CN109543728A - A kind of Thyroid-related Ophthalmopathy detection method based on transfer learning - Google Patents
A kind of Thyroid-related Ophthalmopathy detection method based on transfer learning Download PDFInfo
- Publication number
- CN109543728A CN109543728A CN201811318375.7A CN201811318375A CN109543728A CN 109543728 A CN109543728 A CN 109543728A CN 201811318375 A CN201811318375 A CN 201811318375A CN 109543728 A CN109543728 A CN 109543728A
- Authority
- CN
- China
- Prior art keywords
- bitmap
- network
- migration
- thyroid
- sample
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- 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
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/03—Recognition of patterns in medical or anatomical images
Abstract
The Thyroid-related Ophthalmopathy detection method based on transfer learning that the invention discloses a kind of, medical scanning is carried out to ocular first, 3 kinds of bitmap samples are obtained, the bitmap sample of acquisition is pre-processed, obtain the illness region and right and left eyes position position in bitmap sample；Then pretreated bitmap sample rotated, translated, zoom operations, then all bitmap samples are divided into training set, test set and verifying to scale and collected；Then for every kind of bitmap sample, a corresponding migration network is separately designed, 3 migration networks are trained using the training set of corresponding classification, and with verifying collection come debugging network parameter；The output that 3 migration networks are finally integrated with ballot method obtains the diagnosis output of final system.Present invention feature extraction layer in deep learning model can effectively extract illness feature and multi-level mapping unit carries out the classification of illness characteristic information, have the illness recognition accuracy close to man efficiency.
Description
Technical field
The invention belongs to big data field of cloud computer technology, are related to a kind of Thyroid-related Ophthalmopathy detection method, spy's tool
Body is related to a kind of Thyroid-related Ophthalmopathy detection method based on transfer learning.
Background technique
With the fast development of Medical Imaging Technology, medical image analysis steps into big data era, how from the doctor of magnanimity
It learns in image data and excavates useful information, huge challenge is brought to medical image recognition.It is huge data volume, many kinds of
Imaging device, along with different disease sites and different kinds of Diseases, traditional data analysing method usually cannot
Meet the requirement of people, therefore in medicine big data era, how to excavate useful information from massive medical image data,
Research hotspot as academia and industry.
Deep learning is a frontier of machine learning, and traditional machine learning method cannot effectively excavate medicine
The abundant information contained in image, and deep learning establishes hierarchical mode by simulating human brain, and there is powerful automated characterization to mention
It takes, complex model constructs and efficient feature representation ability, it is often more important that deep learning method can be from the original of Pixel-level
It is extracted step by step in data from bottom to high-rise feature, this provides new think of to solve the new problem that medical image recognition faces
Road.Transfer learning is exactly to make trained model in a problem it is suitable for a new problem by simply adjustment,
There can be fine effect for the less situation of sample data volume, save a large amount of time.
Summary of the invention
For common adult orbit diseases-Thyroid-related Ophthalmopathy, the present invention provides one kind pioneeringly can
Accurately identify the Thyroid-related Ophthalmopathy detection method based on deep learning of illness in eye medical imaging.
The technical scheme adopted by the invention is that: a kind of Thyroid-related Ophthalmopathy detection method based on transfer learning,
Characterized by comprising the following steps:
Step 1: data preparation；
Medical scanning is carried out to ocular, obtains 3 kinds of bitmap samples, including horizontal bit pattern sheet, sagittal bit pattern sheet
With coronal bit pattern sheet；The bitmap sample of acquisition is pre-processed, the illness region and right and left eyes position in bitmap sample are obtained
Position；
Step 2: data enhancing processing；
Pretreated bitmap sample is rotated, is translated, zoom operations, is then drawn all bitmap samples to scale
It is divided into training set, test set and verifying collection；
Step 3: being directed to every kind of bitmap sample, separately design a corresponding migration network, obtain the first migration network, the
Two migration networks, third migrate network, and 3 migration networks constitute major network structure；
Step 4: 3 migration networks of training；
3 migration networks are trained using the training set of corresponding classification, and with verifying collection come debugging network parameter；
Step 5: integrating the output of 3 migration networks with ballot method, obtain the diagnosis output of final system.
Compared with the prior art, the invention has the following beneficial effects: being known by establishing the medical imaging based on transfer learning
Other model, feature extraction layer can effectively extract illness feature in migration models and multi-level mapping unit carries out illness
Characteristic information classification, the illness recognition accuracy with man efficiency.
Detailed description of the invention
Fig. 1 is the flow chart of the embodiment of the present invention.
Specific embodiment
Understand for the ease of those of ordinary skill in the art and implement the present invention, with reference to the accompanying drawings and embodiments to this hair
It is bright to be described in further detail, it should be understood that implementation example described herein is merely to illustrate and explain the present invention, not
For limiting the present invention.
Referring to Fig.1, a kind of Thyroid-related Ophthalmopathy detection method based on transfer learning provided by the invention, including with
Lower step:
Step 1: data preparation；
Medical scanning is carried out to ocular, obtains 3 kinds of bitmap samples, including horizontal bit pattern sheet, sagittal bit pattern sheet
With coronal bit pattern sheet；The bitmap sample of acquisition is pre-processed, the illness region and right and left eyes position in bitmap sample are obtained
Position；
Original eye medical image is pre-processed in the present embodiment, is that left and right eye orbit areas is selected by manual frame,
That is target area removes extra useless region.
Due to there are 3 kinds of scanning directions when ocular carries out medical scanning, 3 kinds of bitmaps, i.e., horizontal bitmap, sagittal have been corresponded to
Bitmap and coronal bitmap.Due to the symmetry of left and right eye socket, selection right eye socket of the eye is standard eye position, can be by horizontal bitmap and coronal
Left eye orbit areas in bitmap carries out flip horizontal operation, obtains left eye orbit areas.
Step 2: data enhancing processing；
Pretreated bitmap sample is rotated, is translated, zoom operations, is then drawn all bitmap samples to scale
It is divided into training set, test set and verifying collection；
Data enhancing refers to a series of method for carrying out expanding to improve data volume to initial data by stochastic transformations.
In conjunction with actual conditions, this system carries out data enhancing using rotation, translation, scaling, wherein the random interval of Pan and Zoom
Range be 0~10%, the random interval range of rotation is 0~10 °.Finally by enhanced bitmap sample according to 4:1:1
Ratio come divide training set, verifying collection and test set.
Step 3: being directed to every kind of bitmap sample, separately design a corresponding migration network, obtain the first migration network, the
Two migration networks, third migrate network, and 3 migration networks constitute major network structure；
The first migration network of the present embodiment corresponds to horizontal bitmap, includes a Google Inception-v3 model, most
The neuron number of full articulamentum is 2048 according to the number of plies afterwards, majorized function Adam；The second corresponding sagittal bitmap of migration network 2,
Comprising a Google Inception-v3 model, the neuron number of last full articulamentum is 2048 according to the number of plies, majorized function
For Adam；Third migrates the corresponding coronal bitmap of network 3, includes a Google Inception-v3 model, last full articulamentum
Neuron number according to the number of plies be respectively 2048 and 1024, majorized function Adam；Learning rate is a hyper parameter, needs to pass through
Verifying collection constantly adjusts；The last layer of 3 migration networks is classification layer, is all made of relu function to construct and intersect loss letter
Number.
Step 4: 3 migration networks of training；
3 migration networks are trained using the training set of corresponding classification, and with verifying collection come debugging network parameter；
In the present embodiment, passed through respectively come training transfer network using backpropagation strategy with the training set of corresponding classification
Adam updates network parameter, and the training sample batch for being sent into network every time is 64, and frequency of training is for 3000 times；It is seen after training each
A migration network is verifying the performance on collection, then regularized learning algorithm rate；Finally compare by the effect on verifying collection, the first migration
The variable optimal learning rate of network is 0.01, and the variable optimal learning rate of the second migration network is 0.01, and third migrates the Optimal Learning of network
Rate is 0.0001.
Step 5: integrating the output of 3 migration networks with ballot method, obtain the diagnosis output of final system.
Since 3 kinds of bitmaps reflect from 3 directions the feature of eye disorders, doctor is usually by the comprehensive detection of 3 kinds of bitmaps
To provide last diagnostic result.Therefore, this system finally integrates the output of 3 migration networks with ballot method, obtains to the end
System export result.Finally carry out the detection effect of checking system using test set, the experimental results showed that the accuracy rate of test reaches
To 90.91%, very close to artificial detection rate.
Medical imaging identifying system provided by the invention based on transfer learning has beneficial below compared with prior art
Effect: by establishing the medical imaging identification model based on deep learning, feature extraction layer can have in deep learning model
Effect extracts illness feature and multi-level mapping unit carries out the classification of illness characteristic information, has the illness close to man efficiency
Recognition accuracy.
It should be understood that the part that this specification does not elaborate belongs to the prior art.
It should be understood that the above-mentioned description for preferred embodiment is more detailed, can not therefore be considered to this
The limitation of invention patent protection range, those skilled in the art under the inspiration of the present invention, are not departing from power of the present invention
Benefit requires to make replacement or deformation under protected ambit, fall within the scope of protection of the present invention, this hair
It is bright range is claimed to be determined by the appended claims.
Claims (7)
1. a kind of Thyroid-related Ophthalmopathy detection method based on transfer learning, which comprises the following steps:
Step 1: data preparation；
Medical scanning is carried out to ocular, obtains 3 kinds of bitmap samples, including horizontal bit pattern sheet, sagittal bit pattern sheet and hat
Shape bitmap sample；The bitmap sample of acquisition is pre-processed, the illness region and right and left eyes position position in bitmap sample are obtained；
Step 2: data enhancing processing；
Pretreated bitmap sample is rotated, is translated, zoom operations, is then divided into all bitmap samples to scale
Training set, test set and verifying collection；
Step 3: being directed to every kind of bitmap sample, separately design a corresponding migration network, the first migration of acquisition network, second move
Move network, third migrates network, 3 migration networks composition major network structures；
Step 4: 3 migration networks of training；
3 migration networks are trained using the training set of corresponding classification, and with verifying collection come debugging network parameter；
Step 5: integrating the output of 3 migration networks with ballot method, obtain the diagnosis output of final system.
2. the Thyroid-related Ophthalmopathy detection method according to claim 1 based on transfer learning, it is characterised in that: step
Original eye medical image is pre-processed described in rapid 1, is that left and right eye orbit areas, i.e. target area are selected by manual frame
Domain removes extra useless region.
3. the Thyroid-related Ophthalmopathy detection method according to claim 2 based on transfer learning, it is characterised in that: choosing
Taking right eye socket of the eye is standard eye position, and the left eye orbit areas in horizontal bitmap and coronal bitmap is carried out flip horizontal operation, is obtained left
Eye orbit areas.
4. the Thyroid-related Ophthalmopathy detection method according to claim 1 based on transfer learning, it is characterised in that: step
In rapid 2, the random interval range of Pan and Zoom is 0~10%, and the random interval range of rotation is 0~10 °；To finally it increase
Bitmap sample after strong divides training set, verifying collection and test set according to the ratio of 4:1:1.
5. the Thyroid-related Ophthalmopathy detection method according to claim 1 based on transfer learning, it is characterised in that: step
In rapid 3, it includes a Google Inception-v3 model that the first migration network, which corresponds to horizontal bitmap, last full articulamentum
Neuron number is 2048 according to the number of plies, majorized function Adam；The second corresponding sagittal bitmap of migration network 2, includes one
Google Inception-v3 model, the neuron number of last full articulamentum are 2048 according to the number of plies, majorized function Adam；
Third migrates the corresponding coronal bitmap of network 3, includes a Google Inception-v3 model, the nerve of last full articulamentum
First number is respectively 2048 and 1024, majorized function Adam according to the number of plies；Learning rate is a hyper parameter, needs to collect by verifying
Constantly adjust；The last layer of 3 migration networks is classification layer, is all made of relu function to construct intersection loss function.
6. the Thyroid-related Ophthalmopathy detection method according to claim 1 based on transfer learning, it is characterised in that: step
In rapid 4, net is updated by Adam using backpropagation strategy come training transfer network with the training set of corresponding classification respectively
Network parameter, the training sample batch for being sent into network every time is 64, and frequency of training is 3000 times；See that each migration network exists after training
The performance on collection is verified, then regularized learning algorithm rate；Finally compare by the effect on verifying collection, optimal of the first migration network
Habit rate is 0.01, and the variable optimal learning rate of the second migration network is 0.01, and the variable optimal learning rate that third migrates network is 0.0001.
7. the Thyroid-related Ophthalmopathy detection method described in -6 any one based on transfer learning according to claim 1,
It is characterized in that: in step 5, examining detection effect using test set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811318375.7A CN109543728A (en) | 2018-11-07 | 2018-11-07 | A kind of Thyroid-related Ophthalmopathy detection method based on transfer learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811318375.7A CN109543728A (en) | 2018-11-07 | 2018-11-07 | A kind of Thyroid-related Ophthalmopathy detection method based on transfer learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109543728A true CN109543728A (en) | 2019-03-29 |
Family
ID=65845014
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811318375.7A Pending CN109543728A (en) | 2018-11-07 | 2018-11-07 | A kind of Thyroid-related Ophthalmopathy detection method based on transfer learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109543728A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110084150A (en) * | 2019-04-09 | 2019-08-02 | 山东师范大学 | A kind of Automated Classification of White Blood Cells method and system based on deep learning |
| CN110246158A (en) * | 2019-07-19 | 2019-09-17 | 上海交通大学医学院附属第九人民医院 | Eye illness detection device, method, electric terminal and storage medium |
| CN111951219A (en) * | 2020-07-09 | 2020-11-17 | 上海交通大学 | Thyroid eye disease screening method, system and equipment based on orbit CT image |
| CN113808738A (en) * | 2021-09-18 | 2021-12-17 | 安徽爱朋科技有限公司 | Disease identification system based on self-identification image |
| KR20220052166A (en) * | 2020-10-20 | 2022-04-27 | 중앙대학교 산학협력단 | Diagnosis method for Graves’orbitopathy using orbital computed tomography (CT) based on neural network-based algorithm |
| WO2023277622A1 (en) * | 2021-06-30 | 2023-01-05 | 주식회사 타이로스코프 | Method for guiding hospital visit for treating active thyroid ophthalmopathy and system for performing same |
| US11663719B2 (en) | 2021-06-30 | 2023-05-30 | Thyroscope Inc. | Method for hospital visit guidance for medical treatment for active thyroid eye disease, and system for performing same |
| US11717160B2 (en) | 2021-06-30 | 2023-08-08 | Thyroscope Inc. | Method and photographing device for acquiring side image for ocular proptosis degree analysis, and recording medium therefor |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106991439A (en) * | 2017-03-28 | 2017-07-28 | 南京天数信息科技有限公司 | Image-recognizing method based on deep learning and transfer learning |
| CN108596900A (en) * | 2018-05-02 | 2018-09-28 | 武汉联合创想科技有限公司 | Thyroid-related Ophthalmopathy medical image data processing unit, method, computer readable storage medium and terminal device |
| CN108764286A (en) * | 2018-04-24 | 2018-11-06 | 电子科技大学 | The classifying identification method of characteristic point in a kind of blood-vessel image based on transfer learning |
-
2018
- 2018-11-07 CN CN201811318375.7A patent/CN109543728A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106991439A (en) * | 2017-03-28 | 2017-07-28 | 南京天数信息科技有限公司 | Image-recognizing method based on deep learning and transfer learning |
| CN108764286A (en) * | 2018-04-24 | 2018-11-06 | 电子科技大学 | The classifying identification method of characteristic point in a kind of blood-vessel image based on transfer learning |
| CN108596900A (en) * | 2018-05-02 | 2018-09-28 | 武汉联合创想科技有限公司 | Thyroid-related Ophthalmopathy medical image data processing unit, method, computer readable storage medium and terminal device |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110084150B (en) * | 2019-04-09 | 2021-05-11 | 山东师范大学 | Automatic white blood cell classification method and system based on deep learning |
| CN110084150A (en) * | 2019-04-09 | 2019-08-02 | 山东师范大学 | A kind of Automated Classification of White Blood Cells method and system based on deep learning |
| CN110246158A (en) * | 2019-07-19 | 2019-09-17 | 上海交通大学医学院附属第九人民医院 | Eye illness detection device, method, electric terminal and storage medium |
| CN111951219B (en) * | 2020-07-09 | 2022-12-20 | 上海交通大学 | Thyroid eye disease screening method, system and equipment based on orbit CT image |
| CN111951219A (en) * | 2020-07-09 | 2020-11-17 | 上海交通大学 | Thyroid eye disease screening method, system and equipment based on orbit CT image |
| KR20220052166A (en) * | 2020-10-20 | 2022-04-27 | 중앙대학교 산학협력단 | Diagnosis method for Graves’orbitopathy using orbital computed tomography (CT) based on neural network-based algorithm |
| KR102537470B1 (en) * | 2020-10-20 | 2023-05-26 | 중앙대학교 산학협력단 | Diagnosis method for Graves’orbitopathy using orbital computed tomography (CT) based on neural network-based algorithm |
| WO2023277622A1 (en) * | 2021-06-30 | 2023-01-05 | 주식회사 타이로스코프 | Method for guiding hospital visit for treating active thyroid ophthalmopathy and system for performing same |
| US11663719B2 (en) | 2021-06-30 | 2023-05-30 | Thyroscope Inc. | Method for hospital visit guidance for medical treatment for active thyroid eye disease, and system for performing same |
| US11717160B2 (en) | 2021-06-30 | 2023-08-08 | Thyroscope Inc. | Method and photographing device for acquiring side image for ocular proptosis degree analysis, and recording medium therefor |
| US11741610B2 (en) | 2021-06-30 | 2023-08-29 | Thyroscope Inc. | Method for hospital visit guidance for medical treatment for active thyroid eye disease, and system for performing same |
| US11748884B2 (en) | 2021-06-30 | 2023-09-05 | Thyroscope Inc. | Method for hospital visit guidance for medical treatment for active thyroid eye disease, and system for performing same |
| CN113808738A (en) * | 2021-09-18 | 2021-12-17 | 安徽爱朋科技有限公司 | Disease identification system based on self-identification image |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109543728A (en) | A kind of Thyroid-related Ophthalmopathy detection method based on transfer learning | |
| CN110097559B (en) | Fundus image focus region labeling method based on deep learning | |
| CN107368670A (en) | Stomach cancer pathology diagnostic support system and method based on big data deep learning | |
| CN109376636A (en) | Eye ground image classification method based on capsule network | |
| CN109034045A (en) | A kind of leucocyte automatic identifying method based on convolutional neural networks | |
| CN106780466A (en) | A kind of cervical cell image-recognizing method based on convolutional neural networks | |
| CN108491077A (en) | A kind of surface electromyogram signal gesture identification method for convolutional neural networks of being divided and ruled based on multithread | |
| CN109598709A (en) | Mammary gland assistant diagnosis system and method based on fusion depth characteristic | |
| CN106778902A (en) | Milk cow individual discrimination method based on depth convolutional neural networks | |
| CN107767935A (en) | Medical image specification processing system and method based on artificial intelligence | |
| CN107368671A (en) | System and method are supported in benign gastritis pathological diagnosis based on big data deep learning | |
| CN106156793A (en) | Extract in conjunction with further feature and the classification method of medical image of shallow-layer feature extraction | |
| CN109389129A (en) | A kind of image processing method, electronic equipment and storage medium | |
| CN109635846A (en) | A kind of multiclass medical image judgment method and system | |
| CN108053398A (en) | A kind of melanoma automatic testing method of semi-supervised feature learning | |
| CN107424145A (en) | The dividing method of nuclear magnetic resonance image based on three-dimensional full convolutional neural networks | |
| CN109920538A (en) | A kind of zero sample learning method based on data enhancing | |
| CN109671068A (en) | A kind of abdominal muscles mask method and device based on deep learning | |
| CN107292314A (en) | A kind of lepidopterous insects species automatic identification method based on CNN | |
| CN108564561A (en) | Pectoralis major region automatic testing method in a kind of molybdenum target image | |
| CN108062749A (en) | Recognition methods, device and the electronic equipment of musculus levator ani ceasma | |
| CN111767952A (en) | Interpretable classification method for benign and malignant pulmonary nodules | |
| CN106096612A (en) | Trypetid image identification system and method | |
| Subhi Al-batah et al. | Multiple adaptive neuro-fuzzy inference system with automatic features extraction algorithm for cervical cancer recognition | |
| CN108345873A (en) | A kind of multiple degrees of freedom body motion information analytic method based on multilayer convolutional neural networks |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190329 |