CN110517256A - A kind of early carcinoma assistant diagnosis system based on artificial intelligence - Google Patents
A kind of early carcinoma assistant diagnosis system based on artificial intelligence Download PDFInfo
- Publication number
- CN110517256A CN110517256A CN201910814399.XA CN201910814399A CN110517256A CN 110517256 A CN110517256 A CN 110517256A CN 201910814399 A CN201910814399 A CN 201910814399A CN 110517256 A CN110517256 A CN 110517256A
- Authority
- CN
- China
- Prior art keywords
- convolutional neural
- neural networks
- training
- image
- networks model
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10068—Endoscopic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30096—Tumor; Lesion
Landscapes
- Engineering & Computer Science (AREA)
- Quality & Reliability (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Endoscopes (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to image analysis technology fields, specifically disclose a kind of early carcinoma assistant diagnosis system based on artificial intelligence, including, image collection module, for obtaining the sample image of the digestive endoscope with mark, pretreatment and randomly ordered, generation training image collection are carried out to sample image；Model construction module is iterated training to convolutional neural networks model for constructing convolutional neural networks model, and based on training image collection, is then tested, the successful convolutional neural networks model of training is exported after the completion of test；Diagnostic module, for obtain digestive endoscope to diagnostic image, diagnostic image is treated based on the successful convolutional neural networks model of training and is judged, judging result is exported.Trained validity can be improved using technical solution of the present invention.
Description
Technical field
The present invention relates to image analysis technology field, in particular to a kind of early carcinoma auxiliary diagnosis system based on artificial intelligence
System.
Background technique
The early discovery early treatment of alimentary tract cancer is of great practical significance.However, alimentary canal gas distribution compared with
More, lesion is smaller, invisible relatively strong, and the morphology of early carcinoma, surface micro-structure, surface capilary and inflammatory and prosthetic pole
Be it is similar, cause the pathological characteristics complicated difficult of digestive system cancer to distinguish that early symptom is not easy to be found.
The diagnosis of alimentary tract cancer mainly acquires people by the optical lens of digestive endoscopy and imaging sensor at present
Intracorporal image, and the image transmitting of acquisition is watched to display terminal for medical staff.Digestive endoscope image is mentioning
For in further detail, accurately diagnostic message while, also increase work load to read tablet doctor, and doctor passes through by diagosis
It tests and qualitative analysis is carried out with biggish subjectivity to lesion.Horizontal by medical level difference and scope doctor diagnosed simultaneously
Limitation, existing diagnosis are lower.
With the rapid development of computer field, different computerized algorithms and the calculating mould based on algorithms of different combination
Type is just constantly developed, and intelligent algorithm is one of them.It is imitate National People's Congress's brain neuron working principle one
Kind mathematical processing methods have the abilities such as self-learning capability, memory capability, predicted events development.But current artificial intelligence
The accuracy rate of algorithm Newly diagnosed is also lower, and the result of output is easy to cause mistake occur.Therefore, it is necessary to utilize a large amount of sample
Image is trained intelligent algorithm.Since there are certain differences for the identification and the visual identity of people of intelligent algorithm
Picture different, that people easily identifies, intelligent algorithm not necessarily easily identify, if the format of sample image, artificial intelligence is calculated
Method is difficult to, then being trained using sample images more again, it is also difficult to reach ideal training effect, cannot guarantee that
The accuracy rate of intelligent algorithm diagnosis.
For this reason, it may be necessary to which a kind of can carry out intelligent algorithm effectively to train, to improve examining for early carcinoma accuracy rate of diagnosis
Disconnected system.
Summary of the invention
In order to solve the above technical problems, the purpose of the present invention is to provide a kind of, the early carcinoma auxiliary based on artificial intelligence is examined
Disconnected system.
Technical solution of the present invention is as follows:
A kind of early carcinoma assistant diagnosis system based on artificial intelligence, including
Image collection module locates sample image for obtaining the sample image of the digestive endoscope with mark in advance
Reason and randomly ordered, generation training image collection；
Model construction module, for constructing convolutional neural networks model, and based on training image collection to convolutional neural networks
Model is iterated training, is then tested, and the successful convolutional neural networks model of training is exported after the completion of test；
Diagnostic module, for obtain digestive endoscope to diagnostic image, based on the successful convolutional neural networks mould of training
Type is treated diagnostic image and is judged, judging result is exported.
It base case principle and has the beneficial effect that:
In the present solution, by sample image carry out pretreatment and it is randomly ordered, so that sample image is met convolutional Neural net
The training requirement of network model facilitates convolutional neural networks model and reaches to improve the training effect of convolutional neural networks model
To expected accuracy rate of diagnosis.The successful convolutional neural networks model of training to digestive endoscope to diagnostic image at
Reason, the speed of processing is fast, accuracy rate is high, avoids due to environment light, visual fatigue, human eye inertia and asks grayscale is insensitive etc.
The problem that inaccuracy is handled caused by topic, alleviates the workload of medical staff's early period, improves medical staff's working efficiency,
Reduce the time that patient waits diagnostic result, improves user experience.
Further, the model construction module further includes construction unit, storage unit and test cell, pre- in storage unit
There is the test image of digestive endoscope, test cell is used in test, test image is inputted into convolutional neural networks model,
Judge the output accuracy rate of convolutional neural networks model, if output accuracy rate is lower than first threshold, test cell is obtained to image
Modulus block sends the information for regenerating training image collection；
Image collection module is adjusted pretreatment mode and is arranged again based on the information for regenerating training image collection
Sequence generates new training image collection；Construction unit is iterated instruction based on new training image collection, to convolutional neural networks model
Practice.
When accuracy rate is relatively low in testing for convolutional neural networks model, pretreated mode may generate shadow to training
It rings, such as pretreated mode is inappropriate, and pretreated sample image is caused to be difficult to be identified.Moreover, machine recognition and people
Visual identity have a certain difference, the picture that people easily identifies, machine not necessarily easily identifies, if held according to people merely
Mode easy to identify pre-processes sample image, is easy for the problem of machine is difficult to occur, namely holds herein
It easily leads to convolutional neural networks model to be difficult to, therefore, when exporting accuracy rate lower than first threshold after training, adjusts pre- place
The mode of reason, then re -training, helps to improve the output accuracy rate of convolutional neural networks model, training for promotion effect.It is logical
Cross the pretreatment to sample image and it is randomly ordered be adjusted, the training for making sample image meet convolutional neural networks model is wanted
It asks, to reduce the demand to sample image.
Further, the sample image includes one of early esophageal cancer, early carcinoma of stomach, early stage colon cancer or a variety of cancers
Disease classification, each cancer class are corresponding with form subclass and invasive depth subclass.
The sample image of form subclass and invasive depth subclass is set, convenient for improving the identification of convolutional neural networks model
Range.
Further, the pretreatment include cut, rotate, stretching, translating, color space adjustment, in setting contrast
It is one or more.
Pretreatment in several ways, it is easier to obtain the sample image convenient for the identification of convolutional neural networks model.
Further, image collection module is also normalized all sample images in training image collection.
By normalized, the identical sample image of format can be obtained, convenient for the identification of convolutional neural networks model.
Further, when the construction unit is iterated trained to convolutional neural networks model, convolutional neural networks model
Weight parameter be random initializtion parameter, by propagated forward algorithm constantly transfer characteristic forward, and calculate residual error, it is then logical
It crosses back-propagation algorithm and constantly transmits residual error backward, correct weight parameter, continuous iteration is until convergence.
By repetitive exercise, the recognition accuracy of convolutional neural networks model can be continuously improved.
Further, it after convolutional neural networks model every iteration 3000-4000 times, is tested.
It after the abundant iteration of convolutional neural networks, then is tested, can exclude to export accuracy rate caused by training is insufficient
The case where lower than first threshold.
Further, when the test cell judges the output accuracy rate of convolutional neural networks model, if output accuracy rate
Greater than first threshold and it is lower than second threshold, continues repetitive exercise.
Illustrate that repetitive exercise has had certain effect at this time, but it is also insufficient, it needs to continue to train.
Further, the first threshold is 30%-40%.
The lower first threshold of accuracy rate is set, it is easy to adjustment in time, convolutional neural networks model is carried out again again
Training.
Further, the second threshold is 70%-80%.
The higher second threshold of accuracy rate is set, convenient for convolutional neural networks model in continuous iteration, improves accuracy rate
The upper limit.
Detailed description of the invention
Fig. 1 is the logic diagram of the early carcinoma assistant diagnosis system embodiment one based on artificial intelligence.
Specific embodiment
It is further described below by specific embodiment:
Embodiment one
As shown in Figure 1, a kind of early carcinoma assistant diagnosis system based on artificial intelligence, including image collection module, model
Construct module and diagnostic module.
Image collection module is used to obtain the sample image of the digestive endoscope with mark, pre-processes to sample image
With it is randomly ordered, generate training image collection, and all sample images in training image collection be normalized.This implementation
In, normalized, which refers to, is normalized to DICOM format, NIfTI format or original binary format for sample image.
In the present embodiment, sample image includes one of early esophageal cancer, early carcinoma of stomach, early stage colon cancer or a variety of cancers
Disease classification, each cancer class are corresponding with form subclass and invasive depth subclass；The sample image of form subclass and
The sample image of invasive depth subclass is no less than 5000, pre-processes including cutting, rotating, stretching, translating, color space
One of adjustment, setting contrast are a variety of.In the present embodiment, pretreatment is including cutting, rotating, stretching, translating, color is empty
Between adjust and setting contrast.
Model construction module includes construction unit, storage unit and test cell；Construction unit is for constructing convolutional Neural
Network model；Construction unit is based on training image collection and is iterated training, convolutional neural networks mould to convolutional neural networks model
The weight parameter of type is random initializtion parameter, by propagated forward algorithm constantly transfer characteristic forward, and calculates residual error, then
Residual error is constantly transmitted backward by back-propagation algorithm, corrects weight parameter, continuous iteration is until convergence.It is pre- in storage unit
There is the test image of digestive endoscope.
In the present embodiment, convolutional neural networks model includes convolutional layer, pond layer, full articulamentum and classification layer, convolutional layer
Quantity be 7-20, the quantity of full articulamentum is 1-10；Layer of classifying uses softmax classifier；Convolutional layer is by specified number
The Convolution Filter of amount is applied to image.For each subregion, which can execute one group of mathematical operation, in output characteristic pattern
Middle generation single value.Then, convolutional layer would generally be exported using ReLU activation primitive, to introduce non-linear rule in a model
Rule.Pond layer carries out down-sampling to the image data that convolutional layer extracts, to reduce the dimension of characteristic pattern, when so as to shorten processing
Between.Common pond algorithm is maximum pond, it can extract the subregion of characteristic pattern, retain the maximum value of subregion, and give up
Other all values.Full articulamentum is to being extracted by convolutional layer and classified by the feature of pond layer down-sampling.In full articulamentum
Each node is all connected to all nodes in preceding layer.The characteristics of classifying layer according to sample image, the parameter of classifier is adjusted,
The classifying quality of classifier under comparative analysis different condition.
Convolutional layer concrete operation process:
The sample image p for being w × w for any given size, utilizes a mapping table of k feature calculation p of extraction
Show y ∈ Rk, y is a k dimensional vector, y(i)It is ith feature φiTo the mapping result of p, i.e. y(i)=φi(p).One width color is logical
Road number is c, and size is the full sample image I of n × n, and feature convolution needs to calculate Feature Mapping to the region of w × w all in I
It indicates, i.e., scans I using the window of w × w size, use some feature φiThe region fallen in window I is reflected
It penetrates, obtains the Feature Mapping figure that a dimension is two.It is M by the Feature Mapping seal to I, then it represents that feature φiTo sample graph
As the mapping result of I (j, k), the dimension of M is k × (n-w+1) × (n-w+1).To result above further genralrlization, training image
Collect the sample image that D includes m width n × n, characteristic window size is w × w, executes the dimension of M after convolution to sample graph image set using φ
Degree is m × k × (n-w+1) × (n-w+1).By utilizing each feature φ in feature set ΦiTo each sample I of data set Di
Convolution is carried out, M ∈ R is obtainedm×k×(n-w+1)×(n-w+1)If n is larger, w is smaller, and the parameter amount in M will be very big, and M is as subsequent process
Input, excessive parameter will make network be difficult to train, and the operation of next step is to carry out to M down-sampled reducing network parameter.
The concrete operation process of pond layer
Using feature set Φ come the feature set M ∈ R of convolution input training image collection Dm×k×(n-w+1)×(n-w+1)Afterwards, theoretically
Subsequent processing can be carried out in obtained all features, however it is inefficient over time and space, need to carry out convolution spy
Sign sampling is to reduce network parameter.If input feature vector indicates that the size of F is n × n, sampling window size is s × s, then carries out to F
The new character representation dimension obtained after sampling is (n/s) × (n/s), and n must be the integral multiple of s here, because of sampling operation
All window scanning modes used are to scan by s step-length great-jump-forward non-overlap, rather than 1 is pressed as convolution operation on F
The continuous overlapping scan of step-length.The feature set being originally inputted has multiple, output of each feature to generating after data set progress convolution
It is multilayer convolution characteristic pattern.
The detailed process of repetitive exercise
It mainly include two stages: 1. propagated forward: i.e. the process that successively carries out of convolution sum pondization, upper one layer of output work
It for next layer of input, and is successively transmitted by activation primitive, finally obtains reality output；2. backpropagation and right value update:
I.e. by the error back propagation of reality output and ideal outlet chamber, obtain the error function of each network layer, then using with
Machine gradient descent method optimizes adjustment to network weight and threshold value.
After convolutional neural networks every iteration 3000-4000 times, tested.Test cell is used to survey in test
Attempt to judge the output accuracy rate of convolutional neural networks model as input convolutional neural networks model；If it is low to export accuracy rate
In first threshold, test cell sends the information for regenerating training image collection to image collection module.Image collection module base
In the information for regenerating training image collection, adjusts pretreatment mode and resequence, generate new training image collection；Structure
Unit is built based on new training image collection, training is iterated to convolutional neural networks model；If exporting accuracy rate lower than the
Two threshold values and be greater than first threshold, continue repetitive exercise；If accuracy rate is greater than or equal to second threshold, construction unit is defeated
Convolution neural network model out.First threshold is 30%-40% in the present embodiment；Second threshold is 70%-80%.Specifically,
First threshold is 30%, second threshold 80%.
Diagnostic module be used to receive digestive endoscope to diagnostic image, based on the successful convolutional neural networks model of training
It treats diagnostic image to be judged, judging result is exported.In the present embodiment, judging result is normal or early carcinoma.
In the present embodiment, image collection module, construction unit, test cell and diagnostic module are using at general computer
Device is managed to run, such as central processing unit (CPU) and graphics processor (GPU) etc..
Embodiment two
A kind of early carcinoma assistant diagnosis system based on artificial intelligence, the difference with embodiment one are that construction unit is defeated
Out when convolution neural network model, pretreatment mode adjusted is labeled as effective pretreatment mode by image collection module.Figure
As obtain module be also used to obtain digestive endoscope after diagnostic image, by effective pretreatment mode treat diagnostic image into
Row pretreatment, is sent to diagnostic module to diagnostic image for pretreated.Diagnostic module be used for receive digestive endoscope to
Diagnostic image is treated diagnostic image based on the successful convolutional neural networks model of training and is judged, judging result is exported.This
In embodiment, judging result is normal or early carcinoma.
Diagnostic image is treated by effective pretreatment mode to be pre-processed, and makes to meet convolutional neural networks to diagnostic image
The input requirements of model improve recognition accuracy.
Embodiment three
A kind of early carcinoma assistant diagnosis system based on artificial intelligence, the difference with embodiment two are, further include judging
Module and environment adjustment module；
Module is judged for obtaining doctor's judging result, by the same judging result to doctor in diagnostic image and diagnosis mould
The judging result of block compares, and judges whether unanimously, if inconsistent, judge module output request and judges information；Volume at present
Product neural network model is still substantially carried out auxiliary diagnosis, the probability of mistaken diagnosis is reduced by double verification, while improving doctor
Diagnosis efficiency.
Judge module, which is also used to receive, has judged information；It is that doctor is correct or convolution is refreshing that information has been judged in the present embodiment
It is correct through network model.When judging that information is inconsistent, judge module output request judge information, can be at this time former doctor into
Row is judged self, is also possible to the step of other doctors carry out third-party judge, pass through judge, can be known it is that doctor examines on earth
Dislocation or convolutional neural networks diagnosis are wrong.
If convolutional neural networks model is correct, the misjudgment information of the doctor is recorded, misjudgment information includes should
To the form subclass or invasive depth subclass under the cancer class and cancer class of diagnostic image.
When misjudgment rate of the same doctor to same form subclass or invasive depth subclass is greater than or equal to third
When threshold value, judges module and export training suggestions information.Third threshold value is 1%-5%, in the present embodiment, specially 2%.It cures at this time
It is raw that error is easy under same subclass, it may be possible to which that certain judgment methods are skilled not yet to be grasped, and is judged module output training and is built
Information is discussed, can be aroused attention, hospital can also arrange to carry out specific aim training to the doctor, to promote diagnosis level.
When misjudgment rate of the same doctor to same form subclass or invasive depth subclass be more than the 4th threshold value and
When less than third threshold value, module is judged to environment adjustment module and sends environment adjusting information.4th threshold value is 0.4%-0.9%,
In the present embodiment, specially 0.5%.Environment adjustment module receive environment adjust information after, to the diagnosis room where doctor into
Row environment is adjusted.Regulative mode is that color space adjustment and the comparison obtained in effective pretreatment mode in mould is obtained from image
Degree adjustment；Based on color space adjustment and setting contrast, brightness and the color of the lighting apparatus of the diagnosis room where doctor are adjusted
Temperature.
Due to the influence of environment light, it may cause a certain form subclass or invasive depth subclass be not easy to be seen by human eye
It observes, to cause the mistaken diagnosis of doctor.If such happen, that shows that the environment light needs of standard are adjusted,
And sample image after imitating the pretreatment of pretreatment mode is easy that convolutional neural networks model is allowed to identify；With reference to effectively pre- place
Reason mode adjusts lighting apparatus, is the habitual thinking in order to get rid of the mankind, allows the thinking of machine to inspire the mankind, allow by machine
The adjusting thinking of verifying is applied in the adjusting to the mankind, can widen the thinking of adjusting.The adjusting thinking of the mankind is not necessarily suitble to
Machine, but the mankind have judgement, can refer to the adjusting thinking of machine, to achieve the effect that reduce the influence of environment light.
Environment adjustment module is also used to adjust the lighting apparatus of the diagnosis room where doctor according to the voice messaging of doctor
Brightness and colour temperature.When the automatic adjustment of environment adjustment module is inappropriate, doctor can actively be intervened, easy to use.This implementation
In example, the lighting apparatus of diagnosis room can use intelligent bulbs, and the means of adjusting can be with reference to day cat spirit, small degree sound equipment or small
Like the adjustings of the intelligent sound boxes to intelligent bulbs such as classmate, belongs to the prior art herein, which is not described herein again.
What has been described above is only an embodiment of the present invention, and the common sense such as well known specific structure and characteristic are not made herein in scheme
Excessive description.It, without departing from the structure of the invention, can be with it should be pointed out that for those skilled in the art
Several modifications and improvements are made, these also should be considered as protection scope of the present invention, these all will not influence what the present invention was implemented
Effect and patent practicability.The scope of protection required by this application should be based on the content of the claims, in specification
The records such as specific embodiment can be used for explaining the content of claim.
Claims (10)
1. a kind of early carcinoma assistant diagnosis system based on artificial intelligence, which is characterized in that including
Image collection module, for obtain with mark digestive endoscope sample image, to sample image carry out pretreatment and
It is randomly ordered, generate training image collection；
Model construction module, for constructing convolutional neural networks model, and based on training image collection to convolutional neural networks model
It is iterated training, is then tested, the successful convolutional neural networks model of training is exported after the completion of test；
Diagnostic module, for obtain digestive endoscope to diagnostic image, based on the successful convolutional neural networks model pair of training
Judged to diagnostic image, judging result is exported.
2. the early carcinoma assistant diagnosis system according to claim 1 based on artificial intelligence, it is characterised in that: the model
Constructing module further includes construction unit, storage unit and test cell, and the test chart of digestive endoscope is prestored in storage unit
Picture, test cell are used in test, and test image is inputted convolutional neural networks model, judges convolutional neural networks model
Accuracy rate is exported, if output accuracy rate is lower than first threshold, test cell regenerates training to image collection module transmission
The information of image set；
Image collection module is adjusted pretreatment mode and is resequenced based on the information for regenerating training image collection, raw
The training image collection of Cheng Xin；Construction unit is iterated training based on new training image collection, to convolutional neural networks model.
3. the early carcinoma assistant diagnosis system according to claim 2 based on artificial intelligence, it is characterised in that: the sample
Image includes one of early esophageal cancer, early carcinoma of stomach, early stage colon cancer or kinds cancer classification, and each cancer class is right
Should there are form subclass and invasive depth subclass.
4. the early carcinoma assistant diagnosis system according to claim 3 based on artificial intelligence, it is characterised in that: the pre- place
Reason include cut, rotate, stretching, translating, color space adjustment, one of setting contrast or a variety of.
5. a kind of early carcinoma assistant diagnosis system based on artificial intelligence according to claim 4, it is characterised in that: image
Module is obtained also all sample images in training image collection are normalized.
6. the early carcinoma assistant diagnosis system according to claim 5 based on artificial intelligence, it is characterised in that: the building
When unit is iterated trained to convolutional neural networks model, the weight parameter of convolutional neural networks model is random initializtion ginseng
Number by propagated forward algorithm constantly transfer characteristic forward, and calculates residual error, then by back-propagation algorithm that residual error is continuous
It transmits backward, corrects weight parameter, continuous iteration is until convergence.
7. the early carcinoma assistant diagnosis system according to claim 6 based on artificial intelligence, it is characterised in that: when convolution mind
After network model every iteration 3000-4000 times, tested.
8. the early carcinoma assistant diagnosis system according to claim 7 based on artificial intelligence, it is characterised in that: the test
When the output accuracy rate of unit judges convolutional neural networks model, if output accuracy rate is greater than first threshold and is lower than the second threshold
Value, continues repetitive exercise.
9. the early carcinoma assistant diagnosis system according to claim 2 based on artificial intelligence, it is characterised in that: described first
Threshold value is 30%-40%.
10. the early carcinoma assistant diagnosis system according to claim 8 based on artificial intelligence, it is characterised in that: described
Two threshold values are 70%-80%.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910814399.XA CN110517256B (en) | 2019-08-30 | 2019-08-30 | Early cancer auxiliary diagnosis system based on artificial intelligence |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910814399.XA CN110517256B (en) | 2019-08-30 | 2019-08-30 | Early cancer auxiliary diagnosis system based on artificial intelligence |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110517256A true CN110517256A (en) | 2019-11-29 |
| CN110517256B CN110517256B (en) | 2022-02-15 |
Family
ID=68628423
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910814399.XA Active CN110517256B (en) | 2019-08-30 | 2019-08-30 | Early cancer auxiliary diagnosis system based on artificial intelligence |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110517256B (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111340094A (en) * | 2020-02-21 | 2020-06-26 | 湘潭大学 | Capsule endoscope image auxiliary classification system and classification method based on deep learning |
| CN111524124A (en) * | 2020-04-27 | 2020-08-11 | 中国人民解放军陆军特色医学中心 | Digestive endoscopy image artificial intelligence auxiliary system for inflammatory bowel disease |
| CN111710394A (en) * | 2020-06-05 | 2020-09-25 | 沈阳智朗科技有限公司 | Artificial intelligence assisted early gastric cancer screening system |
| CN111782529A (en) * | 2020-06-30 | 2020-10-16 | 平安国际智慧城市科技股份有限公司 | Test method and device for auxiliary diagnosis system, computer equipment and storage medium |
| CN112037913A (en) * | 2020-09-15 | 2020-12-04 | 南昌大学 | Intelligent periodontitis detection method and system based on convolutional neural network |
| CN112185548A (en) * | 2020-09-25 | 2021-01-05 | 广州宝荣科技应用有限公司 | Intelligent traditional Chinese medicine diagnosis method and device based on neural network algorithm |
| CN112614128A (en) * | 2020-12-31 | 2021-04-06 | 山东大学齐鲁医院 | System and method for assisting biopsy under endoscope based on machine learning |
| CN112950624A (en) * | 2021-03-30 | 2021-06-11 | 太原理工大学 | Rectal cancer T stage automatic diagnosis method and equipment based on deep convolutional neural network |
| CN113222932A (en) * | 2021-05-12 | 2021-08-06 | 上海理工大学 | Small intestine endoscope image feature extraction method based on multi-convolution neural network integrated learning |
| CN114359279A (en) * | 2022-03-18 | 2022-04-15 | 武汉楚精灵医疗科技有限公司 | Image processing method, image processing device, computer equipment and storage medium |
| CN116705252A (en) * | 2023-06-16 | 2023-09-05 | 脉得智能科技(无锡)有限公司 | Construction method, image classification method, device and medium for prostate cancer diagnosis model |
| CN117132790A (en) * | 2023-10-23 | 2023-11-28 | 南方医科大学南方医院 | Digestive tract tumor diagnosis auxiliary system based on artificial intelligence |
| CN116705252B (en) * | 2023-06-16 | 2024-05-31 | 脉得智能科技(无锡)有限公司 | Construction method, image classification method, device and medium for prostate cancer diagnosis model |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180060722A1 (en) * | 2016-08-30 | 2018-03-01 | Lunit Inc. | Machine learning method and apparatus based on weakly supervised learning |
| CN108229588A (en) * | 2018-02-08 | 2018-06-29 | 重庆师范大学 | A kind of machine learning recognition methods based on deep learning |
| CN108288506A (en) * | 2018-01-23 | 2018-07-17 | 雨声智能科技（上海）有限公司 | A kind of cancer pathology aided diagnosis method based on artificial intelligence technology |
| CN108537791A (en) * | 2018-04-16 | 2018-09-14 | 卡达克科技有限公司 | A kind of stamping parts punching online test method based on machine vision |
| CN109002846A (en) * | 2018-07-04 | 2018-12-14 | 腾讯科技（深圳）有限公司 | A kind of image-recognizing method, device and storage medium |
-
2019
- 2019-08-30 CN CN201910814399.XA patent/CN110517256B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180060722A1 (en) * | 2016-08-30 | 2018-03-01 | Lunit Inc. | Machine learning method and apparatus based on weakly supervised learning |
| CN108288506A (en) * | 2018-01-23 | 2018-07-17 | 雨声智能科技（上海）有限公司 | A kind of cancer pathology aided diagnosis method based on artificial intelligence technology |
| CN108229588A (en) * | 2018-02-08 | 2018-06-29 | 重庆师范大学 | A kind of machine learning recognition methods based on deep learning |
| CN108537791A (en) * | 2018-04-16 | 2018-09-14 | 卡达克科技有限公司 | A kind of stamping parts punching online test method based on machine vision |
| CN109002846A (en) * | 2018-07-04 | 2018-12-14 | 腾讯科技（深圳）有限公司 | A kind of image-recognizing method, device and storage medium |
Non-Patent Citations (2)
| Title |
|---|
| 朱国策: "基于深度卷积神经网络的医学图像肺结节检测方法研究", 《中国优秀硕士学位论文全文数据库医药卫生科技辑》 * |
| 李义: "基于相关学习神经网络的图像识别方法研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111340094A (en) * | 2020-02-21 | 2020-06-26 | 湘潭大学 | Capsule endoscope image auxiliary classification system and classification method based on deep learning |
| CN111524124A (en) * | 2020-04-27 | 2020-08-11 | 中国人民解放军陆军特色医学中心 | Digestive endoscopy image artificial intelligence auxiliary system for inflammatory bowel disease |
| CN111710394A (en) * | 2020-06-05 | 2020-09-25 | 沈阳智朗科技有限公司 | Artificial intelligence assisted early gastric cancer screening system |
| CN111782529B (en) * | 2020-06-30 | 2023-08-01 | 深圳平安智慧医健科技有限公司 | Test method and device for auxiliary diagnosis system, computer equipment and storage medium |
| CN111782529A (en) * | 2020-06-30 | 2020-10-16 | 平安国际智慧城市科技股份有限公司 | Test method and device for auxiliary diagnosis system, computer equipment and storage medium |
| CN112037913A (en) * | 2020-09-15 | 2020-12-04 | 南昌大学 | Intelligent periodontitis detection method and system based on convolutional neural network |
| CN112037913B (en) * | 2020-09-15 | 2022-08-30 | 南昌大学 | Periodontitis intelligent detection system based on convolutional neural network |
| CN112185548A (en) * | 2020-09-25 | 2021-01-05 | 广州宝荣科技应用有限公司 | Intelligent traditional Chinese medicine diagnosis method and device based on neural network algorithm |
| CN112614128A (en) * | 2020-12-31 | 2021-04-06 | 山东大学齐鲁医院 | System and method for assisting biopsy under endoscope based on machine learning |
| CN112950624A (en) * | 2021-03-30 | 2021-06-11 | 太原理工大学 | Rectal cancer T stage automatic diagnosis method and equipment based on deep convolutional neural network |
| CN113222932A (en) * | 2021-05-12 | 2021-08-06 | 上海理工大学 | Small intestine endoscope image feature extraction method based on multi-convolution neural network integrated learning |
| CN114359279A (en) * | 2022-03-18 | 2022-04-15 | 武汉楚精灵医疗科技有限公司 | Image processing method, image processing device, computer equipment and storage medium |
| CN114359279B (en) * | 2022-03-18 | 2022-06-03 | 武汉楚精灵医疗科技有限公司 | Image processing method, image processing device, computer equipment and storage medium |
| CN116705252A (en) * | 2023-06-16 | 2023-09-05 | 脉得智能科技(无锡)有限公司 | Construction method, image classification method, device and medium for prostate cancer diagnosis model |
| CN116705252B (en) * | 2023-06-16 | 2024-05-31 | 脉得智能科技(无锡)有限公司 | Construction method, image classification method, device and medium for prostate cancer diagnosis model |
| CN117132790A (en) * | 2023-10-23 | 2023-11-28 | 南方医科大学南方医院 | Digestive tract tumor diagnosis auxiliary system based on artificial intelligence |
| CN117132790B (en) * | 2023-10-23 | 2024-02-02 | 南方医科大学南方医院 | Digestive tract tumor diagnosis auxiliary system based on artificial intelligence |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110517256B (en) | 2022-02-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110517256A (en) | A kind of early carcinoma assistant diagnosis system based on artificial intelligence | |
| CN110363768A (en) | A kind of early carcinoma lesion horizon prediction auxiliary system based on deep learning | |
| Li et al. | Automatic detection of diabetic retinopathy in retinal fundus photographs based on deep learning algorithm | |
| Wang et al. | Patch-based output space adversarial learning for joint optic disc and cup segmentation | |
| CN109376636B (en) | Capsule network-based eye fundus retina image classification method | |
| Bilal et al. | Diabetic retinopathy detection and classification using mixed models for a disease grading database | |
| CN109272048B (en) | Pattern recognition method based on deep convolutional neural network | |
| CN109635835A (en) | A kind of breast lesion method for detecting area based on deep learning and transfer learning | |
| CN110033456A (en) | A kind of processing method of medical imaging, device, equipment and system | |
| Ran et al. | Cataract detection and grading based on combination of deep convolutional neural network and random forests | |
| CN112101451B (en) | Breast cancer tissue pathological type classification method based on generation of antagonism network screening image block | |
| CN110473192A (en) | Digestive endoscope image recognition model training and recognition methods, apparatus and system | |
| CN109919928A (en) | Detection method, device and the storage medium of medical image | |
| CN109166105A (en) | The malignancy of tumor risk stratification assistant diagnosis system of artificial intelligence medical image | |
| CN110516685A (en) | Lenticular opacities degree detecting method based on convolutional neural networks | |
| Costa et al. | Eyequal: Accurate, explainable, retinal image quality assessment | |
| CN110176295A (en) | A kind of real-time detecting method and its detection device of Gastrointestinal Endoscopes lower portion and lesion | |
| CN109508755A (en) | A kind of Psychological Evaluation method based on image cognition | |
| CN110363751A (en) | A kind of big enteroscope polyp detection method based on generation collaborative network | |
| Tajudin et al. | Deep learning in the grading of diabetic retinopathy: A review | |
| CN111462082A (en) | Focus picture recognition device, method and equipment and readable storage medium | |
| CN109480775A (en) | A kind of icterus neonatorum identification device based on artificial intelligence, equipment, system | |
| Kumar et al. | A multi-thresholding-based discriminative neural classifier for detection of retinoblastoma using CNN models | |
| CN110503636A (en) | Parameter regulation means, lesion prediction technique, parameter adjustment controls and electronic equipment | |
| CN117237711A (en) | Bimodal fundus image classification method based on countermeasure learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |