in the two pictures to be the same, then the eyelid tumor positioning module is called to respectively cut the positions of the eyelid tumors in the first introduced picture and the second introduced picture, the sizes of the two cut eyelid tumor pictures are compared to obtain the growth speed of the tumors, the ulceration judgment submodule conducts ulceration recognition on the reintroduced eyelid pictures to obtain the ulceration condition of the eyelid tumors of the user,
s3-3, generating a recognition result by the tumor growth condition output submodule according to the judgment result of the eyelid tumor growth condition judgment module, outputting a malignant probability by the malignant output submodule when the growth speed of the tumor is more than 30%/month or/and the tumor is broken, and making the user to further diagnose and treat, or outputting a benign probability by the benign output submodule and observing continuously.
Example 2
This embodiment is different from embodiment 1 in that:
as shown in fig. 5, the intelligent identification system for malignant eyelid further includes: as shown in fig. 6, the user information management module for processing user information includes: the personal information management submodule is used for processing the personal information of the user, and the photo management submodule is used for managing the historical eyelid photos of the user.
A database for storing user information and user uploaded photographs of the eyelids.
The working process of the embodiment is as follows:
s1, the user logs in the eyelid malignant tumor intelligent recognition system, fills in the personal information of the user through the personal information management submodule, and then introduces the shot eyelid picture to the eyelid tumor positioning module through the primary introduction submodule;
s2, the eyelid tumor positioning module positions the position of the input picture in the picture of the eyelid tumor through an eyelid tumor positioning model constructed by the Faster-RCNN network, then cuts the picture according to the positioning to obtain a cut picture, and the process is shown in figure 10, and the cut picture is led into the eyelid tumor benign and malignant identification module;
s3, the eyelid tumor benign and malignant identification module identifies the cut photos through an eyelid tumor benign and malignant identification model constructed by a deep learning classification network VGG16, when the result is malignant, the malignant notification submodule of the result output module outputs that the probability is high, the user needs to further diagnose, and when the result is benign, the operation mode comprises the following steps:
s3-1, informing the user through the re-identification informing submodule of the result output module that the eyelid photo is uploaded again after 15 days,
s3-2 and 15 days later, the user uploads the current eyelid picture to the eyelid tumor growth condition judgment module through the reintroduction submodule, the growth speed judgment submodule adjusts the sizes of the two pictures to enable the eyelids in the two pictures to be the same, then the eyelid tumor positioning module is called to respectively cut the positions of the eyelid tumors in the first introduced picture and the second introduced picture, the sizes of the two cut eyelid tumor pictures are compared to obtain the growth speed of the tumors, the ulceration judgment submodule conducts ulceration recognition on the reintroduced eyelid pictures to obtain the ulceration condition of the eyelid tumors of the user,
s3-3, the tumor growth condition output submodule generates a recognition result according to the judgment result of the eyelid tumor growth condition judgment module, when the growth speed of the tumor is more than 30%/month or/and the tumor is broken, the malignant output submodule outputs a high malignant probability, the user needs to further diagnose and treat, otherwise, the benign output submodule outputs a high benign probability, and the observation can be continued;
s4, the user can query the personal information and the previous eyelid photo uploaded by the user information management module.
Claims (8)
1. A photograph-based intelligent eyelid malignancy identification system, comprising:
the photo import module is used for importing photos shot by a common digital camera or a smart phone into the system, and comprises: a first import sub-module for first importing the eyelid picture, a second import sub-module for second importing the eyelid picture,
an eyelid tumor positioning module used for guiding the photo guiding module into the position of the tumor in the photo for positioning, cutting the photo and outputting the cut eyelid tumor photo, wherein the eyelid tumor positioning module is based on an eyelid tumor positioning model trained through positioning deep learning,
the eyelid tumor benign and malignant identification module is used for identifying the cut tumor picture derived by the eyelid tumor positioning module to obtain the tumor type, the eyelid tumor benign and malignant identification module is based on an eyelid tumor benign and malignant identification model trained through classification deep learning,
an eyelid tumor growth condition determining module for comparing the photograph imported again by the photograph importing module with the photograph imported for the first time, wherein the eyelid tumor growth condition determining module comprises: a growth speed judgment submodule for judging the growth speed of the tumor through the picture, a ulceration judgment submodule for judging the ulceration condition of the tumor through the picture, and a tumor growth condition output submodule for obtaining the benign and malignant tumor growth condition of the tumor according to the judgment results of the growth speed judgment submodule and the ulceration judgment submodule,
a result output module for outputting the result determined by the eyelid tumor benign and malignant identification module, wherein the result output module comprises: and the sub-module is used for reminding the user of notifying malignancy of diagnosis and treatment as soon as possible when the result of the judgment of the eyelid tumor benign and malignant identification module or the eyelid tumor growth condition judgment module is malignant, reminding the user of uploading an eyelid picture again on a selected day for further diagnosis when the result of the judgment of the eyelid tumor benign and malignant identification module is benign, and notifying the user of notifying the benign eyelid tumor of the eyelid tumor when the result of the judgment of the eyelid tumor growth condition judgment module is benign.
2. The intelligent photograph-based eyelid malignancy recognition system of claim 1, wherein the intelligent eyelid malignancy recognition system further comprises: a user information management module for processing user information, the user information management module comprising: the personal information management submodule is used for processing the personal information of the user, and the photo management submodule is used for managing the historical eyelid photos of the user.
3. The intelligent photograph-based eyelid malignancy recognition system according to claim 2, wherein the intelligent eyelid malignancy recognition system further comprises: a database for storing user information and user uploaded eyelid photographs.
4. The system of claim 1, wherein the eyelid malignancy localization model is trained by:
SA1, collecting eyelid photographs containing malignant eyelid tumors, benign eyelid tumors, and non-eyelid tumors to generate an image data set, and dividing the image data set into a training data set and a test data set;
SA2, marking the position of the eyelid tumor in the eyelid picture of the image data set by a rectangular frame;
SA3, constructing an eyelid tumor positioning model through a Faster-RCNN network, defining a loss function, selecting an optimizer, and training the eyelid tumor positioning model through a training data set;
SA4, after training, verifying and adjusting the eyelid tumor positioning model through the test data set, so that the eyelid tumor cutting accuracy of the eyelid tumor positioning model in the picture is more than 90%.
5. The intelligent photograph-based eyelid malignancy recognition system according to claim 1, wherein the training process of the eyelid malignancy recognition model is:
SB1, trimming the cut eyelid tumor photos of the eyelid tumor positioning model, dividing the cut eyelid tumor photos into benign and malignant photos according to the pathological result, preprocessing the photos, performing data increment through horizontal and vertical overturning and displacement, taking the processed photos as an image data set, and dividing the image data set into a training data set and a testing data set;
SB2, constructing a good and malignant identification model of the eyelid tumor through a deep learning classification network VGG16, defining a loss function, selecting an optimizer, and training the good and malignant identification model of the eyelid tumor through a training data set;
SB3, after the training is finished, verifying and adjusting the eyelid tumor benign and malignant identification model through the test data set, so that the accuracy of the eyelid tumor benign and malignant identification model in the photo is more than 90%.
6. The system of claim 1, wherein the growth rate determining sub-module operates on the following principle: and adjusting the sizes of the two photos to enable the sizes of the eyelids in the two photos to be the same, calling an eyelid tumor positioning module to respectively cut the positions of the eyelid tumors in the primary imported photo and the secondary imported photo, and comparing the sizes of the two cut eyelid tumor photos to obtain the growth speed of the tumors.
7. The system of claim 6, wherein the breach judgment sub-module is based on a eyelid tumor breach recognition model trained by deep learning with classification.
8. The system as claimed in claim 7, wherein the training process of the eyelid tumor ulceration recognition model is:
SC1, collecting eyelid photos containing eyelid tumor ulceration and eyelid tumor non-ulceration, dividing the eyelid photos into ulceration and non-ulceration, preprocessing the photos, performing data increment through horizontal and vertical overturning and displacement, taking the processed photos as an image data set, and dividing the image data set into a training data set and a test data set;
the method comprises the following steps that SC2, an eyelid tumor ulceration recognition model is built through a deep learning classification network VGG16, a loss function is defined, an optimizer is selected, and then the eyelid tumor ulceration recognition model is trained through a training data set;
and SC3, after the training is finished, verifying and adjusting the eyelid tumor ulceration recognition model through the test data set, so that the accuracy rate of judging the eyelid tumor type in the picture by the eyelid tumor ulceration recognition model is more than 90%.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210156943.8A CN114595747B (en) | 2022-02-21 | 2022-02-21 | Eyelid malignancy intelligent recognition system based on photo |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210156943.8A CN114595747B (en) | 2022-02-21 | 2022-02-21 | Eyelid malignancy intelligent recognition system based on photo |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114595747A true CN114595747A (en) | 2022-06-07 |
| CN114595747B CN114595747B (en) | 2024-05-14 |
Family
ID=81804541
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210156943.8A Active CN114595747B (en) | 2022-02-21 | 2022-02-21 | Eyelid malignancy intelligent recognition system based on photo |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114595747B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2016198379A (en) * | 2015-04-13 | 2016-12-01 | 健一 ▲高▼木 | Eyelid neoplastic tumorous lesion discrimination device and eyelid neoplastic tumorous lesion discrimination program |
| RU2683159C1 (en) * | 2017-10-13 | 2019-03-26 | Федеральное государственное бюджетное военное образовательное учреждение высшего образования Военно-медицинская академия имени С.М. Кирова Министерства обороны Российской Федерации (ВМедА) | Method for removing neoplasm of posterior-medial part of orbit with intra- and extraconical localization |
| CN110638524A (en) * | 2019-09-16 | 2020-01-03 | 山东省肿瘤防治研究院（山东省肿瘤医院） | Tumor puncture real-time simulation system based on VR glasses |
| WO2020210891A1 (en) * | 2019-04-18 | 2020-10-22 | Shelley Boyd | Detection, prediction, and classification for ocular disease |
| CN113139935A (en) * | 2021-03-30 | 2021-07-20 | 宁波市眼科医院 | Slit-lamp picture quality monitoring system |
| CN113449785A (en) * | 2021-06-18 | 2021-09-28 | 浙江大学 | Eyelid tumor digital pathological section image multi-classification method based on deep learning |
-
2022
- 2022-02-21 CN CN202210156943.8A patent/CN114595747B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2016198379A (en) * | 2015-04-13 | 2016-12-01 | 健一 ▲高▼木 | Eyelid neoplastic tumorous lesion discrimination device and eyelid neoplastic tumorous lesion discrimination program |
| RU2683159C1 (en) * | 2017-10-13 | 2019-03-26 | Федеральное государственное бюджетное военное образовательное учреждение высшего образования Военно-медицинская академия имени С.М. Кирова Министерства обороны Российской Федерации (ВМедА) | Method for removing neoplasm of posterior-medial part of orbit with intra- and extraconical localization |
| WO2020210891A1 (en) * | 2019-04-18 | 2020-10-22 | Shelley Boyd | Detection, prediction, and classification for ocular disease |
| CN110638524A (en) * | 2019-09-16 | 2020-01-03 | 山东省肿瘤防治研究院（山东省肿瘤医院） | Tumor puncture real-time simulation system based on VR glasses |
| CN113139935A (en) * | 2021-03-30 | 2021-07-20 | 宁波市眼科医院 | Slit-lamp picture quality monitoring system |
| CN113449785A (en) * | 2021-06-18 | 2021-09-28 | 浙江大学 | Eyelid tumor digital pathological section image multi-classification method based on deep learning |
Non-Patent Citations (4)
| Title |
|---|
| 娆琬·托勒洪: "常规超声及CT诊断眼部良恶性病变的应用价值", 《中国优秀硕士学位论文全文数据库医药卫生科技辑》, 31 January 2019 (2019-01-31), pages 060 - 74 * |
| 李中文;陈蔚;蒋杰伟;刘小天;佟莉杨;杨莉莉: "基于深度学习的裂隙灯眼前段图像质量自动监控系统", 《宁波市眼科医院》, 10 November 2021 (2021-11-10) * |
| 洪瑾;张志强;刘宇;花雷;李浩博;: "212例眼睑肿物临床病理分析", 临床眼科杂志, no. 06, 25 December 2016 (2016-12-25) * |
| 牛膺筠!266003, 刘夫玲!266003, 孟旭霞!266003, 王红云!266003, 林红!266003: "p16和bcl-2蛋白在眼睑恶性肿瘤中的表达及其意义", 中华眼科杂志, no. 04, 12 July 2000 (2000-07-12) * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114595747B (en) | 2024-05-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109492711B (en) | Malignant melanoma and non-malignant melanoma classification system based on deep learning | |
| CN112132833B (en) | Dermatological image focus segmentation method based on deep convolutional neural network | |
| CN103942799B (en) | Breast ultrasounography image segmentation method and system | |
| CN108920634A (en) | The skin disease characteristic analysis system of knowledge based map | |
| KR102162683B1 (en) | Reading aid using atypical skin disease image data | |
| CN112086197A (en) | Mammary nodule detection method and system based on ultrasonic medicine | |
| CN114627067A (en) | Wound area measurement and auxiliary diagnosis and treatment method based on image processing | |
| CN111797901A (en) | Retinal artery and vein classification method and device based on topological structure estimation | |
| CN106370488A (en) | Cycle tumor cell slice-based diagnosis system and method | |
| CN114580558A (en) | Method, system, equipment and storage medium for detecting endometrial cancer cells | |
| CN110189324A (en) | A kind of medical image processing method and processing unit | |
| Nawn et al. | Multifractal alterations in oral sub-epithelial connective tissue during progression of pre-cancer and cancer | |
| Agyenta et al. | Skin lesion classification based on convolutional neural network | |
| JP7404535B2 (en) | Conduit characteristic acquisition method based on computer vision, intelligent microscope, conduit tissue characteristic acquisition device, computer program, and computer equipment | |
| Xu et al. | Application of artificial intelligence technology in medical imaging | |
| CN114595747A (en) | Intelligent identification system for malignant tumors of eyelids based on photos | |
| CN109480775A (en) | A kind of icterus neonatorum identification device based on artificial intelligence, equipment, system | |
| Xue et al. | Automatic detection of oral lesion measurement ruler toward computer-aided image-based oral cancer screening | |
| JP7244974B1 (en) | Pathological image feature extractor training method, training device, electronic device, storage medium, and pathological image classification system based on feature separation | |
| CN110765908A (en) | Cascade type cancer cell detection system based on deep learning | |
| CN109711306A (en) | A kind of method and apparatus obtaining facial characteristics based on depth convolutional neural networks | |
| Praba et al. | Comparing different classifiers for automatic lesion detection in cervix based on colour histogram | |
| Arai et al. | Improvement of automated detection method for clustered microcalcification based on wavelet transformation and support vector machine | |
| Xie et al. | Automated Diagnosis of Melanoma Histopathological Images Based on Deep Learning Using Trust Counting Method | |
| Veeraiah et al. | Medical Internet of Things using Deep Learning Techniques for Skin Cancer Detection |
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 |