CN110176295A - A kind of real-time detecting method and its detection device of Gastrointestinal Endoscopes lower portion and lesion - Google Patents
A kind of real-time detecting method and its detection device of Gastrointestinal Endoscopes lower portion and lesion Download PDFInfo
- Publication number
- CN110176295A CN110176295A CN201910511309.XA CN201910511309A CN110176295A CN 110176295 A CN110176295 A CN 110176295A CN 201910511309 A CN201910511309 A CN 201910511309A CN 110176295 A CN110176295 A CN 110176295A
- Authority
- CN
- China
- Prior art keywords
- video
- image
- focus characteristic
- genius loci
- real
- 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
- 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
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H30/00—ICT specially adapted for the handling or processing of medical images
- G16H30/20—ICT specially adapted for the handling or processing of medical images for handling medical images, e.g. DICOM, HL7 or PACS
-
- 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
- G06T2207/10016—Video; Image sequence
-
- 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
- G06T2207/30028—Colon; Small intestine
-
- 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
- G06T2207/30092—Stomach; Gastric
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Epidemiology (AREA)
- Primary Health Care (AREA)
- Public Health (AREA)
- Endoscopes (AREA)
Abstract
The real-time detecting method of a kind of Gastrointestinal Endoscopes lower portion and illness disclosed by the invention, comprising the following steps: the 1. collected video stream datas of synchronous acquisition stomach and intestine mirror device institute；2. pair preprocessing image data carries out image recognition, and obtains focus characteristic image and genius loci image；3. one side carries out lesion marking to focus characteristic image, position label on the other hand is carried out to genius loci image；4. by lesion detection result and position detection result synchronize be identified expression on the video stream data that gets, and export picture and text result.The invention also discloses a kind of detection devices of real-time detecting method for realizing above-mentioned Gastrointestinal Endoscopes lower portion and illness.The present invention is by synchronizing label to focus characteristic and genius loci in synchronization video flow data, and picture and text result will be exported after the result of sync mark statistical analysis, efficiently reduce and avoid failing to pinpoint a disease in diagnosis in video and mistaken diagnosis, optimization doctor edits six layers of report, greatly reduces the working strength of doctor.
Description
Technical field
The present invention relates to depth learning technology field more particularly to a kind of real-time detection sides of Gastrointestinal Endoscopes lower portion and lesion
Method and its detection device.
Background technique
Conventional gastroscope enteroscopy is the important method and approach that clinician has found people digest's tract disease.In recent years,
With the construction and accumulation of Computing ability being greatly improved with hospital image data base, the spy based on deep learning algorithm
It surveys model and plays powerful advantage in medical imaging analysis field, progress into clinical application.
The work difficulty and intensity that discovery operates doctor when we further investigate the Gastrointestinal Endoscopes operating process in reality are very
It is big.Social background is that the social medtcal insurance range in China expands year by year, and scope doctor's quantity notch is big, and cultivation cycle is long, each big-and-middle
Small hospital's row carries out the doctor of conventional endoscope inspection operation all in overload state, and doctor has been routinely accomplished an endoscope
The time of inspection is about 6~15 minutes.Inspection result is by operation qualified teachers' experience of doctor, qualification, the state of mind, tired
The various factors such as labor degree influence, this will lead to doctor and is difficult to accomplish that the accurate acquisition to lesion and position, scope report are comprehensive
Full and accurate record.It fails to pinpoint a disease in diagnosis unavoidable with mistaken diagnosis.
Presently disclosed related endoscopes secondary diagnostic module and system, it is most of to use depth convolutional neural networks side
Method.For example, number of patent application is that a kind of gastroscope image disclosed in the Chinese invention patent application of 201810690051.X is adopted automatically
Collecting system and method.But with research deeply and extensively, deep learning in versatility, flexibility and adaptability with people
Class vision system is compared, and also there is a big difference.And when encountering complicated natural image, deep learning may further encounter mechanism
Property it is difficult.Therefore, special for the medical operating operating process feature, human organ dynamic change and lesion image specifically studied
Sign, cannot apply simple epidemic algorithms, and need to develop autonomous combinational algorithm.For this purpose, the applicant passes through beneficial spy
Rope and research, have found solution to the problems described above, and technical solution described below is to generate in this background
's.
Summary of the invention
The first technical problem to be solved by the present invention is: provide in view of the deficiencies of the prior art it is a kind of reduction and keep away
Exempt from failing to pinpoint a disease in diagnosis the step of finding with mistaken diagnosis, simplified doctor and acquire lesion image, optimizing the stomach and intestine that doctor edits report in video
The real-time detecting method of mirror lower portion and lesion.
The second technical problem to be solved by the present invention is: providing a kind of above-mentioned Gastrointestinal Endoscopes lower portion and illness realized
The detection device of real-time detecting method.
The real-time detecting method of a kind of Gastrointestinal Endoscopes lower portion and illness as first aspect present invention, including following step
It is rapid:
Step S10, it is synchronous to obtain the collected video data of stomach and intestine mirror device institute, and the video data got is carried out
Pretreatment forms preprocessing image data；
Step S20 carries out image recognition processing to the preprocessing image data using deep neural network built-up pattern,
Identify that the focus characteristic image with focus characteristic and the position with genius loci are special from the preprocessing image data
Levy image；
On the one hand step S30 by type all carries out the single lesion in the focus characteristic image with morphological feature
Screening and classification, and whole labels are carried out using lesion of the Imagelabeline tool to same type, to generate focus characteristic mark
On the other hand note carries out part screening and classification to the single position in the genius loci image with morphological feature by type,
And part label is carried out using position of the Graph Scribbline tool to same type, to generate position signature；
Step S40, by focus characteristic label and genius loci mark synchronize carry out on the video data that gets
Identity expression, and by Jing Guo identity expression video data carry out real-time display, while by focus characteristic label and genius loci
Mark for statistical analysis, output picture and text result.
In a preferred embodiment of the invention, in the step S10, the described pair of video data got is carried out
Pretreatment, comprising the following steps:
Step S11 obtains video-input image according to certain frequency acquisition from video data；
It is normal or abnormal to judge that the content of the video-input image got belongs to by step S12, if be judged as belong to it is different
Often, then S13 is entered step, if being judged as normal, enters step S14；
Step S13 carries out noise reduction process to the video-input image got, subsequently into step S14；
Step S14, the video-input image after the video-input image that will acquire or noise reduction process is as pretreatment
Image data is exported.
In a preferred embodiment of the invention, in the step S12, it is abnormal to judge that video-input image belongs to
Situation includes following one or more contents: the synchronizing information of A. video-input image exists abnormal；B. video-input image
Type is not belonging to detection type；C. the length of video-input image and wide ratio exist abnormal；D. the length of video-input image with
Wide ratio is normal, but specific size exceeds predetermined threshold range；E. the movement rate of video-input image exceeds predetermined threshold
Range；F. the gray scale of video-input image exceeds predetermined threshold range.
In a preferred embodiment of the invention, in the step S20, the deep neural network built-up pattern packet
Include genius loci detection CNN model and focus characteristic detection CNN model；The genius loci detection CNN model for visiting in real time
The probability in preprocessing image data with typical parts feature is surveyed, and exports the detection knot for having the genius loci of predetermined probability
Fruit；The focus characteristic detection CNN model is used for the probability in real-time detection preprocessing image data with typical lesions feature,
And export the detection result for having the genius loci of predetermined probability.
A kind of spy of real-time detecting method for realizing above-mentioned Gastrointestinal Endoscopes lower portion and illness as second aspect of the present invention
Survey device, comprising:
Video acquiring module, the video acquiring module obtain the collected video fluxion of stomach and intestine mirror device institute for synchronous
According to, and the video stream data got is pre-processed；
Video detection module, the video detection module are used for using deep neural network built-up pattern to the pretreatment
Image data carries out image recognition processing, and the focus characteristic figure with focus characteristic is identified from the preprocessing image data
Picture and genius loci image with genius loci；
Image tagged module, described image mark module is on the one hand by type with morphological feature to the focus characteristic
Single lesion in image is all screened and is classified, and is carried out using lesion of the Imagelabeline tool to same type
All labels, to generate focus characteristic label, on the other hand by type with morphological feature to the list in the genius loci image
One position carries out part screening and classification, and carries out part mark using position of the Graph Scribbline tool to same type
Note, to generate position signature；And
Output module is detected, the detection output module is used for focus characteristic label and genius loci label same
It walks and is identified expression on the video data got, and the video data Jing Guo identity expression is subjected to real-time display, simultaneously
Focus characteristic label and genius loci label is for statistical analysis, export picture and text result.
Due to using technical solution as above, the beneficial effects of the present invention are: the present invention passes through in synchronization video fluxion
Label synchronized to focus characteristic and genius loci in, and video data is subjected to real-time display, while by sync mark
Result statistical analysis after export picture and text as a result, efficiently reduce and avoid failing to pinpoint a disease in diagnosis in video and mistaken diagnosis, simplify doctor and seek
The step of looking for and acquire lesion image, optimization doctor edit report, greatly reduce the working strength of doctor.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.
Fig. 1 is flow diagram of the invention.
Fig. 2 is of the invention to the pretreated flow diagram of video data progress.
Specific embodiment
In order to be easy to understand the technical means, the creative features, the aims and the efficiencies achieved by the present invention, tie below
Conjunction is specifically illustrating, and the present invention is further explained.
Referring to Fig. 1, what is provided in figure is the real-time detecting method of a kind of Gastrointestinal Endoscopes lower portion and illness, including following step
It is rapid:
Step S10, it is synchronous to obtain the collected video data of stomach and intestine mirror device institute, and the video data got is carried out
Pretreatment forms preprocessing image data；
Step S20 carries out image recognition processing to the preprocessing image data using deep neural network built-up pattern,
The focus characteristic image with focus characteristic and the genius loci figure with genius loci are identified from preprocessing image data
Picture；
On the one hand step S30 by type all carries out the single lesion in the focus characteristic image with morphological feature
Screening and classification, and whole labels are carried out using lesion of the Imagelabeline tool to same type, to generate focus characteristic mark
On the other hand note carries out part screening and classification to the single position in the genius loci image with morphological feature by type,
And part label is carried out using position of the Graph Scribbline tool to same type, to generate position signature；
Step S40, by focus characteristic label and genius loci mark synchronize carry out on the video data that gets
Identity expression, and by Jing Guo identity expression video data carry out real-time display, while by focus characteristic label and genius loci
Mark for statistical analysis, output picture and text result.
Above-mentioned each step is described in detail below:
In step slo, referring to fig. 2, the video data got is pre-processed, comprising the following steps:
Step S11, according to certain frequency acquisition video-input image from video data.
It is normal or abnormal to judge that the content of the video-input image got belongs to by step S12, if be judged as belong to it is different
Often, then S13 is entered step, if being judged as normal, enters step S14.Wherein, when video-input image include with the next item down or
Then be judged as belonging to abnormal situation when multinomial content: the synchronizing information of A. video-input image exists abnormal；B. video is defeated
The type for entering image is not belonging to detection type；C. the length of video-input image and wide ratio exist abnormal；D. video input figure
The length of picture and wide ratio are normal, but specific size exceeds predetermined threshold range；E. the movement rate of video-input image exceeds
Predetermined threshold range；F. the gray scale of video-input image exceeds predetermined threshold range.
Step S13 carries out noise reduction process to the video-input image got, subsequently into step S14.Here noise reduction
The purpose of processing is to suppress picture noise for authority data, improve the precision of subsequent step.
Step S14, the video-input image after the video-input image that will acquire or noise reduction process is as pretreatment
Image data is exported.
In step S20, deep neural network built-up pattern includes genius loci detection CNN model and focus characteristic detection
CNN model.Genius loci detects CNN model for general with typical parts feature in real-time detection preprocessing image data
Rate, and export the detection result for having the genius loci of predetermined probability.Focus characteristic detection CNN model is located in advance for real-time detection
The probability in image data with typical lesions feature is managed, and exports the detection result for having the genius loci of predetermined probability.
Wherein, predetermined probability refers to the adjusting to detection specificity index, so that the deep neural network built-up pattern is protected
Sensibility that the important indicator for holding appropriate false positive is clinical expert team to be prompted to improve model to lesion and special
Property, on the basis of comparative analysis naked eyes judge the difference of the image of image and the judgement of AI model, CNN mould is detected to focus characteristic
The output image probability of each classification of type or genius loci detection CNN model carries out specific aim adjustment.In the present embodiment, in advance
The target value of probability is determined between 0.801-0.999.The adjustment of predetermined probability is because being known as: the picture number of a. training set and test set
Amount；B. target value of the user (operation doctor) to the sensibility and specificity of model；C. hypersensitivity and appropriate false positive are kept
Ratio between lesion；D. the ratio of high specific is kept；E. other clinical application standards.
In the present embodiment, genius loci detection CNN model and focus characteristic detection CNN model are based on CNN convolution mind
It is constructed through network model, CNN convolutional neural networks model is open source algorithm, belongs to techniques well known, herein no longer
It is described in detail.
Typical parts include 16 kinds of typical parts of 35 kinds of typical parts and inner wall of intestine of oesophagus stomach function regulating inner wall.Oesophagus and
35 kinds of typical parts of stomach inner wall specifically include: pars oralis pharyngis, Upper Esophageal, Esophageal Middle Segment, distal esophagus, cardia closure, cardia
Mouth, dentate line, cardia body of stomach boundary, stomach bottom antetheca (telescope direct+reversing face), stomach bottom rear wall (telescope direct+reversing face), stomach bottom it is big it is curved (telescope direct+
Reversing face), stomach bottom small curved (telescope direct+reversing face), body of stomach upper front wall (telescope direct+reversing face), body of stomach top rear wall (telescope direct+reversing face), stomach
Before body top big curved (telescope direct+reversing face), body of stomach top small curved (telescope direct+reversing face), body of stomach middle part (telescope direct+reversing face), body of stomach lower part
Wall (telescope direct+reversing face), body of stomach lower part rear wall (telescope direct+reversing face), body of stomach lower part big curved (telescope direct+reversing face), body of stomach lower part it is small it is curved (just
Mirror+reversing face), stomach angle antetheca, small curved, the stomach angle antrum in stomach angle rear wall, stomach angle have a common boundary, big curved, the antrum antetheca of antrum, antrum rear wall,
The small curved, lateral pylorus of antrum, pyloric canal, duodenal bulb, duodenal bulb drop boundary, descendant duodenum, duodenum water
Flat portion.16 kinds of typical parts of inner wall of intestine specifically include: crissum, anal canal, Rb, Ra, Rs, sigmoid colon, drop second are had a common boundary, drop is tied
Intestines, splenic flexure, transverse colon, hepatic flexure of colon, colon ascendens, ileocecus, ileocaecal sphineter, appendix hole, terminal ileum.
In step s 30, focus characteristic label is based on supervised learning algorithm (Supervised Learning
Algorithms realized on the basis of), the supervised learning algorithm be the CNN model based on Imagelabeline, be by
A mode of learning (learning model) is acquired or established to training dataset, and mode speculates new example according to this.The calculation
Method requires specific input/output, needs to determine to use which kind of data as example.In this case, each mode of learning,
The classification that should belong to is known.Supervised learning algorithm can shorten learning time, while can also improve nicety of grading.Supervision is learned
Practising algorithm mainly includes neural network, support vector machine, nearest-neighbors method, naive Bayesian method, decision tree etc..
In addition, Recovering also can be used during the focus characteristic in focus characteristic image is marked
Tool judges because of the focus characteristic for being difficult to recognize caused by the reasons such as spherical aberration, accurately holds the state of lesion, mentions
The accuracy rate of height label.Specifically, deterministic process of the Recovering tool to focus characteristic are as follows: 1. distant views label；2. close shot
Laterally (0-15 degree) marks；3. close shot overlooks (70-90 degree) label；4. judging the parting of lesion.
In step s 30, genius loci label is based on semi-supervised learning algorithm (Semi-Supervised Learning
Algorithms it is realized on the basis of), which is based on energy-optimised algorithm Graph
The CNN model of Scribbline is a kind of learning method that supervised learning is combined with unsupervised learning.Semi-supervised learning makes
With a large amount of Unlabeled data, and flag data is used simultaneously, Lai Jinhang pattern-recognition work.When use semi-supervised learning
When, the personnel for being engaged in image tagged will be reduced, while the result of training will be instructed closer to probabilistic realistic situation is full of
Higher accuracy will be obtained by practicing expection.Part screening and part label in genius loci label are semi-supervised learning algorithms
Operating method and feature, it is therefore an objective to weaken the indicative of image tagged, improve the accuracy close to true operation.
During the genius loci in genius loci image is marked using semi-supervised learning algorithm, first by portion
The pixel class information of position characteristic image propagates to other unlabelled pixels via semi-supervised learning algorithm, completes all training
The markers work of image.Wherein, pixel class information is the image information generated through Graph Scribbline label；Pass through
The probability assignments of energy function in Graph Scribbline complete the classification to all training images.Then, then by position mark
The all images of note train CNN model, realize CNN model training and prediction section bit image.And in labeling process, first generate
All super-pixel are marked in super-pixel, the method for being then based on Graph Scribbline.Graph
The energy function of Scribbline are as follows: the energy function of Graph cut
E=∑iψi(yi|X,S)+∑i,jψij(yi,yj,X)
Wherein, i, j are image segmentation node；Y indicates vertex pixel, and X indicates adjacent area pixel, and S indicates corresponding vertex
Weight；∑ i ψ i (yi | X, S) indicate that realm data item, ∑ i, j ψ ij (yi, yj, X) indicate edge smoothing item.
Energy function includes two parts, and a part is the energy-optimised algorithm of Graph Scribbline, and another part is portion
The probability that position feature decision CNN predicts the super-pixel.Entire optimization process is exactly to seek graph cut energy function
The process of optimal value is calculated with genius loci detection CNN parametric joint:
Wherein, i, j are image segmentation node；Y indicates vertex pixel, and X indicates adjacent area pixel, and S indicates corresponding vertex
Weight；The parameter of Ф expression position feature detection CNN model.
The optimization of above formula is realized by the optimal value of alternately summation.
In step s 40, focus characteristic is marked and is identified expression on synchronizing the video data got, including with
Lower four kinds of situations:
1. the CNN model based on Imagelabeline is detected less than lesion, then result is not exported.
2. the CNN model based on Imagelabeline is to certain class lesion, with the movement of video image, detection probability is arrived
When up to predetermined probability, output lesion word content to video pictures.
3. the CNN model based on Imagelabeline is to certain class lesion, curve (solid line or dotted line) in different colors
Lesion word content is exported to video pictures.
4. the CNN model based on Imagelabeline calls corresponding speech text to certain class lesion, through external sound equipment
Equipment is played simultaneously.
Administrator can also be adjusted according to the output frequency and expression way of above-mentioned four kinds of situations.
In step s 40, genius loci is marked and is identified expression on synchronizing the video data got, including with
Lower four kinds of situations:
1. the CNN model based on Graph Scribbline is detected less than position, then result is not exported.
2. based on the CNN model of Graph Scribbline to certain class position, with the movement of video image, detection probability
When reaching predetermined probability, output lesion word content to video pictures.
3. based on the CNN model of Graph Scribbline to certain class position, curve (solid line or void in different colors
Line) lesion word content is exported to video pictures.
4. corresponding speech text is called to certain class position based on the CNN model of Graph Scribbline, through external sound
Equipment is rung to be played simultaneously.
Administrator can also be adjusted according to the output frequency and expression way of above-mentioned four kinds of situations.
The detection device of the real-time detecting method of Gastrointestinal Endoscopes lower portion and illness of the invention, including video acquiring module,
Video detection module, image tagged module and detection output module.
Video acquiring module can connect the 3G-SDI interface of stomach and intestine mirror device by image pick-up card, be used to synchronize acquisition
The collected video stream data of stomach and intestine mirror device institute, and the video stream data got is pre-processed, form pretreatment figure
As data.
Video detection module is used to carry out image recognition to preprocessing image data using deep neural network built-up pattern
Processing identifies that the focus characteristic image with focus characteristic and the position with genius loci are special from preprocessing image data
Levy image.
Image tagged module is on the one hand whole to the single lesion in focus characteristic image with morphological feature by type
It is screened and is classified, and carry out whole labels using lesion of the Imagelabeline tool to same type, to generate lesion spy
On the other hand sign label carries out part screening and classification to the single position in genius loci image with morphological feature by type,
And part label is carried out using position of the Graph Scribbline tool to same type, to generate position signature.
Detection output module be used to marking focus characteristic label and genius loci synchronize the video data that gets
It is identified expression, and the video data Jing Guo identity expression is subjected to real-time display, while by focus characteristic label and position
Signature is for statistical analysis, exports picture and text result.
The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention.The technology of the industry
Personnel are it should be appreciated that the present invention is not limited to the above embodiments, and the above embodiments and description only describe this
The principle of invention, without departing from the spirit and scope of the present invention, various changes and improvements may be made to the invention, these changes
Change and improvement all fall within the protetion scope of the claimed invention.The claimed scope of the invention by appended claims and its
Equivalent thereof.
Claims (5)
1. a kind of real-time detecting method of Gastrointestinal Endoscopes lower portion and illness, which comprises the following steps:
Step S10, it is synchronous to obtain the collected video data of stomach and intestine mirror device institute, and the video data got is located in advance
Reason forms preprocessing image data；
Step S20 carries out image recognition processing to the preprocessing image data using deep neural network built-up pattern, from institute
It states and identifies the image of the focus characteristic with focus characteristic and the genius loci figure with genius loci in preprocessing image data
Picture；
On the one hand step S30 by type all screens the single lesion in the focus characteristic image with morphological feature
And classification, and whole labels are carried out using lesion of the Imagelabeline tool to same type, to generate focus characteristic label,
On the other hand part screening and classification are carried out to the single position in the genius loci image with morphological feature by type, and made
Part label is carried out with position of the Graph Scribbline tool to same type, to generate position signature；
Step S40, by focus characteristic label and genius loci mark synchronize be identified on the video data that gets
Expression, and the video data Jing Guo identity expression is subjected to real-time display, while focus characteristic label and genius loci being marked
It is for statistical analysis, export picture and text result.
2. the real-time detecting method of Gastrointestinal Endoscopes lower portion and illness as described in claim 1, which is characterized in that in the step
In S10, the described pair of video data got is pre-processed, comprising the following steps:
Step S11 obtains video-input image according to certain frequency acquisition from video data；
It is normal or abnormal to judge that the content of the video-input image got belongs to, belongs to exception if being judged as by step S12,
S13 is then entered step, if being judged as normal, enters step S14；
Step S13 carries out noise reduction process to the video-input image got, subsequently into step S14；
Step S14, the video-input image after the video-input image that will acquire or noise reduction process is as pretreatment image
Data are exported.
3. the real-time detecting method of Gastrointestinal Endoscopes lower portion and illness as claimed in claim 2, which is characterized in that in the step
In S12, judging that video-input image belongs to abnormal situation includes following one or more contents: A. video-input image it is same
It walks information and there is exception；B. the type of video-input image is not belonging to detection type；C. the length of video-input image and wide ratio
Example exists abnormal；D. the length of video-input image and wide ratio are normal, but specific size exceeds predetermined threshold range；E. video
The movement rate of input picture exceeds predetermined threshold range；F. the gray scale of video-input image exceeds predetermined threshold range.
4. the real-time detecting method of Gastrointestinal Endoscopes lower portion and illness as described in claim 1, which is characterized in that in the step
In S20, the deep neural network built-up pattern includes genius loci detection CNN model and focus characteristic detection CNN model；Institute
Genius loci detection CNN model is stated for having the probability of typical parts feature in real-time detection preprocessing image data, and defeated
Provide the detection result of the genius loci of standby predetermined probability；The focus characteristic detection CNN model is pre-processed for real-time detection
With the probability of typical lesions feature in image data, and export the detection result for having the genius loci of predetermined probability.
5. a kind of real-time detecting method for realizing Gastrointestinal Endoscopes lower portion and illness according to any one of claims 1 to 4
Detection device characterized by comprising
Video acquiring module, the video acquiring module obtain the collected video stream data of stomach and intestine mirror device institute for synchronous,
And the video stream data got is pre-processed；
Video detection module, the video detection module are used for using deep neural network built-up pattern to the pretreatment image
Data carry out image recognition processing, identified from the preprocessing image data focus characteristic image with focus characteristic and
Genius loci image with genius loci；
Image tagged module, described image mark module is on the one hand by type with morphological feature to the focus characteristic image
In single lesion all screened and classified, and the lesion of same type is carried out all using Imagelabeline tool
Label, to generate focus characteristic label, on the other hand by type with morphological feature to the single portion in the genius loci image
Position carries out part screening and classification, and carries out part label using position of the GraphScribbline tool to same type, with life
It is marked at genius loci；And
Detect output module, the detection output module be used for by the focus characteristic label with genius loci mark synchronize obtain
It is identified expression on the video data got, and the video data Jing Guo identity expression is subjected to real-time display, while will be sick
Stove signature and genius loci label are for statistical analysis, export picture and text result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910511309.XA CN110176295A (en) | 2019-06-13 | 2019-06-13 | A kind of real-time detecting method and its detection device of Gastrointestinal Endoscopes lower portion and lesion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910511309.XA CN110176295A (en) | 2019-06-13 | 2019-06-13 | A kind of real-time detecting method and its detection device of Gastrointestinal Endoscopes lower portion and lesion |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110176295A true CN110176295A (en) | 2019-08-27 |
Family
ID=67697201
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910511309.XA Pending CN110176295A (en) | 2019-06-13 | 2019-06-13 | A kind of real-time detecting method and its detection device of Gastrointestinal Endoscopes lower portion and lesion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110176295A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111739024A (en) * | 2020-08-28 | 2020-10-02 | 安翰科技(武汉)股份有限公司 | Image recognition method, electronic device and readable storage medium |
| CN112364678A (en) * | 2019-09-30 | 2021-02-12 | 山东省科学院海洋仪器仪表研究所 | Buoy identification and positioning method based on NPU board card and shipborne device thereof |
| CN113268623A (en) * | 2021-06-01 | 2021-08-17 | 上海市第一人民医院 | Artificial intelligence gastroscope image recognition processing system |
| CN113744266A (en) * | 2021-11-03 | 2021-12-03 | 武汉楚精灵医疗科技有限公司 | Method and device for displaying focus detection frame, electronic equipment and storage medium |
| CN114494247A (en) * | 2022-04-01 | 2022-05-13 | 武汉大学 | Jaggy line segmentation method, jaggy line segmentation device, computer device, and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107665355A (en) * | 2017-09-27 | 2018-02-06 | 重庆邮电大学 | A kind of agricultural pests detection method based on region convolutional neural networks |
| CN107730489A (en) * | 2017-10-09 | 2018-02-23 | 杭州电子科技大学 | Wireless capsule endoscope small intestine disease variant computer assisted detection system and detection method |
| CN109102491A (en) * | 2018-06-28 | 2018-12-28 | 武汉大学人民医院（湖北省人民医院） | A kind of gastroscope image automated collection systems and method |
-
2019
- 2019-06-13 CN CN201910511309.XA patent/CN110176295A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107665355A (en) * | 2017-09-27 | 2018-02-06 | 重庆邮电大学 | A kind of agricultural pests detection method based on region convolutional neural networks |
| CN107730489A (en) * | 2017-10-09 | 2018-02-23 | 杭州电子科技大学 | Wireless capsule endoscope small intestine disease variant computer assisted detection system and detection method |
| CN109102491A (en) * | 2018-06-28 | 2018-12-28 | 武汉大学人民医院（湖北省人民医院） | A kind of gastroscope image automated collection systems and method |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112364678A (en) * | 2019-09-30 | 2021-02-12 | 山东省科学院海洋仪器仪表研究所 | Buoy identification and positioning method based on NPU board card and shipborne device thereof |
| CN112364678B (en) * | 2019-09-30 | 2023-04-07 | 山东省科学院海洋仪器仪表研究所 | Buoy identification and positioning method based on NPU board card and shipborne device thereof |
| CN111739024A (en) * | 2020-08-28 | 2020-10-02 | 安翰科技(武汉)股份有限公司 | Image recognition method, electronic device and readable storage medium |
| CN111739024B (en) * | 2020-08-28 | 2020-11-24 | 安翰科技(武汉)股份有限公司 | Image recognition method, electronic device and readable storage medium |
| CN113268623A (en) * | 2021-06-01 | 2021-08-17 | 上海市第一人民医院 | Artificial intelligence gastroscope image recognition processing system |
| CN113744266A (en) * | 2021-11-03 | 2021-12-03 | 武汉楚精灵医疗科技有限公司 | Method and device for displaying focus detection frame, electronic equipment and storage medium |
| CN114494247A (en) * | 2022-04-01 | 2022-05-13 | 武汉大学 | Jaggy line segmentation method, jaggy line segmentation device, computer device, and storage medium |
| CN114494247B (en) * | 2022-04-01 | 2022-06-21 | 武汉大学 | Jaggy line segmentation method, jaggy line segmentation device, computer device, and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110176295A (en) | A kind of real-time detecting method and its detection device of Gastrointestinal Endoscopes lower portion and lesion | |
| US10860930B2 (en) | Learning method, image recognition device, and computer-readable storage medium | |
| Tajbakhsh et al. | Automated polyp detection in colonoscopy videos using shape and context information | |
| CN109102491A (en) | A kind of gastroscope image automated collection systems and method | |
| US20190252073A1 (en) | System and method for diagnosing gastrointestinal neoplasm | |
| Mackiewicz et al. | Wireless capsule endoscopy color video segmentation | |
| CN110599448A (en) | Migratory learning lung lesion tissue detection system based on MaskScoring R-CNN network | |
| CN109858540A (en) | A kind of medical image recognition system and method based on multi-modal fusion | |
| CN113379693B (en) | Capsule endoscope key focus image detection method based on video abstraction technology | |
| Ghosh et al. | A statistical feature based novel method to detect bleeding in wireless capsule endoscopy images | |
| CN110517256A (en) | A kind of early carcinoma assistant diagnosis system based on artificial intelligence | |
| Ghosh et al. | Automatic bleeding detection in wireless capsule endoscopy based on RGB pixel intensity ratio | |
| CN110363768A (en) | A kind of early carcinoma lesion horizon prediction auxiliary system based on deep learning | |
| CN109711389A (en) | A kind of milking sow posture conversion identification method based on Faster R-CNN and HMM | |
| Pogorelov et al. | Deep learning and handcrafted feature based approaches for automatic detection of angiectasia | |
| Sadasivan et al. | High accuracy patch-level classification of wireless capsule endoscopy images using a convolutional neural network | |
| Sun et al. | A novel gastric ulcer differentiation system using convolutional neural networks | |
| CN111179252A (en) | Cloud platform-based digestive tract disease focus auxiliary identification and positive feedback system | |
| CN111493805A (en) | State detection device, method, system and readable storage medium | |
| WO2022257314A1 (en) | Image detection method, related training method, related apparatus, device, and medium | |
| KahsayGebreslassie et al. | Automated gastrointestinal disease recognition for endoscopic images | |
| CN112419246A (en) | Depth detection network for quantifying esophageal mucosa IPCLs blood vessel morphological distribution | |
| Al-Gburi et al. | Optical disk segmentation in human retina images with golden eagle optimizer | |
| WO2023216609A1 (en) | Target behavior recognition method and apparatus based on visual-audio feature fusion, and application | |
| Zou et al. | Deep learning and its application in diabetic retinopathy screening |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190827 |
|
| WD01 | Invention patent application deemed withdrawn after publication |